.\" t '\" vim:set syntax=groff:
.\" Copyright (C) 2009-2022 Kaz Kylheku <kaz@kylheku.com>.
.\" All rights reserved.
.\"
.\" Redistribution and use in source and binary forms, with or without
.\" modification, are permitted provided that the following conditions are met:
.\"
.\" 1. Redistributions of source code must retain the above copyright notice, this
.\" list of conditions and the following disclaimer.
.\"
.\" 2. Redistributions in binary form must reproduce the above copyright notice,
.\" this list of conditions and the following disclaimer in the documentation
.\" and/or other materials provided with the distribution.
.\"
.\" THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND
.\" ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
.\" WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
.\" DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE
.\" FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
.\" DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
.\" SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
.\" CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
.\" OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
.\" OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
.\"
.\" Useful groff definitions.
.\"
.\" Some constants that depend on troff/nroff mode:
.ie n \{\
.ds vspc 1
.\}
.el \{\
.ds vspc 0.5
.\}
.\" Mount numeric fonts when not running under man2html
.if !\n(M2 \{\
. fp 4 CR
. fp 5 CI
.\}
.\" Base font
.nr fsav 1
.\" start of code block: switch to monospace font and no format
.de verb
. ft 4
. nf
..
.\" end of code block: restore font and formatting
.de brev
. fi
. ft 1
..
.\" switch to mono font
.de mono
. ft 4
..
.\" switch back from mon font
.de onom
. ft 1
..
.\" typeset argument in monospace
.\" .code x -> \f[CR]x\f[]
.de code
\f[4]\\$1\f[]
..
.\" like .code typesets meta-syntax
.\" which is done in angle brackets + italic in nroff or oblique
.\" courier in PDF/HTML.
.de meta
. ie n \{\
\fI<\\$1>\fP
. \}
. el \{\
\f[5]\\$1\f[]
. \}
..
.\" like .meta but tack on second argument with no space.
.de metn
. ie n \{\
\fI<\\$1>\fP\\$2
. \}
. el \{\
\f[5]\\$1\f[]\\$2
. \}
..
.\" like .code but wraps in quotes
.\" .str x y z -> \f[CR]"x y z"\f[].
.de str
\f[4]"\\$*"\f[]
..
.\" wrap first argument in quotes, tack no second one with no space
.\" .strn x y -> \f[CR]"x"\f[]y.
.de strn
\f[4]"\\$1"\f[]\\$2
..
.\" like .IP but use monospace too
.de coIP
. IP "\\f[4]\\$*\\f[]"
..
.\" Directive heading
.de dir
. NP* The \f[4]\\$1\f[] directive
..
.\" Multiple directive heading
.de dirs
. ds s "
. while (\\n[.$]>2) \{\
. as s \f[4]\\$1\f[],
. shift
. \}
. if (\\n[.$]>1) \{\
. as s \f[4]\\$1\f[]
. shift
. \}
. if (\\n[.$]>0) \{\
. as s and \f[4]\\$1\f[]
. \}
. NP* The \\*s directives
..
.\" heading with code in position 1
.de c1NP
. ds s \\f[4]\\$1\\f[]
. shift
. as s " \\$*
. NP* \\*s
..
.\" utility macro for gathering material into "s" string.
.\" a pair of arguments "@ arg" becomes arg set in code
.\" a pair of arguments "@, arg" becomes "arg," where
.\" arg is set in code, followed by comma not in code.
.de gets
. ds s "
. while (\\n[.$]>0) \{\
. ie "\\$1"@" \{\
. shift
. as s \f[4]\\$1\f[]
. shift
. \}
. el \{\
. ie "\\$1"@," \{\
. shift
. as s \f[4]\\$1\f[],
. shift
. \}
. el \{\
. as s \\$1
. shift
. \}
. \}
. \}
..
.\" a macro for gathering material into "s"
.\" a pair of arguments "< arg" is typeset like
.\" .meta arg. "<< arg arg" is like .metn arg arg.
.ie n \{\
. de getm
. ds s "
. while (\\n[.$]>0) \{\
. ie "\\$1"<" \{\
. shift
. as s \fI<\\$1>\fP
. shift
. \}
. el \{\
. ie "\\$1"<<" \{\
. shift
. as s \fI<\\$1>\fP\\$2
. shift
. shift
. \}
. el \{\
. ie "\\$1">>" \{\
. shift
. as s \\$1\fI<\\$2>\fP
. shift
. shift
. \}
. el \{\
. ie "\\$1"<>" \{\
. shift
. as s \\$1\fI<\\$2>\fP\\$3
. shift
. shift
. shift
. \}
. el \{\
. ie "\\$1"><" \{\
. shift
. as s \fI<\\$1>\fP\\$2\fI<\\$3>\fP
. shift
. shift
. shift
. \}
. el \{\
. ie "\\$1"<2>" \{\
. shift
. as s \\$1\fI<\\$2>\fP\\$3\fI<\\$4>\fP\\$5
. shift
. shift
. shift
. shift
. shift
. \}
. el \{\
. as s \\$1
. shift
. \}
. \}
. \}
. \}
. \}
. \}
. \}
. .
.\}
.el \{\
. de getm
. ds s "
. while (\\n[.$]>0) \{\
. ie "\\$1"<" \{\
. shift
. as s \\f5\\$1\\f4
. shift
. \}
. el \{\
. ie "\\$1"<<" \{\
. shift
. as s \\f5\\$1\\f4\\$2
. shift
. shift
. \}
. el \{\
. ie "\\$1">>" \{\
. shift
. as s \\$1\\f5\\$2\\f4
. shift
. shift
. \}
. el \{\
. ie "\\$1"<>" \{\
. shift
. as s \\$1\\f5\\$2\\f4\\$3
. shift
. shift
. shift
. \}
. el \{\
. ie "\\$1"<2>" \{\
. shift
. as s \\$1\\f5\\$2\\f4\\$3\\f5\\$4\\f4\\$5
. shift
. shift
. shift
. shift
. shift
. \}
. el \{\
. ie "\\$1"><" \{\
. shift
. as s \\f5\\$1\\f4\\$2\f5\\$3\\f4
. shift
. shift
. shift
. \}
. el \{\
. as s \\$1
. shift
. \}
. \}
. \}
. \}
. \}
. \}
. \}
. .
.\}
.\" typeset left argument in monospace, then right one
.\" in previous font, with no space between.
.\" .codn x y \f[CR]x\f[]y
.de codn
\f[4]\\$1\f[]\\$2
..
.\" .cod1 a b c -> abc where a is typeset as code
.de cod1
\&\\$1\f[4]\\$2\f[]\$3
..
.\" .cod2 a b -> ab where b is typeset as code
.de cod2
\&\\$1\f[4]\\$2\f[]
..
.\" .cod3 a b c -> abc where a and c are typeset as code
.de cod3
\f[4]\\$1\f[]\\$2\f[4]\\$3\f[]
..
.\" Syntax section markup
.de synb
. TP* Syntax:
. mono
..
.de syne
. onom
..
.\" Used for meta-variables in syntax blocks
.de mets
. nr fsav \\n[.f]
. getm \\$*
. \"workaround for man2html:
. as s \\f\\n[fsav]
\\*s
. ft \\n[fsav]
..
.\" Used for meta-variables in inline blocks
.de meti
. nr fsav \\n[.f]
. getm \\$*
. \"workaround for man2html:
. as s \\f\\n[fsav]
\&\\*s
. ft \\n[fsav]
..
.\" Used for meta-variables in .coIP
.de meIP
. nr fsav \\n[.f]
. getm \\$*
. \"workaround for man2html:
. as s \\f\\n[fsav]
.coIP \\*s
. ft \\n[fsav]
..
.\" Description section
.de desc
. TP* Description:
..
.\" Section counters: heading, section, paragraph.
.nr shco 0 1
.nr ssco 0 1
.nr spco 0 1
.\" wrapper for .SH
.de SH*
. SH \\n+[shco] \\$*
. rs
. nr ssco 0
. nr spco 0
. sp \*[vspc]
. ns
..
.\" wrapper for .SS
.de SS*
. SS \\n[shco].\\n+[ssco] \\$*
. rs
. nr spco 0
. sp \*[vspc]
. ns
..
.\" wrapper for .TP
.de TP*
. ds s \\$1
. shift
. TP \\$*
\&\\*s
. sp \*[vspc]
. ns
..
.\" numbered paragraph
.de NP*
. ie \n(M2 \{\
. M2SS 2 h4 "\\n[shco].\\n[ssco].\\n+[spco] \\$*"
. \}
. el \{\
. TP* "\f[B]\\n[shco].\\n[ssco].\\n+[spco] \\$*\f[]"
. \}
. PP
..
.\" process arguments using .gets so that some material
.\" is typeset as code. Then pass to .SS* section macro.
.de coSS
. gets \\$*
. SS* \\*s
..
.\" like coSS but targeting NP*
.de coNP
. gets \\$*
. NP* \\*s
..
.\" like coSS but use monospace IP
.de ccIP
. gets \\$*
. IP "\\*s"
..
.\" keystrokes
.ie \n(M2 \{\
.de key
.M2HT <kbd>\\$1</kbd>
..
.de keyn
.M2HT <kbd>\\$1</kbd>\\$2
..
.\}
.el \{\
. ie n \{\
. de key
[\\$1]
. .
. de keyn
[\\$1]\\$2
. .
. \}
. el \{\
. \" Box macro from Groff manual with $2 added
. de box
. nr @wd \w'\\$1'
\h'.2m'\
\h'-.2m'\v'(.2m - \\n[rsb]u)'\
\D'l 0 -(\\n[rst]u - \\n[rsb]u + .4m)'\
\D'l (\\n[@wd]u + .4m) 0'\
\D'l 0 (\\n[rst]u - \\n[rsb]u + .4m)'\
\D'l -(\\n[@wd]u + .4m) 0'\
\h'.2m'\v'-(.2m - \\n[rsb]u)'\
\\$1\
\h'.2m'\\$2
. .
. de key
. box "\\$1" ""
. .
. de keyn
. box "\\$1" "\\$2"
. .
. \}
.\}
.\" TXR name
.ds TX \f[B]TXR\f[]
.ds TL \f[B]TXR Lisp\f[]
.\" Start of man page:
.TH TXR 1 2021-12-28 "Utility Commands" "TXR Programming Language" "Kaz Kylheku"
.SH* NAME
\*(TX \- Programming Language (Version 273)
.SH* SYNOPSIS
.mono
.meti txr [ < options ] [ < script-file [ < arguments ... ]]
.onom
.SH* DESCRIPTION
\*(TX is a general-purpose, multi-paradigm programming language.
It comprises two languages integrated into a single tool: a text
scanning and extraction language referred to as the \*(TX Pattern Language
(sometimes just "TXR"), and a general-purpose dialect of Lisp called \*(TL.
\*(TX can be used for everything from "one liner" data transformation tasks at
the command line, to data scanning and extracting scripts, to full
application development in a wide range of areas.
A script written in the \*(TX Pattern Language, also referred to in this
document as a
.IR query ,
specifies a pattern which matches one or more sources of inputs, such
as text files. Patterns can consist of large chunks of multiline free-form
text, which is matched literally against material in the input sources. Free
variables occurring in the pattern (denoted by the
.code @
symbol) are bound to the pieces of text occurring in the
corresponding positions. Patterns can be arbitrarily complex,
and can be broken down into named pattern functions, which may be mutually
recursive.
In addition to embedded variables which implicitly match text, the
\*(TX pattern language supports a number of directives, for matching text using
regular expressions, for continuing a match in another file, for searching
through a file for the place where an entire subquery matches, for collecting
lists, and for combining subqueries using logical conjunction, disjunction and
negation, and numerous others.
Patterns can contain actions which transform data and generate output.
These actions can be embedded anywhere within the pattern-matching logic.
A common structure for small \*(TX scripts is to perform a complete matching
session at the top of the script, and then deal with processing
and reporting at the bottom.
The \*(TL language can be used from within \*(TX scripts as an
embedded language, or completely standalone. It supports functional,
imperative and object-oriented programming, and provides numerous data types
such as symbols, strings, vectors, hash tables with weak reference support,
lazy lists, and arbitrary-precision ("bignum") integers. It has an expressive
foreign function interface (FFI) for calling into libraries and other software
components that support C-language-style calls.
\*(TL source files as well as individual functions can be optionally compiled
for execution on a virtual machine that is built into \*(TX. Compiled files
execute and load faster, and resist reverse-engineering. Standalone
application delivery is possible.
\*(TX is free software offered under the two-clause BSD license which
places almost no restrictions on redistribution, and allows every conceivable
use, of the whole software or any constituent part, royalty-free, free of
charge, and free of any restrictions.
.SH* ARGUMENTS AND OPTIONS
If \*(TX is given no arguments, it will enter into an interactive
mode. See the INTERACTIVE LISTENER section for a
description of this mode. When \*(TX enters interactive mode this
way, it prints a one-line banner announcing the program name and version,
and one line of help text instructing the user how to exit.
Unless the
.code -c
or
.code -f
options are present, the first non-option argument is treated as a
.meta script-file
which is executed. This is described after the following
descriptions of all of the options.
Any additional arguments have no fixed meaning; they are available to the
\*(TX query or \*(TL application for specifying input files to be processed, or
other meanings under the control of the application.
Options which don't take an argument may be combined together.
The
.code -v
and
.code -q
options are mutually exclusive. Of these two, the one which
occurs in the rightmost position in the argument list dominates.
The
.code -c
and
.code -f
options are also mutually exclusive; if both are specified,
it is a fatal error.
.meIP >> -D var=value
Bind the variable
.meta var
to the value
.meta value
prior to processing the query. The name is in scope over the entire
query, so that all occurrences of the variable are substituted and
match the equivalent text. If the value contains commas, these
are interpreted as separators, which give rise to a list value.
For instance
.code -Dvar=a,b,c
binds
.code var
to the list of the strings
.strn "a" ,
.str "b"
and
.strn "c" .
(See the
.code @(collect)
directive.)
List variables provide a multiple match.
That is to say, if a list variable occurs in a query, a successful
match occurs if any of its values matches the text. If more than one
value matches the text, the first one is taken.
.meIP >> -D var
Binds the variable
.meta var
to an empty string value prior to processing the query.
.coIP -q
Quiet operation during matching. Certain error messages are not reported on the
standard error device (but if the situations occur, they still fail the
query). This option does not suppress error generation during the parsing
of the query, only during its execution.
.coIP -i
If this option is present, then \*(TX will enter into an interactive
interpretation mode after processing all options, and the input query
if one is present. See the INTERACTIVE LISTENER section for a
description of this mode.
.coIP -d
.coIP --debugger
Invoke the interactive \*(TX debugger. See the DEBUGGER section.
Implies
.codn --backtrace .
.coIP --backtrace
Turns on the establishment of backtrace frames for function calls so that a
backtrace can be produced when an unhandled exception occurs, and in other
situations. Backtraces are helpful in identifying the causes of errors, but
require extra stack space and slow down execution.
.coIP -n
.coIP --noninteractive
This option affects behavior related to \*(TX's
.code *stdin*
stream. It also has a another, unrelated effect, on the
behavior of the interactive listener; see below.
Normally, if this stream is connected to a terminal device, it is
automatically marked as having the real-time property when \*(TX starts up (see
the functions
.code stream-set-prop
and
.codn real-time-stream-p ).
The
.code -n
option suppresses this behavior; the
.code *stdin*
stream remains ordinary.
The \*(TX pattern language reads standard input via
a lazy list, created by applying the
.code lazy-stream-cons
function to the
.code *stdin*
stream. If that stream is marked real-time, then the lazy list which is
returned by that function has behaviors that are better suited for scanning
interactive input. A more detailed explanation is given under the description
of this function.
If the
.code -n
option is effect and \*(TX enters into the interactive listener,
the listener operates in
.IR "plain mode" .
The listener reads buffered lines
from the operating system without any character-based editing features
or history navigation. In plain mode, no prompts appear and no
terminal control escape sequences are generated. The only output is
the results of evaluation, related diagnostic messages, and any output
generated by the evaluated expressions themselves.
.coIP -v
Verbose operation. Detailed logging is enabled.
.coIP -b
This option binds a Lisp global lexical variable (as if by the
.code defparml
function) to an object described by Lisp syntax. It requires an argument
of the form
.meta sym=value
where
.meta sym
must be, syntactically, a token denoting a bindable symbol, and
.meta value
is arbitrary \*(TL syntax. The
.meta sym
syntax is converted to the symbol it denotes, which is bound as a global
lexical variable, if it is not already a variable.
The
.meta value
syntax is parsed to the Lisp object it denotes. This object is not subject
to evaluation; the object itself is stored into the variable binding denoted
by
.metn sym .
Note that if
.meta sym
already exists as a global variable, then it is simply overwritten. If
.meta sym
is marked special, then it stays special.
.coIP -B
If the query is successful, print the variable bindings as a sequence
of assignments in shell syntax that can be
.IR eval -ed
by a POSIX shell.
II the query fails, print the word "false". Evaluation of this word
by the shell has the effect of producing an unsuccessful termination
status from the shell's
.I eval
command.
.coIP -l
.coIP --lisp-bindings
This option implies
.codn -B .
Print the variable bindings in Lisp syntax instead of
shell syntax.
.meIP -a < num
This option implies
.codn -B .
The decimal integer argument
.meta num
specifies the maximum
number of array dimensions to use for list-valued variable bindings.
The default is 1. Additional dimensions are expressed using numeric suffixes
in the generated variable names.
For instance, consider the three-dimensional list arising out of a triply
nested collect:
.mono
((("a" "b") ("c" "d")) (("e" "f") ("g" "h"))).
.onom
Suppose this is bound to a variable
.metn V .
With
.codn "-a 1" ,
this will be
reported as:
.verb
V_0_0[0]="a"
V_0_1[0]="b"
V_1_0[0]="c"
V_1_1[0]="d"
V_0_0[1]="e"
V_0_1[1]="f"
V_1_0[1]="g"
V_1_1[1]="h"
.brev
With
.codn "-a 2" ,
it comes out as:
.verb
V_0[0][0]="a"
V_1[0][0]="b"
V_0[0][1]="c"
V_1[0][1]="d"
V_0[1][0]="e"
V_1[1][0]="f"
V_0[1][1]="g"
V_1[1][1]="h"
.brev
The leftmost bracketed index is the most major index. That is to say,
the dimension order is:
.codn "NAME_m_m+1_..._n[1][2]...[m-1]" .
.meIP -c < query
Specifies the query in the form of a command-line argument. If this option is
used, the
.meta script-file
argument is omitted. The first non-option argument,
if there is one, now specifies the first input source rather than a query.
Unlike queries read from a file, (nonempty) queries specified as arguments
using -c do not have to properly end in a newline. Internally,
\*(TX adds the missing newline before parsing the query. Thus
.code -c
.str @a
is a valid query which matches a line.
Example:
Shell script which uses \*(TX to read two lines
.str 1
and
.str 2
from standard input,
binding them to variables
.code a
and
.codn b .
Standard
input is specified as
.code -
and the data
comes from shell "here document" redirection:
.RS
.IP code:
.mono
\ #!/bin/sh
txr -B -c "@a
@b" - <<!
1
2
!
.onom
.IP output:
.mono
\ a=1
b=2
.onom
.PP
The
.code @;
comment syntax can be used for better formatting:
.verb
txr -B -c "@;
@a
@b"
.brev
.RE
.meIP -f < script-file
Provides a way to specify the file from which the query is to be read,
as an alternative to using the main
.meta script-file
argument. This is useful in
.code #!
("hash bang") scripts. (See Hash-Bang Support below.)
Use of this option does not affect the order of processing. All of the options
are processed first, before the
.meta script-file
is read, as if it were specified by the main
.meta script-file
argument.
If the argument to
.code -f
is
.code -
(dash) then the script will be read from standard input instead
of a file.
If this option is used, the first non-option argument, if there is one,
no longer specifies the
.metn script-file .
It is an argument to the script, such as the name of an input source.
.meIP -e < expression
Evaluates a \*(TL expression for its side effects, without printing
its value. Can be specified more than once. The
.meta script-file
argument becomes optional if at least one
.codn -e ,
.codn -p ,
.code -P
or
.code -t
option is processed. If the evaluation of every
.meta expression
evaluated this way terminates normally, and there is no
.meta script-file
argument, then \*(TX terminates with a successful status.
.meIP -p < expression
Just like
.code -e
but prints the value of
.meta expression
using the
.code prinl
function.
.meIP -P < expression
Like
.code -p
but prints using the
.code pprinl
function.
.meIP -t < expression
Like
.code -p
but prints using the
.code tprint
function.
.meIP -C < number
.meIP >> --compat= number
Requests \*(TX to behave in a manner that is compatible with the specified
version of \*(TX. This makes a difference in situations when a release of
\*(TX breaks backward compatibility. If some version N+1 deliberately introduces
a change which is backward incompatible, then
.code "-C N"
can be used to request the old behavior.
The requested value of N can be too low, in which case \*(TX will
complain and exit with an unsuccessful termination status. This indicates
that \*(TX refuses to be compatible with such an old version. Users requiring
the behavior of that version will have to install an older version of \*(TX which
supports that behavior, or even that exact version.
If the option is specified more than once, the behavior is not specified.
Compatibility can also be requested via the
.code TXR_COMPAT
environment variable instead of the
.code -C
option.
For more information, see the COMPATIBILITY section.
.meIP >> --gc-delta= number
The
.meta number
argument to this option must be a decimal integer. It represents
a megabyte value, the "GC delta": one megabyte is 1048576 bytes. The "GC
delta" controls an aspect of the garbage collector behavior.
See the
.code gc-set-delta
function for a description.
.meIP --debug-autoload
This option turns on debugging, like
.code --debugger
but also requests stepping into the autoload processing of
\*(TL library code. Normally, debugging through the evaluations
triggered by autoloading is suppressed.
Implies
.codn --backtrace .
.meIP --debug-expansion
This option turns on debugging, like
.code --debugger
but also requests stepping into the parse-time macro-expansion
of \*(TL code embedded in \*(TX queries. Normally, this is suppressed.
Implies
.codn --backtrace .
.coIP --help
Prints usage summary on standard output, and terminates successfully.
.coIP --license
Prints the software license. This depends on the software being
installed such that the LICENSE file is in the data directory.
Use of \*(TX implies agreement with the liability disclaimer in the license.
.coIP --version
Prints a message on standard output which includes the program version,
and then immediately causes \*(TX to terminate with a successful status.
.coIP --build-id
If \*(TX was built with an embedded build ID string, this
option prints that string. Otherwise nothing is printed.
In either case, \*(TX then immediately terminates with a successful
status.
.coIP --args
The
.code --args
option provides a way to encode multiple arguments as a single
argument, which is useful on some systems which have limitations in
their implementation of the hash-bang mechanism. For details about
its special syntax, see Hash-Bang Support below. It is also useful in
standalone application deployment. See the section
STANDALONE APPLICATION SUPPORT, in which example uses of
.code --args
are shown.
.coIP --eargs
The
.code --eargs
option (extended
.codn --args )
is like
.code --args
but must be followed by an argument. The argument is removed from
the argument list and substituted in place of occurrences of
.code {}
among the arguments expanded from the
.code --eargs
syntax.
.coIP --lisp
.coIP --compiled
These options influence the treatment of query files which do not have
a suffix indicating their type. The
.code --lisp
option causes an unsuffixed file to be treated as Lisp source;
.code --compiled
causes it to be treated as a compiled file.
Moreover, if
.code --lisp
is specified, and an unsuffixed file does not exist, then \*(TX
will add the
.str .tl
suffix and try the file again; and
.code --compiled
will similarly add the
.str .tlo
suffix and try opening the file again.
In the same situation, if neither
.code --lisp
nor
.code --compiled
has been specified, \*(TX will first try adding the
.str .txr
suffix. If that fails,
then the
.str .tlo
suffix will be tried and finally
.strn .tl .
Note that
.code --lisp
and
.code --compiled
influence how the argument of the
.code -f
option is treated, but only if they precede that option.
.coIP --reexec
On platforms which support the POSIX
.code exec
family of functions, this option causes \*(TX to re-execute itself.
The re-executed image receives the remaining arguments which follow
the
.code --reexec
argument. Note: this option is useful for supporting setuid operation in
hash-hang scripts. On some platforms, the interpreter designated by
a hash-bang script runs without altered privilege, even if that
interpreter is installed setuid. If the interpreter is executed directly,
then setuid applies to it, but not if it is executed via hash bang.
If the
.code --reexec
option is used in the interpreter command line of such a script, the
interpreter will re-execute itself, thereby gaining the setuid privilege.
The re-executed image will then obtain the script name from the arguments
which are passed to it and determine whether that script will run setuid.
See the section SETUID/SETGID OPERATION.
.coIP --noprofile
If entering the interactive listener, suppress the reading of the
.code .txr_profile
in the home directory. See the Interactive Profile File subsection in the
INTERACTIVE LISTENER section of the manual.
.coIP --gc-debug
This option enables a behavior which stresses the garbage collector with
frequent garbage collection requests. The purpose is to make it more likely
to reproduce certain kinds of bugs. Use of this option severely degrades
the performance of \*(TX.
.coIP --vg-debug
If \*(TX is enabled with Valgrind support, then this option is available.
It enables code which uses the Valgrind API to integrate with the Valgrind
debugger, for more accurate tracking of garbage collected objects. For
example, objects which have been reclaimed by the garbage collector
are marked as inaccessible, and marked as uninitialized when they are
allocated again.
.coIP --free-all
This option specifies that all memory allocated by \*(TX should be freed upon
normal termination. This behavior is useful for debugging memory leaks.
An accurate leak detection tool, such as the one built into Valgrind,
should report zero leaked or still reachable memory if
.code --free-all
has been used and \*(TX has terminated normally.
that indicates either a leak in \*(TX, a leak or global object retention
in a platform library, or else a a leak introduced due to misuse of FFI.
.coIP --dv-regex
If this option is used, then regular expressions are all treated using the
derivative-based back-end. The NFA-based regex implementation is disabled.
Normally, only regular expressions which require the intersection and
complement operators are handled using the derivative back-end.
This option makes it possible to test that back-end on test cases that it
wouldn't normally receive.
.coIP --
Signifies the end of the option list.
.coIP -
This argument is not interpreted as an option, but treated as a filename
argument. After the first such argument, no more options are recognized. Even
if another argument looks like an option, it is treated as a name.
This special argument
.code -
means "read from standard input" instead of a file.
The
.metn script-file ,
or any of the data files, may be specified using this option.
If two or more files are specified as
.codn - ,
the behavior is system-dependent.
It may be possible to indicate EOF from the interactive terminal, and
then specify more input which is interpreted as the second file, and so forth.
.PP
After the options, the remaining arguments are treated as follows.
If neither the
.code -f
nor the
.code -c
options were specified, then the first argument is treated as the
.metn script-file .
If no arguments are present, then \*(TX
enters interactive mode, provided that none of the
.codn -e ,
.codn -p ,
.code -P
or
.code -t
options had been processed, in which case it instead terminates.
The \*(TX Pattern Language has features for implicitly treating
the subsequent command-line arguments as input files.
It follows the convention that an argument consisting of a single
.code -
(dash) character specifies that standard input is to be used,
instead of opening a file. If the query does not use the
.code @(next)
directive to select an alternative data source, and a pattern-matching
construct is processed which demands data, then the first argument
will be opened as a data source. Arguments not opened as data sources
can be assigned alternative meanings and uses, or can be ignored
entirely, under control of the query.
Specifying standard input as a source with an explicit
.code -
argument is unnecessary. If no arguments are present, then
\*(TX scans standard input by default. This was not true in versions of \*(TX
prior to 171; see the COMPATIBILITY section.
.PP
\*(TX begins by reading the script, which is given as the
contents of the argument of the
.code -c
option, or else as the contents of an input source specified by the
.code -f
option or by the
.meta script-file
argument. If
.code -f
or the
.meta script-file
argument specify
.code -
(dash) then the script is read from standard input.
In the case of the \*(TX pattern language,
the entire query is scanned, internalized, and then begins executing, if it is
free of syntax errors. (\*(TL is processed differently, form by form.) On the
other hand, the pattern language reads data files in a lazy manner. A file
isn't opened until the query demands material from that file, and then the
contents are read on demand, not all at once.
The suffix of the
.meta script-file
is significant. If the name has no suffix, or if it has a
.str .txr
suffix, then it is assumed to be in the \*(TX pattern language. If it has
the
.str .tl
suffix, then it is assumed to be \*(TL. The
.code --lisp
option changes the treatment of unsuffixed script file names, causing them
to be interpreted as \*(TL.
If an unsuffixed script file name is specified, and cannot be opened, then
\*(TX will add the
.str .txr
suffix and try again. If that fails, it will be tried with the
.str .tl
suffix, and treated as \*(TL.
If the
.code --lisp
option has been specified, then \*(TX tries only the
.str .tl
suffix.
A \*(TL file is processed as if by the
.code load
macro: forms from the file are read and evaluated. If the forms do not terminate
the \*(TX process or throw an exception, and there are no syntax errors, then
\*(TX terminates successfully after evaluating the last form. If syntax errors
are encountered in a form, then \*(TX terminates unsuccessfully.
\*(TL is documented in the section TXR LISP.
If a query file is specified, but no file arguments,
it is up to the query to open a file, pipe or standard input via the
.code @(next)
directive
prior to attempting to make a match. If a query attempts to match text,
but has run out of files to process, the match fails.
.SH* STATUS AND ERROR REPORTING
\*(TX sends errors and verbose logs to the standard error device. The following
paragraphs apply when \*(TX is run without enabling verbose mode with
.codn -v ,
or the printing of variable
bindings with
.code -B
or
.codn -a .
If the command-line arguments are incorrect, \*(TX issues an error diagnostic
and terminates with a failed status.
If the
.meta script-file
specifies a query, and the query has a malformed syntax, \*(TX likewise
issues error diagnostics and terminates with a failed status.
If the query fails due to a mismatch, \*(TX terminates
with a failed status. No diagnostics are issued.
If the query is well-formed, and matches, then \*(TX issues
no diagnostics, and terminates with a successful status.
In verbose mode (option
.codn -v ),
\*(TX issues diagnostics on the standard error device even in situations which
are not erroneous.
In bindings-printing mode (options
.code -B
or
.codn -a ),
\*(TX prints the word
.code false
if the query fails, and exits with a failed
termination status. If the query succeeds, the variable bindings, if any,
are output on standard output.
If the
.meta script-file
is \*(TL, then it is processed form by form. Each top-level Lisp form
is evaluated after it is read. If any form is syntactically malformed,
\*(TX issues diagnostics and terminates unsuccessfully. This is somewhat
different from how the pattern language is treated: a script in the pattern
language is parsed in its entirety before being executed.
.SH* BASIC TXR SYNTAX
.SS* Comments
A query may contain comments which are delimited by the sequence
.code @;
and extend to the end of the line. Whitespace can occur between the
.code @
and
.codn ; .
A comment which begins on a line swallows that entire line, as well as the
newline which terminates it. In essence, the entire comment line disappears.
If the comment follows some material in a line, then it does not consume
the newline. Thus, the following two queries are equivalent:
.IP 1.
.mono
\ @a@; comment: match whole line against variable @a
@; this comment disappears entirely
@b
.onom
.IP 2.
.mono
\ @a
@b
.onom
.PP
The comment after the
.code @a
does not consume the newline, but the
comment which follows does. Without this intuitive behavior,
line comment would give rise to empty lines that must match empty
lines in the data, leading to spurious mismatches.
Instead of the
.code ;
character, the
.code #
character can be used. This is an obsolescent feature.
.SS* Hash-Bang Support
\*(TX has several features which support use of the hash-bang convention
for creating apparently standalone executable programs.
.NP* Basic Hash Bang
Special processing is applied to \*(TX query or \*(TL script files that are
specified on the command line via the
.code -f
option or as the first non-option argument. If the first line of such
a file begins with the characters
.codn #! ,
that entire line is consumed and processed specially.
This removal allows
for \*(TX queries to be turned into standalone executable programs in the POSIX
environment using the hash-bang mechanism. Unlike most interpreters,
\*(TX applies special processing to the
.code #!
line, which is described below, in the section
.BR "Argument Generation with the Null Hack" .
Shell session example: create a simple executable program called
.str "twoline.txr"
and
run it. This assumes \*(TX is installed in
.codn /usr/bin .
.verb
$ cat > hello.txr
#!/usr/bin/txr
@(bind a "Hey")
@(output)
Hello, world!
@(end)
$ chmod a+x hello.txr
$ ./hello.txr
Hello, world!
.brev
When this plain hash-bang line is used, \*(TX receives the name of the script
as an argument. Therefore, it is not possible to pass additional options
to \*(TX. For instance, if the above script is invoked like this
.verb
$ ./hello.txr -B
.brev
the
.code -B
option isn't processed by \*(TX, but treated as an additional argument,
just as if
.mono
.meti txr < script-file -B
.onom
had been executed directly.
This behavior is useful if the script author wants not to expose the
\*(TX options to the user of the script.
However, the hash-bang line can use the
.code -f
option:
.verb
#!/usr/bin/txr -f
.brev
Now, the name of the script is passed as an argument to the
.code -f
option, and \*(TX will look for more options after that, so that the resulting
program appears to accept \*(TX options. Now we can run
.verb
$ ./hello.txr -B
Hello, world!
a="Hey"
.brev
The
.code -B
option is honored.
.coNP Argument Generation with @ --args and @ --eargs
On some operating systems, it is not possible to pass more than one
argument through the hash-bang mechanism. That is to say, this will
not work.
.verb
#!/usr/bin/txr -B -f
.brev
To support systems like this, \*(TX supports the special argument
.codn --args ,
as well as an extended version,
.codn --eargs .
With
.codn --args ,
it is possible to encode multiple arguments
into one argument. The
.code --args
option must be followed by a separator
character, chosen by the programmer. The characters after that are
split into multiple arguments on the separator character. The
.code --args
option is then removed from the argument list and replaced with these
arguments, which are processed in its place.
Example:
.verb
#!/usr/bin/txr --args:-B:-f
.brev
The above has the same behavior as
.verb
#!/usr/bin/txr -B -f
.brev
on a system which supports multiple arguments in the hash-bang line.
The separator character is the colon, and so the remainder
of that argument,
.codn -B:-f ,
is split into the two arguments
.codn "-B -f" .
The
.code --eargs
option is similar to
.codn --args ,
but must be followed by one more argument.
After
.code --eargs
performs the argument splitting in the same manner as
.codn --args ,
any of the arguments which it produces which are the
two-character sequence
.code {}
are replaced with that following argument. Whether
or not the replacement occurs, that following argument
is then removed.
Example:
.verb
#!/usr/bin/txr --eargs:-B:{}:--foo:42
.brev
This has an effect which cannot be replicated in any known
implementation of the hash-bang mechanism. Suppose
that this hash-bang line is placed in a script called
.codn script.txr .
When this script is invoked with arguments, as in:
.verb
script.txr a b c
.brev
then \*(TX is invoked similarly to:
.verb
/usr/bin/txr --eargs:-B:{}:--foo:42 script.txr a b c
.brev
Then, when
.code --eargs
processing takes place, firstly the argument sequence
.verb
-B {} --foo 42
.brev
is produced by splitting into four fields using the
.code :
(colon) character as the separator.
Then, within these four fields, all occurrences of
.code {}
are replaced with the following argument
.codn script.txr ,
resulting in:
.verb
-B script.txr --foo 42
.brev
Furthermore, that
.code script.txr
argument is removed from the remaining argument list.
The four arguments are then substituted in place of the original
.code --eargs:-B:{}:--foo:42
syntax.
The resulting \*(TX invocation is, therefore:
.verb
/usr/bin/txr -B script.txr --foo 42 a b c
.brev
Thus,
.code --eargs
allows some arguments to be encoded into the interpreter script, such that
script name is inserted anywhere among them, possibly multiple times. Arguments
for the interpreter can be encoded, as well as arguments to be processed by the
script.
.coNP Argument Generation with the Null Hack
The
.code --args
and
.code --eargs
mechanisms do not solve the following problem: the POSIX
.code env
utility is often exploited for its
.code PATH
searching capability, and used to express hash-bang scripts in the following
way:
.verb
#!/usr/bin/env txr
.brev
Here, the
.code env
utility searches for the
.code txr
program in the directories indicated by the
.code PATH
variable, which liberates the script from having to encode the exact location
where the program is installed. However, if the operating system allows only
one argument in the hash-bang mechanism, then no arguments can be passed
to the program.
To mitigate this problem,
\*(TX
supports a special feature in its hash-bang support. If the hash-bang
line contains a null byte, then the text from after the null byte
until the end of the line
is split into fields using the space character as a separator, and these
fields are inserted into the command line. This manipulation happens during
command-line processing, i.e. prior to the execution of the file.
If this processing is applied to a file that is specified using the
.code -f
option, then the arguments which arise from the special processing
are inserted after that option and its argument. If this processing is
applied to the file which is the first non-option argument, then the
options are inserted before that argument. However, care is taken not
to process that argument a second time.
In either situation, processing of the command-line options continues, and the
arguments which are processed next are the ones which were just inserted. This
is true even if the options had been inserted as a result of processing the
first non-option argument, which would ordinarily signal the termination of
option processing.
In the following examples, it is assumed that the script is
named, and invoked, as
.codn /home/jenny/foo.txr ,
and is given arguments
.codn "--bar abc" ,
and that
.code txr
resolves to
.codn /usr/bin/txr .
The
.code <NUL>
code indicates a literal ASCII NUL character (the zero byte).
Basic example:
.verb
#!/usr/bin/env txr<NUL>-a 3
.brev
Here,
.code env
searches for
.codn txr ,
finding it in
.codn /usr/bin .
Thus, including the executable name, \*(TX receives this full argument list:
.verb
/usr/bin/txr /home/jenny/foo.txr --bar abc
.brev
The first non-option argument is the name of the script. \*(TX opens
the script, and notices that it begins with a hash-bang line.
It consumes the hash-bang line and finds the null byte inside it,
retrieving the character string after it, which is
.strn "-a 3" .
This is split into the two arguments
.code -a
and
.codn 3 ,
which are then inserted into the command line ahead of the
the script name. The effective command line then becomes:
.verb
/usr/bin/txr -a 3 /home/jenny/foo.txr --bar abc
.brev
Command-line option processing continues, beginning with the
.code -a
option. After the option is processed,
.code /home/jenny/foo.txr
is encountered again. This time it is not opened a second time;
it signals the end of option processing, exactly as it would immediately
do if it hadn't triggered the insertion of any arguments.
Advanced example: use
.code env
to invoke
.codn txr ,
passing options to the interpreter and to the script:
.verb
#!/usr/bin/env txr<NUL>--eargs:-C:175:{}:--debug
.brev
This example shows how
.code --eargs
can be used in conjunction with the null hack. When
.code txr
begins executing, it receives the arguments
.verb
/usr/bin/txr /home/jenny/foo.txr
.brev
The script file is opened, and the arguments delimited by the
null character in the hash-bang line are inserted, resulting
in the effective command line:
.verb
/usr/bin/txr --eargs:-C:175:{}:--debug /home/jenny/foo.txr
.brev
Next,
.code --eargs
is processed in the ordinary way, transforming the command line
into:
.verb
/usr/bin/txr -C 175 /home/jenny/foo.txr --debug
.brev
The name of the script file is encountered, and signals the end
of option processing. Thus
.code txr
receives the
.code -C
option, instructing it to emulate some behaviors from version 175,
and the
.code /home/jenny/foo.txr
script receives
.code --debug
as
.B its
argument: it executes with the
.code *args*
list containing one element, the character string
.strn --debug .
The hash-bang null-hack feature was introduced in \*(TX 177.
Previous versions ignore the hash-bang line, performing no special
processing. Where a risk exists that programs which depend on the
feature might be executed by an older version of \*(TX, care must
be taken to detect and handle that situation, either by means of the
.code txr-version
variable, or else by some logic which infers that the processing of the
hash-bang line hasn't been performed.
.coNP Passing Options to \*(TX via Hash-Bang Null Hack
It is possible to use the Hash-Bang Null Hack, such that the resulting
executable program recognizes \*(TX options. This is made possible by
a special behavior in the processing of the
.code -f
option.
For instance, suppose that the effect of the following familiar hash-bang line
is required:
.verb
#!/path/to/txr -f
.brev
However, suppose there is also a requirement to use the
.code env
utility to find \*(TX. Furthermore, the operating system allows only one
hash-bang argument. Using the Null Hack, this is rewritten as:
.verb
#!/usr/bin/env txr<NUL>-f
.brev
then if the script is invoked with arguments
.codn "-a b c" ,
the command line will ultimately be transformed into:
.verb
/path/to/txr -f /path/to/scriptfile -i a b c
.brev
which allows \*(TX to process the
.code -i
option, leaving
.codn a ,
.code b
and
.code c
as arguments for the script.
However, note that there is a subtle issue with the
.code -f
option that has been inserted via the Null Hack: namely, this
insertion happens after
\*(TX has opened the script file and read the hash-bang line from it.
This means that when the inserted
.code -f
option is being processed, the script file is already open.
A special behavior occurs. The
.code -f
option processing notices that the argument to
.code -f
is identical to the pathname of name of the script file that \*(TX has
already opened for processing. The
.code -f
option and its argument are then skipped.
.NP* Hash Bang and Setuid
\*(TX supports setuid hash-bang scripting, even on platforms that do not
support setuid and setgid attributes on hash-bang scripts. On such
platforms, \*(TX has to be installed setuid/setgid. See the section
SETUID/SETGID OPERATION. On some platforms, it may also be necessary to
to use the
.code --reexec
option.
.SS* Whitespace
Outside of directives, whitespace is significant in \*(TX queries, and represents
a pattern match for whitespace in the input. An extent of text consisting of
an undivided mixture of tabs and spaces is a whitespace token.
Whitespace tokens match a precisely identical piece of whitespace in the input,
with one exception: a whitespace token consisting of precisely one space has a
special meaning. It is equivalent to the regular expression
.codn "@/[ ]+/" :
match an extent of one or more spaces (but not tabs!). Multiple consecutive
spaces do not have this meaning.
Thus, the query line
.str "a b"
(one space between
.code a
and
.codn b )
matches
.str "a b"
with any number of spaces between the two letters.
For matching a single space, the syntax
.code "@\e "
can be used (backslash-escaped space).
It is more often necessary to match multiple spaces than to
match exactly one space, so this rule simplifies many queries
and inconveniences only a few.
In output clauses, string and character literals and quasiliterals, a space
token denotes a space.
.SS* Text
Query material which is not escaped by the special character
.code @
is literal text, which matches input character for character. Text which occurs at
the beginning of a line matches the beginning of a line. Text which starts in
the middle of a line, other than following a variable, must match exactly at
the current position, where the previous match left off. Moreover, if the text
is the last element in the line, its match is anchored to the end of the line.
An empty query line matches an empty line in the input. Note that an
empty input stream does not contain any lines, and therefore is not matched
by an empty line. An empty line in the input is represented by a newline
character which is either the first character of the file, or follows
a previous newline-terminated line.
Input streams which end without terminating their last line with a newline are
tolerated, and are treated as if they had the terminator.
Text which follows a variable has special semantics, described in the
section Variables below.
A query may not leave a line of input partially matched. If any portion of a
line of input is matched, it must be entirely matched, otherwise a matching
failure results. However, a query may leave unmatched lines. Matching only
four lines of a ten-line file is not a matching failure. The
.code eof
directive can be used to explicitly match the end of a file.
In the following example, the query matches the text, even though
the text has an extra line.
.IP code:
.mono
\ Four score and seven
years ago our
.onom
.IP data:
.mono
\ Four score and seven
years ago our
forefathers
.onom
.PP
In the following example, the query
.B fails
to match the text, because the text has extra material on one
line that is not matched:
.IP code:
.mono
\ I can carry nearly eighty gigs
in my head
.onom
.IP data:
.mono
\ I can carry nearly eighty gigs of data
in my head
.onom
.PP
Needless to say, if the text has insufficient material relative
to the query, that is a failure also.
To match arbitrary material from the current position to the end
of a line, the "match any sequence of characters, including empty"
regular expression
.code @/.*/
can be used. Example:
.IP code:
.mono
\ I can carry nearly eighty gigs@/.*/
.onom
.IP data:
.mono
\ I can carry nearly eighty gigs of data
.onom
.PP
In this example, the query matches, since the regular expression
matches the string "of data". (See the Regular Expressions section below.)
Another way to do this is:
.IP code:
.mono
\ I can carry nearly eighty gigs@(skip)
.onom
.SS* Special Characters in Text
Control characters may be embedded directly in a query (with the exception of
newline characters). An alternative to embedding is to use escape syntax.
The following escapes are supported:
.meIP >> @\e newline
A backslash immediately followed by a newline introduces a physical line
break without breaking up the logical line. Material following this sequence
continues to be interpreted as a continuation of the previous line, so
that indentation can be introduced to show the continuation without appearing
in the data.
.meIP >> @\e space
A backslash followed by a space encodes a space. This is useful in line
continuations when it is necessary for some or all of the leading spaces to be
preserved. For instance the two line sequence
.verb
abcd@\e
@\e efg
.brev
is equivalent to the line
.verb
abcd efg
.brev
The two spaces before the
.code @\e
in the second line are consumed. The spaces after are preserved.
.coIP @\ea
Alert character (ASCII 7, BEL).
.coIP @\eb
Backspace (ASCII 8, BS).
.coIP @\et
Horizontal tab (ASCII 9, HT).
.coIP @\en
Line feed (ASCII 10, LF). Serves as abstract newline on POSIX systems.
.coIP @\ev
Vertical tab (ASCII 11, VT).
.coIP @\ef
Form feed (ASCII 12, FF). This character clears the screen on many
kinds of terminals, or ejects a page of text from a line printer.
.coIP @\er
Carriage return (ASCII 13, CR).
.coIP @\ee
Escape (ASCII 27, ESC)
.meIP >> @\ex hex-digits
A
.code @\ex
immediately followed by a sequence of hex digits is interpreted as a hexadecimal
numeric character code. For instance
.code @\ex41
is the ASCII character A. If a semicolon character immediately follows the
hex digits, it is consumed, and characters which follow are not considered
part of the hex escape even if they are hex digits.
.meIP >> @\e octal-digits
A
.code @\e
immediately followed by a sequence of octal digits (0 through 7) is interpreted
as an octal character code. For instance
.code @\e010
is character 8, same as
.codn @\eb .
If a semicolon character immediately follows the octal digits, it is consumed,
and subsequent characters are not treated as part of the octal escape,
even if they are octal digits.
.PP
Note that if a newline is embedded into a query line with
.code @\en,
this does not split the line into two; it's embedded into the line and thus
cannot match anything. However,
.code @\en
may be useful in the
.code @(cat)
directive and
in
.codn @(output) .
.SS* Character Handling and International Characters
\*(TX represents text internally using wide characters, which are used to
represent Unicode code points. Script source code, as well as all data sources,
are assumed to be in the UTF-8 encoding. In \*(TX and \*(TL source, extended
characters can be used directly in comments, literal text, string literals,
quasiliterals and regular expressions. Extended characters can also be
expressed indirectly using hexadecimal or octal escapes.
On some platforms, wide characters may be restricted to 16 bits, so that
\*(TX can only work with characters in the BMP (Basic Multilingual Plane)
subset of Unicode.
\*(TX does not use the localization features of the system library;
its handling of extended characters is not affected by environment variables
like
.code LANG
and
.codn L_CTYPE .
The program reads and writes only the UTF-8 encoding.
\*(TX deals with UTF-8 separately in its parser and in its I/O streams
implementation.
\*(TX's text streams perform UTF-8 conversion internally,
such that \*(TX applications use Unicode code points.
In text streams, invalid UTF-8 bytes are treated as follows. When an invalid
byte is encountered in the middle of a multibyte character, or if the input
ends in the middle of a multibyte character, or if an invalid character is decoded,
such as an overlong from, or code in the range U+DC00 through U+DCFF, the UTF-8
decoder returns to the starting byte of the ill-formed multibyte character, and
extracts just one byte, mapping that byte to the Unicode character range U+DC00
through U+DCFF, producing that code point as the decoded result. The decoder
is then reset to its initial state and begins decoding at the following byte,
where the same algorithm is repeated.
Furthermore, because \*(TX internally uses a null-terminated character
representation of strings which easily interoperates with C language
interfaces, when a null character is read from a stream, \*(TX converts it to
the code U+DC00. On output, this code converts back to a null byte,
as explained in the previous paragraph. By means of this representational
trick, \*(TX can handle textual data containing null bytes.
In contrast to the above, the \*(TX parser scans raw UTF-8 bytes from a binary
stream, rather than using a text stream. The parser performing its own
recognition of UTF-8 sequences in certain language constructs, using a UTF-8
decoder only when processing certain kinds of tokens.
Comments are read without regard for encoding, so invalid encoding bytes in
comments are not detected. A comment is simply a sequence of bytes terminated
by a newline.
Invalid UTF-8 encountered while scanning identifiers and character names in
character literal (hash-backslash) syntax is diagnosed as a syntax error.
UTF-8 in string literals is treated in the same way as UTF-8 in text streams.
Invalid UTF-8 bytes are mapped into code points in the U+DC000 through U+DCFF
range, and incorporated as such into the resulting string object which the
literal denotes. The same remarks apply to regular-expression literals.
.SS* Regular Expression Directives
In place of a piece of text (see section Text above), a regular-expression
directive may be used, which has the following syntax:
.verb
@/RE/
.brev
where the RE part enclosed in slashes represents regular-expression
syntax (described in the section Regular Expressions below).
Long regular expressions can be broken into multiple lines using a
backslash-newline sequence. Whitespace before the sequence or after the
sequence is not significant, so the following two are equivalent:
.verb
@/reg \e
ular/
@/regular/
.brev
There may not be whitespace between the backslash and newline.
Whereas literal text simply represents itself, regular expression denotes a
(potentially infinite) set of texts. The regular-expression directive
matches the longest piece of text (possibly empty) which belongs to the set
denoted by the regular expression. The match is anchored to the current
position; thus if the directive is the first element of a line, the match is
anchored to the start of a line. If the regular-expression directive is the
last element of a line, it is anchored to the end of the line also: the regular
expression must match the text from the current position to the end of the
line.
Even if the regular expression matches the empty string, the match will fail if
the input is empty, or has run out of data. For instance suppose the third line
of the query is the regular expression
.codn @/.*/ ,
but the input is a file which has
only two lines. This will fail: the data has no line for the regular expression to
match. A line containing no characters is not the same thing as the absence of
a line, even though both abstractions imply an absence of characters.
Like text which follows a variable, a regular-expression directive which
follows a variable has special semantics, described in the section Variables
below.
.SS* Variables
Much of the query syntax consists of arbitrary text, which matches file data
character for character. Embedded within the query may be variables and
directives which are introduced by a
.code @
character. Two consecutive
.code @@
characters encode a literal
.codn @ .
A variable-matching or substitution directive is written in one of several
ways:
.mono
.mets >> @ sident
.mets <> @{ bident }
.mets >> @* sident
.mets <> @*{ bident }
.mets >> @{ bident <> / regex /}
.mets >> @{ bident >> ( fun >> [ arg ...])}
.mets >> @{ bident << number }
.mets >> @{ bident << bident }
.onom
The forms with an
.code *
indicate a long match, see Longest Match below.
The forms with the embedded regexp
.mono
.meti <> / regex /
.onom
or function or
.meta number
have special semantics; see Positive Match below.
The identifier
.code t
cannot be used as a name; it is a reserved symbol which
denotes the value true. An attempt to use the variable
.code @t
will result in an exception. The symbol
.code nil
can be used where a variable name is required syntactically,
but it has special semantics, described in a section below.
A
.meta sident
is a "simple identifier" form which is not delimited by
braces.
A
.meta sident
consists of any combination of
one or more letters, numbers, and underscores. It may not look like a number,
so that for instance
.code 123
is not a valid
.metn sident ,
but
.code 12A
is valid. Case is
sensitive, so that
.code FOO
is different from
.codn foo ,
which is different from
.codn Foo .
The braces around an identifier can be used when material which follows would
otherwise be interpreted as being part of the identifier. When a name is
enclosed in braces it is a
.metn bident .
The following additional characters may be used as part of a
.meta bident
which are not allowed in a
.metn sident :
.verb
! $ % & * + - < = > ? \e ~
.brev
Moreover, most Unicode characters beyond U+007F may appear in a
.metn bident ,
with certain exceptions. A character may not be used if it is any of the
Unicode space characters, a member of the high or low surrogate region,
a member of any Unicode private-use area, or is either of the two characters
U+FFFE and U+FFFF. These situations produce a syntax error. Invalid UTF-8
in an identifier is also a syntax error.
The rule still holds that a name cannot look like a number so
.code +123
is not a valid
.meta bident
but these are valid:
.codn a->b ,
.codn *xyz* ,
.codn foo-bar .
The syntax
.code @FOO_bar
introduces the name
.codn FOO_bar ,
whereas
.code @{FOO}_bar
means the
variable named
.str FOO
followed by the text
.strn _bar .
There may be whitespace
between the
.code @
and the name, or opening brace. Whitespace is also allowed in the
interior of the braces. It is not significant.
If a variable has no prior binding, then it specifies a match. The
match is determined from some current position in the data: the
character which immediately follows all that has been matched previously.
If a variable occurs at the start of a line, it matches some text
at the start of the line. If it occurs at the end of a line, it matches
everything from the current position to the end of the line.
.SS* Negative Match
If a variable is one of the plain forms
.mono
.mets >> @ sident
.mets <> @{ bident }
.mets >> @* sident
.mets <> @*{ bident }
.onom
then this is a "negative match". The extent of the matched text (the text
bound to the variable) is determined by looking at what follows the variable,
and ranges from the current position to some position where the following
material finds a match. This is why this is called a "negative match": the
spanned text which ends up bound to the variable is that in which the match for
the trailing material did not occur.
A variable may be followed by a piece of text, a regular-expression directive,
a function call, a directive, another variable, or nothing (i.e. occurs at the
end of a line). These cases are described in detail below.
.NP* Variable Followed by Nothing
If the variable is followed by nothing, the negative match extends from the
current position in the data, to the end of the line. Example:
.IP code:
.mono
\ a b c @FOO
.onom
.IP data:
.mono
\ a b c defghijk
.onom
.IP result:
.mono
\ FOO="defghijk"
.onom
.NP* Variable Followed by Text
For the purposes of determining the negative match, text is defined as a
sequence of literal text and regular expressions, not divided by a directive.
So for instance in this example:
.verb
@a:@/foo/bcd e@(maybe)f@(end)
.brev
.PP
the variable
.code a
is considered to be followed by
.strn ":@/foo/bcd e" .
If a variable is followed by text, then the extent of the negative match is
determined by searching for the first occurrence of that text within the line,
starting at the current position.
The variable matches everything between the current position and the matching
position (not including the matching position). Any whitespace which follows
the variable (and is not enclosed inside braces that surround the variable
name) is part of the text. For example:
.IP code:
.mono
\ a b @FOO e f
.onom
.IP data:
.mono
\ a b c d e f
.onom
.IP result:
.mono
\ FOO="c d"
.onom
.PP
In the above example, the pattern text
.str "a b "
matches the
data
.strn "a b " .
So when the
.code @FOO
variable is processed, the data being
matched is the remaining
.strn "c d e f" .
The text which follows
.code @FOO
is
.strn " e f" .
This is found within the data
.str "c d e f"
at position 3 (counting from 0). So positions 0\(en2
.mono
("c d")
.onom
constitute the matching text which is bound to FOO.
.NP* Variable Followed by a Function Call or Directive
If the variable is followed by a function call, or a directive, the extent is
determined by scanning the text for the first position where a match occurs
for the entire remainder of the line. (For a description of functions,
see Functions.)
For example:
.verb
@foo@(bind a "abc")xyz
.brev
Here,
.code @foo
will match the text from the current position to where
.str "xyz"
occurs, even though there is a
.code @(bind)
directive. Furthermore, if
more material is added after the
.strn "xyz" ,
it is part of the search.
Note the difference between the following two:
.verb
@foo@/abc/@(func)
@foo@(func)@/abc/
.brev
In the first example,
.code @foo
matches the text from the current
position until the match for the regular expression
.strn "abc" .
.code @(func)
is not
considered when processing
.codn @foo .
In the second example,
.code @foo
matches the text from the current position until the position which matches
the function call, followed by a match for the regular expression.
The entire sequence
.code @(func)@/abc/
is considered.
.NP* Consecutive Variables
If an unbound variable specifies a fixed-width match or a regular expression,
then the issue of consecutive variables does not arise. Such a variable
consumes text regardless of any context which follows it.
However, what if an unbound variable with no modifier is followed by another
variable? The behavior depends on the nature of the other variable.
If the other variable is also unbound, and also has no modifier, this is a
semantic error which will cause the query to fail. A diagnostic message will
be issued, unless operating in quiet mode via
.codn -q .
The reason is that there is no way to bind two
consecutive variables to an extent of text; this is an ambiguous situation,
since there is no matching criterion for dividing the text between two
variables. (In theory, a repetition of the same variable, like
.codn @FOO@FOO ,
could find a solution by dividing the match extent in half, which would work
only in the case when it contains an even number of characters. This behavior
seems to have dubious value.)
An unbound variable may be followed by one which is bound. The bound
variable is effectively replaced by the text which it denotes, and the logic
proceeds accordingly.
It is possible for a variable to be bound to a regular expression.
If
.code x
is an unbound variable and
.code y
is bound to a regular expression
.codn RE ,
then
.code @x@y
means
.codn @x@/RE/ .
A variable
.code v
can be bound to a regular expression using, for example,
.codn "@(bind v #/RE/)" .
The
.code @*
syntax for longest match is available. Example:
.IP code:
.mono
\ @FOO:@BAR@FOO
.onom
.IP data:
.mono
\ xyz:defxyz
.onom
.IP result:
.mono
\ FOO=xyz, BAR=def
.onom
.PP
Here,
.code FOO
is matched with
.strn "xyz" ,
based on the delimiting around the
colon. The colon in the pattern then matches the colon in the data,
so that
.code BAR
is considered for matching against
.strn "defxyz" .
.code BAR
is followed by
.codn FOO ,
which is already bound to
.strn "xyz" .
Thus
.str "xyz"
is located in the
.str "defxyz"
data following
.strn "def" ,
and so BAR is bound to
.strn "def" .
If an unbound variable is followed by a variable which is bound to a list, or
nested list, then each character string in the list is tried in turn to produce
a match. The first match is taken.
An unbound variable may be followed by another unbound variable which specifies
a regular expression or function call match. This is a special case called a
"double variable match". What happens is that the text is searched using the
regular expression or function. If the search fails, then neither variable is
bound: it is a matching failure. If the search succeeds, then the first
variable is bound to the text which is skipped by the search. The second
variable is bound to the text matched by the regular expression or function.
Example:
.IP code:
.mono
\ @foo@{bar /abc/}
.onom
.IP data:
.mono
\ xyz@#abc
.onom
.IP result:
.mono
\ foo="xyz@#", BAR="abc"
.onom
.PP
.NP* Consecutive Variables via Directive
Two variables can be de facto consecutive in a manner shown in the
following example:
.verb
@var1@(all)@var2@(end)
.brev
This is treated just like the variable followed by directive. No semantic
error is identified, even if both variables are unbound. Here,
.code @var2
matches everything at the current position, and so
.code @var1
ends up bound to the empty string.
Example 1:
.code b
matches at position 0 and
.code a
binds the empty string:
.IP code:
.mono
\ @a@(all)@b@(end)
.onom
.IP data:
.mono
\ abc
.onom
.IP result:
.mono
\ a=""
b="abc"
.onom
.PP
Example 2:
.code *a
specifies longest match (see Longest Match below), and so it takes
everything:
.IP code:
.mono
\ @*a@(all)@b@(end)
.onom
.IP data:
.mono
\ abc
.onom
.IP result:
.mono
\ a="abc"
b=""
.onom
.PP
.NP* Longest Match
The closest-match behavior for the negative match can be overridden to longest
match behavior. A special syntax is provided for this: an asterisk between the
.code @
and the variable, e.g.:
.IP code:
.mono
\ a @*{FOO}cd
.onom
.IP data:
.mono
\ a b cdcdcdcd
.onom
.IP result:
.mono
\ FOO="b cdcdcd"
.onom
.PP
.IP code:
.mono
\ a @{FOO}cd
.onom
.IP data:
.mono
\ a b cdcdcd
.onom
.IP result:
.mono
\ FOO="b "
.onom
.PP
In the former example, the match extends to the rightmost occurrence of
.strn "cd" ,
and so
.code FOO
receives
.strn "b cdcdcd" .
In the latter example, the
.code *
syntax isn't used, and so a leftmost match takes place. The extent
covers only the
.strn "b " ,
stopping at the first
.str "cd"
occurrence.
.SS* Positive Match
There are syntactic variants of variable syntax which have an embedded expression
enclosed with the variable in braces:
.mono
.mets >> @{ bident <> / regex /}
.mets >> @{ bident >> ( fun >> [ args ...])}
.mets >> @{ bident << number }
.mets >> @{ bident << bident }
.onom
These specify a variable binding that is driven by a positive match derived
from a regular expression, function or character count, rather than from
trailing material (which is regarded as a "negative" match, since the
variable is bound to material which is
.B skipped
in order to match the trailing material).
The positive match syntax is processed without considering any following
syntax, and therefore may be followed by an unbound variable.
In the
.mono
.meti >> @{ bident <> / regex /}
.onom
form, the match
extends over all characters from the current position which match
the regular expression
.metn regex .
(See the Regular Expressions section below.)
If the variable already has a value, the text extracted by the regular
expression must exactly match the variable.
In the
.mono
.meti >> @{ bident >> ( fun >> [ args ...])}
.onom
form, the match extends over lines or characters which
are matched by the call to the function, if the call
succeeds. Thus
.code "@{x (y z w)}"
is just like
.codn "@(y z w)" ,
except that the region of
text skipped over by
.code "@(y z w)"
is also bound to the variable
.codn x .
Except in one special case, the matching takes place horizontally within the
current line, and the spanned range of text is treated as a string.
The exception is that if the
.mono
.meti >> @{ bident >> ( fun >> [ args ...])}
.onom
appears as the only element of a line, and
.meta fun
has a binding as a vertical function, then the function is invoked in
the same manner as it would be by the
.mono
.meti >> @( fun >> [ args ...])
.onom
syntax. Then the variable indicated by
.meta bident
is bound to the list of lines matched by the function call.
Pattern functions are described in the Functions section below.
The function is invoked even if the variable already has a value.
The text matched by the function must match the variable.
In the
.mono
.meti >> @{ bident << number }
.onom
form, the match processes a field of text which
consists of the specified number of characters, which must be a nonnegative
number. If the data line doesn't have that many characters starting at the
current position, the match fails. A match for zero characters produces an
empty string. The text which is actually bound to the variable
is all text within the specified field, but excluding leading and
trailing whitespace. If the field contains only spaces, then an empty
string is extracted. This fixed-field extraction takes place whether or not the
variable already has a binding. If it already has a binding, then it must match
the extracted, trimmed text.
The
.mono
.meti >> @{ bident << bident }
.onom
syntax allows the
.meta number
or
.meta regex
modifier to come from a variable. The variable must be bound and contain
a nonnegative integer or regular expression.
For example,
.code "@{x y}"
behaves like
.code "@{x 3}"
if
.code y
is bound to the integer 3. It is an error if
.code y
is unbound.
.coSS Special Symbols @ nil and @ t
Just like in the Common Lisp language, the names
.code nil
and
.code t
are special.
.code nil
symbol stands for the empty
list object, an object which marks the end of a list, and Boolean false. It is
synonymous with the syntax
.code ()
which may be used interchangeably with
.code nil
in most constructs.
In \*(TL,
.code nil
and
.code t
cannot be used as variables. When evaluated, they evaluate to themselves.
In the \*(TX pattern language,
.code nil
can be used in the variable binding syntax, but does not create a binding;
it has a special meaning. It allows the variable-matching syntax to be used to
skip material, in ways similar to the
.code skip
directive.
The
.code nil
symbol is also used as a
.code block
name, both in the \*(TX pattern language and in \*(TL.
A block named
.code nil
is considered to be anonymous.
.SS* Keyword Symbols
Names beginning with the
.code :
(colon) character are keyword symbols. These also
stand for themselves and may not be used as variables. Keywords are
useful for labeling information and situations.
.SS* Regular Expressions
Regular expressions are a language for specifying sets of character strings.
Through the use of pattern-matching elements, a regular expression is
able to denote an infinite set of texts.
\*(TX contains an original implementation of regular expressions, which
supports the following syntax:
.coIP .
The period is a "wildcard" that matches any character.
.coIP []
Character class: matches a single character, from the set specified by
special syntax written between the square brackets.
This supports basic regexp character class syntax. POSIX
notation like
.code [:digit:]
is not supported.
The regex tokens
.codn \es ,
.code \ed
and
.code \ew
are permitted in character classes, but not their complementing counterparts.
These tokens simply contribute their characters to the class.
The class
.code [a-zA-Z]
means match an uppercase
or lowercase letter; the class
.code [0-9a-f]
means match a digit or
a lowercase letter; the class
.code [^0-9]
means match a non-digit, and so forth.
There are no locale-specific behaviors in \*(TX regular expressions;
.code [A-Z]
denotes an ASCII/Unicode range of characters.
The class
.code [\ed.]
means match a digit or the period character.
A
.code ]
or
.code -
can be used within a character class, but must be escaped
with a backslash. A
.code ^
in the first position denotes a complemented
class, unless it is escaped by backslash. In any other position, it denotes
itself. Two backslashes code for one backslash. So for instance
.code [\e[\e-]
means match a
.code [
or
.code -
character,
.code [^^]
means match any character other
than
.codn ^ ,
and
.code [\e^\e\e]
means match either a
.code ^
or a backslash. Regex operators
such as
.codn * ,
.code +
and
.code &
appearing in a character class represent ordinary
characters. The characters
.codn - ,
.code ]
and
.code ^
occurring outside of a character class
are ordinary. Unescaped
.code /
characters can appear within a character class. The
empty character class
.code []
matches no character at all, and its complement
.code [^]
matches any character, and is treated as a synonym for the
.code .
(period) wildcard operator.
.ccIP @, \es @ \ew and @ \ed
These regex tokens each match a single character.
The
.code \es
regex token matches a wide variety of ASCII whitespace characters
and Unicode spaces. The
.code \ew
token matches alphabetic word characters; it
is equivalent to the character class
.codn [A-Za-z_] .
The
.code \ed
token matches a digit, and is equivalent to
.codn [0-9] .
.ccIP @, \eS @ \eW and @ \eD
These regex tokens are the complemented counterparts of
.codn \es ,
.code \ew
and
.codn \ed .
The
.code \eS
token matches all those characters which
.code \es
does not match,
.code \eW
matches all characters that
.code \ew
does not match and
.code \eD
matches nondigits.
.coIP empty
An empty expression is a regular expression. It represents the set of strings
consisting of the empty string; i.e. it matches just the empty string. The
empty regex can appear alone as a full regular expression (for instance the
\*(TX syntax
.code @//
with nothing between the slashes)
and can also be passed as a subexpression to operators, though this
may require the use of parentheses to make the empty regex explicit. For
example, the expression
.code a|
means: match either
.codn a ,
or nothing. The forms
.code *
and
.code (*)
are syntax errors; though not useful, the correct way to match the
empty expression zero or more times is the syntax
.codn ()* .
.coIP nomatch
The nomatch regular expression represents the
empty set: it matches no strings at all, not even the empty string.
There is no dedicated syntax to directly express nomatch in the regex language.
However, the empty character class
.code []
is equivalent to nomatch, and may be
considered to be a notation for it. Other representations of nomatch are
possible: for instance, the regex
.code ~.*
which is the complement of the regex that
denotes the set of all possible strings, and thus denotes the empty set. A
nomatch has uses; for instance, it can be used to temporarily "comment out"
regular expressions. The regex
.code ([]abc|xyz)
is equivalent to
.codn (xyz) ,
since the
.code []abc
branch cannot match anything. Using
.code []
to "block" a subexpression allows
you to leave it in place, then enable it later by removing the "block".
.coIP (R)
If
.code R
is a regular expression, then so is
.codn (R) .
The contents of parentheses denote one regular expression unit, so that for
instance in
.codn (RE)* ,
the
.code *
operator applies to the entire parenthesized group.
The syntax
.code ()
is valid and equivalent to the empty regular expression.
.coIP R?
Optionally match the preceding regular expression
.codn R .
.coIP R*
Match the expression
.code R
zero or more times. This
operator is sometimes called the "Kleene star", or "Kleene closure".
The Kleene closure favors the longest match. Roughly speaking, if there are two
or more ways in which
.code R1*R2
can match, then that match occurs in which
.code R1*
matches the longest possible text.
.coIP R+
Match the preceding expression
.code R
one or more times. Like
.codn R* ,
this favors the longest possible match:
.code R+
is equivalent to
.codn RR* .
.coIP R1%R2
Match
.code R1
zero or more times, then match
.codn R2 .
If this match can occur in
more than one way, then it occurs such that
.code R1
is matched the fewest
number of times, which is opposite from the behavior of
.codn R1*R2 .
Repetitions of
.code R1
terminate at the earliest
point in the text where a nonempty match for
.code R2
occurs. Because
it favors shorter matches,
.code %
is termed a non-greedy operator. If
.code R2
is the empty expression, or equivalent to it, then
.code R1%R2
reduces to
. codn R1* .
So for
instance
.code (R%)
is equivalent to
.codn (R*) ,
since the missing right operand is
interpreted as the empty regex. Note that whereas the expression
.code (R1*R2)
is equivalent to
.codn (R1*)R2 ,
the expression
.code (R1%R2)
is
.B not
equivalent to
.codn (R1%)R2 .
Also note that
.code A(XY%Z)B
is equivalent to
.codn AX(Y%Z)B .
This is because the precedence of
.code %
is higher than that of catenation on its left side; this rule prevents the given
syntax from expressing the
.code XY
catenation. The expression may be understood as:
.code A(X(Y%Z))B
where the inner parentheses clarify how the syntax surrounding the
.code %
operator is being parsed, and the outer parentheses are superfluous.
The correct way to assert catenation of
.code XY
as the left operand of
.code %
is
.codn A(XY)%ZB .
To specify
.code XY
as the left operand, and limit the right operand to just
.codn Z ,
the correct syntax is
.codn A((XY)%Z)B .
By contrast, the expression
.code A(X%YZ)B
is not equivalent to
.code A(X%Y)ZB
because the precedence of
.code %
is lower than that of catenation on its right side. The operator is
effectively "bi-precedential".
.coIP ~R
Match the opposite of the following expression
.codn R ;
that is, match exactly
those texts that
.code R
does not match. This operator is called complement,
or logical not.
.coIP R1R2
Two consecutive regular expressions denote catenation:
the left expression must match, and then the right.
.coIP R1|R2
Match either the expression
.code R1
or
.codn R2 .
This operator is known by
a number of names: union, logical or, disjunction, branch, or alternative.
.coIP R1&R2
Match both the expression
.code R1
and
.code R2
simultaneously; i.e. the
matching text must be one of the texts which are in the intersection of the set
of texts matched by
.code R1
and the set matched by
.codn R2 .
This operator is called intersection, logical and, or conjunction.
.PP
Any character which is not a regular-expression operator, a backslash escape,
or the slash delimiter, denotes a one-position match of that character itself.
Any of the special characters, including the delimiting
.codn / ,
and the backslash, can be escaped with a backslash to suppress its
meaning and denote the character itself.
Furthermore, all of the same escapes that are described in the section Special
Characters in Text above are supported \(em the difference is that in regular
expressions, the
.code @
character is not required, so for example a tab is coded as
.code \et
rather than
.codn @\et .
Octal and hex character escapes can be optionally
terminated by a semicolon, which is useful if the following characters are
octal or hex digits not intended to be part of the escape.
Only the above escapes are supported. Unlike in some other regular-expression
implementations, if a backlash appears before a character which isn't a regex
special character or one of the supported escape sequences, it is an error.
This wasn't true of historic versions of \*(TX. See the COMPATIBILITY section.
.IP "Precedence table, highest to lowest:"
.TS
tab(!);
l l l.
Operators!Class!Associativity
\f[4](R) []\f[]!primary!
\f[4]R? R+ R* R%...\f[]!postfix!left-to-right
\f[4]R1R2\f[]!catenation!left-to-right
\f[4]~R ...%R\f[]\f[]\f[]!unary!right-to-left
\f[4]R1&R2\f[]!intersection!left-to-right
\f[4]R1|R2\f[]!union!left-to-right
.TE
.PP
The
.code %
operator is like a postfix operator with respect to its left
operand, but like a unary operator with respect to its right operand.
Thus
.code a~b%c~d
is
.codn a(~(b%(c(~d)))) ,
demonstrating right-to-left associativity,
where all of
.code b%
may be regarded as a unary operator being applied to
.codn c~d .
Similarly,
.code a?*+%b
means
.codn (((a?)*)+)%b ,
where the trailing
.code %b
behaves like a postfix operator.
In
\*(TX, regular expression matches do not span multiple lines. The regex
language has no feature for multiline matching. However, the
.code @(freeform)
directive
allows the remaining portion of the input to be treated as one string
in which line terminators appear as explicit characters. Regular expressions
may freely match through this sequence.
It's possible for a regular expression to match an empty string.
For instance, if the next input character is
.codn z ,
facing the regular expression
.codn /a?/ ,
there is a zero-character match:
the regular expression's state machine can reach an acceptance
state without consuming any characters. Examples:
.IP code:
.mono
\ @A@/a?/@/.*/
.onom
.IP data:
.mono
\ zzzzz
.onom
.IP result:
.mono
\ A=""
.onom
.PP
.IP code:
.mono
\ @{A /a?/}@B
.onom
.IP data:
.mono
\ zzzzz
.onom
.IP result:
.mono
\ A="", B="zzzz"
.onom
.PP
.IP code:
.mono
\ @*A@/a?/
.onom
.IP data:
.mono
\ zzzzz
.onom
.IP result:
.mono
\ A="zzzzz"
.onom
.PP
In the first example, variable
.code @A
is followed by a regular expression
which can match an empty string. The expression faces the letter
.code "z"
at position 0 in the data line. A zero-character match occurs there,
therefore the variable
.code A
takes on the empty string. The
.code @/.*/
regular expression then consumes the line.
Similarly, in the second example, the
.code /a?/
regular expression faces a
.codn "z" ,
and thus yields an empty string which is bound to
.codn A .
Variable
.code @B
consumes the entire line.
The third example requests the longest match for the variable binding.
Thus, a search takes place for the rightmost position where the
regular expression matches. The regular expression matches anywhere,
including the empty string after the last character, which is
the rightmost place. Thus variable
.code A
fetches the entire line.
For additional information about the advanced regular-expression
operators, see NOTES ON EXOTIC REGULAR EXPRESSIONS below.
.SS* Compound Expressions
If the
.code @
escape character is followed by an open parenthesis or square bracket,
this is taken to be the start of a \*(TL compound expression.
The \*(TX language has the unusual property that its syntactic elements,
so-called
.IR directives ,
are Lisp compound expressions. These expressions not only enclose syntax, but
expressions which begin with certain symbols de facto
behave as tokens in a phrase structure grammar. For instance, the expression
.code @(collect)
begins a block which must be terminated by the expression
.codn @(end) ,
otherwise there is a syntax error. The
.code collect
expression can contain arguments which modify the behavior of the construct,
for instance
.codn "@(collect :gap 0 :vars (a b))" .
In some ways, this situation might be compared to HTML, in which
an element such as
.code <a>
must be terminated by
.code </a>
and can have attributes such as
.codn "<a href=\(dq...\(dq>" .
Compound expressions contain subexpressions which are
other compound expressions or literal objects of various kinds.
Among these are: symbols, numbers, string literals, character
literals, quasiliterals and regular expressions. These are described in the
following sections. Additional kinds of literal objects exist, which are
discussed in the TXR LISP section of the manual.
Some examples of compound expressions are:
.verb
(banana)
(a b c (d e f))
( a (b (c d) (e ) ))
("apple" #\eb #\espace 3)
(a #/[a-z]*/ b)
(_ `@file.txt`)
.brev
Symbols occurring in a compound expression follow a slightly more permissive
lexical syntax than the
.meta bident
in the syntax
.mono
.meti <> @{ bident }
.onom
introduced earlier. The
.code /
(slash) character may be part of an identifier, or even
constitute an entire identifier. In fact a symbol inside a
directive is a
.metn lident .
This is described in the Symbol Tokens section under TXR LISP.
A symbol must not be a number; tokens that look like numbers are treated as
numbers and not symbols.
.SS* Character Literals
Character literals are introduced by the
.code #\e
(hash-backslash) syntax, which is either
followed by a character name, the letter
.code x
followed by hex digits,
the letter
.code o
followed by octal digits, or a single character. Valid character
names are:
.verb
nul linefeed return
alarm newline esc
backspace vtab space
tab page pnul
.brev
For instance
.code #\eesc
denotes the escape character.
This convention for character literals is similar to that
of the Scheme language. Note that
.code #\elinefeed
and
.code #\enewline
are the same
character. The
.code #\epnul
character is specific to \*(TX and denotes the
.code U+DC00
code in Unicode; the name stands for "pseudo-null", which is related to
its special function. For more information about this, see the section
"Character Handling and International Characters".
.SS* String Literals
String literals are delimited by double quotes.
A double quote within a string literal is encoded using
.mono
\e"
.onom
and a backslash is encoded as
.codn \e\e .
Backslash escapes like
.code \en
and
.code \et
are recognized, as are hexadecimal escapes like
.code \exFF
or
.code \exabc
and octal escapes like
.codn \e123 .
Ambiguity between an escape and subsequent
text can be resolved by adding a semicolon delimiter after the escape:
.str "\exabc;d"
is a string consisting of the character
.code "U+0ABC"
followed by
.strn "d" .
The semicolon
delimiter disappears. To write a literal semicolon immediately after a hex
or octal escape, write two semicolons, the first of which will be interpreted
as a delimiter. Thus,
.str "\ex21;;"
represents
.strn "!;" .
Note that the source code syntax of \*(TX string literals is specified
in UTF-8, which is decoded into an internal string representation consisting
of code points. The numeric escape sequences are an abstract syntax for
specifying code points, not for specifying bytes to be inserted into the
UTF-8 representation, even if they lie in the 8-bit range. Bytes cannot be
directly specified, other than literally. However, when a \*(TX string object
is encoded to UTF-8, every code point lying in the range U+DC00 through U+DCFF
is converted to a single byte by taking the low-order eight bits of its
value. By manipulating code points in this special range, \*(TX programs can
reproduce arbitrary byte sequences in text streams. Also note that the
.code \eu
escape sequence for specifying code points found in some languages is
unnecessary and absent, since the existing hexadecimal and octal escapes
satisfy this requirement. More detailed information is given in the earlier
section Character Handling and International Characters.
If the line ends in the middle of a literal, it is an error, unless the
last character is a backslash. This backslash is a special escape which does
not denote a character; rather, it indicates that the string literal continues
on the next line. The backslash is deleted, along with whitespace which
immediately precedes it, as well as leading whitespace in the following line.
The escape sequence
.str "\e "
(backslash space) can be used to encode a significant space.
Example:
.verb
"foo \e
bar"
"foo \e
\e bar"
"foo\e \e
bar"
.brev
The first string literal is the string
.strn "foobar" .
The second two are
.strn "foo bar" .
.SS* Word List Literals
A word list literal (WLL) provides a convenient way to write a list of strings
when such a list can be given as whitespace-delimited words.
There are two flavors of the WLL: the regular WLL which begins with
.mono
#"
.onom
(hash, double quote) and the splicing list literal which begins with
.mono
#*"
.onom
(hash, star, double quote).
Both types are terminated by a double quote, which may be escaped
as
.mono
\e"
.onom
in order to include it as a character. All the escaping conventions
used in string literals can be used in word literals.
Unlike in string literals, whitespace (tabs and spaces) is not
significant in word literals: it separates words.
A whitespace character may be escaped with a backslash
in order to include it as a literal character.
Just like in string literals, an unescaped newline character is not allowed.
A newline preceded by a backslash is permitted. Such an escaped backslash,
together with any leading and trailing unescaped whitespace, is removed
and replaced with a single space.
Example:
.verb
#"abc def ghi" --> notates ("abc" "def" "ghi")
#"abc def \e
ghi" --> notates ("abc" "def" "ghi")
#"abc\e def ghi" --> notates ("abc def" "ghi")
#"abc\e def\e \e
\e ghi" --> notates ("abc def " " ghi")
.brev
A splicing word literal differs from a word literal in that it does not
produce a list of string literals, but rather it produces a sequence of string
literals that is merged into the surrounding syntax. Thus, the following two
notations are equivalent:
.verb
(1 2 3 #*"abc def" 4 5 #"abc def")
(1 2 3 "abc" "def" 4 5 ("abc" "def"))
.brev
The regular WLL produced a single list object, but the splicing
WLL expanded into multiple string literal objects.
.SS* String Quasiliterals
Quasiliterals are similar to string literals, except that they may
contain variable references denoted by the usual
.code @
syntax. The quasiliteral
represents a string formed by substituting the values of those variables
into the literal template. If
.code a
is bound to
.str "apple"
and
.code b
to
.strn "banana" ,
the quasiliteral
.code "`one @a and two @{b}s`"
represents the string
.strn "one apple and two bananas" .
A backquote escaped by a backslash represents
itself. Unlike in directive syntax, two consecutive
.code @
characters do not code for a literal
.codn @ ,
but cause a syntax error. The reason for this is that compounding of the
.code @
syntax is meaningful.
Instead, there is a
.code \e@
escape for encoding a literal
.code @
character. Quasiliterals support the full output variable
syntax. Expressions within variable substitutions follow the evaluation rules
of \*(TL. This hasn't always been the case: see the COMPATIBILITY section.
Quasiliterals can be split into multiple lines in the same way as ordinary
string literals.
.SS* Quasiword List Literals
The quasiword list literals (QLLs) are to quasiliterals what WLLs are to
ordinary literals. (See the above section Word List Literals.)
A QLL combines the convenience of the WLL
with the power of quasistrings.
Just as in the case of WLLs, there are two flavors of the
QLL: the regular QLL which begins with
.code #`
(hash, backquote) and the splicing QLL which begins with
.code #*`
(hash, star, backquote).
Both types are terminated by a backquote, which may be escaped
as
.code \e`
in order to include it as a character. All the escaping conventions
used in quasiliterals can be used in QLLs.
Unlike in quasiliterals, whitespace (tabs and spaces) is not
significant in QLLs: it separates words.
A whitespace character may be escaped with a backslash
in order to include it as a literal character.
A newline is not permitted unless escaped. An escaped newline works exactly the
same way as it does in WLLs.
Note that the delimiting into words is done before the variable
substitution. If the variable
.code a
contains spaces, then
.code #`@a`
nevertheless
expands into a list of one item: the string derived from
.codn a .
Examples:
.verb
#`abc @a ghi` --> notates (`abc` `@a` `ghi`)
#`abc @d@e@f \e
ghi` --> notates (`abc` `@d@e@f` `ghi`)
#`@a\e @b @c` --> notates (`@a @b` `@c`)
.brev
A splicing QLL differs from an ordinary QLL in that it does not produce a list
of quasiliterals, but rather it produces a sequence of quasiliterals that is
merged into the surrounding syntax.
.SS* Numbers
\*(TX supports integers and floating-point numbers.
An integer constant is made up of digits
.code 0
through
.codn 9 ,
optionally preceded by a
.code +
or
.code -
sign.
Examples:
.verb
123
-34
+0
-0
+234483527304983792384729384723234
.brev
An integer constant can also be specified in hexadecimal using the prefix
.code #x
followed by an optional sign, followed by hexadecimal digits:
.code 0
through
.code 9
and the uppercase or lowercase letters
.code A
through
.codn F :
.verb
#xFF ;; 255
#x-ABC ;; -2748
.brev
Similarly, octal numbers are supported with the prefix
.code #o
followed by octal digits:
.verb
#o777 ;; 511
.brev
and binary numbers can be written with a
.code #b
prefix:
.verb
#b1110 ;; 14
.brev
Note that the
.code #b
prefix is also used for buffer literals.
A floating-point constant is marked by the inclusion of a decimal point, the
scientific E notation, or both. It is an optional sign, followed
by a mantissa consisting of digits, a decimal point, more digits, and then an
optional E notation consisting of the letter
.code "e"
or
.codn "E" ,
an optional
.code +
or
.code -
sign, and then digits indicating the exponent value.
In the mantissa, the digits are not optional. At least one digit must either
precede the decimal point or follow it.
That is to say, a decimal point by itself is not a floating-point constant.
Examples:
.verb
.123
123.
1E-3
20E40
.9E1
9.E19
-.5
+3E+3
1.E5
.brev
Examples which are not floating-point constant tokens:
.verb
. ;; dot token, not a number
123E ;; the symbol 123E
1.0E- ;; syntax error: invalid floating point constant
1.0E ;; syntax error: invalid floating point constant
1.E ;; syntax error: invalid floating point literal
.e ;; syntax error: dot token followed by symbol
.brev
In \*(TX there is a special "dotdot" token consisting of two consecutive periods.
An integer constant followed immediately by dotdot is recognized as such; it is
not treated as a floating constant followed by a dot. That is to say,
.code 123..
does not mean
.code "123. ."
(floating point
.code 123.0
value followed by dot token). It means
.code "123 .."
(integer
.code 123
followed by
.code ..
token).
Dialect Note: unlike in Common Lisp,
.code 123.
is not an integer, but the floating-point number
.codn 123.0 .
.SS* Comments
Comments of the form
.code @;
were introduced earlier. Inside compound expressions, another convention for
comments exists: Lisp comments, which are introduced by the
.code ;
(semicolon) character and span to the end of the line.
Example:
.verb
@(foo ; this is a comment
bar ; this is another comment
)
.brev
This is equivalent to
.codn "@(foo bar)" .
.SH* DIRECTIVES
.SS* Overview
When a \*(TL compound expression occurs in \*(TX preceded by a
.codn @ ,
it is a
.IR directive .
Directives which are based on certain symbols are, additionally,
involved in a phrase-structure syntax which uses Lisp expressions
as if they were tokens.
For instance, the directive
.verb
@(collect)
.brev
not only denotes a compound expression with the
.code collect
symbol in its head position, but it also introduces a syntactic phrase which
requires a matching
.code @(end)
directive. In other words,
.code @(collect)
is not only
an expression, but serves as a kind of token in a higher-level, phrase-structure
grammar.
Effectively,
.code collect
is a reserved symbol in the \*(TX language. A \*(TX program cannot use
this symbol as the name of a pattern function due to its role in the syntax.
The symbol has no reserved role in \*(TL.
Usually if this type of directive occurs alone in a line, not
preceded or followed by other material, it is involved in a "vertical"
(or line-oriented) syntax.
If such a directive is embedded in a line (has preceding or trailing material)
then it is in a horizontal syntactic and semantic context (character-oriented).
There is an exception: the definition of a horizontal function looks like this:
.verb
@(define name (arg))body material@(end)
.brev
Yet, this is considered one vertical item, which means that it does not match
a line of data. (This is necessary because all horizontal syntax matches
something within a line of data, which is undesirable for definitions.)
Many directives exhibit both horizontal and vertical syntax, with different but
closely related semantics. Some are vertical only, some are
horizontal only.
A summary of the available directives follows:
.coIP @(eof)
Explicitly match the end of file. Fails if unmatched data remains in
the input stream. Can capture or match the termination status of a pipe.
.coIP @(eol)
Explicitly match the end of line. Fails if the current position is not the
end of a line. Also fails if no data remains (there is no current line).
.coIP @(next)
Continue matching in another file or data source.
.coIP @(block)
Groups together a sequence of directives into a logical name block,
which can be explicitly terminated from within by using
the
.code @(accept)
and
.code @(fail)
directives.
Blocks are described in the section Blocks below.
.coIP @(skip)
Treat the remaining query as a subquery unit, and search the lines (or
characters) of the input file until that subquery matches somewhere.
A
.code skip
is also an anonymous block.
.coIP @(trailer)
Treat the remaining query or subquery as a match for a trailing context. That
is to say, if the remainder matches, the data position is not advanced.
.coIP @(freeform)
Treat the remainder of the input as one big string, and apply the following
query line to that string. The newline characters (or custom separators) appear
explicitly in that string.
.coIP @(fuzz)
The
.code fuzz
directive, inspired by the patch utility, specifies a partial
match for some lines.
.ccIP @ @(line) and @ @(chr)
These directives match a variable or expression against the current line
number or character position.
.coIP @(name)
Match a variable against the name of the current data source.
.coIP @(data)
Match a variable against the remaining data (a lazy list of strings).
.coIP @(some)
Multiple clauses are each applied to the same input. Succeeds if at least one
of the clauses matches the input. The bindings established by earlier
successful clauses are visible to the later clauses.
.coIP @(all)
Multiple clauses are applied to the same input. Succeeds if and only if each
one of the clauses matches. The clauses are applied in sequence, and evaluation
stops on the first failure. The bindings established by earlier successful
clauses are visible to the later clauses.
.coIP @(none)
Multiple clauses are applied to the same input. Succeeds if and only if none of
them match. The clauses are applied in sequence, and evaluation stops on the
first success. No bindings are ever produced by this construct.
.coIP @(maybe)
Multiple clauses are applied to the same input. No failure occurs if none of
them match. The bindings established by earlier successful clauses are visible
to the later clauses.
.coIP @(cases)
Multiple clauses are applied to the same input. Evaluation stops on the
first successful clause.
.coIP @(require)
The
.code require
directive is similar to the
.code do
directive in that it evaluates one or more
\*(TL expressions. If the result of the rightmost expression is
.codn nil ,
then
.code require
triggers a match failure. See the TXR LISP section far below.
.ccIP @, @(if) @, @(elif) and @ @(else)
The
.code if
directive with optional
.code elif
and
.code else
clauses allows one of multiple bodies of pattern-matching directives to be
conditionally selected by testing the values of Lisp expressions. It is
also available inside
.code @(output)
for conditionally selecting output clauses.
.coIP @(choose)
Multiple clauses are applied to the same input. The one whose effect persists
is the one which maximizes or minimizes the length of a particular variable.
.coIP @(empty)
The
.code @(empty)
directive matches the empty string. It is useful in certain
situations, such as expressing an empty match in a directive that doesn't
accept an empty clause. The
.code @(empty)
syntax has another meaning in
.code @(output)
clauses, in conjunction with
.codn @(repeat) .
.meIP @(define < name >> ( args ...))
Introduces a function. Functions are described in the Functions section below.
.meIP @(call < expr << arg *)
Performs function indirection. Evaluates
.metn expr ,
which must produce a symbol that names a pattern function. Then that
pattern function is invoked.
.coIP @(gather)
Searches text for matches for multiple clauses which may occur in arbitrary
order. For convenience, lines of the first clause are treated as separate
clauses.
.coIP @(collect)
Search the data for multiple matches of a clause. Collect the
bindings in the clause into lists, which are output as array variables.
The
.code @(collect)
directive is line-oriented. It works with a multiline
pattern and scans line by line. A similar directive called
.code @(coll)
works within one line.
A collect is an anonymous block.
.coIP @(and)
Separator of clauses for
.codn @(some) ,
.codn @(all) ,
.codn @(none) ,
.code @(maybe)
and
.codn @(cases) .
Equivalent to
.codn @(or) .
The choice is stylistic.
.coIP @(or)
Separator of clauses for
.codn @(some) ,
.codn @(all) ,
.codn @(none) ,
.code @(maybe)
and
.codn @(cases) .
Equivalent to
.codn @(and) .
The choice is stylistic.
.coIP @(end)
Required terminator for
.codn @(some) ,
.codn @(all) ,
.codn @(none) ,
.codn @(maybe) ,
.codn @(cases) ,
.codn @(if) ,
.codn @(collect) ,
.codn @(coll) ,
.codn @(output) ,
.codn @(repeat) ,
.codn @(rep) ,
.codn @(try) ,
.code @(block)
and
.codn @(define) .
.coIP @(fail)
Terminate the processing of a block, as if it were a failed match.
Blocks are described in the section Blocks below.
.coIP @(accept)
Terminate the processing of a block, as if it were a successful match.
What bindings emerge may depend on the kind of block:
.code collect
has special semantics. Blocks are described in the section Blocks below.
.coIP @(try)
Indicates the start of a try block, which is related to exception
handling, described in the Exceptions section below.
.ccIP @ @(catch) and @ @(finally)
Special clauses within
.codn @(try) .
See Exceptions below.
.ccIP @ @(defex) and @ @(throw)
Define custom exception types; throw an exception. See Exceptions below.
.coIP @(assert)
The
.code assert
directive requires the following material to match, otherwise
it throws an exception. It is useful for catching mistakes or omissions
in parts of a query that are surefire matches.
.coIP @(flatten)
Normalizes a set of specified variables to one-dimensional lists. Those
variables which have a scalar value are reduced to lists of that value.
Those which are lists of lists (to an arbitrary level of nesting) are converted
to flat lists of their leaf values.
.coIP @(merge)
Binds a new variable which is the result of merging two or more
other variables. Merging has somewhat complicated semantics.
.coIP @(cat)
Decimates a list (any number of dimensions) to a string, by catenating its
constituent strings, with an optional separator string between all of the
values.
.coIP @(bind)
Binds one or more variables against a value using a structural
pattern match. A limited form of unification takes place which can cause a
match to fail.
.coIP @(set)
Destructively assigns one or more existing variables using a structural
pattern, using syntax similar to bind. Assignment to unbound
variables triggers an error.
.coIP @(rebind)
Evaluates an expression in the current binding environment, and
then creates new bindings for the variables in the structural pattern.
Useful for temporarily overriding variable values in a scope.
.coIP @(forget)
Removes variable bindings.
.coIP @(local)
Synonym of
.codn @(forget) .
.coIP @(output)
A directive which encloses an output clause in the query. An output section
does not match text, but produces text. The directives above are not
understood in an output clause.
.coIP @(repeat)
A directive understood within an
.code @(output)
section, for repeating multiline
text, with successive substitutions pulled from lists. The directive
.code @(rep)
produces iteration over lists horizontally within one line. These directives
have a different meaning in matching clauses, providing a shorthand
notation for
.code "@(collect :vars nil)"
and
.codn "@(coll :vars nil)" ,
respectively.
.coIP @(deffilter)
The
.code deffilter
directive is used for defining named filters, which are useful
for filtering variable substitutions in output blocks. Filters are useful
when data must be translated between different representations that
have different special characters or other syntax, requiring escaping
or similar treatment. Note that it is also possible to use a function
as a filter. See Function Filters below.
Named filters are stored in the hash table held in the Lisp special variable
.codn *filters* .
.coIP @(filter)
The
.code filter
directive passes one or more variables through a given
filter or chain or filters, updating them with the filtered values.
.ccIP @ @(load) and @ @(include)
The
.code load
and
.code include
directives allow \*(TX programs to be modularized. They bring in
code from a file, in two different ways.
.coIP @(do)
The
.code do
directive is used to evaluate \*(TL expressions, discarding their
result values. See the TXR LISP section far below.
.coIP @(mdo)
The
.code mdo
(macro
.codn do )
directive evaluates \*(TL expressions immediately, during the parsing
of the \*(TX syntax in which it occurs.
.coIP @(in-package)
The
.code in-package
directive is used to switch to a different symbol package.
It mirrors the \*(TL macro of the same name.
.PP
.SS* Subexpression Evaluation
Some directives contain subexpressions which are evaluated. Two distinct
styles of evaluations occur in \*(TX: bind expressions and Lisp expressions.
Which semantics applies to an expression depends on the syntactic
context in which it occurs: which position in which directive.
The evaluation of \*(TL expressions is described in the TXR LISP section of the manual.
Bind expressions are so named because they occur in the
.code @(bind)
directive. \*(TX pattern function invocations also treat argument expressions
as bind expressions.
The
.codn @(rebind) ,
.codn @(set) ,
.codn @(merge) ,
and
.code @(deffilter)
directives also use bind expression evaluation. Bind expression evaluation
also occurs in the argument position of the
.code :tlist
keyword in the
.code @(next)
directive.
Unlike Lisp expressions, bind expressions do not support operators. If a bind
expression is a nested list structure, it is a template denoting that
structure. Any symbol in any position of that structure is interpreted as a
variable. When the bind expression is evaluated, those corresponding positions
in the template are replaced by the values of the variables.
Anywhere where a variable can appear in a bind expression's nested list
structure, a Lisp expression can appear preceded by the
.code @
character. That Lisp expression is evaluated and its value is substituted
into the bind expression's template.
Moreover, a Lisp expression preceded by
.code @
can be used as an entire bind expression. The value of that Lisp
expression is then taken as the bind expression value.
Any object in a bind expression which is not a nested list structure containing
Lisp expressions or variables denotes itself literally.
.TP* Examples:
In the following examples, the variables
.code a
and
.code b
are assumed to have the string values
.str foo
and
.strn bar ,
respectively.
The
.code ->
notation indicates the value of each expression.
.verb
a -> "foo"
(a b) -> ("foo" "bar")
((a) ((b) b)) -> (("foo") (("bar") "bar"))
(list a b) -> error: unbound variable list
@(list a b) -> ("foo" "bar") ;; Lisp expression
(a @[b 1..:]) -> ("foo" "ar") ;; Lisp eval of [b 1..:]
(a @(+ 2 2)) -> ("foo" 4) ;; Lisp eval of (+ 2 2)
#(a b) -> (a b) ;; Vector literal, not list.
[a b] -> error: unbound variable dwim
.brev
The last example above
.code "[a b]"
is a notation equivalent to
.code "(dwim a b)"
and so follows similarly to the example involving
.codn list .
.SS* Input Scanning and Data Manipulation
.dir next
The
.code next
directive indicates that the remaining directives in the current block
are to be applied against a new input source.
It can only occur by itself as the only element in a query line,
and takes various arguments, according to these possibilities:
.mono
.mets @(next)
.mets @(next << source )
.mets @(next < source :nothrow)
.mets @(next :args)
.mets @(next :env)
.mets @(next :list << lisp-expr )
.mets @(next :tlist << bind-expr )
.mets @(next :string << lisp-expr )
.mets @(next :var << var )
.mets @(next nil)
.onom
The lone
.code @(next)
without arguments specifies that subsequent directives
will match inside the next file in the argument list which was passed
to \*(TX on the command line.
If
.meta source
is given, it must be a \*(TL expression which denotes an
input source. Its value may be a string or an input stream.
For instance, if variable
.code A
contains the text
.strn "data" ,
then
.code "@(next A)"
means switch to the file called
.strn "data" ,
and
.code "@(next `@A.txt`)"
means to switch to the file
.strn "data.txt" .
The directive
.code "@(next (open-command `git log`))"
switches to the input stream connected to the output of the
.code "git log"
command.
If the input source cannot be opened for whatever reason,
\*(TX throws an exception (see Exceptions below). An unhandled exception will
terminate the program. Often, such a drastic measure is inconvenient;
if
.code @(next)
is invoked with the
.code :nothrow
keyword, then if the input
source cannot be opened, the situation is treated as a simple
match failure. The
.code :nothrow
keyword also ensures that when the stream is later closed,
which occurs when the lazy list reads all of the available data,
the implicit call to the
.code close-stream
function specifies
.code nil
as the argument value to that function's
.meta throw-on-error-p
parameter. This
.code :nothrow
mechanism does not suppress all exceptions related to the processing
of that stream; unusual conditions encountered during the reading of
data from the stream may throw exceptions.
The variant
.code "@(next :args)"
means that the remaining command-line arguments are to
be treated as a data source. For this purpose, each argument is considered to
be a line of text. The argument list does include that argument which specifies
the file that is currently being processed or was most recently processed.
As the arguments are matched, they are consumed. This means that if a
.code @(next)
directive without
arguments is executed in the scope of
.codn "@(next :args)" ,
it opens the file named
by the first unconsumed argument.
To process arguments, and then continue with the original file and argument
list, wrap the argument processing in a
.codn @(block) .
When the block terminates, the input source and argument list are restored
to what they were before the block.
The variant
.code "@(next :env)"
means that the list of process environment variables
is treated as a source of data. It looks like a text file stream
consisting of lines of the form
.strn "name=value" .
If this feature is not available
on a given platform, an exception is thrown.
The syntax
.mono
.meti @(next :list << lisp-expr )
.onom
treats \*(TL expression
.meta lisp-expr
as a source of
text. The value of
.meta lisp-expr
is flattened to a simple list in a way similar to the
.code @(flatten)
directive. The resulting list is treated as if it were the
lines of a text file: each element of the list must be a string,
which represents a line. If the strings happen contain embedded newline
characters, they are a visible constituent of the line, and do not act as line
separators.
The syntax
.mono
.meti @(next :tlist << bind-expr )
.onom
is similar to
.code "@(next :list ...)"
except that
.meta bind-expr
is not a \*(TL expression, but a \*(TX bind expression.
The syntax
.mono
.meti @(next :var << var )
.onom
requires
.meta var
to be a previously bound variable. The value of the
variable is retrieved and treated like a list, in the
same manner as under
.codn "@(next :list ...)" .
Note that
.code "@(next :var x)"
is not always the same as
.codn "@(next :tlist x)" ,
because
.code ":var x"
strictly requires
.code x
to be a \*(TX variable, whereas the
.code x
in
.code ":tlist x"
is an expression which can potentially refer to Lisp variable.
The syntax
.mono
.meti @(next :string << lisp-expr )
.onom
treats expression
.meta lisp-expr
as a source of text. The value of the expression must be a string. Newlines in
the string are interpreted as line terminators.
A string which is not terminated by a newline is tolerated, so that:
.verb
@(next :string "abc")
@a
.brev
binds
.code a
to
.strn "abc" .
Likewise, this is also the case with input files and other
streams whose last line is not terminated by a newline.
However, watch out for empty strings, which are analogous to a correctly formed
empty file which contains no lines:
.verb
@(next :string "")
@a
.brev
This will not bind
.code a
to
.strn "" ;
it is a matching failure. The behavior of
.code :list
is
different. The query
.verb
@(next :list "")
@a
.brev
binds
.code a
to
.strn "" .
The reason is that under
.code :list
the string
.str ""
is flattened to
the list
.mono
("")
.onom
which is not an empty input stream, but a stream consisting of
one empty line.
The
.code "@(next nil)"
variant indicates that the following subquery is applied to empty data,
and the list of data sources from the command line is considered empty.
This directive is useful in front of \*(TX code which doesn't process data
sources from the command line, but takes command-line arguments.
The
.code "@(next nil)"
incantation absolutely prevents \*(TX from trying to open the
first command-line argument as a data source.
Note that the
.code @(next)
directive only redirects the source of input over the scope of subquery in which
the that directive appears. For
example, the following query looks for the line starting with
.str "xyz"
at the top of the file
.strn "foo.txt" ,
within a
.code some
directive. After the
.code @(end)
which terminates the
.codn @(some) ,
the
.str "abc"
is matched in the previous input stream which was in effect before
the
.code @(next)
directive:
.verb
@(some)
@(next "foo.txt")
xyz@suffix
@(end)
abc
.brev
However, if the
.code @(some)
subquery successfully matched
.str "xyz@suffix"
within the
file
.codn foo.text ,
there is now a binding for the
.code suffix
variable, which
is visible to the remainder of the entire query. The variable bindings
survive beyond the clause, but the data stream does not.
.dir skip
The
.code skip
directive considers the remainder of the query as a search
pattern. The remainder is no longer required to strictly match at the
current line in the current input stream. Rather, the current stream is searched,
starting with the current line, for the first line where the entire remainder
of the query will successfully match. If no such line is found, the
.code skip
directive fails. If a matching position is found, the remainder of
the query is processed from that point.
The remainder of the query can itself contain
.code skip
directives.
Each such directive performs a recursive subsearch.
Skip comes in vertical and horizontal flavors. For instance, skip and match the
last line:
.verb
@(skip)
@last
@(eof)
.brev
Skip and match the last character of the line:
.verb
@(skip)@{last 1}@(eol)
.brev
The
.code skip
directive has two optional arguments, which are evaluated as \*(TL
expressions. If the first argument evaluates to an integer,
its value limits the range of lines scanned for a match. Judicious use
of this feature can improve the performance of queries.
Example: scan until
.str "size: @SIZE"
matches, which must happen within
the next 15 lines:
.verb
@(skip 15)
size: @SIZE
.brev
Without the range limitation,
.code skip
will keep searching until it consumes the entire input source.
In a horizontal
.codn skip ,
the range-limiting numeric argument is expressed in characters, so that
.verb
abc@(skip 5)def
.brev
means: there must be a match for
.str "abc"
at the start of the line, and then within the next five characters,
there must be a match for
.strn "def" .
Sometimes a skip is nested within a
.codn collect ,
or
following another skip. For instance, consider:
.verb
@(collect)
begin @BEG_SYMBOL
@(skip)
end @BEG_SYMBOL
@(end)
.brev
The above
.code collect
iterates over the entire input. But, potentially, so does
the embedded
.codn skip .
Suppose that
.str "begin x"
is matched, but the data has no
matching
.strn "end x" .
The skip will search in vain all the way to the end of the
data, and then the collect will try another iteration back at the
beginning, just one line down from the original starting point. If it is a
reasonable expectation that an
.code "end x"
occurs 15 lines of a
.strn "begin x" ,
this can be specified instead:
.verb
@(collect)
begin @BEG_SYMBOL
@(skip 15)
end @BEG_SYMBOL
@(end)
.brev
If the symbol
.code nil
is used in place of a number, it means to scan
an unlimited range of lines; thus,
.code "@(skip nil)"
is equivalent to
.codn @(skip) .
If the symbol
.code :greedy
is used, it changes the semantics of the skip
to longest match semantics. For instance, match the last three space-separated
tokens of the line:
.verb
@(skip :greedy) @a @b @c
.brev
Without
.codn :greedy ,
the variable
.code @c
may match multiple tokens,
and end up with spaces in it, because nothing follows
.code @c
and so it matches from any position which follows a space to the
end of the line. Also note the space in front of
.codn @a .
Without this
space,
.code @a
will get an empty string.
A line-oriented example of greedy skip: match the last line without
using
.codn @(eof) :
.verb
@(skip :greedy)
@last_line
.brev
There may be a second numeric argument. This specifies a minimum
number of lines to skip before looking for a match. For instance,
skip 15 lines and then search indefinitely for
.codn "begin ..." :
.verb
@(skip nil 15)
begin @BEG_SYMBOL
.brev
The two arguments may be used together. For instance, the following
matches if and only if the 15th line of input starts with
.codn "begin " :
.verb
@(skip 1 15)
begin @BEG_SYMBOL
.brev
Essentially,
.mono
.meti @(skip 1 << n )
.onom
means "hard skip by
.meta n
lines".
.code "@(skip 1 0)"
is the same as
.codn "@(skip 1)" ,
which is a noop, because it means: "the remainder of the query must match
starting on the next line", or, more briefly, "skip exactly zero lines",
which is the behavior if the
.code skip
directive is omitted altogether.
Here is one trick for grabbing the fourth line from the bottom of the input:
.verb
@(skip)
@fourth_from_bottom
@(skip 1 3)
@(eof)
.brev
Or using greedy skip:
.verb
@(skip :greedy)
@fourth_from_bottom
@(skip 1 3)
.brev
Non-greedy skip with the
.code @(eof)
directive has a slight advantage because the greedy skip
will keep scanning even though it has found the correct match, then backtrack
to the last good match once it runs out of data. The regular skip with explicit
.code @(eof)
will stop when the
.code @(eof)
matches.
.NP* Reducing Backtracking with Blocks
The
.code skip
directive can consume considerable CPU time when multiple skips are nested.
Consider:
.verb
@(skip)
A
@(skip)
B
@(skip)
C
.brev
This is actually nesting: the second and third skips occur within the body of the
first one, and thus this creates nested iteration. \*(TX is searching for the
combination of skips which match the pattern of lines
.codn A ,
.code B
and
.code C
with
backtracking behavior. The outermost skip marches through the data until it
finds
.code A
followed by a pattern match for the second skip. The second skip
iterates to find
.code B
followed by the third skip, and the third skip
iterates to find
.codn C .
If
.code A
and
.code B
are only one line each,
then this is reasonably fast. But suppose there are many lines matching
.code A
and
.codn B ,
giving rise to a large number of combinations of skips which match
.code A
and
.codn B ,
and yet do not find a match for
.codn C ,
triggering backtracking. The nested stepping which tries
the combinations of
.code A
and
.code B
can give rise to a considerable running time.
One way to deal with the problem is to unravel the nesting with the help of
blocks. For example:
.verb
@(block)
@ (skip)
A
@(end)
@(block)
@ (skip)
B
@(end)
@(skip)
C
.brev
Now the scope of each skip is just the remainder of the block in which it
occurs. The first skip finds
.codn A ,
and then the block ends. Control passes to the
next block, and backtracking will not take place to a block which completed
(unless all these blocks are enclosed in some larger construct which
backtracks, causing the blocks to be re-executed.
This rewrite is not equivalent, and cannot be used for instance
in backreferencing situations such as:
.verb
@;
@; Find three lines anywhere in the input which are identical.
@;
@(skip)
@line
@(skip)
@line
@(skip)
@line
.brev
This example depends on the nested search-within-search semantics.
.dir trailer
The
.code trailer
directive introduces a trailing portion of a query or subquery
which matches input material normally, but in the event of a successful match,
does not advance the current position. This can be used, for instance, to
cause
.code @(collect)
to match partially overlapping regions.
Trailer can be used in vertical context:
.mono
.mets @(trailer)
.mets < directives
.mets ...
.onom
or horizontal:
.mono
.mets @(trailer) < directives ...
.onom
A vertical
.code trailer
prevents the vertical input position from
advancing as it is matched by
.metn directives ,
whereas a horizontal
.code trailer
prevents the horizontal position from advancing. In other words,
.code trailer
performs matching without consuming the input, providing a
lookahead mechanism.
Example:
.verb
@(collect)
@line
@(trailer)
@(skip)
@line
@(end)
.brev
This script collects each line which has a duplicate somewhere later
in the input. Without the
.code @(trailer)
directive, this does not work properly
for inputs like:
.verb
111
222
111
222
.brev
Without
.codn @(trailer) ,
the first duplicate pair constitutes a match which
spans over the
.codn 222 .
After that pair is found, the matching continues
after the second
.codn 111 .
With the
.code @(trailer)
directive in place, the collect body, on each
iteration, only consumes the lines matched prior to
.codn @(trailer) .
.dir freeform
The
.code freeform
directive provides a useful alternative to
\*(TX's line-oriented matching discipline. The
.code freeform
directive treats all
remaining input from the current input source as one big line. The query line
which immediately follows freeform is applied to that line.
The syntax variations are:
.verb
@(freeform)
... query line ..
.mets @(freeform << number )
... query line ..
.mets @(freeform << string )
... query line ..
.mets @(freeform < number << string )
... query line ..
.brev
where
.meta number
and
.meta string
denote \*(TL expressions which evaluate to an integer or string
value, respectively.
If
.meta number
and
.meta string
are both present, they may be given in either order.
If the
.meta number
argument is given, its value limits the range of lines which are combined
together. For instance
.code "@(freeform 5)"
means to only consider the next five lines
to be one big line. Without this argument,
.code freeform
is "bottomless".
It can match the entire file,
which creates the risk of allocating a large amount of memory.
If the
.meta string
argument is given, it specifies a custom line terminator. The
default terminator is
.strn "\en" .
The terminator does not have to be one character long.
Freeform does not convert the entire remainder of the input into one big line
all at once, but does so in a dynamic, lazy fashion, which takes place as the
data is accessed. So at any time, only some prefix of the data exists as a flat
line in which newlines are replaced by the terminator string, and the remainder
of the data still remains as a list of lines.
After the subquery is applied to the virtual line, the unmatched remainder
of that line is broken up into multiple lines again, by looking for and
removing all occurrences of the terminator string within the flattened portion.
Care must be taken if the terminator is other than the default
.strn "\en" .
All occurrences of the terminator string are treated as line terminators in
the flattened portion of the data, so extra line breaks may be introduced.
Likewise, in the yet unflattened portion, no breaking takes place, even if
the text contains occurrences of the terminator string. The extent of data which
is flattened, and the amount of it which remains, depends entirely on the
query line underneath
.codn @(flatten) .
In the following example, lines of data are flattened using $ as the line
terminator.
.IP code:
.mono
\ @(freeform "$")
@a$@b:
@c
@d
.onom
.IP data:
.mono
\ 1
2:3
4
.onom
.IP "output (\f[4]-B\f[]):"
.mono
\ a="1"
b="2"
c="3"
d="4"
.onom
.PP
The data is turned into the virtual line
.codn 1$2:3$4$ .
The
.code @a$@b:
subquery matches
the
.code 1$2:
portion, binding
.code a
to
.strn 1 ,
and
.code b
to
.strn 2 .
The remaining portion
.code 3$4$
is then split into separate lines again according to the line terminator
.codn $i :
.verb
3
4
.brev
Thus the remainder of the query
.verb
@c
@d
.brev
faces these lines, binding
.code c
to
.code 3
and
.code d
to
.codn 4 .
Note that since the data
does not contain dollar signs, there is no ambiguity; the meaning may be
understood in terms of the entire data being flattened and split again.
In the following example,
.code freeform
is used to solve a tokenizing problem. The
Unix password file has fields separated by colons. Some fields may be empty.
Using freeform, we can join the password file using
.str ":"
as a terminator.
By restricting freeform to one line, we can obtain each line of the password
file with a terminating
.strn ":" ,
allowing for a simple tokenization, because
now the fields are colon-terminated rather than colon-separated.
Example:
.verb
@(next "/etc/passwd")
@(collect)
@(freeform 1 ":")
@(coll)@{token /[^:]*/}:@(end)
@(end)
.brev
.dir fuzz
The
.code fuzz
directive allows for an imperfect match spanning a set number of
lines. It takes two arguments, both of which are \*(TL expressions that should
evaluate to integers:
.mono
.meti @(fuzz m n)
...
.onom
This expresses that over the next
.meta n
query lines, the matching strictness
is relaxed a little bit. Only
.meta m
out of those
.meta n
lines have to match.
Afterward, the rest of the query follows normal, strict processing.
In the degenerate situation where there are fewer than
.meta n
query lines following the
.code fuzz
directive, then
.meta m
of them must succeed anyway. (If there
are fewer than
.metn m ,
then this is impossible.)
.dirs line chr
The
.code line
and
.code chr
directives perform binding between the current input line number or character
position within a line, against an expression or variable:
.verb
@(line 42)
@(line x)
abc@(chr 3)def@(chr y)
.brev
The directive
.code "@(line 42)"
means "match the current input line number against the integer 42". If
the current line is 42, then the directive matches, otherwise it fails.
.code line
is a vertical directive which doesn't consume a line of input. Thus,
the following matches at the beginning of an input stream, and
.code x
ends up bound to the first line of input:
.verb
@(line 1)
@(line 1)
@(line 1)
@x
.brev
The directive
.code "@(line x)"
binds variable
.code x
to the current input line number, if
.code x
is an unbound variable. If
.code x
is already bound, then the value of
.code x
must match the current line number, otherwise the directive fails.
The
.code chr
directive is similar to
.code line
except that it's a horizontal directive, and matches the character position
rather than the line position. Character positions are measured from zero,
rather than one.
.code chr
does not consume a character. Hence the two occurrences of
.code chr
in the following example both match, and
.code x
takes the entire line of input:
.verb
@(chr 0)@(chr 0)@x
.brev
The argument of
.code line
or
.code chr
may be an
.codn @ -delimited
Lisp expression. This is useful for matching computed lines or
character positions:
.verb
@(line @(+ a (* b c)))
.brev
.dir name
The
.code name
directive performs a binding between the name of the current
data source and a variable or bind expression:
.verb
@(name na)
@(name "data.txt")
.brev
If
.code na
is an unbound variable, it is bound and takes on the name
of the data source, such as a file name. If
.code na
is bound, then it has to match the name of the data source,
otherwise the directive fails.
The directive
.mono
@(name "data.txt")
.onom
fails unless the current data source has that name.
.dir data
The
.code data
directive performs a binding between the unmatched data
at the current position, and and a variable or bind expression.
The unmatched data takes the form of a list of strings:
.verb
@(data d)
.brev
The binding is performed on object equality. If
.code d
is already bound, a matching failure occurs unless
.code d
contains the current unmatched data.
Matching the current data has various uses.
For instance, two branches of pattern matching can, at some point, bind the
current data into different variables. When those paths join, the variables can
be bound together to create the assertion that the current data had been the
same at those points:
.verb
@(all)
@ (skip)
foo
@ (skip)
bar
@ (data x)
@(or)
@ (skip)
xyzzy
@ (skip)
bar
@ (data y)
@(end)
@(require (eq x y))
.brev
Here, two branches of the
.code @(all)
match some material which ends in the line
.codn "bar" .
However, it is possible that this is a different line. The
.code data
directives are used to create an assertion that the data regions matched
by the two branches are identical. That is to say, the unmatched data
.code x
captured after the first
.code "bar"
and the unmatched data
.code y
captured after the second
.code "bar"
must be the same object in order for
.code "@(require (eq x y))"
to succeed, which implies that the same
.code "bar"
was matched in both branches of the
.codn @(all) .
Another use of
.code data
is simply to gain access to the trailing remainder of the unmatched
input in order to print it, or do some special processing on it.
The
.code tprint
Lisp function is useful for printing the unmatched data as newline-terminated
lines:
.verb
@(data remainder)
@(do (tprint remainder))
.brev
.dir eof
The
.code eof
directive, if not given any argument, matches successfully when no more input
is available from the current input source.
In the following example, the
.meta line
variable captures the text
.str "One-line file"
and then since that is the last line of input, the
.code eof
directive matches:
.IP code:
.mono
\ @line
@(eof)
.onom
.IP data:
.mono
\ One-line file
.onom
.PP
If the data consisted of two or more lines,
.code eof
would fail.
The
.code eof
directive may be given a single argument, which is a pattern that matches the
termination status of the input source. This is useful when the input source
is a process pipe. For the purposes of
.codn eof ,
sources which are not process pipes have the symbol
.code t
as their termination status.
In the following example, which assumes the availability of a POSIX shell
command interpreter in the host system, the variable
.meta a
captures the string
.str a
and the
.meta status
variable captures the integer value
.codn 5 ,
which is the termination status of the command:
.verb
@(next (open-command "echo a; exit 5"))
@a
@(eof status)
.brev
.dirs some all none maybe cases choose
These directives, called the parallel directives, combine multiple subqueries,
which are applied at the same input position, rather than to consecutive input.
They come in vertical (line mode) and horizontal (character mode) flavors.
In horizontal mode, the current position is understood to be a character
position in the line being processed. The clauses advance this character
position by moving it to the right. In vertical mode, the current position is
understood to be a line of text within the stream. A clause advances the
position by some whole number of lines.
The syntax of these parallel directives follows this example:
.verb
@(some)
subquery1
.
.
.
@(and)
subquery2
.
.
.
@(and)
subquery3
.
.
.
@(end)
.brev
And in horizontal mode:
.verb
@(some)subquery1...@(and)subquery2...@(and)subquery3...@(end)
.brev
Long horizontal lines can be broken up with line continuations, allowing the
above example to be written like this, which is considered a single logical
line:
.verb
@(some)@\e
subquery1...@\e
@(and)@\e
subquery2...@\e
@(and)@\e
subquery3...@\e
@(end)
.brev
The
.codn @(some) ,
.codn @(all) ,
.codn @(none) ,
.codn @(maybe) ,
.code @(cases)
or
.code @(choose)
must be followed
by at least one subquery clause, and be terminated by
.codn @(end) .
If there are two
or more subqueries, these additional clauses are indicated by
.code @(and)
or
.codn @(or) ,
which are interchangeable. The separator and terminator directives also must
appear as the only element in a query line.
The
.code choose
directive requires keyword arguments. See below.
The syntax supports arbitrary nesting. For example:
.verb
QUERY: SYNTAX TREE:
@(all) all -+
@ (skip) +- skip -+
@ (some) | +- some -+
it | | +- TEXT
@ (and) | | +- and
@ (none) | | +- none -+
was | | | +- TEXT
@ (end) | | | +- end
@ (end) | | +- end
a dark | +- TEXT
@(end) *- end
.brev
nesting can be indicated using whitespace between
.code @
and the
directive expression. Thus, the above is an
.code @(all)
query containing a
.code @(skip)
clause which applies to a
.code @(some)
that is followed by the text line
.strn "a dark" .
The
.code @(some)
clause combines the text line
.strn it ,
and a
.code @(none)
clause which contains just one clause consisting of the line
.strn was .
The semantics of the parallel directives is:
.coIP @(all)
Each of the clauses is matched at the current position. If any of the
clauses fails to match, the directive fails (and thus does not produce
any variable bindings). Clauses following the failed directive are not
evaluated. Bindings extracted by a successful clause are visible to the clauses
which follow, and if the directive succeeds, all of the combined bindings
emerge.
.meIP @(some [ :resolve >> ( var ...) ])
Each of the clauses is matched at the current position. If any of the clauses
succeed, the directive succeeds, retaining the bindings accumulated by the
successfully matching clauses. Evaluation does not stop on the first successful
clause. Bindings extracted by a successful clause are visible to the clauses
which follow.
The
.code :resolve
parameter is for situations when the
.code @(some)
directive has
multiple clauses that need to bind some common variables to different
values: for instance, output parameters in functions. Resolve takes
a list of variable name symbols as an argument. This is called the
resolve set. If the clauses of
.code @(some)
bind variables in the resolve
set, those bindings are not visible to later clauses. However, those
bindings do emerge out of the
.code @(some)
directive as a whole.
This creates a conflict: what if two or more clauses introduce
different bindings for a variable in the resolve set?
This is why it is called the resolve set: conflicts for variables in the
resolve set are automatically resolved in favor of later directives.
Example:
.verb
@(some :resolve (x))
@ (bind a "a")
@ (bind x "x1")
@(or)
@ (bind b "b")
@ (bind x "x2")
@(end)
.brev
Here, the two clauses both introduce a binding for
.codn x .
Without the
.code :resolve
parameter, this would mean that the second clause fails, because
.code x
comes in
with the value
.strn x1 ,
which does not bind with
.strn x2 .
But because
.code x
is placed
into the resolve set, the second clause does not see the
.str x1
binding. Both
clauses establish their bindings independently creating a conflict over
.codn x .
The conflict is resolved in favor of the second clause, and so the bindings
which emerge from the directive are:
.verb
a="a"
b="b"
x="x2"
.brev
.coIP @(none)
Each of the clauses is matched at the current position. The
directive succeeds only if all of the clauses fail. If
any clause succeeds, the directive fails, and subsequent clauses are not
evaluated. Thus, this directive never produces variable bindings, only matching
success or failure.
.coIP @(maybe)
Each of the clauses is matched at the current position. The directive always
succeeds, even if all of the clauses fail. Whatever bindings are found in any
of the clauses are retained. Bindings extracted by any successful clause are
visible to the clauses which follow.
.coIP @(cases)
Each of the clauses is matched at the current position.
The clauses are matched, in order, at the current position.
If any clause matches, the matching stops and the bindings
collected from that clause are retained. Any remaining clauses
after that one are not processed. If no clause matches, the
directive fails, and produces no bindings.
.meIP @(choose [ :longest < var | :shortest < var ])
Each of the clauses is matched at the current position in order. In this
construct, bindings established by an earlier clause are not visible to later
clauses. Although any or all of the clauses can potentially match, the clause
which succeeds is the one which maximizes or minimizes the length of the
text bound to the specified variable. The other clauses have no effect.
For all of the parallel directives other than
.code @(none)
and
.codn @(choose) ,
the query
advances the input position by the greatest number of lines that match in any
of the successfully matching subclauses that are evaluated. The
.code @(none)
directive does not advance the input position.
For instance if there are two subclauses, and one of them matches three lines,
but the other one matches five lines, then the overall clause is considered to
have made a five line match at its position. If more directives follow, they
begin matching five lines down from that position.
.dir require
The syntax of
.code @(require)
is:
.mono
.mets @(require << lisp-expression )
.onom
The
.code require
directive evaluates a \*(TL expression. (See TXR LISP far
below.) If the expression yields a true value, then it succeeds, and matching
continues with the directives which follow. Otherwise the directive fails.
In the context of the
.code require
directive, the expression should not be introduced by the
.code @
symbol; it is expected to be a Lisp expression.
Example:
.verb
@; require that 4 is greater than 3
@; This succeeds; therefore, @a is processed
@(require (> (+ 2 2) 3))
@a
.brev
.dir if
The
.code if
directive allows for conditional selection of pattern-matching clauses,
based on the Boolean results of Lisp expressions.
A variant of the
.code if
directive is also available for use inside an
.code output
clauses, where it similarly allows for the conditional selection of output
clauses.
The syntax of the
.code if
directive can be exemplified as follows:
.mono
.mets @(if << lisp-expr )
.
.
.
.mets @(elif << lisp-expr )
.
.
.
.mets @(elif << lisp-expr )
.
.
.
@(else)
.
.
.
@(end)
.onom
The
.code @(elif)
and
.code @(else)
clauses are all optional. If
.code @(else)
is present, it must be
last, before
.codn @(end) ,
after any
.code @(elif)
clauses. Any of the clauses may be empty.
.TP* "Example:"
.verb
@(if (> (length str) 42))
foo: @a @b
@(else)
{@c}
@(end)
.brev
In this example, if the length of the variable
.code str
is greater than
.codn 42 ,
then matching continues with
.strn "foo: @a b" ,
otherwise it proceeds with
.codn {@c} .
.PP
More precisely, how the
.code if
directive works is as follows. The Lisp expressions are evaluated in order,
starting with the
.code if
expression, then the
.code elif
expressions if any are present. If any Lisp expression yields a true result
(any value other than
.codn nil )
then evaluation of Lisp expressions stops. The corresponding clause of that
Lisp expression is selected and pattern matching continues
with that clause. The result of that clause (its success or failure,
and any newly bound variables) is then taken as the result of the
.code if
directive. If none of the Lisp expressions yield true, and an
.code else
clause is present, then that clause is processed and its result determines
the result of the
.code if
directive. If none of the Lisp expressions
yield true, and there is no
.code else
clause, then the
.code if
directive is deemed to have trivially succeeded, allowing matching to continue
with whatever directive follows it.
.coNP The Lisp @ if versus TXR @ if
The
.code @(output)
directive supports the embedding of Lisp expressions, whose values are
interpolated into the output. In particular, Lisp
.code if
expressions are useful. For instance
.code "@(if expr \(dqA\(dq \(dqB\(dq)"
reproduces
.code A
if
.code expr
yields a true value, otherwise
.codn B .
Yet the
.code @(if)
directive is also supported in
.codn @(output) .
How the apparent conflict between the two is resolved is that the two take
different numbers of arguments. An
.code @(if)
which has no arguments at all is a syntax error. One that has one argument
is the head of the
.code if
directive syntax which must be terminated by
.code @(end)
and which takes the optional
.code @(elif)
and
.code @(else)
clauses. An
.code @(if)
which has two or more arguments is parsed as a self-contained Lisp expression.
.dir gather
Sometimes text is structured as items that can appear in an arbitrary order.
When multiple matches need to be extracted, there is a combinatorial explosion
of possible orders, making it impractical to write pattern matches for all
the possible orders.
The
.code gather
directive is for these situations. It specifies multiple clauses
which all have to match somewhere in the data, but in any order.
For further convenience, the lines of the first clause of the
.code gather
directive
are implicitly treated as separate clauses.
The syntax follows this pattern:
.verb
@(gather)
one-line-query1
one-line-query2
.
.
.
one-line-queryN
@(and)
multi
line
query1
.
.
.
@(and)
multi
line
query2
.
.
.
@(end)
.brev
The multiline clauses are optional. The
.code gather
directive takes
keyword parameters, see below.
.coNP The @ until / @ last clause in @ gather
Similarly to
.codn collect ,
.code gather
has an optional
.cod3 until / last
clause:
.verb
@(gather)
...
@(until)
...
@(end)
.brev
How
.code gather
works is that the text is searched for matches for the single-line
and multiline queries. The clauses are applied in the order in which they appear.
Whenever one of the clauses matches, any bindings it produces are retained and
it is removed from further consideration. Multiple clauses can match at the
same text position. The position advances by the longest match from among the
clauses which matched. If no clauses match, the position advances by one line.
The search stops when all clauses are eliminated, and then the cumulative
bindings are produced. If the data runs out, but unmatched clauses remain, the
directive fails.
Example: extract several environment variables, which do not appear in a particular
order:
.verb
@(next :env)
@(gather)
USER=@USER
HOME=@HOME
SHELL=@SHELL
@(end)
.brev
If the
.code until
or
.code last
clause is present and a match occurs, then the matches
from the other clauses are discarded and the
.code gather
terminates. The difference between
.cod3 until / last
is that any bindings bindings established in
.code last
are retained,
and the input position is advanced past the matching material.
The
.cod3 until / last
clause has visibility to bindings established in the
previous clauses in that same iteration, even though those bindings
end up thrown away.
For consistency, the
.code :mandatory
keyword is supported in the
.cod3 until / last
clause of
.codn gather .
The semantics of using
.code :mandatory
in this situation is tricky. In particular, if it is in effect, and the
.code gather
terminates successfully by collecting all required matches, it will
trigger a failure. On the other hand, if the
.code until
or
.code last
clause activates before all required matches are gathered, a failure
also occurs, whether or not the clause is
.codn :mandatory .
Meaningful use of
.code :mandatory
requires that the gather be open-ended; it must allow some (or all) variables
not to be required. The presence of the option means that for
.code gather
to succeed, all required variables must be gathered first, but then termination
must be achieved via the
.cod3 until / last
clause before all
.code gather
clauses are satisfied.
.coNP Keyword parameters in @ gather
The
.code gather
directive accepts the keyword parameter
.codn :vars .
The argument to
.code :vars
is a list of required and optional variables. A required variable is specified
as a symbol. An optional variable is specified as a two element list which
pairs a symbol with a Lisp expression. That Lisp expression is evaluated
and specifies the default value for the variable.
Example:
.verb
@(gather :vars (a b c (d "foo")))
...
@(end)
.brev
Here,
.codn a ,
.code b
and
.code c
are required variables, and
.code d
is optional, with the default value given by the Lisp expression
.strn foo .
The presence of
.code :vars
changes the behavior in three ways.
Firstly, even if all the clauses in the
.code gather
match successfully and are
eliminated, the directive will fail if the required variables do not have
bindings. It doesn't matter whether the bindings are existing, or whether they
are established by
.codn gather .
Secondly, if some of the clauses of
.code gather
did not match, but all
of the required variables have bindings, then the directive succeeds.
Without the presence of
.codn :vars ,
it would fail in this situation.
Thirdly, if
.code gather
succeeds (all required variables have bindings),
then all of the optional variables which do not have bindings are given
bindings to their default values.
The expressions which give the default values are evaluated whenever
the
.code gather
directive is evaluated, whether or not their values are used.
.dir collect
The syntax of the
.code collect
directive is:
.verb
@(collect)
... lines of subquery
@(end)
.brev
or with an
.code until
or
.code last
clause:
.verb
@(collect)
... lines of subquery: main clause
@(until)
... lines of subquery: until clause
@(end)
@(collect)
... lines of subquery: main clause
@(last)
... lines of subquery: last clause
@(end)
.brev
The
.code repeat
symbol may be specified instead of
.codn collect ,
which changes the meaning, see below:
.verb
@(repeat)
... lines of subquery
@(end)
.brev
The subquery is matched repeatedly, starting at the current line.
If it fails to match, it is tried starting at the subsequent line.
If it matches successfully, it is tried at the line following the
entire extent of matched data, if there is one. Thus, the collected regions do
not overlap. (Overlapping behavior can be obtained: see the
.code @(trailer)
directive.)
Unless certain keywords are specified, or unless the collection is explicitly
failed with
.codn @(fail) ,
it always succeeds, even if it collects nothing,
and even if the
.cod3 until / last
clause never finds a match.
If no
.cod3 until / last
clause is specified, and the
.code collect
is not limited
using parameters, the collection is unbounded: it consumes the entire data
file.
.coNP The @ until / @ last clause in @ collect
If an
.cod3 until / last
clause is specified, the collection stops when that clause
matches at the current position.
If an
.code until
clause terminates
.codn collect ,
no bindings are collected at that position,
even if the main clause matches at that position also.
Moreover, the position is not advanced.
The remainder of the query begins matching at that position.
If a
.code last
clause terminates
.codn collect ,
the behavior is different. Any bindings
captured by the main clause are thrown away, just like with the
.code until
clause.
However, the bindings in the
.code last
clause itself survive, and the position is
advanced to skip over that material.
Example:
.IP code:
.mono
\ @(collect)
@a
@(until)
42
@b
@(end)
@c
.onom
.IP data:
.mono
\ 1
2
3
42
5
6
.onom
.IP result:
.mono
\ a[0]="1"
a[1]="2"
a[2]="3"
c="42"
.onom
.PP
The line
.code 42
is not collected, even though it matches
.codn @a .
Furthermore,
the
.code @(until)
does not advance the position, so variable
.code c
takes
.codn 42 .
If the
.code @(until)
is changed to
.code @(last)
the output will be different:
.IP result:
.mono
\ a[0]="1"
a[1]="2"
a[2]="3"
b="5"
c="6"
.onom
.PP
The
.code 42
is not collected into a list, just like before. But now
the binding captured by
.code @b
emerges. Furthermore, the position advances
so variable now takes
.codn 6 .
The binding variables within the clause of a
.code collect
are treated specially.
The multiple matches for each variable are collected into lists,
which then appear as array variables in the final output.
Example:
.IP code:
.mono
\ @(collect)
@a:@b:@c
@(end)
.onom
.IP data:
.mono
\ John:Doe:101
Mary:Jane:202
Bob:Coder:313
.onom
.IP result:
.mono
\ a[0]="John"
a[1]="Mary"
a[2]="Bob"
b[0]="Doe"
b[1]="Jane"
b[2]="Coder"
c[0]="101"
c[1]="202"
c[2]="313"
.onom
.PP
The query matches the data in three places, so each variable becomes
a list of three elements, reported as an array.
Variables with list bindings may be referenced in a query. They denote a
multiple match. The
.code -D
command-line option can establish a one-dimensional
list binding.
The clauses of
.code collect
may be nested.
Variable matches collated into lists in an inner
.code collect
are again collated into nested lists in the outer
.codn collect .
Thus an unbound variable wrapped in N nestings of
.code @(collect)
will
be an N-dimensional list. A one-dimensional list is a list of strings;
a two-dimensional list is a list of lists of strings, etc.
It is important to note that the variables which are bound within the main
clause of a
.codn collect ,
that is, the variables which are subject to collection, appear, within the
.codn collect ,
as normal one-value bindings.
The collation into lists happens outside of the
.codn collect .
So for instance in the query:
.mono
@(collect)
@x=@x
@(end)
.onom
The left
.code @x
establishes a binding for some material preceding an equal sign.
The right
.code @x
refers to that binding. The value of
.code @x
is different in each
iteration, and these values are collected. What finally comes out of the
.code collect
clause is a single variable called
.code x
which holds a list containing each value that
was ever instantiated under that name within the
.code collect
clause.
Also note that the
.code until
clause has visibility over the bindings
established in the main clause. This is true even in the terminating
case when the
.code until
clause matches, and the bindings of the main clause are discarded.
.coNP Keyword parameters in @ collect
By default,
.code collect
searches the rest of the input indefinitely,
or until the
.cod3 until / last
clause matches. It skips arbitrary amounts of
nonmatching material before the first match, and between matches.
Within the
.code @(collect)
syntax, it is possible to specify keyword
parameters for additional control of the behavior. A keyword parameter
consist of a keyword symbol followed by an argument, enclosed within
the
.code @(collect)
syntax. The following are the supported keywords.
.meIP :maxgap < n
The
.code :maxgap
keyword takes a numeric argument
.metn n ,
which is a Lisp expression.
It causes
.code collect
to terminate if it fails to find a match after skipping
.meta n
lines from the starting position,
or more than
.meta n
lines since any successful match. For example,
.verb
@(collect :maxgap 5)
.brev
specifies that the gap between the current position and the first
match for the body of the
.codn collect ,
or between consecutive matches can be no longer than five lines. A
.code :maxgap
value of
.code 0
means that the collected regions must be
adjacent and must match right from the starting position. For instance:
.verb
@(collect :maxgap 0)
M @a
@(end)
.brev
means: from here, collect consecutive lines of the form
.strn "M ..." .
This will not
search for the first such line, nor will it skip lines which do not match this
form.
.meIP :mingap < n
The
.code :mingap
keyword complements
.codn :maxgap ,
though not exactly. Its argument
.metn n ,
a Lisp expression, specifies a minimum number
of lines which must separate consecutive matches. However, it has no effect on
the distance from the starting position to the first match.
.meIP :gap < n
The
.code :gap
keyword effectively specifies
.code :mingap
and
.code :maxgap
at the same time, and can only be
used if these other two are not used. Thus:
.verb
@(collect :gap 1)
@a
@(end)
.brev
means: collect every other line starting with the current line.
.meIP :times < n
This shorthand means the same thing as if
.meIP :mintimes < n :maxtimes < n
were specified. This means that exactly
.meta n
matches must occur. If fewer occur, then
.code collect
fails.
The
.code collect
stops once it achieves
.code n
matches.
.meIP :mintimes < n
The argument
.meta n
of the
.code :mintimes
keyword is a Lisp expression which specifies that at least
.meta n
matches must occur, or else
.code collect
fails.
.meIP :mintimes < n
The Lisp argument expression
.meta n
of the
.code :mintimes
keyword specifies that at most
.meta n
matches are collected.
.meIP :lines < n
The argument
.meta n
of the
.code :lines
keyword parameter
is a Lisp expression which specifies the upper bound on how many lines
should be scanned by
.codn collect ,
measuring from the starting position.
The extent of the
.code collect
body is not counted. Example:
.verb
@(collect :lines 2)
foo: @a
bar: @b
baz: @c
@(end)
.brev
The above
.code collect
will look for a match only twice: at the current position,
and one line down.
.meIP :vars >> ({ variable | >> ( variable << default-value)}*)
The
.code :vars
keyword specifies a restriction on what variables will emanate
from the
.codn collect .
Its argument is a list of variable names.
An empty list may be specified using empty parentheses
or, equivalently, the symbol
.codn nil .
The
.meta default-value
element of the syntax is a Lisp expression.
The behavior of the
.code :vars
keyword is specified in the following section, "Specifying variables in
.codn collect \(dq.
.meIP :lists <> ( variable *)
The
.code :lists
keyword indicates a list of variables. After the
.code collect
terminates, each
.meta variable
in the list which does not have a binding is bound to the empty
list symbol
.codn nil .
Unlike
.code :vars
the
.code :lists
mechanism doesn't assert that only the listed variables may emanate
from the
.codn collect .
It also doesn't assert that each iteration of the
.code collect
must bind each of those variables.
.meIP :counter >> { variable | >> ( variable << starting-value )}
The
.code :counter
keyword's argument is a variable name symbol,
or a compound expression consisting of a variable name symbol
and the \*(TL expression
.metn starting-value .
If this keyword argument is specified, then a binding for
.meta variable
is established prior to each repetition of the
.code collect
body, to an integer value representing the repetition count.
By default, repetition counts begin at zero.
If
.meta starting-value
is specified, it must evaluate to a number. This number is
then added to each repetition count, and
.meta variable
takes on the resulting displaced value.
If there is an existing binding for
.meta variable
prior to the processing of the
.codn collect ,
then the variable is shadowed.
The binding is collected in the same way as other bindings
that are established in the
.code collect
body.
The repetition count only increments after a successful match.
The
.code variable
is visible to the
.codn collect 's
.cod3 until / last
clause. If that clause is being processed after a successful match
of the body, then
.meta variable
holds an integer value. If the body fails to match, then the
.cod3 until / last
clause sees a binding for
.code variable
with a value of
.codn nil .
.PP
.coNP Specifying variables in @ collect
Normally, any variable for which a new binding occurs in a
.code collect
block is collected. A
.code collect
clause may be "sloppy": it can neglect to collect
some variables on some iterations, or bind some variables which are intended to
behave like local temporaries, but end up collated into lists. Another issue is
that the
.code collect
clause might not match anything at all, and then none of the
variables are bound.
The
.code :vars
keyword allows the query writer to add discipline the
.code collect
body.
The argument to
.code :vars
is a list of variable specs. A variable spec is either a
symbol, denoting a required variable, or a
.mono
.meti >> ( symbol << default-value )
.onom
pair, where
.meta default-value
is a Lisp expression whose value specifies a default value
for the variable, which is optional.
When a
.code :vars
list is specified, it means that only the given variables can
emerge from the successful
.codn collect .
Any newly introduced bindings for other variables do not propagate.
More precisely, whenever the
.code collect
body matches successfully, the following three rules apply:
.IP 1.
If
.code :vars
specifies required variables, the
.code collect
body must bind all of them,
or else must not bind any variable at all, whether listed in
.code :vars
or not, otherwise an exception of type
.code query-error
is thrown.
.IP 2.
If
.code :vars
specifies required variables, and also specifies default variables,
and the
.code collect
body binds no variable at all, then the default variables
are not bound to their default values.
.IP 3.
If
.code :vars
specifies optional variables, and all required variables are bound by
the
.code collect
body, then all those optional variables that are not bound
by the
.code collect
body are bound to their default values. Under this rule, if
.code :vars
specifies no required variables, that is deemed to be
logically equivalent to all required variables being bound.
.PP
In the event that
.code collect
does not match anything, the variables
specified in
.codn :vars ,
whether required or optional, are all bound to
empty lists. These bindings are established after the processing of the
.cod3 until / last
clause, if present.
Example:
.verb
@(collect :vars (a b (c "foo")))
@a @c
@(end)
.brev
Here, if the body
.str @a @c
matches, an error will be thrown because one of the
mandatory variables is
.codn b ,
and the body neglects to produce a binding for
.codn b .
Example:
.verb
@(collect :vars (a (c "foo")))
@a @b
@(end)
.brev
Here, if
.str @a @b
matches, only
.code a
will be collected, but not
.codn b ,
because
.code b
is not
in the variable list. Furthermore, because there is no binding for
.code c
in the
body, a binding is created with the value
.strn foo ,
exactly as if
.code c
matched
such a piece of text.
In the following example, the assumption is that
.code "THIS NEVER MATCHES"
is not found anywhere in the input but the line
.code "THIS DOES MATCH"
is found and has a successor which is bound to
.codn a .
Because the body did not
match, the
.code :vars
.code a
and
.code b
should be bound to empty lists. But
.code a
is bound by the last clause to some text, so this takes precedence. Only
.code b
is bound to an empty list.
.verb
@(collect :vars (a b))
THIS NEVER MATCHES
@(last)
THIS DOES MATCH
@a
@(end)
.brev
The following means: do not allow any variables to propagate out of any
iteration of the
.code collect
and therefore collect nothing:
.verb
@(collect :vars nil)
...
@(end)
.brev
Instead of writing
.codn "@(collect :vars nil)" ,
it is possible to write
.codn @(repeat) .
.code @(repeat)
takes all
.code collect
keywords, except for
.codn :vars .
There is a
.code @(repeat)
directive used in
.code @(output)
clauses; that is a different directive.
.coNP Mandatory @ until and @ last
The
.cod3 until / last
clause supports the option keyword
.codn :mandatory ,
exemplified by the following:
.verb
@(collect)
...
@(last :mandatory)
...
@(end)
.brev
This means that the
.code collect
.B must
be terminated by a match for the
.cod3 until / last
clause, or else by an explicit
.codn @(accept) .
Specifically, the
.code collect
cannot terminate due to simply running out of data,
or exceeding a limit on the number of matches that may be collected. In
those situations, if an
.code until
or
.code last
clause is present with
.codn :mandatory ,
the
.code collect
is deemed to have failed.
.dir coll
The
.code coll
directive is the horizontal version of
.codn collect .
Whereas
.code collect
works with multiline clauses on line-oriented
material,
.code coll
works within a single line. With
.codn coll ,
it is possible to
recognize repeating regularities within a line and collect lists.
Regular-expression-based Positive Match variables work well with
.codn coll .
Example: collect a comma-separated list, terminated by a space.
.IP code:
.mono
\ @(coll)@{A /[^, ]+/}@(until) @(end)@B
.onom
.IP data:
.mono
\ foo,bar,xyzzy blorch
.onom
.IP result:
.mono
\ A[0]="foo"
A[1]="bar"
A[2]="xyzzy"
B=blorch
.onom
.PP
Here, the variable
.code A
is bound to tokens which match the regular
expression
.codn "/[^, ]+/" :
nonempty sequence of characters other than commas or
spaces.
Like
.codn collect ,
.code coll
searches for matches. If no match
occurs at the current character position, it tries at the next character
position. Whenever a match occurs, it continues at the character position which
follows the last character of the match, if such a position exists.
If not bounded by an until clause, it will exhaust the entire line. If the
until clause matches, then the collection stops at that position,
and any bindings from that iteration are discarded.
Like collect, coll also supports an
.cod3 until / last
clause, which propagates variable
bindings and advances the position. The
.code :mandatory
keyword is supported.
.code coll
clauses nest, and variables bound within a coll are available to clauses
within the rest of the
.code coll
clause, including the
.cod3 until / last
clause, and appear as
single values. The final list aggregation is only visible after the
.code coll
clause.
The behavior of
.code coll
leads to difficulties when a delimited variable are used
to match material which is delimiter separated rather than terminated.
For instance, entries in a comma-separated files usually do
not appear as
.str a,b,c,
but rather
.strn a,b,c .
So for instance, the following result is not satisfactory:
.IP code:
.mono
\ @(coll)@a @(end)
.onom
.IP data:
.mono
\ 1 2 3 4 5
.onom
.IP result:
.mono
\ a[0]="1"
a[1]="2"
a[2]="3"
a[3]="4"
.onom
.PP
The
.code 5
is missing because it isn't followed by a space, which the text-delimited
variable match
.str "@a "
looks for. After matching "4 ", coll continues to look for
matches, and doesn't find any. It is tempting to try to fix it like this:
.IP code:
.mono
\ @(coll)@a@/ ?/@(end)
.onom
.IP data:
.mono
\ 1 2 3 4 5
.onom
.IP result:
.mono
\ a[0]=""
a[1]=""
a[2]=""
a[3]=""
a[4]=""
a[5]=""
a[6]=""
a[7]=""
a[8]=""
.onom
.PP
The problem now is that the regular expression
.code "/ ?/"
(match either a space or nothing), matches at any position.
So when it is used as a variable
delimiter, it matches at the current position, which binds the empty string to
the variable, the extent of the match being zero. In this situation, the
.code coll
directive proceeds character by character. The solution is to use
positive matching: specify the regular expression which matches the item,
rather than a trying to match whatever follows. The
.code collect
directive will
recognize all items which match the regular expression:
.IP code:
.mono
\ @(coll)@{a /[^ ]+/}@(end)
.onom
.IP data:
.mono
\ 1 2 3 4 5
.onom
.IP result:
.mono
\ a[0]="1"
a[1]="2"
a[2]="3"
a[3]="4"
a[4]="5"
.onom
.PP
The
.code until
clause can specify a pattern which, when recognized, terminates
the collection. So for instance, suppose that the list of items may
or may not be terminated by a semicolon. We must exclude
the semicolon from being a valid character inside an item, and
add an until clause which recognizes a semicolon:
.IP code:
.mono
\ @(coll)@{a /[^ ;]+/}@(until);@(end);
.onom
.IP data:
.mono
\ 1 2 3 4 5;
.onom
.IP result:
.mono
\ a[0]="1"
a[1]="2"
a[2]="3"
a[3]="4"
a[4]="5"
.onom
.PP
Whether followed by the semicolon or not, the items are collected properly.
Note that the
.code @(end)
is followed by a semicolon. That's because
when the
.code @(until)
clause meets a match, the matching material
is not consumed.
This repetition can be avoided by using
.code @(last)
instead of
.code @(until)
since
.code @(last)
consumes the terminating material.
Instead of the above regular-expression-based approach, this extraction problem
can also be solved with
.codn cases :
.IP code:
.mono
\ @(coll)@(cases)@a @(or)@a@(end)@(end)
.onom
.IP data:
.mono
\ 1 2 3 4 5
.onom
.IP result:
.mono
\ a[0]="1"
a[1]="2"
a[2]="3"
a[3]="4"
a[4]="5"
.onom
.PP
.coNP Keyword parameters in @ coll
The
.code @(coll)
directive takes most of the same parameters as
.codn @(collect) .
See the section Keyword parameters in
.code collect
above.
So for instance
.code "@(coll :gap 0)"
means that the collects must be
consecutive, and
.code "@(coll :maxtimes 2)"
means that at most two matches
will be collected. The
.code :lines
keyword does not exist, but there is
an analogous
.code :chars
keyword.
The
.code @(coll)
directive takes the
.code :vars
keyword.
The shorthand
.code @(rep)
may be used instead of
.codn "@(coll :vars nil)" .
.code @(rep)
takes all keywords, except
.codn :vars .
.dir flatten
The
.code flatten
directive can be used to convert variables to one-dimensional
lists. Variables which have a scalar value are converted to lists containing
that value. Variables which are multidimensional lists are flattened to
one-dimensional lists.
Example (without
.codn @(flatten) ):
.IP code:
.mono
\ @b
@(collect)
@(collect)
@a
@(end)
@(end)
.onom
.IP data:
.mono
\ 0
1
2
3
4
5
.onom
.IP result:
.mono
\ b="0"
a_0[0]="1"
a_1[0]="2"
a_2[0]="3"
a_3[0]="4"
a_4[0]="5"
.onom
.PP
Example (with
.codn @(flatten) ):
.IP code:
.mono
\ @b
@(collect)
@(collect)
@a
@(end)
@(end)
@(flatten a b)
.onom
.IP data:
.mono
\ 0
1
2
3
4
5
.onom
.IP result:
.mono
\ b="0"
a[0]="1"
a[1]="2"
a[2]="3"
a[3]="4"
a[4]="5"
.onom
.PP
.dir merge
The syntax of
.code merge
follows the pattern:
.mono
.meti @(merge < destination >> [ sources ...])
.onom
.meta destination
is a variable, which receives a new binding.
.meta sources
are bind expressions.
The
.code merge
directive provides a way of combining collected data from multiple
nested lists in a way which normalizes different nesting levels
among the sources. This directive is useful for combining the results from
collects at different levels of nesting into a single nested list such that
parallel elements are at equal depth.
A new binding is created for the
.meta destination
variable, which holds the result of the operation.
The
.code merge
directive performs its special function if invoked with at least three
arguments: a destination and two sources.
The one-argument case
.code "@(merge x)"
binds a new variable
.code x
and initializes it with the empty list and is thus equivalent to
.codn "@(bind x)" .
Likewise, the two-argument case
.code "@(merge x y)"
is equivalent to
.codn "@(bind x y)" ,
establishing a binding for
.code x
which is initialized with the value of
.codn y .
To understand what merge does when two sources are given, as in
.codn "@(merge C A B)" ,
we first have
to define a property called depth. The depth of an atom such as a string is
defined as
.codn 1 .
The depth of an empty
list is
.codn 0 .
The depth of a nonempty list is one plus the depth of its deepest
element. So for instance
.str foo
has depth 1,
.mono
("foo")
.onom
has depth 2, and
.mono
("foo" ("bar"))
.onom
has depth three.
We can now define a binary (two argument) merge(A, B) function as follows.
First, merge(A, B) normalizes the values A and B
to produce a pair of values which have equal depth, as defined above.
If either value is an atom it is first converted
to a one-element list containing that atom. After this step, both
values are lists; and the only way an argument has depth zero is if it is an
empty list. Next, if either value has a smaller depth than the
other, it is wrapped in a list as many times as needed to give it equal depth.
For instance if A is
.code ("a")
and B is
.code "((((\(dqb\(dq \(dqc\(dq) (\(dqd\(dq \(dqe))))"
then A is converted to
.codn "((((\(dqa\(dq))))" .
Finally, the list values are appended together to produce the merged result.
In the case of the preceding two example values, the result is:
.codn "((((\(dqa\(dq))) (((\(dqb\(dq \(dqc\(dq) (\(dqd\(dq \(dqe))))" .
The result is stored into a the newly bound destination variable
.codn C .
If more than two source arguments are given, these are merged by a left-associative
reduction, which is to say that a three argument
.code "merge(X, Y, Z)"
is defined as
.codn "merge(merge(X, Y), Z)" .
The leftmost two values are merged, and then this result is merged with the third
value, and so on.
.dir cat
The
.code cat
directive converts a list variable into a single
piece of text. The syntax is:
.mono
.mets @(cat < var <> [ sep ])
.onom
The
.meta sep
argument is a Lisp expression whose value specifies a separating piece of text.
If it is omitted, then a single space is used as the separator.
Example:
.IP code:
.mono
\ @(coll)@{a /[^ ]+/}@(end)
@(cat a ":")
.onom
.IP data:
.mono
\ 1 2 3 4 5
.onom
.IP result:
.mono
\ a="1:2:3:4:5"
.onom
.PP
.dir bind
The syntax of the
.code bind
directive is:
.mono
.mets @(bind < pattern < bind-expression >> { keyword << value }*)
.onom
The
.code bind
directive is a kind of pattern match, which matches one or more
variables given in
.meta pattern
against a value produced by the
.meta bind-expression
on the right.
Variable names occurring in the
.meta pattern
expression may refer to bound or unbound variables.
All variable references occurring in
.meta bind-expression
must have a value.
Binding occurs as follows. The tree structure of
.meta pattern
and the value of
.meta bind-expression
are considered to be parallel structures.
Any variables in
.meta pattern
which are unbound receive a new binding, which is initialized with
the structurally corresponding piece of the object produced by
.metn bind-expression .
Any variables in
.meta pattern
which are already bound must match the corresponding part of the
value of
.metn bind-expression ,
or else
the
.code bind
directive fails. Variables which are already bound are not altered,
retaining their current values even if the matching is inexact.
The simplest
.code bind
is of one variable against itself, for instance binding
.code A
against
.codn A :
.verb
@(bind A A)
.brev
This will throw an exception if
.code A
is not bound. If
.code A
is bound, it
succeeds, since
.code A
matches itself.
The next simplest
.code bind
binds one variable to another:
.verb
@(bind A B)
.brev
Here, if
.code A
is unbound, it takes on the same value as
.codn B .
If
.code A
is bound, it has
to match
.codn B ,
or the
.code bind
fails. Matching means that either
.IP -
.code A
and
. code B
are the same text
.IP -
.code A
is text,
.code B
is a list, and
.code A
occurs within
.codn B .
.IP -
vice versa:
.code B
is text,
.code A
is a list, and
.code B
occurs within
.codn A .
.IP -
.code A
and
.code B
are lists and are either identical, or one is
found as a substructure within the other.
.PP
The right-hand side does not have to be a variable. It may be some other
object, like a string, quasiliteral, regexp, or list of strings, etc.
For instance,
.verb
@(bind A "ab\etc")
.brev
will bind the string
.str ab\etc
to the variable
.code A
if
.code A
is unbound. If
.code A
is bound, this will fail unless
.code A
already contains an identical string. However, the right-hand side of a
.code bind
cannot be an unbound variable, nor a complex expression that contains unbound
variables.
The left-hand side of
.code bind
can be a nested list pattern containing variables.
The last item of a list at any nesting level can be preceded by a
.code .
(dot), which means that the variable matches the rest of the list from that
position.
.TP* "Example 1:"
Suppose that the list A contains
.mono
("now" "now" "brown" "cow").
.onom
Then the
directive
.codn "@(bind (H N . C) A)" ,
assuming that
.codn H ,
.code N
and
.code C
are unbound variables,
will bind
.code H
to
.strn how ,
code N
to
.strn now ,
and
.code C
to the remainder of the list
.mono
("brown" "cow").
.onom
Example: suppose that the list
.code A
is nested to two dimensions and contains
.mono
(("how" "now") ("brown" "cow")).
.onom
Then
.code "@(bind ((H N) (B C)) A)"
binds
.code H
to
.strn how ,
.code N
to
.strn now ,
.code B
to
.str brown
and
.code C
to
.strn cow .
The dot notation may be used at any nesting level. it must be followed
by an item. The forms
.code (.)
and
.code "(X .)"
are invalid, but
.code "(. X)"
is valid and equivalent to
.codn X .
The number of items in a left pattern match must match the number of items in
the corresponding right side object. So the pattern
.code ()
only matches
an empty list. The notations
.code ()
and
.code nil
mean exactly the same thing.
The symbols
.codn nil ,
.code t
and keyword symbols may be used on either side.
They represent themselves. For example
.code "@(bind :foo :bar)"
fails,
but
.code "@(bind :foo :foo)"
succeeds since the two sides denote the same
keyword symbol object.
.TP* "Example 2:"
In this example, suppose
.code A
contains
.str foo
and
.code B
contains
bar. Then
.code "@(bind (X (Y Z)) (A (B \(dqhey\(dq)))"
binds
.code X
to
.strn foo ,
.code Y
to
.str bar
and
.code Z
to
.strn hey .
This is because the
.meta bind-expression
produces the object
.mono
("foo" ("bar" "hey"))
.onom
which is then structurally matched against the pattern
.codn "(X (Y Z))" ,
and the variables receive the corresponding pieces.
.coNP Keywords in the @ bind directive
The
.code bind
directive accepts these keywords:
.coIP :lfilt
The argument to
.code :lfilt
is a filter specification. When the left side pattern
contains a binding which is therefore matched against its counterpart from the
right side expression, the left side is filtered through the filter specified
by
.code :lfilt
for the purposes of the comparison. For example:
.verb
@(bind "a" "A" :lfilt :upcase)
.brev
produces a match, since the left side is the same as the right after
filtering through the :upcase filter.
.coIP :rfilt
The argument to
.code :rfilt
is a filter specification. The specified filter is
applied to the right-hand-side material prior to matching it against
the left side. The filter is not applied if the left side is a variable
with no binding. It is only applied to determine a match. Binding takes
place the unmodified right-hand-side object.
For example, the following produces a match:
.verb
@(bind "A" "a" :rfilt :upcase)
.brev
.coIP :filter
This keyword is a shorthand to specify both filters to the same value.
For instance
.code ":filter :upcase"
is equivalent to
.codn ":lfilt :upcase :rfilt :upcase" .
For a description of filters, see Output Filtering below.
Compound filters like
.code "(:fromhtml :upcase)"
are supported with all these keywords. The filters apply across arbitrary
patterns and nested data.
Example:
.verb
@(bind (a b c) ("A" "B" "C"))
@(bind (a b c) (("z" "a") "b" "c") :rfilt :upcase)
.brev
Here, the first bind establishes the values for
.codn a ,
.code b
and
.codn c ,
and the second bind
succeeds, because the value of a matches the second element of the list
.mono
("z" "a")
.onom
if it is upcased, and likewise
.code b
matches
.str b
and
.code c
matches
.str c
if these are upcased.
.coNP Lisp forms in the @ bind directive
\*(TL forms, introduced by
.code @
may be used in the
.meta bind-expression
argument of
.codn bind ,
or as the entire form. This is consistent with the rules for bind expressions.
\*(TL forms can be used in the
.meta pattern
expression also.
Example:
.verb
@(bind a @(+ 2 2))
@(bind @(+ 2 2) @(* 2 2))
.brev
Here,
.code a
is bound to the integer
.codn 4 .
The second
.code bind
then succeeds because the forms
.code "(+ 2 2)"
and
.code "(* 2 2)"
produce equal values.
.dir set
The syntax of the
.code set
directive is:
.mono
.mets @(set < pattern << bind-expression )
.onom
The
.code set
directive syntactically resembles
.codn bind ,
but is not a pattern match. It overwrites
the previous values of variables with new values from the right-hand side.
Each variable that is assigned must have an existing binding:
.code set
will not induce binding.
Examples follow.
Store the value of
.code A
back into
.codn A ,
an operation with no effect:
.verb
@(set A A)
.brev
Exchange the values of
.code A
and
.codn B :
.verb
@(set (A B) (B A))
.brev
Store a string into
.codn A :
.verb
@(set A "text")
.brev
Store a list into
.codn A :
.verb
@(set A ("line1" "line2"))
.brev
Destructuring assignment.
.code A
ends up with
.strn A ,
.code B
ends up with
.mono
("B1" "B2")
.onom
and
.code C
binds to
.mono
("C1" "C2").
.onom
.verb
@(bind D ("A" ("B1" "B2") "C1" "C2"))
@(bind (A B C) (() () ()))
@(set (A B . C) D)
.brev
Note that
.code set
does not support a \*(TL expression on the left side, so the following
are invalid syntax:
.verb
@(set @(+ 1 1) @(* 2 2))
@(set @b @(list "a"))
.brev
The second one is erroneous even though there is a variable on the left.
Because it is preceded by the
.code @
escape, it is a Lisp variable, and not a pattern variable.
The
.code set
directive also doesn't support Lisp expressions in the
.metn pattern ,
which must consist only of variables.
.dir rebind
The syntax of the
.code rebind
directive is:
.mono
.mets @(rebind < pattern << bind-expression )
.onom
The
.code rebind
directive resembles
.codn bind .
It combines the semantics of
.code local
and
.code bind
into a single directive.
The
.meta bind-expression
is evaluated in the current
environment, and its value remembered. Then a new
environment is produced in which all the variables specified in
.meta pattern
are absent. Then, the pattern is newly bound in
that environment against the previously produced value, as if using
.codn bind .
The old environment with the previous variables is not modified;
it continues to exist. This is in contrast with the
.code set
directive, which mutates existing bindings.
.code rebind
makes it easy to create temporary bindings based on existing bindings.
.verb
@(define pattern-function (arg))
@;; inside a pattern function:
@(rebind recursion-level @(+ recursion-level 1))
@;; ...
@(end)
.brev
When the function terminates, the previous value of recursion-level
is restored. The effect is less verbose and more efficient than
the following equivalent
.verb
@(define pattern-function (arg))
@;; inside a pattern function:
@(local temp)
@(set temp recursion-level)
@(local recursion-level)
@(set recursion-level @(+ temp 1))
@;; ...
@(end)
.brev
Like
.codn bind ,
.code rebind
supports nested patterns, such as
.verb
@(rebind (a (b c)) (1 (2 3))
.brev
but it does not support any keyword arguments. The filtering
features of
.code bind
do not make sense in
.code rebind
because the variables are always reintroduced into an environment
in which they don't exist, whereas filtering applies in
situations when bound variables are matched against values.
The
.code rebind
directive also doesn't support Lisp expressions in the
.metn pattern ,
which must consist only of variables.
.dir forget
The
.code forget
has two spellings:
.code @(forget)
and
.codn @(local) .
The arguments are one or more symbols, for example:
.verb
@(forget a)
@(local a b c)
.brev
this can be written
.verb
@(local a)
@(local a b c)
.brev
Directives which follow the
.code forget
or
.code local
directive no longer see
any bindings for the symbols mentioned in that directive, and
can establish new bindings.
It is not an error if the bindings do not exist.
It is strongly recommended to use the
.code @(local)
spelling in functions,
because the forgetting action simulates local variables:
for the given symbols, the machine forgets any earlier variables
from outside of the function, and consequently, any new bindings
for those variables belong to the function. (Furthermore,
functions suppress the propagation of variables that are not
in their parameter list, so these locals will be automatically
forgotten when the function terminates.)
.dir do
The syntax of
.code @(do)
is:
.mono
.mets @(do << lisp-expression *)
.onom
The
.code do
directive evaluates zero or more \*(TL expressions. (See TXR LISP far
below.) The value of the expression is ignored, and matching
continues with the directives which follow the
.code do
directive, if any.
In the context of the
.code do
directive, the expression should not be introduced by the
.code @
symbol; it is expected to be a Lisp expression.
Example:
.verb
@; match text into variables a and b, then insert into hash table h
@(bind h @(hash))
@a:@b
@(do (set [h a] b))
.brev
.dir mdo
The syntax of
.code @(mdo)
is:
.mono
.mets @(mdo << lisp-expression *)
.onom
Like the
.code do
directive,
.code mdo
(macro-time
.codn do )
evaluates zero or more \*(TL expressions. Unlike
.codn do ,
.code mdo
performs this evaluation immediately upon being parsed.
Then it disappears from the syntax.
The effect of
.code "@(mdo e0 e1 e2 ...)"
is exactly like
.code "@(do (macro-time e0 e1 e2 ...))"
except that
.code do
doesn't disappear from the syntax.
Another difference is that
.code do
can be used as a horizontal or vertical directive, whereas
.code mdo
is only vertical.
.dir in-package
The
.code in-package
directive shares the same syntax and semantics as the \*(TL macro
of the same name:
.mono
.mets (in-package << name )
.onom
The
.code in-package
directive is evaluated immediately upon being parsed,
leaving no trace in the syntax tree of the surrounding \*(TX
query.
It causes the
.code *package*
special variable to take on the package denoted by
.metn name .
The directive that
.meta name
is either a string or symbol. An error exception is thrown if
this isn't the case. Otherwise it searches for the package.
If the package is not found, an error exception is thrown.
.SS* Blocks
.NP* Overview
Blocks are sections of a query which are either denoted by a name,
or are anonymous. They may nest: blocks can occur within blocks
and other constructs.
Blocks are useful for terminating parts of a pattern-matching search
prematurely, and escaping to a higher level. This makes blocks not only
useful for simplifying the semantics of certain pattern matches,
but also an optimization tool.
Judicious use of blocks and escapes can reduce or eliminate the amount of
backtracking that \*(TX performs.
.dir block
The
.mono
.meti @(block << name )
.onom
directive introduces a named block, except when
.meta name
is the symbol
.codn nil .
The
.code @(block)
directive introduces an unnamed block, equivalent
to
.codn "@(block nil)" .
The
.code @(skip)
and
.code @(collect)
directives introduce implicit anonymous blocks,
as do function bodies.
Blocks must be terminated by
.code "@(end)"
and can be vertical:
.mono
.mets @(block <> [ name ])
...
.mets @(end)
.onom
or horizontal:
.mono
.mets @(block <> [ name ])...@(end)
.onom
.NP* Block Scope
The names of blocks are in a distinct namespace from the variable binding
space. So
.code "@(block foo)"
is unrelated to the variable
.codn @foo .
A block extends from the
.code "@(block ...)"
directive which introduces it,
until the matching
.codn @(end) ,
and may be empty. For instance:
.verb
@(some)
abc
@(block foo)
xyz
@(end)
@(end)
.brev
Here, the block foo occurs in a
.code @(some)
clause, and so it extends to the
.code @(end)
which terminates the block. After that
.codn @(end) ,
the name foo is not
associated with a block (is not "in scope"). The second
.code @(end)
terminates
the
.code @(some)
block.
The implicit anonymous block introduced by
.code @(skip)
has the same scope
as the
.codn @(skip) :
it extends over all of the material which follows the skip,
to the end of the containing subquery.
.NP* Block Nesting
Blocks may nest, and nested blocks may have the same names as blocks in
which they are nested. For instance:
.verb
@(block)
@(block)
...
@(end)
@(end)
.brev
is a nesting of two anonymous blocks, and
.verb
@(block foo)
@(block foo)
@(end)
@(end)
.brev
is a nesting of two named blocks which happen to have the same name.
When a nested block has the same name as an outer block, it creates
a block scope in which the outer block is "shadowed"; that is to say,
directives which refer to that block name within the nested block refer to the
inner block, and not to the outer one.
.NP* Block Semantics
A block normally does nothing. The query material in the block is evaluated
normally. However, a block serves as a termination point for
.code @(fail)
and
.code @(accept)
directives which are in scope of that block and refer to it.
The precise meaning of these directives is:
.meIP @(fail << name )
Immediately terminate the enclosing query block called
.metn name ,
as if that block
failed to match anything. If more than one block by that name encloses
the directive, the innermost block is terminated. No bindings emerge from
a failed block.
.coIP @(fail)
Immediately terminate the innermost enclosing anonymous block, as if
that block failed to match.
The
.code @(fail)
directive has a vertical and horizontal form.
If the implicit block introduced by
.code @(skip)
is terminated in this manner,
this has the effect of causing
.code skip
itself to fail. In other words, the behavior
is as if
.codn @(skip) 's
search did not find a match for the trailing material,
except that it takes place prematurely (before the end of the available
data source is reached).
If the implicit block associated with a
.code @(collect)
is terminated this way,
then the entire
.code collect
fails. This is a special behavior, because a
.code collect
normally does not fail, even if it matches nothing and collects nothing!
To prematurely terminate a
.code collect
by means of its anonymous block, without failing it, use
.codn @(accept) .
.meIP @(accept << name )
Immediately terminate the enclosing query block called
.metn name ,
as if that block
successfully matched. If more than one block by that name encloses the
directive, the innermost block is terminated.
.coIP @(accept)
Immediately terminate the innermost enclosing anonymous block, as if
that block successfully matched.
.code @(accept)
communicates the current bindings and input position to the terminated
block. These bindings and current position may be altered by special
interactions between certain directives and
.codn @(accept) ,
described in the following section. Communicating the current bindings
and input position means that the block which is terminated by
.code @(accept)
exhibits the bindings which were collected just prior to the execution
of that
.code @(accept)
and the input position which was in effect at that time.
.code @(accept)
has a vertical and horizontal form. In the horizontal form,
it communicates a horizontal input position. A horizontal input
position thus communicated will only take effect if the block being
terminated had been suspended on the same line of input.
If the implicit block introduced by
.code @(skip)
is terminated by
.codn @(accept) ,
this has the effect of causing the skip itself to succeed, as if
all of the trailing material had successfully matched.
If the implicit block associated with a
.code @(collect)
is terminated by
.codn @(accept) ,
then the collection stops. All bindings collected in the current iteration of
the collect are discarded. Bindings collected in previous iterations are
retained, and collated into lists in accordance with the semantics of collect.
Example: alternative way to achieve
.code @(until)
termination:
.verb
@(collect)
@ (maybe)
---
@ (accept)
@ (end)
@LINE
@(end)
.brev
This query will collect entire lines into a list called
.codn LINE .
However,
if the line
.code ---
is matched (by the embedded
.codn @(maybe) ),
the collection
is terminated. Only the lines up to, and not including the
.code ---
line, are collected. The effect is identical to:
.verb
@(collect)
@LINE
@(until)
---
@(end)
.brev
The difference (not relevant in these examples) is that the until clause has
visibility into the bindings set up by the main clause.
However, the following example has a different meaning:
.verb
@(collect)
@LINE
@ (maybe)
---
@ (accept)
@ (end)
@(end)
.brev
Now, lines are collected until the end of the data source, or until a line is
found which is followed by a
.code ---
line. If such a line is found,
the collection stops, and that line is not included in the collection!
The
.code @(accept)
terminates the process of the collect body, and so the
action of collecting the last
.code @LINE
binding into the list is not performed.
.PP
Example: communication of bindings and input position:
.IP code:
.mono
\ @(some)
@(block foo)
@first
@(accept foo)
@ignored
@(end)
@second
.onom
.IP data:
.mono
\ 1
2
3
.onom
.IP result:
.mono
\ first="1"
second="2"
.onom
.PP
At the point where the
.code accept
occurs, the foo block has matched the first line,
bound the text
.str 1
to the variable
.codn @first .
The block is then terminated.
Not only does the
.code @first
binding emerge from this terminated block, but
what also emerges is that the block advanced the data past the first line to
the second line. Next, the
.code @(some)
directive ends, and propagates the
bindings and position. Thus the
.code @second
which follows then matches the second
line and takes the text
.strn 2 .
Example: abandonment of
.code @(some)
clause by
.codn @(accept) :
In the following query, the foo block occurs inside a maybe clause.
Inside the foo block there is a
.code @(some)
clause. Its first subclause
matches variable
.code @first
and then terminates block foo. Since block foo is
outside of the
.code @(some)
directive, this has the effect of terminating the
.code @(some)
clause:
.IP code:
.mono
\ @(maybe)
@(block foo)
@ (some)
@first
@ (accept foo)
@ (or)
@one
@two
@three
@four
@ (end)
@(end)
@second
.onom
.IP data:
.mono
\ 1
2
3
4
5
.onom
.IP result:
.mono
\ first="1"
second="2"
.onom
.PP
The second clause of the
.code @(some)
directive, namely:
.verb
@one
@two
@three
@four
.brev
is never processed. The reason is that subclauses are processed in top
to bottom order, but the processing was aborted within the
first clause the
.codn "@(accept foo)" .
The
.code @(some)
construct never gets the
opportunity to match four lines.
If the
.code "@(accept foo)"
line is removed from the above query, the output
is different:
.IP code:
.mono
\ @(maybe)
@(block foo)
@ (some)
@first
@# <-- @(accept foo) removed from here!!!
@ (or)
@one
@two
@three
@four
@ (end)
@(end)
@second
.onom
.IP data:
.mono
\ 1
2
3
4
5
.onom
.IP result:
.mono
\ first="1"
one="1"
two="2"
three="3"
four="4"
second="5"
.onom
.PP
Now, all clauses of the
.code @(some)
directive have the opportunity to match.
The second clause grabs four lines, which is the longest match.
And so, the next line of input available for matching is
.codn 5 ,
which goes
to the
.code @second
variable.
.coNP Interaction Between the @ trailer and @ accept Directives
If one of the clauses which follow a
.code @(trailer)
requests a successful
termination to an outer block via
.codn @(accept) ,
then
.code @(trailer)
intercepts the escape and adjusts the data extent to the position
that it was given.
Example:
.IP code:
.mono
\ @(block)
@(trailer)
@line1
@line2
@(accept)
@(end)
@line3
.onom
.IP data:
.mono
\ 1
2
3
.onom
.IP result:
.mono
\ line1="1"
line2="2"
line3="1"
.onom
.PP
The variable
.code line3
is bound to
.str 1
because although
.code @(accept)
yields a data
position which has advanced to the third line, this is intercepted by
.code @(trailer)
and adjusted back to the first line. Neglecting to do this adjustment
would violate the semantics of
.codn trailer .
.coNP Interaction Between the @ next and @ accept Directives
When the clauses under a
.code next
directive are terminated by an
.codn accept ,
such that control passes to a block which surrounds that
.codn next ,
the
.code accept
is intercepted by
.codn next .
The input position being communicated by the
.code accept
is replaced with the original input position in the original
stream which is in effect prior to the
.code next
directive. The
.code accept
transfer is then resumed.
In other words,
.code accept
cannot be used to "leak" the new stream out of a
.code next
scope.
However,
.code next
has no effect on the bindings being communicated.
Example:
.mono
\ @(next "file-x")
@(block b)
@(next "file-y")
@line
@(accept b)
@(end)
.onom
Here, the variable
.code line
matches the first line of the file
.strn file-y ,
after which an
.code accept
transfer is initiated, targeting block
.codn b .
This transfer communicates the
.code line
binding, as well as the position within
.codn file-y ,
pointing at the second line.
However, the
.code accept
traverses the
.code next
directive, causing it to be abandoned. The special unwinding
action within that directive detects this transfer and rewrites
the input position to be the original one within the stream
associated with
.strn file-x .
Note that this special handling exists in order for the behavior to be
consistent with what would happen if the
.code "@(accept b)"
were removed, and the block
.code b
terminated normally: because the inner
.code next
is nested within that block, \*(TX would backtrack to the
previous input position within
.strn file-x .
.coNP Interaction Between Functions and the @ accept directive
If a pattern function is terminated due to
.codn accept ,
the function return mechanism intercepts the
.codn accept .
The bindings being communicated by that
.code accept
are then subject to the special resolution with respect to the
function parameters, exactly as if the bindings were being
returned normally out of the function. The resolved bindings
then replace those being communicated by the
.code accept
and the
.code accept
transfer is resumed.
Example:
.mono
\ @(define fun (a))
@ (bind a "a")
@ (bind b "b")
@ (accept blk)
@(end)
@(block blk)
@(fun x)
this line is skipped by accept
@(end)
.onom
Here, the
.code accept
initiates a control transfer which communicates the
.code a
and
.code b
variable bindings which are visible in that scope. This transfer is
intercepted by the function, and the treatment of the bindings follows
to the same rules as a normal return (which, in the given function, would
readily take place if the
.code accept
directive were removed). The
.code b
variable is suppressed, because
.code b
isn't a parameter of the function. Because
.code a
.B is
a parameter, and the argument to that parameter is the unbound
variable
.codn x ,
the effect is that
.code x
is bound to the value of
.codn a .
When the accept transfer reaches block
.code blk
and terminates it, all that emerges is the
.code x
binding carrying
.strn a .
If the
.code accept
invocation is removed from
.codn fun ,
then the function returns normally, producing the
.code x
binding. In that case, the line
.code "this line is skipped by accept"
isn't skipped since the block isn't being terminated; that
line must match something.
.coNP Interaction Between @ finally and the @ accept directive
If the exception handling
.code try
directive protected body is terminated by an
.code accept
transfer, and if that
.code try
has a
.code finally
block, then there is a special interaction between the
.code finally
block and the
.code accept
transfer.
The processing of the
.code finally
block detects that it has been triggered by an
.code accept
transfer. Consequently, it retrieves the current input position
and bindings from that transfer, and uses that position and those
bindings for the processing of the
.code finally
clauses.
If the
.code finally
clauses succeed, then the new input position and new bindings
are installed into the
.code accept
control transfer and that transfer resumes.
If the
.code finally
clauses fail, then the
.code accept
transfer is converted to a
.codn fail ,
with exactly the same block as its destination.
.coNP Vertical-Horizontal Mismatch Between @ block and @ accept
The
.codn block ,
.code accept
and
.code fail
directives comes in horizontal and vertical forms.
This creates the possibility that an
.code accept
in horizontal context targets a vertical
.code block
or vice versa,
raising the question of how the input position
is treated. The semantics of this is defined.
If a horizontal-context
.code accept
targets a vertical block, the current position at the target block will be the
following line. That is to say, when the horizontal
.code accept
occurs, there is a current input line which may have unconsumed
material past the current position. If the
.code accept
communicates its input position to a vertical context, that unconsumed
material is skipped, as if it had been matched and the vertical position
is advanced to the next line.
If a horizontal block catches a vertical accept, it rejects that
.codn accept 's
position and stays at the current backtracking position for that block.
Only the bindings from the
.code accept
are retained.
.coNP Horizontal-Horizontal Mismatch between @ block and @ accept
It is possible for a horizontal
.code accept
to terminate in a horizontal block which is processing
a different line of input (or even a different input stream).
This situation is treated the same way as vertical accept terminating
in a horizontal block: the position communicated by
.code accept
is ignored, and only the bindings are taken.
.SS* Functions
.NP* Overview
\*(TX functions allow a query to be structured to avoid repetition.
On a theoretical note, because
\*(TX functions support recursion, functions enable \*(TX to match some
kinds of patterns which exhibit self-embedding, or nesting,
and thus cannot be matched by a regular language.
Functions in \*(TX are not exactly like functions in mathematics or functional
languages, and are not like procedures in imperative programming languages.
They are not exactly like macros either. What it means for a
\*(TX function to take arguments and produce a result is different from
the conventional notion of a function.
A \*(TX function may have one or more parameters. When such a function is
invoked, an argument must be specified for each parameter. However, a special
behavior is at play here. Namely, some or all of the argument expressions may
be unbound variables. In that case, the corresponding parameters behave like
unbound variables also. Thus \*(TX function calls can transmit the "unbound"
state from argument to parameter.
It should be mentioned that functions have access to all bindings that are
visible in the caller; functions may refer to variables which are not
mentioned in their parameter list.
With regard to returning, \*(TX functions are also unconventional. If the
function fails, then the function call is considered to have failed. The
function call behaves like a kind of match; if the function fails, then the
call is like a failed match.
When a function call succeeds, then the bindings emanating from that function
are processed specially. Firstly, any bindings for variables which do not
correspond to one of the function's parameters are thrown away. Functions may
internally bind arbitrary variables in order to get their job done, but only
those variables which are named in the function argument list may propagate out
of the function call. Thus, a function with no arguments can only indicate
matching success or failure, but not produce any bindings. Secondly,
variables do not propagate out of the function directly, but undergo
a renaming. For each parameter which went into the function as an unbound
variable (because its corresponding argument was an unbound variable),
if that parameter now has a value, that value is bound onto the corresponding
argument.
Example:
.verb
@(define collect-words (list))
@(coll)@{list /[^ \et]+/}@(end)
@(end)
.brev
The above function
.code collect-words
contains a query which collects words from a
line (sequences of characters other than space or tab), into the list variable
called
.codn list .
This variable is named in the parameter list of the function,
therefore, its value, if it has one, is permitted to escape from the function
call.
Suppose the input data is:
.verb
Fine summer day
.brev
and the function is called like this:
.verb
@(collect-words wordlist)
.brev
The result (with
.codn "txr -B" )
is:
.verb
wordlist[0]=Fine
wordlist[1]=summer
wordlist[1]=day
.brev
How it works is that in the function call
.codn "@(collect-words wordlist)" ,
.code wordlist
is an unbound variable. The parameter corresponding to that
unbound variable is the parameter
.codn list .
Therefore, that parameter
is unbound over the body of the function. The function body collects the
words of
.str Fine summer day
into the variable
.codn list ,
and then
yields the that binding. Then the function call completes by
noticing that the function parameter
.code list
now has a binding, and
that the corresponding argument
.code wordlist
has no binding. The binding
is thus transferred to the
.code wordlist
variable. After that, the
bindings produced by the function are thrown away. The only enduring
effects are:
.IP -
the function matched and consumed some input; and
.IP -
the function succeeded; and
.IP -
the
.code wordlist
variable now has a binding.
.PP
Another way to understand the parameter behavior is that function
parameters behave like proxies which represent their arguments. If an argument
is an established value, such as a character string or bound variable, the
parameter is a proxy for that value and behaves just like that value. If an
argument is an unbound variable, the function parameter acts as a proxy
representing that unbound variable. The effect of binding the proxy is
that the variable becomes bound, an effect which is settled when the
function goes out of scope.
Within the function, both the original variable and the proxy are
visible simultaneously, and are independent. What if a function binds both of
them? Suppose a function has a parameter called
.codn P ,
which is called with an argument
.codn A ,
which is an unbound variable, and then, in the function, both
.code A
and
.code P
bound. This is
permitted, and they can even be bound to different values. However, when the
function terminates, the local binding of A simply disappears (because
the symbol
.code A
is not among the parameters of the function).
Only the value bound to
.code P
emerges, and is bound to
.codn A ,
which still appears unbound at that point. The
.code P
binding disappears also, and the net effect is that
.code A
is now bound. The "proxy" binding of
.code A
through the parameter
.code P
"wins" the conflict with the direct binding.
.NP* Definition Syntax
Function definition syntax comes in two flavors: vertical and horizontal.
Horizontal definitions actually come in two forms, the distinction
between which is hardly noticeable, and the need for which is
made clear below.
A function definition begins with a
.code "@(define ...)"
directive. For vertical
functions, this is the only element in a line.
The
.code define
symbol must be followed by a symbol, which is the name of the
function being defined. After the symbol, there is a parenthesized optional
argument list. If there is no such list, or if the list is specified as
.code ()
or
the symbol
.code nil
then the function has no parameters. Examples of valid
.code define
syntax are:
.verb
@(define foo)
@(define bar ())
@(define match (a b c))
.brev
If the
.code define
directive is followed by more material on the same line, then
it defines a horizontal function:
.verb
@(define match-x)x@(end)
.brev
If the define is the sole element in a line, then it
is a vertical function, and the function definition continues below:
.verb
@(define match-x)
x
@(end)
.brev
The difference between the two is that a horizontal function matches
characters within a line, whereas a vertical function matches lines
within a stream. The former
.code match-x
matches the character
.codn x ,
advancing
to the next character position. The latter
.code match-x
matches a line consisting
of the character
.codn x ,
advancing to the next line.
Material between
.code @(define)
and
.code @(end)
is the function body. The define
directive may be followed directly by the
.code @(end)
directive, in which case the
function has an empty body.
Functions may be nested within function bodies. Such local functions have
dynamic scope. They are visible in the function body in which they are defined,
and in any functions invoked from that body.
The body of a function is an anonymous block. (See Blocks above.)
.NP* Two Forms of The Horizontal Function
If a horizontal function is defined as the only element of a line,
it may not be followed by additional material. The following
construct is erroneous:
.verb
@(define horiz (x))@foo:@bar@(end)lalala
.brev
This kind of definition is actually considered to be in the vertical context,
and like other directives that have special effects and that do not match
anything, it does not consume a line of input. If the above syntax were
allowed, it would mean that the line would not only define a function but also
match
.codn "lalala" .
This would, in turn, would mean that the
.code @(define)...@(end)
is
actually in horizontal mode, and so it matches a span of zero characters within
a line (which means that is would require a line of input to match: a
surprising behavior for a nonmatching directive!)
A horizontal function can be defined in an actual horizontal context. This
occurs if its is in a line where it is preceded by other material.
For instance:
.verb
X@(define fun)...@(end)Y
.brev
This is a query line which must match the text
.codn XY .
It also defines the function
.codn fun .
The main use of this form is for nested horizontal functions:
.verb
@(define fun)@(define local_fun)...@(end)@(end)
.brev
.NP* Vertical-Horizontal Overloading
A function of the same name may be defined as both vertical and horizontal.
Both functions are available at the same time. Which one is used by
a call is resolved by context. See the section Vertical Versus Horizontal Calls
below.
.NP* Call Syntax
A function is invoked by compound directive whose first symbol is the name of
that function. Additional elements in the directive are the arguments.
Arguments may be symbols, or other objects like string and character
literals, quasiliterals ore regular expressions.
Example:
.IP code:
.mono
\ @(define pair (a b))
@a @b
@(end)
@(pair first second)
@(pair "ice" cream)
.onom
.IP data:
.mono
\ one two
ice milk
.onom
.IP result:
.mono
\ first="one"
second="two"
cream="milk"
.onom
.PP
The first call to the function takes the line
.strn "one two" .
The parameter
.code a
takes
.str one
and parameter
.code b
takes
.strn two .
These are rebound to the arguments
.code first
and
.codn second .
The second call to the function binds the a parameter to the word
.strn ice ,
and the
.code b
is unbound, because the
corresponding argument
.code cream
is unbound. Thus inside the function,
.code a
is forced to match
.codn "ice" .
Then a space is matched and
.code b
collects the text
.strn milk .
When the function returns, the unbound
.str cream
variable gets this value.
If a symbol occurs multiple times in the argument list, it constrains
both parameters to bind to the same value. That is to say, all parameters
which, in the body of the function, bind a value, and which are all derived
from the same argument symbol must bind to the same value. This is settled when
the function terminates, not while it is matching. Example:
.IP code:
.mono
\ @(define pair (a b))
@a @b
@(end)
@(pair same same)
.onom
.IP data:
.mono
\ one two
.onom
.IP result:
.mono
\ [query fails]
.onom
.PP
Here the query fails because
.code a
and
.code b
are effectively proxies for the same unbound variable
.code same
and are bound to different values, creating a conflict which
constitutes a match failure.
.NP* Vertical Versus Horizontal Calls
A function call which is the only element of the query line in
which it occurs is ambiguous. It can go either to a vertical
function or to the horizontal one. If both are defined, then
it goes to the vertical one.
Example:
.IP code:
.mono
\ @(define which (x))@(bind x "horizontal")@(end)
@(define which (x))
@(bind x "vertical")
@(end)
@(which fun)
.onom
.IP result:
.mono
\ fun="vertical"
.onom
.PP
Not only does this call go to the vertical function, but
it is in a vertical context.
If only a horizontal function is defined, then that is the one which is called,
even if the call is the only element in the line. This takes place in a
horizontal character-matching context, which requires a line of input which can
be traversed:
Example:
.IP code:
.mono
\ @(define which (x))@(bind x "horizontal")@(end)
@(which fun)
.onom
.IP data:
.mono
\ ABC
.onom
.IP result:
.mono
\ [query fails]
.onom
.PP
The query fails because since
.code "@(which fun)"
is in horizontal mode,
it matches characters in a line. Since the function body consists
only of
.code "@(bind ...)"
which doesn't match any characters, the function
call requires an empty line to match. The line
.code ABC
is not empty,
and so there is a matching failure. The following
example corrects this:
Example:
.IP code:
.mono
\ @(define which (x))@(bind x "horizontal")@(end)
@(which fun)
.onom
.IP data:
.mono
\ [empty line]
.onom
.IP result:
.mono
\ fun="horizontal"
.onom
.PP
A call made in a clearly horizontal context will prefer the
horizontal function, and only fall back on the vertical one
if the horizontal one doesn't exist. (In this fallback case,
the vertical function is called with empty data; it is useful
for calling vertical functions which process arguments and
produce values.)
In the next example, the call is followed by trailing material,
placing it in a horizontal context. Leading material will
do the same thing:
Example:
.IP code:
.mono
\ @(define which (x))@(bind x "horizontal")@(end)
@(define which (x))
@(bind x "vertical")
@(end)
@(which fun)B
.onom
.IP data:
.mono
\ B
.onom
.IP result:
.mono
\ fun="horizontal"
.onom
.PP
.NP* Local Variables
As described earlier, variables bound in a function body which are not
parameters of the function are discarded when the function returns. However,
that, by itself, doesn't make these variables local, because pattern functions
have visibility to all variables in their calling environment. If a variable
.code x
exists already when a function is called, then an attempt to bind it inside a
function may result in a failure. The
.code local
directive must be used in a
pattern function to list which variables are local.
Example:
.verb
@(define path (path))@\e
@(local x y)@\e
@(cases)@\e
(@(path x))@(path y)@(bind path `(@x)@y`)@\e
@(or)@\e
@{x /[.,;'!?][^ \et\ef\ev]/}@(path y)@(bind path `@x@y`)@\e
@(or)@\e
@{x /[^ .,;'!?()\et\ef\ev]/}@(path y)@(bind path `@x@y`)@\e
@(or)@\e
@(bind path "")@\e
@(end)@\e
@(end)
.brev
This is a horizontal function which matches a path, which lands into four
recursive cases. A path can be parenthesized path followed by a path; it can be
a certain character followed by a path, or it can be empty
This function ensures that the variables it uses internally,
.code x
and
.codn y ,
do not have anything to do with any inherited bindings for
.code x
and
.codn y .
Note that the function is recursive, which cannot work without
.code x
and
.code y
being local, even if no such bindings exist prior to the top-level invocation of the
function. The invocation
.code "@(path x)"
causes
.code x
to be bound, which is
visible inside the invocation
.codn "@(path y)" ,
but that invocation needs to have its own binding of
.code x
for local use.
.NP* Nested Functions
Function definitions may appear in a function. Such definitions
are visible in all functions which are invoked from the body
(and not necessarily enclosed in the body). In other words, the
scope is dynamic, not lexical. Inner definitions shadow outer
definitions. This means that a caller can redirect the function
calls that take place in a callee, by defining local functions
which capture the references.
Example:
.IP code:
.mono
\ @(define which)
@ (fun)
@(end)
@(define fun)
@ (output)
top-level fun!
@ (end)
@(end)
@(define callee)
@ (define fun)
@ (output)
local fun!
@ (end)
@ (end)
@ (which)
@(end)
@(callee)
@(which)
.onom
.IP output:
.mono
\ local fun!
top-level fun!
.onom
.PP
Here, the function
.code which
is defined which calls
.codn fun .
A top-level definition of
.code fun
is introduced which
outputs
.strn "top-level fun!" .
The function
.code callee
provides its own local
definition of
.code fun
which outputs
.str "local fun!"
before calling
.codn which .
When
.code callee
is invoked, it calls
.codn which ,
whose
.code @(fun)
call is routed to callee's
local definition. When
.code which
is called directly from the top level, its
.code fun
call goes to the top-level definition.
.NP* Indirect Calls
Function indirection may be performed using the
.code call
directive. If
.meta fun-expr
is an Lisp expression which evaluates to a symbol, and
that symbol names a function which takes no arguments, then
.verb
@(call fun-expr)
.brev
may be used to invoke the function. Additional
expressions may be supplied which specify arguments.
Example 1:
.mono
\ @(define foo (arg))
@(bind arg "abc")
@(end)
@(call 'foo b)
.onom
In this example, the effect is that
.code foo
is invoked, and
.code b
ends up bound to
.strn abc .
The
.code call
directive here uses the
.code 'foo
expression to calculate the name of the function to be invoked.
(See the
.code quote
operator).
This particular
.code call
expression can just be replaced by the direct invocation
syntax
.codn "@(foo b)" .
The power of
.code call
lies in being able to specify the function as a value which
comes from elsewhere in the program, as in the following example.
.mono
\ @(define foo (arg))
@(bind arg "abc")
@(end)
@(bind f @'foo)
@(call f b)
.onom
Here the
.code call
directive obtains the name of the function from the
.code f
variable.
Note that function names are resolved to functions in the environment
that is apparent at the point in execution where the
.code call
takes place. The directive
.code "@(call f args ...)"
is precisely equivalent to
.code "@(s args ...)"
if, at the point of the call,
.code f
is a variable which holds the symbol
.code s
and symbol
.code s
is defined as a function. Otherwise it is erroneous.
.SS* Modularization
.dirs load include
The syntax of the
.code load
and
.code include
directives is:
.mono
.mets @(load << expr )
.mets @(include << expr )
.onom
Where
.meta expr
is a Lisp expression that
evaluates to a string giving the path of the file to load.
Firstly, the path given by
.meta expr
is converted to an effective path, as follows.
If the value of the
.code *load-path*
variable has a current value which is not
.code nil
and the path given in
.meta expr
is pure relative according to the
.code pure-rel-path-p
function, then the effective path is interpreted taken relative
to the directory portion of the path which is stored in
.codn *load-path* .
If
.code *load-path*
is
.codn nil ,
or the load path is not pure relative, then the
path is taken as-is as the effective path.
Next, an attempt is made to open the file for processing, in
almost exactly the same manner as by the \*(TL function
.codn load .
The difference is that if the effective path is unsuffixed,
then the
.code .txr
suffix is added to it, and that resulting path is tried first,
and if it succeeds, then the file is treated as \*(TX Pattern
Language syntax.
If that fails, then the suffix
.code .tlo
is tried, and so forth, as described for the
.code load
function.
Both the
.code load
and
.code include
directives bind the
.code *load-path*
variable to the path of the loaded file just before parsing syntax from it,
The
.code *package*
variable is also given a new dynamic binding, whose value is the
same as the existing binding. These bindings are removed when the
load operation completes, restoring the prior values of these
variables.
The
.code *load-hooks*
variable is given a new dynamic binding, with a
.code nil
value.
If the file opened for processing is \*(TL source, or
a compiled \*(TL file, then it is processed in the manner
described for the
.code load
function.
Different requirements apply to the processing of the file under the
.code load
and
.code include
directives.
The
.code include
directive performs the processing of the file at parse time. If the
file being processed is \*(TX Pattern Language, then it is parsed,
and then its syntax replaces the
.code include
directive, as if it had originally appeared in its place.
If a \*(TL source or a compiled \*(TL file is processed by
.code include
then the
.code include
directive is removed from the syntax.
The
.code load
directive performs the processing of the file at evaluation time.
Evaluation time
occurs after a \*(TX program is read from beginning to end and parsed.
That is to say, when a \*(TX query is parsed, any embedded
.code "@(load ...)"
forms in it are parsed and constitute part of its syntax tree.
They are executed when that query is executed, whenever its execution
reaches those
.code load
directives. When the
.code load
directive processes \*(TX Pattern Language syntax, it parses
the file in its entirety and then executes that file's directives
against the current input position. Repeated executions of the
same
.code load
directive result in repeated processing of the file.
Note: the
.code include
directive is useful for loading \*(TX files which contain Lisp macros
which are needed by the parent program. The parent program cannot use
.code load
to bring in macros because macros are required during expansion, which
takes place prior to evaluation time, whereas
.code load
doesn't execute until evaluation time.
See also: the
.codn self-path ,
.code stdlib
and
.code *load-path*
variables in \*(TL.
.SS* Output
.NP* Introduction
A \*(TX query may perform custom output. Output is performed by
.code output
clauses,
which may be embedded anywhere in the query, or placed at the end. Output
occurs as a side effect of producing a part of a query which contains an
.code @(output)
directive, and is executed even if that part of the query ultimately
fails to find a match. Thus output can be useful for debugging.
An
.code output
clause specifies that its output goes to a file, pipe, or (by
default) standard output. If any output clause is executed whose destination is
standard output, \*(TX makes a note of this, and later, just prior to
termination, suppresses the usual printing of the variable bindings or the word
false.
.dir output
The syntax of the
.code @(output)
directive is:
.mono
.mets @(output [ < destination ] { < bool-keyword | < keyword < value }* )
.
. one or more output directives or lines
.
@(end)
.onom
If the directive has arguments, then the first one is evaluated.
If it is an object other than a keyword symbol, then it specifies
the optional
.metn destination .
Any remaining arguments after the optional destination are
the keyword list. If the destination is missing, then the
entire argument list is a keyword list.
The
.meta destination
argument,
if present,
is treated as a \*(TL expression and evaluated.
The resulting value is taken as the output destination. The value may be a
string which gives the pathname of a file to open for output. Otherwise,
the destination must be a stream object.
The keyword list consists of a mixture of Boolean keywords which
do not have an argument, or keywords with arguments.
The following Boolean keywords are supported:
.coIP :nothrow
The
.code output
directive throws an exception if the output destination
cannot be opened, unless the
.code :nothrow
keyword is present, in which
case the situation is treated as a match failure.
Note that since command pipes are processes that report errors
asynchronously, a failing command will not throw an immediate exception that
can be suppressed with
.codn :nothrow .
This is for synchronous errors, like
trying to open a destination file, but not having permissions, etc.
.coIP :append
This keyword is meaningful for files, specifying append mode: the output is to
be added to the end of the file rather than overwriting the file.
.PP
The following value keywords are supported:
.coIP :filter
The argument can be a symbol, which specifies a filter to be applied to
the variable substitutions occurring within the
.code output
clause.
The argument can also be a list of filter symbols, which specifies
that multiple filters are to be applied, in left-to-right order.
See the later sections Output Filtering below, and The Deffilter Directive.
.coIP :into
The argument of
.code :into
is a symbol which denotes a variable.
The output will go into that variable. If the variable is unbound,
it will be created. Otherwise, its contents are overwritten
unless the
.code :append
keyword is used. If
.code :append
is used, then
the new content will be appended to the previous content of
the variable, after flattening the content to a list,
as if by the
.code flatten
directive.
.coIP :named
The argument of
.code :named
is a symbol which denotes a variable.
The file or pipe stream which is opened for the output is
stored in this variable, and is not closed at the end of the
output block. This allows a subsequent output block to continue
output on the same stream, which is possible using the
next two keywords,
.code :continue
or
.codn :finish .
A new binding is established for the variable, even if it
already has an existing binding.
.coIP :continue
A destination should not be specified if
.code :continue
is used. The argument of
.code :continue
is an expression, such as a variable name, that evaluates to a
stream object. That stream object is used for the output block.
At the end of the output block, the stream is flushed, but not
closed. A usage example is given in the documentation for the Close Directive
below.
.coIP :finish
A destination should not be specified if
.code :finish
is used. The argument of
.code :finish
is an expression, such as a variable name, that evaluates to a
stream object. That stream object is used for the output block.
At the end of the output block, the stream is closed.
An example is given in the documentation for the Close Directive
below.
.NP* Output Text
Text in an output clause is not matched against anything, but is output
verbatim to the destination file, device or command pipe.
.NP* Output Variables
Variables occurring in an output clause do not match anything; instead
their contents are output.
A variable being output can be any object. If it is of a type other
than a list or string, it will be converted to a string as if by the
.code tostring
function in \*(TL.
A list is converted to a string in a special way: the elements are
individually converted to a string and then they are catenated together.
The default separator string is a single space: an alternate separation
can be specified as an argument in the brace substitution syntax.
Empty lists turn into an empty string.
Lists may be output within
.code @(repeat)
or
.code @(rep)
clauses. Each nesting of
these constructs removes one level of nesting from the list variables
that it contains.
In an output clause, the
.mono
.meti >> @{ name << number }
.onom
variable syntax generates fixed-width
field, which contains the variable's text. The absolute value of the
number specifies the field width. For instance
.code -20
and
.code 20
both specify a field
width of twenty. If the text is longer than the field, then it overflows the
field. If the text is shorter than the field, then it is left-adjusted within
that field, if the width is specified as a positive number, and right-adjusted
if the width is specified as negative.
An output variable may specify a filter which overrides any filter established
for the output clause. The syntax for this is
.mono
.meti @{NAME :filter << filterspec }.
.onom
The filter specification syntax is the same as in the output clause.
See Output Filtering below.
.NP* Output Variables: Indexing
Additional syntax is supported in output variables that does not appear
in pattern-matching variables.
A square bracket index notation may be used to extract elements or
ranges from a variable, which works with strings, vectors and lists. Elements
are indexed from zero. This notation is only available in brace-enclosed
syntax, and looks like this:
.meIP <> @{name[ expr ]}
Extract the element at the position given by
.metn expr .
.meIP <> @{name[ expr1..expr2 ]}
Extract a range of elements from the position given by
.metn expr1 ,
up to
one position less than the position given by
.metn expr2 .
If the variable is a list, it is treated as a list substitution,
exactly as if it were the value of an unsubscripted list variable.
The elements of the list are converted to strings and catenated
together with a separator string between them, the default one being
a single space.
An alternate character may be given as a string argument in the brace
notation.
.PP
Example:
.verb
@(bind a ("a" "b" "c" "d"))
@(output)
@{a[1..3] "," 10}
@(end)
.brev
The above produces the text
.str b,c
in a field
.code 10
spaces wide. The
.code [1..3]
argument extracts a range of
.codn a ;
the
.str ,
argument specifies an alternate
separator string, and
.code 10
specifies the field width.
.NP* Output Substitutions
The brace syntax has another syntactic and semantic extension in
.code output
clauses. In place
of the symbol, an expression may appear. The value of that expression
is substituted.
Example:
.mono
@(bind a "foo")
@(output)
@{`@a:` -10}
.onom
Here, the quasiliteral expression
.code `@a:`
is evaluated, producing the string
.strn foo: .
This string is printed right-adjusted in a
.code 10
character field.
.dir repeat
The
.code repeat
directive generates repeated text from a "boilerplate",
by taking successive elements from lists. The syntax of repeat is
like this:
.verb
@(repeat)
.
.
main clause material, required
.
.
special clauses, optional
.
.
@(end)
.brev
.code repeat
has four types of special clauses, any of which may be
specified with empty contents, or omitted entirely. They are described
below.
.code repeat
takes arguments, also described below.
All of the material in the main clause and optional clauses
is examined for the presence of variables. If none of the variables
hold lists which contain at least one item, then no output is performed,
(unless the repeat specifies an
.code @(empty)
clause, see below).
Otherwise, among those variables which contain nonempty lists, repeat finds
the length of the longest list. This length of this list determines the number
of repetitions, R.
If the
.code repeat
contains only a main clause, then the lines of this clause is
output R times. Over the first repetition, all of the variables which, outside
of the repeat, contain lists are locally rebound to just their first item. Over
the second repetition, all of the list variables are bound to their second
item, and so forth. Any variables which hold shorter lists than the longest
list eventually end up with empty values over some repetitions.
Example: if the list
.code A
holds
.strn 1 ,
.str 2
and
.strn 3 ;
the list
.code B
holds
.strn A ,
.strn B ;
and the variable
.code C
holds
.strn X ,
then
.verb
@(repeat)
>> @C
>> @A @B
@(end)
.brev
will produce three repetitions (since there are two lists, the longest
of which has three items). The output is:
.verb
>> X
>> 1 A
>> X
>> 2 B
>> X
>> 3
.brev
The last line has a trailing space, since it is produced by
.strn "@A @B" ,
where
.code B
has an empty value. Since
.code C
is not a list variable, it
produces the same value in each repetition.
The special clauses are:
.coIP @(single)
If the
.code repeat
produces exactly one repetition, then the contents of this clause
are processed for that one and only repetition, instead of the main clause
or any other clause which would otherwise be processed.
.coIP @(first)
The body of this clause specifies an alternative body to be used for the first
repetition, instead of the material from the main clause.
.coIP @(last)
The body of this clause is used instead of the main clause for the last
repetition.
.coIP @(empty)
If the repeat produces no repetitions, then the body of this clause is output.
If this clause is absent or empty, the repeat produces no output.
.coIP "@(mod n m)"
The forms
.code n
and
.code m
are Lisp expressions that evaluate to integers. The value of
.code m
should be nonzero. The clause denoted this way is active if the repetition
modulo
.code m
is equal to
.codn n .
The first repetition is numbered zero.
For instance the clause headed by
.code "@(mod 0 2)"
will be used on repetitions
0, 2, 4, 6, ... and
.code "@(mod 1 2)"
will be used on repetitions 1, 3, 5, 7, ...
.coIP "@(modlast n m)"
The meaning of
.code n
and
.code m
is the same as in
.codn "@(mod n m)" ,
but one more condition
is imposed. This clause is used if the repetition modulo
.code m
is equal to
.codn n ,
and if it is the last repetition.
.PP
The precedence among the clauses which take an iteration is:
.codn "single > first > modlast > last > mod > main" .
That is, whenever two or more of these
clauses can apply to a repetition, then the leftmost one in this precedence
list will be selected. It is possible for all these clauses to be viable
for processing the same repetition. If a
.code repeat
occurs which has only one repetition, then that repetition is simultaneously
the first, only and last repetition. Moreover, it also matches
.code "(mod 0 m)"
and, because it is the last repetition, it matches
.codn "(modlast 0 m)" .
In this situation, if there is a
.code @(single)
clause present, then the repetition shall be processed using that
clause. Otherwise, if there is a
.code @(first)
clause present, that clause is activated. Failing
that,
.code @(modlast)
is used if there is such a clause, featuring an
.code n
argument of zero. If there isn't, then the
.code @(last)
clause is considered, if present. Otherwise, the
.code @(mod)
clause is considered if present with an
.code n
argument of zero. Otherwise, none of these clauses are present or applicable,
and the repetition is processed using the main clause.
The
.code @(empty)
clause does not appear in the above precedence list because it is mutually
exclusive with respect to the others: it is processed only when there are no
iterations, in which case even the main clause isn't active.
The
.code @(repeat)
clause supports arguments.
.mono
.mets @(repeat
.mets \ \ \ [:counter >> { symbol | >> ( symbol << expr )}]
.mets \ \ \ [:vars >> ({ symbol | >> ( symbol << expr )}*)])
.onom
The
.code :counter
argument designates a symbol which will behave as an integer
variable over the scope of the clauses inside the repeat. The variable provides
access to the repetition count, starting at zero, incrementing with each
repetition. If the argument is given as
.mono
.meti >> ( symbol << expr )
.onom
then
.meta expr
is a Lisp expression whose value is taken as a displacement value which
is added to each iteration of the counter. For instance
.code ":counter (c 1)"
specifies a counter
.code c
which counts from 1.
The
.code :vars
argument specifies a list of variable name symbols
.meta symbol
or else pairs of the form
.mono
.meti >> ( symbol << init-form )
.onom
consisting of a variable name and Lisp expression. Historically, the former
syntax informed
.code repeat
about references to variables contained in Lisp code. This usage is no
longer necessary as of \*(TX 243, since the
.code repeat
construct walks Lisp code, identifying all free variables.
The latter syntax introduces a new pattern variable binding for
.meta symbol
over the scope of the
.code repeat
construct. The
.meta init-form
specifies a Lisp expression which is evaluated to produce the
binding's value.
The
.code repeat
directive then processes the list of variables, selecting from it
those which have a binding, either a previously existing binding or the
one just introduced. For each selected variable, repeat
will assume that the variable occurs in the repeat block and contains
a list to be iterated.
The variable binding syntax supported by
.code :vars
of the form
.mono
.meti >> ( symbol << init-form )
.onom
provides a solution for situations when it is necessary to iterate
over some list, but that list is the result of an expression, and not stored in
any variable. A repeat block iterates only over lists emanating from variables;
it does not iterate over lists pulled from arbitrary expressions.
Example: output all file names matching the
.code *.txr
pattern in the current directory:
.verb
@(output)
@(repeat :vars ((name (glob "*.txr"))))
@name
@(end)
@(end)
.brev
Prior to \*(TX 243, the simple variable-binding syntax supported by
.code :vars
of the form
.meta symbol
was needed for situations in which \*(TL expressions which
referenced variables were embedded in
.code @(repeat)
blocks. Variable references embedded in Lisp code were not identified in
.codn @(repeat) .
For instance, the following produced no output, because no variables
were found in the
.code repeat
body:
.verb
@(bind trigraph ("abc" "def" "ghi"))
@(output)
@(repeat)
@(reverse trigraph)
@(end)
@(end)
.brev
There is a reference to
.meta trigraph
but it's inside the
.code "(reverse trigraph)"
Lisp expression that was not processed by
.codn repeat .
The solution was to mention
.meta trigraph
in the
.code :vars
construct:
.verb
@(bind trigraph ("abc" "def" "ghi"))
@(output)
@(repeat :vars (trigraph))
@(reverse trigraph)
@(end)
@(end)
.brev
Then the
.code repeat
block would iterate over
.metn trigraph ,
producing the output
.verb
cba
fed
igh
.brev
This workaround is no longer required as of \*(TX 243; the output
is produced by the first example, without
.codn :vars .
.coNP Nested @ repeat directives
If a
.code repeat
clause encloses variables which hold multidimensional lists,
those lists require additional nesting levels of
.code repeat
(or
.codn rep ).
It is an error to attempt to output a list variable which has not been
decimated into primary elements via a
.code repeat
construct.
Suppose that a variable
.code X
is two-dimensional (contains a list of lists).
.code X
must be nested twice in a
.codn repeat .
The outer
.code repeat
will traverse the lists contained in
.codn X .
The inner
.code repeat
will traverse the elements of each of these lists.
A nested
.code repeat
may be embedded in any of the clauses of a
.codn repeat ,
not only in the main clause.
.dir rep
The
.code rep
directive is similar to
.codn repeat .
Whereas
.code repeat
is line-oriented,
.code rep
generates material within a line. It has all the same clauses,
but everything is specified within one line:
.verb
@(rep)... main material ... .... special clauses ...@(end)
.brev
More than one
.code @(rep)
can occur within a line, mixed with other material.
A
.code @(rep)
can be nested within a
.code @(repeat)
or within another
.codn @(rep) .
Also,
.code @(rep)
accepts the same
.code :counter
and
.code :vars
arguments.
.coNP @ repeat and @ rep Examples
Example 1: show the list
.code L
in parentheses, with spaces between
the elements, or the word
.code EMPTY
if the list is empty:
.verb
@(output)
@(rep)@L @(single)(@L)@(first)(@L @(last)@L)@(empty)EMPTY@(end)
@(end)
.brev
Here, the
.code @(empty)
clause specifies
.codn EMPTY .
So if there are no repetitions,
the text
.code EMPTY
is produced. If there is a single item in the list
.codn L ,
then
.code @(single)(@L)
produces that item between parentheses. Otherwise
if there are two or more items, the first item is produced with
a leading parenthesis followed by a space by
.code @(first)(@L
and the last item is produced with a closing parenthesis:
.codn @(last)@L) .
All items in between are emitted with a trailing space by
the main clause:
.codn @(rep)@L .
Example 2: show the list L like Example 1 above, but the empty list is
.codn () .
.verb
@(output)
(@(rep)@L @(last)@L@(end))
@(end)
.brev
This is simpler. The parentheses are part of the text which
surrounds the
.code @(rep)
construct, produced unconditionally.
If the list
.code L
is empty, then
.code @(rep)
produces no output, resulting in
.codn () .
If the list
.code L
has one or more items, then they are produced with
spaces each one, except the last which has no space.
If the list has exactly one item, then the
.code @(last)
applies to it
instead of the main clause: it is produced with no trailing space.
.dir close
The syntax of the
.code close
directive is:
.mono
.mets @(close << expr )
.onom
Where
.meta expr
evaluates to a stream. The
.code close
directive can be
used to explicitly close streams created using
.mono
.meti @(output ... :named << var )
.onom
syntax, as an alternative to
.mono
.meti @(output :finish << expr ).
.onom
Examples:
Write two lines to
.str foo.txt
over two output blocks using
a single stream:
.verb
@(output "foo.txt" :named foo)
Hello,
@(end)
@(output :continue foo)
world!
@(end)
@(close foo)
.brev
The same as above, using
.code :finish
rather than
.code :continue
so that the stream is closed at the end of the second block:
.verb
@(output "foo.txt" :named foo)
Hello,
@(end)
@(output :finish foo)
world!
@(end)
.brev
.NP* Output Filtering
Often it is necessary to transform the output to preserve its meaning
under the convention of a given data format. For instance, if a piece of
text contains the characters
.code <
or
.codn > ,
then if that text is being
substituted into HTML, these should be replaced by
.code <
and
.codn > .
This is what filtering is for. Filtering is applied to the contents of output
variables, not to any template text.
\*(TX implements named filters. Built-in filters are named by keywords, given
below. User-defined filters are possible, however. See notes on the deffilter
directive below.
Instead of a filter name, the syntax
.mono
.meti (fun << name )
.onom
can be used. This
denotes that the function called
.meta name
is to be used as a filter.
This is described in the next section Function Filters below.
Built-in filters named by keywords:
.coIP :tohtml
Filter text to HTML, representing special characters using HTML
ampersand sequences. For instance
.code >
is replaced by
.codn > .
.coIP :tohtml*
Filter text to HTML, representing special characters using HTML
ampersand sequences. Unlike
.codn :tohtml ,
this filter doesn't treat the single and double quote characters.
It is not suitable for preparing HTML fragments which end up
inserted into HTML tag attributes.
.coIP :fromhtml
Filter text with HTML codes into text in which the codes are replaced by the
corresponding characters. For instance
.code >
is replaced by
.codn > .
.coIP :upcase
Convert the 26 lowercase letters of the English alphabet to uppercase.
.coIP :downcase
Convert the 26 uppercase letters of the English alphabet to lowercase.
.coIP :frompercent
Decode percent-encoded text. Character triplets consisting
of the
.code %
character followed by a pair of hexadecimal digits (case insensitive)
are are converted to bytes having the value represented by the hexadecimal
digits (most significant nybble first). Sequences of one or more such bytes are
treated as UTF-8 data and decoded to characters.
.coIP :topercent
Convert to percent encoding according to RFC 3986. The text is first converted
to UTF-8 bytes. The bytes are then converted back to text as follows.
Bytes in the range 0 to 32, and 127 to 255 (note: including the ASCII DEL),
bytes whose values correspond to ASCII characters which are listed by RFC 3986
as being in the "reserved set", and the byte value corresponding to the
ASCII
.code %
character are encoded as a three-character sequence consisting
of the
.code %
character followed by two hexadecimal digits derived from the
byte value (most significant nybble first, upper case). All other bytes
are converted directly to characters of the same value without any such
encoding.
.coIP :fromurl
Decode from URL encoding, which is like percent encoding, except that
if the unencoded
.code +
character occurs, it is decoded to a space character. The
.code %20
sequence still decodes to space, and
.code %2B
to the
.code +
character.
.coIP :tourl
Encode to URL encoding, which is like percent encoding except that
a space maps to
.code +
rather than
.codn %20 .
The
.code +
character, being in the
reserved set, encodes to
.codn %2B .
.coIP :frombase64
Decode from the Base 64 encoding described in RFC 4648, section 5.
.coIP :tobase64
Encode to the Base 64 encoding described in RFC 4648, section 5.
.coIP :frombase64url
Decode from the Base64 encoding described in RFC 4648, section 6.
This uses the URL and filename safe alphabet, in which the
.code +
(plus) and
.code /
(slash) characters used in regular Base 64 are respectively replaced with
.code -
(minus) and
.code _
(underscore).
.coIP :tobase64url
Encode to the Base 64 encoding described in RFC 4648, section 6. See
.code :frombase64url
above.
.coIP :tonumber
Converts strings to numbers. Strings that contain a period,
.code e
or
.code E
are converted to floating point as if by the Lisp function
.codn flo-str .
Otherwise they are converted to integer as if using
.code int-str
with a radix of 10.
Non-numeric junk results in the object
.codn nil .
.coIP :toint
Converts strings to integers as if using
.code int-str
with a radix of 10.
Non-numeric junk results in the object
.codn nil .
.coIP :tofloat
Converts strings to floating-point values as if using the function
.codn flo-str .
Non-numeric junk results in the object
.codn nil .
.coIP :hextoint
Converts strings to integers as if using
.code int-str
with a radix of 16.
Non-numeric junk results in the object
.codn nil .
.PP
Examples:
To escape HTML characters in all variable substitutions occurring in an
output clause, specify
.code ":filter :tohtml"
in the directive:
.verb
@(output :filter :tohtml)
...
@(end)
.brev
To filter an individual variable, add the syntax to the variable spec:
.verb
@(output)
@{x :filter :tohtml}
@(end)
.brev
Multiple filters can be applied at the same time. For instance:
.verb
@(output)
@{x :filter (:upcase :tohtml)}
@(end)
.brev
This will fold the contents of
.code x
to uppercase, and then encode any special
characters into HTML. Beware of combinations that do not make sense.
For instance, suppose the original text is HTML, containing codes
like
.codn " .
The compound filter
.code "(:upcase :fromhtml)"
will not work
because
.code "
will turn to
.code "
which no longer be recognized by the
.code :fromhtml
filter, since the entity names in HTML codes
are case-sensitive.
Capture some numeric variables and convert to numbers:
.verb
@date @time @temperature @pressure
@(filter :tofloat temperature pressure)
@;; temperature and pressure can now be used in calculations
.brev
.NP* Function Filters
A function can be used as a filter. For this to be possible, the function must
conform to certain rules:
.IP 1.
The function must take two special arguments, which may be followed
by additional arguments.
.IP 2.
When the function is called, the first argument will be bound to a string,
and the second argument will be unbound. The function must produce a
value by binding it to the second argument. If the filter is to be used
as the final filter in a chain, it must produce a string.
.PP
For instance, the following is a valid filter function:
.verb
@(define foo_to_bar (in out))
@ (next :string in)
@ (cases)
foo
@ (bind out "bar")
@ (or)
@ (bind out in)
@ (end)
@(end)
.brev
This function binds the
.code out
parameter to
.str bar
if the in parameter
is
.strn foo ,
otherwise it binds the
.code out
parameter to a copy of the
.code in
parameter.
This is a simple filter.
To use the filter, use the syntax
.code "(:fun foo_to_bar)"
in place of a filter name.
For instance in the
.code bind
directive:
.verb
@(bind "foo" "bar" :lfilt (:fun foo_to_bar))
.brev
The above should succeed since the left side is filtered from
.str foo
to
.strn bar ,
so that there is a match.
Function filters can be used in a chain:
.verb
@(output :filter (:downcase (:fun foo_to_bar) :upcase))
...
@(end)
.brev
Here is a split function which takes an extra argument
which specifies the separator:
.verb
@(define split (in out sep))
@ (next :list in)
@ (coll)@(maybe)@token@sep@(or)@token@(end)@(end)
@ (bind out token)
@(end)
.brev
Furthermore, note that it produces a list rather than a string.
This function separates the argument in into tokens according to the
separator text carried in the variable
.codn sep .
Here is another function,
.codn join ,
which catenates a list:
.verb
@(define join (in out sep))
@ (output :into out)
@ (rep)@in@sep@(last)@in@(end)
@ (end)
@(end)
.brev
Now here is these two being used in a chain:
.verb
@(bind text "how,are,you")
@(output :filter (:fun split ",") (:fun join "-"))
@text
@(end)
.brev
Output:
.verb
how-are-you
.brev
When the filter invokes a function, it generates the first two arguments
internally to pass in the input value and capture the output. The remaining
arguments from the
.code "(:fun ...)"
construct are also passed to the function.
Thus the string objects
.str ","
and
.str "-"
are passed as the
.code sep
argument to
.code split
and
.codn join .
Note that
.code split
puts out a list, which
.code join
accepts. So the overall filter
chain operates on a string: a string goes into split, and a string comes out of
join.
.dir deffilter
The
.code deffilter
directive allows a query to define a custom filter, which
can then be used in
.code output
clauses to transform substituted data.
The syntax of
.code deffilter
is illustrated in this example:
.IP code:
.mono
\ @(deffilter rot13
("a" "n")
("b" "o")
("c" "p")
("d" "q")
("e" "r")
("f" "s")
("g" "t")
("h" "u")
("i" "v")
("j" "w")
("k" "x")
("l" "y")
("m" "z")
("n" "a")
("o" "b")
("p" "c")
("q" "d")
("r" "e")
("s" "f")
("t" "g")
("u" "h")
("v" "i")
("w" "j")
("x" "k")
("y" "l")
("z" "m"))
@(collect)
@line
@(end)
@(output :filter rot13)
@(repeat)
@line
@(end)
@(end)
.onom
.IP data:
.mono
\ hey there!
.onom
.IP output:
.mono
\ url gurer!
.onom
.PP
The
.code deffilter
symbol must be followed by the name of the filter to be defined,
followed by bind expressions which evaluate to lists of strings. Each list must
be at least two elements long and specifies one or more texts which are mapped
to a replacement text. For instance, the following specifies a telephone keypad
mapping from uppercase letters to digits.
.verb
@(deffilter alpha_to_phone ("E" "0")
("J" "N" "Q" "1")
("R" "W" "X" "2")
("D" "S" "Y" "3")
("F" "T" "4")
("A" "M" "5")
("C" "I" "V" "6")
("B" "K" "U" "7")
("L" "O" "P" "8")
("G" "H" "Z" "9"))
@(deffilter foo (`@a` `@b`) ("c" `->@d`))
@(bind x ("from" "to"))
@(bind y ("---" "+++"))
@(deffilter sub x y)
.brev
The last
.code deffilter
has the same effect as the
.mono
@(deffilter sub ("from" "to") ("---" "+++"))
.onom
directive.
Filtering works using a longest match algorithm. The input is scanned from left
to right, and the longest piece of text is identified at every character
position which matches a string on the left-hand side, and that text is
replaced with its associated replacement text. The scanning then continues
at the first character after the matched text.
If none of the strings matches at a given character position, then that
character is passed through the filter untranslated, and the scan continues at
the next character in the input.
Filtering is not in-place but rather instantiates a new text, and so
replacement text is not re-scanned for more replacements.
If a filter definition accidentally contains two or more repetitions of the
same left-hand string with different right-hand translations, the later ones
take precedence. No warning is issued.
.dir filter
The syntax of the
.code filter
directive is:
.verb
@(filter FILTER { VAR }+ )
.brev
A filter is specified, followed by one or more variables whose values
are filtered and stored back into each variable.
Example: convert
.codn a ,
.codn b ,
and
.code c
to uppercase and HTML encode:
.verb
@(filter (:upcase :tohtml) a b c)
.brev
.SS* Exceptions
.NP* Introduction
The exceptions mechanism in \*(TX is another
disciplined form of nonlocal transfer, in addition to the blocks
mechanism (see Blocks above). Like blocks, exceptions provide a construct
which serves as the target for a dynamic exit. Both blocks and exceptions
can be used to bail out of deep nesting when some condition occurs.
However, exceptions provide more complexity. Exceptions are useful for
error handling, and \*(TX in fact maps certain error situations to exception
control transfers. However, exceptions are not inherently an error-handling
mechanism; they are a structured dynamic control transfer mechanism, one
of whose applications is error handling.
An exception control transfer (simply called an exception) is always identified
by a symbol, which is its type. Types are organized in a subtype-supertype
hierarchy. For instance, the
.code file-error
exception type is a subtype of the
.code error
type. This means that a file error is a kind of error. An exception
handling block which catches exceptions of type
.code error
will catch exceptions of
type
.codn file-error ,
but a block which catches
.code file-error
will not catch all
exceptions of type
.codn error .
A
.code query-error
is a kind of error, but not a kind of
.codn file-error .
The symbol
.code t
is the supertype of every type: every exception type
is considered to be a kind of
.codn t .
(Mnemonic:
.code t
stands for type, as in any type).
Exceptions are handled using
.code @(catch)
clauses within a
.code @(try)
directive.
In addition to being useful for exception handling, the
.code @(try)
directive
also provides unwind protection by means of a
.code @(finally)
clause,
which specifies query material to be executed unconditionally when the
.code try
clause terminates, no matter how it terminates.
.dir try
The general syntax of the
.code try
directive is
.verb
@(try)
... main clause, required ...
... optional catch clauses ...
... optional finally clause
@(end)
.brev
A
.code catch
clause looks like:
.verb
@(catch TYPE [ PARAMETERS ])
.
.
.
.brev
and also this simple form:
.verb
@(catch)
.
.
.
.brev
which catches all exceptions, and is equivalent
to
.codn "@(catch t)" .
A
.code finally
clause looks like:
.verb
@(finally)
...
.
.
.brev
The main clause may not be empty, but the catch and finally may be.
A try clause is surrounded by an implicit anonymous block (see Blocks section
above). So for instance, the following is a no-op (an operation with no effect,
other than successful execution):
.verb
@(try)
@(accept)
@(end)
.brev
The
.code @(accept)
causes a successful termination of the implicit anonymous block.
Execution resumes with query lines or directives which follow, if any.
.code try
clauses and blocks interact. For instance, an
.code accept
from within
a try clause invokes a
.codn finally .
.IP code:
.mono
\ @(block foo)
@ (try)
@ (accept foo)
@ (finally)
@ (output)
bye!
@ (end)
@ (end)
.onom
.IP output:
.mono
\ bye!
.onom
.PP
How this works: the
.code try
block's main clause is
.codn "@(accept foo)" .
This causes
the enclosing block named
.code foo
to terminate, as a successful match.
Since the
.code try
is nested within this block, it too must terminate
in order for the block to terminate. But the try has a
.code finally
clause,
which executes unconditionally, no matter how the try block
terminates. The
.code finally
clause performs some output, which is seen.
Note that
.code finally
interacts with
.code accept
in subtle ways not revealed in this example; they are documented in
the description of
.code accept
under the
.code block
directive documentation.
.coNP The @ finally clause
A
.code try
directive can terminate in one of three ways. The main clause
may match successfully, and possibly yield some new variable bindings.
The main clause may fail to match. Or the main clause may be terminated
by a nonlocal control transfer, like an exception being thrown or a block
return (like the block foo example in the previous section).
No matter how the
.code try
clause terminates, the
.code finally
clause is processed.
The
.code finally
clause is itself a query which binds variables, which leads to
questions: what happens to such variables? What if the
.code finally
block fails
as a query? As well as: what if a
.code finally
clause itself initiates a
control transfer? Answers follow.
Firstly, a
.code finally
clause will contribute variable bindings only if the main
clause terminates normally (either as a successful or failed match).
If the main clause of the
.code try
block successfully matches, then the
.code finally
block continues
matching at the next position in the data, and contributes bindings.
If the main clause fails, then the
.code finally
block tries to match at the same position where the main clause failed.
The overall
.code try
directive succeeds as a match if either the main clause
or the
.code finally
clause succeed. If both fail, then the
.code try
directive is a failed match.
Example:
.IP code:
.mono
\ @(try)
@a
@(finally)
@b
@(end)
@c
.onom
.IP data:
.mono
\ 1
2
3
.onom
.IP result:
.mono
\ a="1"
b="2"
c="3"
.onom
.PP
In this example, the main clause of the
.code try
captures line
.str 1
of the data as
variable
.codn a ,
then the finally clause captures
.str 2
as
.codn b ,
and then the query continues with the
.code @c
line after try block, so that
.code c
captures
.strn "3" .
Example:
.IP code:
.mono
\ @(try)
hello @a
@(finally)
@b
@(end)
@c
.onom
.IP data:
.mono
\ 1
2
.onom
.IP result:
.mono
\ b="1"
c="2"
.onom
.PP
In this example, the main clause of the
.code try
fails to match, because
the input is not prefixed with
.strn "hello " .
However, the
.code finally
clause
matches, binding
.code b
to
.strn "1" .
This means that the try block is a successful
match, and so processing continues with
.code @c
which captures
.strn "2" .
When
.code finally
clauses are processed during a nonlocal return,
they have no externally visible effect if they do not bind variables.
However, their execution makes itself known if they perform side effects,
such as output.
A
.code finally
clause guards only the main clause and the
.code catch
clauses. It does not
guard itself. Once the finally clause is executing, the
.code try
block is no
longer guarded. This means if a nonlocal transfer, such as a block accept
or exception, is initiated within the finally clause, it will not re-execute
the
.code finally
clause. The
.code finally
clause is simply abandoned.
The disestablishment of blocks and
.code try
clauses is properly interleaved
with the execution of
.code finally
clauses. This means that all surrounding
exit points are visible in a
.code finally
clause, even if the
.code finally
clause
is being invoked as part of a transfer to a distant exit point.
The finally clause can make a control transfer to an exit point which
is more near than the original one, thereby "hijacking" the control
transfer. Also, the anonymous block established by the
.code try
directive
is visible in the
.code finally
clause.
Example:
.verb
@(try)
@ (try)
@ (next "nonexistent-file")
@ (finally)
@ (accept)
@ (end)
@(catch file-error)
@ (output)
file error caught
@ (end)
@(end)
.brev
In this example, the
.code @(next)
directive throws an exception of type
.codn file-error ,
because the given file does not exist. The exit point for this exception is the
.code "@(catch file-error)"
clause in the outermost
.code try
block. The inner block is
not eligible because it contains no catch clauses at all. However, the inner
try block has a finally clause, and so during the processing of this
exception which is headed for
.codn "@(catch file-error)" ,
the
.code finally
clause performs an anonymous
.codn accept .
The exit point for that
.code accept
is the anonymous block
surrounding the inner
.codn try .
So the original
transfer to the
.code catch
clause is thereby abandoned. The inner
.code try
terminates
successfully due to the
.codn accept ,
and since it constitutes the main clause of the outer try,
that also terminates successfully. The
.str "file error caught"
message is never printed.
.c1NP catch clauses
.code catch
clauses establish their associated
.code try
blocks as potential exit points for
exception-induced control transfers (called "throws").
A
.code catch
clause specifies an optional list of symbols which represent
the exception types which it catches. The
.code catch
clause will catch
exceptions which are a subtype of any one of those exception types.
If a
.code try
block has more than one
.code catch
clause which can match a given
exception, the first one will be invoked.
When a
.code catch
is invoked, it is understood that the main clause did
not terminate normally, and so the main clause could not have produced any
bindings.
.code catch
clauses are processed prior to
.codn finally .
If a
.code catch
clause itself throws an exception, that exception cannot
be caught by that same clause or its siblings in the same try block.
The
.code catch
clauses of that block are no longer visible at that point.
Nevertheless, the
.code catch
clauses are still protected by the finally block.
If a catch clause throws, or otherwise terminates, the
.code finally
block is still processed.
If a
.code finally
block throws an exception, then it is simply aborted;
the remaining directives in that block are not processed.
So the success or failure of the
.code try
block depends on the behavior of the
.code catch
clause or the
.code finally
clause, if there is one. If either of them succeed, then the try
block is considered a successful match.
Example:
.IP code:
.mono
\ @(try)
@ (next "nonexistent-file")
@ x
@ (catch file-error)
@a
@(finally)
@b
@(end)
@c
.onom
.IP data:
.mono
\ 1
2
3
.onom
.IP result:
.mono
\ a="1"
b="2"
c="3"
.onom
.PP
Here, the
.code try
block's main clause is terminated abruptly by a
.code file-error
exception from the
.code @(next)
directive. This is handled by the
.code catch
clause, which binds variable
.code a
to the input line
.strn 1 .
Then the
.code finally
clause executes, binding
.code b
to
.strn 2 .
The
.code try
block then terminates successfully, and so
.code @c
takes
.strn "3" .
.coNP @ catch Clauses with Parameters
A
.code catch
clause may have parameters following the type name, like this:
.verb
@(catch pair (a b))
.brev
To write a catch-all with parameters, explicitly write the
master supertype t:
.verb
@(catch t (arg ...))
.brev
Parameters are useful in conjunction with
.codn throw .
The built-in
.code error
exceptions carry one argument, which is a string containing
the error message. Using
.codn throw ,
arbitrary parameters can be passed
from the throw site to the catch site.
.dir throw
The
.code throw
directive generates an exception. A type must be specified,
followed by optional arguments, which are bind expressions. For example,
.verb
@(throw pair "a" `@file.txt`)
.brev
throws an exception of type
.codn pair ,
with two arguments, being
.str a
and the expansion of the quasiliteral
.codn `@file.txt` .
The selection of the target
.code catch
is performed purely using the type
name; the parameters are not involved in the selection.
Binding takes place between the arguments given in
.code throw
and the target
.codn catch .
If any
.code catch
parameter, for which a
.code throw
argument is given, is a bound
variable, it has to be identical to the argument, otherwise the catch fails.
(Control still passes to the
.codn catch ,
but the catch is a failed match).
.IP code:
.mono
\ @(bind a "apple")
@(try)
@(throw e "banana")
@(catch e (a))
@(end)
.onom
.IP result:
.mono
\ [query fails]
.onom
.PP
If any argument is an unbound variable, the corresponding parameter
in the
.code catch
is left alone: if it is an unbound variable, it remains
unbound, and if it is bound, it stays as is.
.IP code:
.mono
\ @(try)
@(throw e "honda" unbound)
@(catch e (car1 car2))
@car1 @car2
@(end)
.onom
.IP data:
.mono
\ honda toyota
.onom
.IP result:
.mono
\ car1="honda"
car2="toyota"
.onom
.PP
If a
.code catch
has fewer parameters than there are throw arguments,
the excess arguments are ignored:
.IP code:
.mono
\ @(try)
@(throw e "banana" "apple" "pear")
@(catch e (fruit))
@(end)
.onom
.IP result:
.mono
\ fruit="banana"
.onom
.PP
If a
.code catch
has more parameters than there are throw arguments, the excess
parameters are left alone. They may be bound or unbound variables.
.IP code:
.mono
\ @(try)
@(throw e "honda")
@(catch e (car1 car2))
@car1 @car2
@(end)
.onom
.IP data:
.mono
\ honda toyota
.onom
.IP result:
.mono
\ car1="honda"
car2="toyota"
.onom
.PP
A
.code throw
argument passing a value to a
.code catch
parameter which is unbound causes
that parameter to be bound to that value.
.code throw
arguments are evaluated in the context of the
.codn throw ,
and the bindings
which are available there. Consideration of what parameters are bound
is done in the context of the catch.
.IP code:
.mono
\ @(bind c "c")
@(try)
@(forget c)
@(bind (a c) ("a" "lc"))
@(throw e a c)
@(catch e (b a))
@(end)
.onom
.IP result:
.mono
\ c="c"
b="a"
a="lc"
.onom
.PP
In the above example,
.code c
has a top-level binding to the string
.strn "c" ,
but then becomes unbound
via
.code forget
within the
.code try
construct, and rebound to the value
.strn lc .
Since the
.code try
construct is terminated by a
.codn throw ,
these modifications of the
binding environment are discarded. Hence, at the end of the query, variable
.code c
ends up bound to the original value
.strn c .
The
.code throw
still takes place
within the scope of the bindings set up by the
.code try
clause, so the values of
.code a
and
.code c
that are thrown are
.str a
and
.strn lc .
However, at the
.code catch
site, variable
.code a
does not have a binding. At that point, the binding to
.str a
established in
the
.code try
has disappeared already. Being unbound, the
.code catch
parameter
.code a
can take
whatever value the corresponding throw argument provides, so it ends up with
.strn lc .
There is a horizontal form of
.codn throw .
For instance:
.verb
abc@(throw e 1)
.brev
throws exception
.code e
if
.code abc
matches.
If
.code throw
is used to generate an exception derived from type
.code error
and that exception is not handled, \*(TX will issue diagnostics on the
.code *stderr*
stream and terminate. If an exception derived from
.code warning
is not handled, \*(TX will generate diagnostics on the
.code *stderr*
stream, after which control returns to the
.code throw
directive, and proceeds with the next directive.
If an exception not derived from
.code error
is thrown, control returns to the
.code throw
directive and proceeds with the next directive.
.dir defex
The
.code defex
directive allows the query writer to invent custom exception types,
which are arranged in a type hierarchy (meaning that some exception types are
considered subtypes of other types).
Subtyping means that if an exception type
.code B
is a subtype of
.codn A ,
then every
exception of type
.code B
is also considered to be of type
.codn A .
So a catch for type
.code A
will also catch exceptions of type
.codn B .
Every type is a supertype of itself: an
.code A
is a kind of
.codn A .
This implies that every type is a subtype of itself
also. Furthermore, every type is a subtype of the type
.codn t ,
which has no
supertype other than itself. Type
.code nil
is a subtype of every type, including
itself. The subtyping relationship is transitive also. If
.code A
is a subtype
of
.codn B ,
and
.code B
is a subtype of
.codn C ,
then
.code A
is a subtype of
.codn C .
.code defex
may be invoked with no arguments, in which case it does nothing:
.verb
@(defex)
.brev
It may be invoked with one argument, which must be a symbol. This introduces a
new exception type. Strictly speaking, such an introduction is not necessary;
any symbol may be used as an exception type without being introduced by
.codn @(defex) :
.verb
@(defex a)
.brev
Therefore, this also does nothing, other than document the intent to use
a as an exception.
If two or more argument symbols are given, the symbols are all introduced as
types, engaged in a subtype-supertype relationship from left to right.
That is to say, the first (leftmost) symbol is a subtype of the next one,
which is a subtype of the next one and so on. The last symbol, if it
had not been already defined as a subtype of some type, becomes a
direct subtype of the master supertype
.codn t .
Example:
.verb
@(defex d e)
@(defex a b c d)
.brev
The first directive defines
.code d
as a subtype of
.codn e ,
and
.code e
as a subtype of
.codn t .
The second defines
.code a
as a subtype of
.codn b ,
.code b
as a subtype of
.codn c ,
and
.code c
as a subtype of
.codn d ,
which is already defined as a subtype of
.codn e .
Thus
.code a
is now a subtype of
.codn e .
The above can be condensed to:
.verb
@(defex a b c d e)
.brev
Example:
.IP code:
.mono
\ @(defex gorilla ape primate)
@(defex monkey primate)
@(defex human primate)
@(collect)
@(try)
@(skip)
@(cases)
gorilla @name
@(throw gorilla name)
@(or)
monkey @name
@(throw monkey name)
@(or)
human @name
@(throw human name)
@(end)@#cases
@(catch primate (name))
@kind @name
@(output)
we have a primate @name of kind @kind
@(end)@#output
@(end)@#try
@(end)@#collect
.onom
.IP data:
.mono
\ gorilla joe
human bob
monkey alice
.onom
.IP output:
.mono
\ we have a primate joe of kind gorilla
we have a primate bob of kind human
we have a primate alice of kind monkey
.onom
.PP
Exception types have a pervasive scope. Once a type relationship is introduced,
it is visible everywhere. Moreover, the
.code defex
directive is destructive,
meaning that the supertype of a type can be redefined. This is necessary so
that something like the following works right:
.verb
@(defex gorilla ape)
@(defex ape primate)
.brev
These directives are evaluated in sequence. So after the first one, the
.code ape
type has the type
.code t
as its immediate supertype. But in the second directive,
.code ape
appears again, and is assigned the
.code primate
supertype, while retaining
.code gorilla
as a subtype. This situation could be diagnosed as an
error, forcing the programmer to reorder the statements, but instead
\*(TX obliges. However, there are limitations. It is an error to define a
subtype-supertype relationship between two types if they are already connected
by such a relationship, directly or transitively. So the following
definitions are in error:
.verb
@(defex a b)
@(defex b c)
@(defex a c)@# error: a is already a subtype of c, through b
@(defex x y)
@(defex y x)@# error: circularity; y is already a supertype of x.
.brev
.dir assert
The
.code assert
directive requires the remaining query or subquery which follows it
to match. If the remainder fails to match, the
.code assert
directive throws an exception. If the directive is simply
.verb
@(assert)
.brev
Then it throws an assertion of type assert, which is a subtype of error.
The
.code assert
directive also takes arguments similar to the
.code throw
directive: an exception symbol and additional arguments which are bind
expressions, and may be unbound variables. The following assert directive, if
it triggers, will throw an exception of type
.codn foo ,
with arguments
.code 1
and
.strn 2 :
.verb
@(assert foo 1 "2")
.brev
Example:
.verb
@(collect)
Important Header
----------------
@(assert)
Foo: @a, @b
@(end)
.brev
Without the assertion in places, if the
.code "Foo: @a, @b"
part does not
match, then the entire interior of the
.code @(collect)
clause fails,
and the collect continues searching for another match.
With the assertion in place, if the text
.str "Important Header"
and its
underline match, then the remainder of the collect body must
match, otherwise an exception is thrown. Now the program will not
silently skip over any Important Header sections due to a problem
in its matching logic. This is particularly useful when the matching is varied
with numerous cases, and they must all be handled.
There is a horizontal
.code assert
directive also. For instance:
.verb
abc@(assert)d@x
.brev
asserts that if the prefix
.str abc
is matched, then it must be
followed by a successful match for
.strn "d@x" ,
or else an exception is thrown.
If the exception is not handled, and is derived from
.code error
then \*(TX issues diagnostics on the
.code *stderr*
stream and terminates. If the exception is derived from
.code warning
and not handled, \*(TX issues a diagnostic on
.code *stderr*
after which control returns to the
.code assert
directive. Control silently returns to the
.code assert
directive if an exception of any other kind is not handled.
When control returns to
.code assert
due to an unhandled exception, it behaves like a failed match,
similarly to the require directive.
.SH* TXR LISP
The \*(TX language contains an embedded Lisp dialect called \*(TL.
This language is exposed in \*(TX in a number of ways.
In any situation that calls for an expression, a Lisp
expression can be used, if it is preceded by the
.code @
character. The Lisp expression
is evaluated and its value becomes the value of that expression.
Thus, \*(TX directives are embedded in literal text using
.codn @ ,
and Lisp expressions
are embedded in directives using
.code @
also.
Furthermore, certain directives evaluate Lisp expressions without
requiring
.codn @ .
These are
.codn @(do) ,
.codn @(require) ,
.codn @(assert) ,
.code @(if)
and
.codn @(next) .
\*(TL code can be placed into files. On the command
line, \*(TX treats files with a
.str ".tl"
suffix as \*(TL code, and the
.code @(load)
directive does also.
\*(TX also provides an interactive listener for Lisp evaluation.
Lastly, \*(TL expressions can be evaluated via the
command line, using the
.code -e
and
.code -p
options.
.B
Examples:
Bind variable
.code a
to the integer 4:
.verb
@(bind a @(+ 2 2))
.brev
Bind variable
.code b
to the standard input stream. Note that
.code @
is not required on a Lisp variable:
.verb
@(bind a *stdin*)
.brev
Define several Lisp functions inside
.codn @(do) :
.verb
@(do
(defun add (x y) (+ x y))
(defun occurs (item list)
(cond ((null list) nil)
((atom list) (eql item list))
(t (or (eq (first list) item)
(occurs item (rest list)))))))
.brev
Trigger a failure unless previously bound variable
.code answer
is greater than 42:
.verb
@(require (> (int-str answer) 42)
.brev
.SS* Overview
\*(TL is a small and simple dialect, like Scheme, but much more similar to
Common Lisp than Scheme. It has separate value and function binding namespaces,
like Common Lisp (and thus is a Lisp-2 type dialect), and represents Boolean
.B true
and
.B false
with the symbols
.code t
and
.code nil
(note the case sensitivity of
identifiers denoting symbols!). Furthermore, the symbol
.code nil
is also the empty list, which terminates nonempty lists.
\*(TL has lexically scoped local variables and dynamic global variables,
similarly to Common Lisp, including the convention that
.code defvar
marks symbols for dynamic binding in local scopes. Lexical closures
are supported. \*(TL also supports global lexical variables via
.codn defvarl .
Functions are lexically scoped in \*(TL; they can be
defined in the pervasive global environment using
.code defun
or in local scopes using
.code flet
and
.codn labels .
.SS* Additional Syntax
Much of the \*(TL syntax has been introduced in the previous sections of the
manual, since directive forms are based on it. There is some additional syntax
that is useful in \*(TL programming.
.NP* Symbol Tokens
The symbol tokens in \*(TL,
called a
.meta lident
(Lisp identifier) has a similar syntax to the
.meta bident
(braced identifier) in the \*(TX pattern language. It may consist of
all the same characters, as well as the
.code /
(slash) character which may not be used in a
.metn bident .
Thus a
.meta lident
may consist of these characters, in addition to letters, numbers and
underscores:
.mono
! $ % & * + - < = > ? \e ~ /
.onom
and may not look like a number.
A
.meta lident
may also include all of the Unicode characters which are permitted in a
.metn bident .
The one character which is allowed in a
.meta lident
but not in a
.meta bident
is
.code /
(forward slash).
A lone
.code /
is a valid
.meta lident
and consequently a symbol token in \*(TL. The token
.code /abc/
is also a symbol, and, unlike in a braced expression, is not a regular
expression. In \*(TL expressions, regular expressions are written with
a leading
.codn # .
.NP* Package Prefixes
If a symbol name contains a colon, the
.I lident
characters, if any, before that colon constitute the package prefix.
For example, the syntax
.code foo:bar
denotes
.code bar
symbol in the
.code foo
package.
It is a syntax error to read a symbol whose package doesn't exist.
If the package exists, but the symbol name doesn't exist in that package,
then the symbol is interned in that package.
If the package name is an empty string (the colon is preceded by nothing), the
package is understood to be the
.code keyword
package. The symbol is interned in that package.
The syntax
.code :test
denotes the symbol
.code test
in the
.code keyword
package, the same as
.codn keyword:test .
Symbols in the keyword package are self-evaluating. This means that
when a keyword symbol is evaluated as a form, the value of that form
is the keyword symbol itself. Exactly two non-keyword symbols also
have this special self-evaluating behavior:
the symbols
.code t
and
.code nil
in the user package, whose fully qualified names are
.code usr:t
and
.codn usr:nil .
The syntax
.code @foo:bar
denotes the meta prefix
.code @
being applied to the
.code foo:bar
symbol, not to a symbol in the
.code @foo
package.
The syntax
.code #:bar
denotes an uninterned symbol named
.codn bar ,
described in the next section.
.TP* "Dialect Note:"
In ANSI Common Lisp, the
.code foo:bar
syntax does not intern the symbol
.code bar
in the
.code foo
package; the symbol must exist and be an exported symbol, or else the syntax is
erroneous. In ANSI Common Lisp, the syntax
.code foo::bar
does intern
.code foo
in the
.code bar
package. \*(TX's package system has no double-colon syntax, and lacks the concept of exported symbols.
.NP* Uninterned Symbols
Uninterned symbols are written with the
.code #:
prefix, followed by zero or more
.I lident
characters.
When an uninterned symbol is read, a new, unique symbol is constructed,
with the specified name. Even if two uninterned symbols have the same name,
they are different objects. The
.code make-sym
and
.code gensym
functions produce uninterned symbols.
"Uninterned" means "not entered into a package". Interning refers to a
process which combines package lookup with symbol creation, which ensures
that multiple occurrences of a symbol name in written syntax are all converted
to the same object: the first occurrence creates the symbol and associates it
with its name in a package. Subsequent occurrences do not create a new symbol,
but retrieve the existing one.
.NP* Meta-Atoms and Meta-Expressions
An expression may be preceded by the
.code @
(at sign) character. If the expression is an
.codn atom ,
then this is a meta-atom, otherwise it is a meta-expression.
When the atom is a symbol, this is also called a meta-symbol and in situations
when such a symbol behaves like a variable, it is also referred to as a
meta-variable.
When the atom is an integer, the meta-atom expression is called a meta-number.
Meta-atom and meta-expression expressions have no evaluation semantics;
evaluating them throws an exception. They play a syntactic role in the
.code op
operator, which makes use of meta-variables and meta-numbers, and in structural
pattern matching, which uses meta-variables as pattern variables and whose
operator vocabulary is based on meta-expressions.
Meta-expressions also appear in the quasiliteral notation.
In other situations, application code may assign meaning to meta syntax as the
programmer sees fit.
Meta syntax is defined as a shorthand notation, as follows:
If
.code X
is the syntax of an atom, such as a symbol, string or vector, then
.code @X
is a shorthand for the expression
.codn "(sys:var X)" .
Here,
.code sys:var
refers to the
.code var
symbol in the
.codn system-package .
If
.code X
is a compound expression, either
.code "(...)"
or
.codn "[...]" ,
then
.code @X
is a shorthand for the expression
.codn "(sys:expr X)" .
The behavior of
.code @
followed by the syntax of a floating-point constant introduced by a leading
decimal point, not preceded by digits, is unspecified. Examples of this
are
.code "@.123"
and
.codn "@.123E+5" .
The behavior of
.code @
followed by the syntax of a floating-point expression in E notation,
which lacks a decimal point, is also unspecified. An example of this is
.codn @12E5 .
It is a syntax error for
.code @
to be followed by what appears to be a floating-point constant consisting
of a decimal point flanked by digits on both sides. For instance
.code @1.2
is rejected.
A meta-expression followed by a period, and the syntax of another object is
otherwise interpreted as a referencing dot expression. For instance
.code @1.E3
denotes
.code "(qref @1 E3)"
which, in turn, denotes
.codn "(qref (sys:var 1) E3)" ,
even though the unprefixed character sequence
.code 1.E3
is otherwise a floating-point constant.
.NP* Consing Dot
Unlike other major Lisp dialects, \*(TL allows a consing dot with no forms
preceding it. This construct simply denotes the form which follows the dot.
That is to say, the parser implements the following transformation:
.verb
(. expr) -> expr
.brev
This is convenient in writing function argument lists that only take
variable arguments. Instead of the syntax:
.verb
(defun fun args ...)
.brev
the following syntax can be used:
.verb
(defun fun (. args) ...)
.brev
When a
.code lambda
form is printed, it is printed in the following style.
.verb
(lambda nil ...) -> (lambda () ...)
(lambda sym ...) -> (lambda (. sym) ...)
(lambda (sym) ...) -> (lambda (sym) ...)
.brev
In no other circumstances is
.code nil
printed as
.codn () ,
or an atom
.meta sym
as
.codn "(. sym)" .
.NP* Referencing Dot
A dot token which is flanked by expressions on both sides, without any
intervening whitespace, is the referencing dot, and not the consing dot.
The referencing dot is a syntactic sugar which translated to the
.code qref
syntax ("quoted ref"). When evaluated as a form, this syntax denotes structure
access; see Structures. However, it is possible to put this syntax to use for
other purposes, in other contexts.
.verb
;; a.b may be almost any expressions
a.b <--> (qref a b)
a.b.c <--> (qref a b c)
a.(qref b c) <--> (qref a b c)
(qref a b).c <--> (qref (qref a b) c)
.brev
That is to say, this dot operator constructs a
.code qref
expression out of its left and right arguments. If the right argument
of the dot is already a qref expression (whether produced by another instance
of the dot operator, or expressed directly) it is merged. This requires
the qref dot operator to be right-to-left associative, so that
.code a.b.c
works by first translating
.code b.c
to
.codn "(qref b c)" ,
and then adjoining
.code a
to produce
.codn "(qref a b c)" .
If the referencing dot is immediately followed by a question mark, it forms
a single token, which produces the following syntactic variation,
in which the following item is annotated as a list headed by
the symbol
.codn t :
.verb
a.?b <--> (t a).b <--> (qref (t a) b)
a.?b.?c <--> (t a).(t b).c <--> (qref (t a) (t b) c)
a.?(b) <--> (t a).(b) <--> (qref (t a) (b))
(a).?b <--> (t (a)).b <--> (qref (t (a)) b)
.brev
This syntax denotes
.I null-safe
access to structure slots and methods.
.code a.?b
means that
.code a
may evaluate to
.codn nil ,
in which case the expression yields
.codn nil ;
otherwise,
.code a
must evaluate to a
.code struct
which has a slot
.codn b ,
and the expression denotes access to that slot.
Similarly,
.code "a.?(b 1)"
means that if
.code a
evaluates to
.codn nil ,
the expression yields
.codn nil ;
otherwise,
.code a
is treated as a struct object whose method
.code b
is invoked with argument
.codn 1 ,
and the value returned by that method becomes the value of
the expression.
Integer tokens cannot be involved in this syntax, because they
form floating-point constants when juxtaposed with a dot.
Such ambiguous uses of floating-point tokens are diagnosed as syntax errors:
.verb
(a.4) ;; error: cramped floating-point literal
(a .4) ;; good: a followed by 0.4
.brev
.NP* Unbound Referencing Dot
Closely related to the referencing dot syntax is the unbound
referencing dot. This is a dot which is flanked by an expression on the right,
without any intervening whitespace, but is not preceded by an expression
Rather, it is preceded by whitespace,
or some punctuation such as
.codn [ ,
.code (
or
.codn ' .
This is a syntactic sugar which translates to
.code uref
syntax:
.verb
.a <--> (uref a)
.a.b <--> (uref a b)
.a.?b <--> (uref (t a) b)
.brev
If the unbound referencing dot is itself combined with a question
mark to form the
.code .?
token, then the translation to
.code uref
is as follows:
.verb
.?a <--> (uref t a)
.?a.b <--> (uref t a b)
.?a.?b <--> (uref t a (t b))
.brev
When the unbound referencing dot is applied to a dotted expression,
this can be understood as a conversion of
.code qref
to
.codn uref .
Indeed, this is exactly what happens if the unbound dot is applied to an
explicit
.code qref
expression:
.verb
.(qref a b) <--> (uref a b)
.brev
The unbound referencing dot takes its name from the semantics of the
.code uref
macro, which produces a function that implements late binding of an
object to a method slot. Whereas the expression
.code obj.a.b
denotes accessing object
.code obj
to retrieve slot
.code a
and then accessing slot
.code b
of the object from that slot, the expression
.code .a.b.
represents a "disembodied" reference: it produces a function which takes an
object as an argument and then performs the implied slot referencing on that
argument. When the function is called, it is said to bind the referencing to
the object. Hence that referencing is "unbound".
Whereas the expression
.code .a
produces a function whose argument must be an object,
.code .?a
produces a function whose argument may be
.codn nil .
The function detects this case and returns
.codn nil .
.NP* Quote and Quasiquote
.RS
.meIP >> ' expr
The quote character in front of an expression is used for suppressing evaluation,
which is useful for forms that evaluate to something other than themselves.
For instance if
.code "'(+ 2 2)"
is evaluated, the value is the three-element list
.codn "(+ 2 2)" ,
whereas if
.code "(+ 2 2)"
is evaluated, the value is
.codn 4 .
Similarly, the value of
.code 'a
is the symbol
.code a
itself, whereas the value of
.code a
is the contents of the variable
.codn a .
.meIP >> ^ qq-template
The caret in front of an expression is a quasiquote. A quasiquote is like
a quote, but with the possibility of substitution of material.
Under a quasiquote, form is considered to be a quasiquote template. The template
is considered to be a literal structure, except that it may contain
the notations
.mono
.meti >> , expr
.onom
and
.mono
.meti >> ,* expr
.onom
which denote non-constant parts.
A quasiquote gets translated into code which, when evaluated, constructs
the structure implied by
.metn qq-template ,
taking into account the unquotes and splices.
A quasiquote also processes nested quasiquotes specially.
If
.meta qq-template
does not contain any unquotes or splices (which match its
level of nesting), or is simply an atom, then
.mono
.meti >> ^ qq-template
.onom
is equivalent to
.mono
.meti >> ' qq-template .
.onom
in other words, it is like an ordinary quote. For instance
.code "^(a b ^(c ,d))"
is equivalent to
.codn "'(a b ^(c ,d))" .
Although there is an unquote ,d it
belongs to the inner quasiquote
.codn "^(c ,d)" ,
and the outer quasiquote does not have
any unquotes of its own, making it equivalent to a quote.
Dialect Note: in Common Lisp and Scheme,
.code ^form
is written
.codn `form ,
and
quasiquotes are also informally known as backquotes. In \*(TX, the backquote
character
.code `
used for quasistring literals.
.meIP >> , expr
The comma character is used within a
.meta qq-template
to denote an unquote. Whereas the quasiquote suppresses evaluation,
similarly to the quote, the comma introduces an exception: an element
of a form which is evaluated. For example, list
.code "^(a b c ,(+ 2 2) (+ 2 2))"
is the list
.codn "(a b c 4 (+ 2 2))" .
Everything
in the quasiquote stands for itself, except for the
.code ",(+ 2 2)"
which is evaluated.
Note: if a variable is called
.codn *x* ,
then the syntax
.code ,*x*
means
.codn ",* x*" :
splice
the value of
.codn x* .
In this situation, whitespace between the comma and the
variable name must be used:
.codn ", *x*" .
.meIP >> ,* expr
The comma-star operator is used within quasiquote list to denote a splicing
unquote. The form which follows
.code ,*
must evaluate to a list. That list is spliced into
the structure which the quasiquote denotes. For example:
.code "'(a b c ,*(list (+ 3 3) (+ 4 4) d))"
evaluates to
.codn "(a b c 6 8 d)" .
The expression
.code "(list (+ 3 3) (+ 4 4))"
is evaluated to produce the list
.codn "(6 8)" ,
and this list is spliced into the quoted template.
.meIP >> @,* expr
This syntax is not a distinct quasiquoting operator, but rather the combination of
an unquote occurring as a meta-expression, denoting the structure
.codn "(sys:expr ,expr)" .
This structure is treated specially by the quasiquote expander. Code is generated
for it such that if
.meta expr
evaluates to an value
.meta val
which is an
.codn atom ,
then the result will be the
.mono
.meti (sys:var << val )
.onom
structure. If
.meta val
is a
.code cons
rather than an
.codn atom ,
then the result is the
.mono
.meti (sys:expr << val )
.onom
structure. In other words, when quasiquoting is used to insert a value under the
.code @
meta prefix, the expander generates code to analyze the type of the value, and
produce to the form which is most likely intended.
.RE
.TP* "Dialect Notes:"
In other Lisp dialects, like Scheme and ANSI Common Lisp, the equivalent syntax
is usually
.code ,@
(comma at). The
.code @
character already has an assigned meaning in \*(TX, so
.code *
is used.
However,
.code *
is also a character that may appear in a symbol name, which creates
a potential for ambiguity. The syntax
.code ,*abc
denotes the application of the
.code ,*
splicing operator to the symbolic expression
.codn abc ;
to apply the ordinary non-splicing unquote to the symbol
.codn *abc ,
whitespace must be used:
.codn ", *abc" .
In \*(TX, the unquoting and splicing forms may freely appear outside of
a quasiquote template. If they are evaluated as forms, however, they
throw an exception:
.verb
,(+ 2 2) ;; error!
',(+ 2 2) --> ,(+ 2 2)
.brev
In other Lisp dialects, a comma not enclosed by backquote syntax is
treated as a syntax error by the reader.
.NP* Quasiquoting non-List Objects
Quasiquoting is supported over hash table and vector literals (see Vectors
and Hashes below). A hash table or vector literal can be quoted, like any
object, for instance:
.verb
'#(1 2 3)
.brev
The
.code "#(1 2 3)"
literal is turned into a vector atom right in the \*(TX parser,
and this atom is being quoted: this is
.mono
.meti (quote << atom )
.onom
syntactically, which evaluates to
.metn atom .
When a vector is quasi-quoted, this is a case of
.mono
.meti >> ^ atom
.onom
which evaluates to
.metn atom .
A vector can be quasiquoted, for example:
.verb
^#(1 2 3)
.brev
Unquotes can occur within a quasiquoted vector:
.verb
(let ((a 42))
^#(1 ,a 3)) ; value is #(1 42 3)
.brev
In this situation, the
.code ^#(...)
notation produces code which constructs a vector.
The vector in the following example is also a quasivector. It contains
unquotes, and though the quasiquote is not directly applied to it,
it is embedded in a quasiquote:
.verb
(let ((a 42))
^(a b c #(d ,a))) ; value is (a b c #(d 42))
.brev
Hash-table literals have two parts: the list of hash construction
arguments and the key-value pairs. For instance:
.verb
#H((:eql-based) (a 1) (b 2))
.brev
where
.code (:eql-based)
indicates that this hash table's keys are treated using
.code eql
equality, and
.code "(a 1)"
and
.code "(b 2)"
are the key/value entries. Hash literals may be quasiquoted. In
quasiquoting, the arguments and pairs are treated as separate syntax; it is not
one big list. So the following is not a possible way to express the above
hash:
.verb
;; not supported: splicing across the entire syntax
(let ((hash-syntax '((:eql-based) (a 1) (b 2))))
^#H(,*hash-syntax))
.brev
This is correct:
.verb
;; fine: splicing hash arguments and contents separately
(let ((hash-args '(:eql-based))
(hash-contents '((a 1) (b 2))))
^#H(,hash-args ,*hash-contents))
.brev
.NP* Quasiquoting combined with Quasiliterals
When a quasiliteral is embedded in a quasiquote, it is possible to use
splicing to insert material into the quasiliteral.
Example:
.verb
(eval (let ((a 3)) ^`abc @,a @{,a} @{(list 1 2 ,a)}`))
-> "abc 3 3 1 2 3"
.brev
.NP* Vector Literals
.coIP "#(...)"
A hash token followed by a list denotes a vector. For example
.code "#(1 2 a)"
is a three-element vector containing the numbers
.code 1
and
.codn 2 ,
and the symbol
.codn a .
.NP* Struct Literals
.meIP >> #S( name >> { slot << value }*)
The notation
.code #S
followed by a nested list syntax denotes a struct literal.
The first item in the syntax is a symbol denoting the struct type
name. This must be the name of a struct type, otherwise the
literal is erroneous. Followed by the struct type are slot names
interleaved with their values. The values are literal expressions,
not subject to evaluation.
Each slot name which is present in the
literal must name a slot in the struct type, though not
all slots in the struct type must be present in the literal.
When a struct literal is read, the denoted struct type is
constructed as if by a call to
.code make-struct
with an empty
.meta plist
argument, followed by a sequence of assignments which store into each
.meta slot
the corresponding
.meta value
expression.
.NP* Hash Literals
.meIP <> #H(( hash-argument *) >> ( key << value )*)
The notation
.code #H
followed by list syntax denotes a hash-table literal.
The first item in the syntax is a list of keywords. These are the same
keywords as are used when calling the function hash to construct
a hash table. Allowed keywords are:
.codn :equal-based ,
.codn :eql-based ,
.codn :eq-based ,
.codn :weak-keys ,
.codn :weak-vals ,
and
.codn :userdata .
If the
.code :userdata
keyword is present,
it must be followed by an object; that object
specifies the hash table's user data, which
can be retrieved using the
.code hash-userdata
function.
The
.codn :equal-based ,
.code :eql-based
and
.code :eq-based
keywords are mutually exclusive.
An empty list can be specified as
.code nil
or
.codn () ,
which defaults to a
hash table based on the
.code eql
function, with no weak semantics or user data.
The entire syntax following
.code #H
may be an empty list; however, that empty list may not
be specified as
.codn nil ;
the empty parentheses notation is required.
The hash table's key-value contents are specified as zero or more
two-element lists, whose first element specifies the
.meta key
and whose second specifies the
.metn value .
Both expressions are literal objects, not subject to evaluation.
.NP* Range Literals
.meIP >> #R( from << to )
The notation
.code #R
followed by a two-element list syntax denotes a range literal.
It combines
.meta from
and
.meta to
expressions, themselves literals not subject to
evaluation, producing the range object whose corresponding
.code to
and
.code from
fields are the objects denoted by these expressions.
.NP* Buffer Literals
.meIP <> #b' hex-data '
The notation
.code #b'
introduces a buffer object: a data representation for a
block of bytes. This
.code #b'
prefix must be followed
by a data section and a closing quote.
The data section consists of hexadecimal digits, among which
may be interspersed whitespace: tabs, spaces and newlines.
There must be an even number of digits, or else the
notation is ill-formed. The whitespace is ignored, and
pairs of successive hex digits specify bytes.
If there are no hex digits, then a zero length buffer
is specified.
Buffers may be constructed by the
.code make-buf
function, and other means such as the
.code ffi-get
function.
Note that the
.code #b
prefix is also used for binary numbers. In that syntax, it
is followed by an optional sign, and then a mixture of one
or more of the digits
.code 0
or
.codn 1 .
.NP* Tree Node Literals
.meIP >> #N([ key >> [ left <> [ right ]]])
The notation
.code #N
followed by list syntax denotes a tree node literal. The list syntax must be a
proper list that has up to three elements. If the list is empty, it may not
be written as
.codn nil .
A tree node is an object of type
.codn tnode .
Every
.code tnode
has three elements: a
.metn key ,
a
.meta left
link and a
.meta right
link. They may be objects of any type.
If the tree node literal syntax omits any of these, they default to
.codn nil .
.NP* Tree Literals
.meIP >> #T([([ keyfun >> [ lessfun <> [ equalfun ]]]) << item *])
The notation
.code #T
followed by list syntax denotes a tree literal, which specifies an
object of type
.codn tree .
Objects of type
.code tree
are search trees.
The list syntax which follows
.code #T
may be empty. If so, it cannot be written as
.codn nil .
The first element of the
.code #T
syntax, if present, must be a list of zero to three elements.
These elements are symbols giving the names of the
.code tree
object's
.IR "key abstraction functions" .
.meta keyfun
specifies the key function which is applied to each element to
retrieve its key. If it is omitted, the object shall use the
.code identity
function as its key.
The
.meta lessfun
specifies the name of the comparison function by which keys are compared
for inequality. It defaults to
.codn less .
The
.meta equalfun
specifies the function by which keys are compared for equality. It
defaults to
.codn equal .
A symbol which is specified as the name of any of these three special
functions must be an element of the list stored in the special variable
.codn *tree-fun-whitelist* ,
otherwise the string literal is diagnosed as erroneous.
Note: this is due to security considerations, since these three
functions are executed during the processing of tree syntax.
A tree object is constructed from a tree literal by first creating an empty
tree endowed with the three key abstraction functions that are indicated in the
syntax, either explicitly or as defaults. Then, every
.meta element
object is constructed from its respective literal syntax and inserted into
the tree.
Duplicate objects are preserved. For instance the tree literal
.code "#T(() 1 1 1)"
specifies a tree with three nodes which have the same key.
Duplicates appear in the tree in the order that they appear in the
literal.
.NP* JSON Literals
.meIP >> #J json-syntax
Introduces a JSON literal.
.meIP >> #J^ json-syntax
Introduces a JSON quasiliteral, allowing unquoting and splicing of Lisp expressions.
The implementation of JSON syntax is based on, and intended to conform with
the IETF RFC 8259 document. Only \*(TX's extensions to JSON syntax are described
in this manual, as well as the correspondence between JSON syntax and Lisp.
The
.meta json-syntax
is translated into a \*(TL object as follows.
A JSON string corresponds to a Lisp string. A JSON number corresponds to a
Lisp floating-point number. A JSON array corresponds to a Lisp vector.
A JSON object corresponds to an
.codn equal -based
hash table.
The JSON Boolean symbols
.code true
and
.code false
translate to the Lisp symbols
.code t
and
.codn nil ,
respectively, those being the standard ones in the
.code usr
package.
The JSON symbol
.code null
maps to the
.code null
symbol in the
.code usr
package.
The
.mono
.meti >> #J json-syntax
.onom
expression produces the object:
.mono
.mets (json quote << lisp-object )
.onom
where
.meta lisp-object
is the Lisp value which corresponds to the
.metn json-syntax .
Similarly, but with a key difference, the
.mono
.meti >> #J^ json-syntax
.onom
expression produces the object:
.mono
.mets (json sys:qquote << lisp-object )
.onom
in which
.code quote
has been replaced with
.codn sys:qquote .
The
.code json
symbol is bound as a macro, which is expanded when a
.code #J
expression is evaluated.
The following remarks indicate special treatment and extensions in the
processing of JSON. Similar remarks regarding the production of JSON are
given under the
.code put-json
function.
When an invalid UTF-8 byte is encountered inside a JSON string, its value is
mapped into the code point range U+DC01 to U+DCFF. That byte is consumed, and
decoding continues with the next byte. This treatment is consistent with the
treatment of invalid UTF-8 bytes in \*(TL literals and I/O streams. If the
valid UTF-8 byte U+0000 (ASCII NUL) occurs in a JSON string, it is also mapped
to U+DC00, \*(TX's pseudo-null character. This treatment is consistent with
\*(TX string literals and I/O streams.
The JSON escape sequence
.code "\eu0000"
denoting the U+0000 NUL character is also converted to U+DC00.
\*(TL does not impose the restriction that the keys in a JSON object
must be strings:
.code "#J{1:2,true:false}"
is accepted.
\*(TL allows the circle notation to occur within JSON syntax. See the section
Notation for Circular and Shared Structure.
\*(TL allows for JSON syntax to be quasiquoted, and provides two extensions
for writing unquotes and splicing unquotes. Within a JSON quasiquote, the
.code ~
(tilde) character introduces a Lisp expression whose value is to be substituted
at that point. Thus, the tilde serves the role of the unquoting comma used
in Lisp quasiquotes. Splicing is indicated by the character sequence
.codn ~* ,
which introduces a Lisp expression that is expected to produce a list, whose
elements are interpolated into the JSON value.
Note: quasiquoting allows Lisp values to be introduced into the resulting
object which are outside of the JSON type system, such as integers, characters,
symbols or structures. These objects have no representation in JSON syntax.
.TP* Examples:
.verb
;; Basic JSON:
#Jtrue -> t
#Jfalse -> nil
(list #J true #Jtrue #Jfalse) -> (t t nil)
#J[1, 2, 3.14] -> #(1.0 2.0 3.14)
#J{"foo":"bar"} -> #H(() ("foo" "bar"))
;; Quoting JSON shows the json expression
'#Jfalse -> (json quote ())
'#Jtrue -> (json quote t)
'#J["a", true, 3.0] -> (json quote #("a" t 3.0))
'#J^[~(+ 2 2), 3] -> (json sys:qquote #(,(+ 2 2) 3.0))
:; Circle notation:
#J[#1="abc", #1#, #1#] -> #("abc" "abc" "abc")
;; JSON Quasiquote:
#J^[~*(list 1.0 2.0 3.0), ~(* 2.0 2), 5.0]
--> #(1.0 2.0 3.0 4.0 5.0)
;; Lisp quasiquote around JSON quote: requires evaluation round.
^#J[~*(list 1.0 2.0 3.0), ~(* 2.0 2), 5.0]
--> (json quote #(1.0 2.0 3.0 4.0 5.0))
(eval ^#J[~*(list 1.0 2.0 3.0), ~(* 2.0 2), 5.0])
--> #(1.0 2.0 3.0 4.0 5.0)
.brev
.coNP The @ .. notation
In \*(TL, there is a special "dotdot" notation consisting of a pair of dots.
This can be written between successive atoms or compound expressions, and is a
shorthand for
.codn rcons .
That is to say,
.code "A .. B"
translates to
.codn "(rcons A B)" ,
and so for instance
.code "(a b .. (c d) e .. f . g)"
means
.codn "(a (rcons b (c d)) (rcons e f) . g)" .
The
.code rcons
function constructs a range object, which denotes a pair of values.
Range objects are most commonly used for referencing subranges of
sequences.
For instance, if
.code L
is a list, then
.code "[L 1 .. 3]"
computes a sublist of
.code L
consisting of
elements 1 through 2 (counting from zero).
Note that if this notation is used in the dot position of an improper
list, the transformation still applies. That is, the syntax
.code "(a . b .. c)"
is valid and produces the object
.code "(a . (rcons b c))"
which is another way of writing
.codn "(a rcons b c)" ,
which is quite probably nonsense.
The notation's
.code ..
operator associates right to left, so that
.code a..b..c
denotes
.codn "(rcons a (rcons b c))" .
Note that range objects are not printed using the dotdot notation.
A range literal has the syntax of a two-element list, prefixed by
.codn #R .
(See Range Literals above.)
In any context where the dotdot notation may be used, and where
it is evaluated to its value, a range literal may also be specified.
If an evaluated dotdot notation specifies two constant expressions, then
an equivalent range literal can replace it. For instance the
form
.code "[L 1 .. 3]"
can also be written
.codn "[L #R(1 3)]" .
The two are syntactically different, and so if these expressions are being
considered for their syntax rather than value, they are not the same.
.NP* The DWIM Brackets
\*(TL has a square bracket notation. The syntax
.code [...]
is a shorthand
way of writing
.codn "(dwim ...)" .
The
.code []
syntax is useful for situations
where the expressive style of a Lisp-1 dialect is useful.
For instance if
.code foo
is a variable which holds a function object, then
.code "[foo 3]"
can be used to call it, instead of
.codn "(call foo 3)" .
If foo is a vector, then
.code "[foo 3]"
retrieves the fourth element, like
.codn "(vecref foo 3)" .
Indexing over lists,
strings and hash tables is possible, and the notation is assignable.
Furthermore, any arguments enclosed in
.code []
which are symbols are treated
according to a modified namespace lookup rule.
More details are given in the documentation for the
.code dwim
operator.
.NP* Compound Forms
In \*(TL, there are two types of compound forms: the Lisp-2 style
compound forms, denoted by ordinary lists that are expressed with parentheses.
There are Lisp-1 style compound forms denoted by the DWIM Brackets, described
in the previous section.
The first position of an ordinary Lisp-2 style compound form, is expected to
have a function or operator name. Then arguments follow. There may
also be an expression in the dotted position, if the form is a function call.
If the form is a function call then the arguments are evaluated. If any of the
arguments are symbols, they are treated according to Lisp-2 namespacing rules.
A function name may be a symbol, or else any of the syntactic forms given in the
description of the function
.codn func-get-name .
.NP* Dot Position in Function Calls
If there is an expression in the dotted position of a function call
expression, it is also evaluated, and the resulting value is involved in the
function call in a special way.
Firstly, note that a compound form cannot be used in the dot position,
for obvious reasons, namely that
.code "(a b c . (foo z))"
does not mean that there is
a compound form in the dot position, but denotes an alternate spelling for
.codn "(a b c foo z)" ,
where foo behaves as a variable.
If the dot position of a compound form is an atom, then the behavior
may be understood according to the following transformations:
.verb
(f a b c ... . x) --> (apply (fun f) a b c ... x)
[f a b c ... . x] --> [apply f a b c ... x]
.brev
In addition to atoms, meta-expressions and meta-symbols can appear in the dot
position, even though their underlying syntax is comprised of a compound
expression. This appears to work according to a transformation pattern
which superficially appears to be the same as that for atoms:
.verb
(f a b c ... . @x) --> (apply (fun f) a b c ... @x)
.brev
However, in this situation, the
.code @x
is actually the form
.code "(sys:var x)"
and the dotted form is actually a proper list. The transformation is
in fact taking place over a proper list, like this:
.verb
(f a b c ... sys:var x) --> (apply (fun f) a b c ... (sys:var @x))
.brev
That is to say, the \*(TL form expander reacts to the presence of a
.code sys:var
or
.code sys:expr
atom in embedded in the form. That symbol and the items which follow it
are wrapped in an additional level of nesting, converted into a single
compound form element.
Effectively, in all these cases, the dot notation constitutes a shorthand for
.codn apply .
Examples:
.verb
;; a contains 3
;; b contains 4
;; c contains #(5 6 7)
;; s contains "xyz"
(foo a b . c) ;; calls (foo 3 4 5 6 7)
(foo a) ;; calls (foo 3)
(foo . s) ;; calls (foo #\ex #\ey #\ez)
(list . a) ;; yields 3
(list a . b) ;; yields (3 . 4)
(list a . c) ;; yields (3 5 6 7)
(list* a c) ;; yields (3 . #(5 6 7))
(cons a . b) ;; error: cons isn't variadic.
(cons a b . c) ;; error: cons requires exactly two arguments.
[foo a b . c] ;; calls (foo 3 4 5 6 7)
[c 1] ;; indexes into vector #(5 6 7) to yield 6
(call (op list 1 . @1) 2) ;; yields (1 . 2)
.brev
Note that the atom in the dot position of a function call may
be a symbol macro. Since the semantics works as if by
transformation to an apply form in which the original dot
position atom is an ordinary argument, the symbol macro
may produce a compound form.
Thus:
.verb
(symacrolet ((x 2))
(list 1 . x)) ;; yields (1 . 2)
(symacrolet ((x (list 1 2)))
(list 1 . x)) ;; yields (1 1 2)
.brev
That is to say, the expansion of
.code x
is not substituted into the form
.code "(list 1 . x)"
but rather the transformation to
.code apply
syntax takes place first, and
so the substitution of
.code x
takes place in a form resembling
.codn "(apply (fun list) 1 x)" .
Dialect Note:
In some other Lisp dialects like ANSI Common Lisp, the improper list syntax may
not be used as a function call; a function called apply (or similar) must be
used for application even if the expression which gives the trailing arguments
is a symbol. Moreover, applying sequences other than lists is not supported.
.NP* Improper Lists as Macro Calls
\*(TL allows macros to be called using forms which are improper lists.
These forms are simply destructured by the usual macro parameter list
destructuring. To be callable this way, the macro must have an argument
list which specifies a parameter match in the dot position. This dot position
must either match the terminating atom of the improper list form,
or else match the trailing portion of the improper list form.
For instance if a macro mac is defined as
.verb
(defmacro mac (a b . c) ...)
.brev
then it may not be invoked as
.code "(mac 1 . 2)"
because the required argument
.code b
is not satisfied, and so the
.code 2
argument cannot match the dot position
.code c
as required. The macro may be called as
.code "(mac 1 2 . 3)"
in which case
.code c
receives the form
.codn 3 .
If it is called as
.code "(mac 1 2 3 . 4)"
then
.code c
receives the improper list form
.codn "3 . 4" .
.NP* Regular-Expression Literals
In \*(TL, the
.code /
character can occur in symbol names, and the
.code /
token
is a symbol. Therefore the
.code /regex/
syntax is not used for denoting regular expressions; rather, the
.code #/regex/
syntax is used.
.NP* Notation for Circular and Shared Structure
\*(TL supports a printed notation called
.I "circle notation"
which accurately articulates
the representation of objects which contain shared substructures as well
as circular references. The notation is supported as a means of
input, and is also optionally produced as output, controlled by the
.code *print-circle*
variable.
Ordinarily, shared substructure in printed objects is not
evident, except in the case of multiple occurrences of interned symbols,
in whose semantics it is implicit that they refer to the same object.
Other shared structure is printed as separate copies which look like
distinct objects. For instance, the object produced by
.code "(let ((shared '(1 2))) (list shared shared))"
is printed as
.codn "((1 2) (1 2))" ,
where it is not clear that the two occurrences of
.code "(1 2)"
are actually the same object. Under the circle notation, this object
can be represented as
.codn "(#5=(1 2) #5#)" .
The
.code #5=
part introduces a reference label, associating the arbitrarily
chosen nonnegative integer 5 with the object which follows.
The subsequent notation
.code #5#
simply refers to the object labeled by 5, reproducing that object
by reference. The result is a two-element list which has the same
.code "(1 2)"
in two places.
Circular structure presents a greater challenge to printing: namely, if it is
printed by a naive recursive descent, it results in infinite output, and
possibly stack exhaustion due to recursion. The circle notation detects
and handles circular references. For instance, the object produced by
.code "(let ((c (list 1))) (rplacd c c))"
produces a circular list which looks like an infinite list of 1's:
.codn "(1 1 1 1 ...)" .
This cannot be printed. However, under the circle notation, it can
be represented as
.codn "#1=(1 . #1#)" .
The entire object itself is labeled by the integer 1. Then, enclosed
within the syntax of that labeled object itself, a reference occurs
to the label. This circular label reference represents the corresponding
circular reference in the object.
A detailed description of the notational elements follows:
.meIP <> # digits = < object
The
.code #=
syntax introduces an object label which denotes the
object whose printed representation follows. The label is identified by
the integer value arising from digits
.meta digits
which are one or more decimal digits. Note: the value zero is permitted;
even though when the notation is produced by the \*(TL printer, labeling
begins at 1. Negative values are not possible because a leading sign
is not part of the syntax.
There may be no more than one definition for a given label within the syntactic
scope being parsed, otherwise a syntax error occurs.
In \*(TX pattern language code,
an entire source file is parsed as one unit, and so scope for the circular
notation's references is the entire source file. Files processed by
.code @(include)
have their own scope. The scope for labels in \*(TL source code is the
top-level expression in which they appear. Consequently, references
in one \*(TL top-level expression cannot reach definitions in another.
.meIP <> # digits #
The
.code ##
syntax denotes a label reference: the repetition of an object that was
previously labeled by the integer given by
.metn digits .
If no such label had been introduced in the syntactic scope,
a syntax error occurs.
An object was previously labeled by
.meta digits
if a
.code #=
definition occurs in the same syntactic scope as the reference,
and is applied to an object which either encloses the reference,
or lexically precedes the reference. Forward references such as
.code "(#1# #1=(1 2))"
are not supported.
.PP
Note:
Circular notation can span hash-table literals. The syntax
.code "#1=#H((:eql-based) (#1# #1#))"
denotes an
.codn eql -based
hash table which contains one entry, in which that
same table itself is both the key and value. This kind of
circularity is not supported for
.codn equal -based
hash tables. The analogous syntax
.code "#1=#H(() (#1# #1#))"
produces a hash table in an inconsistent state.
Dialect Note:
Circle notation is taken from Common Lisp,
intended to be unsurprising to users familiar with that
language.
The implementation is based on descriptions in the ANSI Common Lisp
document, judiciously taking into account the content of the X3J13 Cleanup
Issues named PRINT-CIRCLE-STRUCTURE:USER-FUNCTIONS-WORK and
PRINT-CIRCLE-SHARED:RESPECT-PRINT-CIRCLE.
.NP* Notation for Erasing Objects
.meIP #; < expr
The \*(TL notation
.code #;
in TXR Lisp indicates that the expression
.meta expr
is to be read and then discarded, as if it were replaced by whitespace.
This is useful for temporarily "commenting out" an expression.
.PP
Notes:
Whereas it is valid for a \*(TL source file to be empty, it is
a syntax error if a \*(TL source file contains nothing but one or more
objects which are each suppressed by a preceding
.codn #; .
In the interactive listener, an input line consisting of nothing but
commented-out objects is similarly a syntax error.
The notation does not cascade; consecutive occurrences of
.code #;
trigger a syntax error.
The notation interacts with the circle notation. Firstly, if an object
which is erased by
.code #;
contains circular-referencing instances of the label notation,
those instances refer to
.codn nil .
Secondly, commented-out objects may introduce labels
which are subsequently referenced in
.metn expr .
An example of the first situation occurs in:
.verb
#;(#1=(#1#))
.brev
Here the
.code #1#
label is a circular reference because it refers to an object which
is a parent of the object which contains that reference. Such a reference
is only satisfied by a "backpatching" process once the entire surrounding syntax
is processed to the top level. The erasure perpetrated by
.code #;
causes the
.code #1#
label reference to be replaced by
.codn nil ,
and therefore the labeled object is the object
.codn (nil) .
An example of the second situation is
.verb
#;(#2=(a b c)) #2#
.brev
Here, even though the expression
.code "(#2=(a b c))"
is suppressed, the label definition which it has introduced persists into the
following object, where the label reference
.code #2#
resolves to
.codn "(a b c)" .
A combination of the two situations occurs in
.verb
#;(#1=(#1#)) #1#
.brev
which yields
.codn "(nil)" .
This is because the
.code #1=
label is available; but the earlier
.code #1#
reference, being a circular reference inside an erased object, had lapsed to
.codn nil .
.SS* Generalization of List Accessors
In ancient Lisp in the 1960's, it was not possible to apply the operations
.code car
and
.code cdr
to the
.code nil
symbol (empty list), because it is not a
.code cons
cell. In
the InterLisp dialect, this restriction was lifted: these operations were
extended to accept
.code nil
(and return
.codn nil ).
The convention was adopted in
other Lisp dialects such as MacLisp and eventually in Common Lisp. Thus there
exists an object which is not a cons, yet which takes
.code car
and
.codn cdr .
In \*(TL, this relaxation is extended further. For the sake of convenience,
the operations
.code car
and
.codn cdr ,
are made to work with strings and vectors:
.verb
(cdr "") -> nil
(car "") -> nil
(car "abc") -> #\ea
(cdr "abc") -> "bc"
(cdr #(1 2 3)) -> #(2 3)
(car #(1 2 3)) -> 1
.brev
Moreover, structure types which define the methods
.codn car ,
.code cdr
and
.code nullify
can also be treated in the same way.
The
.code ldiff
function is also extended in a special way. When the right parameter
a non-list sequence, then it uses the equal equality test rather than eq for
detecting the tail of the list.
.verb
(ldiff "abcd" "cd") -> (#\ea #\eb)
.brev
The
.code ldiff
operation starts with
.str "abcd"
and repeatedly applies
.code cdr
to produce
.str "bcd"
and
.strn "cd" ,
until the suffix is equal to the second argument:
.mono
(equal "cd" "cd")
.onom
yields true.
Operations based on
.codn car ,
.code cdr
and
.codn ldiff ,
such as
.code keep-if
and
.code remq
extend to
strings and vectors.
Most derived list processing operations such as
.code remq
or
.code mapcar
obey the following
rule: the returned object follows the type of the leftmost input list object.
For instance, if one or more sequences are processed by
.codn mapcar ,
and the
leftmost one is a character string, the function is expected to return
characters, which are converted to a character string. However, in the
event that the objects produced cannot be assembled into that type of
sequence, a list is returned instead.
For example
.mono
[mapcar list "ab" "12"]
.onom
returns
.codn "((#\ea #\eb) (#\e1 #\e2))" ,
because a string cannot hold lists of characters. However
.mono
[mappend list "ab" "12"]
.onom
returns
.strn "a1b2" .
The lazy versions of these functions such as
.code mapcar*
do not have this behavior;
they produce lazy lists.
.SS* Generalization of Iteration
\*(TL implements a unified paradigm for iterating over sequence-like
container structures and abstract spaces such as bounded and unbounded ranges
of integers. This concept is based around an iterator abstraction which is
directly compatible with Lisp cons-cell traversal in the sense that when
iteration takes place over lists, the iterator instance is nothing but a cons
cell.
An iterator is created using the constructor function
.code iter-begin
which takes a single argument. The argument denotes a space to be traversed;
the iterator provides the means for that traversal.
When the
.code iter-begin
function is applied to a list (a
.code cons
cell or the
.code nil
object), the return value is that object itself. The remaining functions
in the iterator API then behave like aliases for list processing functions.
The
.code iter-more
function behaves like
.codn identity ,
.code iter-item
behaves like
.code car
and
.code iter-step
behaves like
.codn cdr .
For example, the following loops not only produce identical behavior,
but the
.code iter
variable steps through the
.code cons
cells in the same manner in both:
.verb
;; print all symbols in the list (a b c d):
(let ((iter '(a b c d)))
(while iter
(prinl (car iter))
(set iter (cdr iter))))
;; likewise:
(let ((iter (iter-begin '(a b c d))))
(while (iter-more iter)
(prinl (iter-item iter))
(set iter (iter-step iter))))
.brev
There are three important differences.
Firstly, both examples will still work
if the list
.code "(a b c d)"
is replaced by a different kind of sequence, such as the string
.str abcd
or the vector
.codn "#(a b c d)" .
However, the former example will not execute efficiently on these objects.
The reason is that the
.code cdr
function will construct successive suffixes of the string and list object.
That requires not only the allocation of memory, but changes the running time
complexity of the loop from linear to quadratic.
Secondly, the former example with
.cod3 car / cdr
will not work correctly if the sequence is an empty non-list sequence, like
the null string or empty vector. Rectifying this problem requires the
.code nullify
function to be used:
.verb
;; print all symbols in the list (a b c d):
(let ((iter (nullify "abcd")))
(while iter
(prinl (car iter))
(set iter (cdr iter))))
.brev
The
.code nullify
function converts empty sequences of all kinds into the empty list
.codn nil .
Thirdly, the second
example will work even if the input list is replaced with certain objects
which are not sequences at all:
.verb
;; Print the integers from 0 to 3
(let ((iter (iter-begin 0..4)))
(while (iter-more iter)
(prinl (iter-item iter))
(set iter (iter-step iter))))
;; Print incrementing integers starting at 1,
;; breaking out of the loop after 100.
(let ((iter (iter-begin 1)))
(while (iter-more iter)
(if (eql 100 (prinl (iter-item iter)))
(return))
(set iter (iter-step iter))))
.brev
In \*(TL, numerous functions that appear as list processing functions in other
contemporary Lisp dialects, and historically, are actually sequence processing
functions based on the above iterator paradigm.
.SS* Callable Objects
In \*(TL, sequences (strings, vectors and lists) as well as hashes and
regular expressions can be used as functions everywhere, not just with the DWIM
brackets.
Sequences work as one- or two-argument functions. With a single argument, an
element is selected by position and returned. With two arguments, a range is
extracted and returned.
Moreover, when a sequence is used as a function of one argument, and the
argument is a range object rather than an integer, then the call is equivalent
to the two-argument form. This is the basis for array slice syntax like
.mono
["abc" 0..1] .
.onom
Hashes also work as one or two argument functions, corresponding to the
arguments of the gethash function.
A regular expression behaves as a one, two, or three argument function, which
operates on a string argument.
It returns the leftmost matching substring, or else
.codn nil .
.B Example 1:
.verb
(mapcar "abc" '(2 0 1)) -> (#\ec #\ea #\eb)
.brev
Here,
.code mapcar
treats the string
.str abc
as a function of one argument (since there
is one list argument). This function maps the indices
.codn 0 ,
.code 1
and
.code 2
to the
corresponding characters of string
.strn abc .
Through this function, the list of integer indices
.code "(2 0 1)"
is taken to the list of characters
.codn "(#\ec #\ea #\eb)" .
.B Example 2:
.verb
(call '(1 2 3 4) 1..3) -> (2 3)
.brev
Here, the shorthand
.code "1 .. 3"
denotes
.codn "(rcons 1 3)" .
A range used as an argument
to a sequence performs range extraction: taking a slice starting at
index 1, up to and not including index 3, as if by the call
.codn "(sub '(1 2 3 4) 1 3)" .
.B Example 3:
.verb
(call '(1 2 3 4) '(0 2)) -> (1 2)
.brev
A list of indices applied to a sequence is equivalent to using the
select function, as if
.code "(select '(1 2 3 4) '(0 2))"
were called.
.B Example 4:
.verb
(call #/b./ "abcd") -> "bc"
.brev
Here, the regular expression, called as a function, finds the matching
substring
.str bc
within the argument
.strn abcd .
.SS* Special Variables
Similarly to Common Lisp, \*(TL is lexically scoped by default, but
also has dynamically scoped (a.k.a "special") variables.
When a variable is defined with
.code defvar
or
.codn defparm ,
a binding for the symbol is
introduced in the global name space, regardless of in what scope the
.code defvar
form occurs.
Furthermore, at the time the defvar form is evaluated, the symbol which
names the variable is tagged as special.
When a symbol is tagged as special, it behaves differently when it is used
in a lexical binding construct like
.codn let ,
and all other such constructs
such as function parameter lists. Such a binding is not the usual lexical
binding, but a "rebinding" of the global variable. Over the dynamic scope
of the form, the global variable takes on the value given to it by the
rebinding. When the form terminates, the prior value of the variable
is restored. (This is true no matter how the form terminates; even if by
an exception.)
Because of this "pervasive special" behavior of a symbol that has been
used as the name of a global variable, a good practice is to make global
variables have visually distinct names via the "earmuffs" convention:
beginning and ending the name with an asterisk.
.TP* "Example:"
.verb
(defvar *x* 42) ;; *x* has a value of 42
(defun print-x ()
(format t "~a\en" *x*))
(let ((*x* "abc")) ;; this overrides *x*
(print-x)) ;; *x* is now "abc" and so that is printed
(print-x) ;; *x* is 42 again and so "42" is printed
.brev
.TP* "Dialect Note 1:"
The terms
.I bind
and
.I binding
are used differently in \*(TL compared to ANSI Common Lisp.
In \*(TL binding is an association between a symbol and an abstract storage
location. The association is registered in some namespace, such as the global
namespace or a lexical scope. That storage location, in turn, contains a
value. In ANSI Lisp, a binding of a dynamic variable is the association between
the symbol and a value. It is possible for a dynamic variable to exist, and
not have a value. A value can be assigned, which creates a binding.
In \*(TL, an assignment is an operation which transfers a value into
a binding, not one which creates a binding.
In ANSI Lisp, a dynamic variable can exist which has no value. Accessing
the value signals a condition, but storing a value is permitted; doing so
creates a binding. By contrast, in \*(TL a global variable cannot exist without
a value. If a
.code defvar
form doesn't specify a value, and the variable doesn't exist, it is
created with a value of
.codn nil .
.TP* "Dialect Note 2:"
Unlike ANSI Common Lisp, \*(TL has global lexical variables in addition to
special variables. These are defined using
.code defvarl
and
.codn defparml .
The only difference is that when variables are introduced by these macros,
the symbols are not marked special, so their binding in lexical scopes
is not altered to dynamic binding.
Many variables in \*(TL's standard library are global lexicals.
Those which are special variables obey the "earmuffs" convention
in their naming. For instance
.codn s-ifmt ,
.code log-emerg
and
.code sig-hup
are global lexicals, because they provide constant values
for which overriding doesn't make sense. On the other hand the standard
output stream variable
.code *stdout*
is special. Overriding it over a dynamic scope is useful, as a means of
redirecting the output of functions which write to the
.code *stdout*
stream.
.TP* "Dialect Note 3:"
In Common Lisp,
.code defparm
is known as
.codn defparameter .
.SS* Syntactic Places and Accessors
The \*(TL feature known as
.I syntactic places
allows programs to use
the syntax of a form which is used to
.I access
a value from an environment or
object, as an expression which denotes a
.I place
where a value may be
.I stored.
They are almost exactly the
same concept as "generalized references" in Common Lisp, and are related to
"lvalues" in languages in the C family, or "designators" in Pascal.
.NP* Symbolic Places
A symbol is a is a syntactic place if it names a variable. If
.code a
is a variable, then it may be assigned using the
.code set
operator: the form
.code "(set a 42)"
causes
.code a
to have the integer value 42.
.NP* Compound Places
A compound expression can be a syntactic place, if its leftmost constituent is
as symbol which is specially registered, and if the form has the correct syntax
for that kind of place, and suitable semantics. Such an expression is a compound
place.
An example of a compound place is a
.code car
form. If
.code c
is an expression denoting a
.code cons
cell, then
.code "(car c)"
is not only an expression which retrieves the value of the
.code car
field of the cell. It is also a syntactic place which denotes that field as
a storage location. Consequently, the expression
.mono
(set (car c) "abc")
.onom
stores the character string
.str "abc"
in that location. Although the same effect can be obtained with
.mono
(rplaca c "abc")
.onom
the syntactic place frees the programmer from having to remember
different update functions for different kinds of places.
There are
various other advantages. \*(TL provides a plethora of operators
for modifying a place in addition to
.codn set .
Subject to certain usage restrictions, these operators work uniformly on all
places. For instance, the expression
.code "(rotate (car x) [str 3] y)"
causes three different kinds of places to exchange contents,
while the three expressions denoting those places
are evaluated only once. New kinds of place update macros like
.code rotate
are quite easily defined, as are new kinds of compound places.
.NP* Accessor Functions
When a function call form such as the above
.code "(car x)"
is a syntactic place, then the function is called an
.IR accessor .
This term is used throughout this document to denote functions
which have associated syntactic places.
.NP* Macro Call Syntactic Places
Syntactic places can be macros (global and lexical), including symbol macros.
So for instance in
.code "(set x 42)"
the
.code x
place can actually be a symbolic macro which expands to, say,
.codn "(cdr y)" .
This means that the assignment is effectively
.codn "(set (cdr y) 42)" .
.NP* User-Defined Syntactic Places and Place Operators
Syntactic places, as well as operators upon syntactic places,
are both open-ended. Code can be written quite easily in \*(TL to introduce
new kinds of places, as well as new place-mutating operators.
New places can be introduced with the help of the
.codn defplace ,
.code define-accessor
or
.code defset
macros, or possibly the
.code define-place-macro
macro in simple cases when a new syntactic place can be expressed as a
transformation to the syntax of an existing place.
Three ways exist for developing new place update macros (place operators).
They can be written using the ordinary macro definer
ordinary macro definer
.codn defmacro ,
with the help of special utility macros called
.codn with-update-expander ,
.codn with-clobber-expander ,
and
.codn with-delete-expander .
They can also be written using
.code defmacro
in conjunction with the operators
.code placelet
or
.codn placelet* .
Simple update macros similar to
.code inc
and
.code push
can be written compactly using
.codn define-modify-macro .
.NP* Deletable Places
Unlike generalized references in Common Lisp, \*(TL syntactic
places support the concept of deletion. Some kinds of places
can be deleted, which is an action distinct from (but does not preclude) being
overwritten with a value. What exactly it means for a place to be deleted,
or whether that is even permitted, depends on the kind of place.
For instance a place which denotes a lexical variable may not be deleted,
whereas a global variable may be.
A place which denotes a hash-table entry may be deleted, and results in the
entry being removed from the hash table. Deleting a place in a list
causes the trailing items, if any, or else the terminating atom, to
move in to close the gap. Users may define new kinds of places
which support deletion semantics.
.NP* Evaluation of Places
To bring about their effect, place operators must evaluate one or
more places. Moreover, some of them evaluate additional forms which are not
places. Which arguments of a place operator form are places and which are
ordinary forms depends on its specific syntax. For all the built-in place
operators, the position of an argument in the syntax determines whether it is
treated as (and consequently required to be) a syntactic place, or whether it is
an ordinary form.
All built-in place operators perform the evaluation of place and non-place
argument forms in strict left-to-right order.
Place forms are evaluated not in order to compute a value, but in order to
determine the storage location. In addition to determining a storage location,
the evaluation of a place form may possibly give rise to side effects.
Once a place is fully evaluated, the storage location can then be accessed.
Access to the storage location is not considered part of the evaluation of a
place. To determine a storage location means to compute some hidden referential
object which provides subsequent access to that location without the need for a
reevaluation of the original place form. (The subsequent access to the
place through this referential object may still require a multi-step traversal
of a data structure; minimizing such steps is a matter of optimization.)
Place forms may themselves be compounds, which contain subexpressions that must
be evaluated. All such evaluation for the built-in places takes place in left
to right order.
Certain place operators, such as
.code shift
and
.codn rotate ,
exhibit an unspecified behavior with regard to the timing of the access
of the prior value of a place, relative to the evaluation of places
which occur later in the same place operator form. Access to the prior values
may be delayed until the entire form is evaluated, or it may be interleaved
into the evaluation of the form. For example, in the form
.codn "(shift a b c 1)" ,
the prior value of
.code a
can be accessed and saved as soon as
.code a
is evaluated, prior to the evaluation of
.codn b .
Alternatively,
.code a
may be accessed and saved later, after the evaluation of
.code b
or after the evaluation of all the forms. This issue affects the behavior of
place-modifying forms whose subforms contain side effects. It is recommended
that such forms not be used in programs.
.NP* Nested Places
Certain place forms are required to have one or more arguments which
are themselves places. The prime example of this, and the only example from
among built-in syntactic places, are DWIM forms. A DWIM form has the syntax
.mono
.mets (dwim < obj-place < index <> [ alt ])
.onom
and the square-bracket-notation equivalent:
.mono
.mets >> [ obj-place < index <> [ alt ]]
.onom
Note that not only is the entire form a place, denoting some element or element
range of
.metn obj-place ,
but there is the added constraint that
.meta obj-place
must also itself be a syntactic place.
This requirement is necessary, because it supports the behavior that
when the element or element range is updated, then
.meta obj-place
is also potentially updated.
After the assignment
.mono
(set [obj 0..3] '("forty" "two"))
.onom
not only is the range of places denoted by
.code "[obj 0..3]"
replaced by the list of strings
.mono
("forty" "two")
.onom
but
.code obj
may also be overwritten with a new value.
This behavior is necessary because the DWIM brackets notation maintains
the illusion of an encapsulated array-like container over several dissimilar
types, including Lisp lists. But Lisp lists do not behave as fully
encapsulated containers. Some mutations on Lisp lists return new objects,
which then have to stored (or otherwise accepted) in place of the original
objects in order to maintain the array-like container illusion.
.NP* Built-In Syntactic Places
The following is a summary of the built-in place forms, in addition to symbolic
places denoting variables. New syntactic place forms can be
defined by \*(TX programs.
.mono
.mets (car << object )
.mets (first << object )
.mets (rest << object )
.mets (second << object )
.mets (third << object )
.mets ...
.mets (tenth << object )
.mets (last < object <> [ num ])
.mets (butlast < object <> [ num ])
.mets (cdr << object )
.mets (caar << object )
.mets (cadr << object )
.mets (cdar << object )
.mets (cddr << object )
.mets ...
.mets (cdddddr << object )
.mets (nthcdr < index << obj )
.mets (nthlast < index << obj )
.mets (butlastn < num << obj )
.mets (last < num << obj )
.mets (nth < index << obj )
.mets (ref < seq << idx )
.mets (sub < sequence >> [ from <> [ to ]])
.mets (vecref < vec << idx )
.mets (chr-str < str << idx )
.mets (gethash < hash < key <> [ alt ])
.mets (hash-userdata << hash )
.mets (dwim < obj-place < index <> [ alt ])
.mets (sub-list < obj >> [ from <> [ to ]])
.mets (sub-vec < obj >> [ from <> [ to ]])
.mets (sub-str < str >> [ from <> [ to ]])
.mets >> [ obj-place < index <> [ alt ]] ;; equivalent to dwim
.mets (symbol-value << symbol-valued-form )
.mets (symbol-function << function-name-valued-form )
.mets (symbol-macro << symbol-valued-form )
.mets (fun << function-name )
.mets (force << promise )
.mets (errno)
.mets (slot < struct-obj << slot-name-valued-form )
.mets (qref < struct-obj << slot-name ) ;; by macro-expansion to (slot ...)
.mets >< struct-obj . slot-name ;; equivalent to qref
.mets (sock-peer << socket )
.mets (sock-opt < socket < level < option <> [ ffi-type ])
.mets (carray-sub < carray >> [ from <> [ to ]])
.mets (sub-buf < buf >> [ from <> [ to ]])
.mets (left << node )
.mets (right << node )
.mets (key << node )
.mets (read-once << node )
.onom
.NP* Built-In Place-Mutating Operators
The following is a summary of the built-in place mutating macros.
They are described in detail in their own sections.
.meIP (set >> { place << new-value }*)
Assigns the values of expressions to places, performing assignments in left-to-right order,
returning the value assigned to the rightmost place.
.meIP (pset >> { place << new-value }*)
Assigns the values of expressions to places, performing the determination of
places and evaluation of the expressions left to right, but the assignment
in parallel. Returns the value assigned to the rightmost place.
.meIP (zap < place <> [ new-value ])
Assigns
.meta new-value
to place, defaulting to
.codn nil ,
and returns the prior value.
.meIP (flip << place )
Logically toggles the Boolean value of
.metn place ,
and returns the new value.
.meIP (test-set << place )
If
.meta place
contains
.codn nil ,
stores
.code t
into the place and returns
.code t
to indicate that the store took place.
Otherwise does nothing and returns
.codn nil .
.meIP (test-clear << place )
If
.meta place
contains a Boolean true value, stores
.code nil
into the place and returns
.code t
to indicate that the store took place.
Otherwise does nothing and returns
.codn nil .
.meIP (compare-swap < place < cmp-fun < cmp-val << store-val )
Examines the value of
.meta place
and compares it to
.meta cmp-val
using the comparison function given by the function name
.metn cmp-fun .
If the comparison is false, returns
.codn nil .
Otherwise, stores the
.meta store-val
value into
.meta place
and returns
.codn t .
.meIP (inc < place <> [ delta ])
Increments
.meta place
by
.metn delta ,
which defaults to 1, and returns the new value.
.meIP (dec < place <> [ delta ])
Decrements
.meta place
by
.metn delta ,
which defaults to 1, and returns the new value.
.meIP (pinc < place <> [ delta ])
Increments
.meta place
by
.metn delta ,
which defaults to 1, and returns the old value.
.meIP (pdec < place <> [ delta ])
Decrements
.meta place
by
.metn delta ,
which defaults to 1, and returns the old value.
.meIP (test-inc < place >> [ delta <> [ from-val ]])
Increments
.meta place
by
.meta delta
and returns
.code t
if the previous value was
.code eql
to
.metn from-val ,
where
.meta delta
defaults to 1
and
.meta from-val
defaults to zero.
.meIP (test-dec < place >> [ delta <> [ to-val ]])
Decrements
.meta place
by
.meta delta
and returns
.code t
if the new value is
.code eql
to
.metn to-val ,
where
.meta delta
defaults to 1
and
.meta to-val
defaults to 0.
.meIP (swap < left-place << right-place )
Exchanges the values of
.meta left-place
and
.metn right-place .
.meIP (push < item << place )
Adds
.meta item
to the front of the list which is currently stored in
.codn place ,
then stores the extended list back into
.code place
and returns it.
.meIP (pop << place )
Pop the list stored in
.meta place
and returns the popped value.
.meIP (shift << place + << shift-in-value)
Treats one or more places as a "multi-place shift register".
Values are shifted to the left among the places. The
rightmost place receives
.metn shift-in-value ,
and the value of the leftmost place emerges as the return value.
.meIP (rotate << place *)
Treats zero or more places as a "multi-place rotate register".
The places exchange values among themselves, by a rotation
by one place to the left. The value of the leftmost place
goes to the rightmost place, and that value is returned.
.meIP (del << place )
Deletes a place which supports deletion, and returns
the value which existed in that place prior to deletion.
.meIP (lset <> { place }+ << sequence )
Sets multiple places to values obtained from successive
elements of
.metn sequence .
.meIP (upd < place << opip-arg *)
Applies an
.codn opip -style
operational pipeline to the value of
.meta place
and stores the result back into
.metn place .
.meIP (set-mask < place << integer *)
Sets to 1 the bits in
.meta place
corresponding to bits that are equal to 1 in the mask made up of the
.meta integer
arguments (by combining them together with the inclusive or operation).
.meIP (clear-mask < place << integer *)
Clears (sets to 0) the bits in
.meta place
corresponding to bits that are equal to 1 in the mask made up of the
.meta integer
arguments (by combining them together with the inclusive or operation).
.PP
.SS* Namespaces and Environments
\*(TL is a Lisp-2 dialect: it features separate namespaces for
functions and variables.
.NP* Global Functions and Operator Macros
In \*(TL, global functions and operator macros coexist, meaning that the same
symbol can be defined as both a macro and a function.
There is a global namespace for functions,
into which functions can be introduced with the
.code defun
macro. The global function environment can be inspected and modified using the
.code symbol-function
accessor.
There is a global namespace for macros, into which
macros are introduced with the
.code defmacro
macro. The global function environment can be inspected and modified using the
.code symbol-macro
accessor.
If a name
.code x
is defined as both a function and a macro, then an expression of the form
.code "(x ...)"
is expanded by the macro, whereas an expression of the form
.code "[x ...]"
refers to the function. Moreover, the macro can produce a call to the
function. The expression
.code "(fun x)"
will retrieve the function object.
.NP* Global and Dynamic Variables
There is a global namespace for variables also.
The operators
.code defvar
and
.code defparm
introduce bindings into this namespace. These operators have the
side effect of marking a symbol as a special variable,
of the symbol are treated as dynamic variables, subject to
rebinding. The global variable namespace together with the special dynamic
rebinding is called the dynamic environment.
The dynamic environment can be inspected and modified using the
.code symbol-value
accessor.
The operators
.code defvarl
and
.code defparml
introduce bindings into the global namespace without marking
symbols as special variables. Such bindings are called global lexical
variables.
.NP* Global Symbol Macros
Symbol macros may be defined over the global variable namespace
using
.codn defsymacro .
Note that whereas a symbol may simultaneously have both a function and macro
binding in the global namespace, a symbol may not simultaneously have
a variable and symbol macro binding.
.NP* Lexical Environments
In addition to global and dynamic namespaces, \*(TL provides lexically scoped
binding for functions, variables, macros, and symbol macros.
Lexical variable binding are introduced with
.codn let ,
.code let*
or various binding macros derived from these. Lexical functions are bound
with
.code flet
and
.codn labels .
Lexical macros are established with
.code macrolet
and lexical symbol macros with
.codn symacrolet .
Macros receive an environment parameter with which they may expand
forms in their correct environment, and perform some limited introspection
over that environment in order to determine the nature of bindings,
or the classification of forms in those environments. This introspection
is provided by
.codn lexical-var-p ,
.codn lexical-fun-p ,
and
.codn lexical-lisp1-binding .
Lexical operator macros and lexical functions can also coexist in the
following way. A lexical function shadows a global or lexical macro
completely. However, the reverse is not the case. A lexical macro shadows
only those uses of a function which look like macro calls. This is
succinctly demonstrated by the following form:
.verb
(flet ((foo () 43))
(macrolet ((foo () 44))
(list (fun foo) (foo) [foo])))
-> (#<interpreted fun: lambda nil> 44 43)
.brev
The
.code "(fun foo)"
and
.code [fun]
expressions are oblivious to the macro; the macro expansion process
process the symbol
.code foo
in those contexts. However the form
.code (foo)
is subject to macro-expansion and replaced with
.codn 44 .
If the
.code flet
and
.code macrolet
are reversed, the behavior is different:
.verb
(macrolet ((foo () 44))
(flet ((foo () 43))
(list (fun foo) (foo) [foo])))
-> (#<interpreted fun: lambda nil> 43 43)
.brev
All three forms refer to the function, which lexically shadows the macro.
.NP* Pattern Language and Lisp Scope Nesting
\*(TL expressions can be embedded in the \*(TX pattern language in various
ways. Likewise, the pattern language can be invoked from \*(TL. This
brings about the possibility that Lisp code attempts to access
pattern variables bound in the pattern language. The \*(TX pattern language
can also attempt to access \*(TL variables.
The rules are as follows, but they have undergone historic changes. See the
COMPATIBILITY section, in particular notes under 138 and 121, and also 124.
A Lisp expression evaluated from the \*(TX pattern language executes
in a null lexical environment. The current set of pattern variables captured
up to that point by the pattern language are installed as dynamic variables.
They shadow any Lisp global variables (whether those are defined
by
.code defvar
or
.codn defvarl ).
In the reverse direction, a variable reference from the \*(TX pattern
language searches the pattern variable space first. If a variable doesn't
exist there, then the lookup refers to the \*(TL global variable space.
The pattern language doesn't see Lisp lexical variables.
When Lisp code is evaluated from the pattern language, the pattern variable
bindings are not only installed as dynamic variables for the sake of their
visibility from Lisp, but they are also specially stored in a dynamic
environment frame. When \*(TX pattern code is reentered from Lisp, these
bindings are picked up from the closest such environment frame, allowing the
nested invocation of pattern code to continue with the bindings captured by
outer pattern code.
Concisely, in any context in which a symbol has both a binding as a Lisp global
variable as well as a pattern variable, that symbol refers to the pattern
variable. Pattern variables are propagated through Lisp evaluation into
nested invocations of the pattern language.
The pattern language can also reference
Lisp variables using the
.code @
prefix, which is a consequence of that prefix introducing an expression that is
evaluated as Lisp, the name of a variable being such an expression.
.SH* LISP OPERATOR, FUNCTION AND MACRO REFERENCE
.SS* Conventions
The following sections list all of the special operators, macros
and functions in \*(TL.
In these sections, syntax is indicated using these conventions:
.meIP < word
.ie n \{\
A symbol in angle brackets
.\}
.el \{\
A symbol in
.meta fixed-width-italic
font
.\}
denotes some syntactic unit: it
may be a symbol or compound form. The syntactic unit is explained
in the corresponding Description section.
.meIP {syntax}* << word *
This indicates a repetition of zero or more of the given
syntax enclosed in the braces or syntactic unit.
The curly braces may be omitted if the scope of the
.code *
is clear.
.meIP {syntax}+ << word +
This indicates a repetition of one or more of the given
syntax enclosed in the braces or syntactic unit.
The curly braces may be omitted if the scope of the
.code +
is clear.
.coIP {syntax | syntax | ...}
This indicates a single, mandatory element, which is selected
from among the indicated alternatives.
May be combined with
.code +
or
.code *
repetition.
.meIP [syntax] <> [ word ]
Square brackets indicate optional syntax.
.meIP [syntax | syntax | ...]
Square brackets containing piped elements indicate an optional
element, which, if present, must be chosen from among the indicated
alternatives.
.coIP '[' ']'
The quoted square brackets indicate literal brackets which appear
in the syntax, which they do without quotes. For instance
.code "'['foo [ bar ]']'"
is a pattern denotes the two possible expressions
.code "[foo]"
and
.codn "[foo bar]" .
.meIP syntax -> < result
The arrow notation is used in examples to indicate that the evaluation
of the given syntax produces a value, whose printed representation is
.metn result .
.SS* Form Evaluation
A compound expression with a symbol as its first element, if
intended to be evaluated, denotes either an operator invocation or a function
call. This depends on whether the symbol names an operator or a function.
When the form is an operator invocation, the interpretation of the meaning of
that form is under the complete control of that operator.
If the compound form is a function call, the remaining forms, if any, denote
argument expressions to the function. They are evaluated in left-to-right
order to produce the argument values, which are passed to the function. An
exception is thrown if there are not enough arguments, or too many. Programs
can define named functions with the defun operator
Some operators are macros. There exist predefined macros in the library, and
macro operators can also be user-defined using the macro-defining operator
.codn defmacro .
Operators that are not macros are called special operators.
Macro operators work as functions which are given the source code of the form.
They analyze the form, and translate it to another form which is substituted in
their place. This happens during a code walking phase called the expansion
phase, which is applied to each top-level expression prior to evaluation. All
macros occurring in a form are expanded in the expansion phase, and subsequent
evaluation takes place on a structure which is devoid of macros. All that
remains are the executable forms of special operators, function calls,
symbols denoting either variables or themselves, and atoms such as numeric
and string literals.
Special operators can also perform code transformations during the expansion
phase, but that is not considered macroexpansion, but rather an adjustment
of the representation of the operator into an required executable form.
In effect, it is post-macro compilation phase.
Note that Lisp forms occurring in \*(TX pattern language are not individual
top-level forms. Rather, the entire \*(TX query is parsed at the same time, and
the macros occurring in its Lisp forms are expanded at that time.
.coNP Operator @ quote
.synb
.mets (quote << form )
.syne
.desc
The
.code quote
operator, when evaluated, suppresses the evaluation of
.metn form ,
and instead returns
.meta form
itself as an object. For example, if
.meta form
is a symbol
.metn sym ,
then the value of
.mono
.meti (quote << sym )
.onom
is
.meta sym
itself. Without
.codn quote ,
.meta sym
would evaluate to the value held by the variable which is named
.metn sym ,
or else throw an error if there is no such variable.
The
.code quote
operator never raises an error, if it is given exactly one argument,
as required.
The notation
.mono
.meti >> ' obj
.onom
is translated to the object
.mono
.meti (quote << obj )
.onom
providing a shorthand for quoting. Likewise, when an object of the form
.mono
.meti (quote << obj )
.onom
is printed, it appears as
.codn 'obj .
.TP* Example:
.verb
;; yields symbol a itself, not value of variable a
(quote a) -> a
;; yields three-element list (+ 2 2), not 4.
(quote (+ 2 2)) -> (+ 2 2)
.brev
.SS* Variable Binding
Variables are associations between symbols and storage locations
which hold values. These associations are called
.IR bindings .
Bindings are held in a context called an
.IR environment .
.I Lexical
environments hold local variables, and nest according to the syntactic
structure of the program. Lexical bindings are always introduced by a
some form known as a
.IR "binding construct" ,
and the corresponding environment is instantiated during the evaluation
of that construct. There also exist bindings outside of any binding
construct, in the so-called
.I global environment .
Bindings in the global environment can be temporarily shadowed by
lexically-established binding in the
.I dynamic environment .
See the Special Variables section above.
Certain special symbols cannot be used as variable names, namely the
symbols
.code t
and
.codn nil ,
and all of the keyword symbols (symbols in the keyword package), which are
denoted by a leading colon. When any of these symbols is evaluated as
a form, the resulting value is that symbol itself. It is said that these
special symbols are self-evaluating or self-quoting, similarly to all
other atom objects such as numbers or strings.
When a form consisting of a symbol, other than the above special symbols, is
evaluated, it is treated as a variable, and yields the value of
the variable's storage location. If the variable doesn't exist,
an exception is thrown.
Note: symbol forms may also denote invocations of symbol macros. (See the
operators
.code defsymacro
and
.codn symacrolet ).
All macros, including symbol macros, which occur inside
a form are fully expanded prior to the evaluation of a form, therefore
evaluation does not consider the possibility of a symbol being
a symbol macro.
.coNP Operator @ defvar and macro @ defparm
.synb
.mets (defvar < sym <> [ value ])
.mets (defparm < sym << value )
.syne
.desc
The
.code defvar
operator binds a name in the variable namespace of the global environment.
Binding a name means creating a binding: recording, in some namespace of some
environment, an association between a name and some named entity. In the
case of a variable binding, that entity is a storage location for a value.
The value of a variable is that which has most recently been written into the
storage location, and is also said to be a value of the binding, or stored
in the binding.
If the variable named
.meta sym
already exists in the global environment, the
form has no effect; the
.meta value
form is not evaluated, and the value of the
variable is unchanged.
If the variable does not exist, then a new binding is introduced, with a value
given by evaluating the
.meta value
form. If the form is absent, the variable is initialized
to
.codn nil .
The
.meta value
form is evaluated in the environment
in which the
.code defvar
form occurs, not necessarily in the global environment.
The symbols
.code t
and
.code nil
may not be used as variables,
nor can they be keyword symbols (symbols denoted by a leading colon).
In addition to creating a binding, the
.code defvar
operator also marks
.meta sym
as the name of a special variable. This changes what it means to bind
that symbol in a lexical binding construct such as the
.code let
operator, or a function parameter list. See the section "Special Variables" far
above.
The
.code defparm
macro behaves like
.code defvar
when a variable named
.meta sym
doesn't already exist.
If
.meta sym
already denotes a variable binding in the global namespace,
.code defparm
evaluates the
.meta value
form and assigns the resulting value to the variable.
The following equivalence holds:
.verb
(defparm x y) <--> (prog1 (defvar x) (set x y))
.brev
The
.code defvar
and
.code defparm
forms return
.metn sym .
.coNP Macros @ defvarl and @ defparml
.synb
.mets (defvarl < sym <> [ value ])
.mets (defparml < sym << value )
.syne
.desc
The
.code defvarl
and
.code defparml
macros behave, respectively, almost exactly like
.code defvar
and
.codn defparm .
The difference is that these operators do not mark
.meta sym
as special.
If a global variable
.meta sym
does not previously exist, then after the evaluation of
either of these forms
.mono
.meti (boundp << sym )
.onom
is true, but
.mono
.meti (special-var-p << sym )
.onom
isn't.
If
.meta sym
had been already introduced as a special variable, it stays that way
after the evaluation of
.code defvarl
or
.codn defparml .
.coNP Operators @ let and @ let*
.synb
.mets (let >> ({ sym | >> ( sym << init-form )}*) << body-form *)
.mets (let* >> ({ sym | >> ( sym << init-form )}*) << body-form *)
.syne
.desc
The
.code let
and
.code let*
operators introduce a new scope with variables and
evaluate forms in that scope. The operator symbol, either
.code let
or
.codn let* ,
is followed by a list which can contain any mixture of
.meta sym
or
.mono
.meti >> ( sym << init-form )
.onom
pairs.
Each
.meta sym
must be a symbol, and specifies the name of variable to be instantiated and
initialized.
The
.mono
.meti >> ( sym << init-form )
.onom
variant specifies that the new variable
.meta sym
receives an initial value from the
evaluation of
.metn init-form .
The plain
.meta sym
variant specifies a variable which is initialized to
.codn nil .
The
.metn init-form s
are evaluated in order, by both
.code let
and
.codn let* .
The symbols
.code t
and
.code nil
may not be used as variables, and neither
can be keyword symbols: symbols denoted by a leading colon.
The difference between
.code let
and
.code let*
is that in
.codn let* ,
later
.codn init-form s
are in scope of the variables established by earlier variables in the same
.code let*
construct. In plain
.codn let ,
the
.metn init-form s
are evaluated in a scope which does not include any of the variables.
When the variables are established, the
.metn body-form s
are evaluated in order. The value of the last
.meta body-form
becomes the return value of the
.codn let .
If there are no
.metn body-form s,
then the return value
.code nil
is produced.
The list of variables may be empty.
The list of variables may contain duplicate
.metn sym s
if the operator is
.codn let* .
In that situation, a given
.meta init-form
has in scope the rightmost duplicate of any given
.meta sym
that has been previously established.
The
.metn body-form s
have in scope the rightmost duplicate of any
.meta sym
in the construct.
Therefore, the following form calculates the value 3:
.verb
(let* ((a 1)
(a (succ a))
(a (succ a)))
a)
.brev
Each duplicate is a separately instantiated binding, and may be independently
captured by a lexical closure placed in a subsequent
.codn init-form :
.verb
(let* ((a 0)
(f1 (lambda () (inc a)))
(a 0)
(f2 (lambda () (inc a))))
(list [f1] [f1] [f1] [f2] [f2] [f2]))
--> (1 2 3 1 2 3)
.brev
The preceding example shows that there are two mutable variables named
.code a
in independent scopes, each respectively captured by the separate closures
.code f1
and
.codn f2 .
Three calls to
.code f1
increment the first
.code a
while the second
.code a
retains its initial value.
Under
.codn let ,
the behavior of duplicate variables is unspecified.
Implementation note: the \*(TX compiler diagnoses and rejects duplicate
symbols in
.code let
whereas the interpreter ignores the situation.
When the names of a special variables is specified in
.code let
or
.code let*
remain, a new binding is created for them in the dynamic environment, rather
than the lexical environment.
In
.codn let* ,
later
.metn init-form s
are evaluated in a dynamic scope in which previous dynamic variables
are established, and later dynamic variables are not yet established.
A special variable may appear multiple times in a
.codn let* ,
just like a lexical variable. Each duplicate occurrence extends the
dynamic environment with a new dynamic binding.
All these dynamic environments are removed when the
.code let
or
.code let*
form terminates. Dynamic environments aren't captured by lexical
closures, but are captured in delimited continuations.
.TP* Examples:
.verb
(let ((a 1) (b 2)) (list a b)) -> (1 2)
(let* ((a 1) (b (+ a 1))) (list a b (+ a b))) -> (1 2 3)
(let ()) -> nil
(let (:a nil)) -> error, :a and nil can't be used as variables
.brev
.SS* Functions
.coNP Operator @ defun
.synb
.mets (defun < name <> ( param * [: << opt-param *] [. << rest-param ])
.mets \ \ << body-form )
.syne
.desc
The
.code defun
operator introduces a new function in the global function namespace.
The function is similar to a lambda, and has the same parameter syntax
and semantics as the
.code lambda
operator.
Note that the above syntax synopsis describes only the canonical
parameter syntax which remains after parameter list macros are
expanded. See the section Parameter List Macros.
Unlike in
.codn lambda ,
the
.metn body-form s
of a
.code defun
are surrounded by a block.
The name of this block is the same as the name of the function, making it
possible to terminate the function and return a value using
.mono
.meti (return-from < name << value ).
.onom
For more information, see the definition of the block operator.
A function may call itself by name, allowing for recursion.
The special symbols
.code t
and
.code nil
may not be used as function names. Neither can keyword symbols.
It is possible to define methods as well as macros with
.codn defun ,
as an alternative to the
.code defmeth
and
.code defmacro
forms.
To define a method, the syntax
.mono
.meti (meth < type << name )
.onom
should be used as the argument to the
.meta name
parameter. This gives rise to the syntax
.mono
.meti (defun (meth < type << name ) < args << form *)
.onom
which is equivalent to the
.mono
.meti (defmeth < type < name < args << form *)
.onom
syntax.
Macros can be defined using
.mono
.meti (macro << name )
.onom
as the
.meta name
parameter of
.codn defun .
This way of defining a macro doesn't support destructuring;
it defines the expander as an ordinary function with an ordinary
argument list. To work, the function must accept two arguments:
the entire macro call form that is to be expanded, and the
macro environment. Thus, the macro definition syntax is
.mono
.meti (defun (macro << name ) < form < env << form *)
.onom
which is equivalent to the
.mono
.meti (defmacro < name (:form < form :env << env ) << form *)
.onom
syntax.
.TP* "Dialect Note:"
In ANSI Common Lisp, keywords may be used as function names.
In TXR Lisp, they may not.
.TP* "Dialect Note:"
A function defined by
.code defun
may coexist with a macro defined by
.codn defmacro .
This is not permitted in ANSI Common Lisp.
.coNP Operator @ lambda
.synb
.mets (lambda <> ( param * [: << opt-param *] [. << rest-param ])
.mets \ \ << body-form )
.mets (lambda < rest-param
.mets \ \ << body-form )
.syne
.desc
The
.code lambda
operator produces a value which is a function. Like in most other
Lisps, functions are objects in \*(TL. They can be passed to functions as
arguments, returned from functions, aggregated into lists, stored in variables,
etc.
Note that the above syntax synopsis describes only the canonical
parameter syntax which remains after parameter list macros are
expanded. See the section Parameter List Macros.
The first argument of
.code lambda
is the list of parameters for the function. It
may be empty, and it may also be an improper list (dot notation) where the
terminating atom is a symbol other than
.codn nil .
It can also be a single symbol.
The second and subsequent arguments are the forms making up the function body.
The body may be empty.
When a function is called, the parameters are instantiated as variables that
are visible to the body forms. The variables are initialized from the values of
the argument expressions appearing in the function call.
The dotted notation can be used to write a function that accepts
a variable number of arguments. There are two ways write a function that
accepts only a variable argument list and no required arguments:
.mono
.mets (lambda (. << rest-param ) ...)
.mets (lambda < rest-param ...)
.onom
(These notations are syntactically equivalent because the list notation
.code "(. X)"
actually denotes the object
.meta X
which isn't wrapped in any list).
The keyword symbol
.code :
(colon) can appear in the parameter list. This is
the symbol in the keyword package whose name is the empty string. This
symbol is treated specially: it serves as a separator between
required parameters and optional parameters. Furthermore, the
.code :
symbol has a role to play in function calls: it can be specified as an argument
value to an optional parameter by which the caller indicates that the
optional argument is not being specified. It will be processed exactly
that way.
An optional parameter can also be written in the form
.mono
.meti >> ( name < expr <> [ sym ]).
.onom
In this situation, if the call does not specify a value for the parameter,
or specifies a value as the
.code :
(colon) keyword symbol, then the parameter takes on the
value of the expression
.metn expr .
This expression is only evaluated when its value is required.
If
.meta sym
is specified, then
.meta sym
will be
introduced as an additional binding with a Boolean value which indicates
whether or not the optional parameter had been specified by the caller.
Each
.meta expr
that is evaluated is evaluated in an environment in which
all of the previous parameters are visible, in addition to the surrounding
environment of the
.codn lambda .
For instance:
.verb
(let ((default 0))
(lambda (str : (end (length str)) (counter default))
(list str end counter)))
.brev
In this
.codn lambda ,
the initializing expression for the optional parameter
end is
.codn "(length str)" ,
and the
.meta str
variable it refers to is the previous
argument. The initializer for the optional variable counter is
the expression default, and it refers to the binding established
by the surrounding let. This reference is captured as part of the
.codn lambda 's
lexical closure.
Keyword symbols, and the symbols
.code t
and
.code nil
may not be used as parameter names.
The behavior is unspecified if the same symbol is specified
more than once anywhere in the parameter list, whether as a parameter name or as
the indicator
.meta sym
in an optional parameter or any combination.
Implementation note: the \*(TX compiler diagnoses and rejects duplicate
symbols in
.code lambda
whereas the interpreter ignores the situation.
Note: it is not always necessary to use the
.code lambda
operator directly in order to produce an anonymous function.
In situations when
.code lambda
is being written in order to simulate partial evaluation, it may be possible
to instead make use of the
.code op
macro. For instance the function
.code "(lambda (. args) [apply + a args])"
which adds the values of all of its arguments together, and to the lexically
captured variable
.code a
can be written more succinctly as
.codn "(op + a)" .
The
.code op
operator is the main representative of a family of operators:
.codn lop ,
.codn ap ,
.codn ip ,
.codn do ,
.codn ado ,
.code opip
and
.codn oand .
In situations when functions are simply combined together, the effect
may be achieved using some of the available functional combinators,
instead of a
.codn lambda .
For instance chaining together functions as in
.code "(lambda (x) (square (cos x)))"
is achievable using the
.code chain
function:
.codn "[chain cos square]" .
The
.code opip
operator can also be used:
.codn "(opip cos square)" .
Numerous combinators are available; see the section Partial Evaluation and
Combinators.
When a function is needed which accesses an object, there are also
alternatives. Instead of
.code "(lambda (obj) obj.slot)"
and
.codn "(lambda (obj arg) obj.(slot arg))" ,
it is simpler to use the
.code ".slot"
and
.code ".(slot arg)"
notations. See the section Unbound Referencing Dot.
Also see the functions
.code umethod
and
.code uslot
as well as the related convenience macros
.code umeth
and
.codn usl .
If a function is needed which partially applies,
to some arguments, a method invoked on a specific object, the
.code method
function or
.code meth
macro may be used. For instance, instead of
.codn "(lambda (arg) obj.(method 3 arg))" ,
it is possible to write
.code "(meth obj 3)"
except that the latter produces a variadic function.
.TP* Examples:
.IP "Counting function:"
This function, which takes no arguments, captures the
variable
.codn counter .
Whenever this object is called, it increments
.code counter
by
.code 1
and returns the incremented value.
.verb
(let ((counter 0))
(lambda () (inc counter)))
.brev
.IP "Function that takes two or more arguments:"
The third and subsequent arguments are aggregated into a list passed as the
single parameter
.codn z :
.verb
(lambda (x y . z) (list 'my-arguments-are x y z))
.brev
.IP "Variadic function:"
.verb
(lambda args (list 'my-list-of-arguments args))
.brev
.IP "Optional arguments:"
.verb
[(lambda (x : y) (list x y)) 1] -> (1 nil)
[(lambda (x : y) (list x y)) 1 2] -> (1 2)
.brev
.coNP Macros @ flet and @ labels
.synb
.mets (flet >> ({( name < param-list << function-body-form *)}*)
.mets \ \ << body-form *)
.mets (labels >> ({( name < param-list << function-body-form *)}*)
.mets \ \ << body-form *)
.syne
.desc
The
.code flet
and
.code labels
macros bind local, named functions in the lexical scope.
Note that the above syntax synopsis describes only the canonical
parameter syntax which remains after parameter list macros are
expanded. See the section Parameter List Macros.
The difference between
.code flet
and
.code labels
is that a function defined by
.code labels
can see itself, and therefore recurse directly by name. Moreover, if multiple
functions are defined by the same labels construct, they all have each other's
names in scope of their bodies.
By contrast, a
.codn flet -defined
function does not have itself in scope and cannot recurse.
Multiple functions in the same
.code flet
do not have each other's names in their scopes.
More formally, the
.metn function-body-form s
and
.meta param-list
of the functions defined by
.code labels
are in a scope in which all of the function
names being defined by that same
.code labels
construct are visible.
Under both
.code labels
and
.codn flet ,
the local functions that are defined are
lexically visible to the main
.metn body-form s.
Note that
.code labels
and
.code flet
are properly scoped with regard to macros.
During macro expansion, function bindings introduced by these
macro operators shadow macros defined by
.code macrolet
and
.codn defmacro .
Furthermore, function bindings introduced by
.code labels
and
.code flet
also shadow symbol macros defined by
.codn symacrolet ,
when those symbol macros occur as arguments of a
.code dwim
form.
See also: the
.code macrolet
operator.
.TP* "Dialect Note:"
The
.code flet
and
.code labels
macros do not establish named blocks around the body forms
of the local functions which they bind. This differs from
ANSI Common Lisp, whose local function have implicit named blocks,
allowing for
.code return-from
to be used.
.TP* Examples:
.verb
;; Wastefully slow algorithm for determining evenness.
;; Note:
;; - mutual recursion between labels-defined functions
;; - inner is-even bound by labels shadows the outer
;; one bound by defun so the (is-even n) call goes
;; to the local function.
(defun is-even (n)
(labels ((is-even (n)
(if (zerop n) t (is-odd (- n 1))))
(is-odd (n)
(if (zerop n) nil (is-even (- n 1)))))
(is-even n)))
.brev
.coNP Function @ call
.synb
.mets (call < function << argument *)
.syne
.desc
The
.code call
function invokes
.metn function ,
passing it the given arguments, if any.
.TP* Examples:
Apply
.code lambda
to
.code "1 2"
arguments, adding them to produce
.codn 3 :
.verb
(call (lambda (a b) (+ a b)) 1 2)
.brev
Useless use of
.code call
on a named function; equivalent to
.codn "(list 1 2)" :
.verb
(call (fun list) 1 2)
.brev
.coNP Functions @ apply and @ iapply
.synb
.mets (apply < function <> [ arg * << trailing-args ])
.mets (iapply < function <> [ arg * << trailing-args ])
.syne
.desc
The
.code apply
function invokes
.metn function ,
optionally passing to it an argument
list. The return value of the
.code apply
call is that of
.metn function .
If no arguments are present after
.metn function ,
then
.meta function
is invoked without arguments.
If one argument is present after
.metn function ,
then it is interpreted as
.metn trailing-args .
If this is a sequence (a list, vector or string),
then the elements of the sequence are passed as individual arguments to
.metn function .
If
.meta trailing-args
is not a sequence, then
.meta function
is invoked
with an improper argument list, terminated by the
.meta trailing-args
atom.
If two or more arguments are present after
.metn function ,
then the last of these arguments is interpreted as
.metn trailing-args .
The previous arguments represent leading arguments which are applied to
.metn function ,
prior to the arguments taken from
.metn trailing-args .
Note that if
.meta trailing-args
value is an atom or an improper list, the function is then
invoked with an improper argument list. Only a variadic
function may be invoked with an improper argument list.
Moreover, all of the function's required and optional
parameters must be satisfied by elements of the
improper list, such that the terminating atom either
matches the
.meta rest-param
directly (see the
.code lambda
operator) or else the
.meta rest-param
receives an improper list terminated by that atom.
To treat the terminating atom of an improper list as an
ordinary element which can satisfy a required or optional
function parameter, the
.code iapply
function may be used, described next.
The
.code iapply
function ("improper apply") is similar to
.codn apply ,
except with regard to the treatment of
.metn trailing-args .
Firstly, under
.codn iapply ,
if
.meta trailing-args
is an atom other than
.code nil
(possibly a sequence, such as a vector or string),
then it is treated as an ordinary argument:
.meta function
is invoked with a proper argument list, whose last element is
.metn trailing-args .
Secondly, if
.meta trailing-args
is a list, but an improper list, then the terminating atom of
.meta trailing-args
becomes an individual argument.
This terminating atom is not split into multiple arguments,
even if it is a sequence.
Thus, in all possible cases,
.code iapply
treats an extra
.cod2 non- nil
atom as an argument, and never calls
.meta function
with an improper argument list.
.TP* Examples:
.verb
;; '(1 2 3) becomes arguments to list, thus (list 1 2 3).
(apply (fun list) '(1 2 3)) -> (1 2 3)
;; this effectively invokes (list 1 2 3 4)
(apply (fun list) 1 2 '(3 4)) -> (1 2 3 4)
;; this effectively invokes (list 1 2 . 3)
(apply (fun list) 1 2 3)) -> (1 2 . 3)
;; "abc" is separated into characters
;; which become arguments of list
(apply (fun list) "abc") -> (#\ea #\eb #\ec)
.brev
.TP* "Dialect Note:"
Note that some uses of this function that are necessary in other Lisp dialects
are not necessary in \*(TL. The reason is that in \*(TL, improper list
syntax is accepted as a compound form, and performs application:
.verb
(foo a b . x)
.brev
Here, the variables
.code a
and
.code b
supply the first two arguments for
.codn foo .
In
the dotted position,
.code x
must evaluate to a list or vector. The list or
vector's elements are pulled out and treated as additional arguments for
.codn foo .
This syntax can only be used if
.code x
is a symbolic form or an atom. It
cannot be a compound form, because
.code "(foo a b . (x))"
and
.code "(foo a b x)"
are equivalent structures.
.coNP Operator @ fun
.synb
.mets (fun << function-name )
.syne
.desc
The
.code fun
operator retrieves the function object corresponding to a named
function in the current lexical environment.
The
.meta function-name
may be a symbol denoting a named function: a built in
function, or one defined by
.codn defun .
The
.meta function-name
may also take any of the forms specified in the description of the
.code func-get-name
function. If such a
.meta function-name
refers to a function which exists, then the
.code fun
operator yields that function.
Note: the
.code fun
operator does not see macro bindings via their symbolic names
with which they are defined by
.codn defmacro .
However, the name syntax
.mono
.meti (macro << name )
.onom
may be used to refer to macros. This syntax is documented in the
description of
.codn func-get-name .
It is also possible to
retrieve a global macro expander using the function
.codn symbol-macro .
.coNP Operator @ dwim
.synb
.mets (dwim << argument *)
.mets <> '[' argument *']'
.mets (set (dwim < obj-place < index <> [ alt ]) << new-value )
.mets (set >> '[' obj-place < index <> [ alt ]']' << new-value )
.syne
.desc
The
.code dwim
operator's name is an acronym: DWIM may be taken to mean
"Do What I Mean", or alternatively, "Dispatch, in a Way that is
Intelligent and Meaningful".
The notation
.code [...]
is a shorthand which denotes
.codn "(dwim ...)" .
Note that since the
.code [
and
.code ]
are used in this document for indicating optional syntax,
in the above Syntax synopsis the quoted notation
.code '['
and
.code ']'
denotes bracket tokens which literally appear in the syntax.
The
.code dwim
operator takes a variable number of arguments, which are
treated as expressions to be individually macro-expanded
and evaluated, using the same rules.
This means that the first argument isn't a function name, but an ordinary
expression which can simply compute a function object (or, more generally,
a callable object).
Furthermore, for those arguments of
.code dwim
which are symbols (after all macro-expansion is performed), the evaluation
rules are altered. For the purposes of resolving symbols to values, the
function and variable binding namespaces are considered to be merged into a
single space, creating a situation that is similar to a Lisp-1 style
dialect.
This special Lisp-1 evaluation is not recursively applied. All arguments of
.code dwim
which, after macro expansion, are not symbols are evaluated using the
normal Lisp-2 evaluation rules. Thus, the DWIM operator must be used
in every expression where the Lisp-1 rules for reducing symbols to
values are desired.
If a symbol has bindings both in the variable and function namespace in scope,
and is referenced by a dwim argument, this constitutes a conflict which is
resolved according to two rules. When nested scopes are concerned, then an
inner binding shadows an outer binding, regardless of their kind. An inner
variable binding for a symbol shadows an outer or global function binding, and
vice versa.
If a symbol is bound to both a function and variable in the global namespace,
then the variable binding is favored.
Macros do not participate in the special scope conflation, with one
exception. What this means is that the space of symbol macros is not folded
together with the space of operator macros. An argument of
.code dwim
that is a symbol might be
symbol macro, variable or function, but it cannot be interpreted as the name of
a operator macro.
The exception is this: from the perspective of a
.code dwim
form, function bindings can shadow symbol macros. If a
function binding is defined in an inner scope relative to a symbol macro for
the same symbol,
using
.code flet
or
.codn labels ,
the function hides the symbol macro. In other words, when
macro expansion processes an argument of a
.code dwim
form, and that argument is a symbol, it is treated specially
in order to provide a consistent name lookup behavior. If the innermost
binding for that symbol is a function binding, it refers to that
function binding, even if a more outer symbol macro binding exists,
and so the symbol is not expanded using the symbol macro.
By contrast, in an ordinary form, a symbolic argument never resolves
to a function binding. The symbol refers to either a symbol macro or a
variable, whichever is nested closer.
If, after macro expansion, the leftmost argument of the
.code dwim
is the name of a special operator or macro, the
.code dwim
form doesn't denote an invocation of that operator or macro. A
.code dwim
form is an invocation of the
.code dwim
operator, and the leftmost argument of that operator, if it is a symbol, is
treated as a binding to be resolved in the variable or function namespace,
like any other argument.
Thus
.code "[if x y]"
is an invocation of the
.code if
function, not the
.code if
operator.
How many arguments are required by the
.code dwim
operator depends on the type of
object to which the first argument expression evaluates. The possibilities
are:
.RS
.meIP >> [ function << argument *]
Call the given function object with the given arguments.
.meIP >> [ symbol << argument *]
If the first expression evaluates to a symbol, that symbol
is resolved in the function namespace, and then
the resulting function, if found, is called with the
given arguments.
.meIP >> [ sequence << index ]
Retrieve an element from
.metn sequence ,
at the specified
.metn index ,
which is a nonnegative integer.
This form is also a syntactic place.
If a value is stored to this place, it replaces the
element.
The place may also be deleted, which has the effect of removing the element
from the sequence, shifting the elements at higher indices, if any, down one
element position, and shortening the sequence by one.
If the place is deleted, and if
.meta sequence
is a list, then the
.meta sequence
form itself must be a place.
This form is implemented using the
.code ref
accessor such that, except for the argument evaluation semantics of the DWIM
brackets, it is equivalent to using the
.mono
.meti (ref < sequence << index )
.onom
syntax.
.meIP >> [ sequence << from-index..to-below-index ]
Retrieve the specified range of elements.
The range of elements is specified in the
.code from
and
.code to
fields of a range object. The
.code ..
(dotdot)
syntactic sugar denotes the construction of the range object via the
.code rcons
function. See the section on Range Indexing below.
This form is also a syntactic place. Storing a value in this place
has the effect of replacing the subsequence with
a new subsequence. Deleting the place has the
effect of removing the specified subsequence
from
.metn sequence .
If
.meta sequence
is a list, then the
.meta sequence
form must itself be a place.
The
.meta new-value
argument in a range assignment can be a string, vector or list,
regardless of whether the target is a string, vector or list.
If the target is a string, the replacement sequence must be
a string, or a list or vector of characters.
The semantics is implemented using the
.code sub
accessor, such that the following equivalence holds:
.verb
[seq from..to] <--> (sub seq from..to)
.brev
For this reason,
.meta sequence
may be any object that is iterable by
.codn iter-begin .
.meIP >> [ sequence << index-list ]
Elements specified
by
.metn index-list ,
which may be a list or vector,
are extracted from
.meta sequence
and returned as a sequence
of the same kind as
.metn sequence .
This form is equivalent to
.mono
.meti (select < sequence << where-index )
.onom
except when the target of an assignment operation.
This form is a syntactic place if
.meta sequence
is one. If a sequence is assigned to this place,
then elements of the sequence are distributed to the
specified locations.
The following equivalences hold between index-list-based indexing
and the
.code select
and
.code replace
functions, except that
.code set
always returns the value assigned, whereas
.code replace
returns its first argument:
.verb
[seq idx-list] <--> (select seq idx-list)
(set [seq idx-list] new) <--> (replace seq new idx-list)
.brev
Note that unlike the select function, this does not support
.mono
.meti >> [ hash << index-list ]
.onom
because since hash keys may be lists, that syntax is
indistinguishable from a simple hash lookup where
.meta index-list
is the key.
.meIP >> [ hash < key <> [ alt ]]
Retrieve a value from the hash table corresponding to
.metn key ,
or else return
.meta alt
if there is no such entry. The expression
.meta alt
is always evaluated, whether or not its value is used.
.meIP >> [ search-tree << key ]
Retrieves an element from the search tree as if by applying the
.code tree-lookup
function to
.metn key .
.meIP >> [ search-tree << from-key..to-below-key ]
Retrieves a list of elements from the search tree as if by evaluating the
.mono
.meti (sub-tree < search-tree < from-key << to-below-key )
.onom
expression.
.meIP >> [ regex >> [ start <> [ from-end ]] << string ]
Determine whether regular expression
.meta regex
matches
.metn string ,
and in that case return the
(possibly empty) leftmost matching substring.
Otherwise, return
.codn nil .
If
.meta start
is specified, it gives the starting position where
the search begins, and if
.meta from-end
is given, and has a value other than
.codn nil ,
it specifies a search from right to left. These optional
arguments have the same conventions and semantics as
their equivalents in the
.code search-regst
function.
Note that
.meta string
is always required, and is always the rightmost argument.
.meIP >> [ struct << arg *]
The structure instance
.meta struct
is inquired whether it supports a method named by the symbol
.metn lambda .
If so, that method is invoked on the object. The method
receives
.meta struct
followed by the value of every
.metn arg .
If this form is used as a place, then the object must
support a
.code lambda-set
method.
.meIP >> [ carray << index ]
.meIP >> [ carray << from-index..to-below-index ]
Element and range indexing is possible on object of type
.code carray
which manipulate arrays in a foreign ("C language") representation,
and are closely associated with the Foreign Function Interface (FFI).
Just like in the case of sequences, the semantics of referencing
.code carray
objects with the bracket notation is based on the functions
.codn ref ,
.codn refset ,
.code sub
and
.codn replace .
These, in turn, rely on the specialized functions.
.codn carray-ref ,
.codn carray-refset ,
.code carray-sub
and
.codn carray-replace .
.meIP >> [ buf << index ]
Indexing is supported for objects of type
.codn buf .
This provides a way to access and store the individual bytes
of a buffer.
.RE
.PP
.TP* "Range Indexing:"
Vector and list range indexing is based from zero, meaning
that the first element is numbered zero, the second one
and so on.
zero. Negative values are allowed; the value
.code -1
refers to the last element of the vector or
list, and
.code -2
to the second last and so forth. Thus the range
.code "1 .. -2"
means
"everything except for the first element and the last two".
The symbol
.code t
represents the position one past the end of the vector, string or
list, so
.code 0..t
denotes the entire list or vector, and the range
.code t..t
represents the empty range just beyond the last element.
It is possible to assign to
.codn t..t .
For instance:
.verb
(defvar list '(1 2 3))
(set [list t..t] '(4)) ;; list is now (1 2 3 4)
.brev
The value zero has a "floating" behavior when used as the end of a range.
If the start of the range is a negative value, and the end of the
range is zero, the zero is interpreted as being the position past the
end of the sequence, rather than the first element. For instance the range
.code -1..0
means the same thing as
.codn -1..t .
Zero at the start of a range
always means the first element, so that
.code 0..-1
refers to all the elements except for the last one.
.TP* Notes:
The dwim operator allows for a Lisp-1 flavor of programming in \*(TL,
which is principally a Lisp-2 dialect.
A Lisp-1 dialect is one in which an expression like
.code "(a b)"
treats both a and b
as expressions subject to the same evaluation rules\(emat least, when
.code a
isn't an operator or an operator macro. This means that the symbols
.code a
and
.code b
are resolved to values in the same namespace. The form denotes a function call
if the value of variable
.code a
is a function object. Thus in a Lisp-1, named functions do not exist as
such: they are just variable bindings. In a Lisp-1, the form
.code "(car 1)"
means that there
is a variable called
.codn car ,
which holds a function, which is retrieved from that
variable and the argument
.code 1
is applied to it. In the expression
.codn "(car car)" ,
both occurrences of
.code car
refer to the variable, and so this form applies the
.code car
function to itself. It is almost certainly meaningless.
In a Lisp-2
.code "(car 1)"
means that there is a function called
.codn car ,
in the function namespace. In the expression
.code "(car car)"
the two occurrences refer to different bindings:
one is a function and the other a variable.
Thus there can exist a variable
.code car
which holds a cons-cell object, rather than the
.code car
function, and the form makes sense.
The Lisp-1 approach is useful for functional programming, because it eliminates
cluttering occurrences of the call and fun operators. For instance:
.verb
;; regular notation
(call foo (fun second) '((1 a) (2 b)))
;; [] notation
[foo second '((1 a) (2 b))]
.brev
Lisp-1 dialects can also provide useful extensions by giving a meaning
to objects other than functions in the first position of a form,
and the
.code dwim/[...]
syntax does exactly this.
\*(TL is a Lisp-2 because Lisp-2 also has advantages. Lisp-2 programs
which use macros naturally achieve hygiene because lexical variables do
not interfere with the function namespace. If a Lisp-2 program has
a local variable called
.codn list ,
this does not interfere with the hidden use of the function
.code list
in a macro expansion in the same block of code. Lisp-1 dialects have to
provide hygienic macro systems to attack this problem. Furthermore, even when
not using macros, Lisp-1 programmers have to avoid using the names of functions
as lexical variable names, if the enclosing code might use them.
The two namespaces of a Lisp-2 also naturally accommodate symbol macros and
operator macros. Whereas functions and variables can be represented in a
single namespace readily, because functions are data objects, this is not so
with symbol macros and operator macros, the latter of which are distinguished
syntactically by their position in a form. In a Lisp-1 dialect, given
.codn "(foo bar)" ,
either of the two symbols could be a symbol macro, but only
.code foo
can possibly be an operator macro. Yet, having only a single namespace, a
Lisp-1 doesn't permit
.codn "(foo foo)" ,
where
.code foo
is simultaneously a symbol macro and an operator macro, though the situation is
unambiguous by syntax even in Lisp-1. In other words, Lisp-1 dialects do not
entirely remove the special syntactic recognition given to the leftmost
position of a compound form, yet at the same time they prohibit
the user from taking full advantage of it by providing only one namespace.
\*(TL provides
the "best of both worlds": the DWIM brackets notation provides a model of
Lisp-1 computation that is purer than Lisp-1 dialects (since the leftmost
argument is not given any special syntactic treatment at all)
while the Lisp-2 foundation provides a traditional Lisp environment with its
"natural hygiene".
.coNP Function @ functionp
.synb
.mets (functionp << obj )
.syne
.desc
The
.code functionp
function returns
.code t
if
.meta obj
is a function, otherwise it returns
.codn nil .
.coNP Function @ copy-fun
.synb
.mets (copy-fun << function )
.syne
.desc
The
.code copy-fun
function produces and returns a duplicate of
.metn function ,
which must be a function.
A duplicate of a function is a distinct function object not
.code eq
to the original function, yet which accepts the same arguments
and behaves exactly the same way as the original.
If a function contains no captured environment, then a copy made of that
function by
.code copy-fun
is indistinguishable from the original function in every regard,
except for being a distinct object that compares unequal to the original
under the
.code eq
function.
If a function contains a captured environment, then a copy of that function
made by
.code copy-fun
has its own copy of that environment. If the copied function changes the
values of captured lexical variables, the original function is not affected by
these changes and vice versa.
The entire lexical environment is copied; the copy and original function do not
share any portion of the environment at any level of nesting.
.SS* Sequencing, Selection and Iteration
.coNP Operators/functions @ progn and @ prog1
.synb
.mets (progn << form *)
.mets (prog1 << form *)
.syne
.desc
The
.code progn
operator evaluates each
.meta form
in left-to-right order, and returns the value
of the last form. The value of the form
.code (progn)
is
.codn nil .
The
.code prog1
operator evaluates each
.meta form
in left-to-right order, and returns the value
of the first form. The value of the form
.code (prog1)
is
.codn nil .
Various other operators such as
.code let
also arrange for the evaluation
of a body of forms, the value of the last of which is returned.
These operators are said to feature an implicit
.codn progn .
These special operators are also functions. The
.code progn
function accepts zero or more arguments. It returns its last argument, or
.code nil
if called with no arguments. The
.code prog1
function likewise accepts zero or more arguments. It returns its first argument, or
.code nil
if called with no arguments.
.TP* "Dialect Notes:"
In ANSI Common Lisp,
.code prog1
requires at least one argument. Neither
.code prog
nor
.code prog1
exist as functions.
.coNP Macro/function @ prog2
.synb
.mets (prog2 << form *)
.syne
.desc
The
.code prog2
evaluates each
.meta form
in left-to-right order. The value is that of the second form, if present,
otherwise it is
.codn nil .
The form
.code "(prog2 1 2 3)"
yields
.codn 2 .
The value of
.code "(prog2 1 2)"
is also
.codn 2 ;
.code "(prog2 1)"
and
.code "(prog2)"
yield
.codn nil .
The
.code prog2
symbol also has a function binding. The
.code prog2
function accepts any number of arguments. If invoked with at least two arguments,
it returns the second one. Otherwise it returns
.codn nil .
.TP* "Dialect Notes:"
In ANSI Common Lisp,
.code prog2
requires at least two arguments.
It does not exist as a function.
.coNP Operator @ cond
.synb
.mets (cond >> {( test << form *)}*)
.syne
.desc
The
.code cond
operator provides a multi-branching conditional evaluation of
forms. Enclosed in the cond form are groups of forms expressed as lists.
Each group must be a list of at least one form.
The forms are processed from left to right as follows: the first form,
.metn test ,
in each group is evaluated. If it evaluates true, then the remaining
forms in that group, if any, are also evaluated. Processing then terminates and
the result of the last form in the group is taken as the result of cond.
If
.meta test
is the only form in the group, then result of
.meta test
is taken
as the result of
.codn cond .
If the first form of a group yields
.codn nil ,
then processing continues with the
next group, if any. If all form groups yield
.codn nil ,
then the cond form yields
.codn nil .
This holds in the case that the syntax is empty:
.code (cond)
yields
.codn nil .
.coNP Macros @, caseq @ caseql and @ casequal
.synb
.mets (caseq < test-form << normal-clause * <> [ else-clause ])
.mets (caseql < test-form << normal-clause * <> [ else-clause ])
.mets (casequal < test-form << normal-clause * <> [ else-clause ])
.syne
.desc
These three macros arrange for the evaluation of
.metn test-form ,
whose value is then compared against the key or keys in each
.meta normal-clause
in turn.
When the value matches a key, then the remaining forms of
.meta normal-clause
are evaluated, and the value of the last form is returned; subsequent
clauses are not evaluated. When the value doesn't match any of the keys
of a
.meta normal-clause
then the next
.meta normal-clause
is tested.
If all these clauses are exhausted, and there is no
.metn else-clause ,
then the value nil is returned. Otherwise, the forms in the
.meta else-clause
are evaluated, and the value of the last one is returned.
If there are no forms, then
.code nil
is returned.
The syntax of a
.meta normal-clause
takes on these two forms:
.mono
.mets >> ( key << form *)
.onom
where
.meta key
may be an atom which denotes a single key, or else a list
of keys. There is a restriction that the symbol
.code t
may not be used
as
.metn key .
The form
.code (t)
may be used as a key to match that symbol.
The syntax of an
.meta else-clause
is:
.mono
.mets (t << form *)
.onom
which resembles a form that is often used as the final clause
in the
.code cond
syntax.
The three forms of the case construct differ from what type of
test they apply between the value of
.meta test-form
and the keys.
The
.code caseq
macro generates code which uses the
.code eq
function's
equality. The
.code caseql
macro uses
.codn eql ,
and
.code casequal
uses
.codn equal .
.TP* Example
.verb
(let ((command-symbol (casequal command-string
(("q" "quit") 'quit)
(("a" "add") 'add)
(("d" "del" "delete") 'delete)
(t 'unknown))))
...)
.brev
.coNP Macros @, caseq* @ caseql* and @ casequal*
.synb
.mets (caseq* < test-form << normal-clause * <> [ else-clause ])
.mets (caseql* < test-form << normal-clause * <> [ else-clause ])
.mets (casequal* < test-form << normal-clause * <> [ else-clause ])
.syne
.desc
The
.codn caseq* ,
.codn caseql* ,
and
.code casequal*
macros are similar to the macros
.codn caseq ,
.codn caseql ,
and
.codn casequal ,
differing from them in only the following regard. The
.metn normal-clause ,
of these macros has the form
.mono
.meti >> ( evaluated-key << form *)
.onom
where
.code evaluated-key
is either an atom, which is evaluated to produce a key, or else
else a compound form, whose elements are evaluated as forms, producing
multiple keys. This evaluation takes place at macro-expansion time,
in the global environment.
The
.meta else-clause
works the same way under these macros as under
.code caseq
et al.
Note that although in a
.metn normal-clause ,
.meta evaluated-key
must not be the atom
.codn t ,
there is no restriction against it being
an atom which evaluates to
.codn t .
In this situation, the value
.code t
has no special meaning.
The
.meta evaluated-key
expressions are evaluated in the order in which they appear in
the construct, at the time the
.codn caseq* ,
.code caseql*
or
.code casequal*
macro is expanded.
Note: these macros allow the use of variables and global symbol
macros as case keys.
.TP* Example:
.verb
(defvarl red 0)
(defvarl green 1)
(defvarl blue 2)
(let ((color blue))
(caseql* color
(red "hot")
((green blue) "cool")))
--> "cool"
.brev
.coNP Macros @, ecaseq @, ecaseql @, ecasequal @, ecaseq* @ ecaseql* and @ ecasequal*
.synb
.mets (ecaseq < test-form << normal-clause * <> [ else-clause ])
.mets (ecaseql < test-form << normal-clause * <> [ else-clause ])
.mets (ecasequal < test-form << normal-clause * <> [ else-clause ])
.mets (ecaseq* < test-form << normal-clause * <> [ else-clause ])
.mets (ecaseql* < test-form << normal-clause * <> [ else-clause ])
.mets (ecasequal* < test-form << normal-clause * <> [ else-clause ])
.syne
.desc
These macros are error-catching variants of, respectively,
.codn caseq ,
.codn caseql ,
.codn casequal ,
.codn caseq* ,
.code caseql*
and
.codn casequal* .
If the
.meta else-clause
is present in the invocation of an error-catching case macro, then the the
invocation is precisely equivalent to the corresponding non-error-trapping
variant.
If the
.meta else-clause
is missing in the invocation of an error-catching variant, then a default
.meta else-clause
is inserted which throws an exception of type
.codn case-error ,
derived from
.codn error .
After this insertion, the semantics follows that of the non-error-trapping
variant.
For instance,
.codn "(ecaseql 3)" ,
which has no
.metn else-clause ,
is equivalent to
.mono
.meti (caseql 3 (t << expr ))
.onom
where
.meta expr
indicates the inserted expression which throws
.codn case-error .
However,
.code "(ecaseql 3 (t 42))"
is simply equivalent to
.codn "(caseql 3 (t 42))" ,
since it has an
.metn else-clause .
Note: the error-catching case macros are intended for situations in which it is
a matter of program correctness that every possible value of
.meta test-form
matches a
.metn normal-clause ,
such that if a failure to match occurs, it indicates a software defect.
The error-throwing
.meta else-clause
helps to ensure that the error situation is noticed.
Without this clause, the case macro terminates with a value of
.codn nil ,
which may conceal the defect and delay its identification.
.coNP Operator/function @ if
.synb
.mets (if < cond < t-form <> [ e-form ])
.mets '['if < cond < then <> [ else ]']'
.syne
.desc
There exist both an
.code if
operator and an
.code if
function. A list form with the symbol
.code if
in the first position is interpreted as an invocation of the
.code if
operator.
The function can be accessed using the DWIM bracket notation and in other
ways.
The
.code if
operator provides a simple two-way-selective evaluation control.
The
.meta cond
form is evaluated. If it yields true then
.meta t-form
is evaluated, and that form's return value becomes the return value of the
.codn if .
If
.meta cond
yields false, then
.meta e-form
is evaluated and its return value is taken to be that of
.codn if .
If
.meta e-form
is omitted, then the behavior is as if
.meta e-form
were specified as
.codn nil .
The
.code if
function provides no evaluation control. All of its arguments
are evaluated from left to right. If the
.meta cond
argument is true, then it
returns the
.meta then
argument, otherwise it returns the value of the
.meta else
argument if present, otherwise it returns
.codn nil .
.coNP Operator/function @ and
.synb
.mets (and << form *)
.mets '['and << arg *']'
.syne
.desc
There exist both an
.code and
operator and an
.code and
function. A list form with the
symbol
.code and
in the first position is interpreted as an invocation of the
operator. The function can be accessed using the DWIM bracket notation and in
other ways.
The
.code and
operator provides three functionalities in one. It computes the
logical "and" function over several forms. It controls evaluation (a.k.a.
"short-circuiting"). It also provides an idiom for the convenient substitution
of a value in place of
.code nil
when some other values are all true.
The
.code and
operator evaluates as follows. First, a return value is
established and initialized to the value
.codn t .
The
.metn form s,
if any, are
evaluated from left to right. The return value is overwritten with
the result of each
.metn form .
Evaluation stops when all
.metn form s
are exhausted, or when
.code nil
is stored in the return value.
When evaluation stops, the operator yields the return value.
The
.code and
function provides no evaluation control: it receives all of its
arguments fully evaluated. If it is given no arguments, it returns
.codn t .
If it is given one or more arguments, and any of them are
.codn nil ,
it returns
.codn nil .
Otherwise, it returns the value of the last argument.
.TP* Examples:
.verb
(and) -> t
(and (> 10 5) (stringp "foo")) -> t
(and 1 2 3) -> 3 ;; shorthand for (if (and 1 2) 3).
.brev
.coNP Macro/function @ nand
.synb
.mets (nand << form *)
.mets '['nand << arg *']'
.syne
.desc
There exist both a
.code nand
macro and a
.code nand
function.
A list form with the symbol
.code nand
in the first position is interpreted as an invocation of the macro.
The function can be accessed using the DWIM bracket notation and in
other ways.
The
.code nand
macro and function are the logical negation of the
.code and
operator and function.
They are related according to the following equivalences:
.verb
(nand f0 f1 f2 ...) <--> (not (and f0 f1 f2 ...))
[nand f0 f1 f2 ...] <--> (not [and f0 f1 f2 ...])
.brev
.coNP Operator/function @ or
.synb
.mets (or << form *)
.mets '['or << arg *']'
.syne
.desc
There exist both an
.code or
operator and an
.code or
function. A list form with the
symbol
.code or
in the first position is interpreted as an invocation of the
operator. The function can be accessed using the DWIM bracket notation and in
other ways.
The
.code or
operator provides three functionalities in one. It computes the
logical "or" function over several forms. It controls evaluation (a.k.a.
"short-circuiting"). The behavior of
.code or
also provides an idiom for the selection of the first
.cod2 non- nil
value from a sequence of forms.
The
.code or
operator evaluates as follows. First, a return value is
established and initialized to the value
.codn nil .
The
.metn form s,
if any,
are evaluated from left to right. The return value is overwritten
with the result of each
.metn form .
Evaluation stops when all
.metn form s
are exhausted, or when a true value is stored into the return value.
When evaluation stops, the operator yields the return value.
The
.code or
function provides no evaluation control: it receives all of its
arguments fully evaluated. If it is given no arguments, it returns
.codn nil .
If all of its arguments are
.codn nil ,
it also returns
.codn nil .
Otherwise, it
returns the value of the first argument which isn't
.codn nil .
.TP* Examples:
.verb
(or) -> nil
(or 1 2) -> 1
(or nil 2) -> 2
(or (> 10 20) (stringp "foo")) -> t
.brev
.coNP Macro/function @ nor
.synb
.mets (nor << form *)
.mets '['nor << arg *']'
.syne
.desc
There exist both a
.code nor
macro and a
.code nor
function.
A list form with the symbol
.code nor
in the first position is interpreted as an invocation of the macro.
The function can be accessed using the DWIM bracket notation and in
other ways.
The
.code nor
macro and function are the logical negation of the
.code or
operator and function.
They are related according to the following equivalences:
.verb
(nor f0 f1 f2 ...) <--> (not (or f0 f1 f2 ...))
[nor f0 f1 f2 ...] <--> (not [or f0 f1 f2 ...])
.brev
.coNP Macros @ when and @ unless
.synb
.mets (when < expression << form *)
.mets (unless < expression << form *)
.syne
.desc
The
.code when
macro operator evaluates
.metn expression .
If
.meta expression
yields
true, and there are additional forms, then each
.meta form
is evaluated.
The value of the last form becomes the result value of the
.code when
form.
If there are no forms, then the result is
.codn nil .
The
.code unless
operator is similar to
.codn when ,
except that it reverses the
logic of the test. The forms, if any, are evaluated if and only if
.meta expression
is false.
.coNP Macros @ while and @ until
.synb
.mets (while < expression << form *)
.mets (until < expression << form *)
.syne
.desc
The
.code while
macro operator provides a looping construct. It evaluates
.metn expression .
If
.meta expression
yields
.codn nil ,
then the evaluation of the
.code while
form
terminates, producing the value
.codn nil .
Otherwise, if there are additional forms,
then each
.meta form
is evaluated. Next, evaluation returns to
.metn expression ,
repeating all of the previous steps.
The
.code until
macro operator is similar to
.codn while ,
except that the
.code until
form terminates when
.meta expression
evaluates true, rather than false.
These operators arrange for the evaluation of all their enclosed forms
in an anonymous block. Any of the
.metn form s,
or
.metn expression ,
may use
the
.code return
operator to terminate the loop, and optionally to specify
a result value for the form.
The only way these forms can yield a value other than
.code nil
is if the
.code return
operator is used to terminate the implicit anonymous block,
and is given an argument, which becomes the result value.
.coNP Macros @ while* and @ until*
.synb
.mets (while* < expression << form *)
.mets (until* < expression << form *)
.syne
.desc
The
.code while*
and
.code until*
macros are similar, respectively, to the macros
.code while
and
.codn until .
They differ in one respect: they begin by evaluating the
.metn form s
one time unconditionally, without first evaluating
.metn expression .
After this evaluation, the subsequent behavior is
like that of
.code while
or
.codn until .
Another way to regard the behavior is that that these forms execute
one iteration unconditionally, without evaluating the termination test prior to
the first iteration. Yet another view is that these constructs relocate the
test from the top of the loop to the bottom of the loop.
.coNP Macro @ whilet
.synb
.mets (whilet >> ({ sym | >> ( sym << init-form )}+)
.mets \ \ << body-form *)
.syne
.desc
The
.code whilet
macro provides a construct which combines iteration with variable
binding.
The evaluation of the form takes place as follows. First, fresh bindings are
established for
.metn sym s
as if by the
.code let*
operator.
It is an error for the list of variable bindings to be empty.
After the establishment of the bindings, the value of the last
.meta sym
is tested. If the value is
.codn nil ,
then
.code whilet
terminates. Otherwise,
.metn body-form s
are evaluated in the scope of the variable bindings, and then
.code whilet
iterates from the beginning, again establishing fresh bindings for the
.metn sym s,
and testing the value of the last
.metn sym .
All evaluation takes place in an anonymous block, which can be
terminated with the
.code return
operator. Doing so terminates the loop.
If the
.code whilet
loop is thus terminated by an explicit
.codn return ,
a return value can be specified. Under normal termination, the return value is
.codn nil .
In the syntax, a small convenience is permitted. Instead of the last
.mono
.meti >> ( sym << init-form )
.onom
it is permissible for the syntax
.mono
.meti <> ( init-form )
.onom
to appear, the
.meta sym
being omitted. A machine-generated variable is substituted
in place of the missing
.meta sym
and that variable is then initialized from
.meta init-form
and used as the basis of the test.
.TP* Examples:
.verb
;; read lines of text from *stdin* and print them,
;; until the end-of-stream condition:
(whilet ((line (get-line)))
(put-line line))
;; read lines of text from *stdin* and print them,
;; until the end-of-stream condition occurs or
;; a line is identical to the character string "end".
(whilet ((line (get-line))
(more (and line (nequal line "end"))))
(put-line line))
.brev
.coNP Macros @ iflet and @ whenlet
.synb
.mets (iflet >> {({ sym | >> ( sym << init-form )}+) | << atom-form }
.mets \ \ < then-form <> [ else-form ])
.mets (whenlet >> {({ sym | >> ( sym << init-form )}+) | << atom-form }
.mets \ \ << body-form *)
.syne
.desc
The
.code iflet
and
.code whenlet
macros combine the variable binding of
.code let*
with conditional evaluation of
.code if
and
.codn when ,
respectively.
In either construct's syntax, a non-compound form
.meta atom-form
may appear in place of the variable binding list. In this case,
.meta atom-form
is evaluated as a form, and the construct is equivalent to
its respective ordinary
.code if
or
.code when
counterpart.
If the list of variable bindings is empty, it is interpreted as the atom
.code nil
and treated as an
.metn atom-form .
If one or more bindings are specified rather than
.metn atom-form ,
then the evaluation of these forms takes
place as follows. First, fresh bindings are established for
.metn sym s
as if by the
.code let*
operator.
Then, the last variable's value is tested. If it is not
.code nil
then the test is true, otherwise false.
In the syntax, a small convenience is permitted. Instead of the last
.mono
.meti >> ( sym << init-form )
.onom
it is permissible for the syntax
.mono
.meti <> ( init-form )
.onom
to appear, the
.meta sym
being omitted. A machine-generated variable is substituted
in place of the missing
.meta sym
and that variable is then initialized from
.meta init-form
and used as the basis of the test. This is intended to be useful
in situations in which
.meta then-form
or
.meta else-form
do not require access to the tested value.
In the case of the
.code iflet
operator, if the test is true, the operator evaluates
.meta then-form
and yields its value. Otherwise the test is false, and if the
optional
.meta else-form
is present, that is evaluated instead and its value is returned.
If this form is missing, then
.code nil
is returned.
In the case of the
.code whenlet
operator, if the test is true, then the
.metn body-form s,
if any, are evaluated. The value of the last one is
returned, otherwise
.code nil
if the forms are missing.
If the test is false, then evaluation of
.metn body-form s
is skipped, and
.code nil
is returned.
.TP* Examples:
.verb
;; dispose of foo-resource if present
(whenlet ((foo-res (get-foo-resource obj)))
(foo-shutdown foo-res)
(set-foo-resource obj nil))
;; Contrast with: above, using when and let
(let ((foo-res (get-foo-resource obj)))
(when foo-res
(foo-shutdown foo-res)
(set-foo-resource obj nil)))
;; print frobosity value if it exceeds 150
(whenlet ((fv (get-frobosity-value))
(exceeds-p (> fv 150)))
(format t "frobosity value ~a exceeds 150\en" fv))
;; same as above, taking advantage of the
;; last variable being optional:
(whenlet ((fv (get-frobosity-value))
((> fv 150)))
(format t "frobosity value ~a exceeds 150\en" fv))
;; yield 4: 3 interpreted as atom-form
(whenlet 3 4)
;; yield 4: nil interpreted as atom-form
(iflet () 3 4)
.brev
.coNP Macro @ condlet
.synb
.mets (condlet
.mets \ ([({ sym | >> ( sym << init-form )}+) | << atom-form ]
.mets \ \ << body-form *)*)
.syne
.desc
The
.code condlet
macro generalizes
.codn iflet .
Each argument is a compound consisting of at least one item: a list of
bindings or
.metn atom-form .
This item is followed by zero or more
.metn body-form s.
If there are no
.metn body-form s
then the situation is treated as if there were a single
.meta body-form
specified as
.codn nil .
The arguments of
.code condlet
are considered in sequence, starting with the
leftmost.
If the argument's left item is an
.meta atom-form
then the form is evaluated. If it yields true, then the
.metn body-form s
next to it are evaluated in order, and the
.code condlet
form terminates, yielding the value obtained from the last
.metn body-form .
If
.meta atom-form
yields false, then the next argument is considered, if there is one.
If the argument's left item is a list of bindings, then it is processed
with exactly the same logic as under the
.code iflet
macro. If the last binding contains a true value, then the
adjoining
.metn body-form s
are evaluated in a scope in which all of the bindings are visible, and
.code condlet
terminates, yielding the value of the last
.metn body-form .
Otherwise, the next argument of
.code condlet
is considered (processed in a scope in which the bindings produced
by the current item are no longer visible).
If
.code condlet
runs out of arguments, it terminates and returns
.codn nil .
.TP* Example:
.verb
(let ((l '(1 2 3)))
(condlet
;; first arg
(((a (first l) ;; a binding gets 1
(b (second l)) ;; b binding gets 2
(g (> a b)))) ;; last variable g is nil
'foo) ;; not evaluated
;; second arg
(((b (second l) ;; b gets 2
(c (third l)) ;; c gets 3
(g (> b c)))) ;; last variable g is true
'bar))) ;; condlet terminates
--> bar ;; result is bar
.brev
.coNP Macro @ ifa
.synb
.mets (ifa < cond < then <> [ else ])
.syne
.desc
The
.code ifa
macro provides an anaphoric conditional operator resembling the
.code if
operator. Around the evaluation of the
.meta then
and
.meta else
forms, the symbol
.code it
is implicitly bound to a subexpression of
.metn cond ,
a subexpression which is thereby identified as the
.IR it-form .
This
.code it
alias provides a convenient reference to that place or value, similar to the
word "it" in the English language, and similar anaphoric pronouns in other
languages.
If
.code it
is bound to a place form, the binding is established
as if using the
.code placelet
operator: the form is evaluated only once, even if the
.code it
alias is used multiple times in the
.meta then
or
.meta else
expressions. Furthermore, the place form is implicitly surrounded with
.code read-once
so that the place's value is accessed only once, and multiple references to
.code it
refer to a copy of the value cached in a hidden variable, rather than
generating multiple accesses to the place.
Otherwise, if the form is not a syntactic place
.code it
is bound as an ordinary lexical variable to
the form's value.
An
.I it-candidate
is an an expression viable for having its value or storage location
bound to the
.code it
symbol. An it-candidate is any expression which is not a constant expression
according to the
.code constantp
function, and not a symbol.
The
.code ifa
macro imposes applies several rules to the
.meta cond
expression:
.RS
.IP 1.
The
.meta cond
expression must be either an atom, a function call form,
or a
.code dwim
form. Otherwise the
.code ifa
expression is ill-formed, and throws an exception at
macro-expansion time. For the purposes of these rules,
a
.code dwim
form is considered as a function call expression, whose first
argument is the second element of the form. That is to say,
.code "[f x]"
which is equivalent to
.code "(dwim f x)"
is treated similarly to
.code "(f x)"
as a one-argument call.
.IP 2.
If the
.meta cond
expression is a function call with two or more arguments,
at most one of them may be an it-candidate.
If two or more arguments are it-candidates, the situation
is ambiguous. The
.code ifa
expression is ill-formed and throws an exception at macro-expansion
time.
.IP 3.
If
.meta cond
is an atom, or a function call expression with no arguments,
then the
.code it
symbol is not bound. Effectively,
.code ifa
macro behaves like the ordinary
.code if
operator.
.IP 4.
If
.meta cond
is a function call or
.code dwim
expression with exactly one argument, then the
.code it
variable is bound to the argument expression, except when
the function being called is
.codn not ,
.codn null ,
or
.codn false .
This binding occurs regardless of whether the expression is
an it-candidate.
.IP 5.
If
.meta cond
is a function call with exactly one argument to the Boolean negation
function which goes by one of the three names
.codn not ,
.codn null ,
or
.codn false ,
then that situation is handled by a rewrite according to the following pattern:
.mono
.mets (ifa (not << expr ) < then << else ) -> (ifa < expr < else << then )
.onom
which applies likewise for
.code null
or
.code false
substituted for
.codn not .
The Boolean inverse function is removed, and the
.meta then
and
.meta else
expressions are exchanged.
.IP 6.
If
.meta cond
is a function call with two or more arguments, then it is only
well-formed if at most one of those arguments is an it-candidate.
If there is one such argument, then the
.code it
variable is bound to it.
.IP 7.
Otherwise the variable is bound
to the leftmost argument expression, regardless of whether that
argument expression is an it-candidate.
.RE
.IP
In all other regards, the
.code ifa
macro behaves similarly to
.codn if .
The
.meta cond
expression is evaluated, and, if applicable,
the value of, or storage location denoted by the appropriate argument is
captured and bound to the variable
.code it
whose scope extends over the
.meta then
form, as well as over
.metn else ,
if present.
If
.meta cond
yields a true value, then
.meta then
is evaluated and the resulting value is returned, otherwise
.meta else
is evaluated if present and its value is returned.
A missing
.meta else
is treated as if it were the
.code nil
form.
.TP* Examples:
.verb
(ifa t 1 0) -> 1
;; Rule 6: it binds to (* x x), which is
;; the only it-candidate.
(let ((x 6) (y 49))
(ifa (> y (* x x)) ;; it binds to (* x x)
(list it)))
-> (36)
;; Rule 4: it binds to argument of evenp,
;; even though 4 isn't an it-candidate.
(ifa (evenp 4)
(list it))
-> (4)
;; Rule 5:
(ifa (not (oddp 4))
(list it))
-> (4)
;; Rule 7: no candidates: choose leftmost
(let ((x 6) (y 49))
(ifa (< 1 x y)
(list it)))
-> (1)
-> (4)
;; Violation of Rule 1:
;; while is not a function
(ifa (while t (print 42))
(list it))
--> exception!
;; Violation of Rule 2:
(let ((x 6) (y 49))
(ifa (> (* y y y) (* x x)))
(list it))
--> exception!
.brev
.coNP Macro @ conda
.synb
.mets (conda >> {( test << form *)}*)
.syne
.desc
The
.code conda
operator provides a multi-branching conditional evaluation of
forms, similarly to the
.code cond
operator. Enclosed in the cond form are groups of forms expressed as lists.
Each group must be a list of at least one form.
The
.code conda
operator is anaphoric: it expands into a nested structure of zero or more
.code ifa
invocations, according to these patterns:
.verb
(conda) -> nil
(conda (x y ...) ...) -> (ifa x (progn y ...) (conda ...))
.brev
Thus,
.code conda
inherits all the restrictions on the
.meta test
expressions from
.codn ifa ,
as well as the anaphoric
.code it
variable feature.
.coNP Macro @ whena
.synb
.mets (whena < test << form *)
.syne
.desc
The
.code whena
macro is similar to the
.code when
macro, except that it is anaphoric in exactly the same manner as the
.code ifa
macro. It may be understood as conforming to the following equivalence:
.verb
(whena x f0 f2 ...) <--> (if x (progn f0 f2 ...))
.brev
.coNP Macro @ dotimes
.synb
.mets (dotimes >> ( var < count-form <> [ result-form ])
.mets \ \ << body-form *)
.syne
.desc
The
.code dotimes
macro implements a simple counting loop.
.meta var
is established as a variable, and initialized to zero.
.meta count-form
is evaluated one time to produce a limiting value, which should be a number.
Then, if the value of
.meta var
is less than the limiting value, the
.metn body-form s
are evaluated,
.meta var
is incremented by one, and the process repeats with a new comparison
of
.meta var
against the limiting value possibly leading to another evaluation of
the forms.
If
.meta var
is found to equal or exceed the limiting value, then the loop terminates.
When the loop terminates, its return value is
.code nil
unless a
.meta result-form
is present, in which case the value of that form specifies the return value.
.metn body-form s
as well as
.meta result-form
are evaluated in the scope in which the binding of
.meta var
is visible.
.coNP Operators @, each @, each* @, collect-each @, collect-each* @ append-each and @ append-each*
.synb
.mets (each >> ({( sym << init-form )}*) << body-form *)
.mets (each* >> ({( sym << init-form )}*) << body-form *)
.mets (collect-each >> ({( sym << init-form )}*) << body-form *)
.mets (collect-each* >> ({( sym << init-form )}*) << body-form *)
.mets (append-each >> ({( sym << init-form )}*) << body-form *)
.mets (append-each* >> ({( sym << init-form )}*) << body-form *)
.syne
.desc
These operators establish a loop for iterating over the elements of one or more
sequences. Each
.meta init-form
must evaluate to an iterable object that is suitable as an argument for the
.code iter-begin
function.
The sequences are then iterated in
parallel over repeated evaluations of the
.metn body-form s,
with each
.meta sym
variable being assigned to successive elements of its sequence. The shortest
list determines the number of iterations, so if any of the
.metn init-form s
evaluate to
an empty sequence, the body is not executed.
If the list of
.mono
.meti >> ( sym << init-form )
.onom
pairs itself is empty, then an infinite loop is specified.
The body forms are enclosed in an anonymous block, allowing the
.code return
operator to terminate the loop prematurely and optionally specify
the return value.
The
.code collect-each
and
.code collect-each*
variants are like
.code each
and
.codn each* ,
except that for each iteration, the resulting value of the body is collected
into a list. When the iteration terminates, the return value of the
.code collect-each
or
.code collect-each*
operator is this collection.
The
.code append-each
and
.code append-each*
variants are like
.code each
and
.codn each* ,
except that for each iteration other than the last, the resulting value of the
body must be a list. The last iteration may produce either an atom or a list.
The objects produced by the iterations are combined together as if they
were arguments to the append function, and the resulting value is the
value of the
.code append-each
or
.code append-each*
operator.
The alternate forms denoted by the adorned symbols
.codn each* ,
.code collect-each*
and
.codn append-each* ,
differ from
.codn each ,
.code collect-each
and
.code append-each
in the following way. The plain forms evaluate the
.metn init-form s
in an environment in which none of the
.meta sym
variables are yet visible. By contrast, the alternate
forms evaluate each
.meta init-form
in an environment in which bindings for the
previous
.meta sym
variables are visible. In this phase of evaluation,
.meta sym
variables are list-valued: one by one they are each bound to the list object
emanating from their corresponding
.metn init-form .
Just before the first loop
iteration, however, the
.meta sym
variables are assigned the first item from each
of their lists.
.TP* Note:
The semantics of
.code collect-each
may be understood in terms of an equivalence to a code pattern involving
.codn mapcar :
.mono
(collect-each ((x xinit) (mapcar (lambda (x y)
(y yinit)) <--> body)
body) xinit yinit)
.onom
The
.code collect-each*
variant may be understood in terms of the following equivalence involving
.code let*
for sequential binding and
.codn mapcar :
.mono
(collect-each* ((x xinit) (let* ((x xinit)
(y yinit)) <--> (y yinit))
body) (mapcar (lambda (x y)
body)
x y))
.onom
However, note that the
.code let*
as well as each invocation of the
.code lambda
binds fresh instances of the variables, whereas these operators are permitted
to bind a single instance of the variables, which are first initialized with
the initializing expressions, and then reused as iteration variables which are
stepped by assignment.
The other operators may be understood likewise, with the substitution
of the
.code mapdo
function in the case of
.code each
and
.code each*
and of the
.code mappend
function in the case of
.code append-each
and
.codn append-each* .
.TP* Example:
.mono
;; print numbers from 1 to 10 and whether they are even or odd
(each* ((n 1..11) ;; n is just a range object in this scope
(even (collect-each ((m n)) (evenp m))))
;; n is an integer in this scope
(format t "~s is ~s\en" n (if even "even" "odd")))
.onom
.TP* Output:
.mono
1 is "odd"
2 is "even"
3 is "odd"
4 is "even"
5 is "odd"
6 is "even"
7 is "odd"
8 is "even"
9 is "odd"
10 is "even"
.onom
.coNP Operators @ for and @ for*
.synb
.mets ({for | for*} >> ({ sym | >> ( sym << init-form )}*)
.mets \ \ \ \ \ \ \ \ \ \ \ \ \ >> ([ test-form << result-form *])
.mets \ \ \ \ \ \ \ \ \ \ \ \ \ <> ( inc-form *)
.mets \ \ << body-form *)
.syne
.desc
The
.code for
and
.code for*
operators combine variable binding with loop iteration.
The first argument is a list of variables with optional initializers,
exactly the same as in the
.code let
and
.code let*
operators. Furthermore, the
difference between
.code for
and
.code for*
is like that between
.code let
and
.code let*
with regard to this list of variables.
The
.code for
and
.code for*
operators execute these steps:
.RS
.IP 1.
Establish an anonymous block over the entire form, allowing
the
.code return
operator to be used to terminate the loop.
.IP 2.
Establish bindings for the specified variables similarly to
.code let
and
.codn let* .
The variable bindings are visible over the
.metn test-form ,
each
.metn result-form ,
each
.meta inc-form
and each
.metn body-form .
.IP 3.
Evaluate
.metn test-form .
If
.meta test-form
yields
.codn nil ,
then the loop terminates. Each
.meta result-form
is evaluated, and the value of the last of these
forms is the result value of the loop.
If there are no
.metn result-form s
then the result value is
.codn nil .
If the
.meta test-form
is omitted, then the test
is taken to be true, and the loop does not terminate.
.IP 4.
Otherwise, if
.meta test-form
yields true, then each
.meta body-form
is evaluated in turn. Then, each
.code inc-form
is evaluated in turn and processing resumes at step 2.
.RE
.IP
Furthermore, the
.code for
and
.code for*
operators establish an anonymous block,
allowing the
.code return
operator to be used to terminate at any point.
.coNP Macros @ doloop and @ doloop*
.synb
.mets ({doloop | doloop*}
.mets \ \ ({ sym | >> ( sym >> [ init-form <> [ step-form ])}*)
.mets \ \ >> ([ test-form << result-form *])
.mets \ \ << tagbody-form *)
.syne
.desc
The
.code doloop
and
.code doloop*
macros provide an iteration construct inspired by the ANSI Common Lisp
.code do
and
.code do*
macros.
Each
.meta sym
element in the form must be a symbol suitable for use as a variable name.
The
.metn tagbody-form s
are placed into an implicit
.codn tagbody ,
meaning that a
.meta tagbody-form
which is an integer, character or symbol is interpreted
as a
.code tagbody
label which may be the target of a control transfer via the
.code go
macro.
The
.code doloop
macro binds each
.meta sym
to the value produced by evaluating the adjacent
.metn init-form .
Then, in the environment in which these variables now exist,
.meta test-form
is evaluated. If that form yields
.codn nil ,
then the loop terminates. The
.metn result-form s
are evaluated, and the value of the last one is returned.
If
.metn result-form s
are absent, then
.code nil
is returned.
If
.meta test-form
is also absent, then the loop terminates and returns
.codn nil .
If
.meta test-form
produces a true value, then
.metn result-form s
are not evaluated. Instead, the implicit
.code tagbody
comprised of the
.metn tagbody-form s
is evaluated.
If that evaluation terminates normally, the loop variables are
then updated by assigning to each
.meta sym
the value of
.metn step-form .
The following defaulting behaviors apply in regard to the variable
syntax. For each
.meta sym
which has an associated
.meta init-form
but no
.metn step-form ,
the
.meta init-form
is duplicated and taken as the
.metn step-form .
Thus a variable specification like
.code "(x y)"
is equivalent to
.codn "(x y y)" .
If both forms are omitted, then the
.meta init-form
is taken to be
.codn nil ,
and the
.meta step-form
is taken to be
.metn sym .
This means that the variable form
.code "(x)"
is equivalent to
.code "(x nil x)"
which has the effect that
.code x
retains its current value when the next loop iteration begins.
Lastly, the
.meta sym
variant is equivalent to
.mono
.meti <> ( sym )
.onom
so that
.code x
is also equivalent to
.codn "(x nil x)" .
The differences between
.code doloop
and
.code doloop*
are:
.code doloop
binds the variables in parallel, similarly to
.codn let ,
whereas
.code doloop*
binds sequentially, like
.codn let* ;
moreover,
.code doloop
performs the
.meta step-form
assignments in parallel as if using a single
.mono
.meti (pset < sym0 < step-form-0 < sym1 < step-form-1 ...)
.onom
form, whereas
.code doloop*
performs the assignment sequentially as if using
.code set
rather than
.codn pset .
The
.code doloop
and
.code doloop*
macros establish an anonymous
.codn block ,
allowing early return from the loop, with a value, via the
.code return
operator.
.TP* "Dialect Note:"
These macros are substantially different from the ANSI Common Lisp
.code do
and
.code do*
macros. Firstly, the termination logic is inverted; effectively they
implement "while" loops, whereas their ANSI CL counterparts implement
"until" loops. Secondly, in the ANSI CL macros, the defaulting of
the missing
.meta step-form
is different. Variables with no
.meta step-form
are not updated. In particular, this means that the form
.code "(x y)"
is not equivalent to
.codn "(x y y)" ;
the ANSI CL macros do not feature the automatic replication of
.meta init-form
into the
.meta step-form
position.
.coNP Macros @, sum-each @, sum-each* @ mul-each and @ mul-each*
.synb
.mets (sum-each >> ({( sym << init-form )}*) << body-form *)
.mets (sum-each* >> ({( sym << init-form )}*) << body-form *)
.mets (mul-each >> ({( sym << init-form )}*) << body-form *)
.mets (mul-each* >> ({( sym << init-form )}*) << body-form *)
.syne
.desc
The macros
.codn sum-each ,
and
.code mul-each
behave very similarly to the
.code each
operator. Whereas the
.code each
operator form returns
.code nil
as its result, the
.code sum-each
and
.code mul-each
forms, if they execute to completion and return normally, return
an accumulated value.
The
.code sum-each
macro initializes newly instantiated, hidden accumulator variable
to the value
.codn 0 .
For each iteration of the loop, the
.metn body-form s
are evaluated, and are expected to produce a value. This value
is added to the current value of the hidden accumulator using the
.code +
function, and the result is stored into the accumulator. If
.code sum-each
returns normally, then the value of this accumulator becomes its
resulting value.
The
.code mul-each
macro similarly initializes a hidden accumulator to the value
.codn 1 .
The value from each iteration of the body is multiplied with
the accumulator using the
.code *
function, and the result is stored into the accumulator. If
.code mul-each
returns normally, then the value of this accumulator becomes
its resulting value.
The
.code sum-each*
and
.code mul-each*
variants of the macros implement the sequential scoping rule for
the variable bindings, exactly the way
.code each*
alters the semantics of
.codn each .
The
.metn body-form s
are enclosed in an implicit anonymous block. If the forms terminate
by returning from the anonymous block then these macros terminate
with the specified value.
When
.code sum-each*
and
.code sum-each
are specified with variables whose values specify zero iterations,
or with no variables at all, the form terminates with a value of
.codn 0 .
In this situation,
.code mul-each
and
.code mul-each*
terminate with
.codn 1 .
Note that this behavior differs from
.codn each ,
and its closely-related operators, which loop infinitely when no variables are
specified.
It is unspecified whether
.code mul-each
and
.code mul-each*
continue iterating when the accumulator takes on a value satisfying the
.code zerop
predicate.
.coNP Macros @, each-true @, some-true @ each-false and @ some-false
.synb
.mets (each-true >> ({( sym << init-form )}*) << body-form *)
.mets (some-true >> ({( sym << init-form )}*) << body-form *)
.mets (each-false >> ({( sym << init-form )}*) << body-form *)
.mets (some-false >> ({( sym << init-form )}*) << body-form *)
.syne
.desc
These macros iterate zero or more variables over sequences, similarly to the
.code each
operator, and calculate logical results, with short-circuiting semantics.
The
.code each-true
macro initializes an internal result variable to the
.code t
value. It then evaluates the
.metn body-form s
for each tuple of variable values, replacing the result variable with
the value produced by these forms. If that value is
.codn nil ,
the iteration stops. When the iteration terminates normally, the
value of the result variable is returned.
If no variables are specified, termination occurs immediately.
Note that this is different from the
.code each
operator, which iterates indefinitely if no variables are specified.
The
.metn body-form s
are surrounded by an implicit anonymous block, making it possible
to terminate via
.code return
or
.codn return-from .
In these cases, the form terminates with
.code nil
or the specified return value. The internal result is ignored.
The
.code some-true
macro is similar to
.codn each-true ,
with the following differences.
The internal result variable is initialized to
.code nil
rather than
.codn t .
The iteration stops whenever the
.metn body-form s
produce a true value, and that value is returned.
The
.code each-false
and
.code some-false
macros are, respectively, similar to
.code each-true
and
.codn some-true ,
with one difference. After each iteration, the value produced by the
.metn body-form s
is logically inverted using the
.code not
function prior to being assigned to the result variable.
.TP* Examples:
.verb
(each-true ()) -> t
(each-true ((a ()))) -> t
(each-true ((a '(1 2 3))) a) -> 3
(each-true ((a '(1 2 3))
(b '(4 5 6)))
(< a b))
-> t
(each-true ((a '(1 2 3))
(b '(4 0 6)))
(< a b))
-> nil
(some-true ((a '(1 2 3))) a) -> 1
(some-true ((a '(nil 2 3))) a) -> 2
(some-true ((a '(nil nil nil))) a) -> nil
(some-true ((a '(1 2 3))
(b '(4 0 6)))
(< a b))
-> t
(some-true ((a '(1 2 3))
(b '(0 1 2)))
(< a b))
-> nil
(each-false ((a '(1 2 3))
(b '(4 5 6)))
(> a b))
-> t
(each-false ((a '(1 2 3))
(b '(4 0 6)))
(> a b))
-> nil
(some-false ((a '(1 2 3))
(b '(4 0 6)))
(> a b))
-> t
(some-false ((a '(1 2 3))
(b '(0 1 2)))
(> a b))
-> nil
.brev
.coNP Macros @, each-prod @ collect-each-prod and @ append-each-prod
.synb
.mets (each-prod >> ({( sym << init-form )}*) << body-form *)
.mets (collect-each-prod >> ({( sym << init-form )}*) << body-form *)
.mets (append-each-prod >> ({( sym << init-form )}*) << body-form *)
.syne
.desc
The macros
.codn each-prod ,
.code collect-each-prod
and
.code append-each-prod
have a similar syntax to
.codn each ,
.code collect-each
and
.codn collect-each-prod .
However, instead of iterating over sequences in parallel, they iterate over
the Cartesian product of the elements from the sequences.
The difference between
.code collect-each
and
.code collect-each-prod
is analogous to that between the functions
.code mapcar
and
.codn maprod .
Like in the
.code each
operator family, the
.metn body-form s
are surrounded by an anonymous block. If these forms execute a return from
this block, then these macros terminate with the specified return value.
When no iterations are performed, including in the case when an empty
list of variables is specified, all these macro forms terminate and return
.codn nil .
Note that this behavior differs from
.codn each ,
and its closely-related operators, which loop infinitely when no variables are
specified.
With one caveat noted below, these macros can be understood as providing
syntactic sugar according to the pattern established by the following
equivalences:
.mono
(each-prod (block nil
((x xinit) (let ((#:gx xinit) (#:gy yinit))
(y yinit)) <--> (maprodo (lambda (x y)
body) body)
#:gx #:gy))
(collect-each-prod (block nil
((x xinit) (let ((#:gx xinit) (#:gy yinit))
(y yinit)) <--> (maprod (lambda (x y)
body) body)
#:gx #:gy))
(append-each-prod (block nil
((x xinit) (let ((#:gx xinit) (#:gy yinit))
(y yinit)) <--> (maprend (lambda (x y)
body) body)
#:gx #:gy))
.onom
However, note that each invocation of the
.code lambda
binds fresh instances of the variables, whereas these operators are
permitted to bind a single instance of the variables, which are then stepped by
assignment.
.TP* Example:
.mono
(collect-each-prod ((a '(a b c))
(n #(1 2)))
(cons a n))
--> ((a . 1) (a . 2) (b . 1) (b . 2) (c . 1) (c . 2))
.onom
.coNP Macros @, each-prod* @ collect-each-prod* and @ append-each-prod*
.synb
.mets (each-prod* >> ({( sym << init-form )}*) << body-form *)
.mets (collect-each-prod* >> ({( sym << init-form )}*) << body-form *)
.mets (append-each-prod* >> ({( sym << init-form )}*) << body-form *)
.syne
.desc
The macros
.codn each-prod* ,
.code collect-each-prod*
and
.code append-each-prod*
are variants of
.codn each-prod ,
.code collect-each-prod
and
.code append-each-prod
with sequential binding.
These macros can be understood as providing syntactic sugar according to the
pattern established by the following equivalences:
.mono
(each-prod* (let* ((x xinit)
((x xinit) (y yinit))
(y yinit)) <--> (maprodo (lambda (x y) body)
body) x y)
(collect-each-prod* (let* ((x xinit)
((x xinit) (y yinit))
(y yinit)) <--> (maprod (lambda (x y) body)
body) x y)
(append-each-prod* (let* ((x xinit)
((x xinit) (y yinit))
(y yinit)) <--> (maprend (lambda (x y) body)
body) x y)
.onom
However, note that the
.code let*
as well as each invocation of the
.code lambda
binds fresh instances of the variables, whereas these operators are permitted
to bind a single instance of the variables, which are first initialized with
the initializing expressions, and then reused as iteration variables which are
stepped by assignment.
.TP* Example:
.mono
(collect-each-prod* ((a "abc")
(b (upcase-str a)))
`@a@b`)
--> ("aA" "aB" "aC" "bA" "bB" "bC" "cA" "cB" "cC")
.onom
.coNP Macros @, sum-each-prod @, sum-each-prod* @ mul-each-prod and @ mul-each-prod*
.synb
.mets (sum-each-prod >> ({( sym << init-form )}*) << body-form *)
.mets (sum-each-prod* >> ({( sym << init-form )}*) << body-form *)
.mets (mul-each-prod >> ({( sym << init-form )}*) << body-form *)
.mets (mul-each-prod* >> ({( sym << init-form )}*) << body-form *)
.syne
.desc
The macros
.code sum-each-prod
and
.code mul-each-prod
have a similar syntax to
.code sum-each
and
.codn mul-each .
However, instead of iterating over sequences in parallel, they iterate over
the Cartesian product of the elements from the sequences.
The macros
.code sum-each-prod*
and
.code mul-each-prod*
variants perform sequential variable binding when establishing the initial
values of the variables, similarly to the
.code each*
operator.
The
.metn body-form s
are surrounded by an implicit anonymous block. If these forms execute a return
from this block, then these macros terminate with the specified return value.
When no iterations are specified, including in the case when an empty
list of variables is specified, the summing macros terminate, yielding
.codn 0 ,
and the multiplicative macros terminate with
.codn 1 .
Note that this behavior differs from
.codn each ,
and its closely-related operators, which loop infinitely when no variables are
specified.
.TP* Examples:
.verb
;; Inefficiently calculate (+ (* 1 2 3) (* 4 3 2)).
;; Every value from (1 2 3) is paired with every value
;; from (4 3 2) to form a partial products, and
;; sum-each-prod adds these together implicitly:
(sum-each-prod ((x '(1 2 3))
(y '(4 3 2)))
(* x y))
-> 54
.brev
.coNP Operators @ block and @ block*
.synb
.mets (block < name << body-form *)
.mets (block* < name-form << body-form *)
.syne
.desc
The
.code block
operator introduces a named block around the execution of
some forms. The
.meta name
argument may be any object, though block names are usually symbols.
Two block
.meta name
objects are considered to be the same name according to
.code eq
equality.
Since a block name is not
a variable binding, keyword symbols are permitted, and so are the symbols
.code t
and
.codn nil .
A block named by the symbol nil is slightly special: it is
understood to be an anonymous block.
The
.code block*
operator differs from
.code block
in that it evaluates
.metn name-form ,
which is expected to produce a symbol. The resulting symbol
is used for the name of the block.
A named or anonymous block establishes an exit point for the
.code return-from
or
.code return
operator, respectively. These operators can be invoked within a block
to cause its immediate termination with a specified return value.
A block also establishes a prompt for a
.IR "delimited continuation" .
Anywhere in a block, a continuation can be captured using the
.code sys:capture-cont
function. Delimited continuations are described in the section
Delimited Continuations. A delimited continuation allows an apparently
abandoned block to be restarted at the capture point, with the
entire call chain and dynamic environment between the prompt and the capture
point intact.
Blocks in \*(TL have dynamic scope. This means that the following
situation is allowed:
.verb
(defun func () (return-from foo 42))
(block foo (func))
.brev
The function can return from the
.code foo
block even though the
.code foo
block
does not lexically surround
.codn foo .
It is because blocks are dynamic that the
.code block*
variant exists; for lexically scoped blocks, it would make little
sense to have support a dynamically computed name.
Thus blocks in \*(TL provide dynamic nonlocal returns, as well
as returns out of lexical nesting.
It is permitted for blocks to be aggressively
.codn progn -converted
by compilation. This means that a
.code block
form which meets certain criteria is converted to a
.code progn
form which surrounds the
.metn body-form s
and thus no longer establishes an exit point.
A
.code block
form will be spared from
.codn progn -conversion
by the compiler if it meets the following rules.
.RS
.IP 1.
Any
.meta body-form
references the block's
.meta name
in a
.codn return ,
.codn return-from ,
.code sys:abscond-from
or
.code sys:capture-cont
expression.
.IP 2.
The form contains at least one direct call to a function
not in the standard \*(TL library.
.IP 3.
The form contains at least one direct call to the functions
.codn sys:capture-cont ,
.codn return* ,
.codn sys:abscond* ,
.codn match-fun ,
.codn eval ,
.codn load ,
.codn compile ,
.code compile-file
or
.codn compile-toplevel .
.IP 4.
The form references any of the functions in rules 2 and 3
as a function binding via the
.code dwim
operator (or the DWIM brackets notation) or via the
.code fun
operator.
.IP 5.
The form is a
.code block*
form; these are spared from the optimization.
.RE
.IP
Removal of blocks under the above rules means that some use of blocks which
works in interpreted code will not work in compiled programs. Programs which
adhere to the rules are not affected by such a difference.
Additionally, the compiler may
.codn progn -convert
blocks in contravention of the above rules, but only if doing so makes no
difference to visible program behavior.
.TP* Examples:
.verb
(defun helper ()
(return-from top 42))
;; defun implicitly defines a block named top
(defun top ()
(helper) ;; function returns 42
(prinl 'notreached)) ;; never printed
(defun top2 ()
(let ((h (fun helper)))
(block top (call h)) ;; may progn-convert
(block top (call 'helper)) ;; may progn-convert
(block top (helper)))) ;; not removed
.brev
In the above examples, the block containing
.code "(call h)"
may be converted to
.code progn
because it doesn't express a
.B direct
call to the
.code helper
function. The block which calls
.code helper
using
.code "(call 'helper)"
is also not considered to be making a direct call.
.TP* "Dialect Note:"
In Common Lisp, blocks are lexical. A separate mechanism consisting of
catch and throw operators performs nonlocal transfer based on symbols.
The \*(TL example:
.verb
(defun func () (return-from foo 42))
(block foo (func))
.brev
is not allowed in Common Lisp, but can be transliterated to:
.verb
(defun func () (throw 'foo 42))
(catch 'foo (func))
.brev
Note that foo is quoted in CL. This underscores the dynamic nature of
the construct.
.code throw
itself is a function and not an operator. Also note that the CL
example, in turn, is even more closely transcribed back into
\*(TL simply by replacing its
.code throw
and
.code catch
with
.code return*
and
.codn block* :
.verb
(defun func () (return* 'foo 42))
(block* 'foo (func))
.brev
Common Lisp blocks also do not support delimited continuations.
.coNP Operators @ return and @ return-from
.synb
.mets (return <> [ value ])
.mets (return-from < name <> [ value ])
.syne
.desc
The
.code return
operator must be dynamically enclosed within an anonymous
block (a block named by the symbol
.codn nil ).
It immediately terminates the
evaluation of the innermost anonymous block which encloses it, causing
it to return the specified value. If the value is omitted, the anonymous
block returns
.codn nil .
The
.code return-from
operator must be dynamically enclosed within a named block
whose name matches the
.meta name
argument. It immediately terminates the
evaluation of the innermost such block, causing it to return the specified
value. If the value is omitted, that block returns
.codn nil .
.TP* Example:
.verb
(block foo
(let ((a "abc\en")
(b "def\en"))
(pprint a *stdout*)
(return-from foo 42)
(pprint b *stdout*)))
.brev
Here, the output produced is
.strn "abc" .
The value of
.code b
is not printed
because.
.code return-from
terminates block
.codn foo ,
and so the second pprint form is not evaluated.
.coNP Function @ return*
.synb
.mets (return* < name <> [ value ])
.syne
.desc
The
.code return*
function is similar to the
.code return-from
operator, except that
.code name
is an ordinary function parameter, and so when
.code return*
is used, an argument expression must be specified which evaluates
to a symbol. Thus
.code return*
allows the target block of a return to be dynamically computed.
The following equivalence holds between the operator and function:
.verb
(return-from a b) <--> (return* 'a b)
.brev
Expressions used as
.meta name
arguments to
.code return*
which do not simply quote a symbol have no equivalent in
.codn return-from .
.coNP Macros @ tagbody and @ go
.synb
.mets (tagbody >> { form | << label }*)
.mets (go << label )
.syne
.desc
The
.code tagbody
macro provides a form of the "go to" control construct. The arguments of a
.code tagbody
form are a mixture of zero or more forms and
.IR "go labels" .
The latter consist of those arguments which are symbols, integers or
characters. Labels are not considered by
.code tagbody
and
.code go
to be forms, and are not subject to macro expansion or evaluation.
The
.code go
macro is available inside
.codn tagbody .
It is erroneous for a
.code go
form to occur outside of a
.codn tagbody .
This situation is diagnosed by global macro called
.codn go ,
which unconditionally throws an error.
In the absence of invocations of
.code go
or other control transfers, the
.code tagbody
macro evaluates each
.meta form
in left-to-right order. The
.code go
labels are ignored.
After the last
.meta form
is evaluated, the
.code tagbody
form terminates, and yields
.codn nil .
Any
.meta form
itself, or else any of its subforms, may be the form
.mono
.meti (go << label )
.onom
where
.meta label
matches one of the
.code go
labels of a surrounding
.codn tagbody .
When this
.code go
form is evaluated, then the evaluation of
.meta form
is immediately abandoned, and control transfers to the specified
label. The forms are then evaluated in left-to-right order starting
with the form immediately after that label. If the label is not
followed by any forms, then the
.code tagbody
terminates. If
.meta label
doesn't match to any label in any surrounding
.codn tagbody ,
the
.code go
form is erroneous.
The abandonment of a
.meta form
by invocation of
.code go
is a dynamic transfer. All necessary unwinding inside
.meta form
takes place.
The
.code go
labels are lexically scoped, but dynamically bound. Their scope
being lexical means that the labels are not visible to forms which are not
enclosed within the
.codn tagbody ,
even if their evaluation is invoked from that
.codn tagbody .
The dynamic binding means that the labels of a
.code tagbody
form are established when it begins evaluating, and removed when
that form terminates. Once a label is removed, it is not available
to be the target of a
.code go
control transfer, even if that
.code go
form has the label in its lexical scope. Such an attempted transfer
is erroneous.
It is permitted for
.code tagbody
forms to nest arbitrarily. The labels of an inner
.code tagbody
are not visible to an outer
.codn tagbody .
However, the reverse is true: a
.code go
form in an inner
.code tagbody
may branch to a label in an outer
.codn tagbody ,
in which case the entire inner
.code tagbody
terminates.
In cases where the same objects are used as labels
by an inner and outer
.codn tagbody ,
the inner labels shadow the outer labels.
There is no restriction on what kinds of symbols may be labels.
Symbols in the
.code keyword
package as well as the symbols
.code t
and
.code nil
are valid
.code tagbody
labels.
.TP* "Dialect Note:"
ANSI Common Lisp
.code tagbody
supports only symbols and integers as labels (which are called "go tags");
characters are not supported.
.TP* Examples:
.verb
;; print the numbers 1 to 10
(let ((i 0))
(tagbody
(go skip) ;; forward goto skips 0
again
(prinl i)
skip
(when (<= (inc i) 10)
(go again))))
;; Example of erroneous usage: by the time func is invoked
;; by (call func) the tagbody has already terminated. The
;; lambda body can still "see" the label, but it doesn't
;; have a binding.
(let (func)
(tagbody
(set func (lambda () (go label)))
(go out)
label
(prinl 'never-reached)
out)
(call func))
;; Example of unwinding when the unwind-protect
;; form is abandoned by (go out). Output is:
;; reached
;; cleanup
;; out
(tagbody
(unwind-protect
(progn
(prinl 'reached)
(go out)
(prinl 'notreached))
(prinl 'cleanup))
out
(prinl 'out))
.brev
.coNP Macros @ prog and @ prog*
.synb
.mets (prog >> ({ sym | >> ( sym << init-form )}*)
.mets \ \ >> { body-form | << label }*)
.mets (prog* >> ({ sym | >> ( sym << init-form )}*)
.mets \ \ >> { body-form | << label }*)
.syne
.desc
The
.code prog
and
.code progn*
macros combine the features of
.code let
and
.codn let* ,
respectively,
anonymous block and
.codn tagbody .
The
.code prog
macro treats the
.meta sym
and
.code init-form
expressions similarly to
.codn let ,
establishing variable bindings in parallel.
The
.code prog*
macro treats these expressions in a similar way to
.codn let* .
The forms enclosed are treated like the argument forms of the
.code tagbody
macro: labels are permitted, along with use of
.codn go .
Finally, an anonymous block is established around all of the enclosed
forms (both the
.metn init-form s
and
.metn body-forms s)
allowing the use of
.code return
to terminate evaluation with a value.
The
.code prog
macro may be understood according to the following equivalence:
.verb
(prog vars forms ...) <--> (block nil
(let vars
(tagbody forms ...)))
.brev
Likewise, the
.code prog*
macro follows an analogous equivalence, with
.code let
replaced by
.codn let* .
.SS* Evaluation
.coNP Function @ eval
.synb
.mets (eval < form <> [ env ])
.syne
.desc
The
.code eval
function treats the
.meta form
object as a Lisp expression, which is expanded and
evaluated. The side effects implied by the form are performed, and the value
which it produces is returned. The optional
.meta env
object specifies an environment for
resolving the function and variable references encountered in the expression.
If this argument is omitted,
then evaluation takes place in the global environment.
The
.meta form
is not expanded all at once. Rather, it is treated by the following algorithm:
.RS
.IP 1.
First, if
.meta form
is a macro, it is macro-expanded as if by an application of the function
.codn macroexpand .
.IP 2.
If the resulting expanded form is a
.codn progn ,
.codn compile-only ,
or
.code eval-only
form, then
.code eval
iterates over that form's argument expressions, passing each expression to a
recursive call to
.code eval
using the same
.metn env .
.IP 3.
Otherwise, if the expanded form isn't one of the above three kinds of
expressions, it is subject to a full expansion and evaluation.
.RE
.IP
This algorithm allows a sequence of top-level forms to be combined into a
single top-level form, even when the expansion of forms occurring later in the
sequence depends on the evaluation effects of forms earlier in the sequence.
For instance, a form like
.code "(progn (defmacro foo ()) (foo))"
may be processed with
.codn eval ,
because the above algorithm ensures that the
.code "(defmacro foo ())"
expression is fully evaluated first, thereby providing the
macro definition required by
.codn "(foo)" .
This expansion and evaluation order is important because the semantics of
.code eval
forms the reference model for how the
.code load
function processes top-level forms.
Note that, according to these rules, the constituent body forms of a
.code macrolet
or
.code symacrolet
top-level form are not individual top-level forms, even if the
expansion of the construct combines the expanded versions of those
forms with
.codn progn .
The form
.code "(macrolet () (defmacro foo ()) (foo))"
will therefore not work correctly. However, the specific problem in
this situation can be be resolved by rewriting
.code foo
as a
.code macrolet
macro:
.codn "(macrolet ((foo ())) (foo))" .
See also: the
.code make-env
function.
.coNP Function @ constantp
.synb
.mets (constantp < form <> [ env ])
.syne
.desc
The
.code constantp
function determines whether
.meta form
is a constant form, with respect to environment
.metn env .
If
.meta env
is absent, the global environment is used.
The
.meta env
argument is used for fully expanding
.meta form
prior to analyzing.
Currently,
.code constantp
returns true for any form which, after macro-expansion, is any of
the following: a compound form with the symbol
.code quote
in its first position; a non-symbolic atom; or one of the symbols
which evaluate to themselves and cannot be bound as variables.
These symbols are the keyword symbols, and the symbols
.code t
and
.codn nil .
Additionally,
.code constantp
returns true for a compound form, or a DWIM form, whose symbol is
the member of a set a large number of constant-foldable library functions,
and whose arguments are, recursively,
.code constantp
expressions for the same environment. The arithmetic functions
are members of this set.
For all other inputs,
.code constantp
returns
.codn nil .
Note: some uses of
.code constantp
require manual expansion.
.TP* Examples:
.verb
(constantp nil) -> t
(constantp t) -> t
(constantp :key) -> t
(constantp :) -> t
(constantp 'a) -> nil
(constantp 42) -> t
(constantp '(+ 2 2 [* 3 (/ 4 4)])) -> t
;; symacrolet form expands to 42, which is constant
(constantp '(symacrolet ((a 42)) a))
(defmacro cp (:env e arg)
(constantp arg e))
;; macro call (cp 'a) is replaced by t because
;; the symbol a expands to (+ 2 2) in the given environment,
;; and so (* a a) expands to (* (+ 2 2) (+ 2 2)) which is constantp.
(symacrolet ((a (+ 2 2)))
(cp '(* a a))) -> t
.brev
.coNP Function @ make-env
.synb
.mets (make-env >> [ var-bindings >> [ fun-bindings <> [ next-env ]]])
.syne
.desc
The
.code make-env
function creates an environment object suitable as the
.code env
parameter.
The
.meta var-bindings
and
.meta fun-bindings
parameters, if specified,
should be association lists, mapping symbols to objects. The objects in
.meta fun-bindings
should be functions, or objects callable as functions.
The
.meta next-env
argument, if specified, should be an environment.
Note: bindings can also be added to an environment using the
.code env-vbind
and
.code env-fbind
functions.
.coNP Functions @ env-vbind and @ env-fbind
.synb
.mets (env-vbind < env < symbol << value )
.mets (env-fbind < env < symbol << value )
.syne
.desc
These functions bind a symbol to a value in either the function or variable
space of environment
.codn env .
Values established in the function space should be functions or objects that
can be used as functions such as lists, strings, arrays or hashes.
If
.meta symbol
already exists in the environment, in the given space, then its
value is updated with
.codn value .
If
.meta env
is specified as
.codn nil ,
then the binding takes place in the global environment.
.coNP Functions @, env-vbindings @ env-fbindings and @ env-next
.synb
.mets (env-vbindings << env )
.mets (env-fbindings << env )
.mets (env-next << env )
.syne
.desc
These function retrieve the components of
.metn env ,
which must be an environment. The
.code env-vbindings
function retrieves the association list representing variable
bindings. Similarly, the
.code env-fbindings
retrieves the association list of function bindings.
The
.code env-next
function retrieves the next environment, if
.meta env
has one, otherwise
.codn nil .
If
.code e
is an environment constructed by the expression
.codn "(make-env v f n)" ,
then
.code "(env-vbindings e)"
retrieves
.codn v ,
.code "(env-fbindings e)"
retrieves
.code f
and
.code "(env-next e)"
returns
.codn n .
.SS* Global Environment
.coNP Accessors @, symbol-function @ symbol-macro and @ symbol-value
.synb
.mets (symbol-function >> { symbol | < method-name | << lambda-expr })
.mets (symbol-macro << symbol )
.mets (symbol-value << symbol )
.mets (set (symbol-function >> { symbol | << method-name }) << new-value )
.mets (set (symbol-macro << symbol ) << new-value )
.mets (set (symbol-value << symbol ) << new-value )
.syne
.desc
If given a
.meta symbol
argument, the
.code symbol-function
function retrieves the value of the global function binding of the
given
.meta symbol
if it has one: that is, the function object bound to the
.metn symbol .
If
.meta symbol
has no global function binding, then
.code nil
is returned.
The
.code symbol-function
function supports method names of the form
.mono
.meti (meth < struct << slot )
.onom
where
.meta struct
names a struct type, and
.meta slot
is either a static slot or one of the keyword symbols
.code :init
or
.code :postinit
which refer to special functions associated with a structure type.
Names in this format are returned by the
.meta func-get-name
function. The
.code symbol-function
function also supports names of the form
.mono
.meti (macro << name )
.onom
which denote macros. Thus,
.code symbol-function
provides unified access to functions, methods and macros.
If a
.code lambda
expression is passed to
.codn symbol-function ,
then the expression is macro-expanded and if that is successful, the function
implied by that expression is returned.
It is unspecified whether this function is interpreted or compiled.
The
.code symbol-macro
function retrieves the value of the global macro binding of
.meta symbol
if it has one.
Note: the name of this function has nothing to do with symbol macros;
it is named for consistency with
.code symbol-function
and
.codn symbol-value ,
referring to the "macro-expander binding of the symbol cell".
The value of a macro binding is a function object.
Intrinsic macros are C functions in the \*(TX kernel, which receive
the entire macro call form and macro environment, performing their
own destructuring. Currently, macros written in \*(TL are represented
as curried C functions which carry the following list object in their
environment cell:
.mono
.mets (#<environment object> < macro-parameter-list << body-form *)
.onom
Local macros created by
.code macrolet
have
.code nil
in place of the environment object.
This representation is likely to change or expand to include other
forms in future \*(TX versions.
The
.code symbol-value
function retrieves the value stored in the dynamic binding of
.meta symbol
that is apparent in the current context. If the variable has no dynamic
binding, then
.code symbol-value
retrieves its value in the global environment.
If
.meta symbol
has no variable binding, but is defined as a global symbol macro,
then the value of that symbol macro binding is retrieved.
The value of a symbol macro binding is simply the replacement form.
Rather than throwing an exception, each of these functions returns
.code nil
if the argument symbol doesn't have the binding in the respective
namespace or namespaces which that function searches.
A
.codn symbol-function ,
.codn symbol-macro ,
or
.code symbol-value
form denotes a place, if
.meta symbol
has a binding of the respective kind. This place may be assigned to or
deleted. Assignment to the place causes the denoted binding to have a new
value. Deleting a place with the
.code del
macro removes the binding,
and returns the previous contents of that binding. A binding
denoted by a
.code symbol-function
form is removed using
.codn fmakunbound ,
one denoted by by
.code symbol-macro
is removed using
.code mmakunbound
and a binding denoted by
.code symbol-value
is removed using
.codn makunbound .
Deleting a method via
.code symbol-function
is not possible; an attempt to do so has no effect.
Storing a value, using any one of these three accessors, to a nonexistent
variable, function or macro binding, is not erroneous. It has has the effect of
creating that binding.
Using
.code symbol-function
accessor to assign to a lambda expression is erroneous.
Deleting a binding, using any of these three accessors, when the binding does not
exist, also isn't erroneous. There is no effect and the
.code del
operator yields
.code nil
as the prior value, consistent with the behavior when accessors are used to
retrieve a nonexistent value.
.TP* "Dialect Note:"
In ANSI Common Lisp, the
.code symbol-function
function retrieves a function, macro or special operator binding
of a symbol.
These are all in one space and may not coexist. In \*(TL, it
retrieves a symbol's function binding only. Common Lisp has an accessor
named
.code macro-function
similar to
.codn symbol-macro .
.coNP Functions @, boundp @ fboundp and @ mboundp
.synb
.mets (boundp << symbol )
.mets (fboundp >> { symbol | < method-name | << lambda-expr })
.mets (mboundp << symbol )
.syne
.desc
.code boundp
returns
.code t
if the
.meta symbol
is bound as a variable or symbol macro in the global
environment, otherwise
.codn nil .
.code fboundp
returns
.code t
if the
.meta symbol
has a function binding in the global
environment, the method specified by
.meta method-name
exists, or a lambda expression argument is given.
Otherwise it returns
.codn nil .
.code mboundp
returns
.code t
if the symbol has an operator macro binding in the global environment,
otherwise
.codn nil .
.TP* "Dialect Notes:"
The
.code boundp
function in ANSI Common Lisp doesn't report that
global symbol macros have a binding. They are not considered
bindings. In \*(TL, they are considered bindings.
The ANSI Common Lisp
.code fboundp
yields true if its argument has a function, macro or operator
binding. The behavior of the Common Lisp expression
.code "(fboundp x)"
in Common Lisp can be obtained in \*(TL using the
.verb
(or (fboundp x) (mboundp x) (special-operator-p x))
.brev
expression.
The
.code mboundp
function doesn't exist in ANSI Common Lisp.
.coNP Function @ makunbound
.synb
.mets (makunbound << symbol )
.syne
.desc
The function
.code makunbound
the binding of
.meta symbol
from either the dynamic environment or the global symbol
macro environment. After the call to
.codn makunbound ,
.meta symbol
appears to be unbound.
If the
.code makunbound
call takes place in a scope in which there exists a dynamic rebinding of
.metn symbol ,
the information for restoring the previous binding is not
affected by
.codn makunbound .
When that scope terminates, the previous binding will be restored.
If the
.code makunbound
call takes place in a scope in which the dynamic binding for
.code symbol
is the global binding, then the global binding is removed.
When the global binding is removed, then
if
.meta symbol
was previously marked as special (for instance
by
.codn defvar )
this marking is removed.
Otherwise if
.meta symbol
has a global symbol macro binding, that binding is removed.
If
.meta symbol
has no apparent dynamic binding, and no global symbol macro binding,
.code makunbound
does nothing.
In all cases,
.code makunbound
returns
.metn symbol .
.TP* "Dialect Note:"
The behavior of
.code makunbound
differs from its counterpart in ANSI Common Lisp.
The
.code makunbound
function in Common Lisp only removes a value from a dynamic variable. The
dynamic variable does not cease to exist, it only ceases to have a value
(because a binding is a value). In \*(TL, the variable ceases to exist. The
binding of a variable isn't its value, it is the variable itself: the
association between a name and an abstract storage location, in some
environment. If the binding is undone, the variable disappears.
The
.code makunbound
function in Common Lisp does not remove global symbol macros,
which are not considered to be bindings in the variable namespace.
That is to say, the Common Lisp
.code boundp
does not report true for symbol macros.
The Common Lisp
.code makunbound
also doesn't remove the special attribute from a symbol. If a variable
is introduced with
.code defvar
and then removed with
.codn makunbound ,
the symbol continues to exhibit dynamic binding rather than lexical
in subsequent scopes. In \*(TL, if a global binding is removed, so
is the special attribute.
.coNP Functions @ fmakunbound and @ mmakunbound
.synb
.mets (fmakunbound << symbol )
.mets (mmakunbound << symbol )
.syne
.desc
The function
.code fmakunbound
removes any binding for
.meta symbol
from the function namespace of the global environment. If
.meta symbol
has no such binding, it does nothing.
In either case, it returns
.metn symbol .
The function
.code mmakunbound
removes any binding for
.meta symbol
from the operator macro namespace of the global environment. If
.meta symbol
has no such binding, it does nothing.
In either case, it returns
.metn symbol .
.TP* "Dialect Note:"
The behavior of
.code fmakunbound
differs from its counterpart in ANSI Common Lisp. The
.code fmakunbound
function in Common Lisp removes a function or macro binding, which
do not coexist.
The
.code mmakunbound
function doesn't exist in Common Lisp.
.coNP Function @ func-get-form
.synb
.mets (func-get-form << func )
.syne
.desc
The
.code func-get-form
function retrieves a source code form of
.metn func ,
which must be an interpreted function. The source code form has the syntax
.mono
.meti >> ( name < arglist << body-form *) .
.onom
.coNP Function @ func-get-name
.synb
.mets (func-get-name < func <> [ env ])
.syne
.desc
The
.code func-get-name
tries to resolve the function object
.meta func
to a name. If that is not possible, it returns
.codn nil .
The resolution is performed by an exhaustive search through
up to three spaces.
If an environment is specified by
.metn env ,
then this is searched first. If a binding is found in that
environment which resolves to the function, then the search
terminates and the binding's symbol is returned as the
function's name.
If the search through environment
.meta env
fails, or if that argument is not specified, then the
global environment is searched for a function binding
which resolves to
.metn func .
If such a binding is found, then the search terminates,
and the binding's symbol is returned. If two or more
symbols in the global environment resolve to the function,
it is not specified which one is returned.
If the global function environment search fails,
then the function is considered as a possible macro.
The global macro environment is searched for a macro
binding whose expander function is
.metn func ,
similarly to the way the function environment was
searched. If a binding is found, then the syntax
.mono
.meti (macro << name )
.onom
is returned, where
.meta name
is the name of the global macro binding that was found
which resolves to
.metn func .
If two or more global macro bindings share
.metn func ,
it is not specified which of those bindings provides
.metn name .
If the global macro search fails, then
.meta func
is considered as a possible method.
The static slot space of all struct types is searched for
a slot which contains
.metn func .
If such a slot is found, then the method name is returned,
consisting of the syntax
.mono
.meti (meth < type << name )
.onom
where
.meta type
is a symbol denoting the struct type and
.meta name
is the static slot of the struct type which holds
.metn func .
A check is also performed whether
.meta func
might be equal to one of the two special functions of
a structure type: its
.meta initfun
or
.metn postinitfun ,
in which case it is returned as either the
.mono
.meti (meth < type :init)
.onom
or the
.mono
.meti (meth < type :postinit)
.onom
syntax.
If
.meta func
is an interpreted function not found under any name,
then a lambda expression denoting that function
is returned in the syntax
.mono
.meti (lambda < args << form *)
.onom
If
.meta func
cannot be identified as a function, then
.code nil
is returned.
.coNP Function @ func-get-env
.synb
.mets (func-get-env << func )
.syne
.desc
The
.code func-get-env
function retrieves the environment object associated with
function
.metn func .
The environment object holds the captured bindings of a
lexical closure.
.coNP Functions @ fun-fixparam-count and @ fun-optparam-count
.synb
.mets (fun-fixparam-count << func )
.mets (fun-optparam-count << func )
.syne
.desc
The
.code fun-fixparam-count
reports
.metn func 's
number of fixed parameters. The fixed parameters consist of the required
parameters and the optional parameters. Variadic functions have a parameter
which captures the remaining arguments which are in excess of the fixed
parameters. That parameter is not considered a fixed parameter and therefore
doesn't contribute to this count.
The
.code fun-optparam-count
reports
.metn func 's
number of optional parameters.
The
.meta func
argument must be a function.
Note: if a function isn't variadic (see the
.meta fun-variadic
function) then the value reported by
.code fun-fixparam-count
represents the maximum number of arguments which can be passed to the function.
The minimum number of required arguments can be calculated for any function by
subtracting the value from
.code fun-optparam-count
from the value from
.codn fun-fixparam-count .
.coNP Function @ fun-variadic
.synb
.mets (fun-variadic << func )
.syne
.desc
The
.code fun-variadic
function returns
.code t
if
.meta func
is a variadic function, otherwise
.codn nil .
The
.meta func
argument must be a function.
.coNP Function @ interp-fun-p
.synb
.mets (interp-fun-p << obj )
.syne
.desc
The
.code interp-fun-p
function returns
.code t
if
.meta obj
is an interpreted function, otherwise it returns
.codn nil .
.coNP Function @ vm-fun-p
.synb
.mets (vm-fun-p << obj )
.syne
.desc
The
.code vm-fun-p
function returns
.code t
if
.meta obj
a function compiled for the virtual machine: a function representation produced
by means of the functions
.codn compile-file ,
.code compile-toplevel
or
.codn compile .
If
.meta obj
is of any other type, the function returns
.codn nil .
.coNP Function @ special-var-p
.synb
.mets (special-var-p << obj )
.syne
.desc
The
.code special-var-p
function returns
.code t
if
.meta obj
is a symbol marked for special variable binding, otherwise it returns
.codn nil .
Symbols are marked special by
.code defvar
and
.codn defparm .
.coNP Function @ special-operator-p
.synb
.mets (special-operator-p << obj )
.syne
.desc
The
.code special-operator-p
function returns
.code t
if
.meta obj
is a symbol which names a special operator, otherwise it returns
.codn nil .
.SS* Object Type
In \*(TL, objects obey the following type hierarchy. In this type hierarchy,
the internal nodes denote abstract types: no object is an instance of
an abstract type. Nodes in square brackets indicate an internal structure
in the type graph, invisible to programs, and angle
brackets indicate a plurality of types which are not listed by name:
.verb
t ----+--- [cobj types] ---+--- hash
| |
| +--- hash-iter
| |
| +--- stream
| |
| +--- random-state
| |
| +--- regex
| |
| +--- buf
| |
| +--- tree
| |
| +--- tree-iter
| |
| +--- seq-iter
| |
| +--- cptr
| |
| +--- dir
| |
| +--- struct-type
| |
| +--- <all structures>
| |
| +--- ... others
|
|
+--- sequence ---+--- string ---+--- str
| | |
| | +--- lstr
| | |
| | +--- lit
| |
| +--- list ---+--- null
| | |
| | +--- cons
| | |
| | +--- lcons
| |
| +--- vec
| |
| +--- <structures with car or length methods>
|
+--- number ---+--- float
| |
| +--- integer ---+--- fixnum
| |
| +--- bignum
|
+--- chr
|
+--- sym
|
+--- env
|
+--- range
|
+--- tnode
|
+--- pkg
|
+--- fun
|
+--- args
.brev
In addition to the above hierarchy, the following relationships also exist:
.verb
t ---+--- atom --- <any type other than cons> --- nil
|
+--- cons ---+--- lcons --- nil
|
+--- nil
sym --- null
struct ---- <all structures>
.brev
That is to say, the types are exhaustively partitioned into atoms and conses;
an object is either a
.code cons
or else it isn't, in which case it is the abstract
type
.codn atom .
The
.code cons
type is odd in that it is both an abstract type,
serving as a supertype for the type
.code lcons
and it is also a concrete type in that regular conses are of
this type.
The type
.code nil
is an abstract type which is empty. That is to say, no object is of
type
.codn nil .
This type is considered the abstract subtype of every other type,
including itself.
The type
.code nil
is not to be confused with the type
.code null
which is the type of the
.code nil
symbol.
Because the type of
.code nil
is the type
.code null
and
.code nil
is also a symbol, the
.code null
type is a subtype of
.codn sym .
Lastly, the symbol
.code struct
serves as the supertype of all structures.
.coNP Function @ typeof
.synb
.mets (typeof << value )
.syne
.desc
The
.code typeof
function returns a symbol representing the type of
.metn value .
The core types are identified by the following symbols:
.coIP cons
Cons cell.
.coIP str
String.
.coIP lit
Literal string embedded in the \*(TX executable image.
.coIP chr
Character.
.coIP fixnum
Fixnum integer: an integer that fits into the value word, not having to
be heap-allocated.
.coIP bignum
A bignum integer: arbitrary precision integer that is heap-allocated.
.coIP float
Floating-point number.
.coIP sym
Symbol.
.coIP pkg
Symbol package.
.coIP fun
Function.
.coIP vec
Vector.
.coIP lcons
Lazy cons.
.coIP range
Range object.
.coIP lstr
Lazy string.
.coIP env
Function/variable binding environment.
.coIP hash
Hash table.
.coIP stream
I/O stream of any kind.
.coIP regex
Regular-expression object.
.coIP struct-type
A structure type: the type of any one of the values which represents
a structure type.
.coIP tnode
Binary search tree node.
.coIP tree
Binary search tree.
.coIP args
Function argument list represented as an object.
.PP
There are more kinds of objects, such as user-defined structures.
.coNP Function @ subtypep
.synb
.mets (subtypep < left-type << right-type )
.syne
.desc
The
.code subtypep
function tests whether
.meta left-type
and
.meta right-type
name a pair of types, such that the left type is a subtype of the right
type.
The arguments are either type symbols, or structure type objects, as returned by the
.code find-struct-type
function. Thus, the symbol
.codn time ,
which is the name of a predefined struct type, and the object returned by
.code "(find-struct-type 'time)"
are considered equivalent argument values.
If either argument doesn't name a type, the behavior is
unspecified.
Each type is a subtype of itself. Most other type relationships can be inferred
from the type hierarchy diagrams given in the introduction to this section.
In addition, there are inheritance relationships among structures. If
.meta left-type
and
.meta right-type
are both structure types, then
.code subtypep
yields true if the types are the same struct type, or if the right
type is a direct or indirect supertype of the left.
The type symbol
.code struct
is a supertype of all structure types.
.coNP Function @ typep
.synb
.mets (typep < object << type-symbol )
.syne
.desc
The
.code typep
function tests whether the type of
.meta object
is a subtype of the type named by
.metn type-symbol .
The following equivalence holds:
.verb
(typep a b) --> (subtypep (typeof a) b)
.brev
.coNP Macro @ typecase
.synb
.mets (typecase < test-form >> {( type-sym << clause-form *)}*)
.syne
.desc
The
.code typecase
macro evaluates
.meta test-form
and then successively tests its type against each clause.
Each clause consists of a type symbol
.meta type-sym
and zero or more
.metn clause-form s.
The first clause whose
.meta type-sym
is a supertype of the type of
.metn test-form 's
value is considered to be the matching clause.
That clause's
.metn clause-form s
are evaluated, and the value of the last form is returned.
If there is no matching clause, or there are no clauses present,
or the matching clause has no
.metn clause-form s,
then
.code nil
is returned.
Note: since
.code t
is the supertype of every type, a clause whose
.meta type-sym
is the symbol
.code t
always matches. If such a clause is placed as the last clause of a
.codn typecase ,
it provides a fallback case, whose forms are evaluated if none of the
previous clauses match.
.coNP Macro @ etypecase
.synb
.mets (etypecase < test-form >> {( type-sym << clause-form *)}*)
.syne
.desc
The
.code etypecase
macro is the error-catching variant of
.codn typecase ,
similar to the relationship between the
.code ecaseq
and
.code caseq
families of macros.
If one of the clauses has a
.meta type-sym
which is the symbol
.codn t ,
then
.code etypecase
is precisely equivalent to
.codn typecase .
Otherwise,
a clause with a
.meta type-sym
of
.code t
and which throws an exception of type
.codn case-error ,
derived from
.codn error ,
is appended to the existing clauses,
after which the semantics follows that of
.codn typecase .
.coNP Function @ built-in-type-p
.synb
.mets (built-in-type-p << object )
.syne
.desc
The
.code built-in-type-p
function returns
.code t
if
.meta object
is a symbol which is the name of a built-in type.
For all other objects it returns
.codn nil .
.SS* Object Equivalence
.coNP Functions @, identity @ identity* and @ use
.synb
.mets (identity << value )
.mets (identity* << value *)
.mets (use << value )
.syne
.desc
The
.code identity
function returns its argument.
If the
.code identity*
function is given at least one argument, then it returns its
leftmost argument, otherwise it returns
.codn nil .
The
.code use
function is a synonym of
.codn identity .
.TP* Notes:
The
.code identity
function is useful as a functional argument, when a transformation
function is required, but no transformation is actually desired.
In this role, the
.code use
synonym leads to readable code. For instance:
.verb
;; construct a function which returns its integer argument
;; if it is odd, otherwise it returns its successor.
;; "If it's odd, use it, otherwise take its successor".
[iff oddp use succ]
;; Applications of the function:
[[iff oddp use succ] 3] -> 3 ;; use applied to 3
[[iff oddp use succ] 2] -> 3 ;; succ applied to 2
.brev
.coNP Functions @, null @ not and @ false
.synb
.mets (null << value )
.mets (not << value )
.mets (false << value )
.syne
.desc
The
.codn null ,
.code not
and
.code false
functions are synonyms. They tests whether
.meta value
is
the object
.codn nil .
They return
.code t
if this is the case,
.code nil
otherwise.
.TP* Examples:
.verb
(null '()) -> t
(null nil) -> t
(null ()) -> t
(false t) -> nil
(if (null x) (format t "x is nil!"))
(let ((list '(b c d)))
(if (not (memq 'a list))
(format t "list ~s does not contain the symbol a\en")))
.brev
.coNP Functions @ true and @ have
.synb
.mets (true << value )
.mets (have << value )
.syne
.desc
The
.code true
function is the complement of the
.codn null ,
.code not
and
.code false
functions. The
.code have
function is a synonym for
.codn true .
It return
.code t
if the
.meta value
is any object other than
.codn nil .
If
.meta value
is
.codn nil ,
it returns
.codn nil .
Note: programs should avoid explicitly testing values with true.
For instance
.code "(if x ...)"
should be favored over
.codn "(if (true x) ...)" .
However, the latter is useful with the
.code ifa
macro because
.mono
.meti (ifa (true << expr ) ...)
.onom
binds the
.code it
variable to the value of
.metn expr ,
no matter what kind of form
.meta expr
is, which is not true in the
.mono
.meti (ifa < expr ...)
.onom
form.
.TP* Example:
.verb
;; Compute indices where the list '(1 nil 2 nil 3)
;; has true values:
[where '(1 nil 2 nil 3) true] -> (1 3)
.brev
.coNP Functions @, eq @ eql and @ equal
.synb
.mets (eq < left-obj << right-obj )
.mets (eql < left-obj << right-obj )
.mets (equal < left-obj << right-obj )
.syne
.desc
The principal equality test functions
.codn eq ,
.code eql
and
.code equal
test whether two objects are equivalent, using different criteria. They return
.code t
if the objects are equivalent, and
.code nil
otherwise.
The
.code eq
function uses the strictest equivalence test, called implementation
equality. The eq function returns
.code t
if and only if,
.meta left-obj
and
.meta right-obj
are actually the same object. The
.code eq
test is implemented
by comparing the raw bit pattern of the value, whether or not it is
an immediate value or a pointer to a heaped object.
Two character values are
.code eq
if they are the same character, and two fixnum integers
are
.code eq
if they have the same value. All other object representations are actually
pointers, and are
.code eq
if and only if they point to the same object in memory.
So, for instance, two bignum integers might not be
.code eq
even if they have the same numeric
value, two lists might not be
.code eq
even if all their corresponding elements are
.code eq
and two strings might not be eq even if they hold identical text.
The
.code eql
function is slightly less strict than
.codn eq .
The difference between
.code eql
and
.code eq
is that if
.meta left-obj
and
.meta right-obj
are numbers which are of the same kind and have the same numeric value,
.code eql
returns
.metn t ,
even if they are different objects.
Note that an integers and a floating-point number are not
.code eql
even if one has a value which converts to the other: thus,
.code "(eql 0.0 0)"
yields
.codn nil ;
a comparison expression which finds these numbers equal is
.codn "(= 0.0 0)" .
The
.code eql
function also specially treats range objects. Two distinct range objects are
.code eql
if their corresponding
.meta from
and
.meta to
fields are
.codn eql .
For all other object types,
.code eql
behaves like
.codn eq .
The
.code equal
function is less strict still than
.codn eql .
In general, it recurses into some kinds of aggregate objects to perform a
structural equivalence check. For struct types, it also supports customization
via equality substitution. See the Equality Substitution section under
Structures.
Firstly, if
.meta left-obj
and
.meta right-obj
are
.code eql
then they are also
.codn equal ,
though the converse isn't necessarily the case.
If two objects are both cons cells, then they are equal if their
.code car
fields are
.code equal
and their
.code cdr
fields are
.codn equal .
If two objects are vectors, they are
.code equal
if they have the same length, and
their corresponding elements are
.codn equal .
If two objects are strings, they are
.code equal
if they are textually identical.
If two objects are functions, they are
.code equal
if they have
.code equal
environments,
and if they have
the same code. Two compiled functions are considered to have
the same code if and only if they are pointers to the same function.
Two interpreted functions are considered to have the same
code if their list
structure is
.codn equal .
Two hashes are
.code equal
if they use the same equality (both are
.codn :equal-based ,
or both are
.code :eql-based
or else both are
.codn :eq-based ),
if their associated user data elements are equal (see the function
.codn hash-userdata ),
if their sets of keys are identical, and if the data items associated with
corresponding keys from each respective hash are
.code equal
objects.
Two ranges are
.code equal
if their corresponding
.meta to
and
.meta from
fields are equal.
For some aggregate objects, there is no special semantics. Two arguments
which are symbols, packages, or streams are
.code equal
if and only if they are the same object.
Certain object types have a custom
.code equal
function.
.coNP Functions @, neq @ neql and @ nequal
.synb
.mets (neq < left-obj << right-obj )
.mets (neql < left-obj << right-obj )
.mets (nequal < left-obj << right-obj )
.syne
.desc
The functions
.codn neq ,
.code neql
and
.code nequal
are logically negated counterparts of, respectively,
.codn eq ,
.code eql
and
.codn equal .
If
.code eq
returns
.code t
for a given pair of arguments
.meta left-obj
and
.metn right-obj ,
then
.code neq
returns
.codn nil .
Vice versa, if
.code eq
returns
.codn nil ,
.code neq
returns
.codn t .
The same relationship exits between
.code eql
and
.codn neql ,
and between
.code equal
and
.codn nequal .
.coNP Functions @, meq @ meql and @ mequal
.synb
.mets (meq < left-obj << right-obj *)
.mets (meql < left-obj << right-obj *)
.mets (mequal < left-obj << right-obj *)
.syne
.desc
The functions
.codn meq ,
.code meql
and
.code mequal
("member equal" or "multi-equal")
provide a particular kind of a generalization of the binary
equality functions
.codn eq ,
.code eql
and
.code equal
to multiple arguments.
The
.meta left-obj
value is compared to each
.meta right-obj
value using the corresponding binary equality function.
If a match occurs, then
.code t
is returned, otherwise
.codn nil .
The traversal of the
.meta right-obj
argument values proceeds from left to right, and stops
when a match is found.
.coNP Function @ less
.synb
.mets (less < left-obj << right-obj )
.mets (less < obj << obj *)
.syne
.desc
The
.code less
function, when called with two arguments, determines whether
.meta left-obj
compares less than
.meta right-obj
in a generic way which handles arguments of various types.
The argument syntax of
.code less
is generalized. It can accept one argument, in which case it unconditionally
returns
.code t
regardless of that argument's value. If more than two arguments are
given, then
.code less
generalizes in a way which can be described by the following equivalence
pattern, with the understanding that each argument expression
is evaluated exactly once:
.verb
(less a b c) <--> (and (less a b) (less b c))
(less a b c d) <--> (and (less a b) (less b c) (less c d))
.brev
The
.code less
function is used as the default for the
.meta lessfun
argument of the functions
.code sort
and
.codn merge ,
as well as the
.meta testfun
argument of the
.code pos-min
and
.codn find-min .
The
.code less
function is capable of comparing numbers, characters, symbols, strings,
as well as lists and vectors of these. It can also compare buffers.
If both arguments are the same object so that
.mono
.meti (eq < left-obj << right-obj )
.onom
holds true, then the function returns
.code nil
regardless of the type of
.metn left-obj ,
even if the function doesn't handle comparing different instances
of that type. In other words, no object is less than itself, no matter
what it is.
The
.code less
function pairs with the
.code equal
function. If values
.code a
and
.code b
are objects which are of suitable types to the
.code less
function, then exactly one of the following three expressions must be true:
.codn "(equal a b)" ,
.code "(less a b)"
or
.codn "(less b a)" .
The
.code less
relation is: antisymmetric, such that if
.code "(less a b)"
is true, then
then
.code "(less b a)"
is false; irreflexive, such that
.code "(less a a)"
is false; and transitive, such that
.code "(less a b)"
and
.code "(less b c)"
imply
.codn "(less a c)" .
The following are detailed criteria that
.code less
applies to arguments of different types and combinations thereof.
If both arguments are numbers or characters, they are compared as if using the
.code <
function.
If both arguments are strings, they are compared as if using the
.code string-lt
function.
If both arguments are symbols, the following rules apply.
If the symbols have names which are different, then the result is
that of their names being compared by the
.code string-lt
function. If
.code less
is passed symbols which have the same name, and neither of these
symbols has a home package, then the raw bit patterns of their
values are compared as integers: effectively, the object with the
lower machine address is considered lesser than the other.
If only one of the two same-named symbols has no home package, then if
that symbol is the left argument,
.code less
returns
.codn t ,
otherwise
.codn nil .
If both same-named symbols have home packages, then the result of
.code less
is that of
.code string-lt
applied to the names of their respective packages. Thus
.code a:foo
is less than
.codn z:foo .
If both arguments are conses, then they are compared as follows:
.RS
.IP 1.
The
.code less
function is recursively applied to the
.code car
fields of both arguments. If it yields true, then
.meta left-obj
is deemed to be less than
.metn right-obj .
.IP 2.
Otherwise, if the
.code car
fields are unequal under
the
.code equal
function,
.code less
returns
.codn nil .
.IP 3.
If the
.code car
fields are
.code equal
then
.code less
is recursively applied to the
.code cdr
fields of the arguments, and the result of that comparison is returned.
.RE
.IP
This logic performs a lexicographic comparison on ordinary lists such
that for instance
.code "(1 1)"
is less than
.code "(1 1 1)"
but not less than
.code "(1 0)"
or
.codn (1) .
Note that the empty
.code nil
list nil compared to a cons is handled by type-based precedence, described
below.
Two vectors are compared by
.code less
lexicographically, similarly
to strings. Corresponding elements, starting with element 0, of the
vectors are compared until an index position is found where corresponding
elements of the two vectors are not
.metn equal .
If this differing position is beyond the end of one of the two vectors,
then the shorter vector is considered to be lesser. Otherwise, the result
of
.code less
is the outcome of comparing those differing elements themselves
with
.codn less .
Two buffers are also compared by
.code less
lexicographically, as if they were vectors of integer byte values.
Two ranges are compared by
.code less
using lexicographic logic similar to conses and vectors.
The
.code from
fields of the ranges are first compared. If they are not
.codn equal ,
equal then
.code less
is applied to those fields and the result is returned.
If the
.code from
fields are
.codn equal ,
then
.code less
is applied to the
.code to
fields and that result is returned.
If the two arguments are of the above types, but of different types from
each other, then
.code less
resolves the situation based on the following precedence: numbers and
characters are less than ranges, which are less than strings, which are less
than symbols, which are less than conses, which are less than vectors,
which are less than buffers.
Note that since
.code nil
is a symbol, it is ranked lower than a cons. This interpretation ensures
correct behavior when
.code nil
is regarded as an empty list, since the empty list is lexicographically prior to
a nonempty list.
If either argument is a structure for which the
.code equal
method is defined, the method is invoked on that argument, and the
value returned is used in place of that argument for performing
the comparison. Structures with no
.code equal
method cannot participate in a comparison, resulting in an error.
See the Equality Substitution section under Structures.
Finally, if either of the arguments has a type other than the above
types, the situation is an error.
.coNP Function @ greater
.synb
.mets (greater < left-obj << right-obj )
.mets (greater < obj << obj *)
.syne
.desc
The
.code greater
function is equivalent to
.code less
with the arguments reversed. That is to say, the following
equivalences hold:
.verb
(greater a) <--> (less a) <--> t
(greater a b) <--> (less b a)
(greater a b c ...) <--> (less ... c b a)
.brev
The
.code greater
function is used as the default for the
.meta testfun
argument of the
.code pos-max
and
.code find-max
functions.
.coNP Functions @ lequal and @ gequal
.synb
.mets (lequal < obj << obj *)
.mets (gequal < obj << obj *)
.syne
.desc
The functions
.code lequal
and
.code gequal
are similar to
.code less
and
.code greater
respectively, but differ in the following respect:
when called with two arguments which compare true under the
.code equal
function, the
.code lequal
and
.code gequal
functions return
.codn t .
When called with only one argument, both functions return
.code t
and both functions generalize to three or more arguments
in the same way as do
.code less
and
.codn greater .
.coNP Function @ copy
.synb
.mets (copy << object )
.syne
.desc
The
.code copy
function duplicates objects of various supported types: sequences, hashes,
structures and random states. If
.meta object
is
.codn nil ,
it
returns
.codn nil .
Otherwise,
.code copy
is equivalent to invoking a more specific copying function according to
the type of the argument, as follows:
.RS
.coIP cons
.mono
.meti (copy-list << object )
.onom
.coIP str
.mono
.meti (copy-str << object )
.onom
.coIP vec
.mono
.meti (copy-vec << object )
.onom
.coIP hash
.mono
.meti (copy-hash << object )
.onom
.IP "struct type"
.mono
.meti (copy-struct << object )
.onom
.coIP fun
.mono
.meti (copy-fun << object )
.onom
.coIP buf
.mono
.meti (copy-buf << object )
.onom
.coIP carray
.mono
.meti (copy-carray << object )
.onom
.coIP random-state
.mono
.meti (make-random-state << object )
.onom
.coIP tnode
.mono
.meti (copy-tnode << object )
.onom
.coIP tree
.mono
.meti (copy-search-tree << object )
.onom
.coIP tree-iter
.mono
.meti (copy-tree-iter << object )
.onom
.coIP cptr
.mono
.meti (copy-cptr << object )
.onom
.RE
.IP
For all other types of
.metn object ,
the invocation is erroneous.
Except in the case when
.meta sequence
is
.codn nil ,
.code copy
returns a value that
is distinct from (not
.code eq
to)
.metn sequence .
This is different from
the behavior of
.mono
.meti >> [ sequence 0..t]
.onom
or
.mono
.meti (sub < sequence 0 t)
.onom
which recognize
that they need not make a copy of
.metn sequence ,
and just return it.
Note however, that the elements of the returned sequence may be
eq to elements of the original sequence. In other words, copy is
a deeper copy than just duplicating the
.code sequence
value itself,
but it is not a deep copy.
.SS* List Manipulation
.coNP Function @ cons
.synb
.mets (cons < car-value << cdr-value )
.syne
.desc
The
.code cons
function allocates, initializes and returns a single cons cell.
A cons cell has two fields called
.code car
and
.codn cdr ,
which are accessed by
functions of the same name, or by the functions
.code first
and
.codn rest ,
which are synonyms for these.
Lists are made up of conses. A (proper) list is either the symbol
.code nil
denoting an empty list, or a cons cell which holds the first item of
the list in its
.codn car ,
and the list of the remaining items in
.codn cdr .
The expression
.code "(cons 1 nil)"
allocates and returns a single cons cell which denotes the one-element
list
.codn (1) .
The
.code cdr
is
.codn nil ,
so there are no additional items.
A cons cell whose
.code cdr
is an atom other than
.code nil
is printed with the dotted
pair notation. For example the cell produced by
.code "(cons 1 2)"
is denoted
.codn "(1 . 2)" .
The notation
.code "(1 . nil)"
is perfectly valid as input, but the cell which it denotes
will print back as
.codn (1) .
The notations are equivalent.
The dotted pair notation can be used regardless of what type
of object is the cons cell's
.codn cdr .
so that for instance
.code "(a . (b c))"
denotes the cons cell whose
.code car
is the symbol a
.code a
and whose
.code cdr
is the list
.codn "(b c)" .
This is exactly the same thing as
.codn "(a b c)" .
In other words
.code "(a b ... l m . (n o ... w . (x y z)))"
is exactly the same as
.codn "(a b ... l m n o ... w x y z)" .
Every list, and more generally cons-cell tree structure, can be written
in a "fully dotted" notation, such that there are as many dots as there
are cells. For instance the cons structure of the nested list
.code "(1 (2) (3 4 (5)))"
can be made more explicit using
.codn "(1 . ((2 . nil) . ((3 . (4 . ((5 . nil) . nil))) . nil))))" .
The structure contains eight conses, and so there are eight dots
in the fully dotted notation.
The number of conses in a linear list like
.code "(1 2 3)"
is simply the number of items, so that list in particular is
made of three conses. Additional nestings require additional conses,
so for instance
.code "(1 2 (3))"
requires four conses. A visual way to count the conses from the printed
representation is to count the atoms, then add the count of open parentheses,
and finally subtract one.
A list terminated by an atom other than
.code nil
is called an improper
list, and the dot notation is extended to cover improper lists.
For instance
.code "(1 2 . 3)"
is an improper list of two elements,
terminated by
.codn 3 ,
and can be constructed using
.codn "(cons 1 (cons 2 3))" .
The fully dotted notation for this list is
.codn "(1 . (2 . 3))" .
.coNP Function @ atom
.synb
.mets (atom << value )
.syne
.desc
The
.code atom
function tests whether
.meta value
is an atom. It returns
.code t
if this is the
case,
.code nil
otherwise. All values which are not cons cells are atoms.
.code "(atom x)"
is equivalent to
.codn "(not (consp x))" .
.TP* Examples:
.verb
(atom 3) -> t
(atom (cons 1 2)) -> nil
(atom "abc") -> t
(atom '(3)) -> nil
.brev
.coNP Function @ consp
.synb
.mets (consp << value )
.syne
.desc
The
.code consp
function tests whether
.meta value
is a cons. It returns
.code t
if this is the
case,
.code nil
otherwise.
.code "(consp x)"
is equivalent to
.codn "(not (atom x))" .
Nonempty lists test positive under
.code consp
because a list is represented as a reference to the first cons in a chain of
one or more conses.
Note that a lazy cons is a cons and satisfies the
.code consp
test. See the function
.code make-lazy-cons
and the macro
.codn lcons .
.TP* Examples:
.verb
(consp 3) -> nil
(consp (cons 1 2)) -> t
(consp "abc") -> nil
(consp '(3)) -> t
.brev
.coNP Accessors @ car and @ first
.synb
.mets (car << object )
.mets (first << object )
.mets (set (car << object ) << new-value )
.mets (set (first << object ) << new-value )
.syne
.desc
The functions
.code car
and
.code first
are synonyms.
If
.meta object
is a cons cell, these functions retrieve the
.code car
field of that cons cell.
.code "(car (cons 1 2))"
yields
.codn 1 .
For programming convenience,
.meta object
may be of several other kinds in addition to conses.
.code "(car nil)"
is allowed, and returns
.codn nil .
.meta object
may also be a vector or a string. If it is an empty vector or
string, then
.code nil
is returned. Otherwise the first character of the string or
first element of the vector is returned.
.meta object
may be a structure. The
.code car
operation is possible if the object has a
.code car
method. If so,
.code car
invokes that method and returns whatever the method returns.
If the structure has no
.code car
method, but has a
.code lambda
method, then the
.code car
function calls that method with one argument, that being
the integer zero. Whatever the method returns,
.code car
returns. If neither method is defined, an error
exception is thrown.
A
.code car
form denotes a valid place whenever
.meta object
is a valid argument for the
.code rplaca
function. Modifying the place denoted by the form is equivalent to invoking
.code rplaca
with
.meta object
as the left argument, and the replacement value as the right
argument. It takes place in the manner given under the description
.code rplaca
function, and obeys the same restrictions.
A
.code car
form supports deletion. The following equivalence
then applies:
.verb
(del (car place)) <--> (pop place)
.brev
This implies that deletion requires the argument of the
.code car
form to be a place, rather than the whole form itself.
In this situation, the argument place may have a value
which is
.codn nil ,
because
.code pop
is defined on an empty list.
The abstract concept behind deleting a
.code car
is that physically deleting this field from a cons,
thereby breaking it in half, would result in just the
.code cdr
remaining. Though fragmenting a cons in this manner is impossible,
deletion simulates it by replacing the place which previously held the
cons, with that cons'
.code cdr
field. This semantics happens to coincide with deleting the first element
of a list by a
.code pop
operation.
.coNP Accessors @ cdr and @ rest
.synb
.mets (cdr << object )
.mets (rest << object )
.mets (set (cdr << object ) << new-value )
.mets (set (rest << object ) << new-value )
.syne
.desc
The functions
.code cdr
and
.code rest
are synonyms.
If
.meta object
is a cons cell, these functions retrieve the
.code cdr
field of that cons cell.
.code "(cdr (cons 1 2))"
yields
.codn 2 .
For programming convenience,
.meta object
may be of several other kinds in addition to conses.
.code "(cdr nil)"
is allowed, and returns
.codn nil .
.meta object
may also be a vector or a string. If it is a nonempty string or vector
containing at least two items, then the remaining part of the object is
returned, with the first element removed. For example
.mono
(cdr "abc")
.onom
yields
.strn "bc" .
If
.meta object
is a one-element vector or string, or an empty vector or string,
then
.code nil
is returned. Thus
.mono
(cdr "a")
.onom
and
.mono
(cdr "")
.onom
both result in
.codn nil .
If
.meta object
is a structure, then
.code cdr
requires it to support either the
.code cdr
method or the
.code lambda
method. If both are present,
.code cdr
is used. When the
.code cdr
function uses the
.code cdr
method, it invokes it with no arguments.
Whatever value the method returns becomes the
return value of
.codn cdr .
When
.code cdr
invokes a structure's
.code lambda
method, it passes as the argument the range object
.codn "#R(1 t)" .
Whatever the
.code lambda
method returns becomes the return value of
.codn cdr .
The invocation syntax of a
.code cdr
or
.code rest
form is a syntactic place.
The place is semantically correct if
.meta object
is a valid argument for the
.code rplacd
function. Modifying the place denoted by the form is equivalent to invoking
.code rplacd
with
.meta object
as the left argument, and the replacement value as the right
argument. It takes place in the manner given under the description
.code rplacd
function, and obeys the same restrictions.
A
.code cdr
place supports deletion, according to the following near equivalence:
.verb
(del (cdr place)) <--> (prog1 (cdr place)
(set place (car place)))
.brev
The
.code place
expression is evaluated only once.
Note that this is symmetric with the delete semantics of
.code car
in that the cons stored in
.code place
goes away, as does the
.code cdr
field, leaving just the
.codn car ,
which takes the place of the original cons.
.TP*
Example:
Walk every element of the list
.code "(1 2 3)"
using a
.code for
loop:
.verb
(for ((i '(1 2 3))) (i) ((set i (cdr i)))
(print (car i) *stdout*)
(print #\enewline *stdout*))
.brev
The variable
.code i
marches over the cons cells which make up the "backbone"
of the list. The elements are retrieved using the
.code car
function.
Advancing to the next cell is achieved using
.codn "(cdr i)" .
If
.code i
is the
last cell in a (proper) list,
.code "(cdr i)"
yields
.code nil
and so
.code i
becomes
.codn nil ,
the loop guard expression
.code i
fails and the loop terminates.
.coNP Functions @ rplaca and @ rplacd
.synb
.mets (rplaca < object << new-car-value )
.mets (rplacd < object << new-cdr-value )
.syne
.desc
If
.code object
is a cons cell or lazy cons cell, then
.code rplaca
and
.code rplacd
functions assign new values into the
.code car
and
.code cdr
fields of the
.metn object .
In addition, these functions are meaningful for other kinds of objects also.
Note that, except for the difference in return value,
.code "(rplaca x y)"
is the same as the more generic
.codn "(set (car x) y)" ,
and likewise
.code "(rplacd x y)"
can be written as
.codn "(set (cdr x) y)" .
The
.code rplaca
and
.code rplacd
functions return
.metn cons .
Note: In \*(TX versions 89 and earlier, these functions returned the new value.
The behavior was undocumented.
The
.meta cons
argument does not have to be a cons cell. Both functions support meaningful
semantics for vectors and strings. If
.meta cons
is a string, it must be modifiable.
The
.code rplaca
function replaces the first element of a vector or first character
of a string. The vector or string must be at least one element long.
The
.code rplacd
function replaces the suffix of a vector or string after the first element
with a new suffix. The
.meta new-cdr-value
must be a sequence, and if the suffix of a string is being replaced,
it must be a sequence of characters. The suffix here refers to the portion of
the vector or string after the first element.
It is permissible to use
.code rplacd
on an empty string or vector. In this case,
.meta new-cdr-value
specifies the contents of the entire string or vector, as if the operation
were done on a nonempty vector or string, followed by the deletion of the
first element.
The
.meta object
argument may be a structure. In the case of
.codn rplaca ,
the structure must have a defined
.code rplaca
method or else, failing that, a
.code lambda-set
method. The first of these methods which is available, in the given order, is
used to perform the operation. Whatever the respective method returns,
If the
.code lambda-set
method is used, it is called with two arguments (in addition to
.codn object ):
the integer zero, and
.metn new-car-value .
In the case of
.codn rplacd ,
the structure must have a defined
.code rplacd
method or else, failing that, a
.code lambda-set
method. The first of these methods which is available, in the given order, is
used to perform the operation. Whatever the respective method returns,
If the
.code lambda-set
method is used, it is called with two arguments (in addition to
.codn object ):
the range value
.code "#R(1 t)"
and
.metn new-car-value .
.coNP Accessors @, second @, third @, fourth @, fifth @, sixth @, seventh @, eighth @ ninth and @ tenth
.synb
.mets (first << object )
.mets (second << object )
.mets (third << object )
.mets (fourth << object )
.mets (fifth << object )
.mets (sixth << object )
.mets (seventh << object )
.mets (eighth << object )
.mets (ninth << object )
.mets (tenth << object )
.mets (set (first << object ) << new-value )
.mets (set (second << object ) << new-value )
.mets ...
.mets (set (tenth << object ) << new-value )
.syne
.desc
Used as functions, these accessors retrieve the elements of a sequence by
position. If the sequence is shorter than implied by the position, these
functions return
.codn nil .
When used as syntactic places, these accessors denote the storage locations
by position. The location must exist, otherwise an error exception results.
The places support deletion.
.TP* Examples:
.verb
(third '(1 2)) -> nil
(second "ab") -> #\eb
(third '(1 2 . 3)) -> **error, improper list*
(let ((x (copy "abcd")))
(inc (third x))
x) -> "abce"
.brev
.coNP Functions @ append and @ nconc
.synb
.mets (append <> [ sequence *])
.mets (nconc <> [ sequence *])
.syne
.desc
The
.code append
function creates a new object which is a catenation of the
.meta list
arguments. All arguments are optional;
.code append
produces the empty list, and if
a single argument is specified, that argument
is returned.
If two or more arguments are present, then the situation
is identified as one or more
.meta sequence
arguments followed by
.metn last-arg .
The
.meta sequence
arguments must be sequences;
.meta last-arg
may be a sequence or atom.
The
.code append
operation over three or more arguments is left-associative, such that
.code "(append x y z)"
is equivalent to both
.code "(append (append x y) z)"
and
.codn "(append x (append z y))" .
This allows the catenation of an arbitrary number of arguments
to be understood in terms of a repeated application of the two-argument
case, whose semantics is given by these rules:
.RS
.IP 1.
.code nil
catenates with
.code nil
to produce
.codn nil :
.verb
(append nil nil) -> nil
.brev
.IP 2.
.code nil
catenates with a proper or improper list, producing that list itself:
.verb
(append nil '(1 2)) -> (1 2)
(append nil '(1 2 . 3)) -> (1 2 . 3)
.brev
.IP 3.
A proper list catenates with
.codn nil ,
producing that list itself:
.verb
(append '(1 2) nil) -> (1 2)
.brev
.IP 4.
A proper list catenates with an atom,
producing an improper list terminated by that atom,
whether or not that atom is a sequence:
.verb
(append '(1 2) #(3)) -> (1 2 . #(3))
(append '(1 2) 3) -> (1 2 . 3)
.brev
.IP 5.
A non-list sequence catenates with another sequence into a sequence,
producing a sequence which contains the elements of both,
of the same kind as the left sequence. The elements must be
compatible; a string can only catenate with a sequence of characters.
.verb
(append #(1 2) #(3 4)) -> #(1 2 3 4)
(append "ab" "cd") -> "abcd"
(append "ab" #(#\ec #\ed)) -> "abcd"
(append "ab" #(3 4)) -> ;; error
.brev
.IP 6.
A non-list sequence catenates with an atom if it is a suitable element
type for that kind of sequence. The resulting sequence is of the same
kind, and includes that atom:
.verb
(append #(1 2) 3) -> #(1 2 3)
(append "ab" #\c) -> "abc"
(append "ab" 3) -> ;; error
.brev
.IP 7.
If an improper list is catenated with any object, the catenation
takes place between the terminating atom of that list and that object. This
requires the terminating atom to be a sequence. If the catenation is possible,
then the result is a new improper list which is a copy of the original, but
with the terminating atom replaced by a catenation of that atom and the object:
.verb
(append '(1 2 . "ab") "c") -> (1 2 . "abc")
(append '(1 2 . "ab") '(2 3)) -> ;; error
.brev
.IP 8.
A non-sequence atom doesn't catenate; the situation is erroneous:
.verb
(append 1 2) -> ;; error
(append '(1 . 2) 3) -> ;; error
.brev
.RE
.IP
If N arguments are specified, where N > 1, then the first N-1 arguments must be
proper lists. Copies of these lists are catenated together. The last argument
N, shown in the above syntax as
.metn last-arg ,
may be any kind of object. It is
installed into the
.code cdr
field of the last cons cell of the resulting list.
Thus, if argument N is also a list, it is catenated onto the resulting list,
but without being copied. Argument N may be an atom other than
.codn nil ;
in that case
.code append
produces an improper list.
The
.code nconc
function works like
.codn append ,
but may destructively manipulate any of the input objects.
.TP* Examples:
.verb
;; An atom is returned.
(append 3) -> 3
;; A list is also just returned: no copying takes place.
;; The eq function can verify that the same object emerges
;; from append that went in.
(let ((list '(1 2 3)))
(eq (append list) list)) -> t
(append '(1 2 3) '(4 5 6) 7) -> '(1 2 3 4 5 6 . 7))
;; the (4 5 6) tail of the resulting list is the original
;; (4 5 6) object, shared with that list.
(append '(1 2 3) '(4 5 6)) -> '(1 2 3 4 5 6)
(append nil) -> nil
;; (1 2 3) is copied: it is not the last argument
(append '(1 2 3) nil) -> (1 2 3)
;; empty lists disappear
(append nil '(1 2 3) nil '(4 5 6)) -> (1 2 3 4 5 6)
(append nil nil nil) -> nil
;; atoms and improper lists other than in the last position
;; are erroneous
(append '(a . b) 3 '(1 2 3)) -> **error**
;; sequences other than lists can be catenated.
(append "abc" "def" "g" #\eh) -> "abcdefgh"
;; lists followed by non-list sequences end with non-list
;; sequences catenated in the terminating atom:
(append '(1 2) '(3 4) "abc" "def") -> (1 2 3 4 . "abcdef")
.brev
.coNP Function @ append*
.synb
.mets (append* <> [ list *])
.syne
.desc
The
.code append*
function lazily catenates lists.
If invoked with no arguments, it returns
.codn nil .
If invoked with a single argument, it returns that argument.
Otherwise, it returns a lazy list consisting of the elements of every
.meta list
argument from left to right.
Arguments other than the last are treated as lists, and traversed using
.code car
and
.code cdr
functions to visit their elements.
The last argument isn't traversed: rather, that object itself becomes the
.code cdr
field of the last cons cell of the lazy list constructed from the
previous arguments.
.coNP Functions @ revappend and @ nreconc
.synb
.mets (revappend < list1 << list2 )
.mets (nreconc < list1 << list2 )
.syne
.desc
The
.code revappend
function returns a list consisting of
.code list2
appended to a reversed copy of
.metn list1 .
The returned object shares structure
with
.metn list2 ,
which is unmodified.
The
.code nreconc
function behaves similarly, except
that the returned object may share
structure with not only
.meta list2
but also
.metn list1 ,
which is modified.
.coNP Function @ list
.synb
.mets (list << value *)
.syne
.desc
The
.code list
function creates a new list, whose elements are the
argument values.
.TP* Examples:
.verb
(list) -> nil
(list 1) -> (1)
(list 'a 'b) -> (a b)
.brev
.coNP Function @ list*
.synb
.mets (list* << value *)
.syne
.desc
The
.code list*
function is a generalization of cons. If called with exactly
two arguments, it behaves exactly like cons:
.code "(list* x y)"
is identical to
.codn "(cons x y)" .
If three or more arguments are specified,
the leading arguments specify additional atoms to be consed to the
front of the list. So for instance
.code "(list* 1 2 3)"
is the same as
.code "(cons 1 (cons 2 3))"
and produces the improper list
.codn "(1 2 . 3)" .
Generalizing in the other direction,
.code list*
can be called with just
one argument, in which case it returns that argument, and
can also be called with no arguments in which case it returns
.codn nil .
.TP* Examples:
.verb
(list*) -> nil
(list* 1) -> 1
(list* 'a 'b) -> (a . b)
(list* 'a 'b 'c) -> (a b . c)
.brev
.TP* "Dialect Note:"
Note that unlike in some other Lisp dialects, the effect
of
.code "(list* 1 2 x)"
can also be obtained using
.codn "(list 1 2 . x)" .
However,
.code "(list* 1 2 (func 3))"
cannot be rewritten as
.code "(list 1 2 . (func 3))"
because the latter is equivalent to
.codn "(list 1 2 func 3)" .
.coNP Accessor @ sub-list
.synb
.mets (sub-list < list >> [ from <> [ to ]])
.mets (set (sub-list < list >> [ from <> [ to ]]) << new-value )
.syne
.desc
The
.code sub-list
function has the same parameters and semantics as the
.code sub
function, except that it operates on its
.meta list
argument using list operations, and assumes that
.meta list
is terminated by
.codn nil .
If a
.code sub-list
form is used as a place, then the
.meta list
argument form must also be a place.
The
.code sub-list
place denotes a subrange of
.meta list
as if it were a storage location. The previous value of this location,
if needed, is fetched by a call to
.codn sub-list .
Storing
.meta new-value
to the place is performed by a call to
.codn replace-list .
The return value of
.meta replace-list
is stored into
.metn list .
In an update operation which accesses the prior value and stores a new value,
the arguments
.metn list ,
.metn from ,
.meta to
and
.meta new-value
are evaluated once.
.coNP Function @ replace-list
.synb
.mets (replace-list < list < item-sequence >> [ from <> [ to ]])
.syne
.desc
The
.code replace-list
function is like the
.code replace
function, except that it operates on its
.meta list
argument using list operations. It assumes that
.meta list
is terminated by
.codn nil ,
and that it is made of cells which can be mutated using
.codn rplaca .
.coNP Functions @ listp and @ proper-list-p
.synb
.mets (listp << value )
.mets (proper-list-p << value )
.syne
.desc
The
.code listp
and
.code proper-list-p
functions test, respectively, whether
.meta value
is a list, or a proper list, and return
.code t
or
.code nil
accordingly.
The
.code listp
test is weaker, and executes without having to traverse
the object.
The value produced by the expression
.code "(listp x)"
is the same as that of
.codn "(or (null x) (consp x))" ,
except that
.code x
is evaluated only once.
The empty list
.code nil
is a list, and a cons cell is a list.
The
.code proper-list-p
function returns
.code t
only for proper lists. A proper list is
either
.codn nil ,
or a cons whose
.code cdr
is a proper list.
.code proper-list-p
traverses the
list, and its execution will not terminate if the list is circular.
These functions return
.code nil
for list-like sequences that are not made of actual
.code cons
cells.
Dialect Note: in \*(TX 137 and older,
.code proper-list-p
is called
.codn proper-listp .
The name was changed for adherence to conventions and compatibility with other
Lisp dialects, like Common Lisp. However, the function continues to be
available under the old name. Code that must run on \*(TX 137 and older
installations should use
.codn proper-listp ,
but its use going forward is deprecated.
.coNP Function @ endp
.synb
.mets (endp << object )
.syne
.desc
The
.code endp
function returns
.code t
if
.meta object
is the object
.codn nil .
If
.meta object
is a cons cell, then
.code endp
returns
.codn t .
Otherwise,
.code endp
function throws an exception.
.coNP Function @ length-list
.synb
.mets (length-list << list )
.syne
.desc
The
.code length-list
function returns the length of
.metn list ,
which may be
a proper or improper list. The length of a list is the number of conses in that
list.
.coNP Function @ copy-list
.synb
.mets (copy-list << list )
.syne
.desc
The
.code copy-list
function which returns a list similar to
.metn list ,
but with
a newly allocated cons-cell structure.
If
.meta list
is an atom, it is simply returned.
Otherwise,
.meta list
is a cons cell, and
.code copy-list
returns the same object as the expression
.mono
.meti (cons (car << list ) (copy-list (cdr << list ))).
.onom
Note that the object
.mono
.meti (car << list )
.onom
is not deeply copied, but only
propagated by reference into the new list.
.code copy-list
produces
a new list structure out of the same items that are in
.metn list .
.TP* "Dialect Note:"
Common Lisp does not allow the argument to be an atom, except
for the empty list
.codn nil .
.coNP Function @ copy-cons
.synb
.mets (copy-cons << cons )
.syne
.desc
The
.code copy-cons
function creates and returns a new object that is a replica of
.metn cons .
The
.meta cons
argument must be either a
.code cons
cell, or else a lazy cons: an object of type
.codn lcons .
A new cell of the same type as
.meta cons
is created, and all of its fields are initialized by
copying the corresponding fields from
.metn cons .
If
.meta cons
is lazy, the newly created object is in the same
state as the original. If the original has not yet been updated
and thus has an update function, the copy also has not yet been
updated and has the same update function.
.coNP Function @ copy-tree
.synb
.mets (copy-tree << obj )
.syne
.desc
The
.code copy-tree
function returns a copy of
.meta obj
which represents an arbitrary
.codn cons -cell-based
structure.
The cell structure of
.meta obj
is traversed and a similar structure is constructed, but without regard for
substructure sharing or circularity.
More precisely, if
.meta obj
is an atom, then it is returned.
If it is an ordinary
.code cons
cell, then
.code copy-tree
is recursively applied to the
.code car
and
.code cdr
fields to produce their individual replicas. A new
.code cons
cell is then produced from the replicated
.code car
and
.codn cdr .
If
.meta obj
is a lazy
.codn cons ,
then just like in the ordinary
.code cons
case, the
.code car
and
.code cdr
fields are duplicated with a recursive call to
.codn copy-tree .
Then, a lazy
.code cons
is created from these replicated fields. If
.meta cell
has an update function, then the newly created lazy
.code cons
has the same update function; the function isn't copied.
Like
.codn copy-cons ,
the
.code copy-tree
function doesn't trigger the update of lazy conses.
The copies of lazy conses which have not been updated
are also conses which have not been updated.
.coNP Functions @ reverse and @ nreverse
.synb
.mets (reverse << list )
.mets (nreverse << list )
.syne
.desc
Description:
The functions
.code reverse
and
.code nreverse
produce an object which contains
the same items as proper list
.metn list ,
but in reverse order.
If
.meta list
is
.codn nil ,
then both functions return
.codn nil .
The
.code reverse
function is non-destructive: it creates a new list.
The
.code nreverse
function creates the structure of the reversed list out of the
cons cells of the input list, thereby destructively altering it (if it contains
more than one element). How
.code nreverse
uses the material from the original list
is unspecified. It may rearrange the cons cells into a reverse order, or it may
keep the structure intact, but transfer the
.code car
values among cons cells into
reverse order. Other approaches are possible.
.coNP Accessor @ nthlast
.synb
.mets (nthlast < index << list )
.mets (set (nthlast < index << list ) << new-value )
.syne
.desc
The
.code nthlast
function retrieves the n-th last cons cell of a list,
indexed from one.
The
.meta index
parameter must be a an integer. If
.meta index
is positive and so large that it specifies a nonexistent cons beyond the
beginning of the list,
.code nthlast
returns
.metn list .
Effectively, values of
.meta index
larger than the length of the list are clamped to the length.
If
.meta index
is negative, then
.code nthlast
yields nil. An
.meta index
value of zero retrieves the terminating atom of
.meta list
or else the value
.meta list
itself, if
.meta list
is an atom.
The following equivalences hold:
.verb
(nthlast 1 list) <--> (last list)
.brev
An
.code nthlast
place designates the storage location which holds the n-th cell,
as indicated by the value of
.metn index .
A negative
.meta index
doesn't denote a place.
A positive
.meta index
greater than the length of the list is treated as if it were
equal to the length of the list.
If
.meta list
is itself a syntactic place, then the
.meta index
value
.I n
is permitted for a list of length
.IR n .
This index value denotes the
.meta list
place itself. Storing to this value overwrites
.metn list .
If
.meta list
isn't a syntactic place, then storing to position
.I n
isn't permitted.
If
.meta list
is of length zero, or an atom (in which case its
length is considered to be zero) then the above
remarks about position
.I n
apply to an
.meta index
value of zero: if
.meta list
is a syntactic place, then the position denotes
.meta list
itself, otherwise the position doesn't exist as a place.
If
.meta list
contains one or more elements, then
.meta index
value of zero denotes the
.code cdr
field of its last cons cell. Storing a value to this
place overwrites the terminating atom.
.coNP Accessor @ butlastn
.synb
.mets (butlastn < num << list )
.mets (set (butlastn < num << list ) new-value )
.syne
.desc
The
.code butlastn
function calculates that initial portion of
.meta list
which excludes the last
.meta num
elements.
Note: the
.code butlastn
function doesn't support non-list sequences as sequences;
it treats them as the terminating atom of a zero-length improper list.
The
.code butlast
sequence function supports non-list sequences. If
.code x
is a list, then the following equivalence holds:
.verb
(butlastn n x) <--> (butlast x n)
.brev
If
.meta num
is zero, or negative, then
.code butlastn
returns
.metn list .
If
.meta num
is positive, and meets or exceeds the length of
.metn list ,
then
.code butlastn
returns
.codn nil .
If a
.code butlastn
form is used as a syntactic place, then
.meta list
must be a place. Assigning to the form causes
.meta list
to be replaced with a new list which is a catenation
of the new value and the last
.meta num
elements of the original list, according to the following equivalence:
.verb
(set (butlastn n x) v)
<-->
(progn (set x (append v (nthlast n x))) v)
.brev
except that
.codn n ,
.code x
and
.code v
are evaluated only once, in left-to-right order.
.coNP Accessor @ nth
.synb
.mets (nth < index << object )
.mets (set (nth < index << object ) << new-value )
.syne
.desc
The
.code nth
function performs random access on a list, retrieving the n-th
element indicated by the zero-based index value given by
.metn index .
The
.meta index
argument must be a nonnegative integer.
If
.meta index
indicates an element beyond the end of the list, then
the function returns
.codn nil .
The following equivalences hold:
.verb
(nth 0 list) <--> (car 0) <--> (first list)
(nth 1 list) <--> (cadr list) <--> (second list)
(nth 2 list) <--> (caddr list) <--> (third list)
(nth x y) <--> (car (nthcdr x y))
.brev
.coNP Accessor @ nthcdr
.synb
.mets (nthcdr < index << list )
.mets (set (nthcdr < index << list ) << new-value )
.syne
.desc
The
.code nthcdr
function retrieves the n-th cons cell of a list, indexed from zero.
The
.meta index
parameter must be a nonnegative integer. If
.meta index
specifies a nonexistent cons beyond the end of the list,
then
.code nthcdr
yields nil.
The following equivalences hold:
.verb
(nthcdr 0 list) <--> list
(nthcdr 1 list) <--> (cdr list)
(nthcdr 2 list) <--> (cddr list)
(car (nthcdr x y)) <--> (nth x y)
.brev
An
.code nthcdr
place designates the storage location which holds the n-th cell,
as indicated by the value of
.metn index .
Indices beyond the last cell of
.meta list
do not designate a valid place.
If
.meta list
is itself a place, then the zeroth index is permitted and the
resulting place denotes
.metn list .
Storing a value to
.mono
.meti (nthcdr < 0 << list )
.onom
overwrites
.metn list .
Otherwise if
.meta list
isn't a syntactic place, then the zeroth index does not designate a valid
place;
.meta index
must have a positive value. A
.code nthcdr
place does not support deletion.
.TP* "Dialect Note:"
In Common Lisp,
.code nthcdr
is only a function, not an accessor;
.code nthcdr
forms do not denote places.
.coNP Function @ tailp
.synb
.mets (tailp < object << list )
.syne
.desc
The
.code tailp
function tests whether
.meta object
is a tail of
.metn list .
This means that
.meta object
is either
.meta list
itself, or else one of the
.code cons
cells of
.meta list
or else the terminating atom of
.metn list .
More formally, a recursive definition follows.
If
.meta object
and
.meta list
are the same object (thus equal under the
.code eq
function) then
.code tailp
returns
.codn t .
If
.meta list
is an atom, and is not
.metn object ,
then the function returns
.codn nil .
Otherwise,
.meta list
is a
.code cons
that is not
.meta object
and
.code tailp
yields the same value as the
.mono
.meti "(tailp < object (cdr << list ))"
.onom
expression.
.coNP Accessors @, caar @, cadr @, cdar @, cddr ..., @ cdddddr
.synb
.mets (caar << object )
.mets (cadr << object )
.mets (cdar << object )
.mets (cddr << object )
.mets ...
.mets (cdddr << object )
.mets (set (caar << object ) << new-value )
.mets (set (cadr << object ) << new-value )
.mets ...
.syne
.desc
The
.I a-d accessors
provide a shorthand notation for accessing two to five
levels deep into a cons-cell-based tree structure. For instance, the
the equivalent of the nested function call expression
.mono
.meti (car (car (cdr << object )))
.onom
can be achieved using the single function call
.mono
.meti (caadr << object ).
.onom
The symbol names of the a-d accessors are a generalization of the words
"car" and "cdr". They encode the pattern of
.code car
and
.code cdr
traversal of the structure using a sequence of the letters
.code a
and
.code d
placed between
.code c
and
.codn r .
The traversal is encoded in right-to-left order, so that
.code cadr
indicates a traversal of the
.code cdr
link, followed by the
.codn car .
This order corresponds to the nested function call notation, which also
encodes the traversal right-to-left. The following diagram illustrates
the straightforward relationship:
.verb
(cdr (car (cdr x)))
^ ^ ^
| / |
| / /
| / ____/
|| /
(cdadr x)
.brev
\*(TL provides all possible a-d accessors up to five levels deep, from
.code caar
all the way through
.codn cdddddr .
Expressions involving a-d accessors are places. For example,
.code "(caddr x)"
denotes the same place as
.codn "(car (cddr x))" ,
and
.code "(cdadr x)"
denotes the same place as
.codn "(cdr (cadr x))" .
The a-d accessor places support deletion, with semantics derived from
the deletion semantics of the
.code car
and
.code cdr
places. For example,
.code "(del (caddr x))"
means the same as
.codn "(del (car (cddr x)))" .
.coNP Functions @ cyr and @ cxr
.synb
.mets (cyr < address << object )
.mets (cxr < address << object )
.syne
.desc
The
.code cyr
and
.code cxr
functions provide
.cod3 car / cdr
navigation of tree structure driven by numeric address given by the
.meta address
argument.
The
.meta address
argument can express any combination of the application of
.code car
and
.code cdr
functions, including none at all.
The difference between
.code cyr
and
.code cxr
is the bit order of the encoding. Under
.codn cyr ,
the most significant bit of the encoding given in
.meta address
indicates the initial
.cod3 car / cdr
navigation, and the least significant bit gives the final one.
Under
.codn cxr ,
it is opposite.
Both functions require
.meta address
to be a positive integer. Any other argument raises an error.
Under both functions, the
.meta address
value
.code 1
encodes the
.code identity
operation: no
.cod3 car / cdr
.coNP Functions @ flatten and @ flatten*
.synb
.mets (flatten << list )
.mets (flatten* << list )
.syne
.desc
The
.code flatten
function produces a list whose elements are all of the
.cod2 non- nil
atoms contained in the structure of
.metn list .
The
.code flatten*
function
works like
.code flatten
except that it produces a lazy list. It can be used to lazily flatten an
infinite lazy structure.
.TP* Examples:
.verb
(flatten '(1 2 () (3 4))) -> (1 2 3 4)
;; equivalent to previous, since
;; nil is the same thing as ()
(flatten '(1 2 nil (3 4))) -> (1 2 3 4)
(flatten nil) -> nil
(flatten '(((()) ()))) -> nil
.brev
.coNP Functions @ flatcar and @ flatcar*
.synb
.mets (flatcar << tree )
.mets (flatcar* << tree )
.syne
.desc
The
.code flatcar
function produces a list of all the atoms contained in the
tree structure
.metn tree ,
in the order in which they appear, when the structure is traversed
left to right.
This list includes those
.code nil
atoms which appear in
.code car
fields.
The list excludes
.code nil
atoms which appear in
.code cdr
fields.
The
.code flatcar*
function
works like
.code flatcar
except that it produces a lazy list. It can be used to lazily flatten an
infinite lazy structure.
.TP* Examples:
.verb
(flatcar '(1 2 () (3 4))) -> (1 2 nil 3 4)
(flatcar '(a (b . c) d (e) (((f)) . g) (nil . z) nil . h))
--> (a b c d e f g nil z nil h)
.brev
.coNP Function @ tree-find
.synb
.mets (tree-find < obj < tree << test-function )
.syne
.desc
The
.code tree-find
function searches
.meta tree
for an occurrence of
.metn obj .
Tree can be
any atom, or a cons. If
.meta tree
it is a cons, it is understood to be a proper
list whose elements are also trees.
The equivalence test is performed by
.meta test-function
which must take two
arguments, and has conventions similar to
.codn eq ,
.code eql
or
.codn equal .
.code tree-find
works as follows. If
.meta tree
is equivalent to
.meta obj
under
.metn test-function ,
then
.code t
is returned to announce a successful finding.
If this test fails, and
.meta tree
is an atom,
.code nil
is returned immediately to
indicate that the find failed. Otherwise,
.meta tree
is taken to be a proper list,
and
.code tree-find
is recursively applied to each element of the list in turn, using
the same
.meta obj
and
.meta test-function
arguments, stopping at the first element
which returns a
.cod2 non- nil
value.
.coNP Functions @, memq @ memql and @ memqual
.synb
.mets (memq < object << list )
.mets (memql < object << list )
.mets (memqual < object << list )
.syne
.desc
The
.codn memq ,
.code memql
and
.code memqual
functions search
.meta list
for a member
which is, respectively,
.codn eq ,
.code eql
or
.code equal
to
.metn object .
(See the
.codn eq ,
.code eql
and
.code equal
functions above.)
If no such element found,
.code nil
is returned.
Otherwise, that suffix of
.meta list
is returned whose first element
is the matching object.
.coNP Functions @ member and @ member-if
.synb
.mets (member < key < sequence >> [ testfun <> [ keyfun ]])
.mets (member-if < predfun < sequence <> [ keyfun ])
.syne
.desc
The
.code member
and
.code member-if
functions search through
.meta sequence
for an item which
matches a key, or satisfies a predicate function, respectively.
The
.meta keyfun
argument specifies a function which is applied to the elements
of the sequence to produce the comparison key. If this argument is omitted,
then the untransformed elements of the sequence themselves are examined.
The
.code member
function's
.meta testfun
argument specifies the test function which is
used to compare the comparison keys taken from the sequence to the search key.
If this argument is omitted, then the
.code equal
function is used.
If
.code member
does not find a matching element, it returns
.codn nil .
Otherwise it
returns the suffix of
.meta sequence
which begins with the matching element.
The
.code member-if
function's
.meta predfun
argument specifies a predicate function
which is applied to the successive comparison keys pulled from the sequence
by applying the key function to successive elements. If no match is found,
then
.code nil
is returned, otherwise what is returned is the suffix of
.meta sequence
which begins with the matching element.
.coNP Functions @, rmemq @, rmemql @, rmemqual @ rmember and @ rmember-if
.synb
.mets (rmemq < object << list )
.mets (rmemql < object << list )
.mets (rmemqual < object << list )
.mets (rmember < key < sequence >> [ testfun <> [ keyfun ]])
.mets (rmember-if < predfun < sequence <> [ keyfun ])
.syne
.desc
These functions are counterparts to
.codn memq ,
.codn memql ,
.codn memqual ,
.code member
and
.code member-if
which look for the rightmost
element which matches
.metn object ,
rather than for the leftmost element.
.coNP Functions @ conses and @ conses*
.synb
.mets (conses << list )
.mets (conses* << list )
.syne
.desc
These functions return a list whose elements are the conses which make
up
.metn list .
The
.code conses*
function does this in a lazy way, avoiding the
computation of the entire list: it returns a lazy list of the conses of
.metn list .
The
.code conses
function computes the entire list before returning.
The input
.meta list
may be proper or improper.
The first cons of
.meta list
is that
.meta list
itself. The second cons is the rest
of the list, or
.mono
.meti (cdr << list ).
.onom
The third cons is
.mono
.meti (cdr (cdr << list ))
.onom
and so on.
.TP* Example:
.verb
(conses '(1 2 3)) -> ((1 2 3) (2 3) (3))
.brev
.TP* "Dialect Note:"
These functions are useful for simulating the
.code maplist
function found in other dialects like Common Lisp.
\*(TL's
.code "(conses x)"
can be expressed in Common Lisp as
.codn "(maplist #'identity x)" .
Conversely, the Common Lisp operation
.code "(maplist function list)"
can be computed in \*(TL as
.codn "(mapcar function (conses list))" .
More generally, the Common Lisp operation
.verb
(maplist function list0 list1 ... listn)
.brev
can be expressed as:
.verb
(mapcar function (conses list0)
(conses list1) ... (conses listn))
.brev
.coNP Function @ delcons
.synb
.mets (delcons < cons << list )
.syne
.desc
The
.code delcons
function destructively removes a cons cell from a list. The
.meta list
is searched to see whether one of its cons cells is the same object as
.metn cons .
If so, that cell is removed from the list.
The
.meta list
argument may be a proper or improper list, possibly empty. It may also be an
atom other than
.codn nil ,
which is regarded as being, effectively, an empty improper list terminated by
that atom.
The operation of
.code delcons
is divided into the following three cases. If
.meta cons
is the first cons cell of
.metn list ,
then the
.code cdr
of
.meta list
is returned. If
.meta cons
is the second or subsequent cons of
.metn list ,
then
.meta list
is destructively altered to remove
.meta cons
and then returned. This means that the
.code cdr
field of the predecessor of
.meta cons
is altered from referencing
.meta cons
to referencing
.mono
.meti (cdr << cons )
.onom
instead.
The returned value is the same cons cell as
.metn list .
The third case occurs when
.meta cons
is not found in
.metn list .
In this situation,
.meta list
is returned unchanged.
.TP* Examples:
.verb
(let ((x (list 1 2 3)))
(delcons x x))
-> (2 3)
(let ((x (list 1 2 . 3)))
(delcons (cdr x) x))
-> (1 . 3)
.brev
.SS* Association Lists
Association lists are ordinary lists formed according to a special convention.
Firstly, any empty list is a valid association list. A nonempty association
list contains only cons cells as the key elements. These cons cells are
understood to represent key/value associations, hence the name "association
list".
.coNP Function @ assoc
.synb
.mets (assoc < key << alist )
.syne
.desc
The
.code assoc
function searches an association list
.meta alist
for a cons cell whose
.code car
field is equivalent to
.meta key
under the
.code equal
function.
The first such cons is returned. If no such cons is found,
.code nil
is returned.
.coNP Functions @ assq and @ assql
.synb
.mets (assq < key << alist )
.mets (assql < key << alist )
.syne
.desc
The
.code assq
and
.code assql
functions are very similar to
.codn assoc ,
with the only difference being that they determine equality using,
respectively, the
.code eq
and
.code eql
functions rather than
.codn equal .
.coNP Functions @, rassq @ rassql and @ rassoc
.synb
.mets (rassq < value << alist )
.mets (rassql < value << alist )
.mets (rassoc < value << alist )
.syne
.desc
The
.codn rassq ,
.code rassql
and
.code rassoc
functions are reverse lookup counterparts to
.code assql
and
.codn assoc .
When searching, they examine the
.code cdr
field of the pairs of
.meta alist
rather than the
.code car
field.
The
.code rassoc
function searches association list
.meta alist
for a cons whose
.code cdr
field equivalent to
.meta value
according to the
.code equal
function. If such a cons is found, it is returned.
Otherwise
.code nil
is returned.
The
.code rassq
and
.code rassql
functions search in the same way as
.code rassoc
but compares values using, respectively,
.code eq
and
.codn eql .
.coNP Function @ acons
.synb
.mets (acons < car < cdr << alist )
.syne
.desc
The
.code acons
function constructs a new alist by consing a new cons to the
front of
.metn alist .
The following equivalence holds:
.verb
(acons car cdr alist) <--> (cons (cons car cdr) alist)
.brev
.coNP Function @ acons-new
.synb
.mets (acons-new < car < cdr << alist )
.syne
.desc
The
.code acons-new
function searches
.metn alist ,
as if using the assoc function,
for an existing cell which matches the key provided by the car argument.
If such a cell exists, then its cdr field is overwritten with the
.meta cdr
argument, and then the
.meta alist
is returned. If no such cell exists, then
a new list is returned by adding a new cell to the input list consisting
of the
.meta car
and
.meta cdr
values, as if by the
.code acons
function.
.coNP Function @ aconsql-new
.synb
.mets (aconsql-new < car < cdr << alist )
.syne
.desc
The
.code aconsql-new
function has similar same parameters and semantics as
.codn acons-new ,
except that the
.code eql
function is used
for equality testing. Thus, the list is searched for an existing cell
as if using the
.code assql
function rather than
.codn assoc .
.coNP Function @ alist-remove
.synb
.mets (alist-remove < alist << key *)
.syne
.desc
The
.code alist-remove
function takes association list
.meta alist
and produces a
duplicate from which cells matching any of the specified
.metn key s
have been removed.
.coNP Function @ alist-nremove
.synb
.mets (alist-nremove < alist << key *)
.syne
.desc
The
.code alist-nremove
function is like
.codn alist-remove ,
but potentially destructive.
The input list
.meta alist
may be destroyed and its structural material reused to
form the output list. The application should not retain references to the input
list.
.coNP Function @ copy-alist
.synb
.mets (copy-alist << alist )
.syne
.desc
The
.code copy-alist
function duplicates
.codn alist .
Unlike
.codn copy-list ,
which only duplicates list structure,
.code copy-alist
also duplicates each cons
cell of the input alist. That is to say, each element of the output list
is produced as if by the
.code copy-cons
function applied to the corresponding
element of the input list.
.coNP Function @ pairlis
.synb
.mets (pairlis < keys < values <> [ alist ])
.syne
.desc
The
.code pairlis
function returns an association list consisting of pairs formed from
the elements of
.meta keys
and
.meta values
prepended to the existing
.metn alist .
If an
.meta alist
argument is omitted, it defaults to
.codn nil .
Pairs of elements are formed by taking successive elements from the
.meta keys
and
.meta values
sequences in parallel.
If the sequences are not of equal length, the excess elements from
the longer sequence are ignored.
The pairs appear in the resulting list in the original order in
which their constituents appeared in
.meta keys
and
.metn values .
.TP* "Dialect Note:"
The ANSI CL
.code pairlis
requires
.meta key
and
.meta data
to be lists, not sequences. The behavior of the ANSI CL
.code pairlis
is undefined of those lists are of different lengths. Finally, the elements are
permitted to appear in either the original order or reverse order.
.TP* Examples:
.verb
(pairlis nil nil) -> nil
(pairlis "abc" #(1 2 3 4)) -> ((#\ea . 1) (#\eb . 2) (#\ec . 3))
(pairlis '(1 2 3) '(a b c) '((x . y) (z . w)))
-> ((1 . a) (2 . b) (3 . c) (x . y) (z . w))
.brev
.SS* Property Lists
A
.IR "property list",
also referred to as a
.IR plist ,
is a flat list of even length consisting of interleaved
pairs of property names (usually symbols) and their values (arbitrary
objects). An example property list is (:a 1 :b "two") which contains
two properties, :a having value 1, and :b having value "two".
An
.I "improper plist"
represents Boolean properties in a condensed way, as property
indicators which are not followed by a value. Such properties
only indicate their presence or absence, which is useful for
encoding a Boolean value. If it is absent, then the property
is false. Correctly using an improper plist requires that the
exact set of Boolean keys is established by convention.
In this document, the unqualified terms
.I "property list"
and
.I "plist"
refer strictly to an ordinary plist, not to an improper plist.
.TP* "Dialect Note:"
Unlike in some other Lisp dialects, including ANSI Common Lisp,
symbols do not have property lists in \*(TL. Improper plists
aren't a concept in ANSI CL.
.coNP Function @ prop
.synb
.mets (prop < plist << key )
.syne
.desc
The
.code prop
function searches property list
.meta plist
for key
.metn key .
If the key is found, then the value next to it is returned. Otherwise
.code nil
is returned.
It is ambiguous whether
.code nil
is returned due to the property not being
found, or due to the property being present with a
.code nil
value.
The indicators in
.meta plist
are compared with
.meta key
using
.code eq
equality, allowing them to be symbols, characters or
.code fixnum
integers.
.coNP Function @ memp
.synb
.mets (memp < key << plist )
.syne
.desc
The
.code memp
function searches property list
.meta plist
for key
.metn key ,
using
.code eq
equality.
If the key is found, then the entire suffix of
.meta plist
beginning with the indicator is returned, such that the first
element of the returned list is
.meta key
and the second element is the property value.
Note the reversed argument convention relative to the
.code prop
function, harmonizing with functions in the
.code member
family.
.coNP Functions @ plist-to-alist and @ improper-plist-to-alist
.synb
.mets (plist-to-alist << plist )
.mets (improper-plist-to-alist < imp-plist << bool-keys )
.syne
.desc
The functions
.code plist-to-alist
and
.code improper-plist-to-alist
convert, respectively, a property list and improper property
list to an association list.
The
.code plist-to-alist
function scans
.meta plist
and returns the indicator-property pairs as a list of cons
cells, such that each
.code car
is the indicator, and each
.code cdr
is the value.
The
.code improper-plist-to-alist
is similar, except that it handles the Boolean properties
which, by convention, aren't followed by a value. The list of
all such indicators is specified by the
.code bool-keys
argument.
.TP* "Examples:"
.verb
(plist-to-alist '(a 1 b 2)) --> ((a . 1) (b . 2))
(improper-plist-to-alist '(:x 1 :blue :y 2) '(:blue))
--> ((:x . 1) (:blue) (:y . 2))
.brev
.SS* List Sorting
Note: these functions operate on lists. The principal sorting function
in \*(TL is
.codn sort ,
described under Sequence Manipulation.
The
.code merge
function described here provides access to an elementary step
of the algorithm used internally by
.code sort
when operating on lists.
The
.code multi-sort
operation sorts multiple lists in parallel. It is implemented using
.codn sort .
.coNP Function @ merge
.synb
.mets (merge < seq1 < seq2 >> [ lessfun <> [ keyfun ]])
.syne
.desc
The
.code merge
function merges two sorted sequences
.meta seq1
and
.meta seq2
into a single
sorted sequence. The semantics and defaulting behavior of the
.meta lessfun
and
.meta keyfun
arguments are the same as those of the sort function.
The sequence which is returned is of the same kind as
.metn seq1 .
This function is destructive of any inputs that are lists. If the output
is a list, it is formed out of the structure of the input lists.
.coNP Function @ multi-sort
.synb
.mets (multi-sort < columns < less-funcs <> [ key-funcs ])
.syne
.desc
The
.code multi-sort
function regards a list of lists to be the columns of a
database. The corresponding elements from each list constitute a record.
These records are to be sorted, producing a new list of lists.
The
.meta columns
argument supplies the list of lists which comprise the columns of
the database. The lists should ideally be of the same length. If the lists are
of different lengths, then the shortest list is taken to be the length of the
database. Excess elements in the longer lists are ignored, and do not appear in
the sorted output.
The
.meta less-funcs
argument supplies a list of comparison functions which are
applied to the columns. Successive functions correspond to successive
columns. If
.meta less-funcs
is an empty list, then the sorted database will
emerge in the original order. If
.meta less-funcs
contains exactly one function,
then the rows of the database are sorted according to the first column. The
remaining columns simply follow their row. If
.meta less-funcs
contains more than
one function, then additional columns are taken into consideration if the items
in the previous columns compare
.codn equal .
For instance if two elements from column
one compare
.codn equal ,
then the corresponding second column elements are compared
using the second column comparison function.
The
.meta less-funcs
argument may be a function object, in which case it is treated as if
it were a one-element list containing that function object.
The optional
.meta key-funcs
argument supplies transformation functions through
which column entries are converted to comparison keys, similarly to the single
key function used in the sort function and others. If there are more key
functions than less functions, the excess key functions are ignored.
.SS* Lazy Lists and Lazy Evaluation
.coNP Function @ make-lazy-cons
.synb
.mets (make-lazy-cons < function >> [ car <> [ cdr ]])
.syne
.desc
The function
.code make-lazy-cons
makes a special kind of cons cell called a lazy
cons, whose type is
.codn lcons .
Lazy conses are useful for implementing lazy lists.
Lazy lists are lists which are not allocated all at once. Rather,
the elements of its structure materialize just before they are accessed.
A lazy cons has
.code car
and
.code cdr
fields like a regular cons, and those
fields are initialized to the values of the
.meta car
and
.meta cdr
arguments of
.code make-lazy-cons
when the lazy cons is created. These arguments default to
.meta nil
if omitted. A lazy cons also
has an update function, which is specified by the
.meta function
argument to
.codn make-lazy-cons .
The
.meta function
argument must be a function that may be called with exactly one
parameter.
When either the
.code car
and
.code cdr
fields of a cons are accessed for the first time to retrieve their value,
.meta function
is automatically invoked first, and is given the lazy cons
as a parameter. That function has the opportunity
to store new values into the
.code car
and
.code cdr
fields. Once the function is called, it is removed
from the lazy cons: the lazy cons no longer has an update function.
If the update function itself attempts to retrieve the value of the
lazy cons cell's
.code car
or
.code cdr
field, it will be recursively invoked.
The functions
.code lcons-car
and
.code lcons-cdr
may be used to access the fields of a lazy cons without
triggering the update function.
Storing a value into either the
.code car
or
.code cdr
field does not have the effect of invoking the update function.
If the function terminates by returning normally, the access to the value of
the field then proceeds in the ordinary manner, retrieving whatever value has
most recently been stored.
The return value of the function is ignored.
To perpetuate the growth of a lazy list, the function can make another call to
.code make-lazy-cons
and install the resulting cons as the
.code cdr
of the lazy cons.
.TP* Example:
.verb
;;; lazy list of integers between min and max
(defun integer-range (min max)
(let ((counter min))
;; min is greater than max; just return empty list,
;; otherwise return a lazy list
(if (> min max)
nil
(make-lazy-cons
(lambda (lcons)
;; install next number into car
(rplaca lcons counter)
;; now deal wit cdr field
(cond
;; max reached, terminate list with nil!
((eql counter max)
(rplacd lcons nil))
;; max not reached: increment counter
;; and extend with another lazy cons
(t
(inc counter)
(rplacd lcons
(make-lazy-cons
(lcons-fun lcons))))))))))
.brev
.coNP Function @ lconsp
.synb
.mets (lconsp << value )
.syne
.desc
The
.code lconsp
function returns
.code t
if
.meta value
is a lazy cons cell. Otherwise
it returns
.codn nil ,
even if
.meta value
is an ordinary cons cell.
.coNP Function @ lcons-fun
.synb
.mets (lcons-fun << lazy-cons )
.syne
.desc
The
.code lcons-fun
function retrieves the update function of a lazy cons.
Once a lazy cons has been accessed, it no longer has an update function
and
.code lcons-fun
returns
.codn nil .
While the update function of a lazy cons is
executing, it is still accessible. This allows the update function
to retrieve a reference to itself and propagate itself into
another lazy cons (as in the example under
.codn make-lazy-cons ).
.coNP Functions @ lcons-car and @ lcons-cdr
.synb
.mets (lcons-car << lazy-cons )
.mets (lcons-cdr << lazy-cons )
.syne
.desc
The functions
.code lcons-car
and
.code lcons-cdr
retrieve the
.code car
and
.code cdr
fields of
.metn lazy-cons ,
without triggering the invocation of its associated update function.
The
.meta lazy-cons
argument must be an object of type
.codn lcons .
Unlike the functions
.code car
and
.codn cdr ,
These functions cannot be applied to any other type of object.
Note: these functions may be used by the update function to retrieve
the values which were stored into
.meta lazy-cons
by the
.code make-lazy-cons
constructor, without triggering recursion. The function may then
overwrite either or both of these values. This allows the fields of the lazy
cons to store state information necessary for the propagation of a lazy list.
If that state information consists of no more than two values, then
no additional context object need be allocated.
.coNP Macro @ lcons
.synb
.mets (lcons < car-expression << cdr-expression )
.syne
.desc
The
.code lcons
macro simplifies the construction of structures based on lazy conses.
Syntactically, it resembles the
.code cons
function. However, the arguments are expressions rather than values.
The macro generates code which, when evaluated, immediately produces
a lazy cons. The expressions
.meta car-expression
and
.meta cdr-expression
are not immediately evaluated. Rather, when either the
.code car
or
.code cdr
field of the lazy cons cell is accessed, these expressions are both
evaluated at that time, in the order that they appear in the
.code lcons
expression, and in the original lexical scope in which that
expression was evaluated. The return values of these expressions
are used, respectively, to initialize the corresponding fields
of the lazy cons.
Note: the
.code lcons
macro may be understood in terms of the following reference
implementation, as a syntactic sugar combining the
.code make-lazy-cons
constructor with a lexical closure provided by a
.code lambda
function:
.verb
(defmacro lcons (car-form cdr-form)
(let ((lc (gensym)))
^(make-lazy-cons (lambda (,lc)
(rplaca ,lc ,car-form)
(rplacd ,lc ,cdr-form)))))
.brev
.TP* Example:
.verb
;; Given the following function ...
(defun fib-generator (a b)
(lcons a (fib-generator b (+ a b))))
;; ... the following function call generates the Fibonacci
;; sequence as an infinite lazy list.
(fib-generator 1 1) -> (1 1 2 3 5 8 13 ...)
.brev
.coNP Functions @ lazy-stream-cons and @ get-lines
.synb
.mets (lazy-stream-cons < stream <> [ no-throw-close-p ])
.mets (get-lines >> [ stream <> [ no-throw-close-p ]])
.syne
.desc
The
.code lazy-stream-cons
and
.code get-lines
functions are synonyms, except that the
.meta stream
argument is optional in
.code get-lines
and defaults to
.codn *stdin* .
Thus, the following
description of
.code lazy-stream-cons
also applies to
.codn get-lines .
The
.code lazy-stream-cons
returns a lazy cons which generates a lazy list based on
reading lines of text from input stream
.metn stream ,
which form the elements of
the list. The
.code get-line
function is called on demand to add elements to the
list.
The
.code lazy-stream-cons
function itself makes the first call to
.code get-line
on the stream. If this returns
.codn nil ,
then the stream is closed and
.code nil
is
returned. Otherwise, a lazy cons is returned whose update function will install
that line into the
.code car
field of the lazy cons, and continue the lazy list
by making another call to
.codn lazy-stream-cons ,
installing the result into the
.code cdr
field. When this lazy list obtains an end-of-file indication from the stream,
it closes the stream.
.code lazy-stream-cons
inspects the real-time property of a stream
as if by the
.code real-time-stream-p
function. This determines which of two
styles of lazy list are returned. For an ordinary (non-real-time) stream,
the lazy list treats the end-of-file condition accurately: an empty
file turns into the empty list
.codn nil ,
a one line file into a one-element
list which contains that line and so on. This accuracy requires one
line of lookahead which is not acceptable in real-time streams, and
so a different type of lazy list is used, which generates an extra
.code nil
item after the last line. Under this type of lazy list, an empty input stream
translates to the list
.codn (nil) ;
a one-line stream translates to
.mono
("line" nil)
.onom
and so forth.
If and when
.meta stream
is closed by the function directly, or else by the returned lazy list, the
.meta no-throw-close-p
Boolean argument, defaulting to
.codn nil ,
controls the
.meta throw-on-error-p
argument of the call to the
.code close-stream
function. These arguments have opposite polarity: if
.meta no-throw-close-p
is true, then
.meta throw-on-error-p
shall be false, and vice versa.
.coNP Macro @ delay
.synb
.mets (delay << expression )
.syne
.desc
The delay operator arranges for the delayed (or "lazy") evaluation of
.metn expression .
This means that the expression is not evaluated immediately.
Rather, the delay expression produces a promise object.
The promise object can later be passed to the
.code force
function (described
later in this document). The force function will trigger the evaluation
of the expression and retrieve the value.
The expression is evaluated in the original scope, no matter where
the
.code force
takes place.
The expression is evaluated at most once, by the first call to
.codn force .
Additional calls to
.code force
only retrieve a cached value.
.TP* Example:
.verb
;; list is popped only once: the value is computed
;; just once when force is called on a given promise
;; for the first time.
(defun get-it (promise)
(format t "*list* is ~s\en" *list*)
(format t "item is ~s\en" (force promise))
(format t "item is ~s\en" (force promise))
(format t "*list* is ~s\en" *list*))
(defvar *list* '(1 2 3))
(get-it (delay (pop *list*)))
Output:
*list* is (1 2 3)
item is 1
item is 1
*list* is (2 3)
.brev
.coNP Accessor @ force
.synb
.mets (force << promise )
.mets (set (force << promise ) << new-value )
.syne
.desc
The
.code force
function accepts a promise object produced by the
.code delay
macro.
The first time
.code force
is invoked, the
.meta expression
which was wrapped inside
.meta promise
by the
.code delay
macro is evaluated (in its original lexical environment, regardless of where in
the program the
.code force
call takes place). The value of
.meta expression
is
cached inside
.meta promise
and returned, becoming the return value of the
.code force
function call. If the
.code force
function is invoked additional times on
the same promise, the cached value is retrieved.
A
.code force
form is a syntactic place, denoting the value cache location within
.metn promise .
Storing a value in a
.code force
place causes future accesses to the
.meta promise
to return that value.
If the promise had not yet been forced, then
storing a value into it prevents that from ever happening. The
delayed
.meta expression
will never be evaluated.
If, while a promise is being forced, the evaluation of
.meta expression
itself causes an assignment to the promise, it is not specified whether
the promise will take on the value of
.meta expression
or the assigned value.
.coNP Function @ promisep
.synb
.mets (promisep << object )
.syne
.desc
The
.code promisep
function returns
.code t
if
.meta object
is a promise object: an object created by the
.code delay
macro. Otherwise it returns
.codn nil .
Note: promise objects are conses. The
.code typeof
function applied to a promise returns
.codn cons .
.coNP Macro @ mlet
.synb
.mets (mlet >> ({ sym | >> ( sym << init-form )}*) << body-form *)
.syne
.desc
The
.code mlet
macro ("magic let" or "mutual let") implements a variable binding construct
similar to
.code let
and
.codn let* .
Under
.codn mlet ,
the scope of the bindings of the
.meta sym
variables extends over the
.metn init-form s,
as well as the
.metn body-form s.
Unlike the
.code let*
construct, each
.meta init-form
has each
.meta sym
in scope. That is to say, an
.meta init-form
can refer not only to previous variables, but also to later variables
as well as to its own variable.
The variables are not initialized until their values are accessed for
the first time. Any
.meta sym
whose value is not accessed is not initialized.
Furthermore, the evaluation of each
.meta init-form
does not take place until the time when its value is needed
to initialize the associated
.metn sym .
This evaluation takes place once. If a given
.meta sym
is not accessed during the evaluation of the
.code mlet
construct, then its
.meta init-form
is never evaluated.
The bound variables may be assigned. If, before initialization, a variable is
updated in such a way that its prior value is not needed, it is unspecified
whether initialization takes place, and thus whether its
.meta init-form
is evaluated.
Direct circular references are erroneous and are diagnosed. This takes
place when the macro-expanded form is evaluated, not during the
expansion of
.codn mlet .
.TP* Examples:
.verb
;; Dependent calculations in arbitrary order
(mlet ((x (+ y 3))
(z (+ x 1))
(y 4))
(+ z 4)) --> 12
;; Error: circular reference:
;; x depends on y, y on z, but z on x again.
(mlet ((x (+ y 1))
(y (+ z 1))
(z (+ x 1)))
z)
;; Okay: lazy circular reference because lcons is used
(mlet ((list (lcons 1 list)))
list) --> (1 1 1 1 1 ...) ;; circular list
.brev
In the last example, the
.code list
variable is accessed for the first time in the body of the
.code mlet
form. This causes the evaluation of the
.code lcons
form. This form evaluates its arguments lazily, which means that it
is not a problem that
.code list
is not yet initialized. The form produces a lazy cons, which is then used
to initialize
.code list.
When the
.code car
or
.code cdr
fields of the lazy cons are accessed, the
.code list
expression in the
.code lcons
argument is accessed. By that time, the variable is initialized
and holds the lazy cons itself, which creates the circular reference,
and a circular list.
.coNP Functions @, generate @ giterate and @ ginterate
.synb
.mets (generate < while-fun << gen-fun )
.mets (giterate < while-fun < gen-fun <> [ value ])
.mets (ginterate < while-fun < gen-fun <> [ value ])
.syne
.desc
The
.code generate
function produces a lazy list which dynamically produces items
according to the following logic.
The arguments to
.code generate
are functions which do not take any arguments. The
return value of generate is a lazy list.
When the lazy list is accessed, for instance with the functions car and cdr, it
produces items on demand. Prior to producing each item,
.meta while-fun
is
called. If it returns a true Boolean value (any value other than
.codn nil ),
then
the
.meta gen-fun
function is called, and its return value is incorporated as
the next item of the lazy list. But if
.meta while-fun
yields
.codn nil ,
then the lazy list immediately terminates.
Prior to returning the lazy list, generate invokes the
.meta while-fun
one time.
If
.code while-fun
yields
.codn nil ,
then
.code generate
returns the empty list
.code nil
instead of a lazy list. Otherwise, it instantiates a lazy list, and invokes the
.code gen-fun
to populate it with the first item.
The
.code giterate
function is similar to
.codn generate ,
except that
.meta while-fun
and
.meta gen-fun
are functions of one argument rather than functions of
no arguments. The optional
.meta value
argument defaults to
.code nil
and is threaded through the function calls. That is to say, the lazy
list returned is
.mono
.meti >> ( value >> [ gen-fun << value ] >> [ gen-fun >> [ gen-fun << value ]] ...).
.onom
The lazy list terminates when a value fails to satisfy
.metn while-fun .
That is to say, prior to generating each value, the lazy list tests
the value using
.metn while-fun .
If that function returns
.codn nil ,
then the item is not added, and the sequence terminates.
Note:
.code giterate
could be written in terms of
.code generate
like this:
.verb
(defun giterate (w g v)
(generate (lambda () [w v])
(lambda () (prog1 v (set v [g v])))))
.brev
The
.code ginterate
function is a variant of
.code giterate
which includes the test-failing item in the generated sequence.
That is to say
.code ginterate
generates the next value and adds it to the lazy list.
The value is then tested using
.metn while-fun .
If that function returns
.codn nil ,
then the list is terminated, and no more items are produced.
.TP* Example:
.verb
(giterate (op > 5) (op + 1) 0) -> (0 1 2 3 4)
(ginterate (op > 5) (op + 1) 0) -> (0 1 2 3 4 5)
.brev
.coNP Function @ expand-right
.synb
.mets (expand-right < gen-fun << value )
.syne
.desc
The
.code expand-right
function is a complement to
.codn reduce-right ,
with lazy semantics.
The
.meta gen-fun
parameter is a function, which must accept a single argument,
and return either a cons pair
or
.codn nil .
The
.meta value
parameter is any value.
The first call to
.meta gen-fun
receives
.metn value .
The return value is interpreted as follows. If
.meta gen-fun
returns a cons-cell pair
.mono
.meti >> ( elem . << next )
.onom
then
.meta elem
specifies the element to be added to the lazy list,
and
.meta next
specifies the value to be passed to the next call
to
.metn gen-fun .
If
.meta gen-fun
returns
.code nil
then the lazy list ends.
.TP* Examples:
.verb
;; Count down from 5 to 1 using explicit lambda
;; for gen-fun:
(expand-right
(lambda (item)
(if (zerop item) nil
(cons item (pred item))))
5)
--> (5 4 3 2 1)
;; Using functional combinators:
[expand-right [iff zerop nilf [callf cons identity pred]] 5]
--> (5 4 3 2 1)
;; Include zero:
[expand-right
[iff null
nilf
[callf cons identity [iff zerop nilf pred]]] 5]
--> (5 4 3 2 1 0)
.brev
.coNP Functions @ expand-left and @ nexpand-left
.synb
.mets (expand-left < gen-fun << value )
.mets (nexpand-left < gen-fun << value )
.syne
.desc
The
.code expand-left
function is a companion to
.codn expand-right .
Unlike
.codn expand-right ,
it has eager semantics: it calls
.code gen-fun
repeatedly and accumulates an output list, not returning
until
.code gen-fun
returns
.codn nil .
The semantics is as follows.
.code expand-left
initializes an empty accumulation list. Then
.meta gen-fun
is called, with
.meta value
as its argument.
If
.meta gen-fun
it returns a cons cell, then the
.code car
of that cons cell is pushed onto the accumulation list,
and the procedure is repeated:
.meta gen-fun
is called again, with
.code cdr
taking the place of
.metn value .
If
.meta gen-fun
returns
.codn nil ,
then the accumulation list is returned.
If the expression
.code "(expand-right f v)"
produces a terminating list, then the following equivalence holds:
.verb
(expand-left f v) <--> (reverse (expand-right f v))
.brev
The equivalence cannot hold for arguments to
.code expand-left
which produce an infinite list.
The
.code nexpand-left
function is a destructive version of
.codn expand-left .
The list returned by
.code nexpand-left
is composed of the cons cells returned by
.code gen-fun
whereas the list returned by
.code expand-left
is composed of freshly allocated cons cells.
.coNP Function @ repeat
.synb
.mets (repeat < list <> [ count ])
.syne
.desc
If
.meta list
is empty, then repeat returns an empty list.
If
.meta count
is omitted, the
.code repeat
function produces an infinite lazy list
formed by catenating together copies of
.metn list .
If
.meta count
is specified and is zero or negative, then an empty list is
returned.
Otherwise a list is returned consisting of
.meta count
repetitions of
.meta list
catenated together.
.coNP Function @ pad
.synb
.mets (pad < sequence < object <> [ count ])
.syne
.desc
The
.code pad
function produces a lazy list which consists of all of the
elements of
.meta sequence
followed by repetitions of
.metn object .
If
.meta count
is omitted, then the repetition of
.meta object
is infinite. Otherwise the specified number of repetitions
occur.
Note that
.meta sequence
may be a lazy list which is infinite. In that case, the repetitions
of
.meta object
will never occur.
.coNP Function @ weave
.synb
.mets (weave <> { sequence }*)
.syne
.desc
The
.code weave
function interleaves elements from the sequences given as arguments.
If called with no arguments, it returns the empty list.
If called with a single sequence, it returns the elements of that sequence
as a new lazy list.
When called with two or more sequences,
.code weave
returns a lazy list which draws elements from the sequences in a round-robin
fashion, repeatedly scanning the sequences from left to right, and
taking an item from each one, removing it from the sequence.
Whenever a sequence runs out of items, it is deleted; the weaving then
continues with the remaining sequences. The weaved sequence terminates
when all sequences are eliminated. (If at least one of the sequences
is an infinite lazy list, then the weaved sequence is infinite.)
.TP* Examples:
.verb
;; Weave negative integers with positive ones:
(weave (range 1) (range -1 : -1)) -> (1 -1 2 -2 3 -3 ...)
(weave "abcd" (range 1 3) '(x x x x x x x))
--> (#\ea 1 x #\eb 2 x #\ec 3 x #\ed x x x x)
.brev
.coNP Macros @ gen and @ gun
.synb
.mets (gen < while-expression << produce-item-expression )
.mets (gun << produce-item-expression )
.syne
.desc
The
.code gen
macro operator produces a lazy list, in a manner similar to the
.code generate
function. Whereas the
.code generate
function takes functional arguments,
the
.code gen
operator takes two expressions, which is often more convenient.
The return value of
.code gen
is a lazy list. When the lazy list is accessed, for
instance with the functions
.code car
and
.codn cdr ,
it produces items on demand. Prior to
producing each item, the
.meta while-expression
is evaluated, in its original
lexical scope. If the expression yields a
.cod2 non- nil
value, then
.meta produce-item-expression
is evaluated, and its return value is incorporated as
the next item of the lazy list. If the expression yields
.codn nil ,
then the lazy list immediately terminates.
The
.code gen
operator itself immediately evaluates
.meta while-expression
before
producing the lazy list. If the expression yields
.codn nil ,
then the operator
returns the empty list
.codn nil .
Otherwise, it instantiates the lazy list and
invokes the
.meta produce-item-expression
to force the first item.
The
.code gun
macro similarly creates a lazy list according to the following
rules. Each successive item of the lazy list is obtained as a result of
evaluating
.metn produce-item-expression .
However, when
.meta produce-item-expression
yields
.codn nil ,
then the list terminates (without adding that
.code nil
as an item).
Note 1: the form
.code gun
can be implemented as a macro-expanding to
an instance of the
.code gen
operator, like this:
.verb
(defmacro gun (expr)
(let ((var (gensym)))
^(let (,var)
(gen (set ,var ,expr)
,var))))
.brev
This exploits the fact that the
.code set
operator returns the value that is
assigned, so the set expression is tested as a condition by
.codn gen ,
while having the side effect of storing the next item temporarily
in a hidden variable.
In turn,
.code gen
can be implemented as a macro expanding to some
.code lambda
functions which are passed to the
.code generate
function:
.verb
(defmacro gen (while-expr produce-expr)
^(generate (lambda () ,while-expr)
(lambda () ,produce-expr)))
.brev
Note 2:
.code gen
can be considered as an acronym for Generate, testing Expression
before Next item, whereas
.code gun
stands for Generate Until Null.
.TP* Example:
.verb
;; Make a lazy list of integers up to 1000
;; access and print the first three.
(let* ((counter 0)
(list (gen (< counter 1000) (inc counter))))
(format t "~s ~s ~s\en" (pop list) (pop list) (pop list)))
Output:
1 2 3
.brev
.coNP Functions @ range and @ range*
.synb
.mets (range >> [ from >> [ to <> [ step ]]])
.mets (range* >> [ from >> [ to <> [ step ]]])
.syne
.desc
The
.code range
and
.code range*
functions generate a lazy sequence of integers, with a
fixed step between successive values.
The difference between
.code range
and
.code range*
is that
.code range*
excludes the endpoint.
For instance
.code "(range 0 3)"
generates the list
.codn "(0 1 2 3)" ,
whereas
.code "(range* 0 3)"
generates
.codn "(0 1 2)" .
All arguments are optional. If the
.meta step
argument is omitted, then it defaults
to
.codn 1 :
each value in the sequence is greater than the previous one by
.codn 1 .
Positive or negative step sizes are allowed. There is no check for a step size
of zero, or for a step direction which cannot meet the endpoint.
The
.meta to
argument specifies the endpoint value, which, if it occurs in the
sequence, is excluded from it by the
.code range*
function, but included by the range
function. If
.meta to
is missing, or specified as
.codn nil ,
then there is no endpoint,
and the sequence which is generated is infinite, regardless of
.metn step .
If
.meta from
is omitted, then the sequence begins at zero, otherwise
.meta from
must be an integer which specifies the initial value.
The sequence stops if it reaches the endpoint value (which is included in the
case of
.codn range ,
and excluded in the case of
.codn range *).
However, a sequence with a stepsize greater than
.code 1
or less than
.code -1
might step over the endpoint value, and
therefore never attain it. In this situation, the sequence also stops, and the
excess value which surpasses the endpoint is excluded from the sequence.
.coNP Functions @ rlist and @ rlist*
.synb
.mets (rlist << item *)
.mets (rlist* << item *)
.syne
.desc
The
.code rlist
("range list") function is useful for producing a list consisting of a mixture
of discontinuous numeric or character ranges and individual items.
The function returns a lazy list of elements. The items are produced
by converting the function's successive
.meta item
arguments into lists, which are lazily catenated together to form the
output list.
Each
.meta item
is transformed into a list as follows. Any item which is
.B not
a range object is trivially turned into a one-element list as if by the
.mono
.meti (list << item *)
.onom
expression.
Any item which is a range object, whose
.code to
field
.B isn't
a range is turned into a lazy list as if by evaluating the
.mono
.meti (range (from << item) (to << item))
.onom
expression. Thus for instance the argument
.code 1..10
turns into the (lazy) list
.codn "(1 2 3 4 5 6 7 8 9 10)" .
Any item which is a range object such that its
.code to
field is also a range is turned into a lazy list as if by evaluating the
.mono
.meti (range (from << item) (from (to << item)) (to (to << item)))
.onom
expression. Thus for instance the argument expression
.code 1..10..2
produces an
.meta item
which
.code rlist
turns into the lazy list
.code "(1 3 5 7 9)"
as if by the call
.codn "(range 1 10 2)" .
Note that the expression
.code 1..10..2
stands for the expression
.code "(rcons 1 (rcons 10 2))"
which evaluates to
.codn "#R(1 #R(10 2))" .
The
.code "#R(1 #R(10 2))"
range literal syntax can be passed as an argument to
.code rlist
with the same result as
.codn 1..10..2 .
The
.code rlist*
function differs from
.code rlist
in one regard: under
.codn rlist* ,
the ranges denoted by the range notation exclude the endpoint. That is,
the ranges are generated as if by the
.code range*
function rather than
.codn range .
Note: it is permissible for
.meta item
objects to specify infinite ranges.
It is also permissible to apply
.code rlist
to an infinite argument list.
.TP* Examples:
.verb
(rlist 1 "two" :three) -> (1 "two" :three)
(rlist 10 15..16 #\ea..#\ed 2) -> (10 15 16 #\ea #\eb #\ec #\ed 2)
(take 7 (rlist 1 2 5..:)) -> (1 2 5 6 7 8 9)
.brev
.SS* Ranges
Ranges are objects that aggregate two values, not unlike
.code cons
cells. However, they are atoms, and are primarily intended to hold numeric or
character values in their two fields. These fields are called
.code from
and
.code to
which are the names of the functions which access them. These fields
are not mutable; a new value cannot be stored into either field of
a range.
The printed notation for a range object consists of the prefix
.code #R
(hash R) followed by the two values expressed as a two-element
list. Ranges can be constructed using the
.code rcons
function. The notation
.code x..y
corresponds to
.codn "(rcons x y)" .
Ranges behave as a numeric type and support a subset of the numeric
operations. Two ranges can be added or subtracted, which obeys
these equivalences:
.verb
(+ a..b c..d) <--> (+ a c)..(+ b d)
(- a..b c..d) <--> (- a c)..(- b d)
.brev
A range
.code a..b
can be combined with a character or number
.code n
using addition or subtractions, which obeys these equivalences:
.verb
(+ a..b n) <--> (+ n a..b) <--> (+ a n)..(+ b n)
(- a..b n) <--> (- a n)..(- b n)
(- n a..b) <--> (- n a)..(- n b)
.brev
A range can be multiplied by a number:
.verb
(* a..b n) <--> (* n a..b) <--> (* a n)..(* b n)
.brev
A range can be divided by a number using the
.code /
or
.code trunc
functions, but a number cannot be divided by a range:
.verb
(trunc a..b n) <--> (trunc a n)..(trunc b n)
(/ a..b n) <--> (/ a n)..(/ b n)
.brev
Ranges can be compared using the equality and inequality functions
.codn = ,
.codn < ,
.codn > ,
.code <=
and
.codn >= .
Equality obeys this equivalence:
.verb
(= a..b c..d) <--> (and (= a c) (= b d))
.brev
Inequality comparisons treat the
.code from
component with precedence over
.code to
such that only if the
.code from
components of the two ranges are not equal under the
.code =
function, then the inequality is based solely on them.
If they are equal, then the inequality is based on the
.code to
components. This gives rise to the following equivalences:
.verb
(< a..b c..d) <--> (if (= a c) (< b d) (< a c))
(> a..b c..d) <--> (if (= a c) (> b d) (> a c))
(>= a..b c..d) <--> (if (= a c) (>= b d) (> a c))
(<= a..b c..d) <--> (if (= a c) (<= b d) (< a c))
.brev
Ranges can be negated with the one-argument form of the
.code -
function, which is equivalent to subtraction from zero:
the negation distributes over the two range components.
The
.code abs
function also applies to ranges and distributes into
their components.
The
.code succ
and
.code pred
family of functions also operate on ranges.
The length of a range may be obtained with the
.code length
function;
The length of the range
.code a..b
is defined as
.codn "(- b a)" ,
and may be obtained using the
.code length
function. The
.code empty
function accepts ranges and tests them for zero length.
.coNP Function @ rcons
.synb
.mets (rcons < from << to )
.syne
.desc
The
.code rcons
function constructs a range object which holds the values
.meta from
and
.metn to .
Though range objects are effectively binary cells like conses, they are atoms.
They also aren't considered sequences, nor are they structures.
Range objects are used for indicating numeric ranges, such as substrings of
lists, arrays and strings. The dotdot notation serves as a syntactic sugar for
.codn rcons .
The syntax
.code a..b
denotes the expression
.codn "(rcons a b)" .
Note that ranges are immutable, meaning that it is not possible to
replace the values in a range.
.coNP Function @ rangep
.synb
.mets (rangep << value )
.syne
.desc
The
.code rangep
function returns
.code t
if
.meta value
is a range. Otherwise it returns
.codn nil .
.coNP Functions @ from and @ to
.synb
.mets (from << range )
.mets (to << range )
.syne
.desc
The
.code from
and
.code to
functions retrieve, respectively, the from and to fields
of a range.
Note that these functions are not accessors, which is because
ranges are immutable.
.coNP Functions @ in-range and @ in-range*
.synb
.mets (in-range < range << value )
.mets (in-range* < range << value )
.syne
.desc
The
.code in-range
and
.code in-range*
functions test whether the
.meta value
argument lies in the range represented by the
.meta range
argument, indicating the Boolean result using one of the values
.code t
or
.codn nil .
The
.meta range
argument must be a range object.
It is expected that the range object's
.code from
value does not exceed the
.code to
value; a reversed range is considered empty.
The
.code in-range*
function differs from
.code in-range
in that it excludes the
upper endpoint.
The implicit comparison against the range endpoints is performed
using the
.code less
and
.code lequal
functions, as appropriate.
The following equivalences hold:
.verb
(in-range r x) <--> (and (lequal (from r) x)
(lequal x (to r)))
(in-range* r x) <--> (and (lequal (from r) x)
(less x (to r)))
.brev
.SS* Characters and Strings
.coNP Function @ mkstring
.synb
.mets (mkstring < length <> [ char ])
.syne
.desc
The
.code mkstring
function constructs a string object of a length specified
by the
.meta length
parameter. Every position in the string is initialized
with
.metn char ,
which must be a character value.
If the optional argument
.meta char
is not specified, it defaults to the space character.
.coNP Function @ copy-str
.synb
.mets (copy-str << string )
.syne
.desc
The
.code copy-str
function constructs a new string whose contents are identical
to
.metn string .
If
.meta string
is a lazy string, then a lazy string is constructed with the
same attributes as
.metn string .
The new lazy string has its own copy of the prefix portion of
.meta string
which has been forced so far. The unforced list and separator
string are shared between
.meta string
and the newly constructed lazy string.
.coNP Function @ upcase-str
.synb
.mets (upcase-str << string )
.syne
.desc
The
.code upcase-str
function produces a copy of
.meta string
such that all lowercase
characters of the English alphabet are mapped to their uppercase counterparts.
.coNP Function @ downcase-str
.synb
.mets (downcase-str << string )
.syne
.desc
The
.code downcase-str
function produces a copy of
.meta string
such that
all uppercase characters of the English alphabet are mapped to their
lowercase counterparts.
.coNP Function @ string-extend
.synb
.mets (string-extend < string < tail <> [ final ])
.syne
.desc
The
.code string-extend
function destructively increases the length of
.metn string ,
which must be an ordinary dynamic string. It is an error to invoke this
function on a literal string or a lazy string.
The
.meta tail
argument can be a character, string or integer. If it is a string or
character, it specifies material which is to be added to the end of the string:
either a single character or a sequence of characters. If it is an integer, it
specifies the number of characters to be added to the string.
If
.meta tail
is an integer, the newly added characters have indeterminate contents.
The string appears to be the original one because of an internal terminating
null character remains in place, but the characters beyond the terminating zero
are indeterminate.
The optional Boolean argument
.metn final ,
defaulting to
.codn nil ,
is a hint which indicates whether this
.code string-extend
call is expected to be the last time that the function
is invoked on the given
.metn string .
If
.meta final
is true, then the
.meta string
object's underlying memory allocation is trimmed to fit the actual
string data. If the argument is false, the object may be given a larger
allocation intended to improves the performance of subsequent
.code string-extend
calls.
.coNP Function @ string-finish
.synb
.mets (string-finish << string )
.syne
.desc
The
.code string-finish
function removes excess allocation from
.meta string
that may have been produced by previous calls to
.codn string-extend .
Note: if the most recent call to string
.code string-extend
specified a true value for the
.meta final
parameter, then calling
.code string-finish
is unnecessary and does nothing.
.coNP Function @ stringp
.synb
.mets (stringp << obj )
.syne
.desc
The
.code stringp
function returns
.code t
if
.meta obj
is one of the several
kinds of strings. Otherwise it returns
.codn nil .
.coNP Function @ length-str
.synb
.mets (length-str << string )
.syne
.desc
The
.code length-str
function returns the length
.meta string
in characters. The argument must be a string.
.coNP Function @ coded-length
.synb
.mets (coded-length << string )
.syne
.desc
The
.code coded-length
function returns the number of bytes required to encode
.meta string
in UTF-8.
The argument must be a character string.
If the string contains only characters in the ASCII range U+0001 to U+007F
range, then the value returned shall be the same as that returned by the
.code length-str
function.
.coNP Function @ search-str
.synb
.mets (search-str < haystack < needle >> [ start <> [ from-end ]])
.syne
.desc
The
.code search-str
function finds an occurrence of the string
.meta needle
inside
the
.meta haystack
string and returns its position. If no such occurrence exists,
it returns
.codn nil .
If a
.meta start
argument is not specified, it defaults to zero. If it is
a nonnegative integer, it specifies the starting character position for
the search. Negative values of
.meta start
indicate positions from the end of the
string, such that
.code -1
is the last character of the string.
If the
.meta from-end
argument is specified and is not
.codn nil ,
it means
that the search is conducted right-to-left. If multiple matches are possible,
it will find the rightmost one rather than the leftmost one.
.coNP Function @ search-str-tree
.synb
.mets (search-str-tree < haystack < tree >> [ start <> [ from-end ]])
.syne
.desc
The
.code search-str-tree
function is similar to
.codn search-str ,
except that instead of
searching
.meta haystack
for the occurrence of a single needle string, it searches
for the occurrence of numerous strings at the same time. These search strings
are specified, via the
.meta tree
argument, as an arbitrarily structured tree whose
leaves are strings.
The function finds the earliest possible match, in the given search direction,
from among all of the needle strings.
If
.meta tree
is a single string, the semantics is equivalent to
.codn search-str .
.coNP Function @ match-str
.synb
.mets (match-str < bigstring < littlestring <> [ start ])
.syne
.desc
Without the
.meta start
argument, the
.code match-str
function determines whether
.meta littlestring
is a prefix of
.metn bigstring .
If the
.meta start
argument is specified, and is a nonnegative integer, then the
function tests whether
.meta littlestring
matches a prefix of that portion of
.meta bigstring
which starts at the given position.
If the
.meta start
argument is a negative integer, then
.code match-str
determines
whether
.meta littlestring
is a suffix of
.metn bigstring ,
ending on that position
of bigstring, where
.code -1
denotes the last character of
.metn bigstring ,
.code -2
the second last one and so on.
If
.meta start
is
.codn -1 ,
then this corresponds to testing whether
.meta littlestring
is a suffix of
.metn bigstring .
The
.code match-str
function returns
.code nil
if there is no match.
If a prefix match is successful,
then an integer value is returned indicating the position, inside
.metn bigstring ,
one character past the matching prefix. If the entire string is matched, then
this value corresponds to the length of
.metn bigstring .
If a suffix match is successful, the return value is the position within
.meta bigstring
where the leftmost character of
.meta littlestring
matched.
.coNP Function @ match-str-tree
.synb
.mets (match-str-tree < bigstring < tree <> [ start ])
.syne
.desc
The
.code match-str-tree
function is a generalization of
.code match-str
which matches multiple test strings against
.meta bigstring
at the same time. The value
reported is the longest match from among any of the strings.
The strings are specified as an arbitrarily shaped tree structure which has
strings at the leaves.
If
.meta tree
is a single string atom, then the function behaves exactly like
.codn match-str .
.coNP Accessor @ sub-str
.synb
.mets (sub-str < str >> [ from <> [ to ]])
.mets (set (sub-str < str >> [ from <> [ to ]]) << new-value )
.syne
.desc
The
.code sub-str
function has the same parameters and semantics as the
.code sub
function,
function, except that the first argument is operated upon
using string operations.
If a
.code sub-str
form is used as a place, it denotes a subrange of
.meta list
as if it were a storage location. The previous value of this location,
if needed, is fetched by a call to
.codn sub-str .
Storing
.meta new-value
to the place is performed by a call to
.codn replace-str .
In an update operation which accesses the prior value and stores a new value,
the arguments
.metn str ,
.metn from ,
.meta to
and
.meta new-value
are evaluated once.
The
.meta str
argument is not itself required to be a place; it is not updated
when a value is written to the
.code sub-str
storage location.
.coNP Function @ replace-str
.synb
.mets (replace-str < string < item-sequence >> [ from <> [ to ]])
.syne
.desc
The
.code replace-str
function has the same parameters and semantics as the
.code replace
function, except that the first argument is operated upon
using string operations.
.coNP Functions @, cat-str @ join-with and @ join
.synb
.mets (cat-str < item-seq <> [ sep ])
.mets (join-with < sep << item *)
.mets (join << item *)
.syne
.desc
The
.codn cat-str ,
.code join-with
and
.code join
functions combine items into a single string, which is returned.
Every
.meta item
argument must be a character or string object. The same is true of the
.meta sep
argument, if present.
The
.meta item-seq
argument must be a sequence of any mixture of characters or strings.
Note that this means that if
.meta item-seq
is a character string, it is a valid argument, since it is a sequence
of characters.
If
.meta item-seq
is empty, or no
.meta item
arguments are present, then all three functions return an
empty string.
The
.code cat-str
function receives the items as a single list. If the
.meta sep
argument is present, the items are catenated together such that
.meta sep
is interposed between them. If
.meta item-seq
contains
.I n
items, then
.I "n - 1"
copies of
.meta sep
occur in the resulting string.
If
.meta sep
is absent, then
.code cat-str
catenates the items together directly, without any separator.
Copies of the items appear in the resulting string in the same
order as the items appear in
.metn item-seq .
The
.code join-with
function receives the items as arguments rather than a single
.meta item-seq
arguments. The arguments are joined into a single character string
in order, with
.meta sep
interposed between them.
The
.code join
function takes no
.meta sep
argument. It joins all of its argument items into a single
string, in order.
.coNP Function @ split-str
.synb
.mets (split-str < string < sep <> [ keep-between ])
.syne
.desc
The
.code split-str
function breaks the
.meta string
into pieces, returning a list
thereof. The
.meta sep
argument must be one of three types: a string, a character
or a regular expression. It determines the separator character
sequences within
.metn string .
All non-overlapping matches for
.meta sep
within
.meta string
are identified in left-to-right order, and are removed from
.metn string .
The string is broken into pieces
according to the gaps left behind by the removed separators, and a list
of the remaining pieces is returned.
If
.meta sep
is the empty string, then the separator pieces removed from the
string are considered to be the empty strings between its
characters. In this case, if
.meta string
is of length one or zero, then it is considered to have no such pieces, and a
list of one element is returned containing the original string.
These remarks also apply to the situation when
.meta sep
is a regular expression which matches only an empty
substring of
.metn string .
If a match for
.meta sep
is not found in the string at all (not even an empty match), then the string is
not split at all: a list of one element is returned containing the original
string.
If
.meta sep
matches the entire string, then a list of two empty strings is
returned, except in the case that the original string is empty, in which case a
list of one element is returned, containing the empty string.
Whenever two adjacent matches for
.meta sep
occur, they are considered separate
cuts with an empty piece between them.
This operation is nondestructive:
.meta string
is not modified in any way.
If the optional
.meta keep-between
argument is specified and is not
.codn nil ,
If an argument is given and is true, then
.meta split-str
incorporates the matching separating pieces of
.meta string
into the resulting list, such that if the resulting
list is catenated, a string equivalent to the original
string will be produced.
Note: to split a string into pieces of length one such that an empty string
produces
.code nil
rather than
.codn ("") ,
use the
.mono
.meti (tok-str < string #/./)
.onom
pattern.
Note: the function call
.code "(split-str s r t)"
produces a resulting list identical to
.codn "(tok-str s r t)" ,
for all values of
.code r
and
.codn s ,
provided that
.code r
does not match empty strings. If
.code r
matches empty strings, then the
.code tok-str
call returns extra elements compared to
.codn split-str ,
because
.code tok-str
allows empty matches to take place and extract empty tokens
before the first character of the string, and after the
last character, whereas
.code split-str
does not recognize empty separators at these outer limits
of the string.
.coNP Function @ spl
.synb
.mets (spl < sep <> [ keep-between ] << string )
.syne
.desc
The
.code spl
function performs the same computation as
.codn split-str .
The same-named parameters of
.code spl
and
.code split-str
have the same semantics. The difference is the argument order.
The
.code spl
function takes the
.meta sep
argument first.
The last argument is always
.meta string
whether or not there are two arguments or three. If there are
three arguments, then
.meta keep-between
is the middle one.
Note: the argument conventions of
.code spl
facilitate less verbose partial application, such as with macros in the
.code op
family, in the common situation when
.meta string
is the unbound argument.
.coNP Functions @ split-str-set and @ sspl
.synb
.mets (split-str-set < string << set )
.mets (sspl < set << string )
.syne
.desc
The
.code split-str-set
function breaks the
.meta string
into pieces, returning a list
thereof. The
.meta set
argument must be a string. It specifies a set of
characters. All occurrences of any of these characters within
.meta string
are
identified, and are removed from
.metn string .
The string is broken into pieces
according to the gaps left behind by the removed separators.
Adjacent occurrences of characters from
.meta set
within
.meta string
are considered to
be separate gaps which come between empty strings.
This operation is nondestructive:
.meta string
is not modified in any way.
The
.code sspl
function performs the same operation; the only difference between
.code sspl
and
.code split-str-set
is argument order.
.coNP Functions @ tok-str and @ tok-where
.synb
.mets (tok-str < string < regex <> [ keep-between ])
.mets (tok-where < string << regex )
.syne
.desc
The
.code tok-str
function searches
.meta string
for tokens, which are defined as
substrings of
.meta string
which match the regular expression
.meta regex
in the
longest possible way, and do not overlap. These tokens are extracted from the
string and returned as a list.
Whenever
.meta regex
matches an empty string, then an empty token is returned, and
the search for another token within
.meta string
resumes after advancing by one
character position. However, if an empty match occurs immediately
after a nonempty token, that empty match is not turned into
a token.
So for instance,
.mono
(tok-str "abc" #/a?/)
.onom
returns
.mono
("a" "" "").
.onom
After the token
.str "a"
is extracted from a nonempty match
for the regex, an empty match for the regex occurs just
before the character
.codn b .
This match is discarded because it is an empty match which
immediately follows the nonempty match. The character
.code b
is skipped. The next match is an empty match between the
.code b
and
.code c
characters. This match causes an empty token to be
extracted. The character
.code c
is skipped, and one more empty match occurs after that
character and is extracted.
If the
.meta keep-between
argument is specified, and is not
.codn nil ,
then the behavior
of
.code tok-str
changes in the following way. The pieces of
.meta string
which are
skipped by the search for tokens are included in the output. If no token is
found in
.metn string ,
then a list of one element is returned, containing
.metn string .
Generally, if N tokens are found, then the returned list consists of 2N + 1
elements. The first element of the list is the (possibly empty) substring which
had to be skipped to find the first token. Then the token follows. The next
element is the next skipped substring and so on. The last element is the
substring of
.meta string
between the last token and the end.
The
.code tok-where
function works similarly to
.codn tok-str ,
but instead of returning
the extracted tokens themselves, it returns a list of the character position
ranges within
.meta string
where matches for
.meta regex
occur. The ranges
are pairs of numbers, represented as cons cells, where the first number
of the pair gives the starting character position, and the second number
is one position past the end of the match. If a match is empty, then the
two numbers are equal.
The
.code tok-where
function does not support the
.meta keep-between
parameter.
.coNP Function @ tok
.synb
.mets (tok < regex <> [ keep-between ] << string )
.syne
.desc
The
.code tok
function performs the same computation as
.codn tok-str .
The same-named parameters of
.code tok
and
.code tok-str
have the same semantics. The difference is the argument order.
The
.code tok
function takes the
.meta regex
argument first.
The last argument is always
.meta string
whether or not there are two arguments or three. If there are
three arguments, then
.meta keep-between
is the middle one.
Note: the argument conventions of
.code tok
facilitate less verbose partial application, such as with macros in the
.code op
family, in the common situation when
.meta string
is the unbound argument.
.coNP Function @ list-str
.synb
.mets (list-str << string )
.syne
.desc
The
.code list-str
function converts a string into a list of characters.
.coNP Function @ trim-str
.synb
.mets (trim-str << string )
.syne
.desc
The
.code trim-str
function produces a copy of
.meta string
from which leading and
trailing tabs, spaces and newlines are removed.
.coNP Functions @ string-set-code and @ string-get-code
.synb
.mets (string-set-code < string << value )
.mets (string-get-code << string )
.syne
.desc
The
.code string-set-code
and
.code string-get-code
functions provide a mechanism for associating an integer code
with a string.
Note: this mechanism is the basis for associating system error messages passed
in exceptions with the
.code errno
values of the failed system library calls which precipitated these error
exceptions.
Not all string types can have an integer code: lazy strings and literal
strings do not have this capability. The
.meta string
argument must be of type
.codn str .
The
.meta value
argument must be an integer or character. It is recommended that its
value be confined to the non-negative range of the platform's
.code int
C type. Otherwise it is unspecified whether the same value shall be
observed by
.code string-get-code
as what was stored with
.codn string-set-code .
The
.code string-set-code
function associates the integer
.meta value
with the given
.codn string ,
and returns
.codn string .
Any previously associated value is overwritten.
The
.code string-get-code
function retrieves the value most recently associated with
.metn string .
If
.meta string
has no associated value, then
.code nil
is returned.
If the
.code string-extend
is invoked on a
.meta string
then it is unspecified whether or not
.meta string
has an associated value and, if so, what value that is, except in the
following case: if
.code string-extend
is invoked with a
.meta final
argument which is true, then
.meta string
is caused not to have an associated value.
If the
.code string-finish
function is invoked on a
.metn string ,
that string is caused not to have an associated value.
.coNP Function @ chrp
.synb
.mets (chrp << obj )
.syne
.desc
Returns
.code t
if
.meta obj
is a character, otherwise nil.
.coNP Function @ chr-isalnum
.synb
.mets (chr-isalnum << char )
.syne
.desc
Returns
.code t
if
.meta char
is an alphanumeric character, otherwise nil. Alphanumeric
means one of the uppercase or lowercase letters of the English alphabet found in
ASCII, or an ASCII digit. This function is not affected by locale.
.coNP Function @ chr-isalpha
.synb
.mets (chr-isalpha << char )
.syne
.desc
Returns
.code t
if
.meta char
is an alphabetic character, otherwise
.codn nil .
Alphabetic
means one of the uppercase or lowercase letters of the English alphabet found in
ASCII. This function is not affected by locale.
.coNP Function @ chr-isascii
.synb
.mets (chr-isascii << char )
.syne
.desc
The
.code chr-isascii
function returns
.code t
if the code of character
.meta char
is in the range 0 to 127 inclusive. For characters outside of this range, it
returns
.codn nil .
.coNP Function @ chr-iscntrl
.synb
.mets (chr-iscntrl << char )
.syne
.desc
The
.code chr-iscntrl
function returns
.code t
if the character
.meta char
is a control character. For all other character, it returns
.codn nil .
A control character is one which belongs to the Unicode C0 or C1 block.
C0 consists of the characters U+0000 through U+001F, plus the
character U+007F. These are the original ASCII control characters.
Block C1 consists of U+0080 through U+009F.
.coNP Functions @ chr-isdigit and @ chr-digit
.synb
.mets (chr-isdigit << char )
.mets (chr-digit << char )
.syne
.desc
If
.meta char
is an ASCII decimal digit character,
.code chr-isdigit
returns the value
.code t
and
.code chr-digit
returns the integer value corresponding to that digit character,
a value in the range 0 to 9. Otherwise, both functions return
.codn nil .
.coNP Function @ chr-isgraph
.synb
.mets (chr-isgraph << char )
.syne
.desc
The
.code chr-isgraph
function returns
.code t
if
.meta char
is a non-space printable ASCII character.
It returns
.code nil
if it is a space or control character.
It also returns
.code nil
for non-ASCII characters: Unicode characters with a code above 127.
.coNP Function @ chr-islower
.synb
.mets (chr-islower << char )
.syne
.desc
The
.code chr-islower
function returns
.code t
if
.meta char
is an ASCII lowercase letter. Otherwise it returns
.codn nil .
.coNP Function @ chr-isprint
.synb
.mets (chr-isprint << char )
.syne
.desc
The
.code chr-isprint
function returns
.code t
if
.meta char
is an ASCII character which is not a
control character. It also returns
.code nil
for all non-ASCII characters: Unicode
characters with a code above 127.
.coNP Function @ chr-ispunct
.synb
.mets (chr-ispunct << char )
.syne
.desc
The
.code chr-ispunct
function returns
.code t
if
.meta char
is an ASCII character which is not a
control character. It also returns
.code nil
for all non-ASCII characters: Unicode characters with a code above 127.
.coNP Function @ chr-isspace
.synb
.mets (chr-isspace << char )
.syne
.desc
The
.code chr-isspace
function returns
.code t
if
.meta char
is an ASCII whitespace character: any of the
characters in the set
.codn #\espace ,
.codn #\etab ,
.codn #\elinefeed ,
.codn #\enewline ,
.codn #\ereturn ,
.code #\evtab
and
.codn #\epage .
For all other characters, it returns
.codn nil .
.coNP Function @ chr-isblank
.synb
.mets (chr-isblank << char )
.syne
.desc
The
.code chr-isblank
function returns
.code t
if
.meta char
is a space or tab: the character
.code #\espace
or
.codn #\etab .
For all other characters, it returns
.codn nil .
.coNP Function @ chr-isunisp
.synb
.mets (chr-isunisp << char )
.syne
.desc
The
.code chr-isunisp
function returns
.code t
if
.meta char
is a Unicode whitespace character. This the case for
all the characters for which
.code chr-isspace
returns
.codn t .
It also returns
.code t
for these additional characters:
.codn #\exa0 ,
.codn #\ex1680 ,
.codn #\ex180e ,
.codn #\ex2000 ,
.codn #\ex2001 ,
.codn #\ex2002 ,
.codn #\ex2003 ,
.codn #\ex2004 ,
.codn #\ex2005 ,
.codn #\ex2006 ,
.codn #\ex2007 ,
.codn #\ex2008 ,
.codn #\ex2009 ,
.codn #\ex200a ,
.codn #\ex2028 ,
.codn #\ex2029 ,
.codn #\ex205f ,
and
.codn #\ex3000 .
For all other characters, it returns
.codn nil .
.coNP Function @ chr-isupper
.synb
.mets (chr-isupper << char )
.syne
.desc
The
.code chr-isupper
function returns
.code t
if
.meta char
is an ASCII uppercase letter. Otherwise it returns
.codn nil .
.coNP Functions @ chr-isxdigit and @ chr-xdigit
.synb
.mets (chr-isxdigit << char )
.mets (chr-xdigit << char )
.syne
.desc
If
.meta char
is a hexadecimal digit character,
.code chr-isxdigit
returns the value
.code t
and
.code chr-xdigit
returns the integer value corresponding to that digit character,
a value in the range 0 to 15. Otherwise, both functions returns
.codn nil .
A hexadecimal digit is one of the ASCII
digit characters
.code 0
through
.codn 9 ,
or else one of the letters
.code A
through
.code F
or their lowercase equivalents
.code a
through
.code f
denoting the values 10 to 15.
.coNP Function @ chr-toupper
.synb
.mets (chr-toupper << char )
.syne
.desc
If character
.meta char
is a lowercase ASCII letter character, this function
returns the uppercase equivalent character. If it is some other
character, then it just returns
.metn char .
.coNP Function @ chr-tolower
.synb
.mets (chr-tolower << char )
.syne
.desc
If character
.meta char
is an uppercase ASCII letter character, this function
returns the lowercase equivalent character. If it is some other
character, then it just returns
.metn char .
.coNP Functions @ int-chr and @ chr-int
.synb
.mets (int-chr << char )
.mets (chr-int << num )
.syne
.desc
The
.meta char
argument must be a character. The
.code int-chr
function returns that
character's Unicode code point value as an integer.
The
.meta num
argument must be a fixnum integer in the range
.code 0
to
.codn #\ex10FFFF .
The
.code chr-int
function interprets
.meta num
as a Unicode code point value and returns the
corresponding character object.
Note: these functions are also known by the obsolescent names
.code num-chr
and
.codn chr-num .
.coNP Accessor @ chr-str
.synb
.mets (chr-str < str << idx )
.mets (set (chr-str < str << idx ) << new-value )
.syne
.desc
The
.code chr-str
function performs random access on string
.meta str
to retrieve
the character whose position is given by integer
.metn idx ,
which must
be within range of the string.
The index value 0 corresponds to the first (leftmost) character of the string
and so nonnegative values up to one less than the length are possible.
Negative index values are also allowed, such that -1 corresponds to the
last (rightmost) character of the string, and so negative values down to
the additive inverse of the string length are possible.
An empty string cannot be indexed. A string of length one supports index 0 and
index -1. A string of length two is indexed left to right by the values 0 and
1, and from right to left by -1 and -2.
If the element
.meta idx
of string
.meta str
exists, and the string is modifiable, then the
.code chr-str
form denotes a place.
A
.code chr-str
place
supports deletion. When a deletion takes place,
then the character at
.meta idx
is removed from the string. Any characters
after that position move by one position
to close the gap, and the length of the string
decreases by one.
.TP* Notes:
Direct use of
.code chr-str
is equivalent to the DWIM bracket notation except
that
.code str
must be a string. The following relation holds:
.verb
(chr-str s i) --> [s i]
.brev
since
.codn "[s i] <--> (ref s i)" ,
this also holds:
.verb
(chr-str s i) --> (ref s i)
.brev
However, note the following difference. When the expression
.code "[s i]"
is used as a place, then the subexpression
.code s
must be a place. When
.code "(chr-str s i)"
is used as a place,
.code s
need not be a place.
.coNP Function @ chr-str-set
.synb
.mets (chr-str-set < str < idx << char )
.syne
.desc
The
.code chr-str
function performs random access on string
.meta str
to overwrite
the character whose position is given by integer
.metn idx ,
which must
be within range of the string. The character at
.meta idx
is overwritten
with character
.metn char .
The
.meta idx
argument works exactly as in
.codn chr-str .
The
.meta str
argument must be a modifiable string.
.TP* Notes:
Direct use of
.code chr-str
is equivalent to the DWIM bracket notation provided that
.meta str
is a string and
.meta idx
an integer. The following relation holds:
.verb
(chr-str-set s i c) --> (set [s i] c)
.brev
Since
.code "(set [s i] c) <--> (refset s i c)"
for an integer index
.codn i ,
this also holds:
.verb
(chr-str s i) --> (refset s i c)
.brev
.coNP Function @ span-str
.synb
.mets (span-str < str << set )
.syne
.desc
The
.code span-str
function determines the longest prefix of string
.meta str
which
consists only of the characters in string
.metn set ,
in any combination.
.coNP Function @ compl-span-str
.synb
.mets (compl-span-str < str << set )
.syne
.desc
The
.code compl-span-str
function determines the longest prefix of string
.meta str
which
consists only of the characters which do not appear in
.metn set ,
in any combination.
.coNP Function @ break-str
.synb
.mets (break-str < str << set )
.syne
.desc
The
.code break-str
function returns an integer which represents the position of the
first character in string
.meta str
which appears in string
.metn set .
If there is no such character, then
.code nil
is returned.
.SS* Lazy Strings
Lazy strings are objects that were developed for the \*(TX pattern-matching
language, and are exposed via \*(TL. Lazy strings behave much like strings,
and can be substituted for strings. However, unlike regular strings, which
exist in their entirety, first to last character, from the moment they are
created, lazy strings do not exist all at once, but are created on demand. If
character at index N of a lazy string is accessed, then characters 0 through N
of that string are forced into existence. However, characters at indices
beyond N need not necessarily exist.
A lazy string dynamically grows by acquiring new text from a list of strings
which is attached to that lazy string object. When the lazy string is accessed
beyond the end of its hitherto materialized prefix, it takes enough strings
from the list in order to materialize the index. If the list doesn't have
enough material, then the access fails, just like an access beyond the end of a
regular string. A lazy string always takes whole strings from the attached
list.
Lazy string growth is achieved via the
.code lazy-str-force-upto
function which
forces a string to exist up to a given character position. This function is
used internally to handle various situations.
The
.code lazy-str-force
function forces the entire string to materialize. If the
string is connected to an infinite lazy list, this will exhaust all memory.
Lazy strings are specially recognized in many of the regular string functions,
which do the right thing with lazy strings. For instance when
.code sub-str
is invoked on a lazy string, a special version of the
.code sub-str
logic is
used which handles various lazy string cases, and can potentially return
another lazy string. Taking a
.code sub-str
of a lazy string from a given character position
to the end does not force the entire lazy string to exist,
and in fact the operation will work on a lazy string that is infinite.
Furthermore, special lazy string functions are provided which allow programs to
be written carefully to take better advantage of lazy strings. What carefully
means is code that avoids unnecessarily forcing the lazy string. For instance,
in many situations it is necessary to obtain the length of a string, only to
test it for equality or inequality with some number. But it is not necessary to
compute the length of a string in order to know that it is greater than some
value.
.coNP Function @ lazy-str
.synb
.mets (lazy-str < string-list >> [ terminator <> [ limit-count ]])
.syne
.desc
The
.code lazy-str
function constructs a lazy string which draws material from
.meta string-list
which is a list of strings.
If the optional
.meta terminator
argument is given, then it specifies a string
which is appended to every string from
.metn string-list ,
before that string is
incorporated into the lazy string. If
.meta terminator
is not given,
then it defaults to the string
.strn "\en" ,
and so the strings from
.meta string-list
are effectively treated as lines which get terminated by newlines
as they accumulate into the growing prefix of the lazy string.
To avoid the use of a terminator string, a null string
.meta terminator
argument
must be explicitly passed. In that case, the lazy string grows simply
by catenating elements from
.metn string-list .
If the
.meta limit-count
argument is specified, it must be a positive integer. It
expresses a maximum limit on how many elements will be consumed from
.meta string-list
in order to feed the lazy string. Once that many elements are
drawn, the string ends, even if the list has not been exhausted.
However, that remaining list, though not contributing to the string, is still
incorporated into the value returned by
.codn lazy-str-get-trailing-list .
.coNP Function @ lazy-stringp
.synb
.mets (lazy-stringp << obj )
.syne
.desc
The
.code lazy-stringp
function returns
.code t
if
.meta obj
is a lazy
string. Otherwise it returns
.codn nil .
.coNP Function @ lazy-str-force-upto
.synb
.mets (lazy-str-force-upto < lazy-str << index )
.syne
.desc
The
.code lazy-str-force-upto
function tries to instantiate the lazy string such that
the position given by
.meta index
materializes. The
.meta index
is a character
position, exactly as used in the
.code chr-str
function.
It is an error if the
.meta lazy-str
argument isn't a lazy string.
Some positions beyond
.meta index
may also materialize, as a side effect, because the operation
takes only whole strings from the internal list, according
to the algorithm described below.
If the string is already materialized through to at least
.metn index ,
or if it is
possible to materialize the string that far, then the value
.code t
is returned to indicate success.
If there is sufficient material to force the lazy string through to the
.meta index
position, then
.code t
is returned, otherwise
.codn nil .
The
.meta lazy-str
object's
.meta limit-count
is observed: a total of no more than
.meta limit-count
elements are taken from the object's list.
The algorithm is as follows:
.RS
.IP 1.
While the length of the materialized prefix of the string is less than or equal to
.meta index
and while elements are available in the list, subject to observance of the
.metn limit-count ,
perform the following steps 2 and 3:
.IP 2.
Remove the next available string from the list, and add it as a suffix to the materialized prefix.
.IP 3.
Add the
.meta terminator
string to the materialized prefix.
.IP 4.
Return
.code t
if the length of the materialized prefix exceeds
.metn index ,
otherwise
.codn nil .
.RE
.IP
The algorithm does not take portions of strings from the list, and always adds the terminator
after incorporating each piece into the materialized prefix.
.coNP Function @ lazy-str-force
.synb
.mets (lazy-str-force << lazy-str )
.syne
.desc
The
.meta lazy-str
argument must be a lazy string. The lazy string is forced
to fully materialize.
The return value is an ordinary, non-lazy string equivalent to the fully
materialized lazy string.
The
.meta lazy-str
object's
.meta limit-count
is observed: a total of no more than
.meta limit-count
elements are taken from the object's list.
The algorithm that is followed by
.code lazy-str-force
is similar to the one followed by
.codn lazy-str-force-upto ,
with only the following modification. The test in step 1 isn't concerned with
the length of the materialized prefix, since the goal is to materialize all available
characters. Steps 2 and 3 are performed while elements are available in
the list, subject to observance of the
.metn limit-count .
.coNP Function @ lazy-str-get-trailing-list
.synb
.mets (lazy-str-get-trailing-list < string << index )
.syne
.desc
The
.code lazy-str-get-trailing-list
function can be considered, in some way, an inverse operation to
the production of the lazy string from its associated list.
Note: the behavior of this function changed in \*(TX 274. This is subject
to a note in the COMPATIBILITY section.
First, the lazy string
.meta string
is forced up through the position
.metn index ,
as if by a call to
.metn lazy-str-force-upto .
If
.meta string
consists of
.meta index
or more characters, then after the forcing operation, it is guaranteed that
at least
.meta index
characters of the string have been materialized into a single string, called the
.IR "materialized prefix"
of the lazy string. If fewer than
.meta index
characters are available, taking into account the contribution of the
terminator string, then the number of characters in the materialized prefix fall short of
.metn index .
The materialized prefix never takes fractional strings from the lazy string's
list, and is always terminated by the terminator string.
Next, the materialized prefix is split into pieces on occurrences of
.metn string 's
terminator string, as if by using
.code spl
function. If the terminator string is empty, it is split into individual characters,
in accordance with the semantics of that function.
Then, if the last piece of the split prefix is an empty string, it is removed.
This situation occurs in two cases: the materialized prefix is empty, or else
it ends in the terminating string. For example, if the terminating
string is a single newline, and the prefix is
.strn "foo\en" .
In this case,
.code "(spl \(dq\en\(dq \(dqfoo\en\(dq)"
produces
.code "(\(dqfoo\(dq \(dq\(dq)"
from which the trailing empty string is removed, leaving
.codn "(\(dqfoo\(dq)" .
Finally, a list is formed by appending the split piece of the materialized prefix,
calculated as described above, with
.metn string 's
remaining list of strings which have not been pulled into the materialized
prefix. This list is returned.
.coNP Functions @, length-str-> @, length-str->= @ length-str-< and @ length-str-<=
.synb
.mets (length-str-> < string << len )
.mets (length-str->= < string << len )
.mets (length-str-< < string << len )
.mets (length-str-<= < string << len )
.syne
.desc
These functions compare the lengths of two strings. The following
equivalences hold, as far as the resulting value is concerned:
.verb
(length-str-> s l) <--> (> (length-str s) l)
(length-str->= s l) <--> (>= (length-str s) l)
(length-str-< s l) <--> (< (length-str s) l)
(length-str-<= s l) <--> (<= (length-str s) l)
.brev
The difference between the functions and the equivalent forms is that if the
string is lazy, the
.code length-str
function will fully force it in order to
calculate and return its length.
These functions only force a string up to position
.metn len ,
so they are not
only more efficient, but on infinitely long lazy strings they are usable.
.code length-str
cannot compute the length of a lazy string with an unbounded
length; it will exhaust all memory trying to force the string.
These functions can be used to test such as string whether it is longer
or shorter than a given length, without forcing the string beyond
that length.
.coNP Function @ cmp-str
.synb
.mets (cmp-str < left-string << right-string )
.syne
.desc
The
.code cmp-str
function returns -1 if
.meta left-string
is lexicographically prior to
.metn right-string .
If the reverse relationship holds, it returns 1.
Otherwise the strings are equal
and zero is returned.
If either or both of the strings are lazy, then they are only forced to the
minimum extent necessary for the function to reach a conclusion and return the
appropriate value, since there is no need to look beyond the first character
position in which they differ.
The lexicographic ordering is naive, based on the character code point
values in Unicode taken as integers, without regard for locale-specific
collation orders.
Note: in \*(TX 232 and earlier versions,
.code cmp-str
conforms to a weaker requirements: any negative integer value
may be returned rather than -1, and any positive integer value
can be returned instead of 1.
.coNP Functions @, str= @, str< @, str> @ str>= and @ str<=
.synb
.mets (str= < left-string << right-string )
.mets (str< < left-string << right-string )
.mets (str> < left-string << right-string )
.mets (str<= < left-string << right-string )
.mets (str>= < left-string << right-string )
.syne
.desc
These functions compare
.meta left-string
and
.meta right-string
lexicographically,
as if by the
.code cmp-str
function.
The
.code str=
function returns
.code t
if the two strings are exactly the same, character
for character, otherwise it returns
.codn nil .
The
.code str<
function returns
.code t
if
.meta left-string
is lexicographically before
.metn right-string ,
otherwise nil.
The
.code str>
function returns
.code t
if
.meta left-string
is lexicographically after
.metn right-string ,
otherwise
.codn nil .
The
.code str<
function returns
.code t
if
.meta left-string
is lexicographically before
.metn right-string ,
or if they are exactly the same, otherwise
.codn nil .
The
.code str<
function returns
.code t
if
.meta left-string
is lexicographically after
.metn right-string ,
or if they are exactly the same, otherwise
.codn nil .
.coNP Function @ string-lt
.synb
.mets (string-lt < left-str << right-str )
.syne
.desc
The
.code string-lt
is a deprecated alias for
.codn str< .
.SS* Vectors
.coNP Function @ vector
.synb
.mets (vector < length <> [ initval ])
.syne
.desc
The
.code vector
function creates and returns a vector object of the specified
length. The elements of the vector are initialized to
.metn initval ,
or to nil if
.meta initval
is omitted.
.coNP Function @ vec
.synb
.mets (vec << arg *)
.syne
.desc
The
.code vec
function creates a vector out of its arguments.
.coNP Function @ vectorp
.synb
.mets (vectorp << obj )
.syne
.desc
The
.code vectorp
function returns
.code t
if
.meta obj
is a vector, otherwise it returns
.codn nil .
.coNP Function @ vec-set-length
.synb
.mets (vec-set-length < vec << len )
.syne
.desc
The
.code vec-set-length
modifies the length of
.metn vec ,
making it longer or
shorter. If the vector is made longer, then the newly added elements
are initialized to nil. The
.meta len
argument must be nonnegative.
The return value is
.metn vec .
.coNP Accessor @ vecref
.synb
.mets (vecref < vec << idx )
.mets (set (vecref < vec << idx ) << new-value )
.syne
.desc
The
.code vecref
function performs indexing into a vector. It retrieves
an element of
.meta vec
at position
.metn idx ,
counted from zero.
The
.meta idx
value must range from 0 to one less than the
length of the vector. The specified element is returned.
If the element
.meta idx
of vector
.meta vec
exists, then the
.code vecref
form denotes a place.
A
.code vecref
place
supports deletion. When a deletion takes place,
then if
.meta idx
denotes the last element in the vector, the
vector's length is decreased by one, so that
the vector no longer has that element.
Otherwise, if
.meta idx
isn't the last element, then each elements
values at a higher index than
.meta idx
shifts by one one element position to the
adjacent lower index. Then, the length of the
vector is decreased by one, so that the last
element position disappears.
.coNP Function @ vec-push
.synb
.mets (vec-push < vec << elem )
.syne
.desc
The
.code vec-push
function extends the length of a vector
.meta vec
by one element, and
sets the new element to the value
.metn elem .
The previous length of the vector (which is also the position of
.metn elem )
is returned.
.coNP Function @ length-vec
.synb
.mets (length-vec << vec )
.syne
.desc
The
.code length-vec
function returns the length of vector
.metn vec .
It performs
similarly to the generic
.code length
function, except that the argument must
be a vector.
.coNP Function @ size-vec
.synb
.mets (size-vec << vec )
.syne
.desc
The
.code size-vec
function returns the number of elements for which storage
is reserved in the vector
.metn vec .
.TP* Notes:
The
.code length
of the vector can be extended up to this size without any memory
allocation operations having to be performed.
.coNP Function @ vec-list
.synb
.mets (vec-list << list )
.syne
.desc
The
.code vec-list
function returns a vector which contains all of the same elements
and in the same order as list
.metn list .
Note: this function is also known by the obsolescent name
.codn vector-list .
.coNP Function @ list-vec
.synb
.mets (list-vec << vec )
.syne
.desc
The
.code list-vec
function returns a list of the elements of vector
.metn vec .
Note: this function is also known by the obsolescent name
.codn list-vector .
.coNP Function @ copy-vec
.synb
.mets (copy-vec << vec )
.syne
.desc
The
.code copy-vec
function returns a new vector object of the same length
as
.meta vec
and containing the same elements in the same order.
.coNP Accessor @ sub-vec
.synb
.mets (sub-vec < vec >> [ from <> [ to ]])
.mets (set (sub-vec < vec >> [ from <> [ to ]]) << new-value )
.syne
.desc
The
.code sub-vec
function has the same parameters and semantics as the
function
.codn sub ,
except that the
.meta vec
argument must be a vector.
If a
.code sub-vec
form is used as a place, it denotes a subrange of
.meta list
as if it were a storage location. The previous value of this location,
if needed, is fetched by a call to
.codn sub-vec .
Storing
.meta new-value
to the place is performed by a call to
.codn replace-vec .
In an update operation which accesses the prior value and stores a new value,
the arguments
.metn vec ,
.metn from ,
.meta to
and
.meta new-value
are evaluated once.
The
.meta vec
argument is not itself required to be a place; it is not updated
when a value is written to the
.code sub-vec
storage location.
.coNP Function @ replace-vec
.synb
.mets (replace-vec < vec < item-sequence >> [ from <> [ to ]])
.syne
.desc
The
.code replace-vec
is like the
.code replace
function except that the
.meta vec
argument must be a vector.
.coNP Function @ fill-vec
.synb
.mets (fill-vec < vec < elem >> [ from <> [ to ]])
.syne
.desc
The
.code fill-vec
function overwrites a range of the vector with copies of the
.meta elem
value.
The
.meta from
and
.meta to
index arguments follow the same range indexing conventions as the
.meta replace
and
.meta sub
functions.
If
.meta from
is omitted, it defaults to zero.
If
.meta to
is omitted, it defaults to the length of
.metn vec .
Negative values of
.meta from
and
.meta to
are adjusted by adding the length of the vector to them, once.
If the adjusted value of either
.meta from
or
.meta to
is negative, or exceeds the length of
.metn vec ,
an error exception is thrown.
The adjusted values of
.meta to
and
.meta from
specify a range of vec starting at the
.meta from
index, and ending at the
.meta to
index, which is excluded from the range.
If the adjusted
.meta to
is less than or equal to the adjusted
.metn from ,
then
.meta vec
is unaltered.
Otherwise, copies of element are stored into
.meta vec
starting at the
.meta from
index, ending just before the
.meta to
index is reached.
The
.code fill-vec
function returns
.metn vec .
.TP* Examples:
.verb
(defvarl v (vec 1 2 3))
v --> #(1 2 3)
(fill-vec v 0) --> #(0 0 0)
(fill-vec v 3 1) --> #(0 3 3)
(fill-vec v 4 -1) --> #(0 3 4)
(fill-vec v 5 -3 -1) --> #(5 5 4)
.brev
.coNP Function @ cat-vec
.synb
.mets (cat-vec << vec-list )
.syne
.desc
The
.meta vec-list
argument is a list of vectors. The
.code cat-vec
function
produces a catenation of the vectors listed in
.metn vec-list .
It returns
a single large vector formed by catenating those vectors together in
order.
.SS* Buffers
.coNP The @ buf type
Object of the type
.code buf
are
.IR buffers :
vector-like objects specialized for holding binary data represented as
a sequence of 8-bit bytes. Buffers support operations specialized toward the
encoding of Lisp values into machine-oriented data types, and decoding such
data types into Lisp values.
Buffers are particularly useful in conjunction with the Foreign Function
Interface (FFI), since they can be used to prepare arbitrary data which
can be passed into and out of a function by pointer. They are also useful for
binary I/O.
.coNP Conventions Used by the @ buf-put- Functions
Buffers support a number of similar functions for converting Lisp numeric
values into common data types, which are placed into the buffer. These
functions are named starting with the
.code buf-put-
prefix, followed by an abbreviated type name.
Each of these functions takes three arguments:
.meta buf
specifies the buffer,
.meta pos
specifies the byte offset position into the buffer which receives
the low-order byte of the data transfer, and
.meta val
indicates the value.
If
.meta pos
has a value such that any portion of the data transfer would
like outside of the buffer, the buffer is automatically extended
in length to contain the data transfer. If this extension causes
any padding bytes to appear between the previous length of the
buffer and
.metn pos ,
those bytes are initialized to zero.
The argument
.meta val
giving the value to be stored must be an integer or character,
except in the case of the types
.meta float
and
.metn double (the
functions
.code buf-put-float
and
.codn buf-put-double )
for which it is required to be of type
.codn float ,
and in case of the function
.code buf-put-cptr
which expects the
.meta val
argument to be a
.code cptr
object.
The
.meta val
argument must be in range for the data type, or an exception
results.
Unless otherwise indicated, the stored datum is in the local format
used by the machine with regard to byte order and other representational
details.
.coNP Conventions Used by the @ buf-get- Functions
Buffers support a number of similar functions for extracting
common data types, and converting them into Lisp values.
These functions are named starting with the
.code buf-get-
prefix, followed by an abbreviated type name.
Each of these functions takes two arguments:
.meta buf
specifies the buffer and
.meta pos
specifies the byte offset position into the buffer which holds
the low-order byte of the datum to be extracted.
If any portion of requested datum lies outside of the boundaries
of the buffer, an error exception is thrown.
The extracted value is converted to a Lisp datum. For the
majority of these functions, the returned value is of type
integer. The
.code buf-get-float
and
.code buf-get-double
return a floating-point value.
The
.code buf-get-cptr
function returns a value of type
.codn cptr .
.coNP Function @ make-buf
.synb
.mets (make-buf < len >> [ init-val <> [ alloc-size ]])
.syne
.desc
The
.code make-buf
function creates a new buffer object which holds
.meta len
bytes. This argument may be zero.
If
.meta init-val
is present, it specifies the value with which the first
.meta len
bytes of the buffer are initialized. If omitted, it
defaults to zero.
The value of
.meta init-val
must lie in the range 0 to 255.
The
.meta alloc-size
parameter indicates how much memory to actually allocate for
the buffer.
If an argument is not given, the parameter takes on
the same value as
.metn len .
If an argument is given, its value must not be less than
.metn len .
.coNP Function @ bufp
.synb
.mets (bufp << object )
.syne
.desc
The
.code bufp
function returns
.code t
if
.meta object
is a
.codn buf ,
otherwise it returns
.codn nil .
.coNP Function @ length-buf
.synb
.mets (length-buf << buf )
.syne
.desc
The
.code length-buf
function retrieves the buffer length: how many bytes are stored
in the buffer.
Note: the generic
.code length
function is also applicable to buffers.
.coNP Function @ buf-alloc-size
.synb
.mets (buf-alloc-size << buf )
.syne
.desc
The
.code buf-alloc-size
function retrieves the allocation size of the buffer.
.coNP Function @ buf-trim
.synb
.mets (buf-trim << buf )
.syne
.desc
The
.code buf-trim
function reduces the amount of memory allocated to the buffer
to the minimum required to hold it contents, effectively
setting the allocation size to the current length.
The previous allocation size is returned.
.coNP Function @ buf-set-length
.synb
.mets (buf-set-length < buf < len <> [ init-val ])
.syne
.desc
The
.code buf-set-length
function changes the length of the buffer. If the buffer
is made longer, the newly added bytes appear at the end,
and are initialized to the value given by
.metn init-val .
If
.meta init-val
is specified, its value must be in the range 0 to 255.
It defaults to zero.
.coNP Function @ copy-buf
.synb
.mets (copy-buf << buf )
.syne
.desc
The
.code copy-buf
function returns a duplicate of
.metn buf :
an object distinct from
.meta buf
which has the same length and contents, and compares
.code equal
to
.metn buf .
.coNP Accessor @ sub-buf
.synb
.mets (sub-buf < buf >> [ from <> [ to ]])
.mets (set (sub-buf < buf >> [ from <> [ to ]]) << new-val )
.syne
.desc
The
.code sub-buf
function has the same semantics as the
.code sub
function, except that the first argument must be a buffer.
The extracted sub-range of a buffer is itself a buffer object.
If
.code sub-buf
is used as a syntactic place, the argument expressions
.metn buf ,
.metn from ,
.meta to
and
.meta new-val
are evaluated just once. The prior value, if required, is accessed by calling
.code sub-buf
and
.meta new-val
is then stored via
.codn replace-buf .
.coNP Function @ replace-buf
.synb
.mets (replace-buf < buf < item-sequence >> [ from <> [ to ]])
.syne
.desc
The
.code replace-buf
function has the same semantics as the
.code replace
function, except that the first argument must be a buffer.
The elements of
.code item-sequence
are stored into
.meta buf
as if using the
.code buf-put-u8
function and therefore must be suitable
.meta val
arguments for that function.
The of the arguments, semantics and return value given for
.code replace
apply to
.codn replace-buf .
.coNP Function @ buf-list
.synb
.mets (buf-list << list )
.syne
.desc
The
.code buf-list
function creates and returns a new buffer, whose contents are derived from the
elements of
.metn list ,
which may be any kind of sequence.
The elements of
.meta list
must be integers whose values lie in the range 0 to 255, or else
characters whose code point values lie in that range.
These values are placed into the newly created buffer, which
therefore has the same length as
.metn list .
.coNP Function @ buf-put-buf
.synb
.mets (buf-put-buf < dst-buf < pos << src-buf )
.syne
.desc
The
.code buf-put-buf
function stores a copy of buffer
.meta src-buf
into
.meta dst-buf
at the offset indicated by
.metn pos .
The source and destination memory regions may overlap.
The return value is
.metn src-buf .
Note: the effect of a
.code buf-put-buf
operation may also be performed by a suitable call to
.codn replace-buf ;
however,
.code buf-put-buf
is less general: it doesn't insert or delete by replacing
destination ranges with data of differing length,
and requires a source operand of buffer type.
.coNP Function @ buf-put-i8
.synb
.mets (buf-put-i8 < buf < pos << val )
.syne
.desc
The
.code buf-put-i8
converts
.meta val
into an 8-bit signed integer, and stores it into the buffer at
the offset indicated by
.metn pos .
The return value is
.metn val .
.coNP Function @ buf-put-u8
.synb
.mets (buf-put-u8 < buf < pos << val )
.syne
.desc
The
.code buf-put-u8
converts
.meta val
into an 8-bit unsigned integer, and stores it into the buffer at
the offset indicated by
.metn pos .
The return value is
.metn val .
.coNP Function @ buf-put-i16
.synb
.mets (buf-put-i16 < buf < pos << val )
.syne
.desc
The
.code buf-put-i16
converts
.meta val
into a sixteen bit signed integer, and stores it into the buffer at
the offset indicated by
.metn pos .
The return value is
.metn val .
.coNP Function @ buf-put-u16
.synb
.mets (buf-put-u16 < buf < pos << val )
.syne
.desc
The
.code buf-put-u16
converts
.meta val
into a sixteen bit unsigned integer, and stores it into the buffer at
the offset indicated by
.metn pos .
The return value is
.metn val .
.coNP Function @ buf-put-i32
.synb
.mets (buf-put-i32 < buf < pos << val )
.syne
.desc
The
.code buf-put-i32
converts
.meta val
into a 32-bit signed integer, and stores it into the buffer at
the offset indicated by
.metn pos .
The return value is
.metn val .
.coNP Function @ buf-put-u32
.synb
.mets (buf-put-u32 < buf < pos << val )
.syne
.desc
The
.code buf-put-u32
converts
.meta val
into a 32-bit unsigned integer, and stores it into the buffer at
the offset indicated by
.metn pos .
The return value is
.metn val .
.coNP Function @ buf-put-i64
.synb
.mets (buf-put-i64 < buf < pos << val )
.syne
.desc
The
.code buf-put-i64
converts
.meta val
into a 64-bit signed integer, and stores it into the buffer at
the offset indicated by
.metn pos .
The return value is
.metn val .
.coNP Function @ buf-put-u64
.synb
.mets (buf-put-u64 < buf < pos << val )
.syne
.desc
The
.code buf-put-u64
converts the value
.meta val
into a 64-bit unsigned integer, and stores it into the buffer at
the offset indicated by
.metn pos .
The return value is
.metn val .
.coNP Function @ buf-put-char
.synb
.mets (buf-put-char < buf < pos << val )
.syne
.desc
The
.code buf-put-char
converts
.meta val
into a value of the C type
.code char
and stores it into the buffer at
the offset indicated by
.metn pos .
The return value is
.metn val .
Note that the
.code char
type may be signed or unsigned.
.coNP Function @ buf-put-uchar
.synb
.mets (buf-put-uchar < buf < pos << val )
.syne
.desc
The
.code buf-put-uchar
converts
.meta val
into a value of the C type
.code "unsigned char"
and stores it into the buffer at
the offset indicated by
.metn pos .
.coNP Function @ buf-put-short
.synb
.mets (buf-put-short < buf < pos << val )
.syne
.desc
The
.code buf-put-short
converts
.meta val
into a value of the C type
.code short
and stores it into the buffer at
the offset indicated by
.metn pos .
.coNP Function @ buf-put-ushort
.synb
.mets (buf-put-ushort < buf < pos << val )
.syne
.desc
The
.code buf-put-ushort
converts
.meta val
into a value of the C type
.code "unsigned short"
and stores it into the buffer at
the offset indicated by
.metn pos .
.coNP Function @ buf-put-int
.synb
.mets (buf-put-int < buf < pos << val )
.syne
.desc
The
.code buf-put-int
converts
.meta val
into a value of the C type
.code int
and stores it into the buffer at
the offset indicated by
.metn pos .
.coNP Function @ buf-put-uint
.synb
.mets (buf-put-uint < buf < pos << val )
.syne
.desc
The
.code buf-put-uint
converts
.meta val
into a value of the C type
.code "unsigned int"
and stores it into the buffer at
the offset indicated by
.metn pos .
.coNP Function @ buf-put-long
.synb
.mets (buf-put-long < buf < pos << val )
.syne
.desc
The
.code buf-put-long
converts
.meta val
into a value of the C type
.code long
and stores it into the buffer at
the offset indicated by
.metn pos .
.coNP Function @ buf-put-ulong
.synb
.mets (buf-put-ulong < buf < pos << val )
.syne
.desc
The
.code buf-put-ulong
converts
.meta val
into a value of the C type
.code "unsigned long"
and stores it into the buffer at
the offset indicated by
.metn pos .
.coNP Function @ buf-put-float
.synb
.mets (buf-put-float < buf < pos << val )
.syne
.desc
The
.code buf-put-float
converts
.meta val
into a value of the C type
.code float
and stores it into the buffer at
the offset indicated by
.metn pos .
Note: the conversion of a \*(TL floating-point value
to the C type float may be inexact, reducing the
numeric precision.
.coNP Function @ buf-put-double
.synb
.mets (buf-put-double < buf < pos << val )
.syne
.desc
The
.code buf-put-double
converts
.meta val
into a value of the C type
.code double
and stores it into the buffer at
the offset indicated by
.metn pos .
.coNP Function @ buf-put-cptr
.synb
.mets (buf-put-cptr < buf < pos << val )
.syne
.desc
The
.code buf-put-cptr
expects
.meta val
to be of type
.codn cptr .
It stores the object's pointer value into the buffer at
the offset indicated by
.metn pos .
.coNP Function @ buf-get-i8
.synb
.mets (buf-get-i8 < buf << pos )
.syne
.desc
The
.code buf-get-i8
function extracts and returns signed 8-bit integer from
.meta buf
at the offset given by
.metn pos .
.coNP Function @ buf-get-u8
.synb
.mets (buf-get-u8 < buf << pos )
.syne
.desc
The
.code buf-get-u8
function extracts and returns an unsigned 8-bit integer from
.meta buf
at the offset given by
.metn pos .
.coNP Function @ buf-get-i16
.synb
.mets (buf-get-i16 < buf << pos )
.syne
.desc
The
.code buf-get-i16
function extracts and returns a signed 16-bit integer from
.meta buf
at the offset given by
.metn pos .
.coNP Function @ buf-get-u16
.synb
.mets (buf-get-u16 < buf << pos )
.syne
.desc
The
.code buf-get-u16
function extracts and returns an unsigned 16-bit integer from
.meta buf
at the offset given by
.metn pos .
.coNP Function @ buf-get-i32
.synb
.mets (buf-get-i32 < buf << pos )
.syne
.desc
The
.code buf-get-i32
function extracts and returns a signed 32-bit integer from
.meta buf
at the offset given by
.metn pos .
.coNP Function @ buf-get-u32
.synb
.mets (buf-get-u32 < buf << pos )
.syne
.desc
The
.code buf-get-u32
function extracts and returns an unsigned 32-bit integer from
.meta buf
at the offset given by
.metn pos .
.coNP Function @ buf-get-i64
.synb
.mets (buf-get-i64 < buf << pos )
.syne
.desc
The
.code buf-get-i64
function extracts and returns a signed 64-bit integer from
.meta buf
at the offset given by
.metn pos .
.coNP Function @ buf-get-u64
.synb
.mets (buf-get-u64 < buf << pos )
.syne
.desc
The
.code buf-get-u64
function extracts and returns an unsigned 64-bit integer from
.meta buf
at the offset given by
.metn pos .
.coNP Function @ buf-get-char
.synb
.mets (buf-get-char < buf << pos )
.syne
.desc
The
.code buf-get-char
function extracts and returns a value of the C type
.code char
from
.meta buf
at the offset given by
.metn pos .
Note that
.code char
may be signed or unsigned.
.coNP Function @ buf-get-uchar
.synb
.mets (buf-get-uchar < buf << pos )
.syne
.desc
The
.code buf-get-uchar
function extracts and returns a value of the C type
.code "unsigned char"
from
.meta buf
at the offset given by
.metn pos .
.coNP Function @ buf-get-short
.synb
.mets (buf-get-short < buf << pos )
.syne
.desc
The
.code buf-get-short
function extracts and returns a value of the C type
.code short
from
.meta buf
at the offset given by
.metn pos .
.coNP Function @ buf-get-ushort
.synb
.mets (buf-get-ushort < buf << pos )
.syne
.desc
The
.code buf-get-ushort
function extracts and returns a value of the C type
.code "unsigned short"
from
.meta buf
at the offset given by
.metn pos .
.coNP Function @ buf-get-int
.synb
.mets (buf-get-int < buf << pos )
.syne
.desc
The
.code buf-get-int
function extracts and returns a value of the C type
.code int
from
.meta buf
at the offset given by
.metn pos .
.coNP Function @ buf-get-uint
.synb
.mets (buf-get-uint < buf << pos )
.syne
.desc
The
.code buf-get-uint
function extracts and returns a value of the C type
.code "unsigned int"
from
.meta buf
at the offset given by
.metn pos .
.coNP Function @ buf-get-long
.synb
.mets (buf-get-long < buf << pos )
.syne
.desc
The
.code buf-get-long
function extracts and returns a value of the C type
.code long
from
.meta buf
at the offset given by
.metn pos .
.coNP Function @ buf-get-ulong
.synb
.mets (buf-get-ulong < buf << pos )
.syne
.desc
The
.code buf-get-ulong
function extracts and returns a value of the C type
.code "unsigned long"
from
.meta buf
at the offset given by
.metn pos .
.coNP Function @ buf-get-float
.synb
.mets (buf-get-float < buf << pos )
.syne
.desc
The
.code buf-get-float
function extracts and returns a value of the C type
.code float
from
.meta buf
at the offset given by
.metn pos ,
returning that value as a Lisp floating-point number.
.coNP Function @ buf-get-double
.synb
.mets (buf-get-double < buf << pos )
.syne
.desc
The
.code buf-get-double
function extracts and returns a value of the C type
.code double
from
.meta buf
at the offset given by
.metn pos ,
returning that value as a Lisp floating-point number.
.coNP Function @ buf-get-cptr
.synb
.mets (buf-get-cptr < buf << pos )
.syne
.desc
The
.code buf-get-cptr
function extracts a C pointer from
.meta buf
at the offset given by
.metn pos ,
returning that value as a Lisp object of type
.codn cnum .
.coNP Function @ put-buf
.synb
.mets (put-buf < buf >> [ pos <> [ stream ]])
.syne
.desc
The
.code put-buf
function writes the contents of buffer
.metn buf ,
starting at position
.meta pos
to a stream, through to the last byte, if possible.
Successive bytes from the buffer are written to the stream as if by a
.code put-byte
operation.
If
.meta stream
is omitted, it defaults to
.codn *stdout* .
If
.meta pos
is omitted, it defaults to zero.
It indicates the starting position within the buffer.
The stream must support the
.code put-byte
operation. Streams which support
.code put-byte
can be expected to support
.code put-buf
and, conversely, streams which do not support
.code put-byte
do not support
.codn put-buf .
The
.code put-buf
function returns the position of the last byte that was successfully written.
If the buffer was written through to the end, then this value corresponds to
the length of the buffer.
If an error occurs before any bytes are written, the function
throws an error.
.coNP Functions @ fill-buf and @ fill-buf-adjust
.synb
.mets (fill-buf < buf >> [ pos <> [ stream ]])
.mets (fill-buf-adjust < buf >> [ pos <> [ stream ]])
.syne
.desc
The
.code fill-buf
reads bytes from
.meta stream
and writes them into consecutive locations in buffer
.meta buf
starting at position
.metn pos .
The bytes are read as if using the
.code get-byte
function.
If the
.meta stream
argument is omitted, it defaults to
.codn *stdin* .
If
.meta pos
is omitted, it defaults to zero.
It indicates the starting position within the buffer.
The stream must support the
.code get-byte
operation. Buffers which support
.code get-byte
can be expected to support
.code fill-buf
and, conversely, streams which do not support
.code get-byte
do not support
.codn fill-buf .
The
.code fill-buf
function returns the position that is one byte past the last byte that
was successfully read.
If an end-of-file or other error condition occurs before the buffer is filled
through to the end, then the value returned is smaller than the buffer length.
In this case, the area of the buffer beyond the read size retains its previous
content.
If an error situation occurs other than a premature end-of-file before
any bytes are read, then an exception is thrown.
If an end-of-file condition occurs before any bytes are read, then zero
is returned.
The
.code fill-buf-adjust
differs usefully from
.code fill-buf
as follows. Whereas
.code fill-buf
doesn't manipulate the length of the buffer at any stage of the operation, the
.code fill-buf-adjust
begins by adjusting the length of the buffer to the underlying allocated size.
Then it performs the fill operation in
exactly the same manner as
.codn fill-buf .
Finally, if the operation succeeds, then
.code fill-buf-adjust
adjusts the length of the buffer to match the position that is returned.
.coNP Function @ get-line-as-buf
.synb
.mets (get-line-as-buf <> [ stream ])
.syne
.desc
The
.code get-line-as-buf
reads bytes from
.meta stream
as if using the
.code get-byte
function, until either a the newline character is encountered, or else the end
of input is encountered. The bytes which are read, exclusive of the newline
character, are returned in a new buffer object. The newline character, if it
occurs, is consumed.
If
.meta stream
is omitted, it defaults to
.codn *stdin* .
The stream is required to support byte input.
.coNP Functions @ file-get-buf and @ command-get-buf
.synb
.mets (file-get-buf < name >> [ max-bytes <> [ skip-bytes ]])
.mets (command-get-buf < cmd >> [ max-bytes <> [ skip-bytes ]])
.syne
.desc
The
.code file-get-buf
function opens a binary stream over the file indicated by the string argument
.meta name
for reading. By default, the entire file is read and its contents are returned as a
buffer object. The buffer's length corresponds to the number of bytes
read from the file.
The
.code command-get-buf
function opens a binary stream over an input command pipe created for
the command string
.metn cmd ,
as if by the
.code open-command
function. It read bytes from the pipe until the indication that no more
input is available. The bytes are returned aggregated into a buffer object.
If the
.meta max-bytes
parameter is given an argument, it must be a nonnegative integer.
That value specifies a limit on the number of bytes to read. A buffer
no longer than
.meta max-bytes
shall be returned.
If the
.meta skip-bytes
parameter is given an argument, it must be a nonnegative integer.
That value specifies how many initial bytes of the input should be
discarded before accumulation of the buffer begins.
If possible, the semantics of this parameter is achieved by performing a
.code seek-stream
operation, falling back on reading and discarding bytes if the
stream doesn't support seeking.
If
.meta max-bytes
is specified, then the stream is opened in unbuffered mode, so that bytes
beyond the specified range shall not be requested from the underlying file,
device or process.
.coNP Functions @, file-put-buf @ file-append-buf and @ command-put-buf
.synb
.mets (file-put-buf < name < buf << skip-bytes )
.mets (file-place-buf < name < buf << skip-bytes )
.mets (file-append-buf < name << buf )
.mets (command-put-buf < cmd << buf )
.syne
.desc
The
.code file-put-buf
function opens a text stream over the file indicated by the string argument
.metn name ,
writes the contents of the buffer object
.meta buf
into the file, and then closes the file. If the file doesn't exist, it is
created. If it exists, it is truncated to zero length and overwritten.
The default value of the optional
.meta skip-bytes
parameter is zero. If an argument is given, it must be a nonnegative integer.
If it is nonzero, then after opening the file, before writing the buffer,
the function will seek to an offset of that many bytes from the start of the
file. The contents of
.meta buf
will be written at that offset.
The
.code file-place-buf
function does not truncate an existing file to zero length.
In all other regards, it is equivalent to
.codn file-put-buf .
The
.code file-append-buf
function is similar to
.code file-put-buf
except that if the file exists, it isn't overwritten. Rather, the buffer
is appended to the file.
The
.code command-put-buf
function opens an output text stream over an output command pipe created
for the command specified in the string argument
.metn cmd ,
as if by the
.code open-command
function.
It then writes the contents of buffer
.meta buf
into the stream and closes the stream.
The return value of all three functions is that of the
.code put-buf
operation which is implicitly performed.
.coNP Functions @ buf-str and @ str-buf
.synb
.mets (buf-str < str <> [ null-term-p ])
.mets (str-buf < buf <> [ null-term-p ])
.syne
.desc
The
.code buf-str
and
.code str-buf
functions perform UTF-8 conversion between the character string and buffer
data types.
The
.code buf-str
function UTF-8-encodes
.meta str
and returns a buffer containing the converted representation.
If a true argument is given to the
.meta null-term-p
parameter, then a null terminating byte is added to the buffer.
This byte is added even if the previous byte is already a null byte
from the conversion of a pseudo-null character occurring in
.metn str .
The
.code str-buf
function takes the contents of buffer
.meta buf
to be UTF-8 data, which is converted to a character string and returned.
Null bytes in the buffer are mapped to the pseudo-null character
.codn #\exDC00 .
If a true argument is given to the
.meta null-term-p
parameter, then if the contents of
.meta buf
end in a null byte, that byte is not included in the conversion.
.coNP Functions @ buf-int and @ buf-uint
.synb
.mets (buf-int << integer )
.mets (buf-uint << integer )
.syne
.desc
The
.code buf-int
and
.code buf-uint
functions convert a signed and unsigned integer, respectively, or else a
character, into a binary representation, which is returned as a buffer object.
Under both functions, the representation uses big endian byte order: most
significant byte first.
The
.code buf-uint
function requires a nonnegative
.meta integer
argument, which may be a character. The representation stored in the
buffer is a pure binary representation of the value using the smallest
number of bytes required for the given
.meta integer
value.
The
.code buf-int
function requires an integer or character argument. The representation
stored in the buffer is a two's complement representation of
.meta integer
using the smallest number of bytes which can represent that value.
If
.meta integer
is nonnegative, then the first byte of the buffer lies in the range
0 to 127.
If
.meta integer
is negative, then the first byte of the buffer lies in the range 128 to 255.
The integer 255 therefore doesn't convert to the buffer
.code #b'ff'
but rather
.codn #b'00ff' .
The buffer
.code #b'ff'
represents -1.
If the
.meta integer
argument is a character object, it is taken to be its Unicode code
point value, as returned by the
.code int-chr
function.
.coNP Functions @ int-buf and @ uint-buf
.synb
.mets (int-buf << buf )
.mets (uint-buf << buf )
.syne
.desc
The
.code int-buf
and
.code uint-buf
functions recover an integer value from its binary form which appears
inside
.metn buf ,
which must be a buffer object. These functions expect
.meta buf
to contain the representation produced by, respectively, the functions
.code buf-int
and
.codn buf-uint .
If
.meta buf
holds the representation of an integer value
.metn n ,
as produced by
.mono
.meti (buf-int << n )
.onom
then
.mono
.meti (int-buf << buf )
.onom
returns
.metn n .
The same relationship holds between
.code buf-uint
and
.codn uint-buf .
Thus, these equalities hold:
.verb
.mets (= (int-buf (buf-int << n )) << n )
.mets (= (uint-buf (buf-uint << n )) << n )
.brev
provided that
.meta n
is of integer type and, in the case of
.codn buf-uint ,
nonnegative.
.SS* Structures
\*(TX supports user-defined types in the form of structures. Structures
are objects which hold multiple storage locations called slots, which are named
by symbols. Structures can be related to each other by inheritance. Multiple
inheritance is permitted.
The type of a structure is itself an object, of type
.codn struct-type .
When the program defines a new structure type, it does so by creating a new
.code struct-type
instance, with properties which describe the new structure type: its
name, its list of slots, its initialization and "boa constructor" functions,
and the structures type it inherits from (the
.IR supertypes ).
The
.code struct-type
object is then used to generate instances.
Structures instances are not only containers which hold named slots, but they
also indicate their struct type. Two structures which have the same number of
slots having the same names are not necessarily of the same type.
Structure types and structures may be created and manipulated using
a programming interface based on functions.
For more convenient and clutter-free expression of structure-based
program code, macros are also provided.
Furthermore, concise and expressive slot access syntax is provided courtesy of
the referencing dot and unbound referencing dot syntax, a syntactic sugar
for the
.code qref
and
.code uref
macros.
Structure types have a name, which is a symbol. The
.code typeof
function, when applied to any struct type, returns the symbol
.codn struct-type .
When
.code typeof
is applied to a struct instance, it returns the name of
the struct type. Effectively, struct names are types.
The consequences are unspecified if an existing struct name is reused for a
different struct type, or an existing type name is used for a struct type.
.NP* Static Slots
Structure slots can be of two kinds: they can be the ordinary instance slots or
they can be static slots. The instances of a given structure type have their
own instance of a given instance slot. However, they all share a single
instance of a static slot.
Static slots are allocated in a global area associated with a structure type
and are initialized when the structure type is created. They are useful for
efficiently representing properties which have the same value for all instances
of a struct. These properties don't have to occupy space in each instance, and
time doesn't have to be wasted initializing them each time a new instance is
created. Static slots are also useful for struct-specific global variables.
Lastly, static slots are also useful for holding methods and functions.
Although structures can have methods and functions in their instances, usually,
all structures of the same type share the same functions. The
.code defstruct
macro supports a special syntax for defining methods and struct-specific
functions at the same time when a new structure type is defined.
The
.code defmeth
macro can be used for adding new methods and functions to an existing
structure and its descendants.
Static slots may be assigned just like instance slots. Changing a static
slot changes that slot in every structure of the same type.
Static slots are not listed in the
.code #S(...)
notation when a structure is printed. When the structure notation is
read from a stream, if static slots are present, they will be processed
and their values stored in the static locations they represent, thus
changing their values for all instances.
Static slots are inherited just like instance slots. The following
simplified discussion is restricted to single inheritance. A detailed
description of multiple inheritance is given in the Multiple Inheritance
section below. If a given structure
.meta B
has some static slot
.metn s ,
and a new structure
.meta D
is derived from
.metn B ,
using
.codn defstruct ,
and does not define a slot
.metn s ,
then
.meta D
inherits
.metn s .
This means that
.meta D
shares the static slot with
.metn B :
both types share a single instance of that slot.
On the other hand if
.code D
defines a static slot
.meta s
then that slot will have its own instance in the
.meta D
structure type;
.meta D
will not inherit the
.meta B
instance of slot
.metn s .
Moreover, if the definition of
.code D
omits the
.meta init-form
for slot
.metn s ,
then that slot will be initialized with a copy of the current value of slot
.meta s
of the
.meta B
base type, which allows derived types to obtain the value of base type's
static slot, yet have that in their own instance.
The slot type can be overridden. A structure type deriving from another
type can introduce slots which have the same names as the supertype,
but are of a different kind: an instance slot in the supertype
can be replaced by a static slot in the derived type or vice versa.
Note that, in light of the above type overriding possibility, the static slot
value propagation happens only from the immediate supertype.
If
.code D
is derived from
.code G
which has a static slot
.codn s ,
whereas
.code D
specifies
.code s
as an instance slot, but then
.code B
again specifies a static slot
.codn s ,
then
.codn B 's
slot
.code s
will not inherit the value from
.codn G 's
.code s
slot.
Simply,
.codn B 's
supertype is
.code D
and that supertype is not considered to have a static slot
.codn s .
A structure type is associated with a static initialization function
which may be used to store initial values into static slots. This function
is invoked once in a type's life time, when the type is created.
The function is also inherited by derived struct types and invoked when
they are created.
.NP* Multiple Inheritance
When a structure type is defined, two or more supertypes may be specified. The
new structure type then potentially inherits instance and static slots from all
of the specified supertypes, and is considered to be a subtype of all of them.
This situation with two or more supertypes is called
.IR "multiple inheritance" .
The contrasting term is
.IR "single inheritance" ,
denoting the situation when a structure has exactly one supertype.
\*(TL's struct types initially permitted only single inheritance.
Multiple inheritance support was introduced in version 229, as a
straightforward extension of single inheritance semantics.
In the
.code make-struct-type
function and
.code defstruct
macro, a list of supertypes can be given instead of just one.
The type then inherits slots from all of the specified types.
If any conflicts arise among the supertypes due to slots having the same name,
the leftmost supertype dominates: that type's slot will be inherited.
If the leftmost slot is static, then that static slot will be inherited.
Otherwise, the instance slot will be inherited.
Of course, any slot which is specified in the newly defined type itself
dominates over any same-named slots among the supertypes.
The new structure type inherits all of the slot initializing expressions, as
well as
.code :init
and
.code :postinit
methods of all of its supertypes.
Each time the structure is instantiated, the
.code :init
initializing expressions inherited from the supertypes, together with the slot
initializing expressions, are all evaluated, in right-to-left order:
the initializations contributed by each supertype are performed before
considering the next supertype to the left.
The
.code :postinit
methods are similarly invoked in right-to-left order, before the
.code :postinit
methods of the new type itself.
Thus the order is: supertype inits, own inits, supertype post-inits,
own post-inits.
.NP* Duplicate Supertypes
Multiple inheritance makes it possible for a type to inherit the
same supertype more than once, either directly (by naming it more than
once as a direct supertype) or indirectly (by inheriting two or
more different types, which have a common ancestor).
The latter situation is sometimes referred to as the
.IR "diamond problem" .
Until \*(TX 242, the situation of duplicate supertypes was
ignored for the purposes of object initialization. It was documented that if a
supertype is referenced by inheritance, directly or indirectly, two or more
times, then its initializing expressions are evaluated that many times.
Starting in \*(TX 243, duplicate supertypes no longer give rise to duplicate
initialization. When an object is instantiated, only one initialization of a
duplicated supertype occurs. The subsequent initializations that would take
place in the absence of duplicate detection are suppressed.
Note also that the
.code :fini
mechanism is tied to initialization. Initialization of an object
registers the finalizers, and so in \*(TX 242,
.code :fini
finalizers are also executed multiple times, if
.code :init
initializers are.
.TP* Examples:
Consider following program:
.verb
(defstruct base ()
(:init (me) (put-line "base init"))
(:fini (me) (put-line "base fini")))
(defstruct d1 (base)
(:init (me) (put-line "d1 init"))
(:fini (me) (put-line "d1 fini")))
(defstruct d2 (base)
(:init (me) (put-line "d2 init"))
(:fini (me) (put-line "d2 fini")))
(defstruct s (d1 d2))
(call-finalizers (new s))
.brev
Under \*(TX 242, and earlier versions that support multiple inheritance, it
produces the output:
.verb
base init
d2 init
base init
d1 init
d1 fini
base fini
d2 fini
base fini
.brev
The supertypes are initialized in a right-to-left traversal of the
type lattice, without regard for
.code base
being duplicated.
Starting with \*(TX 243, the output is:
.verb
base init
d2 init
d1 init
d1 fini
d2 fini
base fini
.brev
The rightmost duplicate of the base is initialized, so that the initialization
is complete prior to the initializations of any dependent types.
Likewise, the same rightmost duplicate of the base is finalized, so that
finalization takes place after that of any dependent struct types.
Note, however, that the
.code derived
function function mechanism is not required to detect duplicated direct
supertypes.
If a supertype implements the
.code derived
function to detect situations when it is the target of inheritance,
and some subtype inherits that type more than once, that function
may be called more than once. The behavior is unspecified.
.NP* Dirty Flags
All structure instances contain a Boolean flag called the
.IR "dirty flag" .
This flag is not a slot, but rather a meta-data property that is exposed
to program access. When the flag is set, an object is said to be dirty;
otherwise it is clean.
Newly constructed objects come into existence dirty. The dirty flag
state can be tested with the function
.codn test-dirty .
An object can be marked as clean by clearing its dirty flag with
.codn clear-dirty .
A combined operation
.code test-clear-dirty
is provided which clears the dirty flag, and
returns its previous value.
The dirty flag is set whenever a new value is stored into the instance
slot of an object.
Note: the dirty flag can be used to support support the caching of values
derived from an object's slots. The derived values don't have to be
recomputed while an object remains clean.
.NP* Equality Substitution
In object-based or object-oriented programming, sometimes it is necessary for a
new data type to provide its own notion of equality: its own requirements for
when two distinct instances of the type are considered equal. Furthermore,
types sometimes have to implement their own notion, also, of inequality: the
requirements for the manner in which one instance is considered lesser or
greater than another.
\*(TL structures implement a concept called
.IR "equality substitution"
which provides a simple, unified way for the implementor of an object to
encode the requirements for both equality and inequality.
Equality substitution allows for objects to be used as keys in a hash
table according to the custom equality, without the programmer being
burdened with the responsibility of developing a custom hashing function.
An object participates in equality substitution by implementing the
.code equal
method. The
.code equal
method takes no arguments other than the object itself. It returns
a representative value which is used in place of that object
for the purposes of
.code equal
comparison.
Whenever an object which supports equality substitution is used as an argument
of any of the functions
.codn equal ,
.codn nequal ,
.codn greater ,
.codn less ,
.codn gequal ,
.code lequal
or
.codn hash-equal ,
the
.code equal
method of that object is invoked, and the return value of that method
is taken in place of that object.
The same is true if an object which supports equality substitution is used as a
key in an
.code :equal-based
hash table.
The substitution is applied repeatedly: if the return value of the object's
.code equal
method is an object which itself supports equality substitution,
than that returned object's method is invoked on that object
to fetch its equality substitute. This repeats as many times as necessary
until an object is determined which isn't a structure that supports
equality substitution.
Once the equality substitute is determined, then the given function proceeds
with the replacement object. Thus for example
.code equal
compares the replacement object in place of the original, and an
.code :equal-based
hash table uses the replacement object as the key for the purposes of
hashing and comparison.
.NP* Custom Slot Expansion
The
.code defstruct
macro has a provision for for application-defined clauses, which may
be defined using the
.code define-struct-clause
macro. This macro associates new clause keywords with custom expansion.
The
.code :delegate
clause of
.code defstruct
is in fact implemented externally to
.code defstruct
using
.codn define-struct-clause .
.coNP Macro @ defstruct
.synb
.mets (defstruct >> { name | >> ( name << arg *)} < super
.mets \ \ << slot-specifier *)
.syne
.desc
The
.code defstruct
macro defines a new structure type and registers it under
.metn name ,
which must be a bindable symbol, according to the
.code bindable
function. Likewise, the name of every
.meta slot
must also be a bindable symbol.
The
.meta super
argument must either be
.codn nil ,
or a symbol which names an existing struct type,
or else a list of such symbols.
The newly defined struct type will inherit all slots,
as well as initialization behaviors from the specified
struct types.
The
.code defstruct
macro is implemented using the
.code make-struct-type
function, which is more general. The macro analyzes the
.code defstruct
argument syntax, and synthesizes arguments which are then
used to call the function. Some remarks in the description of
.code defstruct
only apply to structure types defined using that macro.
Slots are specified using zero or more
.IR "slot specifiers" .
Slot specifiers come in the following variety:
.RS
.meIP < name
The simplest slot specifier is just a name, which must be a bindable
symbol, as defined by the
.code bindable
function. This form is a short form for the
.mono
.meti (:instance << name )
.onom
syntax.
.meIP >> ( name << init-form )
This syntax is a short form for the
.mono
.meti (:instance < name << init-form )
.onom
syntax.
.meIP (:instance < name <> [ init-form ])
This syntax specifies an instance slot called
.meta name
whose initial value is obtained by evaluating
.meta init-form
whenever a new instance of the structure is created.
This evaluation takes place in the original lexical environment in which the
.code defstruct
form occurs. If
.meta init-form
is omitted, the slot is initialized to
.codn nil .
.meIP (:static < name <> [ init-form ])
This syntax specifies a static slot called
.meta name
whose initial value is obtained by evaluating
.meta init-form
once, during the evaluation of the
.code defstruct
form in which it occurs, if the
.meta init-form
is present. If
.meta init-form
is absent, and a static slot with the same name
exists in the
.meta super
base type, then this slot is initialized
with the value of that slot.
Otherwise it is initialized to
.codn nil .
The definition of a static slot in a
.code defstruct
causes the new type to have its own instance
that slot, even if a same-named static
slot occurs in the
.meta super
base type, or its bases.
.meIP (:method < name <> ( param +) << body-form *)
This syntax creates a static slot called
.meta name
which is initialized with an anonymous function.
The anonymous function is created during the
evaluation of the
.code defstruct
form. The function takes the arguments specified
by the
.meta param
symbols, and its body consists of the
.metn body-form s.
There must be at least one
.metn param .
When the function is invoked as a method, as intended,
the leftmost
.meta param
receives the structure instance.
The
.metn body-form s
are evaluated in a context in which a block named
.meta name
is visible. Consequently,
.code return-from
may be used to terminate the execution of a method
and return a value.
Methods are invoked
using the
.code "instance.(name arg ...)"
syntax, which implicitly inserts the instance into the argument list.
Due to the semantics of static slots, methods are naturally
inherited from a base structure to a derived one,
and defining a method in a derived class which also exists
in a base class performs OOP-style overriding.
.meIP (:function < name <> ( param *) << body-form *)
This syntax creates a static slot called
.meta name
which is initialized with an anonymous function.
The anonymous function is created during the
evaluation of the
.code defstruct
form. The function takes the arguments specified
by the
.meta param
symbols, and its body consists of the
.metn body-form s.
This specifier differs from
.code :method
only in one respect: there may be zero
parameters. A structure function defined this way is
intended to be used as a utility function which doesn't
receive the structure instance as an argument.
The
.metn body-form s
are evaluated in a context in which a block named
.meta name
is visible. Consequently,
.code return-from
may be used to terminate the execution of the function
and return a value.
Such functions are called using the
.code "(call instance.name arg ...)"
or else the DWIM brackets syntax
.codn "[instance.name arg ...]" .
The remarks about inheritance and overriding
in the description of
.code :method
also apply to
.codn :function .
.meIP (:init <> ( param ) << body-form *)
The
.code :init
specifier doesn't describe a slot. Rather, it specifies code
which is executed when a structure is instantiated, before
the slot initializations specific to the structure type
are performed. The code consists of
.metn body-form s
which are evaluated in order in a lexical scope in
which the variable
.meta param
is bound to the structure object.
The
.code :init
specifier may not appear more than once in a given
.code defstruct
form.
When an object with one or more levels of inheritance
is instantiated, the
.code :init
code of a base structure type, if any, is executed
before any initializations specific to a derived
structure type. Under multiple inheritance, the
.code :init
code of the rightmost base type is executed first,
then that of the remaining bases in right-to-left
order.
The
.code :init
initializations are executed before any other
slot initializations. The argument values passed to the
.code new
or
.code lnew
operator or the
.code make-struct
function are not yet stored in the object's slots,
and are not accessible. Initialization code which needs
these values to be stable can be defined with
.codn :postinit .
Initializers in base structures must be careful about assumptions about slot
kinds, because derived structures can alter static slots to instance slots or
vice versa.
To avoid an unwanted initialization being applied to the
wrong kind of slot, initialization code can be made conditional on the
outcome of
.code static-slot-p
applied to the slot.
(Code generated by
.code defstruct
for initializing instance slots performs this kind of check).
The
.metn body-form s
of an
.code :init
specifier are not surrounded by an implicit
.codn block .
.meIP (:postinit <> ( param ) << body-form *)
The
.code :postinit
specifier is similar to
.codn :init .
Both specify forms which are evaluated during object instantiation.
The difference is that the
.codn body-form s
of a
.code :postinit
are evaluated after other initializations have taken
place, including the
.code :init
initializations, as a second pass. By the time
.code :postinit
initialization runs, the argument material from the
.codn make-struct ,
.code new
or
.code lnew
invocation has already been processed and stored
into slots.
Like
.code :init
actions,
.code :postinit
actions registered at different levels of the type's
inheritance hierarchy are invoked in the base-to-derived
order, and in right-to-left order among multiple bases
at the same level.
.meIP (:fini <> ( param ) << body-form *)
The
.code :fini
specifier doesn't describe a slot. Rather, it specifies
a finalization function which is associated with the
structure instance, as if by use of the
.code finalize
function. This finalization registration takes place
as the first step when an instance of the structure
is created, before the slots are initialized and
the
.code :init
code, if any, has been executed. The registration
takes place as if by the evaluation of the form
.mono
.meti (finalize < obj (lambda <> ( param ) << body-form ...) t)
.onom
where
.meta obj
denotes the structure instance. Note the
.code t
argument which requests reverse order of registration, ensuring that if an
object has multiple finalizers registered at different levels of inheritance
hierarchy, the finalizers specified for a derived structure type are called
before inherited finalizers.
The
.metn body-form s
of a
.code :fini
specifier are not surrounded by an implicit
.codn block .
Note that an object's finalizers can be called explicitly with
.codn call-finalizers .
.RE
.IP
The
.code with-objects
macro arranges for finalizers to be called on objects when the execution
of a scope terminates by any means.
The slot names given in a
.code defstruct
must all be unique among themselves, but they
may match the names of existing slots in the
.meta super
base type.
A given structure type can have only one slot under a given
symbolic name. If a newly specified slot matches the name of an existing slot
in the
.meta super
type or that type's chain of ancestors, it is called a
.IR "repeated slot" .
The kind of the repeated slot (static or instance) is not inherited; it
is established by the
.code defstruct
and may be different from the type of the same-named slot in the
supertype or its ancestors.
If a repeated slot is introduced as a static slot, and
has no
.meta init-form
then it receives the current of the a static of the same name from
the nearest supertype which has such a slot.
If a repeated slot is an instance slot, no such inheritance of value
takes place; only the local
.meta init-form
applies to it; if it is absent, the slot it initialized to
.code nil
in each newly created instance of the new type.
However,
.code :init
and
.code :postinit
initializations are inherited from a base type and they apply to
the repeated slots, regardless of their kind. These initializations
take place on the instantiated object, and the slot references
resolve accordingly.
The initialization for slots which are specified using the
.code :method
or
.code :function
specifiers is reordered with regard to
.code :static
slots. Regardless of their placement in the
.code defstruct
form,
.code :method
and
.code :function
slots are initialized before
.code :static
slots. This ordering is useful, because it means that when the initialization
expression for a given static slot constructs an instance of the struct type,
any instance initialization code executing for that instance can use
all functions and methods of the struct type.
However, note the static slots which follow that slot in the
.code defstruct
syntax are not yet initialized. If it is necessary for a structure's
initialization code to have access to all static slots, even when the
structure is instantiated during the initialization of a static slot,
a possible solution may be to use lazy instantiation using the
.code lnew
operator, rather than ordinary eager instantiation via
.codn new .
It is also necessary to ensure that that the instance isn't accessed until all
static initializations are complete, since access to the instance slots of a
lazily instantiated structure triggers its initialization.
The structure name is specified using two forms, plain
.meta name
or the syntax
.mono
.meti >> ( name << arg *)
.onom
If the second form is used, then the structure type will support
"boa construction", where "boa" stands for "by order of arguments".
The
.metn arg s
specify the list of slot names which are to be initialized in the
by-order-of-arguments style. For instance, if three slot names
are given, then those slots can be optionally initialized by giving three
arguments in the
.code new
macro or the
.code make-struct
function.
Slots are first initialized according to their
.metn init-form s,
regardless of whether they are involved in boa construction.
A slot initialized in this style still has a
.meta init-form
which is processed independently of the existence of, and prior to,
boa construction.
The boa constructor syntax can specify optional parameters, delimited
by a colon, similarly to the
.code lambda
syntax. However, the optional parameters may not be arbitrary symbols;
they must be symbols which name
slots. Moreover, the
.mono
.meti >> ( name < init-form <> [ present-p ])
.onom
optional parameter syntax isn't supported.
When boa construction is invoked with optional arguments missing,
the default values for those arguments come from the
.metn init-form s
in the remaining
.code defstruct
syntax.
.TP* Examples:
.verb
(defvar *counter* 0)
;; New struct type foo with no super type:
;; Slots a and b initialize to nil.
;; Slot c is initialized by value of (inc *counter*).
(defstruct foo nil (a b (c (inc *counter*))))
(new foo) -> #S(foo a nil b nil c 1)
(new foo) -> #S(foo a nil b nil c 2)
;; New struct bar inheriting from foo.
(defstruct bar foo (c 0) (d 100))
(new bar) -> #S(bar a nil b nil c 0 d 100)
(new bar) -> #S(bar a nil b nil c 0 d 100)
;; counter was still incremented during
;; construction of d:
*counter* -> 4
;; override slots with new arguments
(new foo a "str" c 17) -> #S(foo a "str" b nil c 17)
*counter* -> 5
;; boa initialization
(defstruct (point x : y) nil (x 0) (y 0))
(new point) -> #S(point x 0 y 0)
(new (point 1 1)) -> #S(point x 1 y 1)
;; property list style initialization
;; can always be used:
(new point x 4 y 5) -> #S(point x 4 y 5)
;; boa applies last:
(new (point 1 1) x 4 y 5) -> #S(point x 1 y 1)
;; boa with optional argument omitted:
(new (point 1)) -> #S(point x 1 y 0)
;; boa with optional argument omitted and
;; with property list style initialization:
(new (point 1) x 5 y 5) -> #S(point x 1 y 5)
.brev
.coNP Macro @ defmeth
.synb
.mets (defmeth < type-name < name < param-list << body-form *)
.syne
.desc
Unless
.meta name
is one of the two symbols
.code :init
or
.codn :postinit ,
the
.code defmeth
macro installs a function into the static slot named by the symbol
.meta name
in the struct type indicated by
.metn type-name .
If the structure type doesn't already have such a static slot, it is
first added, as if by the
.code static-slot-ensure
function, subject to the same checks.
If the function has at least one argument, it can be used as a method. In that
situation, the leftmost argument passes the structure instance on which the
method is being invoked.
The function takes the arguments specified
by the
.meta param-list
symbols, and its body consists of the
.metn body-form s.
The
.metn body-form s
are placed into a
.code block
named
.codn name .
A method named
.code lambda
allows a structure to be used as if it were a function. When arguments
are applied to the structure as if it were a function, the
.code lambda
method is invoked with those arguments, with the object itself inserted
into the leftmost argument position.
If
.code defmeth
is used to redefine an existing method, the semantics can be inferred
from that of
.codn static-slot-ensure .
In particular, the method will be imposed into all subtypes which inherit
(do not override) the method.
If
.meta name
is the keyword symbol
.codn :init ,
then instead of operating on a static slot, the macro redefines the
.meta initfun
of the given structure type, as if by a call to the function
.codn struct-set-initfun .
Similarly, if
.meta name
is the keyword symbol
.codn :postinit ,
then the macro redefines the
.meta postinitfun
of the given structure type, as if by a call to the function
.codn struct-set-postinitfun .
When redefining
.code :initfun
the admonishments given in the description of
.code struct-set-initfun
apply: if the type has an
.meta initfun
generated by the
.code defstruct
macro, then that
.meta initfun
is what implements all of the slot initializations given in the
slot specifier syntax. These initializations are lost if the
.meta initfun
is overwritten.
The
.code defmeth
macro returns a method name: a unit of syntax of the form
.mono
.meti (meth < type-name << name )
.onom
which can be used as an argument to the accessor
.code symbol-function
and other situations.
.coNP Macros @ new and @ lnew
.synb
.mets (new >> { name | >> ( name << arg *)} >> { slot << init-form }*)
.mets (lnew >> { name | >> ( name << arg *)} >> { slot << init-form }*)
.syne
.desc
The
.code new
macro creates a new instance of the structure type named by
.metn name .
If the structure supports "boa construction", then, optionally, the
arguments may be given using the syntax
.mono
.meti >> ( name << arg *)
.onom
instead of
.metn name .
Slot values may also be specified by the
.meta slot
and
.meta init-form
arguments.
Note: the evaluation order in
.code new
is surprising: namely,
.metn init-form s
are evaluated before
.metn arg s
if both are present.
When the object is constructed, all default initializations take place
first. If the object's structure type has a supertype, then the supertype
initializations take place. Then the type's initializations take
place, followed by the
.meta slot
.meta init-form
overrides from the
.code new
macro, and lastly the "boa constructor" overrides.
If any of the initializations abandon the evaluation of
.code new
by a nonlocal exit such as an exception throw, the object's
finalizers, if any, are invoked.
The macro
.code lnew
differs from new in that it specifies the construction of a
lazy struct, as if by the
.code make-lazy-struct
function.
When
.code lnew
is used to construct an instance, a lazy struct is returned
immediately, without evaluating any of the
.meta arg
and
.meta init-form
expressions.
The expressions are evaluated when any of the object's
instance slots is accessed for the first time. At that time,
these expressions are evaluated (in the same order as under
.codn new )
and initialization proceeds in the same way.
If any of the initializations abandon the delayed initializations steps
arranged by
.code lnew
by a nonlocal exit such as an exception throw, the object's
finalizers, if any, are invoked.
Lazy initialization does not detect cycles. Immediately prior to the lazy
initialization of a struct, the struct is marked as no longer requiring
initialization. Thus, during initialization, its instance slots may be
freely accessed. Slots not yet initialized evaluate as
.codn nil .
.coNP Macros @ new* and @ lnew*
.synb
.mets (new* >> { expr | >> ( expr << arg *)} >> { slot << init-form }*)
.mets (lnew* >> { expr | >> ( expr << arg *)} >> { slot << init-form }*)
.syne
.desc
The
.code new*
and
.code lnew*
macros are variants, respectively, of
.code new
and
.codn lnew .
The difference in behavior in these macros relative to
.code new
and
.code lnew
is that the
.meta name
argument is replaced with an expression
.meta expr
which is evaluated. The value of
.meta expr
must be a struct type, or a symbol which is the name of a struct type.
With one exception, if
.meta expr0
is a compound expression, then
.mono
.meti (new* < expr0 ...)
.onom
is interpreted as
.mono
.meti (new* >> ( expr1 << args... ) ...)
.onom
where the head of
.metn expr0 ,
.metn expr1 ,
is actually the expression which is evaluated to produce the type, and the remaining
constituents of
.metn expr0 ,
.metn args ,
become the boa arguments. The same requirement applies to
.codn lnew* .
The exception is that if
.meta expr1
is the symbol
.codn dwim ,
this interpretation does not apply. Thus
.mono
.meti (new* >> [ fun << args... ] ...)
.onom
evaluates the
.mono
.meti >> [ fun << args... ]
.onom
expression, rather than treating it as
.mono
.meti (dwim < fun << args... )
.onom
where
.code dwim
would be evaluated as a variable reference expected to produce a type.
.TP* Examples:
.verb
;; struct with boa constructor
(defstruct (ab a : b) () a b)
;; error: find-struct-type is interpreted as a variable
(new* (find-struct-type 'ab) a 1) -> ;; error
;; OK: extra nesting.
(new* ((find-struct-type 'ab)) a 1) -> #S(ab a 1 b nil)
;; OK: dwim brackets without nesting.
(new* [find-struct-type 'ab] a 1) -> #S(ab a 1 b nil)
;; boa construction
(new* ([find-struct-type 'ab] 1 2)) -> #S(ab a 1 b 2)
(new* ((find-struct-type 'ab) 1 2)) -> #S(ab a 1 b 2)
;; mixed construction
(new* ([find-struct-type 'ab] 1) b 2) -> #S(ab a 1 b 2)
(let ((type (find-struct-type 'ab)))
(new* type a 3 b 4))
-> #S(ab a 3 b 4)
(let ((type (find-struct-type 'ab)))
(new* (type 3 4)))
-> #S(ab a 3 b 4)
.brev
.coNP Macro @ with-slots
.synb
.mets (with-slots >> ({ slot | >> ( sym << slot )}*) < struct-expr
.mets \ \ << body-form *)
.syne
.desc
The
.code with-slots
binds lexical macros to serve as aliases for the slots of a structure.
The
.meta struct-expr
argument is expected to be an expression which evaluates to a struct
object. It is evaluated once, and its value is retained. The aliases are then
established to the slots of the resulting struct value.
The aliases are specified as zero or more expressions which consist of either
a single symbol
.meta slot
or a
.mono
.meti >> ( sym << slot )
.onom
pair. The simple form binds a macro named
.meta slot
to a slot also named
.metn slot .
The pair form binds a macro named
.meta sym
to a slot named
.metn slot .
The lexical aliases are syntactic places: assigning to an alias causes
the value to be stored into the slot which it denotes.
After evaluating
.meta struct-expr
the
.code with-slots
macro arranges for the evaluation of
.metn body-form s
in the lexical scope in which the aliases are visible.
.TP* "Dialect Notes:"
The intent of the
.code with-slots
macro is to help reduce the verbosity of code which makes multiple
references to the same slot. Use of
.code with-slots
is less necessary in \*(TL than other Lisp dialects
thanks to the dot operator for accessing struct slots.
Lexical aliases to struct places can also be
arranged with considerable convenience using the
.code placelet
operator. However,
.code placelet
will not bind multiple aliases to multiple slots of the same object
such that the expression which produces the object is evaluated only
once.
.TP* Example:
.verb
(defstruct point nil x y)
;; Here, with-slots introduces verbosity because
;; each slot is accessed only once. The function
;; is equivalent to:
;;
;; (defun point-delta (p0 p1)
;; (new point x (- p1.x p0.x) y (- p1.y p0.y)))
;;
;; Also contrast with the use of placelet:
;;
;; (defun point-delta (p0 p1)
;; (placelet ((x0 p0.x) (y0 p0.y)
;; (x1 p1.x) (y1 p1.y))
;; (new point x (- x1 x0) y (- y1 y0)))))
(defun point-delta (p0 p1)
(with-slots ((x0 x) (y0 y)) p0
(with-slots ((x1 x) (y1 y)) p1
(new point x (- x1 x0) y (- y1 y0)))))
.brev
.coNP Macro @ qref
.synb
.mets (qref < object-form
.mets \ \ >> { slot | >> ( slot << arg *) | >> [ slot << arg *]}+)
.syne
.desc
The
.code qref
macro ("quoted reference") performs structure slot access. Structure slot
access is more conveniently expressed using the referencing dot notation, which
works by translating to qref
.code qref
syntax, according to the following equivalence:
.verb
a.b.c.d <--> (qref a b c d) ;; a b c d must not be numbers
.brev
(See the Referencing Dot section under Additional Syntax.)
The leftmost argument of
.code qref
is an expression which is evaluated. This argument is followed by one or more
reference designators.
If there are two or more designators, the following equivalence applies:
.verb
(qref obj d1 d2 ...) <---> (qref (qref obj d1) d2 ...)
.brev
That is to say,
.code qref
is applied to the object and a single designator. This must yield
an object, which to which the next designator is applied as if by
another
.code qref
operation, and so forth.
If the null-safe syntax
.code "(t ...)"
is present, the equivalence becomes more complicated:
.verb
(qref (t obj) d1 d2 ...) <---> (qref (qref (t obj) d1) d2 ...)
(qref obj (t d1) d2 ...) <---> (qref (t (qref obj d1)) d2 ...)
.brev
Thus,
.code qref
can be understood in terms of the semantics of the
binary form
.mono
.meti (qref < object-form << designator )
.onom
Designators come in three basic forms: a lone symbol, an ordinary compound expression
consisting of a symbol followed by arguments, or a DWIM expression
consisting of a symbol followed by arguments.
A lone symbol designator indicates the slot of that name. That is to say, the
following equivalence applies:
.verb
(qref o n) <--> (slot o 'n)
.brev
where
.code slot
is the structure slot accessor function. Because
.code slot
is an accessor, this form denotes the slot as a syntactic place;
slots can be modified via assignment to the
.code qref
form and the referencing dot syntax.
The slot name being implicitly quoted is the basis of the term
"quoted reference", giving rise to the
.code qref
name.
A compound designator indicates that the named slot is a function,
and arguments are to be applied to it. The following equivalence applies
in this case, except that
.code o
is evaluated only once:
.verb
(qref o (n arg ...)) <--> (call (slot o 'n) o arg ...)
.brev
A DWIM designator similarly indicates that the named slot is a function,
and arguments are to be applied to it. The following equivalence applies:
.verb
(qref obj [name arg ...]) <--> [(slot obj 'name) o arg ...]
.brev
Therefore, under this equivalence, this syntax provides the usual Lisp-1-style
evaluation rule via the
.code dwim
operator.
If the
.meta object-form
has the syntax
.mono
.meti (t << expression )
.onom
this indicates null-safe access: if
.meta expression
evaluates to
.code nil
then the entire expression
.mono
.meti (qref (t << expression ) << designator )
.onom
form yields
.codn nil .
This syntax is produced by the
.code .?
notation.
The null-safe access notation prevents not only slot access, but also
method or function calls on
.codn nil .
When a method or function call is suppressed due to the object being
.codn nil ,
no aspect of the method or function call is evaluated; not only
is the slot not accessed, but the argument expressions are not evaluated.
.TP* Example:
.verb
(defstruct foo nil
(array (vec 1 2 3))
(increment (lambda (self index delta)
(inc [self.array index] delta))))
(defvarl s (new foo))
;; access third element of s.array:
[s.array 2] --> 3
;; increment first element of array by 42
s.(increment 0 42) --> 43
;; access array member
s.array --> #(43 2 3)
.brev
Note how
.code increment
behaves much like a single-argument-dispatch object-oriented method.
Firstly, the syntax
.mono
s.(increment 0 42)
.onom
effectively selects the
.code increment
function which is particular to the
.code s
object. Secondly, the object is passed to the selected function as the
leftmost argument, so that the function has access to the object.
.coNP Macro @ uref
.synb
.mets (uref >> { slot | >> ( slot << arg *) | >> [ slot << arg *]}+)
.syne
.desc
The
.code uref
macro ("unbound reference") expands to an expression which evaluates to a
function. The function takes exactly one argument: an object.
When the function is invoked on an object, it references slots
or methods relative to that object.
Note: the
.code uref
syntax may be used directly, but it is also produced by the unbound referencing
dot syntactic sugar:
.verb
.a --> (uref a)
.?a --> (uref t a)
.(f x) --> (uref (f x))
.(f x).b --> (uref (f x) b)
.a.(f x).b --> (uref a (f x) b)
.brev
The macro may be understood in terms of the following translation
scheme:
.verb
(uref a b ...) --> (lambda (o) (qref o a b ...))
(uref t a b ...) --> (lambda (o) (if o (qref o a b ...)))
.brev
where
.code o
is understood to be a unique symbol (for instance, as produced by the
.code gensym
function).
When only one
.code uref
argument is present, these equivalences also hold:
.verb
(uref (f a b c ...)) <--> (umeth f a b c ...)
(uref s) <--> (usl s)
.brev
The terminology "unbound reference" refers to the property that
.code uref
expressions produce a function which isn't bound to a structure
object. The function binds a slot or method; the call to that function then
binds an object to that function, as an argument.
.TP* Examples:
Suppose that the objects in
.code list
have slots
.code a
and
.codn b .
Then, a list of the
.code a
slot values may be obtained using:
.verb
(mapcar .a list)
.brev
because this is equivalent to
.verb
(mapcar (lambda (o) o.a) list)
.brev
Because
.code uref
produces a function, its result can be operated upon by
functional combinators. For instance, we can use the
.code juxt
combinator to produce a list of two-element lists,
which hold the
.code a
and
.code b
slots from each object in
.codn list :
.verb
(mapcar (juxt .a .b) list)
.brev
.coNP Macro @ meth
.synb
.mets (meth < struct < slot << curried-expr *)
.syne
.desc
The
.code meth
macro allows indirection upon a method-like function stored
in a function slot.
The
.code meth
macro binds
.meta struct
as the leftmost argument of the function stored in
.metn slot ,
returning a function which takes the remaining arguments.
That is to say, it returns a function
.meta f
such that
.mono
.meti >> [ f < arg ...]
.onom
calls
.mono
.meti >> [ struct.slot < struct < arg ...]
.onom
except that
.meta struct
is evaluated only once.
If one or more
.meta curried-expr
expressions are present, their values are bound inside
.meta f
also, and when
.meta f
is invoked, these are passed to the function stored in the slot.
Thus if
.meta f
is produced by
.code "(meth struct slot c1 c2 c3 ...)"
then
.mono
.meti >> [ f < arg ...]
.onom
calls
.mono
.meti >> [ struct.slot < struct < c1v < c2v < c3v ... < arg ...]
.onom
except that
.meta struct
is evaluated only once, and
.metn c1v ,
.meta c2v
and
.meta c3v
are the values of expressions
.codn c1 ,
.code c2
and
.codn c3 .
The argument
.meta struct
must be an expression which evaluates to a struct.
The
.meta slot
argument is not evaluated, and must be a symbol denoting a slot.
The syntax can be understood as a translation to a call of the
.code method
function:
.verb
(meth a b) <--> (method a 'b)
.brev
If
.meta curried-arg
expressions are present, the translation may be be understood
as:
.verb
(meth a b c1 c2 ...) <--> [(fun method) a 'b c1 c2 ...]
.brev
In other words the
.meta curried-arg
expressions are evaluated under the
.code dwim
operator evaluation rules.
.TP* Example:
.verb
;; struct for counting atoms eq to key
(defstruct (counter key) nil
key
(count 0)
(:method increment (self key)
(if (eq self.key key)
(inc self.count))))
;; pass all atoms in tree to func
(defun map-tree (tree func)
(if (atom tree)
[func tree]
(progn (map-tree (car tree) func)
(map-tree (cdr tree) func))))
;; count occurrences of symbol a
;; using increment method of counter,
;; passed as func argument to map-tree.
(let ((c (new (counter 'a)))
(tr '(a (b (a a)) c a d)))
(map-tree tr (meth c increment))
c)
--> #S(counter key a count 4
increment #<function: type 0>)
.brev
.coNP Macro @ umeth
.synb
.mets (umeth < slot << curried-expr *)
.syne
.desc
The
.code umeth
macro binds the symbol
.meta slot
to a function and returns that function.
The
.meta curried-expr
arguments, if present, are evaluated as if they were
arguments to the
.code dwim
operator.
When that function is called, it expects at least one argument.
The leftmost argument must be an object of struct type.
The slot named
.meta slot
is retrieved from that object, and is expected to be a function.
That function is called with the object, followed by the values
of the
.metn curried-expr s,
if any, followed by that function's arguments.
The syntax can be understood as a translation to a call of the
.code umethod
function:
.verb
(umeth s ...) <--> [umethod 's ...]
.brev
The macro merely provides the syntactic sugar of not having to quote the
symbol, and automatically treating the curried argument expressions
using Lisp-1 semantics of the
.code dwim
operator.
.TP* Example:
.verb
;; seal and dog are variables which hold structures of
;; different types. Both have a method called bark.
(let ((bark-fun (umeth bark)))
[bark-fun dog] ;; same effect as dog.(bark)
[bark-fun seal]) ;; same effect as seal.(bark)
.brev
The
.code u
in
.code umeth
stands for "unbound". The function produced by
.code umeth
is not bound to any specific object; it binds to an object whenever it is
invoked by retrieving the actual method from the object's slot at call time.
.coNP Macro @ usl
.synb
.mets (usl << slot )
.syne
.desc
The
.code usl
macro binds the symbol
.meta slot
to a function and returns that function.
When that function is called, it expects exactly one argument.
That argument must be an object of struct type.
The slot named
.meta slot
is retrieved from that object and returned.
The name
.code usl
stands for "unbound slot". The term "unbound" refers to the returned
function not being bound to a particular object. The binding of the
slot to an object takes place whenever the function is called.
.coNP Function @ make-struct-type
.synb
.mets (make-struct-type < name < super < static-slots < slots
.mets \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ < static-initfun < initfun << boactor
.mets \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ < boactor << postinitfun )
.syne
.desc
The
.code make-struct-type
function creates a new struct type.
The
.meta name
argument must be a bindable symbol, according to the
.code bindable
function. It specifies the name property of the struct
type as well as the name under which the struct type
is globally registered.
The
.meta super
argument indicates the supertype for the struct type.
It must be either a value of type
.codn struct-type ,
a symbol which names a struct type, or else
.codn nil ,
indicating that the newly created struct type has no supertype.
The
.meta static-slots
argument is a list of symbol which specify static slots.
The symbols must be bindable and the list must not contain duplicates.
The
.meta slots
argument is a list of symbols which specifies the instance slots.
The symbols must be bindable and there must not be any duplicates
within the list, or against entries in the
.meta static-slots
list.
The new struct type's effective list of slots is formed by appending
together
.meta static-slots
and
.metn slots ,
and then appending that to the list of the supertype's slots, and
de-duplicating the resulting list as if by the
.code uniq
function. Thus, any slots which are already present in the supertype are
removed. If the structure has no supertype, then the list of supertype
slots is taken to be empty. When a structure is instantiated, it shall have all
the slots specified in the effective list of slots. Each instance slot
shall be initialized to the value
.codn nil ,
prior to the invocation of
.meta initfun
and
.metn boactor .
The
.meta static-initfun
argument either specifies an initialization function, or is
.codn nil ,
which is equivalent to specifying a function which does nothing.
Prior to the invocation of
.metn static-initfun ,
each new static slot shall be initialized the value
.codn nil .
Inherited static slots retain their values from the supertype.
If specified,
.meta static-initfun
function must
accept one argument. When the structure type is created (before
the
.code make-struct-type
function returns) the
.meta static-initfun
function is invoked, passed the newly created
structure type as its argument.
The
.meta initfun
argument either specifies an initialization function, or is
.codn nil ,
which is equivalent to specifying a function which does nothing.
If specified, this function must
accept one argument. When a structure is instantiated, every
.meta initfun
in its chain of supertype ancestry is invoked, in order of inheritance,
so that the root supertype's
.meta initfun
is called first and the structure's own specific
.meta initfun
is called last. These calls occur before the slots are initialized
from the
.meta arg
arguments
or the
.meta slot-init-plist
of
.codn make-struct .
Each function is passed the newly created structure
object, and may alter its slots.
If multiple inheritance occurs, the
.meta initfun
functions of multiple supertypes are called in right-to-left order.
The
.meta boactor
argument either specifies a by-order-of-arguments initialization
function ("boa constructor") or is
.codn nil ,
which is equivalent to specifying a constructor which does nothing.
If specified, it must be a function which takes at least one argument.
When a structure is instantiated, and boa arguments are given, the
.meta boactor
is invoked, with the structure as the leftmost argument, and
the boa arguments as additional arguments. This takes place
after the processing of
.meta initfun
functions, and after the processing of the
.meta slot-init-plist
specified in the
.code make-struct
call. Note that the
.meta boactor
functions of the supertypes are not called, only the
.meta boactor
specific to the type being constructed.
The
.meta postinitfun
argument either specifies an initialization function, or is
.codn nil ,
which is equivalent to specifying a function which does nothing.
If specified, this function must accept one argument.
The
.meta postinitfun
function is similar to
.metn initfun .
The difference is that
.meta postinitfun
functions are called after all other initialization processing,
rather than before. They are are also called in order of
inheritance: the
.meta postinitfun
of a structure's supertype is called before its own,
and in right-to-left order among multiple supertypes
under multiple inheritance.
.coNP Function @ find-struct-type
.synb
.mets (find-struct-type << name )
.syne
.desc
The
.code find-struct-type
returns a
.code struct-type
object corresponding to the symbol
.metn name .
If no struct type is registered under
.metn name ,
then it returns
.codn nil .
A
.code struct-type
object exists for each structure type and holds information about it.
These objects are not themselves structures and are all of the
same type,
.codn struct-type .
.coNP Function @ struct-type-p
.synb
.mets (struct-type-p << obj )
.syne
.desc
The
.code struct-type-p
function returns
.code t
if
.meta obj
is a structure type, otherwise it returns
.codn nil .
A structure type is an object of type
.codn struct-type ,
returned by
.codn find-struct-type .
.coNP Function @ struct-type-name
.synb
.mets (struct-type-name << type-or-struct )
.syne
.desc
The
.code struct-type-name
function determines a structure type from the
.meta type-or-struct
argument and returns that structure type's symbolic name.
The
.meta type-or-struct
argument must be either a struct type object (such as the return value of
a successful lookup via
.codn find-struct-type ),
a symbol which names a struct type,
or else a struct instance.
.coNP Function @ super
.synb
.mets (super <> [ type-or-struct ])
.syne
.desc
The
.code super
function determines a structure type from the
.meta type-or-struct
argument and returns the struct type object which is
the supertype of that type, or else
.code nil
if that type has no supertype.
The
.meta type-or-struct
argument must be either a struct type object,
a symbol which names a struct type,
or else a struct instance.
.coNP Function @ make-struct
.synb
.mets (make-struct < type < slot-init-plist << arg *)
.syne
.desc
The
.code make-struct
function returns a new object which is an instance of the
structure type
.metn type .
The
.meta type
argument must either be a
.code struct-type
object, or else a symbol which is the name of a structure.
The
.meta slot-init-plist
argument gives a list of slot initializations in the
style of a property list, as defined by the
.code prop
function. It may be empty, in which case
it has no effect. Otherwise, it specifies slot names
and their values. Each slot name which is given must
be a slot of the structure type. The corresponding value
will be stored into the slot of the newly created object. If a slot is
repeated, it is unspecified which value takes effect.
The optional
.metn arg s
specify arguments to the structure type's boa constructor.
If the arguments are omitted, the boa constructor is not invoked.
Otherwise the boa constructor is invoked on the structure object
and those arguments. The argument list must match the trailing parameters of
the boa constructor (the remaining parameters which follow the leftmost
argument which passes the structure to the boa constructor).
When a new structure is instantiated by
.codn make-struct ,
its slot values are first initialized by the structure type's registered
functions as described under
.codn make-struct-type .
Then, the
.meta slot-init-plist
is processed, if not empty, and finally, the
.metn arg s
are processed, if present, and passed to the boa constructor.
If any of the initializations abandon the evaluation of
.code make-struct
by a nonlocal exit such as an exception throw, the object's
finalizers, if any, are invoked.
.coNP Function @ make-lazy-struct
.synb
.mets (make-lazy-struct < type << argfun )
.syne
.desc
The
.code make-lazy-struct
function returns a new object which is an instance of the
structure type
.metn type .
The
.meta type
argument must either be a
.code struct-type
object, or else a symbol which is the name of a structure.
The
.meta argfun
argument should be a function which can be called with no parameters
and returns a cons cell. More requirements are specified below.
The object returned by
.code make-lazy-struct
is a lazily-initialized struct instance, or
.IR "lazy struct" .
A lazy struct remains uninitialized until just before the first access
to any of its instance slots. Just before an instance slot is
accessed, initialization
takes place as follows. The
.meta argfun
function is invoked with no arguments. Its return value must be a cons
cell. The
.code car
of the cons cell is taken to be a property list, as defined by the
.code prop
function. The
.code cdr
field is taken to be a list of arguments. These values are treated
as if they were, respectively, the
.meta slot-init-plist
and the boa constructor arguments given in a
.code make-struct
invocation. Initialization of the structure proceeds as described
in the description of
.codn make-struct .
.coNP Functions @ struct-from-plist and @ struct-from-args
.synb
.mets (struct-from-plist < type >> { slot << value }*)
.mets (struct-from-args < type << arg *)
.syne
.desc
The
.code struct-from-plist
and
.code struct-from-args
functions are interfaces to the
.code make-struct
function.
The
.code struct-from-plist
function passes its
.meta slot
and
.meta value
arguments as the
.meta slot-init-plist
argument of
.codn make-struct .
It passes no boa constructor arguments.
The
.code struct-from-args
function calls
.meta make-struct
with an empty
.metn slot-init-plist ,
passing down the list of
.metn arg s.
The following equivalences hold:
.verb
(struct-from-plist a s0 v0 s1 v1 ...)
<--> (make-struct a (list s0 v0 s1 v1 ...))
(struct-from-args a v0 v1 v2 ...)
<--> (make-struct a nil v0 v1 v2 ...)
.brev
.coNP Function @ allocate-struct
.synb
.mets (allocate-struct << type )
.syne
.desc
The
.code allocate-struct
provides a low-level allocator for structure objects.
The
.meta type
argument must either be a
.code struct-type
object, or else a symbol which is the name of a structure.
The
.code allocate-struct
creates and returns a new instance of
.meta type
all of whose instance slots take on the value
.codn nil .
No initializations are performed. The struct type's
registered initialization functions are not invoked.
.coNP Function @ copy-struct
.synb
.mets (copy-struct << struct-obj )
.syne
.desc
The
.code copy-struct
function creates and returns a new object which is a duplicate of
.metn struct-obj ,
which must be a structure.
The duplicate object is a structure of the same type as
.meta struct-obj
and has the same slot values.
The creation of a duplicate does not involve calling any of the
struct type's initialization functions.
Only instance slots participate in the duplication. Since
the original structure and copy are of the same structure type,
they already share static slots.
This is a low-level, "shallow" copying mechanism. If an object design
calls for a higher level cloning mechanism with deep copying or other
additional semantics, one can be built on top of
.codn copy-struct .
For instance, a structure can have a
.code copy
method similar to the following:
.verb
(:method copy (me)
(let ((my-copy (copy-struct me)))
;; inform the copy that it has been created
;; by invoking its copied method.
my-copy.(copied)
my-copy))
.brev
which can then be invoked on whatever object needs copying.
(Note that this method is not a special structure function, and is thus
not taken into account by the
.code copy
function.)
Since this logic is generic, it can be placed in a base
method. The
.code copied
method which it calls is the means by which the new object is notified that it
is a copy. This method takes on whatever special responsibilities are required
when a copy is produced, such as registering the object in various necessary
associations, or performing a deeper copy of some of the objects held
in the slots.
The
.code copied
handler can be implemented at multiple levels of an inheritance hierarchy. The
initial call to
.code copied
from
.code copy
will call the most derived override of that method.
To call the corresponding method in the base class, a given derived method
can use the
.code call-super-fun
function, or else the
.code "(meth ...)"
syntax in the first position of a compound form, in place of a function name.
Examples of both are given in the documentation for
.codn call-super-fun .
Thus derived structs can inherit the copy handling logic from base structs, and
extend it with their own.
.coNP Accessor @ slot
.synb
.mets (slot < struct-obj << slot-name )
.mets (set (slot < struct-obj << slot-name ) << new-value )
.syne
.desc
The
.code slot
function retrieves a structure's slot. The
.meta struct-obj
argument must be a structure, and
.meta slot-name
must be a symbol which names a slot in that structure.
Because
.code slot
is an accessor, a
.code slot
form is a syntactic place which denotes the
slot's storage location.
A syntactic place expressed by
.code slot
does not support deletion.
.coNP Function @ slotset
.synb
.mets (slotset < struct-obj < slot-name << new-value )
.syne
.desc
The
.code slotset
function stores a value in a structure's slot.
The
.meta struct-obj
argument must be a structure, and
.meta slot-name
must be a symbol which names a slot in that structure.
The
.meta new-value
argument specifies the value to be stored in the slot.
If a successful store takes place to an instance slot of
.metn struct-obj ,
then the dirty flag of that object is set, causing the
.code test-dirty
function to report true for that object.
The
.code slotset
function returns
.metn new-value .
.coNP Functions @, test-dirty @ clear-dirty and @ test-clear-dirty
.synb
.mets (test-dirty << struct-obj )
.mets (clear-dirty << struct-obj )
.mets (test-clear-dirty << struct-obj )
.syne
.desc
The
.codn test-dirty ,
.code clear-dirty
and
.code test-clear-dirty
functions comprise the interface for interacting with structure
dirty flags.
Each structure instance has a dirty flag. When this flag is set, the
structure instance is said to be dirty, otherwise it is said to be clean. A
newly created structure is dirty. A structure remains dirty until its dirty
flag is explicitly reset. If a structure is clean, and one of its instance
slots is overwritten with a new value, it becomes dirty.
The
.code test-dirty
function returns the dirty flag of
.metn struct-obj :
.code t
if
.meta struct-obj
is dirty, otherwise
.codn nil .
The
.code clear-dirty
function clears the dirty flag of
.meta struct-obj
and returns
.meta struct-obj
itself.
The
.code test-clear-dirty
flag combines these operations: it makes a note of the dirty flag of
.meta struct-obj
and clears it. Then it returns the noted value,
.code t
or
.codn nil .
.coNP Function @ structp
.synb
.mets (structp << obj )
.syne
.desc
The
.code structp
function returns
.code t
if
.meta obj
is a structure, otherwise it returns
.codn nil .
.coNP Function @ struct-type
.synb
.mets (struct-type << struct-obj )
.syne
.desc
The
.code struct-type
function returns the structure type object which
represents the type of the structure object instance
.metn struct-obj .
.coNP Function @ clear-struct
.synb
.mets (clear-struct < struct-obj <> [ value ])
.syne
.desc
The
.code clear-struct
replaces all instance slots of
.meta struct-obj
with
.metn value ,
which defaults to
.code nil
if omitted.
Note that finalizers are not executed prior to replacing
the slot values.
.coNP Function @ reset-struct
.synb
.mets (reset-struct << struct-obj )
.syne
.desc
The
.code reset-struct
function reinitializes the structure object
.meta struct-obj
as if it were being newly created.
First, all the slots are set to
.code nil
as if by the
.code clear-struct
function. Then the slots are initialized by invoking the
initialization functions, in order of the supertype ancestry, just as would be
done for a new structure object created by
.code make-struct
with an empty
.meta slot-init-plist
and no boa arguments.
Note that finalizers registered against
.meta struct-obj
are not invoked prior to the reset operation, and remain registered.
If the structure has state which is cleaned up by
finalizers, it is advisable to invoke them using
.code call-finalizers
prior to using
.codn reset-struct ,
or to take other measures to deal with the
situation.
If the structure specifies
.code :fini
handlers, then the reinitialization will cause
these to registered, just like when a new object
it constructed. Thus if
.code call-finalizers
is not used prior to
.codn reset-struct ,
this will result in the existence of duplicate registrations of the
finalization functions.
Finalizers registered against
.meta struct-obj
.B are
invoked if an exception is thrown
during the reinitialization, just like when a new
structure is being constructed.
.coNP Function @ replace-struct
.synb
.mets (replace-struct < target-obj << source-obj )
.syne
.desc
The
.code replace-struct
function causes
.meta target-obj
to take on the attributes of
.meta source-obj
without changing its identity.
The type of
.code target-obj
is changed to that of
.codn source-obj .
All instance slots of
.code target-obj
are discarded, and it is given new slots,
which are copies of the instance slots of
.codn source-obj .
Because of the type change,
.code target-obj
implicitly loses all of its original static slots,
and acquires those of
.codn "source obj" .
Note that finalizers registered against
.meta target-obj
are not invoked, and remain registered.
If
.meta target-obj
has state which is cleaned up by
finalizers, it is advisable to invoke them using
.code call-finalizers
prior to using
.codn replace-struct ,
or to take other measures to handle the situation.
If the
.meta target-obj
and
.meta source-obj
arguments are the same object,
.code replace-struct
has no effect.
The return value is
.metn target-obj .
.coNP Function @ method
.synb
.mets (method < struct-obj < slot-name << curried-arg *)
.syne
.desc
The
.code method
function retrieves a function
.meta m
from a structure's slot
and returns a new function which binds that function's
left argument. If
.meta curried-arg
arguments are present, then they are also stored in
the returned function. These are the
.IR "curried arguments" .
The
.meta struct-obj
argument must be a structure, and
.meta slot-name
must be a symbol denoting a slot in that structure.
The slot must hold a function of at least one
argument.
The function
.meta f
which
.code method
function returns, when invoked,
calls the function
.meta m
previously retrieved from the object's
slot, passing to that function
.meta struct-obj
as the leftmost argument, followed by the curried
arguments, followed by all of
.metn f 's
own arguments.
Note: the
.code meth
macro is an alternative interface which is suitable if
the slot name isn't a computed value.
.coNP Function @ super-method
.synb
.mets (super-method < struct-obj << slot-name )
.syne
.desc
The
.code super-method
function retrieves a function from a static
slot belonging to one of the direct supertypes of the structure type of
.metn struct-obj .
It then returns a function which binds
that function's left argument to the structure.
The
.meta struct-obj
argument must be a structure which has at least one supertype, and
.meta slot-name
must be a symbol denoting a static slot in one of those supertypes.
The slot must hold a function of at least one
argument. The supertypes are searched from left to right for a static
slot named
.metn slot-name ;
when the first such slot is found, its value is used.
The
.code super-method
function returns a function which, when invoked,
calls the function previously retrieved from
the supertype's static slot, passing to that function
.meta struct-obj
as the leftmost argument, followed by the function's
own arguments.
.coNP Function @ umethod
.synb
.mets (umethod < slot-name << curried-arg *)
.syne
.desc
The
.code umethod
returns a function which represents the set of all methods named by
the slot
.meta slot-name
in all structure types, including ones not yet defined.
The
.meta slot-name
argument must be a symbol.
If one or more
.meta curried-arg
argument are present, these values represent the
.I "curried arguments"
which are stored in the function object which is returned.
This returned function must be called with at least one argument. Its leftmost
argument must be an object of structure type, which has a slot named
.metn slot-name .
The function will retrieve the value of the slot from that object,
expecting it to be a function, and calls it, passing to it the following
arguments: the object itself; all of the curried arguments, if any; and
all of its remaining arguments.
Note: the
.code umethod
name stands for "unbound method". Unlike the
.code method
function,
.code umethod
doesn't return a method whose leftmost argument is already bound to
an object; the binding occurs at call time.
.coNP Function @ uslot
.synb
.mets (uslot << slot-name )
.syne
.desc
The
.code uslot
returns a function which represents all slots named
.meta slot-name
in all structure types, including ones not yet defined.
The
.meta slot-name
argument must be a symbol.
The returned function must be called with exactly one argument.
The argument must be a structure which has a slot named
.metn slot-name .
The function will retrieve the value of the slot from that object
and return it.
Note: the
.code uslot
name stands for "unbound slot". The returned function
isn't bound to a particular object. The binding of
.code slot-name
to a slot in the structure object occurs when the function is called.
.coNP Function @ slots
.synb
.mets (slots << type )
.syne
.desc
The
.code slots
function returns a list of all of the slots of struct type
.metn type .
The
.meta type
argument must be a structure type, or else a symbol
which names a structure type.
.coNP Function @ slotp
.synb
.mets (slotp < type << name )
.syne
.desc
The
.code slotp
function returns
.code t
if name
.meta name
is a symbol which names a slot in the structure type
.metn type .
Otherwise it returns
.codn nil .
The
.meta type
argument must be a structure type, or else a symbol
which names a structure type.
.coNP Function @ static-slot-p
.synb
.mets (static-slot-p < type << name )
.syne
.desc
The
.code static-slot-p
function returns
.code t
if name
.meta name
is a symbol which names a slot in the structure type
.metn type ,
and if that slot is a static slot.
Otherwise it returns
.codn nil .
The
.meta type
argument must be a structure type, or else a symbol
which names a structure type.
.coNP Function @ static-slot
.synb
.mets (static-slot < type << name )
.syne
.desc
The
.code static-slot
function retrieves the value of the static slot
named by symbol
.meta name
of the structure type
.metn type .
The
.meta type
argument must be a structure type or a symbol which names a
structure type, and
.meta name
must be a static slot of this type.
.coNP Function @ static-slot-set
.synb
.mets (static-slot-set < type < name << new-value )
.syne
.desc
The
.code static-slot-set
function stores
.meta new-value
into the static slot named by symbol
.meta name
of the structure type
.metn type .
It returns
.metn new-value .
The
.meta type
argument must be a structure type or the name of a structure type, and
.meta name
must be a static slot of this type.
.coNP Function @ static-slot-ensure
.synb
.mets (static-slot-ensure < type < name < new-value <> [ no-error-p ])
.syne
.desc
The
.code static-slot-ensure
ensures, if possible, that the struct type
.metn type ,
as well as possibly one or more struct types derived from it,
have a static slot called
.metn name ,
that this slot is not shared with a supertype,
and that the value stored in it is
.metn new-value .
Note: this function supports the redefinition of methods,
as the implementation underlying the
.code defmeth
macro; its semantics is designed to harmonize with expected
behaviors in that usage.
The function operates as follows.
If
.meta type
itself already has an instance slot called
.meta name
then an error is thrown, and the function has no effect, unless a
true argument is specified for the
.meta no-error-p
Boolean parameter. In that case, in the same situation, the function
has no effect and simply returns
.metn new-value .
If
.meta type
already has a non-inherited static slot called
.meta name
then this slot is overwritten with
.meta new-value
and the function returns
.metn new-value .
Types derived from
.meta type
may also have this slot, via inheritance; consequently, its value
changes in those types also.
If
.meta type
already has an inherited static slot called
.meta name
then its inheritance is severed; the slot is converted
to a non-inherited static slot of
.meta type
and initialized with
.metn new-value .
Then all struct types derived from
.meta type
are scanned. In each such type, if the original inherited
static slot is found, it is replaced with the same
newly converted static slot that was just introduced into
.metn type ,
so that all these types now inherit this new slot from
.meta type
rather than the original slot from some supertype of
.metn type .
These types all share a single instance of the slot with
.metn type ,
but not with supertypes of
.metn type .
In the remaining case,
.meta type
has no slot called
.metn name .
The slot is added as a static slot to
.metn type .
Then it is added to every struct type derived from
.meta type
which does not already have a slot by that name, as if
by inheritance. That is to say, types to which this slot is introduced share a
single instance of that slot. The value of the new slot is
.metn new-value ,
which is also returned from the function. Any subtypes of
.meta type
which already have a slot called
.meta name
are ignored, as are their subtypes.
.coNP Function @ static-slot-home
.synb
.mets (static-slot-home < type << name )
.syne
.desc
The
.code static-slot-home
method determines which structure type actually defines the
static slot
.meta name
present in struct type
.metn type .
If
.meta type
isn't a struct type, or the name of a struct type,
the function throws an error. Likewise, if
.meta name
isn't a static slot of
.metn type .
If
.meta name
is a static slot of
.meta type
then the function returns a struct type name symbol which is either
then name of
.meta type
itself, if the slot is defined specifically for
.meta type
or else the most distant ancestor of
.meta type
from which the slot is inherited.
.coNP Function @ call-super-method
.synb
.mets (call-super-method < struct-obj < name << argument *)
.syne
.desc
The
.code call-super-method
function is deprecated. Solutions involving
.code call-super-method
should be reworked in terms of
.codn call-super-fun .
The
.code call-super-method
retrieves the function stored in the static slot
.meta name
of one of the direct supertypes of
.meta struct-obj
and invokes it, passing to that function
.meta struct-obj
as the leftmost argument, followed by the given
.metn argument s,
if any.
The
.meta struct-obj
argument must be of structure type. Moreover,
that structure type must be derived from one or more supertypes,
and
.meta name
must name a static slot available from at least one of those supertypes.
The supertypes are searched left to right in search of this slot.
The object retrieved from that static slot must be
callable as a function, and accept the arguments.
Note that it is not correct for a method that is defined
against a particular type to use
.code call-super-method
to call the same method (or any other method) in the supertype
of that particular type. This is because
.code call-super-method
refers to the type of the object instance
.metn struct-obj ,
not to the type against which the calling method is defined.
.coNP Function @ call-super-fun
.synb
.mets (call-super-fun < type < name << argument *)
.syne
.desc
The
.code call-super-fun
retrieves the function stored in the slot
.meta name
of one of the supertypes of
.meta type
and invokes it, passing to that function the given
.metn argument s,
if any.
The
.meta type
argument must be a structure type. Moreover,
that structure type must be derived from one or more supertypes,
and
.meta name
must name a static slot available from at least one of those supertypes.
The supertypes are searched left to right in search of this slot.
The object retrieved from that static slot must be
callable as a function, and accept the arguments.
.TP* Example:
Print a message and call supertype method:
.verb
(defstruct base nil)
(defstruct derived base)
(defmeth base fun (obj arg)
(format t "base fun method called with arg ~s\en" arg))
(defmeth derived fun (obj arg)
(format t "derived fun method called with arg ~s\en" arg)
(call-super-fun 'derived 'fun obj arg))
;; Interactive Listener:
1> (new derived).(fun 42)
derived fun method called with arg 42
base fun method called with arg 42
.brev
Note that a static method or function in any structure type
can be invoked by using the
.code "(meth ...)"
name syntax in the first position of a compound form, as
a function name. Thus, the above
.code "derived fun"
can also be written:
.verb
(defmeth derived fun (obj arg)
(format t "derived fun method called with arg ~s\en" arg)
((meth base fun) obj arg))
.brev
.coNP Functions @ struct-get-initfun and @ struct-get-postinitfun
.synb
.mets (struct-get-initfun << type )
.mets (struct-get-postinitfun << type )
.syne
.desc
The
.code struct-get-initfun
and
.code struct-get-postinitfun
functions retrieve, respectively, a structure type's
.meta initfun
and
.meta postinitfun
functions. These are the functions which are initially configured in the call to
.code make-struct-type
via the
.meta initfun
and
.meta postinitfun
arguments.
Either one may be
.codn nil ,
indicating that the type has no
.meta initfun
or
.metn postinitfun .
.coNP Functions @ struct-set-initfun and @ struct-set-postinitfun
.synb
.mets (struct-set-initfun < type << function )
.mets (struct-set-postinitfun < type << function )
.syne
.desc
The
.code struct-set-initfun
and
.code struct-set-postinitfun
functions overwrite, respectively, a structure type's
.meta initfun
and
.meta postinitfun
functions. These are the functions which are initially configured in the call to
.code make-struct-type
via the
.meta initfun
and
.meta postinitfun
arguments.
The
.meta function
argument must either be
.code nil
or else a function which accepts one argument.
Note that
.meta initfun
has the responsibility for all instance slot initializations. The
.code defstruct
syntax compiles the initializing expressions in the slot specifier syntax
into statements which are placed into a function, which becomes the
.meta initfun
of the struct type.
.coNP Macro @ with-objects
.synb
.mets (with-objects >> ({( sym << init-form )}*) << body-form *)
.syne
.desc
The
.code with-objects
macro provides a binding construct similar to
.codn let* .
Each
.meta sym
must be a symbol suitable for use as a variable name.
Each
.meta init-form
is evaluated in sequence, and a binding is established for its
corresponding
.meta sym
which is initialized with the value of that form. The binding
is visible to subsequent
.metn init-form s.
Additionally, the values of the
.metn init-form s
are noted as they are produced. When the
.code with-objects
form terminates, by any means, the
.code call-finalizers
function is invoked on each value which was returned by an
.meta init-form
and had been noted. These calls are performed in the
reverse order relative to the original evaluation of the forms.
After the variables are established and initialized, the
.metn body-form s
are evaluated in the scope of the variables. The value of the
last form is returned, or else
.code nil
if there are no forms. The invocations of
.code call-finalizers
take place just before the value of the last form is returned.
.coNP Macro @ define-struct-clause
.synb
.mets (define-struct-clause < keyword < params <> [ body-form ]*)
.syne
.desc
The
.code define-struct-clause
macro makes available a new, application-defined
.code defstruct
clause. The clause is named by
.metn keyword ,
which must be a keyword symbol, and is implemented as a macro
transformation by the
.meta params
and
.metn body-form s
of the definition. The definition established by
.code define-struct-clause
is called a
.IR "struct clause macro" .
A struct clause macro is invoked when
.code defstruct
syntax is processed which contains one or more clauses which are
headed by the matching
.meta keyword
symbol.
The
.meta params
comprise a macro-style parameter list which must match the
invoking clause, otherwise an error exception is thrown.
When
.meta params
successfully matches the clause parameters, the parameters
are destructured into the parameters and the
.metn body-form s
are evaluated in the scope of those parameters.
The
.metn body-form s
must return a possibly list of
.code defstruct
clauses, not a single clause.
Each of the returned clauses is examined for the possibility that
it may be a struct clause macro; if so, it is expanded.
The built-in clause keywords
.codn :static ,
.codn :instance ,
.codn :function ,
.codn :method ,
.codn :init ,
.code :postinit
and
.code :fini
may not be used as the names of a struct clause macro; if any of these
symbols is used as the
.meta keyword
parameter of
.codn define-struct-clause ,
an error exception is thrown.
The return value of a
.code define-struct-clause
macro invocation is the
.meta keyword
argument.
.TP* Examples:
.verb
;; Trivial struct clause macro which consumes any number of
;; arguments and produces no slots:
(define-struct-clause :nothing (. ignored-args))
;; Consequently, the following defines a struct with one slot, x:
;; The (:nothing ...) clause disappears by producing no clauses.
(defstruct foo ()
(:nothing 1 2 3 beeblebrox)
x)
;; struct clause macro called :multi which takes an initial value
;; and zero or more slot names. It produces instance slot definitions
;; which all use that same initial value.
(define-struct-clause :multi (init-val . names)
(mapcar (lop list init-val) names))
;; define a struct with three slots initialized to zero:
(defstruct bar ()
(:multi 0 a b c)) ;; expands to (a 0) (b 0) (c 0)
;; struct clause macro to define a slot along with a
;; get and set method.
(define-struct-clause :getset (slot getter setter : init-val)
^((,slot ,init-val)
(:method ,getter (obj) obj.,slot)
(:method ,setter (obj new) (set obj.,slot new))))
;; Example use:
(defstruct point ()
(:getset x get-x set-x 0)
(:getset y get-y set-y 0))
;; This has exactly the same effect as the following defstruct:
(defstruct point ()
(x 0)
(y 0)
(:method get-x (obj) obj.x)
(:method set-x (ob new) (set obj.x new))
(:method get-y (obj) obj.y)
(:method set-y (ob new) (set obj.y new)))
.brev
.coNP Struct clause macro @ :delegate
.synb
.mets (:delegate < name <> ( param +) < delegate-expr <> [ target-name ])
.syne
.desc
The
:delegate
struct clause macro provides a way to define a method which is implemented entirely
by delegation to a different object. The name of the method is
.meta name
and its parameter list is specified in the same way as in the
.meta :method
clause. Instead of a method body, the
.code :delegate
clause has an expression
.meta delegate-expr
and an optional
.meta target-name
which defaults to
.metn name .
The
.meta delegate-expr
must be an expression which the delegate method can evaluate to
produce a delegate object. The delegate method then passes its
arguments to the target method, given by the
.meta target-name
argument, invoked on the delegate object. If the delegate method has optional
parameters which have not received an argument value, those parameters
are treated as if they had received the colon symbol
.code :
as their value, and that value is passed on. If the delegate method has
variadic parameters, they are applied to the target. If optional parameters
specified in
.code :delegate
are given argument values, those are discarded, playing no role in the
delegation.
.TP* Example:
Structure definitions:
.verb
(defstruct worker ()
name
(:method work (me)
`worker @{me.name} works`)
(:method relax (me : (min 15))
`worker @{me.name} relaxes for @min min`))
;; "contractor" class has a sub ("subcontractor") slot
;; which is another contractor of the same type.
;; The subcontractor's own sub slot, however is going
;; to be a worker.
(defstruct contractor ()
sub
(:delegate work (me) me.sub.sub)
(:delegate break (me : min) me.sub.sub relax))
.brev
The
.code contractor
structure's
.code work
and
.code break
methods delegate to the sub-subcontractor, which is going to be
instantiated as a
.code worker
object. Note that the
.code break
method delegates to a differently named method
.codn relax .
.verb
;; The objects are set up as described above.
;; general contractor co has a co.sub subcontractor,
;; and co.sub.sub is a worker:
(defvar co (new contractor
sub (new contractor
sub (new worker name "foo"))))
;; Call work method on general contractor:
;; this invokes co.sub.sub.(work) on the worker.
co.(work) -> "worker foo works"
;; Call break method on general contractor with
;; no argument. This causes co.sub.sub.(relax :)
;; to be invoked, triggering argument defaulting:
co.(break) -> "worker foo relaxes for 15 min"
;; Call break method with argument. This
;; invokes co.sub.sub.(relax 5), specifying a
;; value for the default argument:
co.(break 5) -> "worker foo relaxes for 5 min"
.brev
.coNP Struct clause macro @ :mass-delegate
.synb
.mets (:mass-delegate < self-var < delegate-expr
.mets \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ < from-type <> [ * ] <> [ method ]*)
.syne
.desc
The
:mass-delegate
struct macro provides a way to define multiple methods which are implemented
as delegates to corresponding methods on another object.
The implementation of
.code :mass-delegate
depends on the
.code :delegate
macro.
The
.meta self-var
argument must be a bindable symbol. In each generated delegate method,
this symbol will be the first argument. The purpose of this symbol is
to enable the
.meta delegate-expr
to refer to the delegating object.
The
.meta delegate-expr
is an expression which is inserted into every method. It is required
to evaluates to the delegate object. This expression may make a reference to
.meta self-var
in order to retrieve the delegate from the delegating object.
The
.meta from-type
argument is a symbol naming an existing structure type. If no such
structure type has been defined, an error exception is thrown.
After the
.meta from-type
argument, either zero or more slot names appear, optionally preceded by the
.code *
(asterisk) symbol.
If the
.code *
symbol is present, and isn't followed by any other symbols, it indicates
that all methods from
.meta from-type
are to be delegated. If symbols appear after the
.code *
then those specify exceptions: methods not to be delegated.
No validation is performed on the exception list; it may specify
nonexistent method names which have no effect.
If the
.code *
symbol is absent, then every
.meta method
symbol specifies a method to be delegated.
It is consequently expected to name a method of the
.metn from-type :
a static slot which contains a function. If any
.meta method
isn't a static slot of
.metn from-type ,
or not a static slot which contains a function, an error exception is thrown.
The
.code :mass-delegate
struct macro iterates over all of the methods of
.meta from-type
that selected for delegation, and for each one it generates a
.code :delegate
macro clause based on the existing method's parameter list.
For instance, the delegate for a method which has two required arguments and
one optional will itself have two required arguments and one optional.
Delegates are not simply wrapper functions which take any number of arguments
and try to pass them to the target.
The generated
.code :delegate
clauses are then processed by that struct clause macro.
Note: composition with delegation is a useful alternative when
multiple inheritance is not applicable or desired for various reasons.
One such reason is that structures that would be used as multiple inheritance
bases use the same symbols for certain slots, and the semantics of those
slots conflict. Under inheritance, same-named slots coming from different
bases become one slot,
Note: a particular
.meta from-type
being nominated in the
.code :mass-delegate
clause doesn't mean that the specific methods of that type shall be called
by the generated delegates. The methods that shall be called are those
of the calculated delegate object selected by the
.metn delegate-expr .
The
.meta from-type
is used as a source of the argument info, and method existence validation.
It is up to the application to ensure that the delegation using
.meta from-type
makes sense with respect to the delegate object that is selected by the
.metn delegate-expr :
for instance, by ensuring that this object is an instance of
.meta from-type
or a subtype thereof.
.TP* Example:
.verb
(defstruct foo-api ()
name
(:method begin (me) ^(foo ,me.name begin))
(:method increment (me delta) ^(foo ,me.name increment ,delta))
(:method end (me) ^(foo ,me.name end)))
(defstruct bar-api ()
name
(:method open (me) ^(bar ,me.name open))
(:method read (me buf) ^(bar ,me.name read ,buf))
(:method write (me buf) ^(bar ,me.name write ,buf))
(:method close (me) ^(bar ,me.name close)))
;; facade holds the two API objects by composition:
(defstruct facade ()
(foo (new foo-api name "foo"))
(bar (new bar-api name "bar"))
;; delegate foo-api style calls via me.foo
(:mass-delegate me me.foo foo-api *)
;; delegate bar-api style calls via me.bar
;; exclude the write method.
(:mass-delegate me me.bar bar-api * write))
;; instantiate facade as variable fa
(defvar fa (new facade)) -> fa
;; begin call on facade delegates through foo-api object.
fa.(begin) -> (foo "foo" begin)
fa.(increment) -> ;; error: too few arguments
fa.(increment 3) -> (foo "foo" increment 3)
fa.(open) -> (bar "bar" open)
fa.(write 4) -> ;; error: fa has no such method
.brev
.coNP Special variable @ *struct-clause-expander*
.desc
The
.code *struct-clause-expander*
special variable holds the hash table of associations between
keyword symbols and struct clause expander functions, defined by
.codn define-struct-clause .
If the expression
.code "[*struct-clause-expander* :sym]"
yields a function, then symbol
.code :sym
has a binding as a struct clause macro. If that
expression yields
.codn nil ,
then there is no such binding.
The macro expanders in
.code *struct-clause-expander*
are two-parameter functions. The first parameter accepts the
clause to be expanded. The second parameter accepts the
.code defstruct
form in which that clause is found; this is useful for error
reporting.
An expander function returns a list of clauses, which may be any, possibly
empty mixture of primary clauses accepted by
.code defstruct
and clause macros.
.SS* Special Structure Functions
Special structure functions are user-defined methods or structure functions
which are specially recognized by certain functions in \*(TL. They endow
structure objects with the ability to participate in certain usage scenarios,
or to participate in a customized way.
Special functions are required to bound to static slots, which is the
case if the
.code defmeth
macro is used, or when methods or functions are defined using syntax
inside a
.code defstruct
form. If a special function or method is defined as an instance slot,
then the behavior of library functions which depend on this method is
unspecified.
Special functions introduced below by the word "Method" receive an object
instance as an argument. Their syntax is indicated using the same notation
which may be used to invoke them, such as:
.verb
.mets << object .(function-name < arg ...)
.brev
However, those introduced as "Function" do not operate on an instance.
Their syntax is likewise indicated using the notation that may be used
to invoke them:
.verb
.mets <> '[' object .function-name < arg ...']'
.brev
If such a invocation is actually used, the
.meta object
instance only serves for identifying the struct type whose static slot
.code function-name
provides the function;
.meta object
doesn't participate in the call. An object is not strictly required since
the function can be called using
.verb
.mets [(static-slot < type 'function-name) < arg ...]
.brev
which looks up the function in the struct
.meta type
directly.
.coNP Method @ equal
.synb
.mets << object .(equal)
.syne
.desc
Normally, two struct values are not considered the same under the
.code equal
function unless they are the same object.
However, if the
.code equal
method is defined for a structure type, then instances of
that structure type support
.IR "equality substitution" .
The
.code equal
method must not require any arguments other than
.metn object .
Moreover, the method must never return
.codn nil .
When a struct which supports equality substitution is compared using
.codn equal ,
.code less
or
.codn greater ,
its
.code equal
method is invoked, and the return value is used in place of that
structure for the purposes of the comparison.
The same applies when an struct is hashed using the
.code hash-equal
function, or implicitly by an
.code :equal-hash
hash table.
Note: if an
.code equal
method is defined or redefined with different semantics for a struct
type whose instances have already been inserted as keys in an
.code :equal-based
hash table, the behavior of subsequent insertion and lookup operations
on that hash table becomes unspecified.
.coNP Method @ print
.synb
.mets << object .(print < stream << pretty-p )
.syne
.desc
If a method named by the symbol
.code print
is defined for a structure type, then it is used for printing instances
of that type.
The
.meta stream
argument specifies the output stream to which the printed representation
is to be written.
The
.meta pretty-p
argument is a Boolean flag indicating whether pretty-printing
is requested. Its value may simply be passed to recursive calls to
.codn print ,
or used to select between
.code ~s
or
.code ~a
formatting if
.code format
is used.
The value returned by the
.code print
method is significant. If the special keyword symbol
.code :
(colon) is returned, then the system will print the object
in the default way, as if no
.code print
method existed: it is understood that the method declined
the responsibility for printing the object.
If any other value is returned, then it is understood
that the method
.code print
method accepted the responsibility for printing the object,
and the system consequently will generate into
.meta stream
any output output pertaining to
.metn object 's
representation.
.coNP Method @ lambda
.synb
.mets << object .(lambda << arg *)
.syne
.desc
If a structure type provides a method called
.code lambda
then it can be used as a function.
This method can be called by name, using the syntax given
in the above syntactic description.
However, the intended use is that it allows the structure instance itself to be
used as a function. When arguments are applied to a structure object as if it
were a function, this is erroneous, unless that object has a
.code lambda
method. In that case, the arguments are passed to the lambda method.
The leftmost argument of the method is the structure instance
itself.
That is to say, the following equivalences apply, except that
.code s
is evaluated only once:
.verb
(call s args ...) <--> s.(lambda args ...)
[s args ...] <--> [s.lambda s args ...]
(mapcar s list) <--> (mapcar (meth s lambda) list)
.brev
Note: a form such as
.code "[s args ...]"
where
.code s
is a structure can be treated as a place if the method
.code lambda-set
is also implemented.
.coNP Method @ lambda-set
.synb
.mets << object .(lambda-set << arg * << new-value )
.syne
.desc
The
.code lambda-set
method, in conjunction with a
.code lambda
method, allows structures to be used as place accessors. If
structure
.code s
supports a
.meta lambda-set
with four arguments, then the following use of the
.code dwim
operator is possible:
.verb
(set [s a b c d] v)
(set (dwim s a b c d) v) ;; precisely equivalently
.brev
This has an effect which can be described by the following code:
.verb
(progn
s s.(lambda-set a b c d v)
v)
.brev
except that
.code s
and
.code v
are evaluated only once, and
.code a
through
.code d
are evaluated using the Lisp-1 semantics due the
.code dwim
operator.
If a place-mutating operator is used on this form which requires the prior
value, such as the
.code inc
macro, then the structure must support the
.code lambda
function also.
If
.code lambda
takes
.I n
arguments, then
.code lambda-set
should take
.I n+1
arguments. The first
.I n
arguments of these two methods are congruent; the extra rightmost argument
of
.code lambda-set
is the new value to be stored into the place denoted by the prior
arguments.
The return value of
.code lambda-set
is ignored.
Note: the
.code lambda-set
method is also used by the
.code rplaca
function, if no
.code rplaca
method exists.
.TP* Example
The following defines a structure with a single instance
slot
.code hash
which holds a hash table, as well as
.code lambda
and
.code lambda-set
methods:
.verb
(defstruct hash-wrapper nil
(hash (hash))
(:method lambda (self key)
[self.hash key])
(:method lambda-set (self key new-val)
(set [self.hash key] new-val) self))
.brev
An instance of this structure can now be used as follows:
.verb
(let ((s (new hash-wrapper)))
(set [s "apple"] 3
[s "orange] 4)
[s "apple"]) -> 3
.brev
.coNP Method @ length
.synb
.mets << object .(length)
.syne
.desc
If a structure has
.code length
method, then it can be used as an argument to the
.code length
function.
Structures which implement the methods
.codn lambda ,
.code lambda-set
and
.code length
can be treated as abstract vector-like sequences, because such
structures support the
.codn ref ,
.code refset
and
.code length
functions.
For instance, the
.code nreverse
function will operate on such objects.
Note: a structure which supports the
.code car
method also supports the
.code length
function, in a different way. Such a structure is treated by
.code length
as a list-like sequence, and its length is measured by walking the
sequence with
.code cdr
operations. If a structure supports both
.code length
and
.codn car ,
preference is given to
.codn length ,
which is likely to be much more efficient.
.coNP Methods @, car @ cdr and @ nullify
.synb
.mets << object .(car)
.mets << object .(cdr)
.mets << object .(nullify)
.syne
.desc
Structures may be treated as sequences if they define methods named
by the symbols
.codn car ,
.codn cdr ,
and
.codn nullify .
If a structure supports these methods, then these methods are used
by the functions
.codn car ,
.codn cdr ,
.codn nullify ,
.code empty
and various other sequence manipulating functions derived from them, when those
functions are applied to that object.
An object which implements these three methods can be considered to represent a
.I list-like
abstract sequence.
The object's
.code car
method should return the first value in that abstract sequence, or else
.code nil
if that sequence is empty.
The object's
.code cdr
method should return an object denoting the remainder of the sequence,
or else
.code nil
if the sequence is empty or contains only one value. This returned object can
be of any type: it may be of the same structure type as that object, a
different structure type, a list, or whatever else. If a non-sequence object
is returned.
The
.code nullify
method should return
.code nil
if the object is considered to denote an empty sequence. Otherwise it
should either return that object itself, or else return the sequence which
that object represents.
.coNP Methods @ rplaca and @ rplacd
.synb
.mets << object .(rplaca << new-car-value )
.mets << object .(rplacd << new-cdr-value )
.syne
.desc
If a structure type defines the methods
.code rplaca
and
.code rplacd
then, respectively, the
.code rplaca
and
.code rplacd
functions will use these methods if they are applied to instances of that type.
That is to say, when the function call
.mono
.meti (rplaca < o << v )
.onom
is evaluated, and
.meta o
is a structure type, the function inquires whether
.meta o
supports a
.code rplaca
method. If so, then, effectively,
.mono
.meti << o . (rplaca << v )
.onom
is invoked. The return value of this method call is ignored;
.code rplaca
returns
.metn o .
The analogous requirements apply to
.code rplacd
in relation to the
.code rplacd
method.
Note: if the
.code rplaca
method doesn't exist, the
.code rplaca
function falls back on trying to store
.meta new-car-value
by means of the structure type's
.code lambda-set
method, using an index of zero. That is to say, if the type has no
.code rplaca
method, but does have a
.code lambda-set
method, then
.mono
.meti << o . (lambda-set 0 << v )
.onom
is invoked.
.coNP Function @ from-list
.synb
.mets <> '[' object .from-list << list ']'
.syne
.desc
If a
.code from-list
structure function is defined for a structure type, it is called in certain
situations with an argument which is a list object. The function's purpose
is to construct a new instance of the structure type, derived from that
list.
The purpose of this function is to allow sequence processing operations
such as
.code mapcar
and
.code remove
to operate on a structure object as if it were a sequence, and return a
transformed sequence of the same type. This is analogous to the way such
functions can operate on a vector or string, and return a vector or string.
If a structure object behaves as a sequence thanks to providing
.codn car ,
.code cdr
and
.code nullify
methods, but does not have a
.code from-list
function, then those sequence-processing operations which return a sequence
will always return a plain list of items.
.coNP Function @ derived
.synb
.mets <> '[' object .derived < supertype << subtype ']'
.syne
.desc
If a structure type supports a function called
.metn derived ,
this function is called whenever a new type is defined which names
that type as its supertype.
The function is called with two arguments which are both struct types.
The
.meta supertype
argument gives the type that is being inherited from.
The
.meta subtype
gives the new type that is inheriting from
.metn supertype .
When a new structure type is defined, its list of immediate
supertypes is considered. For each of those supertypes which defines the
.code derived
function, the function is invoked.
The function is not retroactively invoked. If it is defined for
a structure type from which subtypes have already been derived,
it is not invoked for those existing subtypes.
If
.meta derived
directly inherits
.meta supertype
more than once, it is not specified whether this function is called
once, or multiple times.
Note: the
.meta supertype
parameter exists because the
.code derived
function is itself inherited. If the same version of this function is shared by
multiple structure types due to inheritance, this argument informs the function
which of those types it is being invoked for.
.coNP Methods @ iter-begin and @ iter-reset
.synb
.mets << object .(iter-begin)
.mets << object .(iter-reset << iter )
.syne
.desc
If an object supports the
.code iter-begin
method, it is considered iterable; the
.code iterable
function will return
.code t
if invoked on this object.
The responsibility of the
.code iter-begin
method is to return an iterator object: an object which supports
certain special methods related to iteration, according to one of two
protocols, described below.
The
.code iter-reset
method is optional. It is similar to
.code iter-begin
but takes an additional
.meta iter
argument, an iterator object that was previously returned by the
.code iter-begin
method of the same
.metn object .
If
.code iter-reset
determines that
.meta iter
can be reused for a new iteration, then it can suitably mutate the
state of
.meta iter
and return it. Otherwise, it behaves like
.code iter-begin
and returns a new iterator.
There are two protocols for iteration: the fast protocol, and the canonical
protocol.
Both protocols require the iterator object returned by the
.code iter-begin
method to provide the methods
.code iter-item
and
.codn iter-step .
If the iterator also provides the
.code iter-more
method, then the protocol which applies is the canonical protocol. If
that method is absent, then the fast protocol is followed.
Under the fast protocol, the
.code iter-more
method does not exist and is not involved. The iterable object's
.code iter-begin
method must return
.code nil
if the abstract sequence is empty. If an iterator is returned, it is assumed
that an object can be retrieved from the iterator by invoking its
.code iter-item
method. The iterator's
.code iter-next
method should return
.code nil
if there are no more objects in the abstract sequence, or else it should
return an iterator that obeys the fast protocol (possibly itself).
Under the canonical protocol, the iterator implements the
.code iter-more
function. The iterable object's
.code iter-begin
always returns an iterator object. The iterator object's
.code iter-more
method is always invoked to determine whether another item is available
from the sequence. The iterator object's
.code iter-step
method is expected to return an iterator object which conforms to the
canonical protocol.
.coNP Method @ iter-item
.synb
.mets << object .(iter-item)
.syne
.desc
The
.code iter-item
method is invoked on an iterator
.meta object
to retrieve the next item in the sequence.
Under the fast protocol, it
is assumed that if
.meta object
was returned by an iterable object's
.code iter-begin
method, or by an iterator's
.code iter-step
method, that an item is available. This method will be unconditionally invoked.
Under the canonical protocol for iteration, the
.code iter-more
method will be invoked on
.meta object
first. If that method yields true, then
.code iter-item
is expected to yield the next available item in the sequence.
Note: calls to the
.code iter-item
function, with
.meta object
as its argument, invoke the
.code iter-item
method. It is possible for an application to call
.code iter-item
through this function or directly as a method call
without first calling
.codn iter-more .
No iteration mechanism in the \*(TL standard library behaves this way.
If the iterator
.meta object
has no more items available and
.code iter-more
is invoked anyway, no requirements apply to its behavior or return value.
.coNP Method @ iter-step
.synb
.mets << object .(iter-step)
.syne
.desc
The
.code iter-step
method is invoked on an iterator object to produce an iterator object for the
remainder of the sequence, excluding the current item.
Under the fast iteration protocol, this method returns
.code nil
if there are no more items in the sequence.
Under the canonical iteration protocol, this method always returns
an iterator object. If no items remain in the sequence, then that
iterator object's
.code iter-more
method returns
.codn nil .
Furthermore, under this protocol,
.code iter-step
is not called if
.code iter-more
returns
.codn nil .
Note: calls to the
.code iter-step
function, with
.meta object
as its argument, invoke the
.code iter-step
method. It is possible for an application to call
.code iter-step
through this function or directly as a method call
without first calling
.codn iter-more .
No iteration mechanism in the \*(TL standard library behaves this way.
If the iterator
.meta object
has no more items available and
.code iter-step
is invoked anyway, no requirements apply to its behavior or return value.
.coNP Method @ iter-more
.synb
.mets << object .(iter-more)
.syne
.desc
If an iterator
.meta object
returned by
.code iter-begin
supports the
.code iter-more
method, then the canonical iteration protocol applies to that iteration
session. All subsequent iterators that are involved in the iteration
are assumed to conform to the protocol and should implement the
.code iter-more
method also. The behavior is unspecified otherwise.
The
.code iter-more
method is used to interrogate an iterator whether more unvisited items
remain in the sequence. This method does not advance the iteration,
and does not change the state of the iterator. It is idempotent: if it is
called multiple times without any intervening call to any other method,
it yields the same value.
If an iterator does not implement the
.code iter-more
method, then if the
.code iter-more
function is applied to that iterator, it unconditionally returns
.codn t .
.SS* Sequence Manipulation
Functions in this category uniformly manipulate abstract sequences. Lists,
strings and vectors are sequences.
Structure objects can behave
like sequences, either list-like or vector-like sequences, if they have
certain methods: see the previous section Special Structure Functions.
Moreover, hash tables behave like sequences of key-value entries represented by
.code cons
pairs. Not all sequence-processing functions accept hash-table sequences.
Additionally, some sequence-processing functions work not only with sequences
but with all iterable objects: objects that can be used as arguments to the
.code iter-begin
function. Such arguments are called
.meta iterable
rather than
.metn sequence ,
possibly abbreviated to
.meta iter
with or without a numeric suffix.
Hash tables are always supported if they appear as
.meta iterable
arguments.
.coNP Function @ seqp
.synb
.mets (seqp << object )
.syne
.desc
The function
.code seqp
returns
.code t
if
.meta object
is a sequence, otherwise
.codn nil .
Lists, vectors and strings are sequences. The object
.code nil
denotes
the empty list and so is a sequence.
Objects of type
.code buf
and
.code carray
are sequences, as are hash tables.
Structures which implement the
.code length
or
.code car
methods are considered sequences.
No other objects are sequences. However, future revisions of
the language may specify additional objects that are sequences.
.coNP Function @ iterable
.synb
.mets (iterable << object )
.syne
.desc
The
.code iterable
function returns
.code t
if
.meta object
is iterable, otherwise
.codn nil .
If
.meta object
is a sequence according to the
.code seqp
function, then it is iterable.
If
.meta object
is a structure which supports the
.code iter-begin
method, then it is iterable.
Additional objects that are not sequences are also iterable:
numeric or character ranges, and numbers. Future revisions
of the language may specify additional iterable objects.
.coNP Function @ make-like
.synb
.mets (make-like < list << object )
.syne
.desc
The
.meta list
argument must be a list. If
.meta object
is a sequence type,
then
.meta list
is converted to the same type of sequence and returned.
Otherwise the original
.meta list
is returned.
Conversion is supported to string and vector type.
Conversion to a structure type is possible for structures. If
.meta object
is an object of a structure type which has a static function
.codn from-list ,
then
.code make-like
calls that function, passing to it, and the resulting value is returned.
.meta list
and returns whatever value that function returns.
If
.meta object
is a
.codn carray ,
then
.meta list
is passed to the
.code carray-list
function, and the resulting value is returned. The second argument in the
.code carray-list
call is the element type taken from
.metn object .
The third argument is
.codn nil ,
indicating that the resulting
.code carray
is not to be null terminated.
Note: the
.code make-like
function is a helper which supports the development of
unoptimized versions of a generic function that accepts any type of
sequence as input, and produces a sequence of the same type as output.
The implementation of such a function can internally accumulate a list, and
then convert the resulting list to the same type as an input value
by using
.codn make-like .
.coNP Functions @, list-seq @ vec-seq and @ str-seq
.synb
.mets (list-seq << iterable )
.mets (vec-seq << iterable )
.mets (str-seq << iterable )
.syne
.desc
The
.codn list-seq ,
.code vec-seq
and
.code str-seq
functions convert an iterable object of any type into a list, vector
or string, respectively.
The list returned by
.code list-seq
is lazy.
The
.code list-seq
and
.code vec-seq
iterate the items of
.meta iterable
and accumulate these items into a new list or vector.
The
.code str-seq
similarly iterates the items of
.metn iterable ,
requiring them to be a mixture of characters and strings.
.coNP Functions @ length and @ len
.synb
.mets (length << iterable )
.mets (len << iterable )
.syne
.desc
The
.code length
function returns the number of items contained in
.metn iterable .
The
.code len
function is a synonym of
.codn length .
An attempt to calculate the length of infinite lazy lists will not terminate.
Iterable objects representing infinite ranges, such as integers and characters
are invalid arguments.
.coNP Function @ empty
.synb
.mets (empty << iterable )
.syne
.desc
If
.meta iterable
is a suitable argument for the
.code length
function, then the
.code empty
Returns
.code t
if
.mono
.meti (length << iterable )
.onom
is zero, otherwise
.codn nil .
The
.code empty
function also supports certain objects not suitable as arguments for
.codn length .
An infinite lazy list is not empty, and so
.code empty
returns
.code nil
for such an object.
The function also returns
.code nil
for iterable objects representing nonempty spaces, even if
those spaces are infinite. For instance
.code "(empty 0)"
yields
.code nil
because the set of integers beginning with 0 isn't empty.
.coNP Function @ nullify
.synb
.mets (nullify << iterable )
.syne
.desc
The
.code nullify
function returns
.code nil
if
.meta iterable
denotes an empty sequence.
Otherwise, if
.meta iterable
is not an empty sequence, or isn't a sequence, then
.meta iterable
itself is returned.
If
.meta iterable
is a structure object which supports the
.code nullify
method, then that method is called. If it returns
.code nil
then
.code nil
is returned. If the
.code nullify
method returns a substitute object other than the
.meta iterable
object itself, then
.code nullify
is invoked on that returned substitute object.
Note: the
.code nullify
function is a helper to support unoptimized generic
traversal of sequences. Thanks to the generalized behavior of
.codn cdr ,
non-list sequences can be traversed using
.codn cdr ,
similarly to proper lists, by checking for
.code cdr
returning the terminating value
.codn nil .
However, empty non-list sequences are handled incorrectly because
since they are not the
.code nil
object, they look nonempty under this paradigm of traversal.
The
.code nullify
function provides a correction: if the input sequence is filtered
through
.code nullify
then the subsequent list-like iteration works correctly.
Examples:
.verb
;; Incorrect for empty strings:
(defun print-chars (string)
(while string
(prinl (pop string))))
;; Corrected with nullify:
(defun print-chars (string)
(let ((s (nullify string)))
(while s
(prinl (pop s)))))
.brev
Note: optimized generic iteration is available in the form of iteration
based on
.code iter-begin
rather than
.cod3 car / cdr
and
.codn nullify .
Examples:
.verb
;; Efficient with iterators,
;; at the cost of verbosity:
(defun print-chars (string)
(let ((i (iter-begin string)))
(while (iter-more i)
(prinl (iter-item s))
(set s (iter-step s)))))
;; Using mapping function built on iterators:
(defun print-chars (string)
[mapdo prinl string])
.brev
.coNP Accessor @ sub
.synb
.mets (sub < sequence >> [ from <> [ to ]])
.mets (set (sub < sequence >> [ from <> [ to ]]) << new-val )
.syne
.desc
The
.code sub
function extracts a slice from input sequence
.metn sequence .
The slice is
a sequence of the same type as
.metn sequence .
If the
.meta from
argument is omitted, it defaults to
.codn 0 .
If the
.meta to
parameter is
omitted, it defaults to
.codn t .
Thus
.code "(sub a)"
means
.codn "(sub a 0 t)" .
The following semantic equivalence exists between a call to the
.code sub
function and
the DWIM-bracket syntax, except that
.code sub
is an ordinary function call form, which doesn't apply the
Lisp-1 evaluation semantics to its arguments:
.verb
;; from is not a list
(sub seq from to) <--> [seq from..to]
.brev
The description of the
.code dwim
operator\(emin particular, the section
on Range Indexing\(emexplains the semantics of the range specification.
The output sequence may share structure with the input sequence.
If
.meta sequence
is a
.code carray
object, then the function behaves like
.codn carray-sub .
If
.meta sequence
is a
.code buf
object, then the function behaves like
.codn sub-buf .
If
.meta sequence
is a
.code tree
object, then the function behaves like
.codn sub-tree .
Note: because
.code sub-tree
is not an accessor, assigning to the
.code sub
syntax in this case will produce an error.
The
.meta sequence
argument may also be any other object type that is suitable as input to the
.code iter-begin
function. In this situation, assigning to
.code sub
syntax produces an error. The behavior is complex. In cases where the
.meta from
and
.meta to
arguments imply that a suffix of
.meta sequence
is required, an iterator may be returned which traverses the suffix
of the sequence. In other cases, a list of the elements selected by
.code sub
is returned.
If
.meta sequence
is a structure, it must support the
.code lambda
method. The
.code sub
operation is transformed into a call to the
.code lambda
method according to the following equivalence:
.verb
(sub o from to) <--> o.(lambda (rcons from to))
(sub o : to) <--> o.(lambda (rcons : to))
(sub o from) <--> o.(lambda (rcons from :))
(sub o) <--> o.(lambda (rcons : :))
.brev
That is to say, the
.meta from
and
.code to
arguments are converted to range object. If either argument
is missing, the
.code :
(colon) keyword symbol is used for the corresponding element of the range.
When a
.code sub
form is used as a syntactic place, that place denotes a slice of
.metn seq .
The
.meta seq
argument must be itself be syntactic place, because
receives a new value, which may be different from its original value in
cases when
.meta seq
is a list.
Overwriting that slice is equivalent to using the
.code replace
function. The following equivalences give the semantics, except that
.codn x ,
.codn a ,
.code b
and
.code v
are evaluated only once, in left-to-right order:
.verb
(set (sub x a b) v) <--> (progn (set x (replace x v a b))
v)
(del (sub x a b)) <--> (prog1 (sub x a b)
(set x (replace x nil a b)))
.brev
Note that the value of
.code x
is overwritten with the value returned by
.codn replace .
If
.code x
is a vector or string, then the return value of
.code replace
is just
.codn x :
the identity of the object doesn't change under mutation.
However, if
.code x
is a list, its identity changes when items are added to or removed from
the front of the list, and in those cases
.code replace
will return a value different from its first argument.
Similarly, if
.code x
is an object with a
.code lambda-set
method, that method's return value becomes the return value of
.code replace
and must be taken into account.
.coNP Function @ replace
.synb
.mets (replace < sequence < replacement-sequence >> [ from <> [ to ]])
.mets (replace < sequence < replacement-sequence << index-list )
.syne
.desc
The
.meta replace
function modifies
.meta sequence
in the ways described below.
The operation is destructive: it may work "in place" by modifying
the original sequence. The caller should retain the return value
and stop relying on the original input sequence.
The return value of
.code replace
is the modified
version of
.metn sequence .
This may be the same object as
.meta sequence
or it may be a newly allocated object.
Note that the form:
.verb
(set seq (replace seq new fr to))
.brev
has the same effect on the variable
.code seq
as the form:
.verb
(set [seq fr..to] new)
.brev
except that the former
.code set
form returns the entire modified sequence, whereas the latter
returns the value of the
.code new
argument.
The
.code replace
function has two invocation styles, distinguished by the
type of the third argument. If the third argument is a sequence, then it
is deemed to be the
.meta index-list
parameter of the second form.
Otherwise, if the third argument is missing, or is not a list, then
it is deemed to be the
.meta from
argument of the first form.
The first form of the replace function replaces a contiguous subsequence of the
.meta sequence
with
.metn replacement-sequence .
The replaced subsequence may be empty,
in which case an insertion is performed. If
.meta replacement-sequence
is empty
(for example, the empty list
.codn nil ),
then a deletion is performed.
If the
.meta from
and
.meta to
arguments are omitted, their values default
to
.code 0
and
.code t
respectively.
The description of the dwim operator\(emin particular, the section
on Range Indexing\(emexplains the semantics of the range specification.
The second form of the replace function replaces a subsequence of
elements from
.meta sequence
given by
.metn index-list ,
with their counterparts
from
.metn replacement-sequence .
This form of the replace function does not insert
or delete; it simply overwrites elements. If
.meta replacement-sequence
and
.meta index-list
are of different lengths, then the shorter of the two determines
the maximum number of elements which are overwritten.
Whenever a negative value occurs in
.meta index-list
the length of
.meta sequence
is added to that value.
Furthermore, similar restrictions apply on
.meta index-list
as under the
select function. Namely, the replacement stops when an index value
in
.meta index-list
is encountered which is out of range for
.metn sequence .
furthermore, if
.meta sequence
is a list, then
.meta index-list
must
be monotonically increasing, after consideration of the
displacement of negative values.
If
.meta replacement-sequence
shares storage with the target range of
.metn sequence ,
or, in the case when that range is resized by the
.code replace
operation, shares storage with any portion of
.meta sequence
above that range, then the effect of
.code replace
on either object is unspecified.
If
.meta sequence
is a
.code carray
object, then
.code replace
behaves like
.codn carray-replace .
If
.meta sequence
is a
.code buf
object, then
.code replace
behaves like
.codn buf-replace .
If
.meta sequence
is a structure, then the structure must support the
.code lambda-set
method. The
.code replace
operation is translated into a call of the
.code lambda-set
method according to the following equivalences:
.verb
(replace o items from to)
<--> o.(lambda-set (rcons from to) items)
(replace o items index-list)
<--> o.(lambda-set index-list items)
.brev
Thus, the
.meta from
and
.meta to
arguments are converted to single range object,
whereas an
.meta index-list
is passed as-is.
It is an error if the
.code from
argument is a sequence, indicating an
.metn index-list ,
and a
.code to
argument is also given; the situation is diagnosed. If either
.code from
or
.code to
are omitted, the range object contains the
.code :
(colon) keyword symbol in the corresponding place:
.verb
(replace o items from)
<--> o.(lambda-set (rcons from :) items)
(replace o items : to)
<--> o.(lambda-set (rcons : to) items)
(replace o items)
<--> o.(lambda-set (rcons : :) items)
.brev
It is the responsibility of the object's
.code lambda-set
method to implement semantics consistent with the
description of
.codn replace .
.coNP Function @ take
.synb
.mets (take < count << sequence )
.syne
.desc
The
.code take
function returns
.meta sequence
with all except the first
.meta count
items removed.
If
.meta sequence
is a list, then
.code take
returns a lazy list which produces the first
.meta count
items of sequence.
For other kinds of sequences, including lazy strings,
.code take
works eagerly.
If
.meta count
exceeds the length of
.meta sequence
then a sequence is returned which has all the items.
This object may be
.meta sequence
itself, or a copy.
If
.meta count
is negative, it is treated as zero.
.coNP Functions @ take-while and @ take-until
.synb
.mets (take-while < predfun < sequence <> [ keyfun ])
.mets (take-until < predfun < sequence <> [ keyfun ])
.syne
.desc
The
.code take-while
and
.code take-until
functions return a prefix of
.meta sequence
whose items satisfy certain conditions.
The
.code take-while
function returns the longest prefix of
.meta sequence
whose elements, accessed through
.meta keyfun
satisfy the function
.metn predfun .
The
.meta keyfun
argument defaults to the identity function: the elements
of
.meta sequence
are examined themselves.
The
.code take-until
function returns the longest prefix of
.meta sequence
which consists of elements, accessed through
.metn keyfun ,
that do
.B not
satisfy
.meta predfun
followed by an element which does satisfy
.metn predfun .
If
.meta sequence
has no such prefix, then an empty sequence
is returned of the same kind as
.metn sequence .
If
.meta sequence
is a list, then these functions return a lazy list.
.coNP Function @ drop
.synb
.mets (drop < count << sequence )
.syne
.desc
The
.code drop
function returns
.meta sequence
with the first
.meta count
items removed.
If
.meta count
is negative, it is treated as zero.
If
.meta count
is zero, then
.meta sequence
is returned.
If
.meta count
exceeds the length of
.meta sequence
then an empty sequence is returned
of the same kind as
.metn sequence .
.coNP Functions @ drop-while and @ drop-until
.synb
.mets (drop-while < predfun < sequence <> [ keyfun ])
.mets (drop-until < predfun < sequence <> [ keyfun ])
.syne
.desc
The
.code drop-while
and
.code drop-until
functions return
.meta sequence
with a prefix of that sequence removed,
according to conditions involving
.meta predfun
and
.metn keyfun .
The
.code drop-while
function removes the longest prefix of
.meta sequence
whose elements, accessed through
.meta keyfun
satisfy the function
.metn predfun ,
and returns the remaining sequence.
The
.meta keyfun
argument defaults to the identity function: the elements
of
.meta sequence
are examined themselves.
The
.code drop-until
function removes the longest prefix of
.meta sequence
which consists of elements, accessed through
.metn keyfun ,
that do
.B not
satisfy
.meta predfun
followed by an element which does satisfy
.metn predfun .
A sequence of the remaining elements is
returned.
If
.meta sequence
has no such prefix, then a sequence
same as
.meta sequence
is returned, which may be
.meta sequence
itself or a copy.
.coNP Accessor @ last
.synb
.mets (last < sequence <> [ num ])
.mets (set (last < sequence <> [ num ]) << new-value)
.syne
.desc
The
.meta last
function returns a subsequence of
.meta sequence
consisting of the last
.meta num
of its elements, where
.meta num
defaults to 1.
If
.meta num
is zero or negative, then an empty sequence is returned.
If
.meta num
is positive, and greater than or equal to the length of sequence,
then sequence
.meta sequence
is returned.
If a
.code last
form is used as a place, then
.code sequence
must be a place. The following equivalence gives the semantics
of assignment to a
.codn last :
.verb
(set (last x n) v) <--> (set (sub x (- (max n 0)) t) v)
.brev
A
.code last
place is deletable. The semantics of deletion may be understood
in terms of the following equivalence:
.verb
(del (last x n)) <--> (del (sub x (- (max n 0)) t))
.brev
.coNP Accessor @ butlast
.synb
.mets (butlast < sequence <> [ num ])
.mets (set (butlast < sequence <> [ num ]) << new-value )
.syne
.desc
The
.code butlast
function returns the prefix of
.meta sequence
consisting of a copy of it, with the last
.meta num
items removed.
The parameter
.meta num
defaults to 1
if an argument is omitted.
If
.meta sequence
is empty, an empty sequence is returned.
If
.meta num
is zero or negative, then
.meta sequence
is returned.
If
.meta num
is positive, and meets or exceeds the length of
.metn sequence ,
then an empty sequence is returned.
If a
.code butlast
form is used as a place, then
.meta sequence
must itself be a place. The following equivalence gives the semantics
of assignment to a
.codn last :
.verb
(set (butlast x n) v) <--> (set (sub x 0 (- (max n 0))) v)
.brev
A
.code butlast
place is deletable. The semantics of deletion may be understood
in terms of the following equivalence:
.verb
(del (last x n)) <--> (del (sub x 0 (- (max n 0))))
.brev
Note: the \*(TL
.code take
function also computes the prefix of a list; however, it counts items
from the beginning, and provides lazy semantics which allow it
to work with infinite lists.
See also: the
.code butlastn
accessor, which operates on lists. That function has useful semantics for
improper lists and treats an atom as the terminator of a zero-length improper
list.
Dialect Note: a destructive function similar to Common Lisp's
.code nbutlast
isn't provided. Assignment to a
.code butlast
form is destructive; Common Lisp doesn't support
.code butlast
as a place.
.coNP Function @ ldiff
.synb
.mets (ldiff < sequence << tail-sequence )
.syne
.desc
The
.code ldiff
function is a somewhat generalized version of the same-named classic Lisp
function found in traditional Lisp dialects.
The
.code ldiff
function supports the original
.code ldiff
semantics when both inputs are lists. It determines whether the
.meta tail-sequence
list is a structural suffix of
.metn sequence ,
which is to say: is
.meta tail-sequence
one of the
.code cons
cells which comprise
.metn sequence ?
If so, then a list is returned consisting of all the items of
.meta sequence
before
.metn tail-sequence :
a copy of
.meta sequence
with the
.meta tail-sequence
part removed, and replaced by the
.code nil
terminator. If
.meta tail-sequence
is
.code nil
or the lists are unrelated, then
.meta sequence
is returned.
The \*(TL
.code ldiff
function supports the following additional semantics.
.RS
.IP 1.
The basic description of
.code ldiff
is extended to work with list-like sequences, not
merely lists; that is to say, objects which support the
.code car
method.
.IP 2.
If
.meta sequence
is any kind of sequence, and
.meta tail-sequence
is any kind of empty sequence, then
.meta sequence
is returned.
.IP 3.
If either argument is an atom that is not a sequence,
.code ldiff
returns
.metn sequence .
.IP 4.
If
.meta sequence
is a list-like sequence, and
.meta tail-sequence
isn't, then the terminating atom of
.meta sequence
is determined. This atom is compared using
.code equal
to the
.meta tail-sequence
object. If they are equal, then a proper list is
returned containing the items of
.meta sequence
excluding the terminating atom.
.IP 5.
If both arguments are vector-like sequences, then
.code ldiff
determines whether
.meta sequence
has a suffix which is
.code equal
to
.metn tail-sequence .
If this is the case, then a sequence is returned, of the same kind as
.metn sequence ,
consisting of the items of
.meta sequence
before that suffix.
If
.meta tail-sequence
is not
.code equal
to a suffix of
.metn sequence ,
then
.meta sequence
is returned.
.IP 6.
In all other cases,
.meta sequence
and
.meta tail-sequence
are compared with
.codn equal .
If the comparison is true,
.code nil
is returned, otherwise
.meta sequence
is returned.
.RE
.TP* Examples:
.verb
;;; unspecified: the compiler could make
;;; '(2 3) a suffix of '(1 2 3),
;;; or they could be separate objects.
(ldiff '(1 2 3) '(2 3)) -> either (1) or (1 2 3)
;; b is the (1 2) suffix of a, so the ldiff is (1)
(let* ((a '(1 2 3)) (b (cdr a)))
(ldiff a b))
-> (1)
;; Rule 5: strings and vector
(ldiff "abc" "bc") -> "a"
(ldiff "abc" nil) -> "abc"
(ldiff #(1 2 3) #(3)) -> #(1 2)
;; Rule 5: mixed vector kinds
(ldiff "abc" #(#\eb #\ec)) -> "abc"
;; Rule 6:
(ldiff #(1 2 3) '(3)) -> #(1 2 3)
;; Rule 4:
(ldiff '(1 2 3) #(3)) -> '(1 2 3)
(ldiff '(1 2 3 . #(3)) #(3)) -> '(1 2 3)
(ldiff '(1 2 3 . 4) #(3)) -> '(1 2 3 . 4)
;; Rule 6
(ldiff 1 2) -> 1
(ldiff 1 1) -> nil
.brev
.coNP Function @ search
.synb
.mets (search < haystack < needle >> [ testfun <> [ keyfun ]])
.syne
.desc
The
.code search
function determines whether the sequence
.meta needle
occurs as substring
within
.metn haystack ,
under the given comparison function
.meta testfun
and
key function
.metn keyfun .
If this is the case, then the zero-based position of
the leftmost occurrence of
.meta key
within
.meta haystack
is returned. Otherwise
.code nil
is returned to indicate that
.meta key
does not occur within
.metn haystack .
If
.meta key
is empty, then zero is always returned.
The arguments
.meta haystack
and
.meta needle
are sequences. They may not be hash tables.
If
.meta needle
is not empty, then it occurs at some position N within
.meta haystack
if
the first element of
.meta needle
matches the element at position N of
.metn haystack ,
the second element of
.meta needle
matches the element at position N+1 of
.meta haystack
and so forth, for all elements of
.metn needle .
A match between elements
is determined by passing each element through
.metn keyfun ,
and then comparing the resulting values using
.metn testfun .
If
.meta testfun
is supplied, it must be a function which can be
called with two arguments. If it is not supplied, it defaults to
.codn eql .
If
.meta keyfun
is supplied, it must be a function which can be called
with one argument. If it is not supplied, it defaults to
.codn identity .
.TP* Examples:
.verb
;; fails because 3.0 doesn't match 3
;; under the default eql function
[search #(1.0 3.0 4.0 7.0) '(3 4)] -> nil
;; occurrence found at position 1:
;; (3.0 4.0) matches (3 4) under =
[search #(1.0 3.0 4.0 7.0) '(3 4) =] -> 1
;; "even odd odd odd even" pattern
;; matches at position 2
[search #(1 1 2 3 5 7 8) '(2 1 1 1 2) : evenp] -> 2
;; Case insensitive string search
[search "abcd" "CD" : chr-toupper] -> 2
;; Case insensitive string search
;; using vector of characters as key
[search "abcd" #(#\eC #\eD) : chr-toupper] -> 2
.brev
.coNP Function @ contains
.synb
.mets (contains < needle < haystack >> [ testfun <> [ keyfun ]])
.syne
.desc
The syntax of the
.code contains
function differs from that of
.codn search :
that the
.meta needle
and
.meta haystack
arguments are reversed. The semantics is identical.
.coNP Function @ rsearch
.synb
.mets (rsearch < haystack < needle >> [ testfun <> [ keyfun ])
.syne
.desc
The
.code rsearch
function is like
.code search
except for two differences.
Firstly, if
.meta needle
matches
.meta haystack
in multiple places,
.code rsearch
returns the rightmost matching position rather than
the leftmost.
Secondly, if
.meta needle
is an empty sequence, then
.code rsearch
returns the length of
.codn haystack ,
thereby effectively declaring that the rightmost match for an empty
.meta needle
key occurs at the imaginary position past the element of
.metn haystack .
.coNP Functions @ ref and @ refset
.synb
.mets (ref < sequence << index )
.mets (refset < sequence < index << new-value )
.syne
.desc
The
.code ref
and
.code refset
functions perform array-like indexing into sequences, as well as
objects of type
.code buf
and
.codn carray .
If the
.meta sequence
parameter is a hash, then these functions perform
has retrieval and storage; in that case
.meta index
isn't restricted to an integer value.
If
.meta sequence
is a structure, it supports
.code ref
directly if it has a
.code lambda
method. The
.meta index
argument is passed to that method, and the resulting value is
returned.
If a structure lacks a
.code lambda
method, but has a
.code car
method, then
.code ref
treats it as a list, traversing the structure using
.cod3 car / cdr
operations. In the absence of support for these operations,
the function fails with an error exception.
Similarly, a structure supports
.code refset
directly if it has a
.code lambda-set
method. This gets called with
.meta index
and
.meta new-value
as arguments. Then
.meta new-value
is returned.
If a structure lacks a
.code lambda-set
method, then
.code refset
treats it as a list, traversing the structure using
.cod3 car / cdr
operations, and storing
.meta new-value
using
.codn rplaca .
In the absence of support for these operations,
the function fails with an error exception.
The
.code ref
function retrieves an element of
.metn sequence ,
whereas
.code refset
overwrites an
element of
.meta sequence
with a new value.
If
.meta sequence
is a sequence then
.meta index
argument must be an integer. The first element of the sequence
is indexed by zero. Negative values are permitted,
denoting backward indexing from the end of the sequence, such that
the last element is indexed by -1, the second last by -2 and so on.
See also the Range Indexing section under the
description of the
.code dwim
operator.
If
.meta sequence
is a list, then out-of-range indices, whether positive or negative,
are treated leniently by
.codn ref :
such accesses produce the value
.codn nil ,
rather than an error. For other sequence types, such accesses
are erroneous. For hashes, accesses to nonexistent elements
are treated leniently, and produce
.codn nil .
If
.meta sequence
is a search tree, then
.code ref
behaves like
.codn tree-lookup .
The
.code refset
function is not supported by search trees.
The
.code refset
function is strict for out-of-range indices over all sequences,
including lists. In the case of hashes, a
.code refset
of a nonexistent key creates the key.
The
.code refset
function returns
.codn new-value .
The following equivalences hold between
.code ref
and
.codn refset ,
and the DWIM bracket syntax, provided that
.meta idx
is a scalar index and
.meta sequence
is a sequence object, rather than a hash.
.verb
(ref seq idx) <--> [seq idx]
(refset seq idx new) <--> (set [seq idx] new)
.brev
The difference is that
.code ref
and
.code refset
are first class functions which
can be used in functional programming as higher order functions, whereas the
bracket notation is syntactic sugar, and
.code set
is an operator, not a function.
Therefore the brackets cannot replace all uses of
.code ref
and
.codn refset .
.coNP Function @ update
.synb
.mets (update < sequence << function )
.syne
.desc
The
.code update
function replaces each elements in
.meta sequence
in a hash table, with the result of
.meta function
being applied to that element value.
The
.meta sequence
is returned.
The
.meta sequence
may be a hash table. In that case,
.meta function
is invoked with each hash value, which is replaced with the function's return
value.
.coNP Functions @, remq @ remql and @ remqual
.synb
.mets (remq < object < sequence <> [ key-function ])
.mets (remql < object < sequence <> [ key-function ])
.mets (remqual < object < sequence <> [ key-function ])
.syne
.desc
The
.codn remq ,
.code remql
and
.code remqual
functions produce a new sequence based on
.metn sequence ,
removing the elements whose associated keys are
.codn eq ,
.code eql
or
.code equal
to
.metn object .
The input
.meta sequence
is unmodified, but the returned sequence may share substructure
with it. If no items are removed, it is possible that the return value
is
.meta sequence
itself.
If
.meta key-function
is omitted, then the element keys compared to
.meta object
are the elements themselves.
Otherwise,
.meta key-function
is applied to each element and the resulting value
is that element's key which is compared to
.metn object .
.coNP Functions @, remq* @ remql* and @ remqual*
.synb
.mets (remq* < object << sequence )
.mets (remql* < object << sequence )
.mets (remqual* < object << sequence )
.syne
.desc
The
.codn remq* ,
.code remql*
and
.code remqual*
functions are lazy analogs of
.codn remq ,
.code remql
and
.codn remqual .
Rather than computing the entire new sequence
prior to returning, these functions return a lazy list.
Caution: these functions can still get into infinite looping behavior.
For instance, in
.codn "(remql* 0 (repeat '(0)))" ,
.code remql
will keep consuming
the
.code 0
values coming out of the infinite list, looking for the first item that
does not have to be deleted, in order to instantiate the first lazy value.
.TP* Examples:
.verb
;; Return a list of all the natural numbers, excluding 13,
;; then take the first 100 of these.
;; If remql is used, it will loop until memory is exhausted,
;; because (range 1) is an infinite list.
[(remql* 13 (range 1)) 0..100]
.brev
.coNP Functions @, keepq @ keepql and @ keepqual
.synb
.mets (keepq < object < sequence <> [ key-function ])
.mets (keepql < object < sequence <> [ key-function ])
.mets (keepqual < object < sequence <> [ key-function ])
.syne
.desc
The
.codn keepq ,
.code keepql
and
.code keepqual
functions produce a new sequence based on
.metn sequence ,
removing the items whose keys are not
.codn eq ,
.code eql
or
.code equal
to
.metn object .
The input
.meta sequence
is unmodified, but the returned sequence may share substructure
with it. If no items are removed, it is possible that the return value
is
.meta sequence
itself.
The optional
.meta key-function
is applied to each element from the
.meta sequence
to convert it to a key which is compared to
.metn object .
If
.meta key-function
is omitted, then each element itself of
.meta sequence
is compared to
.metn object .
.coNP Functions @, remove-if @, keep-if @, separate @ remove-if* and @ keep-if*
.synb
.mets (remove-if < predicate-function < sequence <> [ key-function ])
.mets (keep-if < predicate-function < sequence <> [ key-function ])
.mets (separate < predicate-function < sequence <> [ key-function ])
.mets (remove-if* < predicate-function < sequence <> [ key-function ])
.mets (keep-if* < predicate-function < sequence <> [ key-function ])
.syne
.desc
The
.code remove-if
function produces a sequence whose contents are those of
.meta sequence
but with those elements removed which satisfy
.metn predicate-function .
Those elements which are not removed appear in the same order.
The result sequence may share substructure with the input sequence,
and may even be the same sequence object if no items are removed.
The optional
.meta key-function
specifies how each element from the
.meta sequence
is transformed to an argument to
.metn predicate-function .
If this argument is omitted
then the predicate function is applied to the elements directly, a behavior
which is identical to
.meta key-function
being
.codn "(fun identity)" .
The
.code keep-if
function is exactly like
.codn remove-if ,
except the sense of
the predicate is inverted. The function
.code keep-if
retains those items
which
.code remove-if
will delete, and removes those that
.code remove-if
will preserve.
The
.code separate
function combines
.code keep-if
and
.code remove-if
into one,
returning a list of two elements whose
.code car
and
.code cadr
are the result of calling
.code keep-if
and
.codn remove-if ,
respectively,
on
.meta sequence
(with the
.meta predicate-function
and
.meta key-function
arguments passed through).
One of the two elements may share substructure with the input sequence,
and may even be the same sequence object if all items are either kept or
removed (in which case the other element will be
.codn nil ).
Note: the
.code separate
function may be understood in terms of the following reference implementation:
.verb
(defun separate (pred seq : (keyfun :))
[(juxt (op keep-if pred @1 keyfun)
(op remove-if pred @1 keyfun))
seq])
.brev
The
.code remove-if*
and
.code keep-if*
functions are like
.code remove-if
and
.codn keep-if ,
but produce lazy lists.
.TP* Examples:
.verb
;; remove any element numerically equal to 3.
(remove-if (op = 3) '(1 2 3 4 3.0 5)) -> (1 2 4 5)
;; remove those pairs whose first element begins with "abc"
[remove-if (op equal [@1 0..3] "abc")
'(("abcd" 4) ("defg" 5))
car]
-> (("defg" 5))
;; equivalent, without test function
(remove-if (op equal [(car @1) 0..3] "abc")
'(("abcd" 4) ("defg" 5)))
-> (("defg" 5))
.brev
.coNP Functions @, countqual @ countql and @ countq
.synb
.mets (countq < object << iterable )
.mets (countql < object << iterable )
.mets (countqual < object << iterable )
.syne
.desc
The
.codn countq ,
.code countql
and
.code countqual
functions count the number of objects
in
.meta iterable
which are
.codn eq ,
.code eql
or
.code equal
to
.metn object ,
and return the count.
.coNP Function @ count-if
.synb
.mets (count-if < predicate-function < iterable <> [ key-function ])
.syne
.desc
The
.code count-if
function counts the number of elements of
.meta iterable
which satisfy
.meta predicate-function
and returns the count.
The optional
.meta key-function
specifies how each element from
.meta iterable
is transformed to an argument to
.metn predicate-function .
If this argument is omitted
then the predicate function is applied to the elements directly, a behavior
which is identical to
.meta key-function
being
.codn "(fun identity)" .
.coNP Functions @, posq @ posql and @ posqual
.synb
.mets (posq < object << sequence )
.mets (posql < object << sequence )
.mets (posqual < object << sequence )
.syne
.desc
The
.codn posq ,
.code posql
and
.code posqual
functions return the zero-based position of the
first item in
.meta sequence
which is, respectively,
.codn eq ,
.code eql
or
.code equal
to
.metn object .
.coNP Functions @ pos and @ pos-if
.synb
.mets (pos < key < sequence >> [ testfun <> [ keyfun ]])
.mets (pos-if < predfun < sequence <> [ keyfun ])
.syne
.desc
The
.code pos
and
.code pos-if
functions search through
.meta sequence
for an item which matches
.metn key ,
or satisfies the predicate function
.metn predfun ,
respectively.
They return the zero-based position of the matching item.
The
.meta keyfun
argument specifies a function which is applied to the elements
of
.meta sequence
to produce the comparison key. If this argument is omitted,
then the untransformed elements of
.meta sequence
are examined.
The
.code pos
function's
.meta testfun
argument specifies the test function which
is used to compare the comparison keys from
.meta sequence
to
.metn key .
If this argument is omitted, then the
.code equal
function is used.
The position of the first element
.meta sequence
whose comparison key (as
retrieved by
.metn keyfun )
matches the search (under
.metn testfun )
is
returned. If no such element is found,
.code nil
is returned.
The
.code pos-if
function's
.meta predfun
argument specifies a predicate function
which is applied to the successive comparison keys taken from
.meta sequence
by applying
.meta keyfun
to successive elements. The position of
the first element for which
.meta predfun
yields true is returned. If
no such element is found,
.code nil
is returned.
.coNP Functions @, rposq @, rposql @, rposqual @ rpos and @ rpos-if
.synb
.mets (rposq < object << sequence )
.mets (rposql < object << sequence )
.mets (rposqual < object << sequence )
.mets (rpos < key < sequence >> [ testfun <> [ keyfun ]])
.mets (rpos-if < predfun < sequence <> [ keyfun ])
.syne
.desc
These functions are counterparts of
.codn rposq ,
.codn rposql ,
.codn rposqual ,
.code rpos
and
.code rpos-if
which report position of the rightmost matching item,
rather than the leftmost.
.coNP Functions @ pos-max and @ pos-min
.synb
.mets (pos-max < sequence >> [ testfun <> [ keyfun ]])
.mets (pos-min < sequence >> [ testfun <> [ keyfun ]])
.syne
.desc
The
.code pos-min
and
.code pos-max
functions implement exactly the same algorithm; they
differ only in their defaulting behavior with regard to the
.meta testfun
argument. If
.meta testfun
is not given, then the
.code pos-max
function defaults
.meta testfun
to the
.code greater
function, whereas
.code pos-min
defaults it to the
.code less
function.
If
.meta sequence
is empty, both functions return
.codn nil .
Without a
.meta testfun
argument, the
.code pos-max
function finds the zero-based
position index of the numerically maximum value occurring in
.metn sequence ,
whereas
.code pos-min
without a
.meta testfun
argument finds the index of the minimum
value.
If a
.meta testfun
argument is given, the two functions are equivalent.
The
.meta testfun
function must be callable with two arguments.
If
.meta testfun
behaves like a greater-than comparison, then
.code pos-max
and
.code pos-min
return the index of the maximum element. If
.meta testfun
behaves like a
.code less-than
comparison, then the functions return
the index of the minimum element.
The
.meta keyfun
argument defaults to the
.code identity
function. Each element
from
.meta sequence
is passed through this one-argument function, and
the resulting value is used in its place.
If a sequence contains multiple equivalent maxima,
whether the position of the leftmost or rightmost such maximum is reported
depends on whether
.meta testfun
compares for strict inequality, or whether it reports true for
equal arguments also. Under the default
.metn testfun ,
which is
.codn less ,
the
.code pos-max
function will return the position leftmost of a duplicate set of maximum
elements. To find the rightmost of the maxima, the
.code lequal
function can be substituted. Analogous reasoning applies to other
test functions.
.coNP Function @ subst
.synb
.mets (subst < old < new < seq >> [ testfun <> [ keyfun ]])
.syne
.desc
The
.code subst
function returns a sequence of the same type as
.meta seq
in which elements of
.meta seq
which match the
.meta old
object have been replaced with the
.meta new
object.
To form the comparison keys, the elements of
.meta seq
are projected through the
.meta testfun
function, which defaults to
.codn identity ,
so the items themselves are used as keys by default.
Keys are compared to the
.meta old
value using
.metn testfun ,
which defaults to
.codn equal .
.TP* Examples:
.verb
(subst "brown" "black" #("how" "now" "brown" "cow"))
-> #("how" "now" "black" "cow"))
;; elements are converted to lower case to form keys
[subst "brown" "black"
#("how" "now" "BROWN" "cow") : downcase-str]
-> #("how" "now" "black" "cow")
;; using < instead of equality, replace elements
;; greater than 5 with 0
[subst 5 0 '(1 2 3 4 5 6 7 8 9 10) <] (1 2 3 4 5 0 0 0 0 0))
.brev
.coNP Functions @, subq @ subql and @ subqual
.synb
.mets (subq < old < new << sequence )
.mets (subql < old < new << sequence )
.mets (subqual < old < new << sequence )
.syne
.desc
The
.codn subq ,
.code subql
and
.code subqual
functions return a sequence of the same kind as
.meta sequence
in which elements matching the
.meta old
object are replaced by
.meta new
object.
The matching elements are identified by comparing with
.meta old
using, respectively, the functions
.codn eq ,
.codn eql ,
and
.codn equal .
.TP* Examples:
.verb
(subq #\eb #\ez "abc") -> "azc"
(subql 1 3 #(0 1 2)) -> #(0 3 2)
(subqual "are" "do" '#"how are you")
-> ("how" "do" "you")
.brev
.coNP Function @ mismatch
.synb
.mets (mismatch < left-seq < right-seq >> [ testfun <> [ keyfun ]])
.syne
.desc
The
.code mismatch
function compares corresponding elements from the sequences
.meta left-seq
and
.metn right-seq ,
returning the position at which the first mismatch occurs.
If the sequences are of the same length, and their corresponding
elements are the same, then
.code nil
is returned.
If one sequence is shorter than the other, and matches a prefix
of the other, then the mismatching position returned is one position
after the last element of the shorter sequence, the same value
as its length. An empty sequence is a prefix of every sequence.
The
.meta keyfun
argument defaults to the
.code identity
function. Each element
from
.meta sequence
is passed to
.meta keyfun
and the resulting value is used in its place.
After being converted through
.metn keyfun ,
items are then compared using
.metn testfun ,
which must accept two arguments, and defaults to
.codn equal .
.coNP Function @ where
.synb
.mets (where < function << iterable )
.syne
.desc
If
.meta iterable
is a sequence, the
.code where
function returns
a lazy list of the numeric indices of those of its elements which satisfy
.metn function .
The numeric indices appear in increasing order.
If
.meta iterable
is a hash, the following special behavior applies:
.code where
returns a lazy list of
of keys which have values which satisfy
.metn function .
These keys are not subject to an order.
.meta function
must be a function that can be called with one argument.
For each element of
.metn iterable ,
.meta function
is called with that element
as an argument. If a
.cod2 non- nil
value is returned, then the zero-based index of
that element is added to a list. Finally, the list is returned.
.coNP Function @ rmismatch
.synb
.mets (rmismatch < left-seq < right-seq >> [ testfun <> [ keyfun ]])
.syne
.desc
Similarly to
.codn mismatch ,
the
.code rmismatch
function compares corresponding elements from the sequences
.meta left-seq
and
.metn right-seq ,
returning the position at which the first mismatch occurs.
All of the arguments have the same semantics as that of
.codn mismatch .
Unlike
.codn mismatch ,
.code rmismatch
compares the sequences right-to-left, finding the suffix
which they have in common, rather than prefix.
If the sequences match, then
.code nil
is returned. Otherwise, a negative index is returned giving the
mismatching position, regarded from the end. If the sequences
match only in the rightmost element, then -1 is returned. If they
match in two elements then -2 and so forth.
.coNP Functions @ starts-with and @ ends-with
.synb
.mets (starts-with < short-seq < long-seq >> [ testfun <> [ keyfun ]])
.mets (ends-with < short-seq < long-seq >> [ testfun <> [ keyfun ]])
.syne
.desc
The
.code starts-with
and
.code ends-with
functions compare corresponding elements from sequences
.meta short-seq
and
.metn long-seq .
The
.code starts-with
function returns
.code t
if
.meta short-seq
is prefix of
.metn long-seq ;
otherwise, it returns
.codn nil .
The
.code ends-with
function returns
.code t
if
.meta short-seq
is suffix of
.metn long-seq ;
otherwise, it returns
.codn nil .
Element from both sequences are mapped to comparison keys using
.metn keyfun ,
which defaults to
.codn identity .
Comparison keys are compared using
.meta testfun
which defaults to
.codn equal .
.coNP Function @ select
.synb
.mets (select < sequence >> { index-list | << function })
.syne
.desc
The
.code select
function returns a sequence, of the same kind as
.metn sequence ,
which consists of those elements of
.meta sequence
which are identified by
the indices in
.metn index-list ,
which may be a list or a vector.
If
.meta function
is given instead of
.metn index-list ,
then
.meta function
is invoked with
.meta sequence
as its argument. The return value is then taken as
if it were the
.meta index-list
argument .
If
.meta sequence
is a sequence, then
.meta index-list
consists of numeric
indices. The length of the sequence, as reported by the
.code length
function, is added to every
.meta index-list
value which is negative.
The
.code select
function stops collecting values upon encountering an index value which is
greater than or equal to the length of the sequence.
(Rationale: without
this strict behavior,
.code select
would not be able to terminate if
.meta index-list
is infinite.)
If
.meta sequence
is, more specifically, a list-like sequence, then
.meta index-list
must contain monotonically increasing
numeric values, even if no value is out of range, since the
.code select
function
makes a single pass through the list based on the assumption that indices
are ordered. (Rationale: optimization.)
This requirement for monotonicity applies to the values which
result after negative indices are displaced by the sequence length
Also, in this list-like sequence case, values taken from
.meta index-list
which are still negative after being displaced by the sequence length are
ignored.
If
.meta sequence
is a hash, then
.meta index-list
is a list of keys. A new hash is
returned which contains those elements of
.meta sequence
whose keys appear
in
.metn index-list .
All of
.meta index-list
is processed, even if it contains
keys which are not in
.metn sequence .
The nonexistent keys are ignored.
The
.code select
function also supports objects of type
.codn carray ,
in a manner similar to vectors. The indicated elements are extracted
from the input sequence, and a new
.code carray
is returned whose storage is initialized by converting the extracted
values back to the foreign representation.
.coNP Function @ reject
.synb
.mets (reject < sequence >> { index-list | << function })
.syne
.desc
The
.code reject
function returns a sequence, of the same kind as
.metn sequence ,
which consists of all those elements of
.meta sequence
which are not identified by the indices in
.metn index-list ,
which may be a list or a vector.
If
.meta function
is given instead of
.metn index-list ,
then
.meta function
is invoked with
.meta sequence
as its argument. The return value is then taken as
if it were the
.meta index-list
argument .
If
.code sequence
is a hash, then
.meta index-list
represents a list of keys. The
.code reject
function returns a duplicate of the hash, in which
the keys specified in
.meta index-list
do not appear.
Otherwise if
.meta sequence
is a vector-like sequence, then the behavior of
.code reject
may be understood by the following equivalence:
.verb
(reject seq idx) --> (make-like
[apply append (split* seq idx)]
seq)
.brev
where it is to be understood that
.meta seq
is evaluated only once.
If
.meta sequence
is a list, then, similarly, the following equivalence applies:
.verb
(reject seq idx) --> (make-like
[apply append* (split* seq idx)]
seq)
.brev
The input sequence is split into pieces at the indicated indices, such that
the elements at the indices are removed and do not appear in the pieces. The
pieces are then appended together in order, and the resulting list is coerced
into the same type of sequence as the input sequence.
.coNP Function @ relate
.synb
.mets (relate < domain-seq < range-seq <> [ default-val ])
.syne
.desc
The
.code relate
function returns a one-argument function which implements the relation formed
by mapping the elements of
.meta domain-seq
to the positionally corresponding elements of
.metn range-seq .
That is to say, the function searches through the sequence
.meta domain-seq
to determine the position where its argument occurs, using
.code equal
as the comparison function.
Then it returns the element from that position in the
.meta range-seq
sequence. This returned function is called the
.IR "relation function" .
If the relation function's argument is not found in
.metn domain-seq ,
then the behavior depends on the optional parameter
.metn default-val .
If an argument is given for
.metn default-val ,
then the relation function returns that value.
Otherwise, the relation function returns its argument.
Note: the
.code relate
function may be understood in terms of the following equivalences:
.verb
(relate d r) <--> (lambda (arg)
(iflet ((p (posqual arg d)))
[r p]
arg))
(relate d r v) <--> (lambda (arg)
(iflet ((p (posqual arg d)))
[r p]
v))
.brev
Note:
.code relate
may return a hash table instead of a function, if such an object
can satisfy the semantics required by the arguments.
.TP* Examples:
.verb
(mapcar (relate "_" "-") "foo_bar") -> "foo-bar"
(mapcar (relate "0123456789" "ABCDEFGHIJ" "X") "139D-345")
-> "BJDXXDEF"
(mapcar (relate '(nil) '(0)) '(nil 1 2 nil 4)) -> (0 1 2 0 4)
.brev
.coNP Function @ in
.synb
.mets (in < sequence < key >> [ testfun <> [ keyfun ]])
.mets (in < hash << key )
.syne
.desc
The
.code in
function tests whether
.meta key
is found inside
.meta sequence
or
.metn hash .
If the
.meta testfun
argument is specified, it specifies the function
which is used to comparison keys from the sequence
to
.metn key .
Otherwise the
.code equal
function is used.
If the
.meta keyfun
argument is specified, it specifies a function which
is applied to the elements of
.meta sequence
to produce the comparison keys. Without this
argument, the elements themselves are taken
as the comparison keys.
If the object being searched is a hash, then if neither of the arguments
.meta keyfun
nor
.meta testfun
is specified,
.code in
performs a hash lookup for
.codn key ,
returning
.code t
if the key is found,
.code nil
otherwise.
If either of
.meta keyfun
or
.meta testfun
is specified, then
.code in
performs an exhaustive search of the hash table, as if it were
a sequence of
.code cons
cells whose
.code car
fields are keys, and whose
.code cdr
keys are values. Thus to search by key, the
.code car
function must be specified as
.metn keyfun .
The
.code in
function returns
.code t
if it finds
.meta key
in
.meta sequence
or
.metn hash ,
otherwise
.codn nil .
.coNP Function @ partition
.synb
.mets (partition < sequence >> { index-list | < index | << function })
.syne
.desc
If
.meta sequence
is empty, then
.code partition
returns an empty list, and the
second argument is ignored; if it is
.metn function ,
it is not called.
Otherwise,
.code partition
returns a lazy list of partitions of
.metn sequence .
Partitions are consecutive, non-overlapping, nonempty substrings of
.metn sequence ,
of the same kind as
.metn sequence ,
such that if these substrings are catenated together in their order
of appearance, a sequence
.code equal
to the original is produced.
If the second argument is of the form
.metn index-list ,
or if an
.meta index-list
was produced from the
.meta index
or
.meta function
arguments, each value in that list must be an integer. Each integer
value which is nonnegative specifies the index position
given by its value. Each integer value which is negative
specifies an index position given by adding the length of
.meta sequence
to its value. The sequence index positions thus denoted by
.meta index-list
shall be strictly nondecreasing. Each successive element
is expected to designate an index position at least as high
as all previous elements, otherwise the behavior is unspecified.
Leading index positions which are (still) negative, or zero, are effectively
ignored.
If
.meta index-list
is empty then a one-element list containing the entire
.meta sequence
is returned.
If
.meta index-list
is an infinite lazy list, the function shall terminate if that
list eventually produces an index position which is greater than or equal to
the length of
.metn sequence .
If the second argument is a function, then this function is applied
to
.metn sequence ,
and the return value of this call is then used in place of the
second argument, which must be a single index value, which is then
taken as if it were the
.meta index
argument, or else a list of indices, which are taken as the
.meta index-list
argument.
If the second argument is an atom other than a function, it is assumed to be
an integer index, and is turned into an
.meta index-list
of one element.
After the
.meta index-list
is obtained as an argument, or determined from the
.meta index
or
.meta function
arguments, the
.code partition
function then divides
.meta sequence
according to the indices given by that list.
The first partition begins with the first element of
.metn sequence .
The second partition begins at the first position in
.metn index-list ,
and so on. Indices beyond the length of the sequence are ignored,
as are indices less than or equal to zero.
.TP* Examples:
.verb
(partition '(1 2 3) 1) -> ((1) (2 3))
;; split the string where there is a "b"
(partition "abcbcbd" (op where (op eql #\eb))) -> ("a" "bc"
"bc" "bd")
.brev
.coNP Functions @ split and @ split*
.synb
.mets (split < sequence >> { index-list | < index | << function })
.mets (split* < sequence >> { index-list | < index | << function })
.syne
.desc
If
.meta sequence
is empty, then both
.code split
and
.code split*
return an empty list, and the
second argument is ignored; if it is
.metn function ,
it is not called.
Otherwise,
.code split
returns a lazy list of pieces of
.metn sequence :
consecutive, non-overlapping, possibly empty substrings of
.metn sequence ,
of the same kind as
.metn sequence .
A catenation of these pieces in the order they appear would produce
a sequence that is
.code equal
to the original sequence.
The
.code split*
function differs from
.code split
in that the elements indicated by the split indices are removed.
The
.metn index ,
.metn index-list ,
and
.meta function
arguments are subject to the same restrictions and treatment
as the corresponding arguments of the
.code partition
function, with the following difference: the index positions indicated by
.code index-list
are required to be strictly increasing, rather than nondecreasing.
If the second argument is of the form
.metn index-list ,
or if an
.meta index-list
was produced from the
.meta index
or
.meta function
arguments, then the
.code split
function divides
.meta sequence
according to the indices indicated in the list. The first piece always begins
with the first element of
.metn sequence .
Each subsequent piece begins with the position indicated by
an element of
.metn index-list .
Negative indices are ignored.
If
.meta index-list
includes index zero,
then an empty first piece is generated.
If
.meta index-list
includes an index greater than or equal to the length of
.meta sequence
(equivalently, an index beyond the last element of the sequence)
then an additional empty last piece is generated.
The length of
.meta sequence
is added to any negative indices. An index which is still negative
after being thus displaced is discarded.
Note: the principal difference between
.code split
and
.code partition
is that
.code partition
does not produce empty pieces.
.TP* Examples:
.verb
(split '(1 2 3) 1) -> ((1) (2 3))
(split "abc" 0) -> ("" "abc")
(split "abc" 3) -> ("abc" "")
(split "abc" 1) -> ("a" "bc")
(split "abc" '(0 1 2 3)) -> ("" "a" "b" "c" "")
(split "abc" '(1 2)) -> ("a" "b" "c")
(split "abc" '(-1 1 2 15)) -> ("a" "b" "c")
;; triple split at makes two additional empty pieces
(split "abc" '(1 1 1)) -> ("a" "" "" "bc")
(split* "abc" 0) -> ("" "bc") ;; "a" is removed
;; all characters removed
(split* "abc" '(0 1 2)) -> ("" "" "" "")
.brev
.coNP Function @ partition*
.synb
.mets (partition* < sequence >> { index-list | < index | << function })
.syne
.desc
If
.meta sequence
is empty, then
.code partition*
returns an empty list, and the
second argument is ignored; if it is
.metn function ,
it is not called.
The
.metn index ,
.metn index-list ,
and
.meta function
arguments are subject to the same restrictions and treatment
as the corresponding arguments of the
.code partition
function, with the following difference: the index positions indicated by
.code index-list
are required to be strictly increasing, rather than nondecreasing.
If the second argument is of the form
.metn index-list ,
then
.code partition*
produces a
lazy list of pieces taken from
.metn sequence .
The pieces are formed by deleting from
.meta sequence
the elements at the positions given
in
.metn index-list ,
such that the pieces are the remaining nonempty substrings from
between the deleted elements, maintaining their order.
If
.meta index-list
is empty then a one-element list containing the entire
.meta sequence
is returned.
.TP* Examples:
.verb
(partition* '(1 2 3 4 5) '(0 2 4)) -> ((2) (4))
(partition* "abcd" '(0 3)) -> "bc"
(partition* "abcd" '(0 1 2 3)) -> nil
.brev
.coNP Functions @, find @ find-if and @ find-true
.synb
.mets (find < key < sequence >> [ testfun <> [ keyfun ]])
.mets (find-if < predfun >> { sequence | << hash } <> [ keyfun ])
.mets (find-true < predfun >> { sequence | << hash } <> [ keyfun ])
.syne
.desc
The
.code find
and
.code find-if
functions search through a sequence for an item which
matches a key, or satisfies a predicate function, respectively.
The
.code find-true
function is a variant of
.code find-if
which returns the value of the predicate function instead
of the item.
The
.meta keyfun
argument specifies a function which is applied to the elements
of
.meta sequence
to produce the comparison key. If this argument is omitted,
then the untransformed elements of the
.meta sequence
are searched.
The
.code find
function's
.meta testfun
argument specifies the test function which
is used to compare the comparison keys from
.meta sequence
to the search key.
If this argument is omitted, then the
.code equal
function is used.
The first element from the list whose comparison key (as retrieved by
.metn keyfun )
matches the search (under
.metn testfun )
is returned. If no such element is found,
.code nil
is returned.
The
.code find-if
function's
.meta predfun
argument specifies a predicate function
which is applied to the successive comparison keys pulled from the list
by applying
.meta keyfun
to successive elements. The first element
for which
.meta predfun
yields true is returned. If no such
element is found,
.code nil
is returned.
In the case of
.codn find-if ,
a hash table may be specified instead of a sequence.
The
.meta hash
is treated as if it were a sequence of hash key and hash
value pairs represented as cons cells, the
.code car
slots of which are the hash keys, and the
.code cdr
of which are the hash values. If the caller doesn't specify a
.meta keyfun
then these cells are taken as their keys.
The
.code find-true
function's argument conventions and search semantics are identical to those of
.codn find-if ,
but the return value is different. Instead of returning the found item,
.code find-true
returns the value which
.meta predfun
returned for the found item's key.
.coNP Functions @ rfind and @ rfind-if
.synb
.mets (rfind < key < sequence >> [ testfun <> [ keyfun ]])
.mets (rfind-if < predfun >> { sequence | << hash } <> [ keyfun ])
.syne
.desc
The
.code rfind
and
.code rfind-if
functions are almost exactly like
.code find
and
.code find-if
except that if there are multiple matches for
.meta key
in
.metn sequence ,
they return the rightmost element rather than
the leftmost.
In the case of
.code rfind-if
when a
.meta hash
is specified instead of a
.metn sequence ,
the function searches through the hash entries in the same order as
.codn find-if ,
but finds the last match rather than the first.
Note: hashes are inherently not ordered; the relative order of items in
a hash table can change when other items are inserted or deleted.
.coNP Functions @ find-max and @ find-min
.synb
.mets (find-max >> { sequence | << hash } >> [ testfun <> [ keyfun ]])
.mets (find-min >> { sequence | << hash } >> [ testfun <> [ keyfun ]])
.syne
.desc
The
.code find-min
and
.code find-max
function implement exactly the same algorithm; they
differ only in their defaulting behavior with regard to the
.meta testfun
argument. If
.meta testfun
is not given, then the
.code find-max
function defaults it to the
.code greater
function, whereas
.code find-min
defaults it to the
.code less
function.
Without a
.meta testfun
argument, the
.code find-max
function finds the numerically
maximum value occurring in
.metn sequence ,
whereas
.code pos-min
without a
.meta testfun
argument finds the minimum value.
If a
.meta testfun
argument is given, the two functions are equivalent.
The
.meta testfun
function must be callable with two arguments.
If
.meta testfun
behaves like a greater-than comparison, then
.code find-max
and
.code find-min
both return the maximum element. If
.meta testfun
behaves like a less-than comparison, then the functions return
the minimum element.
The
.meta keyfun
argument defaults to the
.code identity
function. Each element
from
.meta sequence
is passed through this one-argument function, and
the resulting value is used in its place for the purposes of the
comparison. However, the original element is returned.
A hash table may be specified instead of a sequence.
The
.meta hash
is treated as if it were a sequence of hash key and hash
value pairs represented as cons cells, the
.code car
slots of which are the hash keys, and the
.code cdr
of which are the hash values. If the caller doesn't specify a
.meta keyfun
then these cells are taken as their keys. To find the hash
table's key-value cell with the maximum key, the
.code car
function can be specified as
.metn keyfun .
To find the entry holding the maximum value, the
.code cdr
function can be specified.
If there are multiple equivalent maxima, then under the default
.metn testfun ,
that being
.codn less ,
the leftmost one is reported. See the notes under
.code pos-max
regarding duplicate maxima.
.coNP Functions @, uni @, isec @ diff and @ symdiff
.synb
.mets (uni < iter1 < iter1 >> [ testfun <> [ keyfun ]])
.mets (isec < iter1 < iter1 >> [ testfun <> [ keyfun ]])
.mets (diff < iter1 < iter1 >> [ testfun <> [ keyfun ]])
.mets (symdiff < iter1 < iter2 >> [ testfun <> [ keyfun ]])
.syne
.desc
The functions
.codn uni ,
.codn isec ,
.code diff
and
.code symdiff
treat the sequences
.meta iter1
and
.meta iter2
as if they were sets.
They, respectively, compute the set union, set intersection,
set difference and symmetric difference of
.meta iter1
and
.metn iter2 ,
returning a new sequence.
The arguments
.meta iter1
and
.meta iter2
need not be of the same kind. They may be hash tables.
The returned sequence is of the same kind as
.metn iter1 .
If
.meta iter1
is a hash table, the returned sequence is a list.
For the purposes of these functions, an input which is a hash table
is considered as if it were a sequence of hash key and hash
value pairs represented as cons cells, the
.code car
slots of which are the hash keys, and the
.code cdr
of which are the hash values. This means that if no
.meta keyfun
is specified, these pairs are taken as keys.
Since the input sequences are defined as representing sets, they are assumed
not to contain duplicate elements. These functions are not required, but may,
de-duplicate the sequences.
The union sequence produced by
.code uni
contains all of the elements which occur in both
.meta iter1
and
.metn iter2 .
If a given element occurs exactly once only in
.meta iter1
or exactly once only in
.metn iter2 ,
or exactly once in both sequences, then it occurs exactly once in the union
sequence. If a given element occurs at least once in either
.metn iter1 ,
.meta iter2
or both, then it occurs at least once in the union sequence.
The intersection sequence produced by
.code isec
contains all of the elements which occur in both
.meta iter1
and
.metn iter2 .
If a given element occurs exactly once in
.meta iter1
and exactly once in
.metn iter2 ,
then in occurs exactly once in the intersection sequence.
If a given element occurs at least once in
.meta iter1
and at least once in
.metn iter2 ,
then in occurs at least once in the intersection sequence.
The difference sequence produced by
.code diff
contains all of the elements which occur in
.meta iter1
but do not occur in
.metn iter2 .
If an element occurs exactly once in
.meta iter1
and does not occur in
.metn iter2 ,
then it occurs exactly once in the difference sequence.
If an element occurs at least once in
.meta iter1
and does not occur in
.metn iter2 ,
then it occurs at least once in the difference sequence.
If an element occurs at least once in
.metn iter2 ,
then it does not occur in the difference sequence.
The symmetric difference sequence produced by
.code symdiff
contains all of the elements of
.meta iter1
which do not occur in
.meta iter2
and vice versa:
it also contains all of the elements of
.meta iter2
which do not occur in
.metn iter1 .
Element equivalence is determined by a combination of
.meta testfun
and
.metn keyfun .
Elements are compared pairwise, and each element of a pair is passed through
.meta keyfun
function to produce a comparison value. The comparison values
are compared using
.metn testfun .
If
.meta keyfun
is omitted, then the
untransformed elements themselves are compared, and if
.meta testfun
is omitted,
then the
.code equal
function is used.
Note: a function similar to
.code diff
named
.code set-diff
exists. This became deprecated starting in \*(TX 184.
For the
.code set-diff
function, the requirement was specified to preserve the original
order of items from
.meta iter1
that survive into the output sequence.
This requirement is not documented for the
.code diff
function, but is de facto
honored by the implementation for at as long as the
.code set-diff
synonym continues to be available.
The
.code set-diff
function doesn't support hash tables and is inefficient for vectors and
strings.
Note: these functions are not efficient for the processing of hash tables,
even when both inputs are hashes, the
.meta keyfun
argument is
.codn car ,
and
.meta testfun
matches the equality used by both hash-table inputs.
If applicable, the operations
.codn hash-uni ,
.code hash-isec
and
.code hash-diff
should be used instead.
.coNP Functions @, mapcar @, mappend @ mapcar* and @ mappend*
.synb
.mets (mapcar < function << iterable *)
.mets (mappend < function << iterable *)
.mets (mapcar* < function << iterable *)
.mets (mappend* < function << iterable *)
.syne
.desc
When given only one argument, the
.code mapcar
function returns
.codn nil .
.meta function
is never called.
When given two arguments, the
.code mapcar
function applies
.meta function
to each elements of
.meta iterable
and returns a sequence of the resulting values
in the same order as the original values.
The returned sequence is the same kind as
.metn iterable ,
if possible. If the accumulated values cannot be
elements of that type of sequence, then a list is returned.
When additional sequences are given as arguments, this filtering behavior is
generalized in the following way:
.code mapcar
traverses the sequences in parallel,
taking a value from each sequence as an argument to the function. If there
are two lists,
.meta function
is called with two arguments and so forth.
The traversal is limited by the length of the shortest sequence.
The return values of the function are collected into a new sequence which is
returned. The returned sequence is of the same kind as the leftmost
input sequence, unless the accumulated values cannot be elements of that type of
sequence, in which case a list is returned.
The
.code mappend
function works like
.codn mapcar ,
with the following difference.
Rather than accumulating the values returned by the function into a sequence,
mappend expects the items returned by the function to be sequences which
are catenated with
.codn append ,
and the resulting sequence is returned. The returned sequence is of the same
kind as the leftmost input sequence, unless the values cannot be elements
of that type of sequence, in which case a list is returned.
The
.code mapcar*
and
.code mappend*
functions work like
.code mapcar
and
.codn mappend ,
respectively.
However, they return lazy lists rather than generating the entire
output list prior to returning.
.TP* Caveats:
Like
.codn mappend ,
.code mappend*
must "consume" empty lists. For instance,
if the function being mapped puts out a sequence of
.codn nil s,
then the result must be the empty list
.codn nil ,
because
.code "(append nil nil nil nil ...)"
is
.codn nil .
But suppose that
.code mappend*
is used on inputs which are infinite lazy
lists, such that the function returns
.code nil
values indefinitely.
For instance:
.verb
;; Danger: infinite loop!!!
(mappend* (fun identity) (repeat '(nil)))
.brev
The
.code mappend*
function is caught in a loop trying to consume
and squash an infinite stream of
.codn nil s,
and so doesn't return.
.TP* Examples:
.verb
;; multiply every element by two
(mapcar (lambda (item) (* 2 item)) '(1 2 3)) -> (4 6 8)
;; "zipper" two lists together
(mapcar (lambda (le ri) (list le ri)) '(1 2 3) '(a b c))
-> '((1 a) (2 b) (3 c)))
;; like append, mappend allows a lone atom or a trailing atom:
(mappend (fun identity) 3) -> (3)
(mappend (fun identity) '((1) 2)) -> (1 . 2)
;; take just the even numbers
(mappend (lambda (item) (if (evenp x) (list x))) '(1 2 3 4 5))
-> (2 4)
.brev
.coNP Functions @, maprod @ maprend and @ maprodo
.synb
.mets (maprod < function << iterable *)
.mets (maprend < function << iterable *)
.mets (maprodo < function << iterable *)
.syne
.desc
The
.codn maprod ,
.code maprend
and
.code maprodo
functions resemble
.codn mapcar ,
.code mappend
and
.codn mapdo ,
respectively. When given no
.meta iterable
arguments or exactly one
.meta iterable
argument, they behave exactly like those three functions.
When two or more
.meta iterable
arguments are present,
.code maprod
differs from
.code mapcar
in the following way, as do the remaining functions
from their aforementioned counterparts.
Whereas
.code mapcar
iterates over the
.meta iterable
values in parallel, taking successive tuples of element
values and passing them to
.metn function ,
the
.code maprod
function iterates over all
.I combinations
of elements from the sequences: the Cartesian product. The
.code prod
suffix stands for "product".
If one or more
.meta iterable
arguments specify an empty sequence, then the Cartesian product is empty.
In this situation,
.meta function
is not called. The result of the function is then
.code nil
converted to the same kind of sequence as the leftmost
.metn iterable .
The
.code maprod
function collects the values into a list just as
.code mapcar
does. Just like
.codn mapcar ,
it converts the resulting list into the same kind of sequence
as the leftmost
.meta iterable
argument, if possible. For instance, if the resulting list is
a list or vector of characters, and the leftmost
.meta iterable
is a character string, then the list or vector of characters
is converted to a character string and returned.
The
.code maprend
function ("map product through function and append") iterates the
.meta iterable
element combinations exactly like
.codn maprod ,
passing them as arguments to
.metn function .
The values returned by
.meta function
are then treated exactly as by the
.code mappend
function. The return values are expected to be sequences which
are appended together as if by
.codn append ,
and the final result is converted to the same kind of sequence as the leftmost
.meta iterable
if possible.
The
.code maprodo
function, like
.codn mapdo ,
ignores the result of
.meta function
and returns
.codn nil .
The combination iteration gives priority to the rightmost
.metn iterable ,
which means that the rightmost element of each generated tuple varies
fastest: the tuples are traversed in "rightmost major" order.
This is made clear in the examples.
.TP* Examples
.verb
[maprod list '(0 1 2) '(a b) '(i ii iii)]
->
((0 a i) (0 a ii) (0 a iii) (0 b i) (0 b ii) (0 b iii)
(1 a i) (1 a ii) (1 a iii) (1 b i) (1 b ii) (1 b iii)
(2 a i) (2 a ii) (2 a iii) (2 b i) (2 b ii) (2 b iii))
;; Vectors #(#\ea #\ex) #(#\ea #\ey) ... are appended
;; together resulting in #(#\ea #\ex #\ea #\ey ...)
;; which is converted to a string:
[maprend vec "ab" "xy"] -> "axaybxby"
;; One of the sequences is empty, so the product is an
;; empty sequence of the same kind as the leftmost
;; sequence argument, thus an empty string:
[maprend vec "ab" ""] -> ""
.brev
.coNP Function @ mapdo
.synb
.mets (mapdo < function << iterable *)
.syne
.desc
The
.code mapdo
function is similar to
.codn mapcar ,
but always returns
.codn nil .
It is useful
when
.meta function
performs some kind of side effect, hence the "do" in the name,
which is a mnemonic for the execution of imperative actions.
When only the
.meta function
argument is given,
.meta function
is never called,
and
.code nil
is returned.
If a single
.meta iterable
argument is given, then
.code mapdo
iterates over
.metn iterable ,
invoking
.meta function
on each element.
If two or more
.meta iterable
arguments are given, then
.code mapdo
iterates over
the sequences in parallel, extracting parallel tuples of items. These
tuples are passed as arguments to
.metn function ,
which must accept as many
arguments as there are sequences.
.coNP Functions @ transpose and @ zip
.synb
.mets (transpose << iterable )
.mets (zip << iterable *)
.syne
.desc
The
.code transpose
function performs a transposition on
.metn iterable .
This means that the
elements of
.meta iterable
must be iterable. These iterables are understood to be
columns; transpose exchanges rows and columns, returning a sequence of the rows
which make up the columns. The returned sequence is of the same kind as
.metn iterable ,
and the rows are also the same kind of sequence as the first column
of the original sequence. The number of rows returned is limited by the
shortest column among the sequences.
All of the input sequences (the elements of
.metn iterable )
must have elements
which are compatible with the first sequence. This means that if the first
element of
.meta iterable
is a string, then the remaining sequences must be
strings, or else sequences of characters, or of strings.
The
.code zip
function takes variable arguments, and is equivalent to calling
.code transpose
on a list of the arguments. The following equivalences hold:
.verb
(zip . x) <--> (transpose x)
[apply zip x] <--> (transpose x)
.brev
.TP* Examples:
.verb
;; transpose list of lists
(transpose '((a b c) (c d e))) -> ((a c) (b d) (c e))
;; transpose vector of strings:
;; - string columns become string rows
;; - vector input becomes vector output
(transpose #("abc" "def" "ghij")) -> #("adg" "beh" "cfi")
;; error: transpose wants to make a list of strings
;; but 1 is not a character
(transpose #("abc" "def" '(1 2 3))) ;; error!
;; String elements are catenated:
(transpose #("abc" "def" ("UV" "XY" "WZ")))
-> #("adUV" "beXY" "cfWZ")
;; Transpose list of ranges
(transpose (list 1..4 4..8 8..12))
-> ((1 4 8) (2 5 9) (3 6 10))
(zip '(a b c) '(c d e)) -> ((a c) (b d) (c e))
.brev
.coNP Functions @, window-map @ window-mappend and @ window-mapdo
.synb
.mets (window-map < range < boundary < function << sequence )
.mets (window-mappend < range < boundary < function << sequence )
.mets (window-mapdo < range < boundary < function << sequence )
.syne
.desc
The
.code window-map
and
.code window-mappend
functions process the elements of
.meta sequence
by passing arguments derived from each successive element to
.metn function .
Both functions return, if possible, a sequence of the same kind as
.codn sequence ,
otherwise a list.
Under
.codn window-map ,
values returned by
.meta function
are accumulated into a sequence of the same type as
.meta sequence
and that sequence is returned. Under
.codn window-mappend ,
the values returned by the calls to
.meta function
are expected to be sequence which are appended together to
form the output sequence.
These functions are analogous to
.code mapcar
and
.codn mappend .
Unlike these, they operate only on a single sequence, and over this sequence
they perform a
.IR "sliding window mapping" ,
whose description follows.
The function
.code window-mapdo
avoids accumulating a sequence, and instead returns
.codn nil ;
it is analogous to
.codn mapdo .
The argument to the
.meta range
parameter must be a positive integer, not exceeding 512.
This parameter specifies the amount of ahead/behind context on either
side of each element which is processed. It indirectly determines
the window size for the mapping. The window size is twice
.metn range ,
plus one. For instance if range is 2, then the window size is 5:
the element being processed lies at the center of the window, flanked
by two elements on either side, making five.
The
.meta function
argument must specify a function which accepts a number of arguments
corresponding to the window size. For instance if
.meta range
is 2,
making the window size 5,
then
.meta function
must accept 5 arguments. These arguments constitute the sliding
window being processed. Each time
.meta function
is called, the middle argument is the element being processed,
and the arguments surrounding it are its window.
When an element is processed from somewhere in the interior of
a sequence, where it is flanked on either side by at least
.meta range
elements, then the window is populated by those flanking elements
taken from
.metn sequence .
The
.meta boundary
parameter specifies the window contents which are used for the
processing of elements which are closer than
.meta range
to either end of the sequence. The argument may be a sequence containing
at least twice
.meta range
number of elements (one less than the window size): if it has additional
elements, they are not used. If it is a list, it may be shorter than twice
.metn range .
The argument
may also be one of the two keyword symbols
.code :wrap
or
.codn :reflect ,
described below.
If
.meta boundary
is a sequence, it may be regarded as divided into two pieces of
.meta range
length. If it is a list of insufficient length, then missing elements
are supplied as
.code nil
to make two
.metn range 's
worth of elements. These two pieces then flank
.code sequence
on either end. The left half of
.meta boundary
is effectively prepended to the sequence, and the right half
effectively appended.
When the sliding window extends beyond the boundary of
.meta sequence
near its start or end, the window is populated from these
flanking elements obtained from
.metn boundary .
If
.meta boundary
argument is specified as the keyword
.codn :wrap ,
then the sequence is imagined to be flanked at either end by an infinite
repetition of copies of itself. These flanks are trimmed to the window size to
generate the boundary.
For instance if the sequence is
.code "(1 2 3)"
and the window size is 9 due to the value of
.meta range
being 4, then the behavior of
.code :wrap
is as if
.meta boundary
value of
.code "(3 1 2 3 1 2 3 1)"
were specified.
The left flank is
.codn "(3 1 2 3)" ,
being the last four elements of an infinite repetition of
.codn "1 2 3" ;
and the right flank is similarly
.codn "(1 2 3 1)" ,
being the first four elements of an infinite repetition of
.codn "1 2 3" .
If
.meta boundary
is given as the keyword
.codn :reflect ,
then the sequence is imagined to be flanked at either end by an infinite
repetition of reversed copies of itself. These flanks are trimmed to the window
size to generate the boundary. For instance if the sequence is
.code "(1 2 3)"
and the window size is 9 due to the value of
.meta range
being 4, then the behavior of
.code :reflect
is as if
.meta boundary
value of
.code "(1 3 2 1 3 2 1 3)"
were specified.
The left flank is
.codn "(1 3 2 1)" ,
being the last four elements of an infinite repetition of
.codn "3 2 1" ;
and the right flank is similarly
.codn "(3 2 1 3)" ,
being the first four elements of an infinite repetition of
.codn "3 2 1" .
.TP* Examples:
.verb
;; change characters between angle brackets to upper case.
[window-map 1 nil (lambda (x y z)
(if (and (eq x #\e<)
(eq z #\e>))
(chr-toupper y)
y))
"ab<c>de<f>g"]
--> "ab<C>de<F>g"
;; collect all numbers which are the centre element of
;; a monotonically increasing triplet
[window-mappend 1 :reflect (lambda (x y z)
(if (< x y z)
(list y)))
'(1 2 1 3 4 2 1 9 7 5 7 8 5)]
--> (3 7)
;; calculate a moving average with a five-element
;; window, flanked by zeros at the boundaries:
[window-map 2 #(0 0 0 0)
(lambda (. args) (/ (sum args) 5))
#(4 7 9 13 5 1 6 11 10 3 8)]
--> #(4.0 6.6 7.6 7.0 6.8 7.2 6.6 6.2 7.6 6.4 4.2))
.brev
.coNP Function @ interpose
.synb
.mets (interpose < sep << sequence )
.syne
.desc
The
.code interpose
function returns a sequence of the same type as
.metn sequence ,
in which the elements from
.meta sequence
appear with the
.meta sep
value inserted
between them.
If
.meta sequence
is an empty sequence or a sequence of length 1, then a
sequence identical to
.meta sequence
is returned. It may be a copy of
.meta sequence
or it may be
.meta sequence
itself.
If
.meta sequence
is a character string, then the value
.meta sep
must be a character.
It is permissible for
.metn sequence ,
or for a suffix of
.meta sequence
to be a lazy
list, in which case interpose returns a lazy list, or a list with a lazy
suffix.
.TP* Examples:
.verb
(interpose #\e- "xyz") -> "x-y-z"
(interpose t nil) -> nil
(interpose t #()) -> #()
(interpose #\ea "") -> ""
(interpose t (range 0 0)) -> (0)
(interpose t (range 0 1)) -> (0 t 1)
(interpose t (range 0 2)) -> (0 t 1 t 2)
.brev
.coNP Functions @ reduce-left and @ reduce-right
.synb
.mets (reduce-left < binary-function < list
.mets \ \ \ \ \ \ \ \ \ \ \ \ >> [ init-value <> [ key-function ]])
.mets (reduce-right < binary-function < list
.mets \ \ \ \ \ \ \ \ \ \ \ \ \ >> [ init-value <> [ key-function ]])
.syne
.desc
The
.code reduce-left
and
.code reduce-right
functions reduce lists of operands specified
by
.meta list
and
.meta init-value
to a single value by the repeated application of
.metn binary-function .
In the case of
.codn reduce-left ,
the
.meta list
argument is required to be an object which is iterable according to the
.code iter-begin
function. The
.code reduce-right
function treats the
.meta list
argument using list operations.
An effective list of operands is formed by combining
.meta list
and
.metn init-value .
If
.meta key-function
is specified, then the items of
.meta list
are
mapped to new values through
.metn key-function ,
as if by
.codn mapcar .
If
.meta init-value
is supplied,
then in the case of
.codn reduce-left ,
the effective list of operands is formed by
prepending
.meta init-value
to
.metn list .
In the case of
.codn reduce-right ,
the effective operand list is produced by appending
.meta init-value
to
.metn list .
The
.meta init-value
isn't mapped through
.metn key-function .
The production of the effective list can be expressed like this,
though this is not to be understood as the actual implementation:
.verb
(append (if init-value-present (list init-value))
[mapcar (or key-function identity) list]))))
.brev
In the
.code reduce-right
case, the arguments to
.code append
are reversed.
If the effective list of operands is empty, then
.meta binary-function
is called
with no arguments at all, and its value is returned. This is the only
case in which
.meta binary-function
is called with no arguments; in all
remaining cases, it is called with two arguments.
If the effective list contains one item, then that item is returned.
Otherwise, the effective list contains two or more items, and is decimated as
follows.
Note that an
.meta init-value
specified as
.code nil
is not the same as a missing
.metn init-value ;
this means that the initial value is the object
.codn nil .
Omitting
.meta init-value
is the same as specifying a value of
.code :
(the colon keyword symbol).
It is possible to specify
.meta key-function
while omitting an
.meta init-value
argument. This is achieved by explicitly specifying
.code :
as the
.meta init-value
argument.
Under
.codn reduce-left ,
the leftmost pair of operands is removed
from the list and passed as arguments to
.metn binary-function ,
in the same order
that they appear in the list, and the resulting value initializes an
accumulator. Then, for each remaining item in the list,
.meta binary-function
is invoked on two arguments: the current accumulator value, and the next element
from the list. After each call, the accumulator is updated with the return
value of
.metn binary-function .
The final value of the accumulator is returned.
Under
.codn reduce-right ,
the list is processed right to left. The rightmost
pair of elements in the effective list is removed, and passed as arguments to
.metn binary-function ,
in the same order that they appear in the list. The
resulting value initializes an accumulator. Then, for each remaining item in
the list,
.meta binary-function
is invoked on two arguments: the
next element from the list, in right to left order, and the current
accumulator value. After each call, the accumulator is updated with the return
value of
.metn binary-function .
The final value of the accumulator is returned.
.TP* Examples:
.verb
;;; effective list is (1) so 1 is returned
(reduce-left (fun +) () 1 nil) -> 1
;;; computes (- (- (- 0 1) 2) 3)
(reduce-left (fun -) '(1 2 3) 0 nil) -> -6
;;; computes (- 1 (- 2 (- 3 0)))
(reduce-right (fun -) '(1 2 3) 0 nil) -> 2
;;; computes (* 1 2 3)
(reduce-left (fun *) '((1) (2) (3)) nil (fun first)) -> 6
;;; computes 1 because the effective list is empty
;;; and so * is called with no arguments, which yields 1.
(reduce-left (fun *) nil)
.brev
.coNP Functions @, some @ all and @ none
.synb
.mets (some < sequence >> [ predicate-fun <> [ key-fun ]])
.mets (all < sequence >> [ predicate-fun <> [ key-fun ]])
.mets (none < sequence >> [ predicate-fun <> [ key-fun ]])
.syne
.desc
The
.codn some ,
.code all
and
.code none
functions apply a predicate-test function
.meta predicate-fun
over a list of elements. If the argument
.meta key-fun
is
specified, then elements of
.meta sequence
are passed into
.metn key-fun ,
and
.meta predicate-fun
is
applied to the resulting values. If
.meta key-fun
is omitted, the behavior is
as if
.meta key-fun
were the
.code identity
function. If
.meta predicate-fun
is omitted,
the behavior is as if
.meta predicate-fun
were the
.code identity
function.
These functions have short-circuiting semantics and return conventions similar
to the
.code and
and
.code or
operators.
The
.code some
function applies
.meta predicate-fun
to successive values
produced by retrieving elements of
.meta list
and processing them through
.metn key-fun .
If the list is empty, it returns
.codn nil .
Otherwise it returns the
first
.cod2 non- nil
return value returned by a call to
.meta predicate-fun
and
stops evaluating the elements. If
.meta predicate-fun
returns
.code nil
for all
elements,
.code some
returns
.metn nil .
The
.code all
function applies
.meta predicate-fun
to successive values
produced by retrieving elements of
.meta list
and processing them through
.metn key-fun .
If the list is empty, it returns
.codn t .
Otherwise, if
.meta predicate-fun
yields
.code nil
for any value, the
.code all
function immediately
returns without invoking
.meta predicate-fun
on any more elements.
If all the elements are processed, then the
.code all
function returns the value which
.meta predicate-fun
yielded for the last element.
The
.code none
function applies
.meta predicate-fun
to successive values
produced by retrieving elements of
.meta list
and processing them through
.metn key-fun .
If the list is empty, it returns
.codn t .
Otherwise, if
.meta predicate-fun
yields
.cod2 non- nil
for any value, the
.code none
function immediately returns
.codn nil .
If
.meta predicate-fun
yields
.code nil
for all values, the
.code none
function returns
.codn t .
.TP* Examples:
.verb
;; some of the integers are odd
[some '(2 4 6 9) oddp] -> t
;; none of the integers are even
[none '(1 3 4 7) evenp] -> t
.brev
.coNP Function @ multi
.synb
.mets (multi < function << sequence *)
.syne
.desc
The
.code multi
function distributes an arbitrary list processing function
.meta multi
over multiple sequences given by the
.meta list
arguments.
The
.meta sequence
arguments are first transposed into a single list of tuples. Each
successive element of this transposed list consists of a tuple of the
successive items from the lists. The length of the transposed list is that
of the shortest
.meta list
argument.
The transposed list is then passed to
.meta function
as an argument.
The
.meta function
is expected to produce a list of tuples, which are transposed
again to produce a list of lists which is then returned.
Conceptually, the input sequences are columns and
.meta function
is invoked on
a list of the rows formed from these columns. The output of
.meta function
is a transformed list of rows which is reconstituted into a list of columns.
.TP* Example:
.verb
;; Take three lists in parallel, and remove from all of them the
;; element at all positions where the third list has an element of 20.
(multi (op remove-if (op eql 20) @1 third)
'(1 2 3)
'(a b c)
'(10 20 30))
-> ((1 3) (a c) (10 30))
;; The (2 b 20) "row" is gone from the three "columns".
;; Note that the (op remove if (op eql 20) @1 third)
;; expression can be simplified using the ap operator:
;;
;; (op remove-if (ap eql @3 20))
.brev
.coNP Functions @ sort and @ nsort
.synb
.mets (sort < sequence >> [ lessfun <> [ keyfun ]])
.mets (nsort < sequence >> [ lessfun <> [ keyfun ]])
.syne
.desc
The
.code nsort
function destructively sorts
.metn sequence ,
producing a sequence
which is sorted according to the
.meta lessfun
and
.meta keyfun
arguments.
The
.meta keyfun
argument specifies a function which is applied to elements
of the sequence to obtain the key values which are then compared
using the lessfun. If
.meta keyfun
is omitted, the identity function is used
by default: the sequence elements themselves are their own sort keys.
The
.meta lessfun
argument specifies the comparison function which determines
the sorting order. It must be a binary function which can be invoked
on pairs of keys as produced by the key function. It must
return a
.cod2 non- nil
value if the left argument is considered to be lesser
than the right argument. For instance, if the numeric function
.code <
is used
on numeric keys, it produces an ascending sorted order. If the function
.code >
is used, then a descending sort is produced. If
.meta lessfun
is omitted, then it defaults to the generic
.code less
function.
The
.code sort
function has the same argument requirements as
.code nsort
but is non-destructive: it returns a new object, leaving the input
.meta sequence
unmodified, as if a copy of the input object were made using the
function
.code copy
and then that copy were sorted in-place using
.codn nsort .
The
.code sort
and
.code nsort
functions are stable for sequences which are lists. This means that the
original order of items which are considered identical is preserved.
For strings and vectors,
.code sort
is not stable.
The
.code sort
and
.code nsort
functions can be applied to hashes. It produces meaningful behavior
for a hash table which contains
.I N
keys which are the integers from 0 to
.IR "N - 1" .
Such as hash is treated as if it were a vector. The values are sorted
and reassigned to sorted order to the integer keys.
The behavior of
.code sort
is not specified for hashes whose contents do not conform to this convention.
Note:
.code nsort
was introduced in \*(TX 238. Prior to that version,
.code sort
behaved like
.codn nsort .
.coNP Function @ grade
.synb
.mets (grade < sequence >> [ lessfun <> [ keyfun ]])
.syne
.desc
The
.code grade
function returns a list of integer indices which indicate the position
of the elements of
.meta sequence
in sorted order.
The
.meta lessfun
and
.meta keyfun
arguments behave like those of the
.code sort
function.
The
.meta sequence
object is not modified.
The internal sort performed by
.code grade
is not stable. The indices of any elements considered equivalent under
.code lessfun
may appear in any order in the returned index sequence.
Note: the
.code grade
function is inspired by the "grade up" and "grade down" operators
in the APL language.
.TP* Examples:
.verb
;; Order of the 2 3 positions of the "l"
;; characters is not specified:
[grade "Hello"] -> (0 1 2 3 4)
[grade "Hello" >] -> (4 2 3 1 0)
.brev
.coNP Functions @ shuffle and @ nshuffle
.synb
.mets (shuffle < sequence <> [ random-state ])
.mets (nshuffle < sequence <> [ random-state ])
.syne
.desc
The
.code nshuffle
function pseudorandomly rearranges the elements of
.metn sequence .
This is performed in place:
.meta sequence
object is modified.
The return value is
.meta sequence
itself.
The rearrangement depends on pseudorandom numbers obtained from the
.code rand
function. The
.meta random-state
argument, if present, is passed to that function.
The
.code nshuffle
function supports hash tables in a manner analogous to the way
.code nsort
supports hash tables; the same remarks apply as in the description
of that function.
The
.code shuffle
function has the same argument requirements and
semantics, but differs from
.code nshuffle
in that it avoids in-place modification of
.metn sequence :
a new, shuffled sequence is returned, as if a copy of
.meta sequence
were made using
.code copy
and then that copy were shuffled in-place and returned.
Note:
.code nshuffle
was introduced in \*(TX 238. Prior to that version,
.code shuffle
behaved like
.codn nshuffle .
.coNP Functions @ rot and @ nrot
.synb
.mets (rot < sequence <> [ displacement ])
.mets (nrot < sequence <> [ displacement ])
.syne
.desc
The
.code nrot
and
.code rot
functions rotate the elements of
.metn sequence ,
returning a rotated sequence.
The
.code nrot
function does this destructively; it modifies
.meta sequence
in-place, whereas
.code rot
returns a new sequence without modifying the original.
The
.code rot
function always returns a new sequence. In cases when no rotation
is performed, it copies
.meta sequence
as if using the
.code copy
function.
In cases when no rotation is performed, the
.code nrot
function returns the original sequence, which is unmodified.
The
.meta displacement
parameter, an integer, has a default value of 1.
To rotate elements means to displace their position within the
.meta sequence
by some amount, that being given by the
.meta displacement
parameter, while partially preserving their circular order.
Circular order means that for the purposes of rotation, the sequence
is regarded to be cyclic: the first element of the sequence is
considered to be the successor of the last element and vice versa.
Thus, when an element is displaced past the first or last position, it wraps to
the end or beginning of the sequence.
If the sequence is empty, or contains only one element, then
.code rot
and
.code nrot
terminate, performing no rotation. The following remarks apply to situations when
.meta sequence
has two or more elements.
The
.meta displacement
parameter, which may be negative, is first reduced to the smallest positive
residue modulo the length of the sequence, resulting in a value ranging from
zero to one less than the sequence length. If the resulting value is zero,
then no rotation is performed.
The
.meta displacement
has a negative orientation: each element's position is decreased by this
amount. Those elements whose position would become negative move to the end of
the sequence.
The default displacement of 1 causes the first element to become last,
the second element to become first, and so forth. The opposite rotation can be
obtained using -1 as the displacement.
Note: even though
.code nrot
operates destructively, the returned object may not be the same object as
.metn sequence .
Only the returned object is required to be the rotated sequence. If this
is different from the original
.meta sequence
input, the contents of that original object are unspecified.
Note: the symbol
.code rotate
is the name of a place-mutating macro, which is much older than these functions.
If
.code S
is a three-element sequence, then:
.verb
(set S (nrot S)) ;; alternatively: (upd S nrot)
.brev
has the same effect as:
.verb
(rotate [S 0] [S 1] [S 2])
.brev
.TP* Examples:
.verb
(rot "abc") -> "bca"
(rot #(1 2 3) -1) -> (3 1 2)
;; lower-case rot-13
(mapcar (relate (range #\ea #\ez)
(rot (range #\ea #\ez) 13))
"hello, world!")
-> "uryyb, jbeyq!"
.brev
.coNP Function @ sort-group
.synb
.mets (sort-group < sequence >> [ keyfun <> [ lessfun ]])
.syne
.desc
The
.code sort-group
function sorts
.meta sequence
according to the
.meta keyfun
and
.meta lessfun
arguments, and then breaks the resulting sequence into groups,
based on the equivalence of the elements under
.metn keyfun .
The following equivalence holds:
.verb
(sort-group sq lf kf)
<-->
(partition-by kf (sort (copy sq) kf lf))
.brev
Note the reversed order of
.meta keyfun
and
.meta lessfun
arguments between
.code sort
and
.codn sort-group .
.coNP Function @ uniq
.synb
.mets (uniq << sequence )
.syne
.desc
The
.code uniq
function returns a sequence of the same kind as
.metn sequence ,
but with
duplicates removed. Elements of
.meta sequence
are considered equal under
the
.code equal
function. The first occurrence of each element is retained,
and the subsequent duplicates of that element, of any, are suppressed,
such that the order of the elements is otherwise preserved.
The
.code uniq
function is an alias for the one-argument case of
.codn unique .
That is to say, this equivalence holds:
.verb
(uniq s) <--> (unique s)
.brev
.coNP Function @ unique
.synb
.mets (unique < sequence >> [ keyfun <> { hash-arg }* ])
.syne
.desc
The
.code unique
function is a generalization of
.codn uniq .
It returns a sequence of the same kind as
.metn sequence ,
but with duplicates removed.
If neither
.meta keyfun
nor
.metn hash-arg s
are specified, then elements of sequence are considered equal under the
.code eql
function. The first occurrence of each element is retained,
and the subsequent duplicates of that element, of any, are suppressed,
such that the order of the elements is otherwise preserved.
If
.meta keyfun
is specified, then that function is applied to each element,
and the resulting values are compared for equality.
In other words, the behavior is as if
.meta keyfun
were the
.code identity
function.
If one or more
.metn hash-arg s
are present, these specify the arguments for the construction of
the internal hash table used by
.codn unique .
The arguments are like those of the
.code hash
function.
.coNP Function @ tuples
.synb
.mets (tuples < length < sequence <> [ fill-value ])
.syne
.desc
The
.code tuples
function produces a lazy list which represents a reorganization
of the elements of
.meta sequence
into tuples of
.metn length ,
where
.meta length
must be a positive integer.
The length of the sequence might not be evenly divisible by the tuple length.
In this case, if a
.meta fill-value
argument is specified, then the last tuple
is padded with enough repetitions of
.meta fill-value
to make it have
.meta length
elements. If
.meta fill-value
is not specified, then the last tuple is left
shorter than
.metn length .
The output of the function is a list, but the tuples themselves are sequences
of the same kind as
.metn sequence .
If
.meta sequence
is any kind of list, they
are lists, and not lazy lists.
.TP* Examples:
.verb
(tuples 3 #(1 2 3 4 5 6 7 8) 0) -> (#(1 2 3) #(4 5 6)
#(7 8 0))
(tuples 3 "abc") -> ("abc")
(tuples 3 "abcd") -> ("abc" "d")
(tuples 3 "abcd" #\ez) -> ("abc" "dzz")
(tuples 3 (list 1 2) #\ez) -> ((1 2 #\ez))
.brev
.coNP Function @ tuples*
.synb
.mets (tuples* < length < sequence <> [ fill-value ])
.syne
.desc
The
.code tuples*
function produces a lazy list of overlapping tuples taken from
.metn sequence .
The length of the tuples is given by the
.meta length
argument.
The
.meta length
argument must be a positive integer.
Tuples are subsequences of consecutive items from the input
.metn sequence ,
beginning with consecutive elements. The first tuple in the returned list
begins with the first item of
.metn sequence ;
the second tuple begins with the second item, and so forth.
The output of the function is a list, but the tuples themselves are sequences
of the same kind as
.metn sequence .
If
.meta sequence
is any kind of list, they
are lists, and not lazy lists.
If
.meta sequence
is shorter than
.meta length
then it contains no tuples of that length. In this case, if no
.meta fill-value
argument is specified, then the empty list is returned.
In this same situation, if
.meta fill-value
is specified, then a one-element list is returned, consisting of
a tuple of the required length, consisting of the elements from
.meta sequence
followed by repetitions of
.metn fill-value ,
which must be of a type suitable as an element of the sequence.
The
.meta fill-value
is otherwise ignored.
.TP* Examples:
.verb
.brev
(tuples* 1 "abc") -> ("a" "b" "c")
(tuples* 2 "abc") -> ("ab" "bc")
(tuples* 3 "abc") -> ("abc")
(tuples* 4 "abc") -> nil
(tuples* 4 "abc" #\z) -> ("abcz")
(tuples* 6 "abc" #\z) -> ("abczzz")
(tuples* 6 "abc" 4) -> error
(tuples* 2 '(a b c)) -> ((a b) (b c))
(take 3 (tuples* 3 0)) -> ((0 1 2) (1 2 3) (2 3 4))
.brev
.coNP Function @ partition-by
.synb
.mets (partition-by < function << sequence )
.syne
.desc
If
.meta sequence
is empty, then
.code partition-by
returns an empty list,
and
.meta function
is never called.
Otherwise,
.code partition-by
returns a lazy list of partitions of the sequence
.metn sequence .
Partitions are consecutive, nonempty substrings of
.metn sequence ,
of the same kind as
.metn sequence .
The partitioning begins with the first element of
.meta sequence
being placed into a partition.
The subsequent partitioning is done according to
.metn function ,
which is applied
to each element of
.metn sequence .
Whenever, for the next element, the function
returns the same value as it returned for the previous element, the
element is placed into the same partition. Otherwise, the next element
is placed into, and begins, a new partition.
The return values of the calls to
.meta function
are compared using the
.code equal
function.
.TP* Examples:
.verb
[partition-by identity '(1 2 3 3 4 4 4 5)] -> ((1) (2) (3 3)
(4 4 4) (5))
(partition-by (op = 3) #(1 2 3 4 5 6 7)) -> (#(1 2) #(3)
#(4 5 6 7))
.brev
.SS* Open Sequence Traversal
Functions in this category perform efficient traversal of sequences.
There are two flavors of these functions: functions in the
.code iter-begin
group, and functions in the
.code seq-begin
group. The latter are obsolescent.
User-defined iteration is possible via defining special methods on
structures. An object supports iteration by defining the special method
.code iter-begin
which is different from the
.code iter-begin
function. This special function returns an iterator object which supports
special methods
.codn iter-item ,
.code iter-more
and
.codn iter-step .
Two protocols are supported, one of which is more efficient by eliminating the
.code iter-more
method. Details are specified in the section
.BR "Special Structure Functions" .
.coNP Function @ iter-begin
.synb
.mets (iter-begin << seq )
.syne
.desc
The
.code iter-begin
function returns an iterator object suitable for traversing the
elements of the sequence denoted by the
.meta seq
object.
If
.meta seq
is a list-like sequence, then
.code iter-begin
may return
.meta seq
itself as the iterator. Likewise if
.meta seq
is a number.
If
.meta seq
is an iterator produced by
.code iter-begin
then an iterator similar to that iterator is returned, which can continue
iterating the same sequence. The iterator may be
.meta seq
itself or share state with
.metn seq ,
and thus may not be relied on to produce an independent, parallel iteration.
If
.meta seq
is a structure which supports the
.code iter-begin
method, then that method is called and its return value is returned.
A structure which does not support this method is possibly considered
to be a sequence according to the usual criteria, based on whether
it supports the
.codn nullify ,
.code length
or
.code car
methods. An struct object supporting none of these methods is deemed
not iterable.
In all other cases, if
.meta seq
is iterable, an object of type
.code seq-iter
is returned.
Range objects are iterable if they are numeric. A range consisting of
two strings may also be iterable, as described below.
A range is considered to be a numeric or character range if the
.code from
element is a number or character. The
.code to
is then required to be either a value which is comparable with that number
or character using the
.code <
function, or else it must be one of the two objects
.code t
or
.codn : ,
either of which indicate that the range is unbounded. In this unbounded range
case, the expressions
.code "(iter-begin X..:)"
and
.code "(iter-begin X..t)"
are equivalent to
.codn "(iter-begin X)" .
Numeric ranges are half-open: the
.code to
value of ascending ranges is excluded, as is the
.code from
value of descending ranges, so that
.code 0..10
steps through the values
.code 0
through
.codn 9 ,
and
.code 10..0
steps through the same values in reverse order.
A string range consists of two strings of equal length. If the strings are
of unequal length, an error exception is thrown. For the range to operate
as intended, the strings must meet some additional requirements. If the
.code from
string is lexicographically lesser than the
.code to
string, as determined by the
.code less
function, then the range is ascending, otherwise it is descending. The string
range iterates by incrementing (or decrementing, in the case of a descending range)
the characters of the
.code from
string until they are equal to those of the
.code to
string. The last character has priority. For instance, the range
.code "\(dqAA\(dq..\(dqCC\(dq"
iterates over the strings
.codn "AA" ,
.codn "AB" ,
.codn "AC" ,
.codn "BA" ,
.codn "BB" ,
.codn "BC" ,
.codn "CA" ,
.code "CB"
and
.codn "CC" .
The descending range
.code "\(dqCC\(dq..\(dqAA\(dq"
iterates over the same strings, in reverse order. Whenever the incrementing
character attains the value of the corresponding character in the
.code to
string, that character is reset to its starting value, and its left neighbor,
if it exists, is incremented instead. If no left neighbor exists, the
iteration terminates.
Search trees are iterable. Iteration entails an in-order visits of the elements
of a tree. A tree iterator created by
.code tree-begin
is also iterable. It is unspecified whether iteration over a
.code tree-iter
object modifies that object to perform the traversal, or whether it uses a copy
of the iterator.
If
.code seq
is not an iterable object, an error exception is thrown.
.coNP Function @ iter-more
.synb
.mets (iter-more << iter )
.syne
.desc
The
.code iter-more
function returns
.code t
if there remain more elements to be traversed.
Otherwise it returns
.codn nil .
The
.meta iter
argument must be a valid iterator returned by a call to
.metn iter-begin ,
.meta iter-step
or
.metn iter-reset .
The
.code iter-more
function doesn't change the state of
.metn iter .
If
.code iter
is the object
.code nil
then
.code nil
is returned.
Note: the
.code iter-begin
may return
.code nil
if its argument is
.code nil
or any empty sequence, or an empty range (a range whose
.code to
and
.code from
fields are the same number or character).
If
.meta iter
is a
.code cons
cell, then
.code iter-more
returns
.codn t .
If
.meta iter
is a number, then
.code iter-more
returns
.codn t .
This is the case even if calculating the successor of that number isn't possible
due to floating-point overflow or insufficient system resources.
If
.meta iter
is a character, then
.code iter-more
returns
.code t
if
.meta iter
isn't the highest possible character code, otherwise
.codn nil .
If
.meta iter
was formed from a descending range, meaning that
.code iter-begin
was invoked on a range with a
.code from
fielding exceeding its
.code to
value, then
.code iter-begin
returns true while the current iterator value is greater than the
the limiting value given by the
.code to
field. For an ascending range, it returns true if the current iterator value is
lower than the limiting value. However, note the peculiar semantics of
.code iter-item
with regard to descending range iteration.
If
.meta iter
is a structure, then if it supports an
.code iter-more
method, then that method is called with no arguments, and its return value
is returned. If the structure does not have an
.code iter-more
method, then
.code t
is returned.
.coNP Function @ iter-item
.synb
.mets (iter-item << iter )
.syne
.desc
If the
.code iter-more
function indicates that more items remain to be visited, then
the next item can be retrieved using
.codn iter-item .
The
.meta iter
argument must be a valid iterator returned by a call to
.metn iter-begin ,
.meta iter-step
or
.metn iter-reset .
The
.code iter-more
function doesn't change the state of
.metn iter .
If
.code iter-more
is invoked on an iterator which indicates that no more items
remain to be visited, the return value is
.codn nil .
If
.meta iter
is a
.code cons
cell, then
.code iter-item
returns the
.code car
field of that cell.
If
.meta iter
is a character or number, then
.code iter-item
returns that character or number itself.
If
.meta iter
is based on an ascending numeric or character range, then
.code iter-item
returns the current iteration value, which is initialized by
.code iter-begin
as a copy of the range's
.code from
field. Thus, the range
.code 0..3
traverses the values
.codn 0 ,
.code 1
and
.codn 2 ,
excluding the
.codn 3 .
If
.meta iter
is based on a descending numeric or character range, then
.code iter-item
returns the predecessor of the current iteration value, which is initialized
.code iter-begin
as a copy of the range's
.code from
field.
Thus, the range
.code 3..0
traverses the values
.codn 2 ,
.code 1
and
.codn 0 ,
excluding the
.codn 3 :
exactly the same values are visited as for the range
.code 0..3
only in reverse order.
If
.meta iter
is a structure which supports the
.code iter-item
method, then that method is called and its return value is returned.
.coNP Function @ iter-step
.synb
.mets (iter-step << iter )
.syne
.desc
If the
.code iter-more
function indicates that more items remain to be visited, then the
.code iter-step
function may be used to consume the next item.
The function returns an iterator denoting the traversal of the
remaining items in the sequence.
The
.meta iter
argument must be a valid iterator returned by a call to
.metn iter-begin ,
.meta iter-step
or
.metn iter-reset .
The
.code iter-step
function may return a new object, in which case it avoids
changing the state of
.metn iter ,
or else it may change the state of
.meta iter
and return it.
If the application discontinues the use of
.metn iter ,
and continues the
traversal using the returned iterator, it will work correctly in either
situation.
If
.code iter-step
is invoked on an iterator which indicates that no more items
remain to be visited, the return value is unspecified.
If
.meta iter
is a
.code cons
cell, then
.code iter-step
returns the
.code cdr
field of that cell. That value must itself be a
.code cons
or else
.codn nil ,
otherwise an error is thrown. This is to prevent iteration
from wrongly iterating into the non-null terminators of improper
lists. Without this rule, iteration of a list like
.code "(1 2 . 3)"
would reach the
.code cons
cell
.code "(2 . 3)"
at which point a subsequent
.code iter-step
would return the
.code cdr
field
.codn 3 .
But that value is a valid iterator which will then continue by
stepping through
.codn 4 ,
.code 5
and so on.
If
.meta iter
is a list-like sequence, then
.code cdr
is invoked on it and that value is returned.
The value must also be a list-like sequence, or else
.codn nil .
The reasoning for this is the same as for the similar
restriction imposed in the case when
.meta iter
is a
.codn cons .
If
.meta iter
is a character or number, then
.code iter-step
returns its successor, as if using the
.code succ
function.
If
.meta iter
is a structure which supports the
.code iter-step
method, then that method is called and its return value is returned.
.coNP Function @ iter-reset
.synb
.mets (iter-reset < iter << seq )
.syne
.desc
The
.code iter-reset
function returns an iterator object specialized for the task of traversing
the sequence
.metn seq .
If it is possible for
.meta iter
to be that object, then the function may adjust the state of
.meta iter
and return it.
If
.code iter-reset
doesn't use
.metn iter ,
then it behaves exactly like
.code iter-begin
being invoked on
.metn seq .
If
.meta seq
is a structure which supports the
.code iter-reset
method, then that method is called and its return value is returned.
Note the reversed arguments. The
.code iter-reset
method is of the
.meta seq
object, not of
.metn iter .
That is to say, the call
.mono
.meti (iter-reset < iter << obj )
.onom
results in the
.mono
.meti << obj .(iter-reset << iter )
.onom
call. If
.meta seq
is a structure which doesn't support
.code iter-reset
then
.meta iter
is ignored,
.code iter-begin
is invoked on
.meta seq
and the result is returned.
.coNP Function @ seq-begin
.synb
.mets (seq-begin << object )
.syne
.desc
The obsolescent
.code seq-begin
function returns an iterator object specialized to the task of traversing
the sequence represented by the input
.metn object .
If
.meta object
isn't a sequence, an exception is thrown.
Note that if
.meta object
is a lazy list, the returned iterator maintains a reference to the
head of that list during the traversal; therefore, generic iteration
based on iterators from
.code seq-begin
is not suitable for indefinite iteration over infinite lists.
.coNP Function @ seq-next
.synb
.mets (seq-next < iter << end-value )
.syne
.desc
The obsolescent
.code seq-next
function retrieves the next available item from the sequence iterated by
.metn iter ,
which must be an object returned by
.codn seq-begin .
If the sequence has no more items to be traversed, then
.meta end-value
is returned instead.
Note: to avoid ambiguities, the application should provide an
.meta end-value
which is guaranteed distinct from any item in the sequence, such as a
freshly allocated object.
.coNP Function @ seq-reset
.synb
.mets (seq-reset < iter << object )
.syne
.desc
The obsolescent
.code seq-reset
reinitializes the existing iterator object
.meta iter
to begin a new traversal over the given
.metn object ,
which must be a value of a kind that would be a suitable argument for
.codn seq-begin .
The
.code seq-reset
function returns
.metn iter .
.SS* Procedural List Construction
\*(TL provides an a structure type called
.code list-builder
which encapsulates state and methods for constructing lists procedurally.
Among the advantages of using
.code list-builder
is that lists can be constructed in the left-to-right direction without
requiring multiple traversals or reversal. For example,
.code list-builder
naturally combines with iteration or recursion: items visited in an
iterative or recursive process can be collected easily using
.code list-builder
in the order they are visited.
The
.code list-builder
type provides methods for adding and removing items at either end of
the list, making it suitable where a
.I dequeue
structure is required.
The basic workflow begins with the instantiation of a
.code list-builder
object. This object may be initialized with a piece of list material which
begins the to-be-constructed list, or it may be initialized to begin with an
empty list. Methods such as
.code add
and
.code pend
are invoked on this object to extend the list with new elements. At any point,
the list constructed so far is available using the
.code get
method, which is also how the final version of the list is eventually
retrieved.
The
.code list-builder
methods which add material to the list all return the list builder,
making chaining possible.
.verb
(new list-builder).(add 1).(add 2).(pend '(3 4 5)).(get)
-> (1 2 3 4 5)
.brev
The
.code build
macro is provided which syntactically streamlines the process.
It implicitly creates a
.code list-builder
instance and binds it to a hidden lexical variable.
It then evaluates forms in a lexical scope in which
shorthand macros are available for building the list.
.coNP Structure @ list-builder
.synb
.mets (defstruct list-builder nil
.mets \ \ head tail)
.syne
.desc
The
.code list-builder
structure encapsulates the state for a list building process.
Programs should use the
.code build-list
function for creating an instance of
.codn list-builder .
The
.code head
and
.code tail
slots should be regarded as internal variables.
.coNP Function @ build-list
.synb
.mets (build-list <> [ initial-list ])
.syne
.desc
The
.code build-list
function instantiates and returns an object of struct type
.codn list-builder .
If no
.meta initial-list
argument is supplied, then the object is implicitly
initialized with an empty list.
If the argument is supplied, then it is equivalent
to calling
.code build-list
without an argument to produce an object
.meta obj
by invoking the method call
.mono
.meti << obj .(ncon << initial-list )
.onom
on this object. The object produced by the expression
.meta list
is installed (without being copied) into the
object as the prefix of the list to be constructed.
The
.meta initial-list
argument can be a sequence other than a list.
.TP* Example:
.verb
;; build the list (a b) trivially
(let ((lb (build-list '(a b))))
lb.(get)
-> (a b)
.brev
.coNP Methods @ add and @ add*
.synb
.mets << list-builder .(add << element *)
.mets << list-builder .(add* << element *)
.syne
.desc
The
.code add
and
.code add*
methods extend the list being constructed by a
.code list-builder
object by adding individual
elements to it. The
.code add
method adds elements at the tail of the list,
whereas
.code add*
adds elements at the front.
These methods return the
.meta list-builder
object.
The precise semantics is as follows.
All of the
.meta element
arguments are combined into a list as if by the
.code list
function, and the resulting list combined with the current contents of the
.code list-builder
object as if using the
.code append
function. The resulting list becomes the new contents.
.TP* Examples:
.verb
;; Build the list (1 2 3 4)
(let ((lb (build-list)))
lb.(add 3 4)
lb.(add* 1 2)
lb.(get))
-> (1 2 3 4)
;; Add "c" to "abc"
;; same semantics as (append "abc" #\ec)
(let ((lb (build-list "ab")))
lb.(add #\ec)
lb.(get))
-> "abc"
.brev
.coNP Methods @ pend and @ pend*
.synb
.mets << list-builder .(pend << list *)
.mets << list-builder .(pend* << list *)
.syne
.desc
The
.code pend
and
.code pend*
methods extend the list being constructed by a
.code list-builder
object by adding lists to it. The
.code pend
method catenates the
.code list
arguments together as if by the
.code append
function, then appends the resulting list to
the end of the list being constructed.
The
.code pend*
method is similar, except it prepends the
catenated lists to the front of the list
being constructed.
The
.code pend
and
.code pend*
operations do not mutate the input lists, but may cause the
resulting list to share structure with the input lists.
These functions may mutate the list already contained in
.metn list-builder ;
however, they avoid mutating those parts of the current list
that are shared with inputs that were given in earlier
calls to these functions.
These methods return the
.meta list-builder
object.
.TP* Example:
.verb
;; Build the list (1 2 3 4)
(let ((lb (build-list)))
lb.(pend '(3 4))
lb.(pend* '(1 2))
lb.(get))
-> (1 2 3 4)
.brev
.coNP Methods @ ncon and @ ncon*
.synb
.mets << list-builder .(ncon << list *)
.mets << list-builder .(ncon* << list *)
.syne
.desc
The
.code ncon
and
.code ncon*
methods extend the list being constructed by a
.code list-builder
object by adding lists to it. The
.code ncon
method destructively catenates the
.meta list
arguments as if by the
.code nconc
function. The resulting list is appended
to the list being constructed.
The
.code ncon*
method is similar, except it prepends
the catenated lists to the front of the
list being constructed.
These methods may destructively manipulate the list already contained in the
.meta list-builder
object, and likewise may destructively manipulate the input lists.
They may cause the list being constructed to share substructure with the input
lists.
Additionally, these methods may destructively manipulate the list already
contained in the
.meta list-builder
object without regard for shared structure between that list and inputs
given earlier any of the
.codn pend ,
.codn pend* ,
.code ncon
or
.code ncon*
functions.
The
.code ncon*
function can be called with a single argument
which is an atom. This atom will simply be
installed as the terminating atom of the
list being constructed, if the current list is
an ordinary list.
These methods return the
.meta list-builder
object.
.TP* Example:
.verb
;; Build the list (1 2 3 4 . 5)
(let ((lb (build-list)))
lb.(ncon* (list 1 2))
lb.(ncon (list 3 4))
lb.(ncon 5)
lb.(get))
-> (1 2 3 4 . 5)
.brev
.coNP Method @ get
.synb
.mets << list-builder .(get)
.syne
.desc
The
.code get
method retrieves the list constructed so far by a
.code list-builder
object. It doesn't change the state of the object.
The retrieved list may be passed as an argument
into the construction methods on the same object.
.TP* Examples:
.verb
;; Build the circular list (1 1 1 1 ...)
;; by appending (1) to itself destructively:
(let ((lb (build-list '(1))))
lb.(ncon* lb.(get))
lb.(get))
-> (1 1 1 1 ...)
;; build the list (1 2 1 2 1 2 1 2)
;; by doubling (1 2) twice:
(let ((lb (build-list)))
lb.(add 1 2)
lb.(pend lb.(get))
lb.(pend lb.(get))
lb.(get))
-> (1 2 1 2 1 2 1 2)
.brev
.coNP Methods @ del and @ del*
.synb
.mets << list-builder .(del)
.mets << list-builder .(del*)
.syne
.desc
The
.code del
and
.code del*
methods each remove an element from the list and return it.
If the list is empty, they return
.codn nil .
The
.code del
method removes an element from the front of the list, whereas
.code del*
removes an element from the end of the list.
Note: this orientation is opposite to
.code add
and
.codn add* .
Thus
.code del
pairs with
.code add
to produce FIFO queuing behavior.
.coNP Macros @ build and @ buildn
.synb
.mets (build << form *)
.mets (buildn << form *)
.syne
.desc
The
.code build
and
.code buildn
macros provide a shorthand notation for constructing lists using the
.code list-builder
structure. They eliminate the explicit call to the
.code build-list
function to construct the object, and eliminate the explicit
references to the object.
Both of these macros create a lexical environment in which a
.code list-builder
object is implicitly constructed and bound to a hidden variable.
This lexical environment also provides local functions named
.codn add ,
.codn add* ,
.codn pend ,
.codn pend* ,
.codn ncon ,
.codn ncon* ,
.codn get ,
.code del
and
.codn del* ,
which mimic the
.code list-builder
methods, but operate implicitly on this hidden variable, so that
the object need not be mentioned as an argument.
With the exception of
.codn get ,
.code del
and
.codn del* ,
the local functions return
.codn nil ,
unlike like the same-named
.code list-builder
methods, which return the
.code list-builder
object.
In this lexical environment, each
.meta form
is evaluated in order.
When the last
.meta form
is evaluated,
.code build
returns the constructed list, whereas
.code buildn
returns the value of the last
.metn form .
If no forms are enclosed, both macros return
.codn nil .
Note: because the local function
.code del
has the same name as a global macro, it is implemented as a
.code macrolet.
Inside a
.code build
or
.codn buildn ,
if
.code del
is invoked with no arguments, then it denotes a call to the
.code list-builder
.code del
method. If invoked with an argument, then it resolves to the global
.code del
macro for deleting a place.
.TP* Examples:
.verb
;; Build the circular list (1 1 1 1 ...)
;; by appending (1) to itself destructively:
(build
(add 1)
(ncon* (get))) -> (1 1 1 1 ...)
;; build the list (1 2 1 2 1 2 1 2)
;; by doubling (1 2) twice:
(build
(add 1 2)
(pend (get))
(pend (get))) -> (1 2 1 2 1 2 1 2)
;; build a list by mapping over the local
;; add function:
(build [mapdo add (range 1 3)]) -> (1 2 3)
;; breadth-first traversal of nested list;
(defun bf-map (tree visit-fn)
(buildn
(add tree)
(whilet ((item (del)))
(if (atom item)
[visit-fn item]
(each ((el item))
(add el))))))
(let (flat)
(bf-map '(1 (2 (3 4 (5))) ((6 7) 8)) (do push @1 flat))
(nreverse flat))
-> (1 2 8 3 4 6 7 5)
.brev
.SS* Permutations and Combinations
.coNP Function @ perm
.synb
.mets (perm < seq <> [ len ])
.syne
.desc
The
.code perm
function returns a lazy list which consists of all
length
.meta len
permutations of formed by items taken from
.metn seq .
The permutations do not use any element of
.meta seq
more than once.
Argument
.metn len ,
if present, must be a positive integer, and
.meta seq
must be a sequence.
If
.meta len
is not present, then its value defaults to the length of
.metn seq :
the list of the full permutations of the entire sequence is returned.
The permutations in the returned list are sequences of the same kind as
.codn seq .
If
.meta len
is zero, then a list containing one permutation is returned, and that
permutation is of zero length.
If
.meta len
exceeds the length of
.metn seq ,
then an empty list is returned,
since it is impossible to make a single nonrepeating permutation that
requires more items than are available.
The permutations are lexicographically ordered.
.coNP Function @ rperm
.synb
.mets (rperm < seq << len )
.syne
.desc
The
.code rperm
function returns a lazy list which consists of all the repeating
permutations of length
.meta len
formed by items taken from
.metn seq .
"Repeating" means that the items from
.meta seq
can appear more than
once in the permutations.
The permutations which are returned are sequences of the same kind as
.metn seq .
Argument
.meta len
must be a nonnegative integer, and
.meta seq
must be a sequence.
If
.meta len
is zero, then a single permutation is returned, of zero length.
This is true regardless of whether
.meta seq
is itself empty.
If
.meta seq
is empty and
.meta len
is greater than zero, then no permutations are
returned, since permutations of a positive length require items, and the
sequence has no items. Thus there exist no such permutations.
The first permutation consists of
.meta le
repetitions of the first element of
.metn seq .
The next repetition, if there is one, differs from the first
repetition in that its last element is the second element of
.metn seq .
That is to say, the permutations are lexicographically ordered.
.TP* Examples:
.verb
(rperm "01" 3) -> ("000" "001" "010" "011"
"100" "101" "110" "111")
(rperm #(1) 3) -> (#(1 1 1))
(rperm '(0 1 2) 2) -> ((0 0) (0 1) (0 2) (1 0)
(1 1) (1 2) (2 0) (2 1) (2 2))
.brev
.coNP Function @ comb
.synb
.mets (comb < seq << len )
.syne
.desc
The
.code comb
function returns a lazy list which consists of all
length
.meta len
nonrepeating combinations formed by taking items taken from
.metn seq .
"Nonrepeating combinations" means that the combinations do not use any
element of
.meta seq
more than once. If
.meta seq
contains no duplicates, then
the combinations contain no duplicates.
Argument
.meta len
must be a nonnegative integer, and
.meta seq
must be a sequence or a hash table.
The combinations in the returned list are objects of the same kind as
.metn seq .
If
.meta len
is zero, then a list containing one combination is returned, and that
combination is of zero length.
If
.meta len
exceeds the number of elements in
.metn seq ,
then an empty list is returned, since it is impossible to make a single
nonrepeating combination that requires more items than are available.
If
.meta seq
is a sequence, the returned combinations are lexicographically ordered.
This requirement is not applicable when
.meta seq
is a hash table.
.TP* Example:
.verb
;; powerset function, in terms of comb.
;; Yields a lazy list of all subsets of s,
;; expressed as sequences of the same type as s.
(defun powerset (s)
(mappend* (op comb s) (range 0 (length s))))
.brev
.coNP Function @ rcomb
.synb
.mets (rcomb < seq << len )
.syne
.desc
The
.code comb
function returns a lazy list which consists of all
length
.meta len
repeating combinations formed by taking items taken from
.metn seq .
"Repeating combinations" means that the combinations can use
an element of
.meta seq
more than once.
Argument
.meta len
must be a nonnegative integer, and
.meta seq
must be a sequence.
The combinations in the returned list are sequences of the same kind as
.metn seq .
If
.meta len
is zero, then a list containing one combination is returned, and that
combination is of zero length. This is true even if
.meta seq
is empty.
If
.meta seq
is empty, and
.meta len
is nonzero, then an empty list is returned.
The combinations are lexicographically ordered.
.SS* Macros
Because \*(TL supports structural macros, \*(TX processes \*(TL expressions in
two separate phases: the expansion phase and the evaluation/compilation phase.
During the expansion phase, a top-level expression is recursively traversed,
and all macro invocations in it are expanded. The result is a transformed
expression which contains only function calls and invocations of special
operators. This expanded form is then evaluated or compiled, depending on the
situation.
Macro invocations are compound forms and whose operator symbol has a macro
definition in scope. A macro definition is a kind of function which operates
on syntax during macro-expansion, called upon to calculate a transformation of
the syntax. The return value of a macro replaces its invocation, and is
traversed to look for more opportunities for macro expansion.
Macros differ from ordinary functions in three ways: they are called
at macro-expansion time, they receive pieces of unevaluated syntax as their
arguments, and their parameter lists are macro parameter lists which
support destructuring, as well as certain special parameters.
\*(TL also supports symbol macros. A symbol macro definition associates
a symbol with an expansion. When that symbol appears as a form, the
macro-expander replaces it with the expansion.
\*(TX source files are treated somewhat differently with regard to macro
expansion compared to \*(TL. When \*(TL forms are read from a file by
.code load
or
.code compile
or read by the interactive listener, each form is expanded and evaluated
or compiled before the subsequent form is processed. In contrast,
when a \*(TX file is loaded, expansion of the Lisp forms are its arguments
takes place during the parsing of the entire source file, and is complete
for the entire file before any of the code is executed.
.NP* Macro parameter lists
\*(TX macros support destructuring, similarly to Common Lisp macros.
This means that macro parameter lists are like function argument lists,
but support nesting. A macro parameter list can specify a nested parameter
list in every place where an argument symbol may appear. For instance,
consider this macro parameter list:
.verb
((a (b c)) : (c frm) ((d e) frm2 de-p) . g)
.brev
The top-level of this nested form has the structure
.mono
.meti \ \ >> ( I : < J < K . << L )
.onom
in which we can identify the major constituent positions as
.metn I ,
.metn J ,
.meta K
and
.metn L .
The constituent at position
.meta I
is the mandatory parameter
.codn "(a (b c))" .
Position
.meta J
holds the optional parameter
.code c
(with default init form
.codn frm ).
At
.meta K
is found the optional parameter
.code "(d e)"
(with default init form
.code frm2
and presence-indicating variable
.codn de-p ).
Finally, the parameter in the dot position
.meta L
is
.codn g ,
which captures trailing arguments.
Obviously, some of the parameters are compound expressions rather
than symbols:
.code "(a (b c))"
and
.codn "(d e)" .
These compounds express nested macro parameter lists.
Nested macro parameter lists recursively match the corresponding structure
in the argument object. For instance if a simple argument would capture
the structure
.code "(1 (2 3))"
then we can replace the argument with the nested argument list
.code "(a (b c))"
which destructures the
.code "(1 (2 3))"
such that the parameters
.codn a ,
.code b
and
.code c
will end up bound
to
.codn 1 ,
.code 2
and
.codn 3 ,
respectively.
Nested macro parameter lists have all the features of the top-level
macro parameter lists: they can have optional arguments with default
values, use the dotted position, and contain the
.codn :env ,
.code :whole
and
.code :form
special parameters, which are described below.
In nested parameter lists, the binding strictness is relaxed for optional
parameters. If
.code "(a (b c))"
is optional, and the argument is, say,
.codn (1) ,
then
.code a
gets
.codn 1 ,
and
.code b
and
.code c
receive
.codn nil .
Macro parameter lists also supports three special keywords, namely
.codn :env ,
.code :whole
and
.codn :form .
The parameter list
.code "(:whole x :env y :form z)"
will bind parameter
.code x
to the entire
macro parameter list, bind parameter
.code y
to the macro environment and bind parameter
.code z
to the entire macro form (the original compound form used to invoke the
macro).
The
.codn :env ,
.code :whole
and
.code :form
notations can occur anywhere in a macro parameter list, other than
to the right of the consing dot. They can be used in nested
macro parameter lists also. Note that in a nested macro
parameter list,
.code :form
and
.code :env
do not change meaning: they bind the same object as they would in
the top-level of the macro parameter list.
However the
.code :whole
parameter inside has a restricted scope in a nested parameter
list: its parameter will capture just that part of the argument material which
matches that parameter list, rather than the entire argument list.
The processing of macro parameter lists omits the feature that when the
.code :
(colon) keyword symbol is given as the argument to an optional parameter,
that argument is treated as a missing argument.
This special logic is implemented only
in the function argument passing mechanism, not in the binding of macro
parameters to object structure. If the colon symbol appears in the object
structure and is matched against an optional parameter, it is an
ordinary value. That parameter is considered present, and takes on
the colon symbol as its value.
.TP* "Dialect Note:"
In ANSI Common Lisp, the lambda list keyword
.code &whole
binds its corresponding variable to the entire macro form, whereas
\*(TL's
.code :whole
binds its variable only to the arguments of the macro form.
Note, however, that ANSI CL distinguishes between destructuring and
macro lambda lists, and the
.code &whole
parameter has a different behavior in each. Under
.codn destructuring-bind ,
the
.code &whole
parameter receives just the arguments, just like the behavior
of \*(TL's
.code :whole
parameter.
\*(TL does not distinguish between destructuring and
macro lambda lists;
they are the same and behave the same way. Thus
.code :whole
is treated the same way in macros as in
.code tree-bind
and related binding operators: it binds just the arguments
to the parameter. \*(TL has the special parameter
.code :form
by means of which macros can access their invoking form.
This parameter is also supported in
.code tree-bind
and binds to the entire
.code tree-bind
form.
.coNP Operator @ macro-time
.synb
.mets (macro-time << form *)
.syne
.desc
The
.code macro-time
operator has a syntax similar to the
.code progn
operator. Each
.meta form
is evaluated from left to right, and the resulting value is that of the last
form.
The special behavior of
.code macro-time
is that the evaluation takes place during
the expansion phase, rather than during the evaluation phase.
Also,
.code macro-time
macro-expands each
.meta form
and evaluates it before processing the next
.meta form
in the same way. Thus, for instance, if a
.meta form
introduces a global definition, that definition will be visible not
only during the evaluation of a subsequent
.metn form ,
but also during its macro-expansion time.
During the expansion phase, all
.code macro-time
expressions which occur in a context
that calls for evaluation are evaluated, and replaced by their quoted values.
For instance
.code "(macro-time (list 1 2 3))"
evaluates
.code "(list 1 2 3)"
to the object
.code "(1 2 3)"
and the entire
.code macro-time
form is replaced by that value, quoted:
.codn "'(1 2 3)" .
If the form is evaluated again at evaluation-time, the resulting value will be
that of the quote, in this case
.codn "(1 2 3)" .
.code macro-time
forms do not see the surrounding lexical environment; they see only
global function and variable bindings and macros.
Note:
.code macro-time
supports techniques that require a calculation to be performed in the
environment where the program is being compiled, and inserting the result of
that calculation as a literal into the program source. Possibly, the
calculation can have some useful effect in that environment, or use
as an input information that is available in that environment.
The
.code load-time
operator also inserts a calculated value as a de facto
literal into the program, but it performs that calculation in the
environment where the compiled file is being loaded.
The two operators may be considered complementary in this sense.
Consider the source file:
.verb
(defun host-name-c () (macro-time (uname).nodename))
(defun host-name-l () (load-time (uname).nodename))
.brev
If this is compiled via
.codn compile-file ,
the
.code uname
call in
.code host-name-c
takes place when it is macro-expanded. Thereafter, the compiled version
of the function returns the name of the machine where the
compilation took place, no matter in what environment it is subsequently
loaded and called.
In contrast, the compilation of
.code host-name-l
arranges for that function's
.code uname
call to take place just one time, whenever the compiled file is loaded.
Each time the function is subsequently called, it will
return the name of the machine where it was loaded, without making
any additional calls to
.codn uname .
The
.code macro-time
operator can occasionally be required in order for some constructs to evaluate
or compile. One way that occurs is when a construct that is being fully
expanded itself defines a macro which is later required in that same construct.
For example:
.verb
(progn (defmacro mac () 42) (mac))
.brev
This specific example actually works under
.code eval
or file compilation, because in that situation it isn't fully expanded
all at once. When
.code eval
and
.code compile-file
process a top-level form that is a
.codn progn ,
they treat its argument forms as individual, separate top-level forms. In
general, \*(TL is designed in such a way as to not to require, in most ordinary
programs, extra verbiage to tell the compiler or evaluator that certain
definitions are required by macros. However, somewhat unusual situations can
arise which are not handled in this way.
Also,
.codn macro-time ,
or the related
.code @(mdo)
directive, can be occasionally necessary in \*(TX
queries, which are parsed and subject to macro-expansion in their entirety
before being executed.
.coNP Operator @ defmacro
.synb
.mets (defmacro < name
.mets \ \ \ \ \ \ \ \ \ <> ( param * [: << opt-param * ] [. < rest-param ])
.mets \ \ << body-form *)
.syne
.desc
The
.code defmacro
operator is evaluated at expansion time. It defines a
macro-expander function under the name
.metn name ,
effectively creating a new operator.
Note that the above syntax synopsis describes only the canonical
parameter syntax which remains after parameter list macros are
expanded. See the section Parameter List Macros.
Note that the parameter list is a macro parameter list, and not a
function parameter list. This means that each
.meta param
and
.meta opt-param
can be not only a symbol, but it can itself be a parameter list.
The corresponding argument is then treated as a structure which
matches that parameter list. This nesting of parameter lists
can be carried to an arbitrary depth.
A macro is called like any other operator, and resembles a function. Unlike in
a function call, the macro receives the argument expressions themselves, rather
than their values. Therefore it operates on syntax rather than on values.
Also, unlike a function call, a macro call occurs in the expansion phase,
rather than the evaluation phase.
The return value of the macro is the macro expansion. It is substituted in
place of the entire macro call form. That form is then expanded again;
it may itself be another macro call, or contain more macro calls.
A global macro defined using
.code defmacro
may decline to expand a macro form. Declining to expand is achieved by
returning the original unexpanded form, which may be captured using the
.code :form
parameter. When a global macro declines to expand a form, the form is
taken as-is. At evaluation time, it will be treated as a function call.
Note: when a local macro defined by
.code macrolet
declines, more complicated requirements apply; see the description of
.codn macrolet .
.TP* "Dialect Notes:"
A macro in the global namespace introduced by
.code defmacro
may coexist with a function of the same name introduced by
.codn defun .
This is not permitted in ANSI Common Lisp.
ANSI Common Lisp doesn't describe the concept of declining to expand, except in
the area of compiler macros. Since TXR Lisp allows global macros and functions
of the same name to coexist, ordinary macros can be used to optimize functions
in a manner similar to Common Lisp compiler macros. A macro can be written
of the same name as a function, and can optimize certain cases of the function
call by expanding them to some alternative syntax. Cases which it doesn't
optimize are handled by declining to expand, in which case the form remains
as the original function call.
.TP* Example:
.verb
;; dolist macro similar to Common Lisp's:
;;
;; The following will print 1, 2 and 3
;; on separate lines:
;; and return 42.
;;
;; (dolist (x '(1 2 3) 42)
;; (format t "~s\en" x))
(defmacro dolist ((var list : result) . body)
(let ((i (gensym)))
^(for ((,i ,list)) (,i ,result) ((set ,i (cdr ,i)))
(let ((,var (car ,i)))
,*body))))
.brev
.coNP Operator @ macrolet
.synb
.mets (macrolet >> ({( name < macro-style-params
.mets \ \ \ \ \ \ \ \ \ \ \ \ \ \ << macro-body-form *)}*)
.mets \ \ << body-form *)
.syne
.desc
The
.code macrolet
binding operator extends the macro-time lexical environment
by making zero or more new local macros visible.
The
.code macrolet
symbol is followed by a list of macro definitions.
Each definition is a form which begins with a
.metn name ,
followed by
.meta macro-style-params
which is a macro parameter list, and zero or more
.metn macro-body-form s.
These macro definitions are similar
to those globally defined by the
.code defmacro
operator, except that they
are in a local environment.
The macro definitions are followed by optional
.metn body-forms .
The macros specified in the definitions are visible to these
forms.
Forms inside the macro definitions such as the
.metn macro-body-form s,
and initializer forms appearing in the
.meta macro-style-params
are subject
to macro-expansion in a scope in which none of the new macros being
defined are yet visible. Once the macro definitions are themselves
macro-expanded, they are placed into a new macro environment, which
is then used for macro expanding the
.metn body-form s.
A
.code macrolet
form is fully processed in the expansion phase of a form, and is
effectively replaced by
.code progn
form which contains expanded versions of
.metn body-form s.
This expanded structure shows no evidence that any
macrolet forms ever existed in it. Therefore, it is impossible for the code
evaluated in the bodies and parameter lists of
.code macrolet
macros to have any visibility to any surrounding lexical variable bindings,
which are only instantiated in the evaluation phase, after expansion is done
and macros no longer exist.
A local macro defined using
.code defmacro
may decline to expand a macro form. Declining to expand is achieved by returning the original
unexpanded form, which may be captured using the
.code :form
parameter. When a local macro declines to expand a form, the macro definition
is temporarily hidden, as if it didn't exist in the lexical scope. If another
macro of the same name is thereby revealed (a global macro or another local macro
at a shallower nesting level), then an expansion is tried with that macro. If
no such macro is revealed, or if a lexical function binding of that name is
revealed, then no expansion takes place; the original form is taken as-is.
When another macro is tried, the process repeats, resulting in a search which
proceeds as far as possible through outer lexical scopes and finally the
global scope.
.coNP Function @ macro-form-p
.synb
.mets (macro-form-p < obj <> [ env ])
.syne
.desc
The
.code macro-form-p
function returns
.code t
if
.meta obj
represents the syntax of
a form which is a macro form: either a compound macro or a symbol macro.
Otherwise it returns
.codn nil .
A macro form is one that will transform under
.code macroexpand-1
or
.codn macroexpand ;
an object which isn't a macro form will not undergo expansion.
The optional
.meta env
parameter is a macroexpansion environment.
A macroexpansion environment is passed down to macros and can be received
via their special
.code :env
parameter.
.meta env
is used by
.code macro-form-p
to determine whether
.meta obj
is a macro in a lexical macro environment.
If
.meta env
is not specified or is
.codn nil ,
then
.code macro-form-p
only recognizes global macros.
.TP* Example:
.verb
;; macro which translates to 'yes if its
;; argument is a macro from, or otherwise
;; transforms to the form 'no.
(defmacro ismacro (:env menv form)
(if (macro-form-p form menv)
''yes ''no))
(macrolet ((local ()))
(ismacro (local))) ;; yields yes
(ismacro (local)) ;; yields no
(ismacro (ismacro foo)) ;; yields yes
.brev
During macro expansion, the global macro
.code ismacro
is handed the macro-expansion environment
via
.codn ":env menv" .
When the macro is invoked within the macrolet,
this environment includes the macro-time lexical scope in which the
.code local
macro is defined. So when global checks whether the argument form
.code (local)
is a macro, the conclusion is yes: the (local) form is a macro
call in that environment:
.code macro-form-p
yields
.codn t .
When
.code "(global (local))"
is invoked outside of the macrolet, no local macro is visible is
there, and so
.code macro-form-p
yields
.codn nil .
.coNP Functions @ macroexpand-1 and @ macroexpand
.synb
.mets (macroexpand-1 < obj <> [ env ])
.mets (macroexpand < obj <> [ env ])
.syne
.desc
If
.meta obj
is a macro form (an object for which
.code macro-form-p
returns
.codn t ),
these functions expand the macro form and return the expanded form.
Otherwise, they return
.metn obj .
.code macroexpand-1
performs a single expansion, expanding just the macro
that is referenced by the symbol in the first position of
.metn obj ,
and returns the expansion. That expansion may itself be a macro form.
.code macroexpand
performs an expansion similar to
.codn macroexpand-1 .
If the result is
a macro form, then it expands that form, and keeps repeating this process
until the expansion yields a non-macro-form. That non-macro-form is then
returned.
The optional
.meta env
parameter is a macroexpansion environment.
A macroexpansion environment is passed down to macros and can be received
via their special
.code :env
parameter. The environment they receive is their
lexically apparent macro-time environment in which local macros may be
visible. A macro can use this environment to "manually" expand some
form in the context of that environment.
.TP* Example:
.verb
;; (rem-num x) expands x, and if x begins with a number,
;; it removes the number and returns the resulting
;; form. Otherwise, it returns the entire form.
(defmacro rem-num (:env menv some-form)
(let ((expanded (macroexpand some-form menv)))
(if (numberp (car expanded))
(cdr expanded)
some-form)))
(macrolet ((foo () '(1 list 42))
(bar () '(list 'a)))
(list (rem-num (foo)) (rem-num (bar))))
--> ((42) (a))
.brev
The
.code rem-num
macro is able to expand the
.code (foo)
and
.code (bar)
forms it receives as
the
.code some-form
argument, even though these forms use local macro that are only
visible in their local scope. This is thanks to the macro
environment passed to
.codn rem-num .
It is correctly able to work with the
expansions
.code "(1 list 42)"
and
.code "(list 'a)"
to produce
.code "(list 42)"
and
.code "(list 'a)"
which evaluate to
.code 42
and
.code a
respectively.
.coNP Functions @ macroexpand-1-lisp1 and @ macroexpand-lisp1
.synb
.mets (macroexpand-1-lisp1 < obj <> [ env ])
.mets (macroexpand-lisp1 < obj <> [ env ])
.syne
.desc
The
.code macroexpand-1-lisp1
and
.code macroexpand-lisp1
functions closely resemble, respectively,
.code macroexpand-1
and
.codn macroexpand .
The argument and return value syntax and semantics is almost
identical, except for one difference. These functions consider argument
.meta obj
to be syntax in a Lisp-1 evaluation context, such as any argument
position of the
.code dwim
operator, or the equivalent DWIM Brackets notation.
This makes a difference because in a Lisp-1 evaluation context, an
inner function binding is able to shadow an outer symbol macro binding
of the same name.
The requirements about this language area are given in more
detail in the description of the
.code dwim
operator.
Note: the
.code macroexpand-lisp1
function is useful to the implementor of a macro whose semantics requires
one or more argument forms to be treated in a Lisp-1 context, in situations
when such a macro needs to itself expand the material, rather than merely
insert it as-is into the output code template.
.coNP Functions @ expand and @ expand*
.synb
.mets (expand < form <> [ env ])
.mets (expand* < form <> [ env ])
.syne
.desc
The functions
.code expand
and
.code expand*
both perform a complete expansion of
.meta form
in the macro-environment
.metn env ,
and return that expansion.
If
.meta env
is omitted, the expansion takes place in the global environment in
which only global macros are visible.
The returned object is a structure that
is devoid of any macro calls. Also, all
.code macrolet
and
.code symacrolet
blocks in form
.meta form
are removed in the returned structure, replaced by their fully
expanded bodies.
The difference between
.code expand
and
.code expand*
is that
.code expand
suppresses expansion-time deferred warnings (exceptions of type
.codn defr-warning ),
issued for unbound variables or functions.
To suppress a warning means to intercept the warning exception with a handler
which throws a
.code continue
exception to resume processing.
What this requirement means is that if unbound functions or variables
occur in the
.meta form
being expanded by expand, the warning is effectively squelched. Rationale:
.code expand
is may be used by macros for expanding fragments which contain references to
variables or functions which are not defined in those fragments.
.coNP Function @ expand-with-free-refs
.synb
.mets (expand-with-free-refs < form >> [ inner-env <> [ outer-env ]])
.syne
.desc
The
.code expand-with-free-refs
form performs a full expansion of
.metn form ,
as if by the
.code expand
function and returns a list containing that expansion, plus four additional
items which provide information about variable and function references which
occur in
.metn form .
If both
.meta inner-env
and
.meta outer-env
are provided, then it is expected that
.meta inner-env
is lexically nested within
.metn outer-env .
Note: it is not required that
.meta outer-env
be the immediate parent of
.metn inner-env .
Note: a common usage situation is that
.meta outer-env
is the environment of the invocation of a "parent" macro which generates a form
that contains local macros. The bodies of those local macros use
.codn expand-with-free-refs ,
specifying their own environment as
.meta inner-env
and that of their generating "parent" as
.metn outer-env .
In detail, the five items of the returned list are
.mono
.meti >> ( expansion < fv-inner < ff-inner < fv-outer << ff-outer )
.onom
whose descriptions are:
.RS
.meIP < expansion
The full expansion of
.metn form ,
containing no macro invocations, or
.code symacrolet
or
.code macrolet
forms.
.meIP < fv-inner
A list of the free variables which occur in
.meta form
relative to the
.meta inner-env
environment. That is to say, variables that are not bound inside
.meta form
and are not also bound in
.metn inner-env .
If
.meta inner-env
is omitted, then these are the absolutely free variables
occurring in
.metn form .
.meIP < ff-inner
Exactly like
.meta fv-inner
but informing about function bindings rather than variables.
.meIP < fv-outer
A list of the variables which which occur in
.meta form
which would be free if the environments between
.meta inner-env
and
.meta outer-env
(including the former, excluding the latter)
were removed from consideration. A more detailed description of this semantics
is given below. If
.meta outer-env
is omitted, then these are the absolutely free variables
occurring in
.metn form ,
ignoring the
.metn inner-env .
.meIP < ff-outer
Exactly like
.meta fv-outer
but informing about function bindings rather than variables.
.RE
.IP
The semantics of the treatment of
.meta inner-env
and
.meta outer-env
in the calculation of
.meta fv-outer
and
.meta ff-outer
is as follows. A new environment
.meta diff-env
is calculated from these two environments, and
.meta form
is expanded in this environment. Variables and functions occurring in
.meta form
which are not bound in
.meta diff-env
are listed as
.meta fv-outer
and
.metn ff-outer .
This
.meta diff-env
is calculated as follows. First
.meta diff-env
is initialized as a copy of
.metn outer-env .
Then, all environments below
.meta outer-env
down to
.meta inner-env
are examined for bindings which shadow bindings in
.metn diff-env .
Those shadows are removed from
.metn diff-env .
Therefore, what remains in
.meta diff-env
are those bindings from
.meta outer-env
that are
.I not
shadowed by the environments between
.meta inner-env
and
.metn outer-env .
Within each of the lists of variables returned by
.codn expand-with-free-refs ,
the order of the variables is not specified.
.TP* Example:
Suppose that
.code mac
is a macro which somehow has access to the two indicated lexical environments
in the following code snippet:
.verb
(let (a c) ;; <- outer-env
(let (b)
(let (c) ;; <- inner-env
(mac (list a b c d)))))
.brev
Suppose that
.code mac
invokes the
.code expand-with-free-refs
function, passing in the
.code "(list a b c d)"
argument form as
.code form
and two macro-time environment objects corresponding to the indicated
environments.
Then the following object shall be a correct return value of
.codn expand-with-free-refs :
.verb
((list a b c d) (d) nil (d c b) nil)
.brev
A complete code example of this is given below.
Other correct return values are possible due to permitted variations in the
order of the variables within the four lists. For instance, instead of
.code "(d c b)"
the list
.code "(c b d)"
may appear.
The
.meta fv-inner
list is
.code "(d)"
because this is the only variable that occurs in
.code "(list a b c d)"
which is free with regard to
.metn inner-env .
The
.codn a ,
.code b
and
.code c
variables are not listed because they appear bound inside
.metn inner-env .
The reported
.meta fv-outer
list is
.code "(b c d)"
because the form is considered against
.meta diff-env
which is formed by removing the shadowing bindings from
.metn outer-env .
The difference between
.code "(a c)"
and
.code "(b c)"
is
.code a
and so the form is considered in an environment containing the binding
.code a
which leaves
.code "(b c d)"
free.
The following is a complete code sample demonstrating the above
descriptions:
.verb
;; Given this macro:
(defmacro bigmac (:env out-env big-form)
^(macrolet ((mac (:env in-env little-form)
^',(expand-with-free-refs
little-form in-env ,out-env)))
,big-form))
(let (a c) ;; <- outer-env, surrounding bigmac
(bigmac
(let (b)
(let (c) ;; <- inner-env, surrounding mac
(mac (list a b c d))))))
--> ((list a b c d) (d) nil (d c b) nil)
.brev
Note: this information is useful because a set difference can be calculated
between the two reported sets. The set difference between the
.meta fv-outer
variables
.code "(b c d)"
and the
.meta fv-inner
variables
.code "(d)"
is
.codn "(b c)" .
That set difference
.code "(b c)"
is significant because it precisely informs about the
.I bound
variables which occur in
.code "(list a b c d)"
which appear bound in
.metn inner-env ,
but are not bound due to a binding coming from
.metn outer-env .
In the above example, these are the variables enclosed in the
.code bigmac
macro, but external to the inner
.code mac
macro.
The variable
.code d
is not listed in
.code "(b c)"
because it is not a bound variable.
The variable
.code a
is not in
.code "(b c)"
because though it is bound in
.metn inner-env ,
that binding comes from
.metn outer-env .
The upshot of this logic is that it allows a macro to inspect a form in order
to discover the identities of the variables and functions which are used inside
that form, whose definitions come from a specific, bounded scope surrounding
that form.
.coNP Functions @ lexical-var-p and @ lexical-fun-p
.synb
.mets (lexical-var-p < env << form )
.mets (lexical-fun-p < env << form )
.syne
.desc
These two functions are useful to macro writers. They are intended
to be called from the bodies of macro expanders, such as the bodies of
.code defmacro
or
.code macrolet
forms. The
.meta env
argument is a macro-time environment, which is available to macros
via the special
.code :env
parameter. Using these functions, a macro can enquire whether
a given
.meta form
is a symbol which has a variable binding or a function binding
in the local lexical environment.
This information is known during macro expansion. The macro expander
recognizes lexical function and variable bindings, because these
bindings can shadow macros.
Special variables are not lexical. The function
.code lexical-var-p
returns
.code nil
if
.meta form
satisfies
.code special-var-p
function, indicating that it is the name of a special variable.
The
.code lexical-var-p
function also returns
.code nil
for global lexical variables. If
.meta form
is a symbol for which only a global lexical variable binding is apparent,
.code lexical-var-p
returns
.codn nil .
Testing for the existence for a global variable can be done using
.codn boundp ;
if a symbol is
.code boundp
but not
.codn special-var-p ,
then it is a global lexical variable.
Similarly,
.code lexical-fun-p
returns
.code nil
for global functions.
.TP* Example:
.verb
;;
;; this macro replaces itself with :lexical-var if its
;; argument is a lexical variable, :lexical-fun if
;; its argument is a lexical function, or with
;; :not-lex-fun-var if neither is the case.
;;
(defmacro classify (sym :env e)
(cond
((lexical-var-p e sym) :lexical-var)
((lexical-fun-p e sym) :lexical-fun)
(t :not-lex-fun-var)))
;;
;; Use classify macro above to report classification
;; of the x, y and f symbols in the given scope
;;
(let ((x 1) (y 2))
(symacrolet ((y x))
(flet ((f () (+ 2 2)))
(list (classify x) (classify y) (classify f)))))
--> (:lexical-var :not-lex-fun-var :lexical-fun)
;; Locally bound specials are not lexical
(let ((*stdout* *stdnull*))
(classify *stdout*))
--> :not-lex-fun-var
.brev
.TP* Note:
These functions do not call
.code macroexpand
on the form. In most cases, it is necessary for the macro writers
to do so. Not that in the above example, symbol
.code y
is classified as neither a lexical function nor variable.
However, it can be macro-expanded to
.code x
which is a lexical variable.
.coNP Function @ lexical-lisp1-binding
.synb
.mets (lexical-lisp1-binding < env << symbol )
.syne
.desc
The
.code lexical-lisp1-binding
function inspects the macro-time environment
.meta env
to determine what kind of binding, if any, does
.meta symbol
have in that environment, from a Lisp-1 perspective.
That is to say, it considers function bindings, variable bindings
and symbol macro bindings to be in a single name space and finds
the innermost binding of one of these types for
.metn symbol .
If such a binding is found, then the function returns one of
the three keyword symbols
.codn :var ,
.codn :fun ,
or
.codn :symacro .
If no such lexical binding is found, then the function
returns
.codn nil .
Note that a
.code nil
return doesn't mean that the symbol doesn't have a lexical binding. It could
have an operator macro lexical binding (a macro binding in the function
namespace established by
.codn macrolet ).
.coNP Operator @ defsymacro
.synb
.mets (defsymacro < sym << form )
.syne
.desc
A
.code defsymacro
form introduces a symbol macro. A symbol macro consists of a binding
between a symbol
.meta sym
and and a
.metn form .
The binding denotes the form itself, rather than its value.
The
.meta form
argument is not subject to macro expansion; it is associated with
.meta sym
in its unexpanded state, as it appears in the
.code defmacro
form.
The
.code defsymacro
form must be evaluated for its defining to take place; therefore,
the definition is not available in the top-level form which contains the
.code defsymacro
invocation; it becomes available to a subsequent top-level form.
Subsequent to the evaluation of the
.code defsymacro
definition, whenever the macro expander encounters
.meta sym
sym as a form, it replaces it by
.metn form .
After this replacement takes place,
.meta form
itself is then processed for further replacement of macros and
symbol macros.
Symbol macros are also recognized in contexts
where
.meta sym
denotes a place which is the target of an assignment operation
like
.code set
and similar.
Note: if a symbol macro expands to itself directly, expansion stops. However,
if a symbol macro expands to itself through a chain of expansions,
runaway expansion-time recursion will occur.
If a global variable exists by the name
.metn sym ,
then
.code defsymacro
first removes that variable from the global environment, and if that
variable is special, the symbol's special marking is removed.
.code defsymacro
doesn't alter the dynamic binding of a special variable. Any such
a binding remains intact.
If
.code defsymacro
is evaluated in a scope in which there is any lexical or dynamic binding
of
.meta sym
in the variable namespace, whether as a variable or macro,
the global symbol macro is shadowed by that binding.
.coNP Operator @ symacrolet
.synb
.mets (symacrolet >> ({( sym << form )}*) << body-form *)
.syne
.desc
The
.code symacrolet
operator binds local, lexically scoped macros that are
similar to the global symbol macros introduced by
.codn defsymacro .
Each
.meta sym
in the bindings list is bound to its corresponding form, creating a
new extension of the expansion-time lexical macro environment.
Each
.meta body-form
is subsequently macro-expanded in this new environment
in which the new symbol macros are visible.
Note: ordinary lexical bindings such as those introduced by let or by
function parameters lists shadow symbol macros. If a symbol
.code x
is bound by nested instances of
.code macrolet
and a
.codn let ,
then the scope enclosed by both
constructs will see whichever of the two bindings is more inner,
even though the bindings are active in completely separate phases of
processing.
From the perspective of the arguments of a
.code dwim
form, lexical function bindings also shadow symbol macros.
This is consistent with the Lisp-1-style name resolution which
applies inside a
.code dwim
form. Lexical operator macros do not shadow
symbol macros under any circumstances.
.coNP Macros @ placelet and @ placelet*
.synb
.mets (placelet >> ({( sym << place )}*) << body-form *)
.mets (placelet* >> ({( sym << place )}*) << body-form *)
.syne
.desc
The
.code placelet
macro binds lexically scoped symbol macros in such
a way that they behave as aliases for places
denoted by place forms.
Each
.meta place
must be an expression denoting a syntactic place. The
corresponding
.meta sym
is established as an alias for the storage location which that place denotes,
over the scope of the
.metn body-form s.
This binding takes place in such a way that each
.meta place
is evaluated exactly once, only in order to determine its
storage location. The corresponding
.meta sym
then serves as an alias for that location, over the
scope of the
.metn body-form s.
This means that whenever
.meta sym
is evaluated, it stands for the value of the storage
location, and whenever a value is apparently stored into
.metn sym ,
it is actually the storage location which receives it.
The
.code placelet*
variant implements an alternative scoping rule, which allows a later
.meta place
form to refer to a
.meta sym
bound to an earlier
.meta place
form. In other words, a given
.meta sym
binding is visible not only to the
.metn body-form s
but also to
.meta place
forms which occur later.
Note: certain kinds of places, notably
.mono
.meti (force << promise )
.onom
expressions, must be accessed before they can be stored,
and this restriction continues to hold when those
places are accessed through
.code placelet
aliases.
Note:
.code placelet
differs from
.code symacrolet
in that the forms themselves are not aliased, but the storage
locations which they denote.
.code "(symacrolet ((x y)) z)"
performs the syntactic substitution of symbol
.code x
by form
.codn y ,
wherever
.code x
appears inside
.code z
as an evaluated form, and is not shadowed by any inner binding.
Whereas
.code "(placelet ((x y)) z)"
generates code which arranges for
.code y
to be evaluated to a storage location, and syntactically replaces occurrences
of
.code x
with a form which directly denotes that storage location,
wherever
.code x
appears inside
.code z
as an evaluated form, and is not shadowed by any inner binding.
Also,
.code x
is not necessarily substituted by a single, fixed form,
as in the case of
.codn symacrolet .
Rather it may be substituted by one kind of form when it
is treated as a pure value, and another kind of form
when it is treated as a place.
Note: multiple accesses to an alias created by
.code placelet
denote multiple accesses to the aliased storage location.
That can mean multiple function calls or array indexing operations and such.
If the target of the alias is
.mono
.meti (read-once << place )
.onom
instead of
.metn place ,
then a single access occurs to fetch the prior value of
.meta place
and stored into a hidden variable. All of the multiple occurrences of the
alias then simply retrieve this cached prior value from the hidden
variable, rather than accessing the place. The
.code read-once
macro is independent of
.code placelet
and separately documented.
.TP* "Example:"
Implementation of
.code inc
using
.codn placelet :
.verb
(defmacro inc (place : (delta 1))
(with-gensyms (p)
^(placelet ((,p ,place))
(set ,p (+ ,p ,delta)))))
.brev
The gensym
.code p
is used to avoid accidental capture of references
emanating from the
.code delta
form.
.coNP Macro @ equot
.synb
.mets (equot << form )
.syne
.desc
The
.code equot
macro ("expand and quote") performs a full expansion of
.code form
in the surrounding macro environment. Then it constructs a
.code quote
form whose argument is the expansion. This quote form is
then returned as the macro replacement for the original
.code equot
form.
.TP* Example:
.verb
(symacrolet ((a (+ 2 2)))
(list (quote a) (equot a) a))
--> (a (+ 2 2) 4)
.brev
Above, the expansion of
.code a
is
.codn "(+ 2 2)" .
Thus the macro call
.code "(equot a)"
expands to
.codn "(quote (+ 2 2))" .
When that is evaluated, it yields
.codn "(+ 2 2)" .
If
.code a
is quoted, then the result is
.codn a :
no expansion or evaluation takes place.
Whereas if
.code a
is presented for evaluation, then not only is it expanded to
.codn "(+ 2 2)" ,
but that expansion is reduced to 4.
The
.code equot
operator is a mongrel of these two semantics: it permits expansion to proceed,
but then suppresses evaluation of the result.
.coNP Operators @, tree-bind @ mac-param-bind and @ mac-env-param-bind
.synb
.mets (tree-bind < macro-style-params < expr << form *)
.mets (mac-param-bind < context-expr
.mets \ \ < macro-style-params < expr << form *)
.mets (mac-env-param-bind < context-expr < env-expr
.mets \ \ < macro-style-params < expr << form *)
.syne
.desc
The
.code tree-bind
operator evaluates
.codn expr ,
and then uses the
resulting value as a counterpart to a macro-style parameter list.
If the value has a tree structure which matches the parameters,
then those parameters are established as bindings, and the
.metn form s,
if any, are evaluated in the scope of those bindings. The value
of the last
.meta form
is returned. If there are no forms,
.code nil
is returned.
Under
.codn tree-bind ,
the value of the
.code :form
available to
.meta macro-style-params
is the
.code tree-bind
form itself.
The
.code mac-param-bind
operator is similar to
.code tree-bind
except that it takes an extra argument,
.metn context-expr .
This argument is an expression which is evaluated. It is expected to
evaluate to a compound form. If an error occurs during binding, the error
diagnostic message is based on information obtained from this form.
By contrast, the
.code tree-bind
operator's error diagnostic refers to the
.code tree-bind
form, which is cryptic if the binding is used for the implementation
of some other construct, hidden from the user of that construct.
In addition,
.meta context-expr
specifies the value for the
.code :form
parameter that
.meta macro-style-params
may refer to.
The
.code mac-env-param-bind
is an extension of
.code mac-param-bind
which takes one more argument,
.codn env-expr ,
before the macro parameters. This expression is evaluated,
and becomes the value of the
.code :env
parameter that
.meta macro-style-params
may refer to.
Under
.code tree-bind
and
.codn mac-param-bind ,
the
.code :env
parameter takes on the value
.codn nil .
Under all three operators, the
.code :whole
parameter takes on the value of
.metn expr .
These operators throw an exception if there is a
structural mismatch between the parameters and the value of
.codn expr .
One way to avoid this exception is to use
.codn tree-case ,
which is based on the conventions of
.codn tree-bind .
There exists no
.code tree-case
analog for
.code mac-param-bind
or
.codn mac-env-param-bind .
.coNP Operator @ tree-case
.synb
.mets (tree-case < expr >> {( macro-style-params << form *)}*)
.syne
.desc
The
.code tree-case
operator evaluates
.meta expr
and matches it against a succession
of zero or more cases. Each case defines a pattern match, expressed as a macro
style parameter list
.metn macro-style-params .
If the object produced by
.meta expr
matches
.metn macro-style-params ,
then the parameters are bound, becoming local variables, and the
.metn form s,
if any, are evaluated in order in the environment in which those variables are
visible. If there are forms, the value of the last
.meta form
becomes the result
value of the case, otherwise the result value of the case is nil.
If the result value of a case is the object
.code :
(the colon symbol), then processing continues with the next case. Otherwise the
evaluation of
.code tree-case
terminates, returning the result value.
If the value of
.meta expr
does not match the
.meta macro-style-params
parameter list of a case, processing continues with the next case.
If no cases match, then
.code tree-case
terminates, returning
.codn nil .
.TP* Example:
.verb
;; reverse function implemented using tree-case
(defun tb-reverse (obj)
(tree-case obj
(() ()) ;; the empty list is just returned
((a) obj) ;; one-element list returned
((a . b) ^(,*(tb-reverse b) ,a)) ;; car/cdr recursion
(a a))) ;; atom is just returned
.brev
Note that in this example, the atom case is placed last, because an
argument list which consists of a symbol is a "catch all" match
that matches any object. We know that it matches an atom, because
the previous
.code "(a . b)"
case matches conses. In general, the order of the cases in
.code tree-case
is important: even more so than the order of cases in a
.code cond
or
.codn caseql .
The one-element list case is unnecessary; it can be removed.
.coNP Macro @ tb
.synb
.mets (tb < macro-style-params << form *)
.syne
.desc
The
.code tb
macro is similar to the
.code lambda
operator but its argument binding is based on a macro-style parameter list.
The name is an abbreviation of
.codn tree-bind .
A
.code tb
form evaluates to a function which takes a variable number of
arguments.
When that function is called, those arguments are taken as a list object which
is matched against
.meta macro-style-params
as if by
.metn tree-bind .
If the match is successful, then the parameters are bound to the
corresponding elements from the argument structure and each successive
.meta form
is evaluated in an environment in which those bindings are visible.
The value of the last
.meta form
is the return value of the function. If there are no forms,
the function's return value is
.codn nil .
The following equivalence holds, where
.code args
should be understood to be a globally unique symbol:
.verb
(tb pattern body ...) <--> (lambda (. args)
(tree-bind pattern args body ...))
.brev
.coNP Macro @ tc
.synb
.mets (tc >> {( macro-style-params << form *)}*)
.syne
.desc
The
.code tc
macro produces an anonymous function whose behavior is closely
based on the
.code tree-case
operator. Its name is an abbreviation of
.codn tree-case .
The anonymous function takes a variable number of arguments.
Its argument list is taken to be the value macro is tested
against the multiple pattern clauses of an implicit
.codn tree-case .
The return value of the function is that of the implied
.codn tree-case .
The following equivalence holds, where
.code args
should be understood to be a globally unique symbol:
.verb
(tc clause1 clause2 ...) <--> (lambda (. args)
(tree-case args
clause1 clause2 ...))
.brev
.coNP Macro @ with-gensyms
.synb
.mets (with-gensyms <> ( sym *) << body-form *)
.syne
.desc
The
.code with-gensyms
evaluates the
.metn body-form s
in an environment in which each variable name symbol
.meta sym
is bound to a new uninterned symbol ("gensym").
.TP* "Example:"
The code:
.verb
(let ((x (gensym))
(y (gensym))
(z (gensym)))
^(,x ,y ,z))
.brev
may be expressed more conveniently using the
.code with-gensyms
shorthand:
.verb
(with-gensyms (x y z)
^(,x ,y ,z))
.brev
.SS* Parameter List Macros
Parameter list macros, also more briefly called
.I "parameter macros"
are an original feature of \*(TL.
If the first element of a function or macro parameter list is a keyword
symbol other than
.codn :env ,
.codn :whole ,
.code :form
or
.code :
(the colon symbol),
it denotes a parameter macro. This keyword symbol is expected to
have a binding in the parameter macro namespace: a global namespace
which associates keyword symbols with parameter list expander
functions.
Expansion of a parameter list macro occurs at macro-expansion
time, when a function's parameter list is traversed by the
macro expander. It takes place as follows.
First, the keyword is removed from the parameter list.
The keyword's binding in the parameter macro namespace is
retrieved. If it doesn't exist, an exception is thrown.
Otherwise, the remaining parameter list is first recursively
processed for more occurrences of parameter macros.
This expansion produces a transformed parameter list,
along with a transformed function body. These two artifacts
are then passed to the transformer function retrieved from
the keyword symbol's binding. The function returns a
further transformed version of the parameter list and
body. These are processed for more parameter macros.
The process terminates when no more expansion is
possible, because a parameter list has been produced
which does not begin with a parameter macro. This
final parameter list and its accompanying body are then
taken in place of the original parameter list and
body.
\*(TL provides a two built-in parameter list macros.
The
.code :key
parameter macro endows a function keyword parameters.
The
.code :match
parameter macro allows a function to be expressed using pattern matching,
which requires the body to consist of pattern-matching clauses.
The implementation of both of these macros is written entirely using this
parameter list macro mechanism, by means of the public
.code define-param-expander
macro.
.coNP Special variable @ *param-macro*
.desc
The variable
.code *param-macro*
holds a hash table which associates keyword symbols with
parameter list expander functions.
The functions are expected to conform to the following
syntax:
.mono
.mets (lambda >> ( params < body < env << form ) << form *)
.onom
The
.meta params
parameter receives the parameter list of the function
which is undergoing parameter expansion. All other
parameter macros have already been expanded.
The
.meta body
parameter receives the list of body forms.
The function is expected to return a
.code cons
cell whose
.code car
contains the transformed parameter list, and whose
.code cdr
contains the transformed list of body forms.
Parameter expansion takes place at macro expansion time.
The
.meta env
parameter receives the macro-expansion-time environment
which surrounds the function being expanded.
Note that this environment doesn't take into account the
parameters themselves; therefore, it is not the correct environment
for expanding macros among the
.meta body
forms. For that purpose, it must be extended with
shadowing entries, the manner of doing which is
undocumented. However
.meta env
may be used directly for expanding init forms
for optional parameters occurring in
.metn params .
The
.meta form
parameter receives the overall function-defining
form that is being processes, such as a
.code defun
or
.code lambda
form. This is intended for error reporting.
A parameter transformer returns the transformed parameter list and body as a
single object: a list whose first element is the parameter list,
and whose remaining elements are the forms of the body. Thus, the following
is a correct null transformer:
.verb
(lambda (params body env form)
(cons params body))
.brev
.coNP Macro @ define-param-expander
.synb
.mets (define-param-expander < name >> ( pvar < bvar : < evar << fvar )
.mets \ \ << form *)
.syne
.desc
The
.code define-param-expander
macro provides syntax for defining parameter macros. Invocations
of this macro expand to code which constructs an anonymous
function and installs it into the
.code *param-macro*
hash table, under the key given by
.metn name .
The
.meta name
parameter's argument should be a keyword symbol that is valid for use
as a parameter macro name.
The
.metn pvar ,
.metn bvar ,
.meta evar
and
.meta fvar
arguments must be symbols suitable for variable
binding. These symbols define the parameters of the
expander function which shall, respectively, receive
the parameter list, body forms, macro environment
and function form. If
.meta evar
is omitted, a symbol generated by the
.code gensym
function is used. Likewise if
.meta fvar
is omitted.
The
.meta form
arguments constitute the body of the expander.
The
.code define-param-expander
form returns
.metn name .
The parameter macro returns the transformed parameter list and body as a
single object: a list whose first element is the parameter list,
and whose remaining elements are the forms of the body.
.TP* Example:
The following example shows the implementation
of a parameter macro
.code :memo
which provides rudimentary memoization.
Using the macro is extremely easy. It is a matter
of simply inserting the
.code :memo
keyword at the front of a function's parameter list.
The function is then memoized.
.verb
(defvarl %memo% (hash :weak-keys))
(defun ensure-memo (sym)
(or (gethash %memo% sym)
(sethash %memo% sym (hash))))
(define-param-expander :memo (param body)
(let* ((memo-parm [param 0..(posq : param)])
(hash (gensym))
(key (gensym)))
^(,param (let ((,hash (ensure-memo ',hash))
(,key (list ,*memo-parm)))
(or (gethash ,hash ,key)
(sethash ,hash ,key (progn ,*body)))))))
.brev
The above
.code :memo
macro may be used to define a memoized Fibonacci function
as follows:
.verb
(defun fib (:memo n)
(if (< n 2)
(clamp 0 1 n)
(+ (fib (pred n)) (fib (ppred n)))))
.brev
All that is required is the insertion of the
.code :memo
keyword.
.coNP Parameter list macro @ :key
.synb
.mets (:key << non-key-param *
.mets \ \ [ -- >> { sym | >> ( sym >> [ init-form <> [ p-sym ]])}* ]
.mets \ \ [ . rest-param ])
.syne
.desc
Parameter list macro
.code :key
injects keyword parameter support into functions and macros.
When
.code :key
appears as the first item in a function parameter list, a special syntax is
recognized in the parameter list. After any required and optional parameters,
the symbol
.code --
(two dashes) may appear. Parameters after this symbol are interpreted
as keyword parameters. After the keyword parameters, a rest parameter
may appear in the usual way as a symbol in the dotted position.
Keyword parameters use the same syntax as optional parameters, except
that if used in a macro parameter list, they do not support
destructuring whereas optional parameters do. That is to say, regardless
whether
.code :key
is used in a function or macro, keyword parameters are symbols.
A keyword parameter takes three possible forms:
.RS
.meIP < sym
A keyword parameter may be specified as a simple symbol
.metn sym .
If the argument for such a keyword parameter is missing,
it takes on the value
.codn nil .
.meIP >> ( sym << init-form )
If the keyword parameter symbol
.meta sym
is enclosed in a list, then the second element of that list
specifies a default value, similarly to the default value for
an optional argument. If the function is called in such a way
that the argument for the parameter is missing, the
.meta init-form
is evaluated and the resulting value is bound to the keyword parameter.
The evaluation takes place in a lexical scope in which the
required and optional parameters are are already visible,
and their values are bound. If there is a
.meta rest-param
it is also visible in this scope, even though in the parameter
list it appears to the left.
.meIP >> ( sym < init-form << p-sym )
The three-element form of the keyword parameter specifies
an additional symbol
.metn p-sym ,
which names an argument that implicitly receives a Boolean
argument indicating the presence of the keyword argument.
If an argument is not passed for the keyword parameter
.metn sym ,
then parameter
.meta sym-p
takes on the value
.codn nil .
If an argument is given for
.metn sym ,
then the
.meta sym-p
argument takes on the value
.codn t .
This mechanism also closely resembles the analogous
one supported in optional arguments. See the previous
paragraph regarding the evaluation scope of
.metn init-form .
.RE
.IP
In a call to a
.codn :key -enabled
function, keyword arguments begin after those arguments which satisfy
all of the required and optional parameters. Keyword arguments consist
of interleaved indicators and values, which are separate arguments.
Thus passing a keyword argument actually requires the passing of two
function arguments: an indicator keyword symbol, followed by the
associated value. The indicator keywords are expected to have the
same symbol name as the defined keyword parameters. For instance, the
indicator-value pair
.code ":xyz 42"
passes the value
.code 42
to a keyword parameter that may be named
.code xyz
in any package: it may be
.code usr:xyz
or
.code mypackage:xyz
and so forth.
Arguments specifying unrecognized keywords are ignored.
If the function has a
.metn rest-param ,
then that parameter receives the keyword arguments as a list.
Since that list contains indicators and values, it is a de facto
property list. In detail, the
.code :key
mechanism generates a regular variadic function which receives the keyword
arguments as the trailing argument list. That function
parses the recognized keyword arguments out of the trailing list, and
binds them to the keyword parameter symbols as local variables. If a
.meta rest-param
parameter is defined, then the entire keyword argument list is available
through that parameter, and the keyword argument parsing logic also refers to
the value of that parameter to gain access to the keyword arguments. If
there is no
.meta rest-param
specified, then the
.code :key
macro adds a
.meta rest-param
using a machine-generated symbol. The argument parsing logic then
refers to the value of that symbol.
.TP* Example:
Define a function
.code fun
with two required arguments
.codn "a b" ,
one optional argument
.codn c ,
two keyword arguments
.code foo
and
.codn bar ,
and a rest parameter
.codn klist :
.verb
(defun fun (:key a b : c -- foo bar . klist)
(list a b c foo bar klist))
(fun 1 2 3 :bar 4) -> (1 2 3 nil 4 (:bar 4))
.brev
Define a function with only keyword arguments, with default expressions and
Boolean indicator params:
.verb
(defun keyfun (:key -- (a 10 a-p) (b 20 b-p))
(list a a-p b b-p))
(keyfun :a 3) -> (3 t 20 nil)
(keyfun :b 4) -> (10 nil 4 t)
(keyfun :c 4) -> (10 nil 20 nil)
(keyfun) -> (10 nil 20 nil)
.brev
.SS* Mutation of Syntactic Places
.coNP Macro @ set
.synb
.mets (set >> { place << new-value }*)
.syne
.desc
The
.code set
operator stores the values of expressions in places. It must
be given an even number of arguments.
If there are no arguments, then
.code set
does nothing and returns
.codn nil .
If there are two arguments,
.meta place
and
.metn new-value ,
then
.meta place
is evaluated to determine its storage location, then
.meta new-value
is evaluated to determine the value to be stored there,
and then the value is stored in that location. Finally,
the value is also returned as the result value.
If there are more than two arguments, then
.code set
performs multiple assignments in left-to-right order.
Effectively,
.code "(set v1 e1 v2 e2 ... vn en)"
is precisely equivalent to
.codn "(progn (set v1 e1) (set v2 e2) ... (set vn en))" .
.coNP Macro @ pset
.synb
.mets (pset >> { place << new-value }*)
.syne
.desc
The syntax of
.code pset
is similar to that of
.codn set ,
and the semantics is similar also in that zero or more places are
assigned zero or more values. In fact, if there are no arguments, or
if there is exactly one pair of arguments,
.code pset
is equivalent to
.codn set .
If there are two or more argument pairs, then all of the arguments
are evaluated first, in left-to-right order. No store takes place
until after every
.meta place
is determined, and every
.meta new-value
is calculated. During the calculation, the values to be stored
are retained in hidden, temporary locations. Finally, these values
are moved into the determined places. The rightmost value is returned
as the form's value.
The assignments thus appear to take place in parallel, and
.code pset
is capable of exchanging the values of a pair of places, or rotating
the values among three or more places. (However, there are more convenient
operators for this, namely
.code rotate
and
.codn swap ).
.TP* Example:
.verb
;; exchange x and y
(pset x y y x)
;; exchange elements 0 and 1; and 2 and 3 of vector v:
(let ((v (vec 0 10 20 30))
(i -1))
(pset [vec (inc i)] [vec (inc i)]
[vec (inc i)] [vec (inc i)])
vec)
-> #(10 0 30 20)
.brev
.coNP Macro @ zap
.synb
.mets (zap < place <> [ new-value ])
.syne
.desc
The
.code zap
macro assigns
.meta new-value
to
.meta place
and returns the previous value of
.metn place .
If
.meta new-value
is missing, then
.code nil
is used.
In more detail, first
.code place
is evaluated to determine the storage location.
Then, the location is accessed to retrieve the
previous value. Then, the
.code new-value
expression is evaluated, and that value is
placed into the storage location.
Finally, the previously retrieved value is returned.
.coNP Macro @ flip
.synb
.mets (flip << place )
.syne
.desc
The
.code flip
macro toggles the Boolean value stored in
.metn place .
If
.meta place
previously held
.codn nil ,
it is set to
.codn t ,
and if it previously held a value other than
.codn nil ,
it is set to
.codn nil .
.coNP Macros @ test-set and @ test-clear
.synb
.mets (test-set << place )
.mets (test-clear << place )
.syne
.desc
The
.code test-set
macro examines the value of
.metn place .
If it is
.code nil
then it stores
.code t
into the place, and returns
.codn t .
Otherwise it leaves
.meta place
unchanged and returns
.codn nil .
The
.code test-clear
macro examines the value of
.metn place .
If it is Boolean true (any value except
.codn nil )
then it stores
.code nil
into the place, and returns
.codn t .
Otherwise it leaves
.meta place
unchanged and returns
.codn nil .
.coNP Macro @ compare-swap
.synb
.mets (compare-swap < place < cmp-fun < cmp-val << store-val )
.syne
.desc
The
.code compare-swap
macro examines the value of
.meta place
and compares it to
.meta cmp-val
using the comparison function given by the function name
.metn cmp-fun .
This comparison takes places as if by evaluating the expression
.mono
.meti >> ( cmp-fun < value << cmp-val )
.onom
where
.meta value
denotes the current value of
.metn place .
If the comparison is false,
.meta place
is not modified, the
.meta store-val
expression is not evaluated, and the macro returns
.codn nil .
If the comparison is true, then
.code compare-swap
evaluates the
.meta store-val
expression, stores the resulting value into
.meta place
and returns
.codn t .
.coNP Macros @ inc and @ dec
.synb
.mets (inc < place <> [ delta ])
.mets (dec < place <> [ delta ])
.syne
.desc
The
.code inc
macro increments
.meta place
by adding
.meta delta
to its value.
If
.meta delta
is missing, the value used in its place the integer 1.
First the
.meta place
argument is evaluated as a syntactic place to determine the location.
Then, the value currently stored in that location is retrieved.
Next, the
.meta delta
expression is evaluated. Its value is added to the previously retrieved
value as if by the
.code +
function. The resulting value is stored in the place, and returned.
The macro
.code dec
works exactly like
.code inc
except that addition is replaced by subtraction. The similarly defaulted
.meta delta
value is subtracted from the previous value of the place.
.coNP Macros @ pinc and @ pdec
.synb
.mets (pinc < place <> [ delta ])
.mets (pdec < place <> [ delta ])
.syne
.desc
The macros
.code pinc
and
.code pdec
are similar to
.code inc
and
.codn dec .
The only difference is that they return the previous value of
.meta place
rather than the incremented value.
.coNP Macros @ test-inc and @ test-dec
.synb
.mets (test-inc < place >> [ delta <> [ from-val ]])
.mets (test-dec < place >> [ delta <> [ to-val ]])
.syne
.desc
The
.code test-inc
and
.code test-dec
macros provide combined operations which change the value of a place and
provide a test whether, respectively, a certain previous value was
overwritten, or a certain new value was attained. By default, this tested
value is zero.
The
.code test-inc
macro notes the prior value of
.meta place
and then updates it with that value, plus
.metn delta ,
which defaults to 1. If the prior value is
.code eql
to
.meta from-val
then it returns
.codn t ,
otherwise
.codn nil .
The default value of
.meta from-val
is zero.
The
.code test-dec
macro produces a new value by subtracting
.meta delta
from the value of
.metn place .
The argument
.meta delta
defaults to 1. The new value is stored into
.metn place .
If the new value is
.code eql
to
.meta to-val
then
.code t
is returned, otherwise
.codn nil .
.coNP Macro @ swap
.synb
.mets (swap < left-place << right-place )
.syne
.desc
The
.code swap
macro exchanges the values of
.meta left-place
and
.meta right-place
and returns the value which is thereby transferred to
.metn right-place .
First,
.meta left-place
and
.meta right-place
are evaluated, in that order, to determine their locations.
Then the prior values are retrieved, exchanged and stored back.
The value stored in
.meta right-place
is also returned.
If
.meta left-place
and
.meta right-place
are ranges of the same sequence, the behavior is not specified
if the ranges overlap or are of unequal length.
Note: the
.code rotate
macro's behavior is somewhat more specified in this regard.
Thus, although any correct
.code swap
expression can be expressed using
.codn rotate ,
but the reverse isn't true.
.coNP Macro @ push
.synb
.mets (push < item << place )
.syne
.desc
The
.code push
macro places
.meta item
at the head of the list stored in
.meta place
and returns the updated list which is stored back in
.metn place .
First, the expression
.meta item
is evaluated to produce the push value.
Then,
.meta place
is evaluated to determine its storage location.
Next, the storage location is accessed to retrieve the
list value which is stored there. A new object is
produced as if by invoking
.code cons
function on the push value and list value.
This object is stored into the location,
and returned.
.coNP Macro @ pop
.synb
.mets (pop << place )
.syne
.desc
The
.code pop
macro removes an element from the list stored in
.meta place
and returns it.
First,
.meta place
is evaluated to determine the place. The place is accessed to
retrieve the original value. Then a new value is calculated,
as if by applying the
.code cdr
function to the old value. This new value is stored.
Finally, a return value is calculated and returned, as if by applying the
.code car
function to the original value.
.coNP Macro @ pushnew
.synb
.mets (pushnew < item < place >> [ testfun <> [ keyfun ]])
.syne
.desc
The
.code pushnew
macro inspects the list stored in
.metn place .
If the list already contains the item, then
it returns the list. Otherwise it creates a new list
with the item at the front and stores it back
into
.metn place ,
and returns it.
First, the expression
.meta item
is evaluated to produce the push value.
Then,
.meta place
is evaluated to determine its storage location.
Next, the storage location is accessed to retrieve the
list value which is stored there. The list is
inspected to check whether it already contains the push
value, as if using the
.code member
function. If that is the case, the list
is returned and the operation finishes.
Otherwise, a new object is
produced as if by invoking
.code cons
function on the push value and list value.
This object is stored into the location
and returned.
.coNP Macro @ shift
.synb
.mets (shift << place + << shift-in-value)
.syne
.desc
The
.code shift
macro treats one or more places as a "multi-place shift register".
The values of the places are shifted one place to the left.
The first (leftmost) place receives the value of the second place,
the second receives that of the third, and so on.
The last (rightmost) place receives
.meta shift-in-value
(which is not treated as a place, even if it is a syntactic place form).
The previous value of the first place is returned.
More precisely, all of the argument forms are evaluated left to right, in the
process of which the storage locations of the places are determined,
.meta shift-in-value
is reduced to its value.
The values stored in the places are sampled and saved.
Note that it is not specified whether the places are sampled in a separate
pass after the evaluation of the argument forms, or whether the
sampling is interleaved into the argument evaluation. This affects
the behavior in situations in which the evaluation of any of the
.meta place
forms, or of
.metn shift-in-value ,
has the side effect of modifying later places.
Next, the places are updated by storing the saved value of the second
place into the first place, the third place into the second and so forth,
and the value of
.meta shift-in-value
into the last place.
Finally, the saved original value of the first place is returned.
If any of the places are ranges which index into the same sequence,
and the behavior is not otherwise unspecified due to the issue
noted in an earlier paragraph, the effect upon the multiply-stored
sequence can be inferred from the above-described storage order.
Note that even if stores take place which change the length of
the sequence and move some elements, not-yet-processed stores whose ranges
to refer to these elements are not adjusted.
With regard to the foregoing paragraph, a recommended practice is
that if subranges of the same sequence object are shifted, they be
given to the macro in ascending order of starting index. Furthermore, the
semantics is simpler if the ranges do not overlap.
.coNP Macro @ rotate
.synb
.mets (rotate << place *)
.syne
.desc
Treats zero or more places as a "multi-place rotate register".
If there are no arguments, there is no effect and
.code nil
is returned. Otherwise, the last (rightmost) place receives
the value of the first (leftmost) place. The leftmost place
receives the value of the second place, and so on.
If there are two arguments, this equivalent to
.codn swap .
The prior value of the first place, which is the value
rotated into the last place, is returned.
More precisely, the
.meta place
arguments are evaluated left to right,
and the storage locations are thereby determined. The storage
locations are sampled, and then the sampled values are
stored back into the locations, but rotated by one place
as described above. The saved original value of the leftmost
.meta place
is returned.
It is not specified whether the sampling of the original values
is a separate pass which takes place after the arguments
are evaluated, or whether this sampling it is interleaved into argument
evaluation. This affects
the behavior in situations in which the evaluation of any of the
.meta place
forms has the side effect of modifying the value stored in
a later
.meta place
form.
If any of the places are ranges which index into the same sequence,
and the behavior is not otherwise unspecified due to the issue
noted in the preceding paragraph, the effect upon the multiply-stored
sequence can be inferred from the above-described storage order.
Note that even if stores take place which change the length of
the sequence and move some elements, not-yet-processed stores whose ranges
to refer to these elements are not adjusted.
With regard to the foregoing paragraph, a recommended practice is
that if subranges of the same sequence object are shifted, they be
given to the macro in ascending order of starting index. Furthermore, the
semantics is simpler if the ranges do not overlap.
.coNP Macro @ del
.synb
.mets (del << place )
.syne
.desc
The
.code del
macro requests the deletion of
.codn place .
If
.code place
doesn't support deletion, an exception is thrown.
First
.code place
is evaluated, thereby determining its location.
Then the place is accessed to retrieve its value.
The place is then subject to deletion. Finally, the
previously retrieved value is returned.
Precisely what deletion means depends on the kind of place.
The built-in places in \*(TL have deletion semantics which are
intended to be unsurprising to the programmer familiar with the
data structure which holds the place.
Generally, if a place denotes the element of a sequence, then deletion of the
place implies deletion of the element, and deletion of the element implies that
the gap produced by the element is closed. The deleted element is effectively
replaced by its successor, that successor by its successor and so on. If a
place denotes a value stored in a dynamic data set such as a hash table,
then deletion of that place implies deletion of the entry which holds
that value. If the entry is identified by a key, that key is also removed.
.coNP Macro @ lset
.synb
.mets (lset <> { place }+ << sequence-expr )
.syne
.desc
The
.code lset
operator's parameter list consists of one or more places followed
by an expression
.metn sequence-expr .
The macro evaluates
.codn sequence-expr ,
which is expected to produce a sequence.
Successive elements of the resulting list are then assigned to each
successive
.codn place .
If there are fewer elements in the sequence than places, the
unmatched places receive the value
.codn nil .
Excess elements in the sequence are ignored.
An error exception occurs if the sequence is an improper list with fewer
elements than places.
A
.code lset
form produces the value of
.meta sequence-expr
as its result value.
.coNP Macro @ upd
.synb
.mets (upd < place << opip-arg *)
.syne
.desc
The
.code upd
macro evaluates
.meta place
and passes the value as an argument to the operational pipeline
function formed,
as if by the
.code opip
macro, from the
.meta opip-arg
arguments. The result of this function is then stored back into
.metn place .
The following equivalence holds, except that place
.code p
is evaluated only once:
.verb
(upd p x y z ...) <--> (set p (call (opip x y z ...) p))
.brev
.SS* User-Defined Places and Place Operators
\*(TL provides a number of place-modifying operators such as
.codn set ,
.codn push ,
and
.codn inc .
It also provides a variety of kinds of syntactic places
which may be used with these operators.
Both of these categories are open-ended: \*(TL programs may extend
the set of place-modifying operators, as well as the vocabulary of
forms which are recognized as syntactic places.
Regarding place operators, it might seem obvious that new place operators can
be developed, since they are macros, and macros can expand to uses
of existing place operators. As an example, it may seem that
.code inc
operator could be written as a macro which uses
.codn set :
.verb
(defmacro new-inc (place : (delta 1))
^(set ,place (+ ,place ,delta)))
.brev
However, the above
.code new-inc
macro has a problem: the
.code place
argument form is inserted into two places in the expansion, which
leads to two evaluations. This is visibly incorrect if the place
form contains any side effects. It is also potentially inefficient.
\*(TL provides a framework for writing place update macros which
evaluate their argument forms once, even if they have to access
and update the same places.
The framework also supports the development of new kinds of place forms
as capsules of code which introduce the right kind of material into
the lexical environment of the body of an update macro, to enable
this special evaluation.
.NP* Place-Expander Functions
The central design concept in \*(TL syntactic places are
.IR "place-expander functions" .
Each compound place is defined by up to three place-expander functions,
which are associated with the place via the leftmost operator
symbol of the place form. One place-expander, the
.IR "update expander" ,
is mandatory. Optionally, a place may also provide a
.I "clobber expander"
as well as a
.IR "delete expander" .
An update expander provides the expertise for evaluating a place form once
in its proper run-time context to determine its actual run-time storage
location, and to access and modify the storage location.
A clobber expander provides an optimized mechanism for uses that perform
a one-time store to a place without requiring its prior value.
If a place definition does not supply a clobber expander, then the syntactic
places framework uses the update expander to achieve the functionality.
A delete expander provides the expertise for determining the actual run-time
storage location corresponding to a place, and obliterating it,
returning its prior value. If a place does not supply a delete expander, then
the place does not support deletion. Operators which require deletion, such as
.code del
will raise an error when applied to that place.
The expanders operate independently, and it is expected that place-modifying
operators choose one of the three, and use only that expander. For example,
accessing a place with an update expander and then overwriting its value
with a clobber expander may result in incorrect code which contains multiple
evaluations of the place form.
The programmer who implements a new place does not write expanders directly,
but rather defines them via the
.codn defplace ,
.code define-accessor
or
.code defset
macro.
The programmer who implements a new place update macro likewise does not
call the expanders directly. Usually, they are invoked via the macros
.codn with-update-expander ,
.code with-clobber-expander
and
.codn with-delete-expander .
These are sufficient for most kind of macros.
In certain complicated cases, expanders may be invoked using the wrapper
functions
.codn call-update-expander ,
.code call-clobber-expander
and
.codn call-delete-expander .
These convenience macros and functions perform certain common chores, like
macro-expanding the place in the correct environment, and choosing the
appropriate function.
The expanders are described in the following sections.
.NP* The Update Expander
.synb
.mets (lambda >> ( getter-sym < setter-sym < place-form
.mets \ \ \ \ \ \ \ \ << body-form ) ...)
.syne
.desc
The update expander is a code-writer. It takes a
.meta body-form
argument, representing code, and returns a larger form which surrounds
this code with additional code.
This larger form returned by the update expander can be regarded as having two
abstract actions, when it is substituted and evaluated in the context where
.meta place-form
occurs. The first abstract action is to evaluate
.meta place-form
exactly one time, in order to determine the actual run-time location to which
that form refers.
The second abstract action is to evaluate the caller's
.metn body-form s,
in a lexical environment in which bindings exist for some lexical
functions or (more usually) lexical macros. These lexical macros
are explicitly referenced by the
.metn body-form ;
the update expander just provides their definition, under the names
it is given via the
.meta getter-sym
and
.meta setter-sym
arguments.
The update expander writes local functions or macros under these names: a
getter function and a setter function. Usually, update expanders write
macros rather than functions, possibly in combination with some lexical
anonymous variables which hold temporary objects. Therefore the getter
and setter are henceforth referred to as macros.
The code being generated is with regard to some concrete instance of
.metn place-form .
This argument is the actual form which occurs in a program. For
instance, the update expander for the
.code car
place might be called with an arbitrary variant of the
.meta place-form
which might look like
.codn "(car (inc (third some-list)))" .
In the abstract semantics, upfront code wrapped around the
.meta body-form
by the update expander provides the logic to evaluate this place to
a location, which is retained in some hidden local context.
The getter local macro named by
.meta getter-sym
must provide the logic for retrieving the value of this place.
The getter macro takes no arguments.
The
.meta body-form
makes free use of the getter function; they may call it multiple times,
which must not trigger multiple evaluations of the original place
form.
The setter local macro named by
.meta setter-sym
must generate the logic for storing a new value into the once-evaluated
version of
.metn place-form .
The setter function takes exactly one argument, whose
value specifies the value to be stored into the place.
It is the caller's responsibility to ensure that the
argument form which produces the value to be stored via the setter is evaluated
only once, and in the correct order. The setter does not concern itself with
this form. Multiple calls to the setter can be expected to result in multiple
evaluations of its argument. Thus, if necessary, the caller must supply the code
to evaluate the new value form to a temporary variable, and then pass the
temporary variable to the setter. This code can be embedded in
the
.meta body-form
or can be added to the code returned by a call to the update expander.
The setter local macro or function must return the new value which is stored.
That is to say, when
.meta body-form
invokes this local macro or function, it may rely on it yielding the
new value which was stored, as part of achieving its own semantics.
The update expander does not macro-expand
.codn place-form .
It is assumed that the expander is invoked in such a way that the
place has been expanded in the correct environment. In other words, the
form matches the type of place which the expander handles.
If the expander had to macro-expand the place form, it would sometimes have
to come to the conclusion that the place form must be handled by a different
expander. No such consideration is the case: when an expander is called on
a form, that is final; it is certain that it is the correct expander, which
matches the symbol in the
.code car
position of the form, which is not a macro in the context where it occurs.
An update expander is free to assume that any place which is stored
(the setter local macro is invoked on it) is accessed at least once by
an invocation of the getter. A place update macro which relies on an update
expander, but uses only the store macro, might not work properly.
An example of an update expander which relies on this assumption is the
expander for the
.mono
.meti (force << promise )
.onom
place type. If
.meta promise
has not yet been forced, and only the setter is used, then
.meta promise
might remain unforced as its internal value location is updated.
A subsequent access to the place will incorrectly trigger a force,
which will overwrite the value. The expected behavior is that storing
a value in an unforced
.code force
place changes the place to forced state, preempting the evaluation of
the delayed form. Afterward, the promise exhibits the value which was
thus assigned.
The update expander is not responsible for all issues of evaluation order. A
place update macro may consist of numerous places, as well as numerous
value-producing forms which are not places. Each of the places can provide its
registered update expander which provides code for evaluating just that place,
and a means of accessing and storing the values. The place update macro must
call the place expanders in the correct order, and generate any additional code
in the correct order, so that the macro achieves its required documented
evaluation order.
.TP* "Example Update Expander Call:"
.verb
;; First, capture the update expander
;; function for (car ...) places
;; in a variable, for clarity.
(defvar car-update-expander [*place-update-expander* 'car])
;; Next, call it for the place (car [a 0]).
;; The body form specifies logic for
;; incrementing the place by one and
;; returning the new value.
(call car-update-expander 'getit 'setit '(car [a 0])
'(setit (+ (getit) 1)))
;; --> Resulting code:
(rlet ((#:g0032 [a 0]))
(macrolet ((getit nil
(append (list 'car) (list '#:g0032)))
(setit (val)
(append (list 'sys:rplaca)
(list '#:g0032) (list val))))
(setit (+ (getit) 1))))
;; Same expander call as above, with a call to expand added
;; to show the fully expanded version of the returned code,
;; in which the ;; setit and getit calls have disappeared,
;; replaced by their macro-expansions.
(expand
(call car-update-expander 'getit 'setit '(car [a 0])
'(setit (+ (getit) 1))))
;; --> Resulting code:
(let ((#:g0032 [a 0]))
(sys:rplaca #:g0032 (+ (car #:g0032) 1)))
.brev
The main noteworthy points about the generated code are:
.RS
.IP -
the
.code "(car [a 0])"
place is evaluated by evaluating the embedded form
.code "[a 0]"
and storing storing the resulting object into a hidden local variable.
That's as close a reference as we can make to the
.code car
field.
.IP -
the getter macro expands to code which simply calls the
.code car
function on the cell.
.IP -
the setter uses a system function called
.codn sys:rplaca ,
which differs from
.code rplaca
in that it returns the stored value, rather than the cell.
.RE
.NP* The Clobber Expander
.synb
.mets (lambda >> ( simple-setter-sym < place-form
.mets \ \ \ \ \ \ \ \ << body-form ) ...)
.syne
.desc
The clobber expander is a code-writer similar to the update expander.
It takes a
.meta body-form
argument, and returns a larger form which surrounds this form
with additional program code.
The returned block of code has one main abstract action.
It must arrange for the evaluation of
.meta body-form
in a lexical environment in which a lexical macro or lexical function
exists which has the name requested by the
.meta simple-setter-sym
argument.
The simple setter local macro written by the clobber expander is similar to the
local setter written by the update expander. It has exactly the
same interface, performs the same action of storing a value into
the place, and returns the new value.
The difference is that its logic may be considerably simplified by the
assumption that the place is being subject to exactly one store,
and no access.
A place update macro which uses a clobber expander, and calls it more than
once, break the assumption; doing so may result in multiple evaluations
of the
.metn place-form .
.NP* The Delete Expander
.synb
.mets (lambda >> ( deleter-sym < place-form
.mets \ \ \ \ \ \ \ \ << body-form ) ...)
.syne
.desc
The delete expander is a code-writer similar to clobber expander.
It takes a
.meta body-form
arguments, and returns a larger form which surrounds this form
with additional program code.
The returned block of code has one main abstract action.
It must arrange for the evaluation of
.meta body-form
in a lexical environment in which a lexical macro or lexical function
exists which has the name requested by the
.meta deleter-sym
argument.
The deleter macro written by the clobber expander takes no arguments.
It may be called at most once. It returns the previous value of the
place, and arranges for its obliteration, whatever that means for
that particular kind of place.
.coNP Macro @ with-update-expander
.synb
.mets (with-update-expander >> ( getter << setter ) < place < env
.mets \ << body-form )
.syne
.desc
The
.code with-update-expander
macro evaluates the
.meta body-form
argument, whose result is expected to be a Lisp form.
The macro adds additional code around this code, and the result is returned.
This additional code is called the
.IR "place-access code" .
The
.meta getter
and
.meta setter
arguments must be symbols. Over the evaluation of the
.metn body-form ,
these symbols are bound to the names of local functions which
are provided in the place-access code.
The
.meta place
argument is a form which evaluates to a syntactic place. The generated
place-access code is based on this place.
The
.meta env
argument is a form which evaluates to a macro-expansion-time environment.
The
.code with-update-expander
macro uses this environment to perform macro-expansion on the value of the
.meta place
form, to obtain the correct update expander function for the fully
macro-expanded place.
The place-access code is generated by calling the update expander
for the expanded version of
.codn place .
.TP* "Example:"
The following is an implementation of the
.code swap
macro, which exchanges the contents of two places.
Two places are involved, and, correspondingly, the
.code with-update-expander
macro is used twice, to add two instances of place-update code
to the macro's body.
.verb
(defmacro swap (place-0 place-1 :env env)
(with-gensyms (tmp)
(with-update-expander (getter-0 setter-0) place-0 env
(with-update-expander (getter-1 setter-1) place-1 env
^(let ((,tmp (,getter-0)))
(,setter-0 (,getter-1))
(,setter-1 ,tmp))))))
.brev
The basic logic for swapping two places is contained in the code template:
.verb
^(let ((,tmp (,getter-0)))
(,setter-0 (,getter-1))
(,setter-1 ,tmp))
.brev
The temporary variable named by the
.code gensym
symbol
.code tmp
is initialized by calling the getter function for
.metn place-0 .
Then the setter function of
.meta place-0
is called in order to store the value of
.meta place-1
into
.metn place-0 .
Finally, the setter for
.meta place-1
is invoked to store the previously saved temporary value into
that place.
The name for the temporary variable is provided by the
.code with-gensyms
macro, but establishing the variable is the caller's responsibility;
this is seen as an explicit
.code let
binding in the code template.
The names of the getter and setter functions are similarly provided
by the
.code with-update-expander
macros. However, binding those functions is the responsibility of that
macro. To achieve this, it adds the place-access code to the code generated by
the
.code "^(let ...)"
backquote template. In the following example macro-expansion, the additional
code added around the template is seen. It takes the form of two
.code macrolet
binding blocks, each added by an invocation of
.codn with-update-expander :
.verb
(macroexpand '(swap a b))
-->
(macrolet ((#:g0036 () 'a) ;; getter macro for a
(#:g0037 (val-expr) ;; setter macro for a
(append (list 'sys:setq) (list 'a)
(list val-expr))))
(macrolet ((#:g0038 () 'b) ;; getter macro for b
(#:g0039 (val-expr) ;; setter macro for b
(append (list 'sys:setq) (list 'b)
(list val-expr))))
(let ((#:g0035 (#:g0036))) ;; temp <- a
(#:g0037 (#:g0038)) ;; a <- b
(#:g0039 #:g0035)))) ;; b <- temp
.brev
In this expansion, for example
.code #:g0036
is the generated symbol which forms the value of the
.code getter-0
variable in the
.code swap
macro. The getter is a macro which simply expands to a
.codn a :
straightforward access to the variable a.
The
.code #:g0035
symbol is the value of the
.code tmp
variable. Thus the swap macro's
.mono
^(let ((,tmp (,getter-0))) ...)
.onom
has turned into
.mono
^(let ((#:g0035 (#:g0036))) ...)
.onom
A full expansion, with the
.code macrolet
local macros expanded out:
.verb
(expand '(swap a b))
-->
(let ((#:g0035 a))
(sys:setq a b)
(sys:setq b #:g0035))
.brev
In other words, the original syntax
.mono
(,getter-0)
.onom
became
.mono
(#:g0036)
.onom
and finally just
.codn a .
Similarly,
.mono
(,setter-0 (,getter-1))
.onom
became the
.code macrolet
invocations
.mono
(#:g0037 (#:g0038))
.onom
which finally turned into:
.codn "(sys:setq a b)" .
.coNP Macro @ with-clobber-expander
.synb
.mets (with-clobber-expander <> ( simple-setter ) < place < env
.mets \ << body-form )
.syne
.desc
The
.code with-clobber-expander
macro evaluates
.metn body-form ,
whose result is expected to be a Lisp form. The macro adds additional code
around this form, and the result is returned. This additional code is called
the
.IR "place-access code" .
The
.meta simple-setter
argument must be a symbol. Over the evaluation of the
.metn body-form ,
this symbol is bound to the name of a functions which
are provided in the place-access code.
The
.meta place
argument is a form which evaluates to a syntactic place. The generated
place-access code is based on this place.
The
.meta env
argument is a form which evaluates to a macro-expansion-time environment.
The
.code with-clobber-expander
macro uses this environment to perform macro-expansion on the value of the
.meta place
form, to obtain the correct update expander function for the fully
macro-expanded place.
The place-access code is generated by calling the update expander
for the expanded version of
.codn place .
.TP* "Example:"
The following implements a simple assignment statement, similar to
.code set
except that it only handles exactly two arguments:
.verb
(defmacro assign (place new-value :env env)
(with-clobber-expander (setter) place env
^(,setter ,new-value)))
.brev
Note that the correct evaluation order of
.code place
and
.code new-value
is taken care of, because
.code with-clobber-expander
generates the code which performs all the necessary evaluations of
.codn place .
This evaluation occurs before the code which is generated by
.mono
^(,setter ,new-value)
.onom
part is evaluated, and that code is what evaluates
.codn new-value .
Suppose that a macro were desired which allows assignment to be notated in a right to left
style, as in:
.verb
(assign 42 a) ;; store 42 in variable a
.brev
Now, the new value must be evaluated prior to the place, if left-to-right
evaluation order is to be maintained. The standard
.code push
macro has this property: the push value is on the left, and the place
is on the right.
Now, the code has to explicitly take care of the order, like this:
.verb
;; WRONG! We can't just swap the parameters;
;; place is still evaluated first, then new-value:
(defmacro assign (new-value place :env env)
(with-clobber-expander (setter) place env
^(,setter ,new-value)))
;; Correct: arrange for evaluation of new-value first,
;; then place:
(defmacro assign (new-value place :env env)
(with-gensym (tmp)
^(let ((,tmp ,new-value))
,(with-clobber-expander (setter) place env
^(,setter ,tmp)))))
.brev
.coNP Macro @ with-delete-expander
.synb
.mets (with-delete-expander <> ( deleter ) < place < env
.mets \ << body-form )
.syne
.desc
The
.code with-delete-expander
macro evaluates
.metn body-form ,
whose result is expected to be a Lisp form.
The macro adds additional code
around this code, and the resulting code is returned. This additional code is
called the
.IR "place-access code" .
The
.meta deleter
argument must be a symbol. Over the evaluation of the
.metn body-form ,
this symbol is bound to the name of a functions which
are provided in the place-access code.
The
.meta place
argument is a form which evaluates to a syntactic place. The generated
place-access code is based on this place.
The
.meta env
argument is a form which evaluates to a macro-expansion-time environment.
The
.code with-delete-expander
macro uses this environment to perform macro-expansion on the value of the
.meta place
form, to obtain the correct update expander function for the fully
macro-expanded place.
The place-access code is generated by calling the update expander
for the expanded version of
.codn place .
.TP* "Example:"
The following implements the
.code del
macro:
.verb
(defmacro del (place :env env)
(with-delete-expander (deleter) place env
^(,deleter)))
.brev
.coNP Function @ call-update-expander
.synb
.mets (call-update-expander < getter < setter < place < env
.mets \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ << body-form )
.syne
.desc
The
.code call-update-expander
function provides an alternative interface for making use of an update
expander, complementary to
.codn with-update-expander .
Arguments
.meta getter
and
.meta setter
are symbols, provided by the caller. These are passed to the update
expander function, and are used for naming local functions in the
generated code which the update expander adds to
.metn body-form .
The
.meta place
argument is a place which has not been subject to macro-expansion.
The
.code call-update-expander
function takes on the responsibility for macro-expanding the place.
The
.meta env
parameter is the macro-expansion environment object required to
correctly expand
.code place
in its original environment.
The
.meta body-form
argument represents the source code of a place update operation.
This code makes references to the local functions whose names
are given by
.meta getter
and
.metn setter .
Those arguments allow the update expander to write these functions
with the matching names expected by
.metn body-form .
The return value is an object representing source code which incorporates
the
.metn body-form ,
augmenting it with additional code which evaluates
.code place
to determine its location, and provides place accessor local functions
expected by the
.metn body-form .
.TP* "Example:"
The following shows how to implement a
.code with-update-expander
macro using
.codn call-update-expander :
.verb
(defmacro with-update-expander ((getter setter)
unex-place env body)
^(with-gensyms (,getter ,setter)
(call-update-expander ,getter ,setter
,unex-place ,env ,body)))
.brev
Essentially, all that
.code with-update-expander
does is to choose the names for the local functions, and bind them
to the local variable names it is given as arguments. Then it
calls
.codn call-update-expander .
.TP* "Example:"
Implement the swap macro using
.codn call-update-expander :
.verb
(defmacro swap (place-0 place-1 :env env)
(with-gensyms (tmp getter-0 setter-0 getter-1 setter-1)
(call-update-expander getter-0 setter-0 place-0 env
(call-update-expander getter-1 setter-1 place-1 env
^(let ((,tmp (,getter-0)))
(,setter-0 (,getter-1))
(,setter-1 ,tmp))))))
.brev
.coNP Function @ call-clobber-expander
.synb
.mets (call-clobber-expander < simple-setter < place < env
.mets \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ << body-form )
.syne
.desc
The
.code call-clobber-expander
function provides an alternative interface for making use of a clobber
expander, complementary to
.codn with-clobber-expander .
Argument
.meta simple-setter
is a symbol, provided by the caller. It is passed to the clobber
expander function, and is used for naming a local function in the
generated code which the update expander adds to
.metn body-form .
The
.meta place
argument is a place which has not been subject to macro-expansion.
The
.code call-clobber-expander
function takes on the responsibility for macro-expanding the place.
The
.meta env
parameter is the macro-expansion environment object required to
correctly expand
.code place
in its original environment.
The
.meta body-form
argument represents the source code of a place update operation.
This code makes references to the local function whose name
is given by
.metn simple-setter .
That argument allows the update expander to write this function
with the matching name expected by
.metn body-form .
The return value is an object representing source code which incorporates
the
.metn body-form ,
augmenting it with additional code which evaluates
.code place
to determine its location, and provides the clobber local function
to the
.metn body-form .
.coNP Function @ call-delete-expander
.synb
.mets (call-delete-expander < deleter < place < env << body-form )
.syne
.desc
The
.code call-delete-expander
function provides an alternative interface for making use of a delete
expander, complementary to
.codn with-delete-expander .
Argument
.meta deleter
is a symbol, provided by the caller. It is passed to the delete
expander function, and is used for naming a local function in the
generated code which the update expander adds to
.metn body-form .
The
.meta place
argument is a place which has not been subject to macro-expansion.
The
.code call-delete-expander
function takes on the responsibility for macro-expanding the place.
The
.meta env
parameter is the macro-expansion environment object required to
correctly expand
.code place
in its original environment.
The
.meta body-form
argument represents the source code of a place delete operation.
This code makes references to the local function whose name
is given by
.metn deleter .
That argument allows the update expander to write this function
with the matching name expected by
.metn body-form .
The return value is an object representing source code which incorporates
the
.metn body-form ,
augmenting it with additional code which evaluates
.code place
to determine its location, and provides the delete local function
to the
.metn body-form .
.coNP Macro @ define-modify-macro
.synb
.mets (define-modify-macro < name < parameter-list << function-name )
.syne
.desc
The
.code define-modify-macro
macro provides a simplified way to write certain kinds of place update
macros. Specifically, it provides a way to write place update macros
which modify a place by retrieving the previous value, pass it through
a function (perhaps together with some additional arguments), and then store
the resulting value back into the place and return it.
The
.meta name
parameter specifies the name for the place update macro to be written.
The
.meta function-name
parameter must specify a symbol: the name of the update function.
The update macro and update function both take at least one parameter:
the place to be updated, and its value, respectively.
The
.meta parameter-list
specifies the additional parameters for the update function, which will also
become additional parameters of the macro. Because it is a
function parameter list, it cannot use the special destructuring features of
macro parameter lists, or the
.code :env
or
.code :whole
special parameters. It can use optional parameters, and may be empty.
The
.code define-modify-macro
macro writes a macro called
.metn name .
The leftmost parameter of this macro is a place, followed by the additional arguments
specified by
.metn parameter-list .
The macro will arrange for the evaluation of the place argument to determine
the place location. It will then retrieve and save the prior value of the
place, and evaluate the remaining arguments. The prior value of the
place, and the values of the additional arguments, are all passed to
.meta function
and the resulting value is then stored back into the location previously
determined for
.metn place .
.TP* "Example:"
Some standard place update macros are implementable using
.codn define-modify-macro ,
such as
.codn inc .
The
.code inc
macro reads the old value of the place, then passes it through the
.code +
(plus) function, along with an extra argument: the delta value, which
defaults to one. The
.code inc
macro could be written using
.code define-modify-macro
as follows:
.verb
(define-modify-macro inc (: (delta 1)) +)
.brev
Note that the argument list
.code "(: (delta 1))"
doesn't specify the place, because the place is the implicit leftmost
argument of the macro which isn't given a name. With the above definition
in place, when
.code "(inc (car a))"
is invoked, then
.code "(car a)"
is first reduced to a location, and that location's value is retrieved and
saved. Then the
.code delta
parameter s evaluated to its value, which has defaulted to 1, since
the argument was omitted.
Then these two values are passed to the
.code +
function, and so 1 is added to the value previously retrieved from
.codn "(car a)" .
The resulting sum is then stored back
.code "(car a)"
without evaluating
.code "(car a)"
again.
.coNP Macro @ defplace
.synb
.mets (defplace < place-destructuring-args < body-sym
.mets \ \ >> ( getter-sym < setter-sym << update-body )
.mets \ \ >> [( ssetter-sym << clobber-body )
.mets \ \ \ >> [( deleter-sym << delete-body )]])
.syne
.desc
The
.code defplace
macro is used to introduce a new kind of syntactic place.
It writes the update expander, and optionally clobber and delete
expander functions, from a simpler, more compact specification,
and automatically registers the resulting functions. The compact specification
of a
.code defplace
call contains only code fragments for the expander functions.
The name and syntax of the place is determined by the
.meta place-destructuring-args
argument, which is macro-style parameter list whose structure
mimics that of the place. In particular, its leftmost symbol
gives the name under which the place is registered.
The
.code defplace
macro provides automatic destructuring of the syntactic place,
so that the expander code fragments can refer to the components
of a place by name.
The
.meta body-sym
parameter must be be a symbol. This symbol will capture the
.meta body-forms
parameter which is passed to the update expander, clobber
expander or delete expander. The code fragments then have
access to the body forms via this name.
The
.metn getter-sym ,
.metn setter-sym ,
and
.meta update-body
parenthesized triplet specify the update expander fragment.
The
.code defplace
macro will bind
.meta getter-sym
and
.meta setter-sym
to symbols. The
.meta update-body
must then specify a template of code which evaluates the syntactic place to
determine its storage location, and provides a pair of local functions, using
these two symbols as their name. The template must also insert the
.meta body-sym
forms into the scope of these local functions, and the place determining code.
The
.meta setter-sym
and
.meta clobber-body
arguments similarly specify an optional clobber expander fragment,
as a single optional argument. If specified, the
.meta clobber-body
must generate a local function named using
.meta setter-sym
wrapped around
.meta body-sym
forms.
The
.meta deleter-sym
and
.meta deleter-body
likewise specify a delete expander fragment. If this is omitted,
then the place shall not support deletion.
.TP* "Example:"
Implementation of the place denoting the
.code car
field of
.code cons
cells:
.verb
(defplace (car cell) body
;; the update expander fragment
(getter setter
(with-gensyms (cell-sym) ;; temporary symbol for cell
^(let ((,cell-sym ,cell)) ;; evaluate place to cell
;; getter and setter access cell via temp var
(macrolet ((,getter ()
^(car ,',cell-sym))
(,setter (val)
^(sys:rplaca ,',cell-sym ,val)))
;; insert body form from place update macro
,body))))
;; clobber expander fragment: simpler: no need
;; to evaluate cell to temporary variable.
(ssetter
^(macrolet ((,ssetter (val)
^(sys:rplaca ,',cell ,val)))
,body))
;; deleter: delegate to pop semantics:
;; (del (car a)) == (pop a).
(deleter
^(macrolet ((,deleter () ^(pop ,',cell)))
,body)))
.brev
.coNP Macro @ defset
.synb
.mets (defset < name < params < new-val-sym << set-form )
.mets (defset < get-fun-sym << set-fun-sym )
.syne
.desc
The
.code defset
macro provides a mechanism for introducing a new kind of syntactic place.
It is simpler to use than
.code defplace
and more concise, but not as general.
The
.code defset
macro is designed for situations in which a function or macro which evaluates
all of its arguments is required to serve as a syntactic place.
It provides two flavors of syntax: the long form, indicated by giving
.code defset
five arguments, and a short form, which uses two arguments.
In the long form of
.codn defset ,
the syntactic place is described by
.meta name
and
.metn params .
The
.code defset
form expresses the request that a call to the function or operator named
.meta name
be treated as a syntactic place, which has arguments described by
the parameter list
.metn params .
The
.meta set-form
argument specifies an expression which generates the code for storing a new
value to the place.
The
.code defset
macro makes the necessary arrangements such that when an operator form
named by
.meta name
is treated as a syntactic place, then at macro-expansion time, code is
generated to evaluate all of its argument expressions into machine-generated
variables. The names of those variables are automatically bound to the
corresponding symbols given in the
.meta params
argument list of the
.code defset
syntax. Code is also generated to evaluate the expression which gives the
new value to be stored, and that is bound to a generated variable whose
name is bound to the
.code new-val-sym
symbol. Then arrangements are made to invoke the operator named by
.meta name
and to evaluate the
.code set-form
in an environment in which these symbol bindings are visible.
The operator named
.meta name
is invoked using an altered argument list which uses temporary symbols in place
of the original expressions. The task of
.code set-form
is to insert the values of the symbols from
.meta params
and
.meta new-val-sym
into a suitable code templates that will perform the store actions.
The code generated by
.code set-form
must also take on the responsibility of yielding the new value as its result.
If
.meta params
list contains optional parameters, the default value expressions of those
parameters shall be evaluated in the scope of the
.code defset
definition.
The
.meta params
list may specify a rest parameter. In the expansion, this parameter will
capture a list of temporary symbols, corresponding to the list of variadic
argument expressions. For instance if the
.code defset
parameter list for a place
.code g
is
.codn "(a b . c)" ,
featuring the rest parameter
.codn c ,
and its
.meta set-form
is
.code "^(s ,a ,b ,*c)"
and the place is invoked as
.code "(g (i) (j) (k) (l))"
then parameter
.code c
will be bound to a list of gensyms such as
.code "(#:g0123 #:g0124)"
so that the evaluation of
.meta set-form
will yield syntax resembling
.codn "(s #:g0121 #:g0122 #:g0123 #:g0124)" .
Here, gensyms
.code #:g0123
and
.code #:g0124
are understood to be bound to the values of the expressions
.code (k)
and
.codn (l) ,
the two trailing parameters corresponding to the rest parameter
.codn c .
Syntactic places defined by
.code defset
that have a rest parameter may be invoked with improper syntax such as
.codn "(set (g x y . z) v)" .
In this situation, that rest parameter will be bound to the name of
a temporary variable which holds the value of
.code z
rather than to a list of temporary variable names holding the values
of trailing expressions.
The
.code set-form
must be prepared for this situation. In particular, the rest parameter's value
is an atom, then it cannot be spliced in the backquote syntax, except at the
last position of a list.
Although syntactic places defined by
.code defset
perform macro-parameter-like destructuring of the place form, binding
unevaluated argument expressions to the parameter symbols,
nested macro parameter lists are not supported:
.meta params
specifies a function parameter list.
The parameter list may use parameter macros, keeping in mind that
the parameter expansion is applied at the time the
.code defset
form is processed, specifying an expanded parameter list which
receives unevaluated expressions. The
.meta set-form
may refer to all symbols produced by parameter list expansion, other
than generated symbols. For instance, if a parameter list macro
.code :addx
exists which adds the parameter symbol
.code x
to the parameter list, and this
.code :addx
is invoked in the
.meta params
list of a
.codn defset ,
then
.code x
will be visible to the
.metn set-form .
The short, two-argument form of
.code defset
simply specifies the names of two functions or operators:
.code get-fun-sym
names the operator which accesses the place, and
.code set-fun-sym
names the operator which stores a new value into the place.
It is expected that all arguments of these operators are evaluated
expressions, and that the store operator takes one argument more
than the access operator. The operators are otherwise assumed to be
variadic: each instance of a place based on
.code get-fun-sym
individually determines how many arguments are passed to that operator
and to the one named by
.codn set-fun-sym .
The definition
.code "(defset g s)"
means that
.code "(inc (g x y))"
will generate code which ensures that
.code x
and
.code y
are evaluated exactly once, and then those two values are passed as
arguments to
.code g
which returns the current value of the place. That value is then incremented
by one, and stored into the place by calling the
.code s
function/operator with three arguments: the two values that were passed to
.code g
and the new value. The exact number of arguments is determined by each
individual use of
.code g
as a place; the
.code defset
form doesn't specify the arity of
.code g
and
.codn s ,
only that
.code s
must accept one more argument relative to
.codn g .
The following equivalence holds between the short and long forms:
.verb
(defset g s) <--> (defset g (. r) n ^(g ,*r) ^(s ,*r ,n))
.brev
Note:
the short form of
.code defset
is similar to the
.code define-accessor
macro.
.TP* "Example:"
Implementation of
.code car
as a syntactic place using a long form
.codn defset :
.verb
(defset car (cell) new
(let ((n (gensym)))
^(rlet ((,n ,new))
(progn (rplaca ,cell ,n) ,n))))
.brev
Given such a definition, the expression
.code "(inc (car (abc)))"
expands to code closely resembling:
.verb
(let ((#:g0048 (abc)))
(let ((#:g0050 (succ (car #:g0048))))
(rplaca #:g0048 #:g0050)
#:g0050))
.brev
The
.code defset
macro has arranged for the argument expression
.code (abc)
of
.code car
to be evaluated to a temporary variable
.codn #:g0048 ,
a
.codn gensym .
This, then, holds the
.code cons
cell being operated on.
At macro-expansion time, the variable
.code cell
from the parameter list specified by the
.code defset
is bound to this symbol. The access expression
.code "(car #:0048)"
to retrieve the prior value is automatically generated
by combining the name of the place
.code car
with the gensym to which its argument
.code (abc)
has been evaluated.
The
.code new
variable was bound to the expression giving the new value, namely
.codn "(succ (car #:g0048))" .
The
.meta set-form
is careful to evaluate this only one time, storing its value into
the temporary variable
.codn #:g0050 ,
referenced by the variable
.codn n .
The
.metn set-form 's
.code "(rplaca ,cell ,n)"
fragment thus turned into
.code "(rplaca #:g0048 #:g0050)"
where
.code #:g0048
references the cons cell being operated on, and
.code #:g0050
the calculated new value to be stored into its
.code car
field.
The
.meta set-form
is careful to arrange for the new value
.code #:g0050
to be returned. Those place-mutating operators which yield the new value, such
as
.code set
and
.code inc
rely on this behavior.
.coNP Macro @ define-place-macro
.synb
.mets (define-place-macro < name < macro-style-params
.mets \ \ << body-form *)
.syne
.desc
In some situations, an equivalence exists between two forms, only one
of which is recognized as a place. The
.code define-place-macro
macro can be used to establish a form as a place in terms of a translation to
an equivalent form which is already a place.
The
.code define-place-macro
has the same syntax as
.codn defmacro .
It specifies a macro transformation for a compound form which has the
.meta name
symbol in its leftmost position.
Place macro expansion doesn't use an environment; place macros are in a single
global namespace, special to place macros. There are no lexically scoped place
macros. Such an effect can be achieved by having a place macro expand to
an a form which is the target of a global or local macro, as necessary.
To support place macros, forms which are used as syntactic places are subject
to a modified macro-expansion algorithm:
.RS
.IP 1.
If a place macro exists for a form that is being used as a place, then the
that place macro is invoked to expand the
form, and the expansion is taken in place of the original form. This process
repeats until the form can no longer be expanded as a place macro, or
the place macro declines to expand the form by returning the unexpanded
input.
.IP 2.
A form that has been fully expanded as a place macro is then subject
to a single-round of macro-expansion, as if by
.codn macroexpand-1 ,
which takes place in the original form's lexical environment.
If the form doesn't expand, or the result of expansion is
.code nil
or a
non-symbolic atom, then the process terminates. Otherwise, the
process is repeated from step 1.
.RE
.IP
The
.code define-place-macro
macro does not cause
.meta name
to become
.codn mboundp .
There can exist both an ordinary macro and a place macro of the same name.
In this situation, when the macro call appears as a place form, it is
expanded as a place macro, according to the above steps. When the macro call
appears as an evaluated form, not being used as a place, the form is
expanded using the ordinary macro.
.TP* "Example:"
Implementation of
.code first
in terms of
.codn car :
.verb
(define-place-macro first (obj)
^(car ,obj))
.brev
.coNP Macro @ rlet
.synb
.mets (rlet >> ({( sym << init-form )}*) << body-form *)
.syne
.desc
The macro
.code rlet
is similar to the
.code let
operator. It establishes bindings for one or more
.metn sym s,
which are initialized using the values of
.metn init-form s.
Note that the simplified syntax for a variable which initializes to
.code nil
by default is not supported by
.codn rlet ;
that is to say, the syntax
.meta sym
cannot be used in place of the
.mono
.meti >> ( sym << init-form )
.onom
syntax when
.meta sym
is to be initialized to
.codn nil .
The
.code rlet
macro differs from
.code let
in that
.code rlet
assumes that those
.metn sym s
whose
.metn init-form s,
after macro expansion,
are constant expressions
(according to the
.code constantp
function) may be safely implemented as a symbol macro rather than a lexical
variable.
Therefore
.code rlet
is suitable in situations in which simpler code is desired from the output
of certain kinds of machine-generated code, which binds local symbols:
code with fewer temporary variables.
On the other hand,
.code rlet
is not suitable in some situations when true variables are required, which
are assignable, and provide temporary storage.
.TP* "Example:"
.verb
;; WRONG! Real storage location needed.
(rlet ((flag nil))
(flip flag)) ;; error: flag expands to nil
;; Demonstration of constant-propagation
(let ((a 42))
(rlet ((x 1)
(y a))
(+ x y))) --> 43
(expand
'(let ((a 42))
(rlet ((x 1)
(y a))
(+ x y)))) --> (let ((a 42))
(let ((y a))
(+ 1 y)))
.brev
The last example shows that the
.code x
variable has disappeared in the expansion. The
.code rlet
macro turned it into into a
.code symacrolet
denoting the constant 1, which then propagated to the use site,
turning the expression
.code "(+ x y)"
into
.codn "(+ 1 y)" .
.coNP Macro @ slet
.synb
.mets (slet >> ({( sym << init-form )}*) << body-form *)
.syne
.desc
The macro
.code slet
is a stronger form of the
.code rlet
macro. Just like
.codn rlet ,
.code slet
reduces bindings initialized by constant expressions
to symbol macros. In addition, unlike
.codn rlet ,
.code slet
also reduces to symbol macros those bindings which
are initialized by symbol expressions (values of variables).
.coNP Macro @ alet
.synb
.mets (alet >> ({( sym << init-form )}*) << body-form *)
.syne
.desc
The macro
.code alet
("atomic" or "all") is a stronger form of the
.code slet
macro. All bindings initialized by constant expressions are
turned to symbol macros. Then, if all of the remaining bindings are
all initialized by symbol expressions, they are also turned to
symbol macros. Otherwise, none of the remaining bindings
are turned to symbol macros.
The
.code alet
macro can be used even in situations when it is possible that the initializing
forms the variables may have side effects through which they affect each
others' evaluations. In this situation
.code alet
still propagates constants via symbol macros, and can eliminate the
remaining temporaries if they can all be made symbol macros for
existing variables: i.e. there doesn't exist any initialization form
with interfering side effects.
.coNP Macro @ define-accessor
.synb
.mets (define-accessor < get-function << set-function )
.syne
.desc
The
.code define-accessor
macro is used for turning a function into an accessor,
such that forms which call the function can be treated
as places.
Arguments to
.code define-accessor
are two symbols, which must name functions. When the
.code define-accessor
call is evaluated, the
.meta get-function
symbol is registered as a syntactic place. Stores to the
place are handled via calls to
.metn set-function .
If
.meta get-function
names a function which takes N
arguments,
.meta set-function
must name a function which takes N+1 arguments.
Moreover, in order for the accessor semantics to be correct
.meta set-function
must treat its rightmost argument as the value being stored,
and must also return that value.
When a function call form targeting
.meta get-function
is treated as a place which is subject
to an update operation (for instance an increment via the
.code inc
macro),
the accessor definition created by
.code define-accessor
ensures that the arguments of
.meta get-function
are evaluated only once, even though the update involves
a call to
.meta get-function
and
.meta set-function
with the same arguments. The argument forms are evaluated to
temporary variables, and these temporaries are used as the
arguments in the calls.
No other assurances are provided by
.codn define-accessor .
In particular, if
.meta get-function
and
.meta set-function
internally each perform some redundant calculation over their arguments,
this cannot be optimized. Moreover, if that calculation has a visible effect,
that effect is observed multiple times in an update operation.
If further optimization or suppression of multiple effects is required,
the more general
.code defplace
macro must be used to define the accessor. It may also be possible to
treat the situation in a satisfactory way using a
.code define-place-macro
definition, which effectively then supplies inline code whenever a certain form
is used as a place, and that code itself is treated as a place.
Note:
.code define-accessor
is similar to the short form of
.codn defset .
.coNP Accessor @ read-once
.synb
.mets (read-once << expression )
.mets (set (read-once << place ) << new-value )
.syne
.desc
When the
.code read-once
accessor is invoked as a function, it behaves like
.codn identity ,
simply returning the value of
.metn expression ,
which is not required to be a syntactic place.
If a
.code read-once
form is used as a syntactic place then its argument must also be a
.metn place .
The
.code read-once
syntactic place denotes the same place as the enclosed
.code place
form, but with somewhat altered semantics, which is most useful in conjunction
with
.codn placelet ,
and in writing place-mutating macros which make multiple accesses to a place.
Firstly, if the
.code read-once
place is evaluated, it accesses the existing value of
.meta place
exactly once, even if it occurs in a place-mutating form which
normally doesn't use the prior value, such as the
.code set
macro.
When
.code read-once
accesses
.metn place ,
it stores the value in a hidden variable.
Then, within the same place-mutating form, multiple references to the same
.code read-once
form all access the value of this hidden variable.
Whenever the
.code read-once
form is assigned, both the the hidden variable and the underlying
.meta place
receive the new value.
Multiple references to the same
.code read-once
form can be produced using the
.code placelet
or
.code placelet*
macros, or by making multiple calls to the getter function obtained using
.code with-update-expander
in the implementation of a user-defined place-mutating operator,
or user-defined place.
.TP* Example:
In both of the following two examples, there is no question that the
.code array
and
.code i
expressions are themselves evaluated only once; the issue is the access to the
array itself; under the plain placelet, the array referencing takes place more
times.
.verb
;; without read-once, array element [array i] is
;; accessed twice to fetch its current value: once
;; in the plusp expression, and then once again in
;; the dec expression.
(placelet ((cell [array i]))
(if (plusp cell)
(dec cell)))
;; with read-once, it is accessed once. plusp refers
;; to a hidden lexical variable to obtain the prior
;; value, and so does dec. dec stores the new value
;; through to [array i] and the hidden variable.
(placelet ((cell (read-once [array i])))
(if (plusp cell)
(dec cell)))
.brev
The following is
.B not
an example of multiple references to the same
.code read-once
form:
.verb
(defmacro inc-positive (place)
^(if (plusp (read-once ,place))
(inc (read-once ,place))))
.brev
Here, even though the
.code read-once
forms may be structurally identical, they are separate instances.
The first instance isn't even a syntactic place, but a call to the
.code read-once
function. Multiple references to the same place can only be
generated using
.code placelet
or else by multiple explicit calls to the same getter function or macro
generated for a place by an update expander.
The following is a corrected version of
.codn inc-positive :
.verb
(defmacro inc-positive (place :env env)
(with-update-expander (getter setter) ^(read-once ,place) env
^(if (plusp (,getter))
(,setter (succ (,getter))))))
.brev
To write the macro without
.code read-once
requires that it handles the job of providing a temporary variable
for the value:
.verb
(defmacro inc-positive (place :env e)
(with-update-expander (getter setter) place env
(with-gensym (value)
^(slet ((,value (,getter)))
^(if (plusp ,value)
(,setter (succ ,value)))))))
.brev
The
.code read-once
accessor wrapped around
.meta place
allows
.code inc-positive
to simply make multiple references to
.code "(,getter)"
which will cache the value; the macro doesn't have to introduce its own
hidden caching variable.
.coNP Special variables @, *place-update-expander* @ *place-clobber-expander* and @ *place-delete-expander*
.desc
These variables hold hash tables, by means of which update expanders,
clobber expanders and delete expanders are registered, as associations
between symbols and functions.
If
.code "[*place-update-expander* 'sym]"
yields a function, then symbol
.code sym
is the basis for a syntactic place. If the expression yields
.codn nil ,
then forms beginning with
.code sym
are not syntactic places. (The situation of a clobber accessor or delete
accessor being defined without an update expander is improper).
.coNP Special variable @ *place-macro*
.desc
The
.code *place-macro*
special variable holds the hash table of associations between
symbols and place macro expanders.
If the expression
.code "[*place-macro* 'sym]"
yields a function, then symbol
.code sym
has a binding as a place macro. If that
expression yields
.codn nil ,
then there is no such binding: compound forms beginning with
.code sym
do not undergo place macro expansion.
.SS* Structural Pattern Matching
.NP* Introduction
\*(TL provides a structural pattern-matching system. Structural pattern
matching is a syntax which allows for the succinct expression of code
which classifies objects according to their shape and content, and which
accesses the elements within objects, or both.
The central concept in structural pattern matching is the resolution of a
pattern against an object. The pattern is specified as syntax which is part of
the program code. The object is a run-time value of unknown type, shape and
other properties. The primary pattern-matching decision is Boolean: does the
object match the pattern? If the object matches the pattern, then it is
possible to execute an associated body of code in a scope in which variables
occurring in the pattern take on values from the corresponding parts of the
object.
.NP* Pattern-Matching Operators
Structural pattern matching is available via several different macro
operators, which are:
.codn when-match ,
.codn if-match ,
.codn match ,
.codn match-case ,
.codn match-ecase ,
.code lambda-match
and
.codn defun-match .
Function and macro argument lists may also be augmented with pattern
matching using the
.code :match
parameter macro.
The
.code when-match
macro is the simplest. It tests an object against a pattern, and if there is a
match, evaluates zero or more forms in an environment in which the pattern
variables have bindings to the corresponding elements of the object.
The
.code if-match
macro evaluates a single form if there is a match, in the scope of the
bindings established by the pattern, otherwise an alternative
form evaluated in a scope in which those bindings are absent.
The
.code match
macro tests and object against a pattern, expecting a match. If the match
fails, an exception is thrown. Otherwise, it evaluates zero or more forms
in the scope of the bindings established by the pattern.
The
.code match-case
macro evaluates the same object against multiple clauses, each consisting of a
pattern and zero or more forms. The first case whose pattern matches the object
is selected. The forms associated with a matching clause are evaluated in
the scope the variables bound by that clause's pattern.
The
.code match-ecase
macro is similar to
.code match-case
except that if no matching case is identified, an exception is thrown.
The
.code lambda-match
macro provides a way to express an anonymous function whose argument list
is matched against multiple clauses similarly to
.code match-case
and
.code defun-match
provides a way to define a top-level function using the same concept.
Additionally, there exist
.code each-match
and
.code while-match
macro families.
.NP* Syntax and Key Concepts
\*(TL's structural pattern-matching notation is template-based.
With the exception of structures and hash tables, objects are matched using
patterns which are based on their printed notation. For instance, the pattern
.code "(1 2 @a)"
is a pattern matching the list
.code "(1 2 3)"
binding
.code a
to
.codn 3 .
The notation supports lists, vectors, ranges and atoms. Atoms are compared
using the
.code equal
function. Thus, in the above pattern, the 1 and 2 in the pattern match the
corresponding 1 and 2 atoms in the object using
.codn equal .
All parts of a pattern are static material which matches literally,
except those parts introduced by the meta prefix
.codn @ .
This prefix denotes variables like
.code @a
as well as useful pattern-matching operators like
.mono
.meti @(all << pattern )
.onom
which matches a list or sublist whose elements all match
.metn pattern .
The quasiquote syntax is specially supported for expressing matching,
in an alternative style. For instance the quasiquote
.code "^(1 2 ,a)"
is a pattern equivalent to the
.codn "(1 2 @a)" .
Structure objects are matched using a dedicated
.code "@(struct name ...)"
operator, or else in the quasiquote style using
.code "^#S(name ...)"
syntax. The non-quasiquoted literal syntax
.code "#S(name ...)"
cannot be used for matching.
Similarly, hash objects are matched using a
.code "@(hash ...)"
operator, or else
.code "^#H(...)"
syntax in the quasiquote style.
.code "#H(...)"
cannot be used.
Note: the non-quasiquoted
.code #S
and
.code #H
literals are not and cannot be used for matching because they produce structure
and hash objects which lose important information about how they were specified
in the syntax, and carry restrictions which are unacceptable for pattern
matching. The order of sub-patterns is important in pattern syntax, but struct
and hash objects do not preserve the order in which their elements were
specified. A struct literal is required to specify the name of an existing
struct type, and slot names which are valid for that type, otherwise it is
erroneous. This is not acceptable for pattern matching, because patterns may
appear in place of those elements. The pattern match for a hash may specify the
same key pattern more than once, which means that the key pattern cannot be an
actual key in an actual hash, which requires every key to be unique. Structure
and hash quasiquotes do not have these issues; they are not actually literal
structure and hash objects, but list-based syntax.
.NP* Variables in Patterns
Patterns use meta-symbols for denoting variables. Variables must
be either bindable symbols, or else
.codn nil ,
which has a special meaning: the pattern variable
.code @nil
matches any object, and binds no variable.
Pattern variables are ordinary Lisp variables. Whereas in ordinary non-pattern
matching Lisp code, it is always unambiguous whether a variable is being bound
or referenced, this is deliberately not the case in patterns. A variable
occurring in a pattern may be a fresh variable, or a reference to an existing
one. The difference between these situations is not apparent from the syntax
of the pattern; it depends on the context established by the scope.
With one exception, if a pattern contains a variable which is already in
the surrounding scope, including a global variable, then it refers to that
variable. Otherwise, it freshly binds the variable.
The exception is that pattern operator
.code @(as)
always binds a fresh variable.
When a pattern variable refers to an existing variable, then each occurrence
of that variable must match an object which is
.code equal
to the value of that variable.
For instance, the following function returns the third element of a list, if
the first two elements are repetitions of the
.code x
argument, otherwise
.codn nil :
.verb
(defun x-x-y (list x)
(when-match (@x @x @y) list y))
(x-x-y '(1 1 2) 1) -> 2
(x-x-y '(1 2 3) 1) -> nil ;; no @x @x match
(x-x-y '(1 1 2 r2) 1) -> nil ;; list too long
.brev
If the variable does not exist in the scope surrounding the pattern,
then the leftmost occurrence of the variable establishes a binding,
taking the value from is corresponding object being matched by that
occurrence of the variable. The remaining
occurrences of the variable, if any, must correspond to objects which are
.code equal
to that value, or else there is no match.
For instance, the pattern
.code "(@a @a)"
matches the list like
.code "(1 1)"
as follows. First
.code @a
binds to the leftmost
.code 1
and then the second
.code 1
matches the existing value of that
.codn a .
An input such as
.code "(1 2)"
fails to match because the second occurrence of
.code @a
retrieves an object that is not
.code equal
to that variable's existing value.
A pattern can contain multiple occurrences of the same symbol as a variable.
These may or may not refer to the same variable. Two occurrences of the same
symbol refer to distinct variables if:
.RS
.IP 1.
they are freshly bound in separate
branches of the
.code @(or)
operator; or
.IP 2.
one of the two variables is freshly bound by the
.code @(as)
operator and the other variable occurs outside of that
.codn @(as) ;
or
.IP 3.
or both of the variables are freshly bound using
.codn @(as) .
.RE
Any other two or more occurrences same symbol occurring in the same pattern
refer to the same variable.
.NP* Comparison to Macro Parameter Lists
\*(TL's macro-style parameter lists, appearing in
.code tree-bind
and related macros, also provide a form of structural pattern matching.
Macro-style parameter list pattern matching is limited to objects of
one kind: tree structures made of
.code cons
cells. It is only useful for matching on
shape, not content. For example,
.code tree-bind
cannot express the idea of matching a list whose first element is the symbol
.code a
and whose third element is
.codn 42 .
Moreover, every position in the tree pattern much specify a variable
which captures the corresponding element of the structure. For instance,
a pattern which matches a three-element list must specify three variables,
one for each list position. This is because macro-style parameter lists are
oriented toward writing macros, and macros usually make use of every parameter
position.
.NP* User-defined Patterns
User-defined pattern operators are possible. When the
.meta operator
symbol in the
.mono
.meti >> @( operator << argument *)
.onom
syntax doesn't match any built-in operator, a search takes
place to determine whether
.meta operator
is a pattern macro. If so, the pattern macro is expanded, and
its result of the expansion treated as a pattern to process recursively,
unless it is the original macro form, in which case it is treated
as a predicate pattern. User-defined pattern macros are defined
using the
.code defmatch
macro.
.SS* Pattern-Matching Notation
The pattern-matching notation is documented in the following
sections; a section describing the pattern-matching macros follow.
.NP* Atom match
A pattern consisting of an atom other than a vector
matches a similar object. The similarity is determined using the
.code equal
function.
The atom is not subject to evaluation, which means that a symbolic atom stands
for itself, and not the value of a variable.
.TP* Examples:
.verb
;; the pattern 1 matches the object 1
(if-match 1 1 'yes 'no) --> yes
;; the object 0 does not match
(if-match 1 0 'yes 'no) --> no
;; a matches a, does not match b
(let ((sym 'a))
(list (if-match a sym 'yes 'no)
(if-match b sym 'yes 'no)))
--> (yes no)
.brev
.NP* Variable match
.synb
.mets >> @ symbol
.syne
.desc
A meta-symbol can be used as a pattern expression.
This pattern unconditionally matches an object of any kind.
The
.meta symbol
is required to be a either a bindable symbol according to the
.code bindable
function, or else the symbol
.codn nil .
If
.meta symbol
is a bindable symbol, which has not binding in scope,
then a variable by that name is freshly bound, and takes
on the corresponding object as its value.
If
.meta symbol
is a bindable symbol with an existing binding, then
the corresponding object must be
.code equal
to that variable's existing value, or else the match fails.
If
.meta symbol
is
.codn nil ,
then the match succeeds unconditionally, without binding a variable.
.TP* Examples:
.verb
(when-match @a 42 (list a)) -> (42)
(when-match (@a @b @c) '(1 2 3) (list c b a)) -> (3 2 1)
;; No match: list is longer than pattern
(when-match (@a @b) '(1 2 3) (list a b)) -> nil
;; Use of nil in dot position to match longer list
(when-match (@a @b . @nil) '(1 2 3) (list a b)) -> (1 2)
.brev
.NP* List match
.synb
.mets <> ( pattern +)
.mets <> ( pattern + . << pattern )
.syne
.desc
Pattern syntax consisting of a nonempty, possibly improper list
matches list structure. A pattern expression may be specified in the
dotted position. If it is omitted, then there is an implicit terminating
.code nil
which constitutes an atom expression matching
.codn nil .
A list pattern matches a list of the same shape. For each
.meta pattern
expressions, there must exist an item in the list.
A match occurs when every
.meta pattern
matches the corresponding element of the list, including the
.meta pattern
in the dotted position.
Because the dotted position
.meta pattern
matches a list, it is possible for a short pattern
to match a longer list.
The syntax is indicated as requiring at least one
.meta pattern
because otherwise the list is empty, which corresponds to the
atom pattern
.codn nil .
The syntax
.mono
.meti (. << pattern )
.onom
is valid, but indistinguishable from
.meta pattern
and therefore is not a list pattern.
.TP* Examples:
.verb
(if-match (@a @b @c . @d) '(1 2 3 . 4) (list d c b a))
--> (4 3 2 1)
;; 2 doesn't satisfy oddp
(if-match (@(oddp @a) @b @c . @d) '(2 x y z)
(list a b c d)
:no-match)
--> :no-match
;; 1 and 2 match, a takes (3 4)
(if-match (1 2 . @a) '(1 2 3 4) a) --> (3 4)
;; nesting
(if-match ((1 2 @a) @b) '((1 2 3) 4) (list a b)) -> (3 4)
.brev
.NP* Vector match
.synb
.mets <> #( pattern *)
.syne
.desc
A pattern match for a vector is expressed using vector notation enclosing
pattern expressions. This pattern matches a vector object which contains
exactly as many elements as there are patterns. Each pattern is applied
against the corresponding vector element.
.TP* Examples:
.verb
;; empty vector pattern matches empty vector
(if-match #() #() :yes :no) -> :yes
;; empty vector pattern fails to match nonempty vector
(if-match #() #(1) :yes :no) -> :no
;; match with nested list and vector
(if-match #((1 @a) #(3 @b)) #((1 2) #(3 4)) (list a b))
--> (2 4)
.brev
.NP* Range match
.synb
.mets >> #R( from-pattern << to-pattern )
.syne
.desc
A pattern match for a range can be expressed by embedding pattern
expressions in the
.code #R
notation. The resulting pattern requires the corresponding object
to be a range, otherwise the match fails. If the corresponding
object is a range, then the
.meta from-pattern
is matched against its
.code from
and the
.meta to-pattern
is matched against its
.code to
part.
Note that if the range expression notation
.code a..b
is used as a pattern, that is actually a list pattern, due to
that being a syntactic sugar for
.codn "(rcons a b)" .
.TP* Examples:
.verb
(if-match #R(10 20) 10..20 :yes :no) -> :yes
(if-match #R(10 20) #R(10 20) :yes :no) -> :yes
(if-match #R(10 20) #R(1 2) :yes :no) -> :no
(when-match #R(@a @b) 1..2 (list a b)) -> (1 2)
;; not a range match! rcons syntax match
(when-match @a..@b '1..2 (list a b)) -> (1 2)
;; above, de-sugared:
(when-match (rcons @a @b) '(rcons 1 2) (list a b)) -> (1 2)
.brev
.NP* Quasiliteral match
.synb
.mets <> "`...@" var "...`"
.mets <> @"`...@" var "...`"
.syne
.desc
The quasiliteral syntax is supported as a pattern-matching operator.
The corresponding object is required to be a character string, which
is analyzed according to the structure of the quasiliteral pattern,
and portions of the string are captured by variables. If the corresponding
object isn't a string according to
.code stringp
then the match fails. The quasiliteral pattern must match the entire
input string.
In order that the quasiliteral's syntactic structure is not misinterpreted
as a predicate pattern, and in order to make certain situations work
in quasiquoted pattern matching, a quasiliteral pattern may be specified
as either
.code "`...`"
or
.codn "@`...`" .
The latter form, which is structurally
.code "(sys:expr (sys:quasi ...))"
is specially recognized and treated as equivalent to the unadorned
quasiliteral pattern.
A quasiliteral pattern matches in a linear fashion, from left to right.
Variables bound earlier in the pattern can be referenced later in the pattern
as bound variables.
With one exception, bound variables denote character strings in accordance with the usual
quasiliteral conversion and formatting rules. All of the modifier notations may
be used. For instance, if
.code x
is a bound variable, then
.code "@{x -40}"
denotes the value of
.code x
converted to a string, and right-aligned in a forty-character-wide field.
Consequently, the notation matches exactly such a forty-character text.
The exception is that if a bound variable has a regular expression modifier,
as in
.code "@{x #/re/}"
then it has a special meaning as a pattern. Moreover, this syntax has no
meaning in a quasiliteral.
In the following description of the quasiliteral pattern-matching rules, the
symbols
.metn uv ,
.meta uv0
and
.meta uv1
represent to unbound variables: variables which have no apparent
lexical binding and are not defined as global variables. Unless indicated
otherwise,
.mono
.meti >> @ uv
.onom
refers to a plain variable syntax such as
.code @abc
or else to braced syntax without modifiers, such as
.codn @{abc} .
The same remarks apply to
.meta uv0
and
.metn uv1 .
The symbol
.meta bv
represents a bound variable: a variable which has an existing binding,
which can occur in the form of the ordinary notation, or the braced notation
with or without modifiers.
The notation
.codn {P} ,
.codn {P0} ,
.codn {P1} ...
denotes a substring of the pattern, possibly empty.
.RS
.coIP ``
The empty quasiliteral pattern matches an empty string.
.coIP `text{P}`
A quasiliteral pattern which begins with a portion of text matches a string
which begins with the same text. The remaining portion
.code {P}
of the pattern is then matched against a suffix of the input string which
excludes the matched text.
.meIP <> `@ uv `
A simple unbound variable occurring as the last element of the pattern
matches and binds the entire rest of the input string.
.meIP <> `@ uv text{P}`
A simple unbound variable followed by a text element matches the input string if
.str text
occurs in that string as a substring. In that case,
.meta uv
is bound to the possibly empty prefix of the input string consisting of the
characters before the leftmost match for
.strn text .
The rest of the pattern
.code {P}
is then matched against that suffix of the input string which begins after the
last character of the leftmost match for
.strn text .
.meIP <2> `@ uv @ bv {P}`
The bound variable
.meta bv
is converted to text in the manner of an ordinary quasiliteral substitution.
The situation then reduces to the
.mono
.meti <> `@ uv text{P}`
.onom
pattern, where
.code text
denotes the character string produced by substitution of
.metn bv .
.meIP >> `@{ uv << integer }{P}`
An unbound variable
.meta uv
which uses the brace notation to specify a literal
.meta integer
modifier denotes
a match for that many characters. It is an error if the value is zero or
negative. The match succeeds if the input string has at least that
many characters, in which case the variable
.meta uv
takes on those characters, and the rest of the pattern is matched against
a suffix of the string with those characters removed.
.meIP >> `@{ uv <> #/ regex /}{P}`
An unbound variable
.meta uv
which carries a regular-expression modifier specifies a regular-expression
match. If a prefix of the input string matches
.metn regex ,
then the match is successful and
.meta uv
captures that prefix. The rest of the pattern
.code {P}
is then matched against the rest of the string after the prefix.
.meIP >> `@{ bv <> #/ regex /}{P}`
A bound variable
.meta bv
which carries a regular expression modifier specifies a regular expression
match exactly like an unbound variable. This syntax produces a successful
match if two conditions are met: a prefix of the input string matches
.metn regex ,
and the matched prefix is
.meta equal
to the value of
.metn bv .
The rest of the pattern
.code {P}
is then matched against the rest of the string after the prefix.
.meIP <> `@ bv {P}`
The bound variable
.meta bv
is converted to text the manner of an ordinary quasiliteral substitution.
The situation then reduces to the
.code `text{P}`
pattern, where
.code text
denotes the character string produced by substitution of
.metn bv .
.meIP <2> `@ uv0 @ uv1 {P0}`
Two consecutive unbound variables, where
.meta uv0
is a plain variable with no modifiers, constitutes an invalid pattern.
This situation is diagnosed as an error. If
.meta uv0
is braced, carrying an integer or regular-expression modifier
.metn mod ,
then the situation is treated as the pattern
.mono
.meti >> `@{ uv << mod }{P}`
.onom
where
.code {P}
refers to the
.mono
.meti <> @ uv1 {P0}
.onom
portion.
.RE
.IP
No other quasiliteral syntax, or combination of variable modifiers, is
supported in quasiliteral patterns.
.TP* Examples:
.verb
(when-match `@a-@b` "foo-bar" (list a b)) -> ("foo" "bar")
(when-match `@{a #/\ed+/}@b` "123xy" (list a b)) -> ("123" "xy")
(let ((a 42))
(when-match `[@{a -8}] @b` "[ 42] packets` b))
-> "packets"
.brev
.NP* Quasiquote matching notation
.synb
.mets >> ^ qq-syntax
.syne
.desc
Quasiquoting provides an alternative pattern-matching syntax. It uses a subset
of the quasiquoting notation. Only specific kinds of quasiquoted objects listed
in this description are supported. Within a quasiquote used for
pattern-matching, unquotes indicate operators and variables instead of the
.code @
prefix. Splicing unquote syntax plays no role; its presence produces
unspecified behavior.
The quasiquote matching notation is described, understood and implementing
in terms of a translation to the standard pattern-matching syntax, according
to the following rules. The
.code [X]
notation used here indicates that the element enclosed in brackets is
subject to a recursive translation according to the rules:
.RS
.meIP >> , expr
An unquoted expression occurring in the quasiquote is translated to the
.mono
.meti >> @ expr
.onom
pattern-matching syntax. If
.meta expr
is an symbol, then this is a meta-variable:
.mono
.meti (sys:var << expr )
.onom
otherwise it is translated to the
.mono
.meti (sys:expr << expr )
.onom
syntax.
.coIP ",`...quasilit...`"
An unquoted quasiliteral is treated uniformly as
.mono
.meti >> , expr
.onom
and is therefore translated into
.codn "@`...quasilit...`" .
Since that is equivalent to
.codn "`...quasilit...`" ,
quasiliteral matching is supported within quasiquote notation
in a straightforward way.
.meIP >> ~ expr
In JSON syntax, unquotes are given the same above treatment as
.code ,
(comma) unquotes in ordinary syntax.
.coIP ~`...quasilit...`
Similarly, quasiliterals are supported in JSON syntax.
.meIP #H(() >> ( k0 << v0 ) >> ( k1 << v1 ) ...)
Hash quasiliteral syntax is translated according to the
.mono
.meti @(hash <> ([ k0 ] <> [ v0 ]) <> ([ k0 ] <> [ v0 ]) ...)
.onom
pattern, with each key and value recursively translated.
The syntax must specify
.code ()
for the hash construction arguments part, otherwise an error is diagnosed.
That is to say, it must be of the form
.codn "#H(() ...)" .
where the first element is
.codn () .
.meIP >> #S( type < e0 < e1 ...)
Structure quasiliteral syntax is translated according to the
.mono
.meti @(struct <> [ type ] <> [ e0 ] <> [ e1 ] ...)
.onom
pattern.
.meIP >> #( e0 < e1 ...)
Vector quasiliteral syntax is translated according to the
.mono
.meti <> #([ e0 ] <> [ e1 ] ...)
.onom
pattern: it becomes a vector object containing embedded patterns.
.meIP <> #J[ e0 , << e1 , ...]
A JSON array quasiquote is translated into
.mono
.meti <> #([ e0 ] <> [ e1 ] ...)
.onom
exactly like a vector. Here, the
.code [X]
transformation recognizes JSON
.code ~
(tilde) unquotes, and recursively recognizes and transform JSON syntax not
prefixed by
.codn #J .
.meIP >> #J{ k0 : << v0 , < k1 : << v1 , ...}
A JSON hash quasiquote is translated into
.mono
.meti @(hash <> ([ k0 ] <> [ v0 ]) <> ([ k0 ] <> [ v0 ]) ...)
.onom
exactly like a hash.
.meIP >> ( car . << cdr )
Tree structure is translated according to the
.mono
.meti <> ([ car ] . <> [ cdr ])
.onom
pattern: it is recursively examined for translations.
.meIP >> ^ nested-qq-syntax
A nested quasiquote pattern is diagnosed as an error.
.meIP >> ,* expr
Splicing syntax is diagnosed as an error.
.meIP >> ~* expr
Splicing JSON syntax is diagnosed as an error inside a JSON quasiliteral.
.meIP >> ~* expr
.meIP < obj
Any other quasiquoted object is left untranslated.
.RE
.IP
.TP* Examples:
.verb
;; basic unquote: variables embedded via unquote,
;; not requiring @ prefix.
(when-match ^(,a ,b) '(1 2) (list a b))
--> (1 2)
;; operators embedded via unquote; interior of operators
;; is regular non-quasiquoting pattern syntax.
(when-match ^(,(oddp @a) ,(evenp @b)) '(1 2) (list a b))
--> (1 2)
(when-match ^#(,a ,b) #(1 2) (list a b))
--> (1 2)
(when-match ^#S(,type year ,y) #S(time year 2021)
(list (struct-type-name type) y))
--> (time 2021)
(when-match ^#H(() (x ,y) (,(symbolp @y) ,datum))
#H(() (x k) (k 42))
datum)
--> (42)
;; JSON syntax
(when-match ^#J~a 42.0 a) --> 42.0
(when-match ^#J[~a, ~b] #J[true, false] (list a b)) --> (t nil)
(when-match ^#J{"x" : ~y, ~(symbolp @y) : ~datum}
#J{"x" : true, true : 42}
datum)
--> (42.0)
(when-match ^#J{"foo" : {"x" : ~val}}
#J{"foo" : {"x" : "y"}} val)
--> "y"
.brev
.coNP Pattern operator @ struct
.synb
.mets @(struct < name >> { slot-name << pattern }*)
.mets @(struct < pattern >> { slot-name << pattern }*)
.syne
.desc
The
.code struct
pattern operator matches a structure object. The operator
supports two modes of matching, the choice of which depends on whether the
first argument is a
.meta name
or a
.metn pattern .
The first argument is considered a
.meta name
if it is a bindable symbol according to the
.code bindable
function. In this situation, the operator operates in
strict mode. Otherwise, the operator is in loose mode.
The
.meta name
or
.meta pattern
argument is followed by zero or more
.meta "slot-name pattern"
pairs, which are not enclosed in lists, similarly to the way
slots are presented in the
.code #S
struct syntax and in the argument conventions of the
.code new
macro.
In strict mode,
.meta name
is assumed to be the name of an existing struct type.
The object being matched is tested whether it is a subtype of this type, as
if using the
.code subtypep
function. If it isn't, the match fails.
In loose mode, the object being matched is tested whether it is a structure
object of any structure type. If it isn't, the match fails.
In strict mode, each
.meta "slot-name pattern"
pair requires that the object's slot of that name contain
a value which matches
.metn pattern .
The operator assumes that all the
.metn slot-name s
are slots of the struct type indicated by
.metn name .
In loose mode, no assumption is made that the object actually has the
slots specified by the
.meta slot-name
arguments. The object's structure type is inquired to
determine whether it has each of those slots. If it doesn't, the match fails.
If the object has the required slots, then the values of those slots are
matched against the patterns.
In loose mode, the
.meta pattern
given in the first argument position of the syntax is matched against the
object's structure type: the type itself, rather than its symbolic name.
.TP* Examples:
.verb
;; extract the month from a time structure
;; that is required to have a year of 2021.
(when-match @(struct time year 2021 month @m)
#S(time year 2021 month 1)
m) -> 1
;; match any structure with name and value slots,
;; whose name is foo, and extract the value.
(defstruct widget ()
name
value)
(defstruct grommet ()
name
value)
(append-each ((obj (list (new grommet name "foo" value :grom)
(new widget name "foo" value :widg))))
(when-match @(struct @type name "foo" value @v) obj
(list (list type v))))
--> ((#<struct-type grommet> :grom)
(#<struct-type widget> :widg))
.brev
.coNP Pattern operator @ hash
.synb
.mets @(hash >> {( key-pattern <> [ value-pattern ])}*)
.syne
.desc
The
.code hash
pattern operator matches a hash-table object by means of patterns
which target keys, values or both.
An important concept in the requirements governing the operation of the
.code hash
operator is that of a trivial pattern.
A pattern is nontrivial if it is a variable or operator pattern.
A pattern is also nontrivial if it is a list, vector or range pattern
containing at least one nontrivial pattern. Otherwise, it is trivial.
The
.code hash
operator requires the corresponding object to be a hash table,
otherwise the match fails.
If the corresponding object is a hash table, then matches each
.meta key-pattern
and
.meta value-pattern
pair against that object as described below. Each of
the pairs must successfully match, otherwise the overall
match fails.
The following requirements apply to key-value pattern pairs in which
the value pattern is specified.
If
.meta key-pattern
is a trivial pattern, then the semantics of the match is that
.meta key-pattern
is taken as a literal object representing a hash key. The hash
table is searched for that key. If the key is not found,
the match fails. Otherwise, the value corresponding to
that key is matched against the
.meta value-pattern
which may be trivial or nontrivial.
If
.meta key-pattern
is a simple variable pattern
.mono
.meti >> @ sym
.onom
and if
.meta sym
has an existing binding, then the value of
.meta sym
is looked up in the hash table. If it is not found, then
the match fails, otherwise the corresponding value is matched
against
.metn value-pattern ,
which may be trivial or nontrivial.
If
.meta key-pattern
is a nontrivial pattern other than a variable pattern
for a variable which has an existing binding, and if
.meta value-pattern
is trivial, then
.meta value-pattern
is taken as a literal object, which is used for searching
the hash table for one or more keys, as if it were the
.meta value
argument in a call to the
.code hash-keys-of
function, to find all keys which have a value
.code equal
to that value. If no keys are found, then the match
fails. Otherwise, the
.code key-pattern
is then matched against the retrieved list of hash keys.
Finally, if both
.meta key-pattern
and
.meta value-pattern
are nontrivial, then an exhaustive search is performed of the hash table.
Every key in the hash table is matched against
.meta key-pattern
and if it matches, the value is matched against
.metn value-pattern .
If both match, then the values from the matches are collected into lists.
At least one matching key-value pair must be found, otherwise
the overall match fails.
Note: this situation can be understood as if the hash table were
an association list of
.code cons
cells of the form
.verb
.mets >> ( key . << value )
.brev
and as if the two patterns were combined into a
.code coll
operator against this list in the following way:
.verb
.mets @(coll >> ( key-pattern . << value-pattern ))
.brev
such that the semantics can then be understood in terms of the
.code coll
operator matching against an association list.
The following requirements apply when the
.meta value-pattern
is omitted.
If
.meta key-pattern
is a nontrivial pattern other than a variable pattern
for a variable which has an existing binding, then the pattern
is applied against the list of keys from the hash table, which
are retrieved as if using the
.code hash-keys
function.
If
.meta key-pattern
is a variable pattern referring to an existing binding, then that pattern is
taken as a literal object. The match is successful if that object occurs as a
key in the hash table.
.TP* Example:
.verb
;; First, (x @y) has a trivial key pattern so the x
;; entry from the hash table is retrieved, the
;; value being the symbol k. This k is bound to @y.
;; Because y now a bound variable the pattern (@y @datum)
;; is interpreted as search of the hash table for
;; a single entry matching the value of @y. This
;; is the k entry, whose value is 42. The @datum
;; value match takes this 42.
(when-match @(hash (x @y) (@y @datum))
#H(() (x k) (k 42)) datum)
--> 42
;; Again, (x @y) has a trivial key pattern so the x
;; entry from the hash table is retrieved, the
;; value being the symbol k. This k is bound to @y.
;; This time the second pattern has a @(symbolp)
;; predicate operator. This is not a variable, and
;; so the pattern searches the entire
;; hash table. The @y variable has a binding to k,
;; so only the (k 42) entry is matched. The 42
;; value matches @datum, and is collected into a list.
(when-match @(hash (x @y) (@(symbolp @y) @datum))
#H(() (x k) (k 42)) datum)
--> (42)
.brev
.coNP Pattern operator @ as
.synb
.mets @(as < name << pattern )
.syne
.desc
The
.code as
pattern operator binds the corresponding object to a fresh variable given by
.metn name ,
similarly to the Lisp
.code let
operator. If another variable called
.meta name
exists, it is shadowed; thus, no back-referencing is performed.
The
.meta name
argument must be a bindable symbol, or else
.codn nil .
If
.meta name
is
.codn nil ,
then no name is bound. Thus
.mono
.meti @(as nil << pattern )
.onom
is equivalent to
.metn pattern .
Otherwise,
.meta pattern
processed in a scope in which the new
.meta name
binding is already visible.
The
.code as
operator succeeds if
.meta pattern
matches.
Note: in a situation when it is necessary to bind a variable to an object
in parallel with one or more patterns, such that the variable can back-reference
to an existing occurrence, the
.code and
pattern operator can be used.
.TP* Example:
.verb
;; w captures the entire (1 2 3) list:
(when-match @(as w (@a @b @c)) '(1 2 3) (list w a b c))
--> ((1 2 3) 1 2 3)
;; match a list which has itself as the third element
(when-match @(as a (1 2 @a 4)) '#1=(1 2 #1# 4) :yes)
--> :yes
.brev
.coNP Pattern operator @ with
.synb
.mets @(with <> [ main-pattern ] >> { side-pattern | << name } << expr )
.syne
.desc
The
.code with
pattern operator matches the optional
.meta main-pattern
against a corresponding object, while matching a
.meta side-pattern
or
.meta name
against the value of the expression
.meta expr
which is embedded in the syntax.
First, if
.meta main-pattern
is present in the syntax,
it is matched against its corresponding object. This match must
succeed, or else the
.code with
operator fails to match, in which case
.meta expr
is not evaluated.
Next, if
.meta main-pattern
successfully matched, or is absent,
.meta expr
is evaluated in the scope of earlier pattern variables, including any
which that emanate from
.metn main-pattern .
It is unspecified
whether later pattern variables are visible.
Finally,
.meta side-pattern
is matched against the value of
.metn expr .
If that succeeds, then the operator has successfully matched.
If a
.meta name
is specified instead of a
.metn side-pattern ,
it must be a bindable symbol or else
.metn nil .
.TP* Examples:
.verb
(when-match (@(with @a x 42) @b @c) '(1 2 3) (list a b c x))
--> (1 2 3 42)
(let ((o 3))
(when-match (@(evenp @x) @(with @z @(oddp y) o)) '(4 6)
(list x y z)))
--> (4 3 6)
.brev
.coNP Pattern operator @ require
.synb
.mets @(require < pattern << condition *)
.syne
.desc
The pattern operator
.code require
applies the specified
.meta pattern
to the corresponding object.
If the
.meta pattern
matches, the operator then imposes the additional constraints
specified by zero or more
.meta condition
forms.
Each
.meta condition
is evaluated in a scope in which the variables from
.meta pattern
have already been established.
For the
.code require
operator to be a successful match, every
.meta condition
must evaluate true, otherwise the match fails.
The
.meta condition
forms behave as if they were the arguments of an implicit
.code and
operator, which implies left-to-right evaluation behavior, stopping
evaluation on the first
.meta condition
which produces
.codn nil ,
and defaulting to a result of
.code t
when no
.meta condition
forms are specified.
.TP* Examples:
.verb
;; Match a (+ a b) expression where a and b are similar:
(when-match @(require (+ @a @b) (equal a b)) '(+ z z) (list a b))
--> (z z)
;; Mismatched case
(if-match @(require (+ @a @b) (equal a b)) '(+ y z)
(list a b)
:no-match)
--> :no-match
.brev
.coNP Pattern operators @ all and @ all*
.synb
.mets @(all << pattern )
.mets @(all* << pattern )
.syne
.desc
The
.code all
and
.code all*
pattern operators require the corresponding object to be a sequence.
The specified
.meta pattern
is applied against every element of the sequence. The match is successful if
.meta pattern
matches every element.
Furthermore, in the case of a successful match, each variable that
is freshly bound by
.meta pattern
is converted into a list of all of the objects which that variable
encounters from all elements of the sequence. Those variables which already
have a binding from another
.meta pattern
are not converted to lists. Their existing values are merely required to match
each corresponding object they encounter.
The difference between
.code all
and
.code all*
is as follows. The
.code all
operator respects the vacuous truth of the match when the sequence is empty.
In that case, the match is successful, and the variables are all bound to
the empty list
.codn nil .
In contrast, the alternative
.code all*
operator behaves like a failed match when the sequence is empty.
.TP* Examples:
.verb
;; all elements of list match the pattern (x @a @b)
;; a is bound to (1 2 3); b to (a b c)
(when-match @(all (x @a @b))
'((x 1 a) (x 2 b) (x 3 c))
(list a b))
--> ((1 2 3) (a b c))
;; Match a two element list whose second element
;; consists of nothing but zero or more repetitions
;; of the first element. x is not turned into a list
;; because it has a binding due to @x.
(when-match @(@x @(all x)) '(1 (1 1 1 1)) x) -> 1
;; no match because of the 2
(when-match @(@x @(all x)) '(1 (1 1 1 2)) x) -> nil
.brev
.coNP Pattern operator @ some
.synb
.mets @(some << pattern )
.syne
.desc
The
.code some
pattern operator requires the corresponding object to be a sequence.
The specified
.meta pattern
is applied against every element of the sequence. The match is successful if
.meta pattern
matches at least one element.
Variables are extracted from the first matching which is found.
.TP* Example:
.verb
;; the second (x 2 b) element is the leftmost one
;; which matches the (x @a @b) pattern
(when-match @(some (x @a @b))
'((y 1 a) (x 2 b) (z 3 c))
(list a b))
-> (2 b)
.brev
.coNP Pattern operator @ coll
.synb
.mets @(coll << pattern )
.syne
.desc
The
.code coll
pattern operator requires the corresponding object to be a sequence.
The specified
.meta pattern
is applied against every element of the sequence. The match is successful if
.meta pattern
matches at least one element.
Each variable that is freshly bound by the
.meta pattern
is converted into a list of all of the objects which that variable
encounters from the matching elements of the sequence. Those variables which
already have a binding from another
.meta pattern
are not converted to lists. Their existing values are merely required to match
each corresponding object they encounter.
Variables are extracted from all matching elements, and collected into
parallel lists, just like with the
.code @(all)
operator.
.TP* Example:
.verb
(when-match @(coll (x @a @b))
'((y 1 a) (x 2 b) (z 3 c) (x 4 d))
(list a b))
-> ((2 4) (b d))
.brev
.coNP Pattern operator @ scan
.synb
.mets @(scan << pattern )
.syne
.desc
The
.code scan
operator matches
.meta pattern
against the corresponding object. If the match fails, and the object
is a
.code cons
cell, the match is tried on the
.code cdr
of the cons cell. The
.code cdr
traversal repeats until a successful match is found,
or a match failure occurs against against an atom.
Thus, a list object, possibly improper, matches
.meta pattern
under
.code scan
if any suffix of that object matches.
.TP* Examples:
.verb
;; mismatch: 1 doesn't match 2
(when-match @(scan 2) 1 t) -> t
;; simple atom match: 42 matches 42
(when-match @(scan 42) 42 t) -> t
;; (2 3) is a sublist of (1 2 3 4)
(when-match @(scan (2 3 . @nil)) '(1 2 3 4) t) -> t
;; (2 @x 4 . @nil) matches (2 3 4), binding x to 3:
(when-match @(scan (2 @x 4 . @nil)) '(1 2 3 4 5) x) -> 3
;; The entire matching suffix can be captured.
(when-match @(scan @(as sfx (2 @x 4 . @nil)))
'(1 2 3 4 5)
sfx)
-> (2 3 4 5)
;; Missing . @nil in pattern anchors search to end:
(when-match @(scan (@x 2))
'(1 2 3 2 4 2)
x)
;; Terminating atom anchors to improper end:
(when-match @(scan (@x . 4))
'(1 2 3 . 4)
x)
-> 3
;; Atom pattern matches only terminating atom
(when-match @(scan #(@x @y))
'(1 2 3 . #(4 5))
(list x y))
-> (4 5)
.brev
.coNP Pattern operators @ and and @ or
.synb
.mets @(and << pattern *)
.mets @(or << pattern *)
.syne
.desc
The
.code and
and
.code or
operators match multiple patterns in parallel, against
the same object.
The
.code and
operator matches if every
.meta pattern
matches the object, otherwise there is no match.
The
.code or
operator requires one
.meta pattern
to match. It tries the patterns in left-to-right order, and
stops at the first matching one, declaring failure if none match.
The
.code and
and
.code or
operators have different scoping rules.
Under
.codn and ,
later patterns are processed in the scopes of earlier patterns,
just like with other pattern operators. Duplicate variables
back-reference.
Under
.codn or ,
the patterns are processed in separate, parallel scopes.
No back-referencing takes place among same-named variables
introduced in separate patterns of the same
.codn or .
When the
.code and
matches, the variables from all of the
patterns are bound.
When the
.code or
operator matches, the variables from all of the patterns
are also bound. However, only the variables from the matching
.meta pattern
take on the values implied by that pattern.
The variables from the nonmatching patterns that do not have
the same names as variables in the matching
.metn pattern ,
and that have been newly introduced in the
.code or
operator, take on
.code nil
values.
.TP* Examples
.verb
(if-match @(and (@x 2 3) (1 @y 3) (1 2 @z)) '(1 2 3)
(list x y z)) -> (1 2 3)
(if-match @(or (@x 3 3) (1 @x 3) (1 2 @x)) '(1 2 3)
x) -> 2
.brev
.coNP Pattern operator @ not
.synb
.mets @(not << pattern )
.syne
.desc
The pattern operator
.code not
provides logical inverse semantics. It matches if and only if the
.meta pattern
does not match.
Whether or not the
.code not
operator matches, no variables are bound. If the embedded
.meta pattern
matches, the variables which it binds are suppressed by the
.code not
operator.
.TP* Examples:
.verb
;; @a matches unconditionally, so @(not @a) always fails:
(if-match @(not @a) 1 :yes :no) -> :no
;; error: a is not bound
(if-match @(not @a) 1 :yes a) -> error
(match-case '(1 2 3)
((@(not 1) @b @c) (list :case1 b c))
((@(not 0) @b @c) (list :case2 c b)))
--> (:case2 3 2)
.brev
.NP* Pattern predicate operator
.synb
.mets >> @( function << arg *)
.mets >> @( function << arg * >> @ avar << arg *)
.mets >> @( function << arg * . <> @ avar )
.mets >> @(@ rvar >> ( function << arg *))
.mets >> @(@ rvar >> ( function << arg * >> @ avar << arg *))
.mets >> @(@ rvar >> ( function << arg * . <> @ avar ))
.syne
.desc
Whenever the operator position of a pattern consists of a symbol which is
neither the name of a pattern operator, nor the name of a macro, the expression
denotes a predicate pattern. An expression is also a predicate pattern if
it is handled by a pattern macro which declines to expand it by yielding
the original expression.
An operator pattern is expected to conform to one of the first three
syntactic variations above.
Together, these three variations constitute the
.I "first form"
of the pattern predicate operator.
Whenever the operator position of a pattern consists of a meta-symbol, it is
also a predicate pattern, expected to conform to one of the second three syntax
variations above. These three variations constitute the
.I "second form"
of the operator.
The first form of the predicate pattern consists of a compound form consisting
of an operator and arguments. Exactly one of the arguments may be a pattern
variable
.meta avar
("argument variable") which must be a bindable symbol. The pattern variable
may also appear in the dot position, rather than as an argument. The role of
.meta avar
and the consequences of omitting it are described below.
The second form of the predicate pattern consists of a meta-symbol
.meta rvar
("result variable")
which must be a bindable symbol or else
.codn nil .
This is followed by a compound form which consists of an operator
symbol, followed by arguments, one of which may be a pattern
.code avar
as in the simple form.
If
.meta rvar
is
.codn nil ,
then the predicate pattern is equivalent to the first form. That is to say,
the following are equivalent:
.verb
@(@nil (f ...)) <--> @(f ...)
.brev
The matching of the predicate pattern is processed as follows.
If the
.meta avar
variable is present, then the predicate pattern first binds the
corresponding object to the
.meta avar
variable, performing an ordinary variable match with the potential
back-referencing which that implies. If that succeeds, then the object is
inserted into the compound form, substituted in the position indicated by the
.mono
.meti >> @ avar
.onom
variable, either an ordinary argument position or the dot position. This form
is then evaluated. If it yields true, then the match is successful, otherwise
the match fails.
If the
.meta avar
variable is absent, then no initial variable matching takes place.
The corresponding object is added as an extra rightmost argument into the
compound form, which is evaluated. Its truth value then determines the success
of the match, just like in the case with
.metn avar .
If the second form is being processed, and specifies a
.meta rvar
that is not
.codn nil ,
and if the predicate has succeeded, then then an extra processing step takes
place. A variable match is performed to bind the
.meta rvar
variable to the result of the predicate, with potential back-referencing.
If that match succeeds, then the predicate pattern succeeds.
The compound form may be headed by the
.code dwim
operator, and therefore the DWIM bracket notation may be used.
For instance
.code "@[f @x]"
is equivalent to
.code "@(dwim f @x)"
and is processed accordingly. Similarly,
.code "@(@y [f @x])"
is equivalent to
.codn "@(@y (dwim f @x))" .
The dot position of
.meta avar
in the predicate syntax denotes function application. So that is to say, the
pattern predicate form
.code "(f . @a)"
where
.code @a
is in the dotted position invokes the function
.code f
as if by evaluation of the form
.code "(f . x)"
where
.code x
is hidden temporary variable holding the object corresponding to the pattern.
The form
.code "(f . x)"
is a standard \*(TL notation with the same meaning as
.codn "(apply (fun f) x)" .
.TP* Examples:
.verb
(when-match (@(evenp) @(oddp @x)) '(2 3) x) -> 3
(when-match @(<= 1 @x 10) 4 x) -> 4
(when-match @(@d (chr-digit @c)) #\e5 (list d c)) -> (5 #\e5)
(when-match @(<= 1 @x 10) 11 x) -> nil
;; use hash table as predicate:
(let ((h #H(() (a 1) (b 2))))
(when-match @[h @x] 'a x))
-> a
;; as above, also capture hash value
(let ((h #H(() (a 1) (b 2))))
(when-match @(@y [h @x]) 'a (list x y)))
-> (a 1)
;; apply (1 2 3) to < using dot position
(when-match @(@x (< . @sym)) '(1 2 3) (list x sym))
-> (t (1 2 3))
.brev
.coNP Pattern macro @ sme
.synb
.mets @(sme < spat < mpat < epat >> [ mvar <> [ evar ]])
.syne
.desc
The pattern macro
.code sme
(start, middle, end) is a notation defined using the
.code defmatch
macro.
The
.code sme
macro generates a complex pattern which matches three non-overlapping
parts of a list object using three patterns. The
.meta spat
pattern is required to match a prefix of the input list. If that match is
successful, then the remainder of the list is searched for a match for
.metn mpat ,
using the
.code scan
operator. If that match, in turn, is successful, then the suffix of
the remainder of the list is required to match
.codn epat .
The optional
.meta mvar
and
.meta evar
arguments must be bindable symbols, if they are specified.
These symbols specify lexical variables which are bound to, respectively,
the object matched by
.meta mpat
and
.metn epat ,
using the fresh binding semantics of the
.code as
pattern operator.
The first two patterns,
.meta spat
and
.metn mpat ,
must be possibly dotted list patterns.
The last pattern,
.metn epat ,
must be either an atom or a possibly dotted list pattern.
Important to the semantics of
.code sme
is the concept of the length of a list pattern.
The length of a pattern with a pattern variable or operator
in the dotted position is the number of items before that variable
or operator. The length of
.code "(1 2 . @(and a b))"
is 2; likewise the length of
.code "(1 2 . @nil)"
is also 2.
The length of a pattern which does not have a variable or
operator in the dotted position is simply its list length.
For instance, the pattern
.code "(1 2 3)"
has length 3, and so does the pattern
.codn "(1 2 3 . 4)" .
The length is determined by the list object structure of the
pattern, and not the printed syntax used to express it. Thus,
.code "(1 . (2 3))"
is still a length 3 pattern, because it denotes the same
.code "(1 2 3)"
object, using the dot notation unnecessarily.
The non-overlapping semantics of
.code sme
evolves as follows. In the following description, it is understood
that a match is required at every step. If that match fails, then
the entire
.code sme
operator fails:
.RS
.IP 1.
First,
.meta spat
is required to match a a prefix of the input list. If the match
succeeds, then a
.I "middle suffix"
of the input is calculated by dropping from it leading
elements. The number of elements dropped is equal to the length of
.metn spat .
.IP 2.
The middle suffix is then searched for an occurrence of the middle pattern
.metn mpat ,
as if using the
.code scan
pattern operator. All elements skipped by the search are dropped,
until a match is found.
.IP 3.
At that point, if
.meta mvar
has been specified, it is bound to the remaining input, which still
includes the part which just matched
.metn mpat .
.IP 4.
Next, a number of elements equal to the length of
.metn mpat ,
are dropped from the middle suffix, leaving a residue comprising the
.IR "final suffix" .
.IP 5.
The end pattern
.meta epat
must then match a suffix of the final suffix.
.IP 6.
If the
.meta evar
variable has been specified, it is bound to the entire suffix that
was matched by
.metn epat .
.RE
.TP* Examples:
.verb
(when-match @(sme (1 2) (3 4) (5 . 6) m e)
'(1 2 3 4 5 . 6)
(list m e))
-> ((3 4 5 . 6) (5 . 6))
(when-match @(sme (1 2) (3 4) (5 . 6) m e)
'(1 2 abc 3 4 def 5 . 6)
(list m e))
((3 4 def 5 . 6) (5 . 6))
;; backreferencing
(when-match @(sme (1 @y) (@z @x @y @z) (@x @y)) '(1 2 3 1 2 3 1 2)
(list x y z))
-> (1 2 3))
;; collect odd items starting at 3, before 7
(when-match @(and @(sme (1 @x) (3) (7) m e)
@(with @(coll @(oddp @y)) (ldiff m e)))
'(1 2 3 4 5 6 7)
(list x y))
-> (2 (3 5)))
;; no overlap
(when-match @(sme (1 2) (2 3) (3 4)) '(1 2 3 4) t) -> nil
;; The atom 5 is like a "zero-length improper list".
(when-match @(sme () () 5) 5 t) -> t
.brev
.coNP Pattern macro @ end
.synb
.mets @(end < pattern <> [ var ])
.syne
.desc
The pattern macro
.code end
is a notation defined using the
.code defmatch
macro, which matches
.meta pattern
against the suffix of a corresponding list object,
which may be an improper list or atom.
The optional argument
.meta var
specifies the name of a variable which captures the matched portion of the
object.
The
.code end
macro is related to the
.code sme
macro according to the following equivalence:
.verb
@(end pat var) <--> @(sme () () pat : var)
.brev
All of the requirements given for
.code sme
apply accordingly.
.TP* Examples:
.verb
;; atom match
(when-match @(end 3 x) 3 x) -> 3
;; y captures (2 3)
(when-match @(end (2 @x) y)
'(1 2 3)
(list x y))
-> (3 (2 3))
;; variable in dot position
(when-match @(end (2 . @x) y)
'(1 2 . 3)
(list x y))
-> (3 (2 . 3))
;; z captures entire object
(when-match @(as z @(end (2 @x) y))
'(1 2 3)
(list x y z))
-> (3 (2 3) (1 2 3)))
.brev
.SS* Pattern-Matching Macros
.coNP Macros @, when-match @ match and @ if-match
.synb
.mets (when-match < pattern < expr << form *)
.mets (match < pattern < expr << form *)
.mets (if-match < pattern < expr < then-form <> [ else-form ])
.syne
.desc
The
.codn when-match ,
.code match
and
.code if-match
macros conditionally evaluate code based on whether the value of
.meta expr
matches
.metn pattern .
The
.code when-match
macro arranges for every
.meta form
to be evaluated in the scope of the variables established by
.meta pattern
when it matches the object produced by
.metn expr .
The value of the last
.meta form
is returned, or else
.code nil
if there are no forms.
If the match fails, the forms are not evaluated, and
.code nil
is produced.
The
.code match
macro behaves exactly like
.code when-match
when the match is successful. When the match fails,
.code match
throws an exception of type
.codn match-error .
The
.code if-match
macro evaluates
.meta then-form
in the scope of the variables established by
.meta pattern
if the match is successful, and yields the value of that form.
Otherwise, it evaluates
.metn else-form ,
which defaults to
.code nil
if it is not specified.
.coNP Macros @ match-case and @ match-ecase
.synb
.mets (match-case < expr >> {( pattern << form *)}*)
.mets (match-ecase < expr >> {( pattern << form *)}*)
.syne
.desc
The
.code match-case
macro successively matches the value of
.meta expr
against zero or more patterns.
The syntax of
.code match-case
consists of an expression
.meta expr
followed by zero or more clauses.
Each clause is a compound expression whose first element is
.metn pattern ,
which is followed by zero or more forms.
First,
.meta expr
is evaluated. Then, the value is matched against each
.meta pattern
in succession, stopping at the first pattern which provides
a successful match.
If no pattern provides a successful match, then
.code match-case
terminates and returns
.codn nil .
If a
.meta pattern
matches successfully, then each
.meta form
associated with the pattern is evaluated in the scope of the variable
bindings established by that
.metn pattern .
Then
.code match-case
terminates, returning the value of the last
.meta form
or else
.code nil
if there are no forms.
The
.code match-ecase
macro differs from
.code match-case
as follows. When none of the clauses match under
.codn match-case ,
then that form terminates with a value of
.codn nil .
In the same situation, the
.code match-ecase
form throws an exception of type
.codn match-error .
.TP* Examples:
.verb
;; classify sequence of objects by pattern matching,
;; returning a list of the results
(collect-each ((obj (list '(1 2 3)
'(4 5)
'(3 5)
#S(time year 2021 month 1 day 1)
#(vec tor))))
(match-case obj
(@(struct time year @y) y)
(#(@x @y) (list x y))
((@nil @nil @x) x)
((4 @x) x)
((@x 5) x)))
--> (3 5 3 2021 (vec tor))
;; default case can be represented by a guaranteed match
(match-case 1
(2 :two)
(@x :default)) --> :default
.brev
.coNP Macro @ lambda-match
.synb
.mets (lambda-match >> {( pattern << form *)}*)
.syne
.desc
The
.code lambda-match
is conceptually similar to
.codn match-case .
The arguments of
.code lambda-match
are zero or more clauses similar to those of
.codn match-case ,
each consisting of a compound expression headed by a
.meta pattern
followed by zero or more
.metn form s.
The macro generates a
.code lambda
expression which evaluates to an anonymous function
in the usual way.
When the anonymous function is called, each clause's
.meta pattern
is matched against the function's actual arguments. When a
match occurs, each
.meta form
associated with the
.meta pattern
is evaluated, and the value of the last
.meta form
becomes the return value of the function.
If none of the clauses match, then
.code nil
is returned.
Whenever
.meta pattern
is a list-like pattern, it is not matched against a list object, as is the
usual case with a list-like pattern, but against the actual arguments.
For instance, the pattern
.code "(@a @b @c)"
expects that the function was called with exactly three arguments. If
that is the case, the patterns are then matched to the arguments. The pattern
.code @a
takes the first argument, binding it to variable
.code a
and so forth.
If
.meta pattern
is a dotted list-like pattern, then the dot position is matched
against the remaining arguments. For instance, the pattern
.code "(@a @b . @c)"
requires at least two arguments. The first two are bound to
.code a
and
.codn b ,
respectively. The list of remaining arguments, if any, is bound to
.codn c ,
which will be
.code nil
if there are no remaining arguments.
Any non-list-like
.meta pattern
.code P
is analyzed as an equivalent list-like dotted pattern due to
.code P
syntax being equivalent to
.code "(. P)"
syntax. Such a pattern matches the list of all arguments.
Thus, the following are all equivalent:
.verb
(lambda-match (@a a))
(lambda-match ((. @a) a))
(lambda a a)
(lambda (. a) a)
.brev
The characteristics of the resulting anonymous function are determined as
follows.
If at least one
.meta pattern
specified in a
.meta lambda-match
is a dotted pattern, the function is variadic.
The arity of the resulting anonymous function is determined as follows, from
the lengths of the patterns. The length of a pattern is the number of
elements, not including the dotted element.
The length of the longest pattern determines the number of fixed
arguments. Unless the function is variadic, it may not be called with more
arguments than can be matched by the longest pattern.
The length of the shortest pattern determines the number of required arguments.
The function may not be called with fewer arguments than can be matched
by the shortest pattern.
If these two lengths are unequal, then the function has a number of optional
arguments, equal to the difference.
Note: an anonymous function which takes one argument and matches that
object against clauses using
.code match-case
can be obtained with the
.code do
operator, using the pattern:
.codn "(do match @1 ...)" .
Note: the parameter macro
.code :match
can also define a
.code lambda
with pattern matching. Any
.code "(lambda-match clauses ...)"
form can be written as
.codn "(lambda (:match) clauses ...)" .
The parameter macro offers the additional ability of defining
named arguments which are inserted before the implicit arguments
generated from the clauses, and combining with other parameter
macros.
.TP* Examples:
.verb
(let ((f (lambda-match
(() (list 0 :args))
((@a) (list 1 :arg a))
((@a @b) (list 2 :args a b))
((@a @b . @c) (list* '> 2 :args a b c)))))
(list [f] [f 1] [f 1 2] [f 1 2 3]))
-->
((0 :args) (1 :arg 1) (2 :args 1 2) (> 2 :args 1 2 3))
[(lambda-match
((0 1) :zero-one)
((1 0) :one-zero)
((@x @y) :no-match)) 1 0] --> :one-zero
[(lambda-match
((0 1) :zero-one)
((1 0) :one-zero)
((@x @y) :no-match)) 1 1] --> :no-match
[(lambda-match
((0 1) :zero-one)
((1 0) :one-zero)
((@x @y) :no-match)) 1 2 3] --> ;; error
.brev
.coNP Macro @ defun-match
.synb
.mets (defun-match < name >> {( pattern << form *)}*)
.syne
.desc
The
.code defun-match
macro can be used to define a top-level function in the style of
.codn lambda-match .
It produces a form which has all of the properties of
.codn defun ,
such as a block of the same
.meta name
being established around the implicit
.code match-case
so that
.code return-from
is possible.
The
.mono
.meti >> ( pattern << form *)
.onom
clauses of
.code defun-match
have exactly the same syntax and semantics as those of
.codn lambda-match .
Note: instead of
.codn defun-match ,
the parameter macro
.code :match
may be used. The following equivalence holds:
.verb
(defun name (:match) ...) <--> (defun-match ...)
.brev
The parameter macro offers the additional ability of defining
named arguments which are inserted before the implicit arguments
generated from the clauses, and combining with other parameter
macros.
.TP* Examples:
.verb
;; Fibonacci
(defun-match fib
((0) 1)
((1) 1)
((@x) (+ (fib (pred x)) (fib (ppred x)))))
(fib 0) -> 1
(fib 1) -> 1
(fib 2) -> 2
(fib 3) -> 3
(fib 4) -> 5
(fib 5) -> 8
;; Ackermann
(defun-match ack
((0 @n) (+ n 1))
((@m 0) (ack (- m 1) 1))
((@m @n) (ack (- m 1) (ack m (- n 1)))))
(ack 3 7) -> 1021
(ack 1 1) -> 3
(ack 2 2) -> 7
.brev
.coNP Parameter list macro @ :match
.synb
.mets (:match << left-param * [-- << extra-param *]) << clause *
.syne
.desc
Parameter list macro
.code :match
allows any function to be expressed in the style of
.codn lambda-match ,
with extra features.
The
.code :match
macro expects the body of the function to consists of
.code lambda-match
clauses, which are semantically treated in exactly the same manner as
under
.codn lambda-match .
The following restrictions apply. The parameter list may not include
optional parameters delimited by
.code :
(the colon keyword symbol). The parameter list may not be dotted.
The macro produces a function which the
.meta left-param
parameters, if any, are inserted to the left of the implicit parameters
generated by the
.code lambda-match
transformation.
Furthermore, the
.code :match
parameter macro supports integration with the
.code :key
parameter macro, or any other macro which uses a compatible
.code --
convention for delimiting special arguments.
If the parameter list includes the symbol
.code --
then that portion of the parameter list is set aside and not included in the
.code lambda-match
transformation. Then, that list is integrated into the resulting lambda.
A complete transformation can be described by the following diagram:
.verb
(lambda (:match a b c ... -- s t u ...) clauses ...)
-->
(lambda (a b c ... m n p ... -- s t u ... . z) body ...)
.brev
In this diagram,
.code "a b c ..."
denote the
.meta left-param
parameters.
The
.code "m n p ..."
symbols denote the fixed parameters generated by the
.code lambda-match
transformation from the semantic analysis of
.metn clauses .
The
.code "s t u ..."
symbols denote the original
.meta extra-param
parameters. Finally,
.code z
denotes the dotted parameter generated by the
.code lambda-match
transform. If the transform produces no dotted parameter, then this is
.codn nil .
The dotted parameter is thus separated from the
.code "m n p ..."
group to which it belongs.
When no
.code --
and
.meta extra-params
are present, the transformation reduces to:
.verb
(lambda (:match a b c ...) clauses ...)
-->
(lambda (a b c ... m n p ... . z) body ...)
.brev
Note: these requirements harmonize with the
.code :key
parameter macro. If that is present to the left of
.code :match
it removes the
.code --
and the
.code "s t u ..."
keyword parameters, reuniting the
.code z
parameter with the
.code "m n p"
group. Furthermore, the
.code :key
macro generates code which refers to the existing
.code z
dotted parameter as the source for the keyword parameters, unless
.code z
is
.codn nil ,
in which case it inserts its own generated symbol.
.TP* Examples:
.verb
;; Match-style cond-like macro with unreachability diagnosis.
;; Demonstrates usefulness of :match, which allows the :form
;; parameter to be promoted through to the macro definition.
(defmacro my-cond (:match :form f)
(() nil)
(((@(and @(constantp @test) @(eval))) . @rest)
(when rest
(compile-error f "unreachable code after ~s" test))
test)
(((@(and @(constantp @test) @(eval)) . @forms) . @rest)
(when (and rest)
(compile-error f "unreachable code after ~s" test))
^(progn ,*forms))
(((@test) . @rest)
^(or ,test (my-cond ,*rest)))
(((@test . @forms) . @rest)
^(if ,test (progn ,*forms)
(my-cond ,*rest)))
((@else . @rest) (compile-error f "bad syntax")))
(my-cond (3)) --> 3
(my-cond (3 4)) --> 4
(my-cond (3 4) (5)) --> ;; my-cond: unreachable code after 3
(my-cond 42) --> ;; my-cond: bad syntax
.brev
.verb
;; Keyword parameter example.
(defstruct simple-widget ()
name)
(defstruct widget (simple-widget)
frobosity
luminance)
(defstruct simple-point-widget (simple-widget)
(:static width 0)
(:static height 0))
(defstruct point-widget (widget)
(:static width 0)
(:static height 0))
(defstruct general-widget (widget)
width
height)
;; Note that in clauses with no . @rest parameter, there
;; is a mismatch if keyword arguments are present. The (0 0)
;; clause exploits this to match only when keywords are absent.
(defun make-widget (:key :match name -- frob lum)
((0 0) (new simple-point-widget name name))
((0 0 . @rest) (new point-widget name name
frobosity frob
luminance lum))
((@x @y . @rest) (new general-widget name name
width x
height x
frobosity frob
luminance lum)))
(make-widget "abc" 0 0) --> #S(simple-point-widget name "abc")
(make-widget "abc" 0 0 :frob 42)
--> #S(point-widget name "abc" frobosity 42 luminance nil)
(make-widget "abc" 0 0 :lum 9)
--> #S(point-widget name "abc" frobosity nil luminance 9)
(make-widget "abc" 0 1 :lum 9)
--> #S(general-widget name "abc" frobosity nil luminance 9
width 0 height 0)
.brev
.coNP Macro @ defmatch
.synb
.mets (defmatch < name < macro-style-params
.mets \ \ << body-form *)
.syne
.desc
The
.code defmatch
macro allows for the definition of pattern macros: user-defined pattern
operators which are implemented via expansion into existing operator syntax.
The
.code defmatch
macro has the same syntax as
.codn defmacro .
It specifies a macro transformation for a compound form which has the
.meta name
symbol in its leftmost position.
This macro transformation is performed when
.meta name
is used as a pattern operator: an expression of the form
.mono
.meti >> @( name << argument *)
.onom
occurring in pattern-matching syntax.
The behavior is unspecified if
.meta name
is the name a built-in pattern operator, or a predefined pattern macro.
The pattern macro bindings are stored in a hash table held by the variable
.code *match-macro*
whose keys are symbols, and whose values are expander functions.
There are no lexically scoped pattern macros.
Pattern macros defined with
.code defmatch
may specify the special macro parameters
.code :form
and
.code :env
in their parameter lists. The values of these parameters are determined
in a manner particular to
.codn defmatch .
The
.code :form
parameter captures the pattern-matching form, or a constituent thereof, in
which the macro is being invoked. For instance, if the operator is being
used inside a pattern given to a
.code when-match
macro invocation, then the form will be that entire
.code when-match
form.
The
.code :env
parameter captures a specially constructed macro-time environment object in
which all of the variables to the left of the pattern appear as lexical
variables. The parent of this environment is the surrounding macro environment.
If the pattern macro needs to treat a variable which already has a binding
differently from an unbound variable, it can look up the variable in this
environment.
.TP* Example:
.verb
;; Create an alias called let for the @(as var pattern) operator:
;; Note that the macro produces @(as ...) and not just (as ...)
(defmatch let (var pattern)
^@(as ,var ,pattern))
;; use the macro in matching:
(when-match @(let x @(or foo bar)) 'foo x)
;; Error reporting example using :form
(defmatch foo (sym)
(unless (bindable sym)
(compile-error *match-form*
"~s: bindable symbol expected, not ~s"
'foo sym))
...)
;; Pattern macro which uses = equality to backreference
;; an existing lexical binding, or else binds the variable
;; if it has no existing lexical binding.
(defmatch var= (sym :env e)
(if (lexical-var-p e sym)
(with-gensyms (obj)
^@(require (sys:var ,obj)
(= ,sym ,obj)))
^(sys:var ,sym)))
;; example use:
(when-match (@(var= a) @(var= a)) '(1 1.0) a)
-> 1
;; no match: (equal 1 1.0) is false
(when-match (@a @a) '(1 1.0) a)
-> nil
.brev
.coNP Special variable @ *match-macro*
.desc
The
.code *match-macro*
special variable holds the hash table of associations between
symbols and pattern macro expanders.
If the expression
.code "[*match-macro* 'sym]"
yields a function, then symbol
.code sym
has a binding as a pattern macro. If that
expression yields
.codn nil ,
then there is no such binding: pattern operator forms based on
.code sym
do not undergo place macro expansion.
The macro expanders in
.code *match-macro*
are two-parameter functions. The first argument passes the operator
syntax to be expanded. The second argument is used for passing the
environment object which the expander can capture using
.code :env
in its macro parameter list.
.coNP Macros @ each-match and @ each-match-product
.synb
.mets (each-match >> ({ pattern << seq-form }*) << body-form *)
.mets (each-match-product >> ({ pattern << seq-form }*) << body-form *)
.syne
.desc
The
.code each-match
macro arranges for elements from multiple sequences to be
visited in parallel, and each to be matched against respective patterns.
For each matching tuple of parallel elements, a body of forms is evaluated in
the scope of the variables bound in the patterns.
The first argument of
.code each-match
specifies a list of alternating
.meta pattern
and
.meta seq-form
expressions. Each
.meta pattern
is associated with the sequence which results from evaluating the
immediately following
.metn seq-form .
Items coming from that sequence correspond with that pattern.
The remaining arguments are
.metn body-form s
to evaluated for successful matches.
The
.metn body-form s
are surrounded by an implicit anonymous block. If any of the forms
.code return
invoke a return out of this block, then the iteration terminates, and
the result value of the block becomes the result value of
the loop.
The processing takes place as follows:
.RS
.IP 1.
Every
.meta seq-form
is evaluated in left-to-right order and is expected to produce an
iterable sequence or object that would be a suitable argument to
.code mapcar
or
.codn iter-begin .
This evaluation takes place in the scope surrounding the macro form,
in which none of the variables that are bound in the
.meta pattern
expressions are yet visible.
.IP 2.
The next available item is taken from each of the sequences.
If any of the sequences has no more items available, then
.code each-match
terminates and returns
.codn nil .
.IP 3.
Each item taken in step 2 is matched against the
.meta pattern
which is corresponds with its sequence. Each successive pattern can
refer to the variables bound in the previous patterns in the same
iteration. If any pattern match fails, then the process continues with step 2.
.IP 4.
If all the matches are successful, then
.metn body-form s,
if any, are executed in the scope of variables bound in the
.metn pattern s.
Processing then continues at step 2.
.RE
.IP
The
.code each-match-product
differs from
.code each-match
in that instead of taking parallel tuples of items from the sequences,
it iterates over the tuples of the Cartesian product of the sequences
similarly to the
.code maprod
function. The product tuples are ordered in such a way that the rightmost
element, which always coming coming from sequence produced by the last
.metn seq-form ,
varies the fastest. If there are two sequences
.code "(1 2)"
and
.codn "(a b)" ,
then
.code each-match
iterates over the tuples
.code "(1 a)"
and
.codn "(2 b)" ,
whereas
.code each-match-product
iterates over
.codn "(1 a)" ,
.codn "(1 b)" ,
.code "(2 a)"
and
.codn "(2 b)" .
.TP* Examples:
.verb
;; Number all the .JPG files in the current directory.
;; For instance foo.jpg becomes foo-0001.jpg, if it is
;; the first file.
(each-match (@(as name `@base.jpg`) (glob "*.jpg")
@(@num (fmt "~,04a")) 1)
(rename-path name `@base-@num.jpg`))
;; Iterate over combinations of matching phone
;; numbers and odd integers from the (1 2 3) list
(build
(each-match-product (`(@a) @b-@c` '("x"
""
"(311) 555-5353"
"(604) 923-2323"
"133"
"4-5-6-7")
@(oddp @x) '(1 2 3))
(add (list x a b c))))
-->
((1 "311" "555" "5353") (3 "311" "555" "5353")
(1 "604" "923" "2323") (3 "604" "923" "2323")))
.brev
.coNP Macros @ append-matches and @ append-match-products
.synb
.mets (append-matches >> ({ pattern << seq-form }*) << body-form *)
.mets (append-match-products >> ({ pattern << seq-form }*) << body-form *)
.syne
.desc
The macro
.code append-matches
is subject to all of the requirements specified for
.code each-match
in regard to the argument conventions and semantics,
and the presence of the implicit anonymous block around the
.metn body-form s.
Whereas
.code each-match
returns
.codn nil ,
the
.code append-matches
macro requires, in each iteration which produces a match for each
.metn pattern ,
that the last
.meta body-form
evaluated must produce a list.
These lists are catenated together as if by the
.code append
function and returned.
It is unspecified whether the nonmatching iterations produce
empty lists which are included in the append operation.
If the last tuple of items which produces a match is absolutely the
the last tuple, the corresponding
.meta body-form
evaluation may yield an atom which then becomes the terminator
for the returned list, in keeping with the semantics of
.codn append .
an atom.
The
.code append-match-products
macro differs from
.code append-matches
in that it iterates over the Cartesian product tuples of the sequences,
rather than parallel tuples. The difference is exactly like that between
.code each-match
and
.codn each-match-product .
.TP* Examples:
.verb
(append-matches
((:foo @y) '((:foo a) (:bar b) (:foo c) (:foo d))
(@x :bar) '((1 :bar) (2 :bar) (3 :bar) (4 :foo)))
(list x y))
--> (1 a 3 c)
(append-matches (@x '((1) (2) (3) 4)) x)
--> (1 2 3 . 4)
(append-match-products (@(oddp @x) (range 1 5)
@(evenp @y) (range 1 5))
(list x y))
--> (1 2 1 4 3 2 3 4 5 2 5 4)
.brev
.coNP Macros @ keep-matches and @ keep-match-products
.synb
.mets (keep-matches >> ({ pattern << seq-form }*) << body-form *)
.mets (keep-match-products >> ({ pattern << seq-form }*) << body-form *)
.syne
.desc
The macro
.code keep-matches
is subject to all of the requirements specified for
.code each-match
in regard to the argument conventions and semantics,
and the presence of the implicit anonymous block around the
.metn body-form s.
Whereas
.code each-match
returns
.codn nil ,
the
.code keep-matches
macro returns a list of the values produced by all matching iterations which
led to the execution of the
.metn body-form s.
The
.code keep-match-products
macro differs from
.code keep-matches
in that it iterates over the Cartesian product tuples of the sequences,
rather than parallel tuples. The difference is exactly like that between
.code each-match
and
.codn each-match-product .
.TP* Examples:
.verb
(keep-matches ((:foo @y) '((:foo a) (:bar b) (:foo c) (:foo d))
(@x :bar) '((1 :bar) (2 :bar) (3 :bar) (4 :foo)))
(list x y))
--> ((1 a) (3 c))
(keep-match-products (@(oddp @x) (range 1 5)
@(evenp @y) (range 1 5))
(list x y))
--> ((1 2) (1 4) (3 2) (3 4) (5 2) (5 4))
.brev
.coNP Macro @ while-match
.synb
.mets (while-match < pattern < expr << form *)
.syne
.desc
The
.code while-match
macro evaluates
.meta expr
and matches it against
.meta pattern
similarly to
.codn when-match .
If the match is successful, every
.meta form
is evaluated in an environment in which new bindings from
.meta pattern
are visible. In this case, the process repeats:
.meta expr
is evaluated again, and tested against
.metn pattern .
If the match fails,
.code while-match
terminates and produces
.code nil
as its result value.
Each iteration produces fresh bindings for any variables
that are implicated for binding in
.metn pattern .
The
.meta expr
and
.meta form
expressions are surrounded by an anonymous block.
.coNP Macros @ while-match-case and @ while-true-match-case
.synb
.mets (while-match-case < expr >> {( pattern << form *)}*)
.mets (while-true-match-case < expr >> {( pattern << form *)}*)
.syne
.desc
The macros
.code while-match-case
and
.code while-true-match-case
combine iteration with the semantics of
.codn match-case .
The
.code while-match-case
evaluates
.meta expr
and matches it against zero or more clauses in the manner of
.code match-case.
If there is a match, this process is repeated.
If there is no match,
.code while-match-case
terminates, and returns
.codn nil .
In each iteration, the matching clause produces fresh bindings for any
variables implicated for binding in its respective
.metn pattern .
The
.meta expr
and
.meta form
expressions are surrounded by an anonymous block.
The
.code while-true-match-case
macro is identical in almost every respect to
.codn while-match-case ,
except that it terminates the loop if
.meta expr
evaluates to
.codn nil ,
without attempting to match that value against the clauses.
Note: the semantics of
.code while-true-match-case
can be obtained in
.code while-match-case
by inserting a
.code return
clause. That is to say, a construct of the form
.verb
(while-true-match-case expr
...)
.brev
may be rewritten into
.verb
(while-match-case expr
(nil (return)) ;; match nil and return
...)
.brev
except that
.code while-true-match-case
isn't required to rely on performing a block return.
.SS* Quasiquote Operator Syntax
.coNP Macro @ qquote
.synb
.mets (qquote << form )
.syne
.desc
The
.code qquote
(quasi-quote) macro operator implements a notation for convenient
list construction. If
.meta form
is an atom, or a list structure which
does not contain any
.code unquote
or
.code splice
operators, then
.mono
.meti (qquote << form )
.onom
is equivalent to
.mono
.meti (qquote << form ).
.onom
If
.metn form ,
however, is a list structure which contains
.code unquote
or
.code splice
operators, then the substitutions implied by those operators are performed
on
.metn form ,
and the
.code qquote
operator returns the resulting structure.
Note: how the qquote operator actually works is that it is compiled into
code. It becomes a Lisp expression which, when evaluated, computes the
resulting structure.
A
.code qquote
can contain another
.codn qquote .
If an
.code unquote
or
.code splice
operator occurs
within a nested
.codn qquote ,
it belongs to that
.codn qquote ,
and not to the outer one.
However, an unquote operator which occurs inside another one belongs one level
higher. For instance in
.verb
(qquote (qquote (unquote (unquote x))))
.brev
the leftmost
.code qquote
belongs with the rightmost unquote, and the inner
.code qquote
and
.code unquote
belong together. When the outer
.code qquote
is evaluated,
it will insert the value of
.codn x ,
resulting in the object
.codn "(qquote (unquote [value-of-x]))" .
If this resulting qquote value is evaluated again as Lisp syntax, then it will
yield
.codn [value-of-value-of-x] ,
the value of
.code [value-of-x]
when treated as a Lisp expression and evaluated.
.TP* Examples:
.verb
(qquote a) -> a
(qquote (a b c)) -> (a b c)
(qquote (1 2 3 (unquote (+ 2 2)) (+ 2 3))) -> (1 2 3 4 (+ 2 3))
(qquote (unquote (+ 2 2))) -> 4
.brev
In the second-to-last example, the
.code "1 2 3"
and the
.code "(+ 2 3)"
are quoted verbatim.
Whereas the
.code "(unquote (+ 2 2))"
operator caused the evaluation of
.code "(+ 2 2)"
and the substitution of the resulting value.
The last example shows that
.meta form
can itself (the entire argument of
.codn qquote )
can be an unquote operator.
However, note:
.code "(quote (splice form))"
is not valid.
Note: a way to understand the nesting behavior is a via a possible model of
quasi-quote expansion which recursively compiles any nested quasi quotes first,
and then treats the result of their expansion. For instance, in the processing
of
.verb
(qquote (qquote (unquote (unquote x))))
.brev
the
.code qquote
operator first encounters the
embedded
.code "(qquote ...)"
and compiles it to code. During that recursive
compilation, the syntax
.code "(unquote (unquote x))"
is encountered. The inner quote
processes the outer unquote which belongs to it, and the inner
.code "(unquote x)"
becomes material that is embedded verbatim in the compilation, which will then
be found when the recursion pops back to the outer quasiquote, which will
then traverse the result of the inner compilation and find the
.codn "(unquote x)" .
.TP* "Dialect Note:"
In Lisp dialects which have a published quasiquoting operator syntax, there is
the expectation that the quasiquote read syntax corresponds to it. That is to
say, the read syntax
.code "^(a b ,c)"
is expected to translate to
.codn "(qquote a b (unquote c))" .
In \*(TL, this is not true! Although
.code "^(a b ,c)"
is translated to a
quasiquoting macro, it is an internal one, not based on the public
.codn qquote ,
.code unquote
and
.code splice
symbols being documented here.
This idea exists for hygiene. The quasiquote read syntax is not confused
by the presence of the symbols
.codn qquote ,
.code unquote
or
.code splice
in the template, since it doesn't treat them specially.
This also allows programmers to use the quasiquote read syntax to construct
quasiquote macros. For instance
.verb
^(qquote (unquote ,x)) ;; does not mean ^^,,x !
.brev
To the quasiquote reader, the
.code qquote
and
.code unquote
symbols mean nothing special,
and so this syntax simply means that if the value of
.code x
is
.codn foo ,
the result of evaluating this expression will be
.codn "(qquote (unquote foo))" .
The form's expansion is actually this:
.verb
(sys:qquote (qquote (unquote (sys:unquote x))))
.brev
the
.code sys:qquote
macro recognizes
.code sys:unquote
embedded in the form, and
the other symbols not in the
.code sys:
package are just static template material.
The
.code sys:quote
macro and its associated
.code sys:unquote
and
.code sys:splice
operators work exactly like their ordinary counterparts. So in effect, \*(TX has
two nearly identical, independent quasi-quote implementations, one of which is
tied to the read syntax, and one of which isn't. This is useful for writing
quasiquotes which write quasiquotes.
.coNP Operator @ unquote
.synb
.mets (qquote (... (unquote << form ) ...))
.mets (qquote (unquote << form ))
.syne
.desc
The
.code unquote
operator is not an operator
.I per
.IR se .
The
.code unquote
symbol has no
binding in the global environment. It is a special syntax that is recognized
within a
.code qquote
form, to indicate forms within the quasiquote which are to be
evaluated and inserted into the resulting structure.
The syntax
.mono
.meti (qquote (unquote << form ))
.onom
is equivalent to
.metn form :
the
.code qquote
and
.code unquote
"cancel out".
.coNP Operator @ splice
.synb
.mets (qquote (... (splice << form ) ...))
.syne
.desc
The
.code splice
operator is not an operator
.I per
.IR se .
The
.code splice
symbol has no
binding in the global environment. It is a special syntax that is recognized
within a
.code qquote
form, to indicate forms within the quasiquote which are to be
evaluated and inserted into the resulting structure.
The syntax
.mono
.meti (qquote (splice << form ))
.onom
is not permitted and raises an exception if evaluated. The
.code splice
syntax must occur within a list, and not in the dotted position.
The
.code splice
form differs from unquote in that
.mono
.meti (splice << form )
.onom
requires that
.meta form
must evaluate to a list. That list is
integrated into the surrounding list.
.SS* Math Library
.coNP Functions @ + and @ -
.synb
.mets (+ << number *)
.mets (- < number << number *)
.mets (* << number *)
.syne
.desc
The
.codn + ,
.code -
and
.code *
functions perform addition, subtraction and multiplication,
respectively. Additionally, the
.code -
function performs additive inverse.
The
.code +
function requires zero or more arguments. When called with no
arguments, it produces 0 (the identity element for addition), otherwise it
produces the sum over all of the arguments.
Similarly, the
.code *
function requires zero or more arguments. When called
with no arguments, it produces 1 (the identity element for multiplication).
Otherwise it produces the product of all the arguments.
The semantics of
.code -
changes from subtraction to additive inverse
when there is only one argument. The argument is treated as a subtrahend,
against an implicit minuend of zero. When there are two or more
argument, the first one is the minuend, and the remaining are subtrahends.
When there are three or more operands, these operations are performed as if by
binary operations, in a left-associative way. That is to say,
.code "(+ a b c)"
means
.codn "(+ (+ a b) c)" .
The sum of
.code a
and
.code b
is computed first, and then this is added to
.codn c .
Similarly
.code "(- a b c)"
means
.codn "(- (- a b) c)" .
First,
.code b
is subtracted from
.codn a ,
and then
.code c
is subtracted from that result.
The arithmetic inverse is performed as if it were subtraction from integer 0.
That is,
.code "(- x)"
means the same thing as
.codn "(- 0 x)" .
The operands of
.codn + ,
.code -
and
.code *
can be characters, integers (fixnum and bignum), and
floats, in nearly any combination.
If two operands have different types, then one of them is converted to the
type of the one with the higher rank, according to this ranking:
character < integer < float. For instance if one operand is integer, and the
other float, the integer is converted to a float.
.TP* Restrictions:
Characters are not considered numbers, and participate in these operations in
limited ways. Subtraction can be used to computed the displacement between the
Unicode values of characters, and an integer displacement can be added to a
character, or subtracted from a character. For instance
.codn "(- #\e9 #\e0) is 9" .
The Unicode value of a character
.code C
can be found using
.codn "(- C #\ex0)" :
the displacement from the NUL character.
The rules can be stated as a set of restrictions:
.RS
.IP 1.
Two characters may not be added together.
.IP 2.
A character may not be subtracted from an integer (which also rules out
the possibility of computing the additive inverse of a character).
.IP 3.
A character operand may not be opposite to a floating point operand
in any operation.
.IP 4.
A character may not be an operand of multiplication.
.RE
.coNP Function @ /
.synb
.mets (/ << divisor )
.mets (/ < dividend << divisor *)
.syne
.desc
The
.code /
function performs floating-point division. Each operands is first
converted to floating-point type, if necessary. In the one-argument
form, the
.meta dividend
argument is omitted. An implicit dividend is present, whose value is
.codn 1.0 ,
such that the one-argument form
.code "(/ x)"
is equivalent to the two-argument form
.codn "(/ 1.0 x)" .
If there are two or more arguments, explicitly or by the above equivalence,
then a cumulative division is performed. The
.meta divisor
value is taken into consideration, and divided by the first
.codn divisor .
If another
.code divisor
follows, then that value is divided by that subsequent divisor.
This process repeats until all divisors are exhausted, and the
value of the last division is returned.
A division by zero throws an exception of type
.codn numeric-error .
.coNP Functions @ sum and @ prod
.synb
.mets (sum < sequence <> [ keyfun ])
.mets (prod < sequence <> [ keyfun ])
.syne
.desc
The
.code sum
and
.code prod
functions operate on an effective sequence of numbers derived from
.metn sequence ,
which is an object suitable for iteration according to
.codn seq-begin .
If the
.meta keyfun
argument is omitted, then the effective sequence is the
.meta sequence
argument itself. Otherwise, the effective sequence is understood to be
a projection mapping of the elements of
.meta sequence
through
.meta keyfun
as would be calculated by the
.mono
.meti (mapcar < keyfun << sequence )
.onom
expression.
The
.code sum
function returns the left-associative sum of the elements of
the effective sequence calculated as if using the
.code +
function. Similarly, the
.code prod
function calculates the left-associative product of the elements of
the sequence as if using the
.code *
function.
If
.meta sequence
is empty then
.code sum
returns
.code 0
and
.code prod
returns
.codn 1 .
If the effective sequence contains one number, then both functions
return that number.
.coNP Functions @ wrap and @ wrap*
.synb
.mets (wrap < start < end << number )
.mets (wrap* < start < end << number )
.syne
.desc
The
.code wrap
and
.code wrap*
functions reduce
.meta number
into the range specified by
.meta start
and
.metn end .
Under
.code wrap
the range is inclusive of the
.meta end
value, whereas under
.code wrap*
it is exclusive.
The following equivalence holds
.verb
(wrap a b c) <--> (wrap* a (succ b) c)
.brev
The expression
.code "(wrap* x0 x1 x)"
performs the following calculation:
.mono
.mets (+ (mod (- x x0) (- x1 x0)) x0)
.onom
In other words, first
.meta start
is subtracted from
.metn number .
Then the result is reduced modulo the displacement
between
.code start
and
.codn end .
Finally,
.meta start
is added back to that result, which is returned.
.TP* Example:
.verb
;; perform ROT13 on the string "nop"
[mapcar (opip (+ 13) (wrap #\ea #\ez)) "nop"] -> "abc"
.brev
.coNP Functions @ gcd and @ lcm
.synb
.mets (gcd << number *)
.mets (lcm << number *)
.syne
.desc
The
.code gcd
function computes the greatest common divisor: the largest positive
integer which divides each
.metn number .
The
.code lcm
function computes the lowest common multiple: the smallest positive
integer which is a multiple of
each
.metn number .
Each
.meta number
must be an integer.
Negative integers are replaced by their absolute values, so
.code "(lcm -3 -4)"
is
.code 12
and
.code "(gcd -12 -9)"
yields
.codn 3 .
The value of
.code (gcd)
is
.code 0
and that of
.code (lcm)
is 1 .
The value of
.code "(gcd x)"
and
.code "(lcm x)"
is
.codn "(abs x)" .
Any arguments of
.code gcd
which are zero are effectively ignored so that
.code "(gcd 0)"
and
.code "(gcd 0 0 0)"
are both the same as
.code (gcd)
and
.code "(gcd 1 0 2 0 3)"
is the same as
.codn "(gcd 1 2 3)" .
If
.code lcm
has any argument which is zero, it yields zero.
.coNP Function @ divides
.synb
.mets (divides < d << n )
.syne
.desc
The
.code divides
function tests whether integer
.meta d
divides integer
.metn n .
If this is true,
.code t
is returned, otherwise
.codn nil .
The integers 1 and -1 divide every other integer and themselves.
By established convention, every integer, except zero, divides zero.
For other values,
.meta d
divides
.meta n
if division of
.meta n
by
.meta d
leaves no remainder.
.coNP Function @ abs
.synb
.mets (abs << number )
.syne
.desc
The
.code abs
function computes the absolute value of
.metn number .
If
.meta number
is positive, it is returned. If
.meta number
is negative, its additive inverse is
returned: a positive number of the same type with exactly the same magnitude.
.coNP Function @ signum
.synb
.mets (signum << number )
.syne
.desc
The
.code signum
function calculates a representation of the sign of
.meta number
as a numeric value.
If
.meta number
is an integer, then
.code signum
returns -1 if the integer is negative, 1 if the integer is positive,
or else 0.
If
.meta number
is a floating-point value then
.code signum
returns -1.0 if the value is negative, 1.0 if the value is positive or
else 0.0.
.coNP Functions @, trunc @, floor @ ceil and @ round
.synb
.mets (trunc < dividend <> [ divisor ])
.mets (floor < dividend <> [ divisor ])
.mets (ceil < dividend <> [ divisor ])
.mets (round < dividend <> [ divisor ])
.syne
.desc
The
.codn trunc ,
.codn floor ,
.code ceil
and
.code round
functions perform division of the
.meta dividend
by the
.metn divisor ,
returning an integer quotient.
If the
.meta divisor
is omitted, it defaults to 1.
A zero
.meta divisor
results in an exception of type
.codn numeric-error .
If both inputs are integers,
the result is of type integer.
If all inputs are numbers and at least one of them is
floating-point, the others are converted to floating-point
and the result is floating-point.
The
.meta dividend
input may be a range. In this situation, the operation is
recursively distributed over the
.code from
and
.code to
fields of the range, individually matched against the
.metn divisor ,
and the result is a range composed of these two individual
quotients.
When the quotient is a scalar value,
.code trunc
returns the closest integer, in the zero direction,
from the value of the quotient.
The
.code floor
function returns the highest integer which does not exceed
the value of the quotient. That is to say, the division is
truncated to an integer value toward negative infinity.
The
.code ceil
function the lowest integer which is not below the value
of the quotient.
does not exceed the value of
.metn dividend .
That is to say, the division is truncated to an integer
value toward positive infinity. The
.code round
function returns the nearest integer to the quotient.
Exact halfway cases are rounded to the integer away from
zero so that
.code "(round -1 2)"
yields
.code -1
and
.code "(round 1 2)"
yields 1.
Note that for large floating point values, due to the limited
precision, the integer value corresponding to the mathematical
floor or ceiling may not be available.
.TP* "Dialect Note:"
In ANSI Common Lisp, the
.code round
function chooses the nearest even integer, rather than
rounding halfway cases away from zero. \*(TX's choice
harmonizes with the semantics of the
.code round
function in the C language.
.coNP Function @ mod
.synb
.mets (mod < dividend << divisor )
.syne
.desc
The
.code mod
function performs a modulus operation. Firstly, the absolute value
of
.meta divisor
is taken to be a modulus. Then a residue of
.meta dividend
with respect to
.meta modulus
is calculated. The residue's sign follows
that of the sign of
.metn divisor .
That is, it is the smallest magnitude
(closest to zero) residue of
.meta dividend
with respect to the absolute
value of
.metn divisor ,
having the same sign as
.metn divisor .
If the operands are integer, the result is an integer. If either operand
is of type float, then the result is a float. The modulus operation is
then generalized into the floating point domain. For instance the expression
.code "(mod 0.75 0.5)"
yields a residue of 0.25 because 0.5 "goes into" 0.75 only
once, with a "remainder" of 0.25.
If
.meta divisor
is zero,
.code mod
throws an exception of type
.codn numeric-error .
.coNP Functions @, trunc-rem @, floor-rem @ ceil-rem and @ round-rem
.synb
.mets (trunc-rem < dividend <> [ divisor ])
.mets (floor-rem < dividend <> [ divisor ])
.mets (ceil-rem < dividend <> [ divisor ])
.mets (round-rem < dividend <> [ divisor ])
.syne
.desc
These functions, respectively, perform the same division operation
as
.codn trunc ,
.codn floor ,
.codn ceil ,
and
.codn round ,
referred to here as the respective target functions.
If the
.meta divisor
is missing, it defaults to 1.
Each function returns a list of two values: a
.meta quotient
and a
.metn remainder .
The
.meta quotient
is exactly the same value as what would be returned by the
respective target function for the same inputs.
The
.meta remainder
value obeys the following identity:
.mono
.mets (eql < remainder (- < dividend >> (* divisor << quotient )))
.onom
If
.meta divisor
is zero, these functions throw an exception of type
.codn numeric-error .
.coNP Functions @, sin @, cos @, tan @, asin @, acos @ atan and @ atan2
.synb
.mets (sin << radians )
.mets (cos << radians )
.mets (tan << radians )
.mets (atan << slope )
.mets (atan2 < y << x )
.mets (asin << num )
.mets (acos << num )
.syne
.desc
These trigonometric functions convert their argument to floating point and
return a float result. The
.codn sin ,
.code cos
and
.code tan
functions compute the sine and
cosine and tangent of the
.meta radians
argument which represents an angle
expressed in radians. The
.codn atan ,
.code acos
and
.code asin
are their respective inverse
functions. The
.meta num
argument to
.code asin
and
.code acos
must be in the
range -1.0 to 1.0. The
.code atan2
function converts the rectilinear coordinates
.meta x
and
.meta y
to an angle in polar coordinates in the range [0, 2\(*p).
.coNP Functions @, sinh @, cosh @, tanh @, asinh @ acosh and @ atanh
.synb
.mets (sinh << argument )
.mets (cosh << argument )
.mets (tanh << argument )
.mets (atanh << argument )
.mets (asinh << argument )
.mets (acosh << argument )
.syne
.desc
These functions are the hyperbolic analogs of the trigonometric functions
.codn sin ,
.code cos
and so forth. They convert their argument to floating point and
return a float result.
.coNP Functions @, exp @, log @ log10 and @ log2
.synb
.mets (exp << arg )
.mets (log << arg )
.mets (log10 << arg )
.mets (log2 << arg )
.syne
.desc
The
.code exp
function calculates the value of the transcendental number e raised to
the exponent
.metn arg .
The
.code log
function calculates the base e logarithm of
.metn arg ,
which must be a positive value.
The
.code log10
function calculates the base 10 logarithm of
.metn arg ,
which must be a positive value.
The
.code log2
function calculates the base 2 logarithm of
.metn arg ,
which must be a positive value.
.coNP Functions @, expt @ sqrt and @ isqrt
.synb
.mets (expt < base << exponent *)
.mets (sqrt << arg )
.mets (isqrt << arg )
.syne
.desc
The
.code expt
function raises
.meta base
to zero or more exponents given
by the
.meta exponent
arguments.
.code "(expt x)"
is equivalent to
.codn "(expt x 1)" ,
and yields
.code x
for all
.codn x .
For three or more arguments, the operation is right-associative.
That is to say,
.code "(expt x y z)"
is equivalent to
.codn "(expt x (expt y z))" ,
similarly to the way nested exponents work in standard algebraic
notation.
Exponentiation is done pairwise using a binary operation.
If both operands to this binary operation are nonnegative integers, then the
result is an integer.
If the exponent is negative, and the base is zero, the situation is
treated as a division by zero: an exception of type
.code numeric-error
is thrown. Otherwise, a negative exponent is converted to floating-point,
if it already isn't, and a floating-point exponentiation is performed.
If either operand is a float, then the other
operand is converted to a float, and a floating point exponentiation
is performed. Exponentiation that would produce a complex number is
not supported.
The
.code sqrt
function produces a floating-point square root of
.metn arg ,
which is converted from integer to floating-point if necessary. Negative
operands are not supported.
The
.code isqrt
function computes the integer square root of
.metn arg ,
which must be an integer.
The integer square root is a value which is the
greatest integer that is no greater than the real square root of
.metn arg .
The input value must be an integer.
.coNP Function @ exptmod
.synb
.mets (exptmod < base < exponent << modulus )
.syne
.desc
The
.code exptmod
function performs modular exponentiation and accepts only integer
arguments. Furthermore,
.meta exponent
must be a nonnegative and
.meta modulus
must be positive.
The return value is
.meta base
raised to
.metn exponent ,
and reduced to the
least positive residue modulo
.metn modulus .
.coNP Function @ square
.synb
.mets (square << argument )
.syne
.desc
The
.code square
function returns the product of
.meta argument
with itself. The following
equivalence holds, except that
.code x
is evaluated only once in the
.code square
expression:
.verb
(square x) <--> (* x x)
.brev
.coNP Function @ cum-norm-dist
.synb
.mets (cum-norm-dist << argument )
.syne
.desc
The
.code cum-norm-dist
function calculates an approximation to the cumulative normal
distribution function: the integral, of the normal distribution function, from
negative infinity to the
.metn argument .
.coNP Function @ inv-cum-norm
.synb
.mets (inv-cum-norm << argument )
.syne
.desc
The
.code inv-cum-norm
function calculates an approximate to the inverse of the cumulative
normal distribution function. The argument, a value expected to lie
in the range [0, 1], represents the integral of the normal distribution
function from negative infinity to some domain point
.IR p .
The function calculates the approximate value of
.IR p .
The minimum value returned is -10, and the maximum value returned is 10,
regardless of how closely the argument approaches, respectively,
the 0 or 1 integral endpoints. For values less than zero, or exceeding 1, the
values returned, respectively, are -10 and 10.
.coNP Functions @ n-choose-k and @ n-perm-k
.synb
.mets (n-choose-k < n << k )
.mets (n-perm-k < n << k )
.syne
.desc
The
.code n-choose-k
function computes the binomial coefficient nCk which
expresses the number of combinations of
.meta k
items that can be chosen from
a set of
.metn n ,
where combinations are subsets.
The
.code n-perm-k
function computes nPk: the number of permutations of size
.meta k
that can be drawn from a set of
.metn n ,
where permutations are sequences,
whose order is significant.
The calculations only make sense when
.meta n
and
.meta k
are nonnegative integers, and
.meta k
does not exceed
.metn n .
The behavior is not specified if these conditions
are not met.
.coNP Functions @, fixnump @, bignump @, integerp @ floatp and @ numberp
.synb
.mets (fixnump << object )
.mets (bignump << object )
.mets (integerp << object )
.mets (floatp << object )
.mets (numberp << object )
.syne
.desc
These functions test the type of
.metn object ,
returning
.code t
if it is an object
of the implied type,
.code nil
otherwise. The
.codn fixnump ,
.code bignump
and
.code floatp
functions return
.code t
if the object is of the basic type
.codn fixnum ,
.code bignum
or
.codn float .
The function
.code integerp
returns true of
.meta object
is either a
.code fixnum
or
a
.codn bignum .
The function
.code numberp
returns
.code t
if
.meta object
is either
a
.codn fixnum ,
.code bignum
or
.codn float .
.coNP Functions @ zerop and @ nzerop
.synb
.mets (zerop << number )
.mets (nzerop << number )
.syne
.desc
The
.code zerop
function tests
.meta number
for equivalence to zero. The argument must be
a number or character. It returns
.code t
for the integer value
.code 0
and for the floating-point
value
.codn 0.0 .
For other numbers, it returns
.codn nil .
It returns
.code t
for the null character
.code #\enul
and
.code nil
for all other characters.
If
.meta number
is a range, then
.code zerop
returns
.code t
if both of the range endpoints individually satisfy
.codn zerop .
The
.code nzerop
function is the logical inverse of
.codn zerop :
it returns
.code t
for those arguments for which
.code zerop
returns
.code nil
and vice versa.
.coNP Functions @ plusp and @ minusp
.synb
.mets (plusp << number )
.mets (minusp << number )
.syne
.desc
These functions test whether a number is positive or negative,
returning
.code t
or
.codn nil ,
as the case may be.
The argument may also be a character. All characters other than
the null character
.code #\enul
are positive. No character is negative.
.coNP Functions @ evenp and @ oddp
.synb
.mets (evenp << integer )
.mets (oddp << integer )
.syne
.desc
The
.code evenp
and
.code oddp
functions require integer arguments.
.code evenp
returns
.code t
if
.meta integer
is even (divisible by two), otherwise it returns
.codn nil .
.code oddp
returns
.code t
if
.meta integer
is not divisible by two (odd), otherwise
it returns
.codn nil .
.coNP Functions @, succ @, ssucc @, sssucc @, pred @ ppred and @ pppred
.synb
.mets (succ << number )
.mets (ssucc << number )
.mets (sssucc << number )
.mets (pred << number )
.mets (ppred << number )
.mets (pppred << number )
.syne
.desc
The
.code succ
function adds 1 to its argument and returns the resulting value.
If the argument is an integer, then the return value is the successor
of that integer, and if it is a character, then the return value
is the successor of that character according to Unicode.
The
.code pred
function subtracts 1 from its argument, and under similar considerations
as above, the result represents the predecessor.
The
.code ssucc
and
.code sssucc
functions add 2 and 3, respectively. Similarly,
.code ppred
and
.code pppred
subtract 2 and 3 from their argument.
.coNP Functions @, > @, < @, >= @ <= and @ =
.synb
.mets (> < object << object *)
.mets (< < object << object *)
.mets (>= < object << object *)
.mets (<= < object << object *)
.mets (= < object << object *)
.syne
.desc
These relational functions compare characters, numbers, ranges and sequences of
characters or numbers for numeric equality or inequality. The arguments must be
one or more numbers, characters, ranges, or sequences of these objects,
or, recursively, of sequences.
If just one argument is given, then these functions all return
.codn t .
If two arguments are given then, they are compared as follows.
First, if the numbers do not have the same type, then the one
which has the lower ranking type is converted to the type of
the other, according to this ranking: character < integer < float.
For instance if a character and integer are compared, the character
is converted to its integer character code. Then a numeric comparison
is applied.
Three or more arguments may be given, in which case the comparison proceeds
pairwise from left to right. For instance in
.codn "(< a b c)" ,
the comparison
.code "(< a b)"
is performed in isolation. If the comparison is false, then
.code nil
is returned, otherwise
the comparison
.code "(< b c)"
is performed in isolation, and if that is false,
.code nil
is returned, otherwise
.code t
is returned. Note that it is possible for
.code b
to
undergo two different conversions. For instance in the
.mono
.meti (< < float < character << integer )
.onom
comparison,
.meta character
will first convert to a floating-point representation
of its Unicode value so that it can be compared to
.metn float ,
and if that comparison succeeds, then in the second comparison,
.meta character
will be converted to integer so that it can be compared to
.metn integer .
Ranges may only be compared with ranges. Corresponding
fields of ranges are compared for equality by
.code =
such that
.code "#R(0 1)"
and
.code "#R(0 1.0)"
are reported as equal.
The inequality comparisons are lexicographic, such that the
.code from
field of the range is considered more major than the
.code to
field. For example the inequalities
.code "(< #R(1 2) #R(2 0))"
and
.code "(< #R(1 2) #R(1 3))"
hold.
Sequences may only be compared with sequences, but
mixtures of any kinds of sequences may be compared:
lists with vectors, vectors with strings, and so on.
The
.code =
function considers a pair of sequences of unequal length
to be unequal, reporting
.codn nil .
Sequences are equal if they have the same length
and their corresponding elements are recursively
equal under the
.code =
function.
The inequality functions treat sequences lexicographically.
A pair of sequences is compared by comparing corresponding
elements. The
.code <
function tests each successive pair of corresponding
elements recursively using the
.code <
function. If this recursive comparison reports
.codn t ,
then the function immediately returns
.code t
without considering any more pairs of elements.
Otherwise the same pair of elements is compared again
using the
.code =
function. If that reports false, then the function reports false without
considering any more pairs of elements. Otherwise processing continues with the
next pair, if any. If all corresponding elements are equal, but the right
sequence is longer,
.code <
returns
.codn t ,
otherwise the function reports
.codn nil .
The
.code <=
function tests each successive pair of corresponding
elements recursively using the
.code <=
function. If this returns
.code nil
then the function returns
.code nil
without considering any more pairs. Otherwise processing continues
with the next pair, if any.
If all corresponding elements satisfy the test, but the
left sequence is longer, then
.code nil
is returned. Otherwise
.code t
is returned.
The inequality relations exhibit symmetry, which means that
the functions
.code >
and
.code >=
functions are equivalent, respectively, to
.code <
and
.code <=
with the order of the argument values reversed. For instance, the expression
.code "(< a b c)"
is equivalent to
.code "(> c b a)"
except for the difference in evaluation order of the
.codn a ,
.code b
and
.code c
operands themselves. Any semantic description of
.code <
or
.code <=
applies, respectively, also to
.code >
or
.code >=
with the appropriate adjustment for argument order reversal.
.coNP Function @ /=
.synb
.mets (/= << number *)
.syne
.desc
The arguments to
.code /=
may be numbers or characters. The
.code /=
function returns
.code t
if no two of its arguments are numerically equal. That is to say, if there
exist some
.code a
and
.code b
which are distinct arguments such that
.code "(= a b)"
is true, then
the function returns
.codn nil .
Otherwise it returns
.codn t .
.coNP Functions @ max and @ min
.synb
.mets (max < first-arg << arg *)
.mets (min < first-arg << arg *)
.syne
.desc
The
.code max
and
.code min
functions determine and return the highest or lowest
value from among their arguments.
If only
.meta first-arg
is given, that value is returned.
These functions are type generic, since they compare arguments
using the same semantics as the
.code less
function.
If two or more arguments are given, then
.code "(max a b)"
is equivalent to
.codn "(if (less a b) b a)" ,
and
.code "(min a b)"
is equivalent to
.codn "(if (less a b) a b)" .
If the operands do not
have the same type, then one of them is converted to the type of the other;
however, the original unconverted values are returned. For instance
.code "(max 4 3.0)"
yields the integer
.codn 4 ,
not
.codn 4.0 .
If three or more arguments are given,
.code max
and
.code min
reduce the arguments in a left-associative manner.
Thus
.code "(max a b c)"
means
.codn "(max (max a b) c)" .
.coNP Function @ clamp
.synb
.mets (clamp < low < high << val )
.syne
.desc
The
.code clamp
function clamps value
.meta val
into the range
.meta low
to
.metn high .
The
.code clamp
function returns
.meta low
if
.meta val
is less than
.metn low .
If
.meta val
is greater than or equal to
.metn low ,
but less than
.metn high ,
then it returns
.metn val .
Otherwise it returns
.metn high .
More precisely,
.code "(clamp a b c)"
is equivalent to
.codn "(max a (min b c))" .
.coNP Function @ bracket
.synb
.mets (bracket < value << level *)
.syne
.desc
The
.code bracket
function's arguments consist of one required
.meta value
followed by
.I n
.meta level
arguments.
The
.meta level
arguments are optional; in other words,
.I n
may be zero.
The
.code bracket
function calculates the
.I bracket
of the
.meta value
argument: a zero-based positional index of the value, in relation to the
.meta level
arguments.
Each of the
.meta level
arguments, of which there may be none, is associated with
an integer index, starting at zero, in left-to-right order. The
.meta level
arguments are examined in that order. When a
.meta level
argument is encountered which exceeds
.metn value ,
that
.meta level
argument's index is returned.
If
.meta value
exceeds all of the
.meta level
arguments, then
.I n
is returned.
Determining whether
.meta value
exceeds a
.meta level
is performed using the
.code less
function.
.TP* Examples:
.verb
(bracket 42) -> 0
(bracket 5 10) -> 0
(bracket 15 10) -> 1
(bracket 15 10 20) -> 1
(bracket 15 10 20 30) -> 1
(bracket 20 10 20 30) -> 2
(bracket 35 10 20 30) -> 3
(bracket "a" "aardvark" "zebra") -> 0
(bracket "ant" "aardvark" "zebra") -> 1
(bracket "zebu" "aardvark" "zebra") -> 2
.brev
.coNP Functions @, int-str @ flo-str and @ num-str
.synb
.mets (int-str < string <> [ radix ])
.mets (flo-str << string )
.mets (num-str << string )
.syne
.desc
These functions extract numeric values from character string
.metn string .
Leading whitespace in
.metn string ,
if any, is skipped. If no digits can be successfully extracted, then
.code nil
is returned. Trailing material which does not contribute to the number is
ignored.
The
.code int-str
function converts a string of digits in the specified
.meta radix
to an integer value. If
.meta radix
isn't specified, it defaults to 10.
Otherwise it must be an integer in the range 2 to 36, or else the character
.codn #\ec .
For radices above 10, letters of the alphabet
are used for digits:
.code A
represent a digit whose value is 10,
.code B
represents 11 and
so forth until
.codn Z .
Uppercase and lowercase letters are recognized.
Any character which is not a digit of the specified radix is regarded
as the start of trailing junk at which the extraction of the digits stops.
When
.meta radix
is specified as the character object
.codn #\ec ,
this indicates that a C-language-style integer constant should be
recognized. If, after any optional sign, the remainder of
.meta string
begins with the character pair
.code 0x
then that pair is considered removed from the string, and it is treated
as base 16 (hexadecimal). If, after any optional sign, the remainder of
.meta string
begins with a leading zero not followed by
.codn x ,
then the radix is taken to be 8 (octal). In scanning these formats,
.code int-str
function is not otherwise constrained by C language representational
limitations. Specifically, the input values are taken to be the printed
representation of arbitrary-precision integers and treated accordingly.
The
.code flo-str
function converts a floating-point decimal notation to a nearby
floating point value. The material which contributes to the value
is the longest match for optional leading space, followed by a
mantissa which consists of an optional sign followed by a mixture of at least
one digit, and at most one decimal point, optionally followed by an exponent
part denoted by the letter
.code E
or
.codn e ,
an optional sign and one or more optional exponent digits.
If the value specified by
.meta string
is out of range of the floating-point representation, then
.code nil
is returned.
The
.code num-str
function converts a decimal notation to either an integer as if by
a radix 10 application of
.codn int-str ,
or to a floating point value as if by
.codn flo-str .
The floating point interpretation is chosen if the possibly empty
initial sequence of digits (following any whitespace and optional sign) is
followed by a period, or by
.code e
or
.codn E .
.coNP Functions @ int-flo and @ flo-int
.synb
.mets (int-flo << float )
.mets (flo-int << integer )
.syne
.desc
These functions perform numeric conversion between integer and floating point
type. The
.code int-flo
function returns an integer by truncating toward zero.
The
.code flo-int
function returns an exact floating point value corresponding to
.metn integer ,
if possible, otherwise an approximation using a nearby
floating point value.
.coNP Functions @ tofloat and @ toint
.synb
.mets (tofloat << value )
.mets (toint < value <> [ radix ])
.syne
.desc
These functions convert
.meta value
to floating-point or integer, respectively. The
.meta value
can be of several types, including string.
If a floating-point value is passed into tofloat, or an integer value into
toint, then that value is simply returned.
If
.meta value
is a character, then it is treated as a string of length one
containing that character.
If
.meta value
is a string, then it is converted by
.code tofloat
as if by the function
.metn flo-str ,
and by
.code toint
as if by the function
.codn int-str .
If
.meta value
is an integer, then it is converted by
.code tofloat
as if by the function
.codn flo-int .
If
.meta value
is a floating-point number, then it is converted by
.code toint
as if by the function
.codn int-flo .
.coNP Variables @ fixnum-min and @ fixnum-max
.desc
These variables hold, respectively, the most negative value of the
.code fixnum
integer type, and its most positive value. Integer values
from
.code fixnum-min
to
.code fixnum-max
are all of type
.codn fixnum .
Integers outside of this range are
.code bignum
integers.
.coNP Functions @ tofloatz and @ tointz
.synb
.mets (tofloatz << value )
.mets (tointz < value <> [ radix ])
.syne
.desc
These functions are closely related to, respectively,
.code tofloat
and
.codn toint .
They differ in that these functions return a floating-point
or integer zero, respectively, in some situations
in which those functions would return
.code nil
or throw an error.
Whereas those functions reject a
.meta value
argument of
.codn nil ,
for that same argument
.code tofloatz
function returns 0.0 and
.code tointz
returns 0.
Likewise, in cases when
.code value
contains a string or character which cannot be
converted to a number, and
.code tofloat
and
.code toint
would return
.codn nil ,
these functions return 0.0 and 0, respectively.
In other situations, these functions behave
exactly like
.code tofloat
and
.codn toint .
.coNP Variables @, flo-min @ flo-max and @ flo-epsilon
.desc
These variables hold, respectively: the smallest positive floating-point
value; the largest positive floating-point value; and the difference
between 1.0 and the smallest representable value greater than 1.0.
.code flo-min
and
.code flo-max
define the floating-point range, which consists
of three regions: values from
.code "(- flo-max)"
to
.codn "(- flo-min)" ;
the value 0.0, and values from
.code flo-min
to
.codn flo-max .
.coNP Variable @ flo-dig
.desc
This variable holds an integer representing the number of decimal digits
in a decimal floating-point number such that this number can be converted
to a \*(TX floating-point number, and back to decimal, without a change in any of
the digits. This holds regardless of the value of the number, provided that it
does not exceed the floating-point range.
.coNP Variable @ flo-max-dig
.desc
This variable holds an integer representing the maximum number of
decimal digits required to capture the value of a floating-point number
such that the resulting decimal form will convert back to the same
floating-point number. See also the
.code *print-flo-precision*
variable.
.coNP Variables @ %pi% and @ %e%
.desc
These variables hold an approximation of the mathematical constants \(*p and e.
To four digits of precision, \(*p is 3.142 and e is 2.718. The
.code %pi%
and
.code %e%
approximations are accurate to
.code flo-dig
decimal digits.
.coNP Function @ digits
.synb
.mets (digits < number <> [ radix ])
.syne
.desc
The
.code digits
function returns a list of the digits of
.meta number
represented in the base given by
.metn radix .
The
.meta number
argument must be a nonnegative integer, and
.meta radix
must be an integer greater than one.
If
.meta radix
is omitted, it defaults to 10.
The return value is a list of the digits in descending order of significance:
most significant to least significant.
The digits are integers. For instance, if
.meta radix
is 42, then the digits are integer values in the range 0 to 41.
The returned list always contains at least one element, and
includes no leading zeros, except when
.meta number
is zero. In that case, a one-element list containing zero is returned.
.TP* Examples:
.verb
(digits 1234) -> (1 2 3 4)
(digits 1234567 1000) -> (1 234 567)
(digits 30 2) -> (1 1 1 1 0)
(digits 0) -> (0)
.brev
.coNP Function @ digpow
.synb
.mets (digpow < number <> [ radix ])
.syne
.desc
The
.code digpow
function decomposes the
.meta number
argument into a power series whose terms add up to
.metn number .
The
.meta number
argument must be a nonnegative integer, and
.meta radix
must be an integer greater than one.
The returned power series consists of a list of nonnegative
integers. It is formed from the digits of
.meta number
in the given
.metn radix ,
which serve as coefficients which multiply successive
powers of the
.metn radix ,
starting at the zeroth power (one).
The terms are given in decreasing order of significance:
the term corresponding to the most significant digit of
.metn number ,
multiplying the highest power of
.metn radix ,
is listed first.
The returned list always contains at least one element, and
includes no leading zeros, except when
.meta number
is zero. In that case, a one-element list containing zero is returned.
.verb
(digpow 1234) -> (1000 200 30 4)
(digpow 1234567 1000) -> (1000000 234000 567)
(digpow 30 2) -> (16 8 4 2 0)
(digpow 0) -> (0)
.brev
.coNP Functions @ poly and @ rpoly
.synb
.mets (poly < arg << coeffs )
.mets (rpoly < arg << coeffs )
.syne
.desc
The
.code poly
and
.code rpoly
functions evaluate a polynomial, for the given numeric argument value
.meta arg
and the coefficients given by
.metn coeffs ,
a sequence of numbers.
If
.meta coeffs
is an empty sequence, it denotes the zero polynomial, whose value
is zero everywhere; the functions return zero in this case.
Otherwise, the
.code poly
function considers
.meta coeffs
to hold the coefficients in the conventional order, namely in order
of decreasing degree of polynomial term. The first element of
.meta coeffs
is the leading coefficient, and the constant term appears as the last element.
The
.code rpoly
function takes the coefficients in opposite order: the first element of
.meta coeffs
gives the constant term coefficient, and the last element gives the
leading coefficient.
Note: except in the case of
.code rpoly
operating on a list or list-like sequence of coefficients,
Horner's method of evaluation is
used: a single result accumulator is initialized with zero, and then for each
successive coefficient, in order of decreasing term degree, the accumulator is
multiplied by the argument, and the coefficient is added. When
.code rpoly
operates on a list or list-like sequence, it makes a single
pass through the coefficients in order, thus taking them in increasing
term degree. It maintains two accumulators: one for successive powers of
.meta arg
and one for the resulting value. For each coefficient, the power
accumulator is updated by a multiplication by
.meta arg
and then this value is multiplied by the coefficient, and
that value is then added to the result accumulator.
.TP* Examples:
.verb
;; 2
;; evaluate x + 2x + 3 for x = 10.
(poly 10 '(1 2 3)) -> 123
;; 2
;; evaluate 3x + 2x + 1 for x = 10.
(rpoly 10 '(1 2 3)) -> 321
.brev
.coNP Function @ bignum-len
.synb
.mets (bignum-len << arg )
.syne
.desc
The
.code bignum-len
function reports the machine-specific
.I "bignum order"
of the integer or character argument
.metn arg .
If
.meta arg
is a character or
.code fixnum
integer, the function returns zero.
Otherwise
.meta arg
is expected to be a
.code bignum
integer, and the function returns the number of "limbs" used for its
representation, a positive integer.
Note: the
.code bignum-len
function is intended to be of use in algorithms whose performance
benefits from ordering the operations on multiple integer operands
according to the magnitudes of those operands. The function provides an
estimate of magnitude which trades accuracy for efficiency.
.coNP Function @ quantile
.synb
.mets (quantile < p >> [ group-size <> [ rate ]])
.syne
.desc
The
.code quantile
function returns a function which estimates a specific quantile
of a set of real-valued samples. The desired quantile is indicated
by the
.meta p
parameter, which is a number in the range 0 to 1.0. If
.meta p
is specified as 0.5, then the median is estimated.
The
.meta p
value of 0.9 leads to the estimation of the 90th percentile:
a value such that approximately 90% of the samples are below that value.
If the
.meta group-size
parameter is specified, it must be a positive integer.
The returned function then operates in grouped mode. The
.meta rate
parameter is relevant only to grouped mode. Grouped mode is
described below.
The function returned by
.code quantile
maintains internal state in relation to calculating the quantile.
The function may be called with any number of arguments, including
none. It expects every argument to be either a number, or a sequence
of numbers. These numbers are accumulated into the quantile calculation,
and a revised estimate of the quantile is then returned.
Note: the algorithm used is the P-Squared algorithm invented in 1985 by Raj
Jain and Imrich Chlamtac, which avoids accumulating and sorting the entire data
set, while still obtaining good quality estimates of the quantile.
The algorithm requires an initial seed of five samples. Then additional
samples input into the algorithm produce quantile estimates. To eliminate this
special case from the abstract interface, the \*(TX implementation is capable of
producing an estimate when five or fewer samples have been presented, including
none. In this low situation, the
.meta p
value is ignored in reporting the estimate. When no samples have been given,
the estimate is zero. When one sample has been given, the estimate is that
sample itself. When between two and five samples have been given, the estimate
is their median. Using the median as the estimate ensures a smooth transition
from these early estimates into the estimates produced by the P-Squared
algorithm. This is because the P-Squared algorithm always reports the value of
the middle height accumulator as the estimate, and that accumulator's initial
value is the median of the first five samples.
The function returned by
.codn quantile ,
though not accumulating all of the samples passed to it, nevertheless has
a limited sample capacity, because the registers it uses for tracking the
sample group positions are fixed-width integers. The sample capacity is
approximately 4 times the value of
.codn fixnum-max .
.TP* Example:
.verb
(defparm q (quantile 0.9)) ;; create 90-th percentile accumulator
[q] -> 0.0 ;; no samples given: estimate is 0.
[q 3.14] -> 3.14 ;; one sample: estimate is that sample
[q 13.3 7.9 5.2 6.3] -> 7.9 ;; five samples: estimate is median.
[q 6.8 7.3 9.1 4.0] ;; more than five samples; estimate now
-> 8.44651234567901 ;; from P-Square algorithm
[q #(13.1 5 2.5)] ;; vector argument
-> 9.68660493827161
[q] -> 9.68660493827161 ;; no arguments: repeat current estimate
.brev
If the
.meta group-size
argument is specified, then the quantile accumulator operates in grouped mode.
Grouped mode allows infinite sample operation without overflow: an unlimited
number of samples can be accepted. However, old samples lose their influence
over the estimated value: newer samples are considered more significant than
old samples.
In grouped mode, the quantile accumulator is reset to its initial state whenever
.meta group-size
samples have been accumulated, and begins freshly calculating the quantile.
Prior to the reset, an estimate is obtained and retained in an internal
register. Going forward, this remembered previous estimate is blended in with
the newly calculated estimate values, as described below. The cycle repeats
itself whenever
.meta group-size
samples accumulate: the state is reset, and the current estimate is loaded into
the previous estimate register, which is then blended with newly computed
values.
The
.meta rate
parameter, whose default value is 0.999, controls the estimate blending.
It should be a value between 0 and 1.
Upon each reset, a blend value register is initialized to 1.0. Each time
a new sample is accumulated, the blend register is multiplied by the rate
parameter, and the product is stored back into the blend register.
Thus if the rate is between 0 and 1, exclusive, then the blend register
exponentially decreases as the number of samples grows. The blend register
indicates the fraction of the estimate which comes from the remembered previous
estimate.
For instance, if the current blend value is 0.8, then the returned estimate
value is 0.8 times the remembered previous estimate, plus 0.2 times the newly
computed estimate for the current sample in the new group: the previous and
current estimate are blended 80:20.
The default
.meta rate
value of 0.999 is chosen for a slow transition to the new estimates, which
helps to conceal inaccuracies in the algorithm associated with having
accumulated a small number of samples. At this rate, it requires about 290
samples before the blend value drops to 75% of the old estimate.
If
.code rate
is specified as 0, then no blending of the previous estimate value
takes place, since the blend factor will drop to zero upon the first
sample being received after the group reset, causing the newly calculated
estimates to be returned without blending. The previous sample groups
therefore have no influence over newer estimates. If
.code rate
is specified as 1, then the blend factor will stay at 1, and so the
estimate will forever remain at the previous value, ignoring the
calculations driven by the new samples.
Note: it is recommended that if
.meta group-size
is specified, the value should be at least several hundred. Too small
a group size will prevent the estimation algorithm from settling on
good results. The
.meta rate
parameter should not much smaller than 1. A rate too low will cause
the previous estimate's contribution to the quantile value to diminish,
too quickly, before the new estimation settles.
.coNP Variables @, flo-near @, flo-down @ flo-up and @ flo-zero
.desc
These variables hold integer values suitable as arguments to the
.code flo-set-round-mode
function, which controls the rounding mode for the results of floating-point
operations. These variables are only defined on platforms which support
rounding control.
Their values have the following meanings:
.RS
.coIP flo-near
Round to nearest: the result of an operation is rounded to the nearest
representable value.
.coIP flo-down
Round down: the result of an operation is rounded to the nearest representable
value that lies in the direction of negative infinity.
.coIP flo-up
Round up: the result of an operation is rounded to the nearest representable
value that lies in the direction of positive infinity.
.coIP flo-zero
Round to zero: the result of an operation is rounded to the nearest
representable value that lies in the direction of zero.
.RE
.IP
.coNP Functions @ flo-get-round-mode and @ flo-set-round-mode
.synb
.mets (flo-get-round-mode)
.mets (flo-set-round-mode << mode )
.syne
.desc
Sometimes floating-point operations produce a result which
requires more bits of precision than the floating point representation
can provide. A representable floating-point value must be substituted
for the true result and yielded by the operation.
On platforms which support rounding control, these functions are provided for
selecting the decision procedure by which the floating-point representation
is taken.
The
.code flo-get-round-mode
returns the current rounding mode. The rounding mode is represented by
an integer value which is either equal to one of the four variables
.codn flo-near ,
.codn flo-down ,
.code flo-up
and
.codn flo-zero ,
or else some other value specific to the host environment. Initially,
the value is that of
.codn flo-near .
Otherwise, the value returned is that which was stored by the most
recent successful call to
.codn flo-set-round-mode .
The
.code flo-set-round-mode
function changes the rounding mode. The argument to its
.meta mode
parameter may be the value of one of the above four variables,
or else some other value supported by the host environment's
.code fesetround
C library function.
The
.code flo-set-round-mode
function returns
.code t
if it is successful, otherwise the return value is
.code nil
and the rounding mode is not changed.
If a value is passed to
.code flo-set-round-mode
which is not the value of one of the above
four rounding mode variables, and the function succeeds anyway, then the
rounding behavior of floating-point operations depends on the host
environment's interpretation of that value.
.SS* Bit Operations
In \*(TL, similarly to Common Lisp, bit operations on integers are based
on a concept that might be called "infinite two's complement".
Under infinite two's complement, a positive number is regarded as having
a binary representation prefixed by an infinite stream of zero digits (for
example
.code 1
is
.codn ...00001 ).
A negative number
in infinite two's complement is the bitwise negation of its positive counterpart,
plus one: it carries an infinite prefix of 1 digits. So for instance the number
.code -1
is represented by
.codn ...11111111 :
an infinite sequence of
1
bits. There
is no specific sign bit; any operation which produces such an infinite sequence
of 1 digits on the left gives rise to a negative number. For instance, consider the
operation of computing the bitwise complement of the number
.codn 1 .
Since the
number
.code 1
is represented as
.codn ...0000001 ,
its complement is
.codn ...11111110 .
Each one of the
.code 0
digits in the infinite sequence is replaced by
.codn 1 ,
And this leading sequence means that the number
is negative, in fact corresponding to the two's complement representation of
the value
.codn -2 .
Hence, the infinite digit concept corresponds to an arithmetic
interpretation.
In fact \*(TL's bignum integers do not use a two's complement
representation internally. Numbers are represented as an array which holds a
pure binary number. A separate field indicates the sign: negative,
or nonnegative. That negative numbers appear as two's complement under the
bit operations is merely a carefully maintained illusion (which makes bit
operations on negative numbers more expensive).
The
.code logtrunc
function, as well as a feature of the
.code lognot
function, allow bit
manipulation code to be written which works with positive numbers only, even if
complements are required. The trade off is that the application has to manage a
limit on the number of bits.
.coNP Functions @, logand @ logior and @ logxor
.synb
.mets (logand << integer *)
.mets (logior << integer *)
.mets (logxor < int1 << int2 )
.syne
.desc
These operations perform the familiar bitwise and, inclusive or, and exclusive
or operations, respectively. Positive values inputs are treated as
pure binary numbers. Negative inputs are treated as infinite-bit
two's complement.
For example
.code "(logand -2 7)"
produces
.codn 6 .
This is because
.code -2
is
.code ...111110
in infinite-bit two's complement. And-ing this value with
.code 7
(or
.codn ...000111 )
produces
.codn 110 .
The
.code logand
and
.code logior
functions are variadic, and may be called with zero, one,
two, or more input values. If
.code logand
is called with no arguments, it produces
the value -1 (all bits 1). If
.code logior
is called with no arguments it produces
zero. In the one-argument case, the functions just return their argument value.
In the two-argument case, one of the operands may be a character, if the other
operand is a fixnum integer. The character operand is taken to be an integer
corresponding to the character value's Unicode code point value. The resulting
value is regarded as a Unicode code point and converted to a character value
accordingly.
When three or more arguments are specified, the operation's semantics is
that of a left-associative reduction through two-argument invocations,
so that the three-argument case
.code "(logand a b c)"
is equivalent to the expression
.codn "(logand (logand a b) c)" ,
which features two two-argument cases.
.coNP Function @ logtest
.synb
.mets (logtest < int1 << int2 )
.syne
.desc
The
.code logtest
function returns true if
.meta int1
and
.meta int2
have bits in
common. The following equivalence holds:
.verb
(logtest a b) <--> (not (zerop (logand a b)))
.brev
.coNP Functions @ lognot and @ logtrunc
.synb
.mets (lognot < value <> [ bits ])
.mets (logtrunc < value << bits )
.syne
.desc
The
.code lognot
function performs a bitwise complement of
.metn value .
When the one-argument form of lognot is used, then if
.meta value
is nonnegative,
then the result is negative, and vice versa,
according to the infinite-bit
two's complement representation. For instance
.code "(lognot -2)"
is
.codn 1 ,
and
.code "(lognot 1)"
is
.codn -2 .
The two-argument form of
.code lognot
produces a truncated complement. Conceptually,
a bitwise complement is first calculated, and then the resulting number is
truncated to the number of bits given by
.metn bits ,
which must be a nonnegative integer. The following equivalence holds:
.verb
(lognot a b) <--> (logtrunc (lognot a) b)
.brev
The
.code logtrunc
function truncates the integer
.meta value
to the specified number
of bits. If
.meta value
is negative, then the two's complement representation
is truncated. The return value of
.code logtrunc
is always a nonnegative integer.
.coNP Function @ sign-extend
.synb
.mets (sign-extend < value << bits )
.syne
.desc
The
.code sign-extend
function first truncates the infinite-bit two's complement representation of
the integer
.meta value
to the specified number of bits, similarly to the
.code logtrunc
function. Then, this truncated value is regarded as a
.metn bits -wide
two's complement integer. The value of this integer is
calculated and returned.
.TP* Examples:
.verb
(sign-extend 127 8) -> 127
(sign-extend 128 8) -> -128
(sign-extend 129 8) -> -127
(sign-extend 255 8) -> -1
(sign-extend 256 8) -> 0
(sign-extend -1 8) -> -1
(sign-extend -255 8) -> 0
.brev
.coNP Function @ ash
.synb
.mets (ash < value << bits )
.syne
.desc
The
.code ash
function shifts
.meta value
by the specified number of
.meta bits
producing a
new value. If
.meta bits
is positive, then a left shift takes place. If
.meta bits
is negative, then a right shift takes place. If
.meta bits
is zero, then
.meta value
is returned unaltered. For positive numbers, a left shift by n bits is
equivalent to a multiplication by two to the power of n, or
.codn "(expt 2 n)" .
A right shift by n bits of a positive integer is equivalent to integer
division by
.codn "(expt 2 n)" ,
with truncation toward zero.
For negative numbers, the bit shift is performed as if on the two's complement
representation. Under the infinite two's complement representation,
a right shift does not exhaust the infinite sequence of
.code 1
digits which
extends to the left. Thus if
.code -4
is shifted right it becomes
.code -2
because
the bitwise representations of these values are
.code ...111100
and
.codn ...11110 .
.coNP Function @ bit
.synb
.mets (bit < value << bit )
.syne
.desc
The
.code bit
function tests whether the integer or character
.meta value
has a 1 in bit position
.metn bit .
The
.meta bit
argument must be a nonnegative integer. A value of
.meta bit
of zero indicates the least-significant-bit position of
.metn value .
The
.code bit
function has a Boolean result, returning the symbol
.code t
if bit
.meta bit
of
.meta value
is set, otherwise
.codn nil .
If
.meta value
is negative, it is treated as if it had an infinite-bit
two's complement representation. For instance, if
.meta value
is
.codn -2 ,
then the
.code bit
function returns
.code nil
for a
.meta bit
value of zero, and
.code t
for all other values,
since the infinite bit two's complement representation of
.code -2
is
.codn ...11110 .
.coNP Function @ mask
.synb
.mets (mask << integer *)
.syne
.desc
The
.code mask
function takes zero or more integer arguments, and produces an integer
value which corresponds to a bitmask made up of the bit positions specified by
the integer arguments.
If
.code mask
is called with no arguments, then the return value is zero.
If
.code mask
is called with a single argument
.meta integer
then the return value is the same as
that of the expression
.codn "(ash 1 <integer>)" :
the value 1 shifted left by
.meta integer
bit positions. If
.meta integer
is zero, then the result is
.codn 1 ;
if
.meta integer
is
.codn 1 ,
the
result is
.code 2
and so forth. If
.meta value
is negative, then the result is zero.
If
.code mask
is called with two or more arguments, then the result is a bitwise or of
the masks individually computed for each of the values.
In other words, the following equivalences hold:
.verb
(mask) <--> 0
(mask a) <--> (ash 1 a)
(mask a b c ...) <--> (logior (mask a) (mask b) (mask c) ...)
.brev
.coNP Function @ bitset
.synb
.mets (bitset << integer )
.syne
.desc
The
.code bitset
function returns a list of the positions of bits which have a value of
1 in a positive
.meta integer
argument, or the positions of bits which have a value of zero in a negative
.meta integer
argument. The positions are ordered from least to greatest. The least
significant bit has position zero. If
.meta integer
is zero, the empty list
.code nil
is returned.
A negative integer is treated as an infinite-bit two's complement
representation.
The argument may be a character.
If
.meta integer
.code x
is nonnegative, the following equivalence holds:
.verb
x <--> [apply mask (bitset x)]
.brev
That is to say, the value of
.code x
may be reconstituted by applying the bit positions returned by
.code bitset
as arguments to the
.code mask
function.
The value of a negative
.code x
may be reconstituted from its
.code bitset
as follows:
.verb
x <--> (pred (- [apply mask (bitset x)]))
.brev
also, more trivially, thus:
.verb
x <--> (- [apply mask (bitset (- x))])
.brev
.coNP Function @ width
.synb
.mets (width << integer *)
.syne
.desc
A two's complement representation of an integer consists of a sign bit and a
mantissa field.
The
.code width
function computes the minimum number of bits required for the mantissa portion
of the two's complement representation of the
.meta integer
argument.
For a nonnegative argument, the width also corresponds to the number of bits
required for a natural binary representation of that value.
Two integer values have a width of zero, namely 0 and -1. This means that these
two values can be represented in a one-bit two's complement, consisting of only
a sign bit: the one-bit two's complement bitfield 1 denotes -1, and 0 denotes
0.
Similarly, two integer values have a width of 1: 1 and -2. The two-bit
two's complement bitfield 01 denotes 1, and 10 denotes -2.
The argument may be a character.
.coNP Function @ logcount
.synb
.mets (logcount << integer )
.syne
.desc
The
.code logcount
function considers
.meta integer
to have a two's complement representation. If the integer is positive,
it returns the count of bits in that representation whose value is 1.
If
.meta integer
is negative, it returns the count of zero bits instead. If
.meta integer
is zero, the value returned is zero.
The argument may be a character.
.coNP Macros @ set-mask and @ clear-mask
.synb
.mets (set-mask < place << integer *)
.mets (clear-mask < place << integer *)
.syne
.desc
The
.code set-mask
and
.code clear-mask
macros set to 1 and 0, respectively, the bits in
.meta place
corresponding to bits that are equal to 1 in the mask resulting from
applying the inclusive or operation to the
.meta integer
arguments.
The following equivalences hold:
.verb
(set-mask place integer ...)
<--> (set place (logior place integer ...)
(clear-mask place integer ...)
<--> (set place (logand place (lognot (logior integer ...))))
.brev
.SS* User-Defined Arithmetic Types
\*(TL makes it possible for the user application program to define structure
types which can participate in arithmetic operations as if they were numbers.
Under most arithmetic functions, a structure object may be used instead of a
number, if that structure object implements a specific method which is required
by that arithmetic function.
The following paragraphs give general remarks about the method conventions.
Not all arithmetic and bit manipulation functions have a corresponding
method, and a small number of functions do not follow these conventions.
In the simplest case of arithmetic functions which are unary, the method
takes no argument other than the object itself. Most unary arithmetic functions
expect a structure argument to have a method which has the same name as that
function. For instance, if
.code x
is a structure, then
.code "(cos x)"
will invoke
.codn "x.(cos)" .
If
.code x
has no
.code cos
method, then an
.code error
exception is thrown. A few unary methods are not named after the corresponding function.
The unary case of the
.code -
function expects an object to have a method named
.codn neg ;
thus,
.code "(- x)"
invokes
.codn "x.(neg)" .
Unary division requires a method called
.codn recip ;
thus,
.codn "(/ x)" ,
invokes
.codn "x.(recip)" .
When a structure object is used as an argument in a two-argument (binary)
arithmetic function, there are several cases to consider. If the left argument
to a binary function is an object, then that object is expected to support a
binary method. That method is called with two arguments: the object itself, of
course, and the right argument of the arithmetic operation. In this case, the
method is named after the function. For instance, if
.code x
is an object, then
.code "(+ x 3)"
invokes
.codn "x.(+ 3)" .
If the right argument, and only the right argument, of a binary operation is an
object, then the situation falls into two cases depending on whether the operation
is commutative. If the operation is commutative, then the same method is used
as in the case when the object is the left argument. The arguments are merely reversed.
Thus
.code "(+ 3 x)"
also invokes
.codn "x.(+ 3)" .
If the operation is not commutative, then the object must supply an alternative
method. For most functions, that method is named by a symbol whose name begins
with a
.code r-
prefix. For instance
.code "(mod x 5)"
invokes
.code "x.(mod 5)"
whereas
.code "(mod 5 x)"
invokes
.codn "x.(r-mod 5)" .
Note: the "r" may be remembered as indicating that the object is the
.B right
argument
of the binary operation or that the arguments are
.BR reversed .
Two functions do not follow the
.code r-
convention. These are
.code -
and
.codn / .
For these, the methods used for the object as a right argument, respectively, are
.code --
and
.codn // .
Thus
.code "(/ 5 x)"
invokes
.code "x.(// 5)"
and
.code "(- 5 x)"
invokes
.codn "x.(-- 5)" .
Several binary functions do not support an object as the right argument. These are
.codn sign-extend ,
.code ash
and
.codn bit .
Variadic arithmetic functions, when given three or more arguments, are regarded
as performing a left-associative decimation of the arguments through a binary
function. Thus for instance
.code "(- 1 x 4)"
is understood as
.code "(- (- 1 x) 4)"
where
.code "x.(-- 1)"
is evaluated first. If that method yields an object
.code o
then
.code "o.(- 4)"
is invoked.
Certain variadic arithmetic functions, if invoked with one argument, just
return that argument: for instance,
.code +
and
.code *
are in this category. A special concession exists in these functions: if
their one and only argument is a structure, then that structure is returned
without any error checking, even if it implements no methods related
to arithmetic.
The following sections describe each of the methods that must be implemented
by an object for the associated arithmetic function to work with that object,
either at all, or in a specific argument position, as the case may be.
These methods are not provided by \*(TL; the application is required to provide
them.
.de bmc
. coNP Method @ \\$1
. synb
. mets << obj .(\\$1 << arg )
. syne
. desc
The
. code \\$1
method is invoked when a structure is used as an argument to the
. code \\$1
function.
If an object
. meta obj
is combined with an argument
. metn arg ,
either as
. mono
. meti (\\$1 < obj << arg )
. onom
or as
. mono
. meti (\\$1 < arg << obj )
. onom
then, effectively, the method call
. mono
. meti << obj .(\\$1 << arg )
. onom
takes place, and its return value is taken as the result
of the operation.
..
.de bmcv
. coNP Method @ \\$1
. synb
. mets << obj .(\\$1 << arg )
. syne
. desc
The
. code \\$1
method is invoked when a structure is used as an argument to the
. code \\$1
function together with at least one other operand.
If an object
. meta obj
is combined with an argument
. metn arg ,
either as
. mono
. meti (\\$1 < obj << arg )
. onom
or as
. mono
. meti (\\$1 < arg << obj )
. onom
then, effectively, the method call
. mono
. meti << obj .(\\$1 << arg )
. onom
takes place, and its return value is taken as the result
of the operation.
..
.de bmnl
. coNP Method @ \\$1
. synb
. mets << obj .(\\$1 << arg )
. syne
. desc
The
. code \\$1
method is invoked when the structure
. meta obj
is used as the left argument of the
. code \\$1
function.
If an object
. meta obj
is combined with an argument
. metn arg ,
as
. mono
. meti (\\$1 < obj << arg )
. onom
then, effectively, the method call
. mono
. meti << obj .(\\$1 << arg )
. onom
takes place, and its return value is taken as the result
of the operation.
..
.de bmnr
. coNP Method @ \\$1
. synb
. mets << obj .(\\$1 << arg )
. syne
. desc
The
. code \\$1
method is invoked when the structure
. meta obj
is used as the right argument of the
. code \\$2
function.
If an object
. meta obj
is combined with an argument
. metn arg ,
as
. mono
. meti (\\$2 < arg << obj )
. onom
then, effectively, the method call
. mono
. meti << obj .(\\$1 << arg )
. onom
takes place, and its return value is taken as the result
of the operation.
..
.de umv
. coNP Method @ \\$1
. synb
. mets << obj .(\\$1)
. syne
. desc
The
. code \\$1
method is invoked when the structure
. meta obj
is used as the sole argument to the
. code \\$2
function.
If an object
. meta obj
is passed to the function as
. mono
. meti (\\$2 << obj )
. onom
then, effectively, the method call
. mono
. meti << obj .(\\$1)
. onom
takes place, and its return value is taken as the result
of the operation.
..
.de bma
. coNP Method @ \\$1
. synb
. mets << obj .(\\$1 << arg )
. syne
. desc
The
. code \\$1
method is invoked when the
. code \\$1
function is invoked with two operands, and the structure
. meta obj
is the left operand.
The method is also invoked when the
. code \\$2
function is invoked with two operands, and
.meta obj
is the right operand.
If an object
. meta obj
is combined with an argument
. metn arg ,
either as
. mono
. meti (\\$1 < obj << arg )
. onom
or as
. mono
. meti (\\$2 < arg << obj )
. onom
then, effectively, the method call
. mono
. meti << obj .(\\$1 << arg )
. onom
takes place, and its return value is taken as the result
of the operation.
..
.de um
. coNP Method @ \\$1
. synb
. mets << obj .(\\$1)
. syne
. desc
The
. code \\$1
method is invoked when a structure is used as the argument to the
. code \\$1
function.
If an object
. meta obj
is passed to the function as
. mono
. meti (\\$1 << obj )
. onom
then, effectively, the method call
. mono
. meti << obj .(\\$1)
. onom
takes place, and its return value is taken as the result
of the operation.
..
.de tmnl
. coNP Method @ \\$1
. synb
. mets << obj .(\\$1 < arg1 << arg2 )
. syne
. desc
The
. code \\$1
method is invoked when the structure
. meta obj
is used as the left argument of the
. code \\$1
function.
If an object
. meta obj
is combined with arguments
. meta arg1
and
. metn arg2 ,
as
. mono
. meti (\\$1 < obj < arg1 << arg2 )
. onom
then, effectively, the method call
. mono
. meti << obj .(\\$1 < arg1 << arg2 )
. onom
takes place, and its return value is taken as the result
of the operation.
..
.bmcv +
.bmnl -
.bmnr -- -
.umv neg -
.bmcv *
.bmnl /
.bmnr // /
.umv recip /
.um abs
.um signum
.bmnl trunc
.bmnr r-trunc trunc
.umv trunc1 trunc
.bmnl mod
.bmnr r-mod mod
.bmnl expt
.bmnr r-expt expt
.tmnl exptmod
Note: the
.code exptmod
function doesn't support structure objects in the second and
third argument positions. The
.meta exponent
and
.meta base
arguments must be integers.
.um isqrt
.um square
.bma > <
.bma < >
.bma >= <=
.bma <= >=
.bmc =
.um zerop
.um plusp
.um minusp
.um evenp
.um oddp
.bmnl floor
.bmnr r-floor floor
.umv floor1 floor
.bmnl ceil
.bmnr r-ceil ceil
.umv ceil1 ceil
.bmnl round
.bmnr r-round round
.umv round1 round
.um sin
.um cos
.um tan
.um asin
.um acos
.um atan
.bmnl atan2
.bmnr r-atan2 atan2
.um sinh
.um cosh
.um tanh
.um asinh
.um acosh
.um atanh
.um log
.um log2
.um log10
.um exp
.um sqrt
.bmcv logand
.bmcv logior
.bmnl lognot
.bmnr r-lognot lognot
.umv lognot1 lognot
.bmnl logtrunc
.bmnr r-logtrunc logtrunc
.bmnl sign-extend
Note: the
.code sign-extend
function doesn't support a structure as the right argument,
.metn bits ,
which must be an integer.
.bmnl ash
Note: the
.code ash
function doesn't support a structure as the right argument,
.metn bits ,
which must be an integer.
.bmnl bit
Note: the
.code bit
function doesn't support a structure as the right argument,
.metn bit ,
which must be an integer.
.um width
.um logcount
.um bitset
.SS* Exception Handling
An
.I exception
in \*(TX is a special event in the execution of the program which
potentially results in a transfer of control. An exception is identified by a
symbol, known as the
.IR "exception type" ,
and it carries zero or more arguments, called the
.IR "exception arguments" .
When an exception is initiated, it is said to be
.IR thrown .
This action is initiated by the following functions:
.codn throw ,
.code throwf
and
.codn error ,
and possibly other functions which invoke these.
When an exception is thrown, \*(TX enters into exception processing
mode. Exception processing mode terminates in one of several ways:
.IP -
A
.I catch
is found which matches the exception, and control is transferred
to the catch by a nonlocal transfer which performs unwinding. Catches are
defined by the
.code catch
macro.
.IP -
A
.I handler
is found which matches the exception, and control is transferred to
the handler by invoking its function. The handler function accepts the
exception by performing a nonlocal transfer to a destination of its choice, or
else declines to accept the exception by returning.
Handlers are defined by the
.code handler-bind
operator or
.code handle
macro.
.IP -
If no catch or accepting handler is found for an exception derived from
.code error
and
.code *unhandled-hook*
is
.codn nil ,
then a built-in strategy for handling the exception is invoked,
consisting of unwinding, and then printing some informational messages and
terminating.
If the
.code *unhandled-hook*
variable contains a value that isn't
.codn nil ,
then control is transferred to the function stored in the
that variable first; only if that function returns is the above
built-in strategy invoked.
.IP -
If no catch or accepting handler is found for an exception derived from
.codn warning ,
then a warning diagnostic is issued on the
.code *stderr*
stream and a
.code continue
exception is thrown with no arguments. If no catch or handler is found
for that exception, then control returns normally to the site which
threw the warning exception.
.IP -
If no catch or accepting handler is found for an exception that is
neither derived from
.code error
nor from
.codn warning ,
then no control transfer takes place; control returns to the
.code throw
or
.code throwf
function which returns normally, with a return value of
.codn nil .
.PP
.NP* Catches and Handlers
There are two ways by which exceptions are handled: catches and handlers.
Catches and handlers are similar, but different.
A catch is an exit point associated with an active scope. When an exception is
handled by a catch, the form which threw the exception is abandoned, and unwinding
takes place to the catch site, which receives the exception type and arguments.
A handler is also associated with an active scope. However, it is a function,
and not a dynamic exit point. When an exception is passed to handler,
unwinding does not take place; rather, the function is called. The function then
either completes the exception handling by performing a nonlocal transfer,
or else declines the exception by performing an ordinary return.
Catches and handlers are identified by exception type symbols. A catch or
handler is eligible to process an exception if it handles a type which is
a supertype of the exception which is being processed. Handles and catches
are found by means of a combined search which proceeds from the innermost
nesting of dynamic scope to the outermost, without performing any unwinding.
When an eligible handler is encountered, its registered function is called, thereby suspending the
search. If the handler function returns, the search continues from that scope
to yet unvisited outer scopes. When an eligible catch is encountered rather
than a handler, the search terminates and a control transfer takes place to the
catch site. That control transfer then performs unwinding, which requires it to
make a second pass through the same nestings of dynamic scope that had just
been traversed in order to find that catch.
.NP* Handlers and Sandboxing
Because handlers execute in the dynamic context of the exception origin,
without any unwinding having taken place, they expose a potential route
of sandbox escape via the package system, unless special steps are taken.
The threat is that code at the handler site could take advantage of
the current value of the
.code *package*
and
.code *package-alist*
variables established at the exception throw site to gain inappropriate access
to symbols.
For this reason, when a handler is established, the current values of
.code *package*
and
.code *package-alist*
are recorded into the handler frame.
When that handler is later invoked, it executes in a dynamic environment
in which those variables are bound to the previously noted values.
The catch mechanism doesn't do any such thing because the unwinding
which is performed prior to the invocation of a catch implicitly
restores the values of
.B all
special variables to the values they had at the time the frame was
established.
.NP* Exception Type Hierarchy
Exception type symbols are arranged
in an inheritance hierarchy, at whose top the symbol
.code t
is the supertype of every exception type, and the
.code nil
symbol is at the bottom, the subtype of every exception type.
Keyword symbols may be used as exception types.
Every symbol is its own supertype and subtype. Thus whenever X is known to be a
subtype of Y, it is possible that X is exactly Y.
The
.code defex
macro registers exception supertype/subtype relationships among symbols.
The following tree diagram shows the relationships among \*(TL's built-in
exception symbols. Not shown is the exception symbol
.codn nil ,
subtype of every exception type:
.verb
t ----+--- warning
|
+--- restart ---+--- continue
| |
| +--- retry
| |
| +--- skip
|
+--- error ---+--- type-error
|
+--- internal-error
|
+--- panic
|
+--- numeric-error
|
+--- range-error
|
+--- query-error
|
+--- file-error -------+--- path-not-found
| |
| +--- path-exists
| |
| +--- path-permission
|
+--- process-error
|
+--- socket-error
|
+--- system-error
|
+--- alloc-error
|
+--- timeout-error
|
+--- assert
|
+--- syntax-error
|
+--- eval-error
|
+--- match-error
|
+--- case-error
|
+--- opt-error
.brev
Program designers are encouraged to derive new error exceptions from the
.code error
type. The
.code restart
type is intended to be the root of a hierarchy of exception
types used for denoting restart points: designers are encouraged
to derive restarts from this type.
A catch for the
.code continue
exception should be established around constructs which can throw an
error from which it is possible to recover. That exception provides
the entry point into the recovery which resumes execution.
A catch for
.code retry
should be provided in situations when it is possible and makes sense for a failed
operation to be tried again.
A catch for
.code skip
should be provided in situations when it is possible and sensible to continue
with subsequent operations even though an operation has failed.
.NP* Dialect Notes
Exception handling in \*(TL provides capabilities similar to the condition
system in ANSI Common Lisp. The implementation and terminology differ.
Most obviously, ANSI CL uses the "condition" term, whereas \*(TL uses "exception".
In ANSI CL, a condition is "raised", whereas a \*(TL exception is "thrown".
In ANSI CL, when a condition is raised, a condition object is created. Condition
object are similar to class objects, but are not required to be in the Common Lisp
Object System. They are related by inheritance and can have properties. \*(TL
exceptions are unencapsulated: they consist of a symbol, plus zero or more
arguments. The symbols are related by inheritance.
When a condition is raised in ANSI CL, the dynamic scope is searched for a
handler, which is an ordinary function which receives the condition. No
unwinding or nonlocal transfer takes place. The handler can return, in which
case the search continues. Matching the condition to the handler is by
inheritance. Handler functions are bound to exception type names.
If a handler chooses to actually handle a condition (thereby terminating
the search) it must itself perform some kind of dynamic control transfer,
rather than return normally. ANSI CL provides a dynamic control mechanism
known as restarts which is usually used for this purpose. A condition handler
may invoke a particular restart handler. Restart handlers are similar to
exception handlers: they are functions associated with symbols in the
dynamic environment.
In \*(TL, the special behavior which occurs for exceptions derived from
.code error
and those from
.code warning
is built into the exception handling system, and tied to those types.
When an error or warning exception is unhandled, the exception handling system
itself reacts, so the special behaviors occur no matter how these exceptions
are raised. In ANSI CL, the special behavior for unhandled
.code error
conditions (of invoking the debugger) is implemented only in the
.code error
function;
.code error
conditions signalled other than via that function are not subject to
any special behavior. There is a parallel situation with regard to
warnings: the
ANSI CL
.code warn
function implements a special behavior for unhandled warnings (of emitting
a diagnostic) but warnings not signalled via that function are not
treated that way.
Thus in \*(TL, there is no way to raise an error or warning that is simply
ignored due to being unhandled.
In \*(TL exceptions are a unification of conditions and restarts. From an ANSI CL
perspective, \*(TL exceptions are a lot like CL restarts, except that the
symbols are arranged in an inheritance hierarchy. \*(TL exceptions are used
both as the equivalent of ANSI CL conditions and as restarts.
In \*(TL the terminology "catch" and "handle" is used in a specific way.
To handle an exception means to receive it without unwinding, with the possibility
of declining to handle it, so that the search continues for another handler.
To catch an exception means to match an exception to a catch handler, terminate
the search, unwind and pass control to the handler.
\*(TL provides an operator called
.code handler-bind
for specifying handlers. It has a different syntax from ANSI CL's
.codn handler-bind .
\*(TL provides a macro called
.code handle
which simplifies the use of
.codn handler-bind .
This macro superficially resembles ANSI CL's
.codn handler-case ,
but is semantically different. The most notable difference is that the bodies
of handlers established by
.code handler-bind
execute without any unwinding taking place and may return normally, thereby
declining to take the exception. In other words,
.code handle
has the same semantics as
.codn handler-bind ,
providing only convenient syntax.
\*(TL provides a macro called
.code catch
which has the same syntax as
.code handle
but specifies a catch point for exceptions. If, during an exception search, a
.code catch
clause matches an exception, a dynamic control transfer takes place
from the throw site to the catch site. Then the clause body is executed.
The
.code catch
macro resembles ANSI CL's
.code restart-case
or possibly
.codn handler-case ,
depending on point of view.
\*(TL provides unified introspection over handler and catch frames.
A program can programmatically discover what handler and catches are
available in a given dynamic scope. ANSI CL provides introspection
over restarts only; the standard doesn't specify any mechanism for
inquiring what condition handlers are bound at a given point in
the execution.
.TP* Example:
The following two examples express a similar approach implemented
using ANSI Common Lisp conditions and restarts, and then using \*(TL
exceptions.
.verb
;; Common Lisp
(define-condition foo-error (error)
((arg :initarg :arg :reader foo-error-arg)))
(defun raise-foo-error (arg)
(restart-case
(let ((c (make-condition 'foo-error :arg arg)))
(error c))
(recover (recover-arg)
(format t "recover, arg: ~s~%" recover-arg))))
(handler-bind ((foo-error
(lambda (cond)
(format t "handling foo-error, arg: ~s~%"
(foo-error-arg cond))
(invoke-restart 'recover 100))))
(raise-foo-error 200))
.brev
The output of the above is:
.verb
handling foo-error, arg: 200
recover, arg: 100
.brev
The following is possible \*(TL equivalent for the above Common Lisp example.
It produces identical output.
.verb
(defex foo-error error)
(defex recover restart) ;; recommended practice
(defun raise-foo-error (arg)
(catch
(throw 'foo-error arg)
(recover (recover-arg)
(format t "recover, arg: ~s\en" recover-arg))))
(handle
(raise-foo-error 200)
(foo-error (arg)
(format t "handling foo-error, arg: ~s\en" arg)
(throw 'recover 100)))
.brev
To summarize the differences: exceptions serve as both
conditions and restarts in \*(TX. The same
.code throw
function is used to initiate exception handling for
.code foo-error
and then to transfer control out of the handler
to the recovery code. The handler accepts one exception
by raising another.
When an exception symbol is used for restarting, it is
a recommended practice to insert it into the inheritance
hierarchy rooted at the
.code restart
symbol, either by inheriting directly from
.code restart
or from an exception subtype of that symbol.
.coNP Treatment of @ errno In Built-in Exceptions
Some \*(TL library functions generate exceptions in response to
conditions arising in the operating system, and those conditions
are associated with a numeric code in the POSIX/ISO C variable
.codn errno .
This code isn't represented as an exception argument. Rather,
in many of these situations, the
.code errno
value is attached to the error message string which is passed
as the first and only exception argument. The value can be
retrieved by using the function
.code string-get-code
on the error message string. If this function returns
.codn nil ,
then no such code is available in connection with the given
error.
.TP* Example:
.verb
(catch
(open-file "AsDf")
(error (msg)
;; the value 2 is retrieved from msg
;; 2 is the common value of ENOENT
(list (string-get-code msg) msg)))
-> (2 "error opening \e"AsDf\e": 2/\e"No such file or directory\e"")
.brev
.coNP Functions @, throw @ throwf and @ error
.synb
.mets (throw < symbol << arg *)
.mets (throwf < symbol < format-string << format-arg *)
.mets (error < format-string << format-arg *)
.syne
.desc
These functions generate an exception. The
.code throw
and
.code throwf
functions generate
an exception identified by
.metn symbol ,
whereas
.code error
throws an exception of
type
.codn error .
The call
.code "(error ...)"
can be regarded as a shorthand for
.codn "(throwf 'error ...)" .
The
.code throw
function takes zero or more additional arguments. These arguments
become the arguments of a
.code catch
handler which takes the exception. The
handler will have to be capable of accepting that number of arguments.
The
.code throwf
and
.code error
functions generate an exception which has a single
argument: a character string created by a formatted print to a string stream
using the
.code format
string and additional arguments.
Because
.code error
throws an error exception, it does not return. If an error exception
is not handled, \*(TX will issue diagnostic messages and terminate.
Likewise,
.code throw
or
.code throwf
are used to generate an error exception, they do not return.
If the
.code throw
and
.code throwf
functions are used to generate an exception not derived from
.codn error ,
and no handler is found which accepts the exception, they return normally, with
a value of
.codn nil .
.coNP Macros @, catch @ catch* and @ catch**
.synb
.mets (catch < try-expression
.mets \ \ >> {( symbol <> ( arg *) << body-form *)}*)
.mets (catch* < try-expression
.mets \ \ >> {( symbol >> ( type-arg << arg *) << body-form *)}*)
.mets (catch** < try-expression
.mets \ \ >> {( symbol < desc >> ( type-arg << arg *) << body-form *)}*)
.syne
.desc
The
.code catch
macro establishes an exception catching block around
the
.metn try-expression .
The
.meta try-expression
is followed by zero or more
catch clauses. Each catch clause consists of a symbol which denotes
an exception type, an argument list, and zero or more body forms.
If
.meta try-expression
terminates normally, then the catch clauses
are ignored. The catch itself terminates, and its return value is
that of the
.metn try-expression .
If
.meta try-expression
throws an exception which is a subtype of one or more of
the type symbols given in the exception clauses, then the first (leftmost) such
clause becomes the exit point where the exception is handled.
The exception is converted into arguments for the clause, and the clause
body is executed. When the clause body terminates, the catch terminates,
and the return value of the catch is that of the clause body.
If
.meta try-expression
throws an exception which is not a subtype of any of
the symbols given in the clauses, then the search for an exit point for
the exception continues through the enclosing forms. The catch clauses
are not involved in the handling of that exception.
When a clause catches an exception, the number of arguments in the catch must
match the number of elements in the exception. A catch argument list
resembles a function or lambda argument list, and may be dotted. For instance
the clause
.code "(foo (a . b))"
catches an exception subtyped from
.codn foo ,
with one or
more elements. The first element binds to parameter
.codn a ,
and the rest, if any,
bind to parameter
.codn b .
If there is only one element,
.code b
takes on the value
.codn nil .
The
.code catch*
macro is a variant of
.code catch
with the following difference: when
.code catch*
invokes a clause, it passes the exception symbol as the leftmost argument
.metn type-arg .
Then the exception arguments follow. In contrast,
only the exception arguments are passed to the clauses of
.codn catch .
The
.code catch**
macro is a further variant, which differs from
.code catch*
by requiring each catch clause to provide a description
.metn desc ,
an expression which evaluates to a character string.
The
.meta desc
expressions are evaluated in left-to-right order prior to the
evaluation of
.metn try-expression .
Also see: the
.code unwind-protect
operator, and the functions
.codn throw ,
.code throwf
and
.codn error ,
as well as the
.code handler-bind
operator and
.code handler
macro.
.coNP Operator @ unwind-protect
.synb
.mets (unwind-protect < protected-form << cleanup-form *)
.syne
.desc
The
.code unwind-protect
operator evaluates
.meta protected-form
in such a way that no matter how the execution of
.meta protected-form
terminates, the
.metn cleanup-form s
will be executed.
The
.metn cleanup-form s,
however, are not protected. If a
.meta cleanup-form
terminates via
some nonlocal jump, the subsequent
.metn cleanup-form s
are not evaluated.
.metn cleanup-form s
themselves can "hijack" a nonlocal control transfer such
as an exception. If a
.meta cleanup-form
is evaluated during the processing of
a dynamic control transfer such as an exception, and that
.meta cleanup-form
initiates its own dynamic control transfer, the original control transfer
is aborted and replaced with the new one.
The exit points for dynamic control transfers are removed as unwinding takes
place. That is to say, at the start of a dynamic control transfer, a search
takes place for the target exit point. That search might skip other exit points
which aren't targets of the control transfer. Those skipped exit points are left
undisturbed and are still visible during unwinding until their individual
binding forms are abandoned. Thus at the time of execution of an
.code unwind-protect
.metn cleanup-form ,
all of the exit points of dynamically surrounding forms are still visible, even
ones which are nearer than the targeted exit point.
.TP* Example:
.verb
(block foo
(unwind-protect
(progn (return-from foo 42)
(format t "not reached!\en"))
(format t "cleanup!\en")))
.brev
In this example, the protected
.code progn
form terminates by returning from
block
.codn foo .
Therefore the form does not complete and so the
output
.str not reached!
is not produced. However, the cleanup form
executes, producing the output
.strn cleanup! .
.coNP Macro @ ignerr
.synb
.mets (ignerr << form *)
.syne
.desc
The
.code ignerr
macro operator evaluates each
.meta form
similarly to the
.code progn
operator. If no forms are present, it returns
.codn nil .
Otherwise it evaluates each
.meta form
in turn, yielding the value of the last one.
If the evaluation of any
.meta form
is abandoned due to an exception of type
.codn error ,
the code generated by the
.code ignerr
macro catches this exception. In this situation,
the execution of the
.code ignerr
form terminates without evaluating the remaining
forms, and yields
.codn nil .
.coNP Macro @ ignwarn
.synb
.mets (ignwarn << form *)
.syne
.desc
The
.code ignwarn
macro resembles
.codn ignerr .
It arranges for the evaluation of each
.meta form
in left-to-right order. If all the forms are evaluated, then the
value of the last one is returned. If no forms are present, then
.code nil
is returned.
If any
.meta form
throws an exception of type
.code warning
then this exception is intercepted by a handler established by
.codn ignwarn .
This handler reacts by throwing an exception of type
.codn continue .
The effect is that the warning is ignored, since the handler
doesn't issue any diagnostic, and passes control to the warning's
continue point.
Note: all sites within \*(TX which throw a
.code warning
also provide a nearby catch for a
.code continue
exception, for resuming evaluation at the point where the warning
was issued.
.coNP Operator @ handler-bind
.synb
.mets (handler-bind < function-form < symbol-list << body-form *)
.syne
.desc
The
.code handler-bind
operator establishes a handler for one or more
exception types, and evaluates zero or more
.metn body-form s
in a dynamic scope in which that handler is visible.
When the
.code handler-bind
form terminates normally, the handler is removed. The value of the
last
.meta body-form
is returned, or else
.code nil
if there are no forms.
The
.meta function-form
argument is an expression which must evaluate to a function. The function
must be capable of accepting the exception arguments. All exceptions functions
require at least one argument, since the leftmost argument in an exception handler
call is the exception type symbol.
The
.meta symbol-list
argument is a list of symbols, not evaluated. If it is empty, then the handler
isn't eligible for any exceptions. Otherwise it is eligible for any exception
whose exception type is a subtype of any of the symbols.
If the evaluation of any
.meta body-form
throws an exception which is not handled within that form, and the handler
is eligible for that exception, then the function is invoked. It receives
the exception's type symbol as the leftmost argument. If the exception has
arguments, they appear as additional arguments in the function call.
If the function returns normally, then the exception search continues.
The handler remains established until the exception is handled in such a way
that a dynamic control transfer abandons the
.code handler-bind
form.
Note: while a handler's function is executing, the handler is disabled.
If the function throws an exception for which the handler is eligible,
the handler will not receive that exception; it will be skipped by the
exception search as if it didn't exist. When the handler function terminates,
either via a normal return or a nonlocal control transfer, then the handler is
reenabled.
.coNP Macros @ handle and @ handle*
.synb
.mets (handle < try-expression
.mets \ \ >> {( symbol <> ( arg *) << body-form *)}*)
.mets (handle* < try-expression
.mets \ \ >> {( symbol >> ( type-arg << arg *) << body-form *)}*)
.syne
.desc
The
.code handle
macro is a syntactic sugar for the
.code handler-bind
operator. Its syntax is exactly like that of
.codn catch .
The difference between
.code handle
and
.code catch
is that the clauses in
.code handle
are invoked without unwinding. That is to say,
.code handle
does not establish an exit point for an exception. When control passes to
a clause, it is by means of an ordinary function call and not a dynamic
control transfer. No evaluation frames are yet unwound when this takes place.
The
.code handle
macro establishes a handler, by
.code handler-bind
whose
.meta symbol-list
consists of every
.meta symbol
gathered from every clause.
The handler function established in the generated
.code handler-bind
is synthesized from all of the clauses, together with dispatch logic which
which passes the exception and its arguments to the first
eligible clause.
The
.meta try-expression
is evaluated in the context of this handler.
The clause of the
.code handle
syntax can return normally, like a function, in which case the handler
is understood to have declined the exception, and exception processing
continues. To handle an exception, the clause of the
.code handle
macro must perform a dynamic control transfer, such returning from a block
via
.code return
or throwing an exception.
The
.code handle*
macro is a variant of
.code handle
with the following difference: when
.code handle*
invokes a clause, it passes the exception symbol as the leftmost argument
.metn type-arg .
Then the exception arguments follow. In contrast,
only the exception arguments are passed to the clauses of
.codn handle .
.coNP Macro @ with-resources
.synb
.mets (with-resources >> ({( sym >> [ init-form <> [ cleanup-form *]])}*)
.mets \ \ << body-form *)
.syne
.desc
The
.code with-resources
macro provides a sequential binding construct similar to
.codn let* .
Every
.meta sym
is established as a variable which is visible to the
.metn init-form s
of subsequent variables, to all subsequent
.metn cleanup-form s
including that of the same variable,
and to the
.metn body-form s.
If no
.meta init-form
is supplied, then
.meta sym
is bound to the value
.codn nil .
If an
.meta init-form
is supplied, but no
.metn cleanup-form s,
then
.meta sym
is bound to the value of the
.metn init-form .
If one or more
.metn cleanup-form s
are supplied in addition to
.metn init-form ,
they specify forms to be executed upon the termination of the
.code with-resources
construct.
When an instance of
.code with-resources
terminates, either normally or by a nonlocal control transfer,
then for each
.meta sym
whose
.meta init-form
had executed, thus causing that
.meta sym
to be bound to a value, the
.metn cleanup-form s
corresponding to
.meta sym
are evaluated in the usual left-to-right order.
The
.metn sym s
are cleaned up in reverse (right-to-left) order. The
.metn cleanup-form s
of the most recently bound
.meta sym
are processed first; those of the least recently bound
.meta sym
are processed last.
When the
.code with-resources
form terminates normally, the value of the last
.meta body-form
is returned, or else
.code nil
if no
.metn body-form s
are present.
.TP* Note:
From its inception, until \*(TX 265,
.code with-resources
featured an undocumented behavior. Details are given in the
COMPATIBILITY section's Compatibility Version Values subsection,
in the notes for compatibility value 265.
.TP* "Example:"
The following expression opens a text file and reads a line from it,
returning that line, while ensuring that the stream is closed
immediately:
.verb
(with-resources ((f (open-file "/etc/motd") (close-stream f)))
(whilet ((l (get-line f)))
(put-line l)))
.brev
Note that a better way to initialize exactly one stream resource
is with the
.code with-stream
macro, which implicitly closes the stream when it terminates.
.coNP Special variable @ *unhandled-hook*
The
.code *unhandled-hook*
variable is initialized with
.code nil
by default.
It may instead be assigned a function which is capable of taking
three arguments.
When an exception occurs which has no handler, this function is called,
with the following arguments: the exception type symbol, the exception object,
and a third value which is either
.code nil
or else the form which was being evaluated when the exception was thrown.
The call occurs before any unwinding takes place.
If the variable is
.codn nil ,
or isn't a function, or the function returns after being called,
then unwinding takes place, after which some informational messages are printed
about the exception, and the process exits with a failed termination status.
In the case when the variable contains a object other than
.code nil
which isn't a function, a diagnostic message is printed on the
.code *stderr*
stream prior to unwinding.
Prior to the function being called, the
.code *unhandled-hook*
variable is reset to
.codn nil .
Note: the functions
.code source-loc
or
.code source-loc-str
may be applied to the third argument of the
.code *unhandled-hook*
function to obtain more information about the form.
.coNP Macro @ defex
.synb
.mets (defex <> { symbol }*)
.syne
.desc
The macro
.code defex
records hierarchical relationships among symbols, for the purposes
of the use of those symbols as exceptions. It is closely related to the
.code @(defex)
directive in the \*(TX pattern language, performing the same function.
All symbols are considered to be exception subtypes, and every symbol
is implicitly its own exception subtype. This macro does not introduce
symbols as exception types; it only introduces subtype-supertype
relationships.
If
.code defex
is invoked with no arguments, it has no effect.
If arguments are present, they must be symbols.
If
.code defex
is invoked with only one symbol as its argument, it has no effect.
At least two
symbols must be specified for a useful effect to take place. If exactly two
symbols are specified, then, subject to error checks,
.code defex
makes the left symbol an
.I exception subtype
of the right symbol.
This behavior generalizes to three or more arguments: if three or more symbols
are specified, then each symbol other than the last is registered as a subtype of
the symbol which follows.
If a
.code defex
has three or more arguments, they are processed from left to right.
If errors are encountered during the processing, the correct registrations
already made for prior arguments remain in place.
Every symbol is implicitly considered to be its own exception subtype,
therefore it is erroneous to explicitly register a symbol as its
own subtype.
The symbol
.code nil
is implicitly a subtype of every exception type. Therefore, it is erroneous
to attempt to specify it as a supertype in a registration.
Using
.code nil
as a subtype in a registration is silently permitted, but has no effect.
No explicit registration is recorded between
.code nil
and its successor in the argument list.
The symbol
.code t
is implicitly the supertype of every exception type. Therefore, it
is erroneous to attempt to register it as an exception subtype.
Using
.code t
as a supertype in a registration is also erroneous.
A symbol
.code a
may not be registered as a subtype of a symbol
.code b
if the reverse relationship already exists between those two symbols.
The foregoing rules allow redefinitions to take place, while forbidding cycles
from being created in the exception subtype inheritance graph.
Keyword symbols may be used as exception types.
.coNP Function @ register-exception-subtypes
.synb
.mets (register-exception-subtypes <> { symbol }*)
.syne
.desc
The
.code register-exception-subtypes
function constitutes the underlying implementation for the
.code defex
macro.
The following equivalence applies:
.verb
(defex a b ...) <--> (register-exception-subtypes 'a 'b ...)
.brev
That is, the
.code defex
macro works as if by generating a call to the function, with
the arguments quoted.
The semantics of the function is precisely that of the macro.
.coNP Function @ exception-subtype-p
.synb
.mets (exception-subtype-p < left-symbol << right-symbol )
.syne
.desc
The
.code exception-subtype-p
function tests whether two symbols are in a relationship as exception types,
such that
.meta left-symbol
is a direct or indirect exception subtype of
.metn right-symbol .
If that is the case, then
.code t
is returned, otherwise
.codn nil .
.coNP Function @ exception-subtype-map
.synb
.mets (exception-subtype-map)
.syne
.desc
The
.code exception-subtype-map
function returns a tree structure which captures information
about all registered exception types.
The map appears as an association list which contains an entry
for every exception symbol, paired with that type's supertype path.
The first element in the supertype path is the exception's immediate
supertype. The next element is that type's supertype and so on. The
last element in every path is the grand supertype
.codn t .
For instance, if only the types
.codn a ,
.code b
and
.code c
existed in the system, and were linked according to this inheritance graph:
.verb
t ----+--- b --- a
|
+--- c
.brev
such that the supertype of
.code b
and
.code c
is
.codn t ,
and
.code a
has
.code b
as supertype, then the function might return:
.verb
((a b t) (b t) (c t) (t))
.brev
or any other equivalent permutation.
The returned list may share substructure, so that the
.code "(t)"
sublist is shared among all four entries, and
.code "(b t)"
between the first two.
If the program alters the tree structure returned by
.codn exception-map-p ,
the consequences are unspecified; this structure may be the actual object which
represents the type hierarchy.
.coNP Structures @, frame @ catch-frame and @ handle-frame
.synb
.mets (defstruct frame nil)
.mets (defstruct catch-frame frame types desc jump)
.mets (defstruct handle-frame frame types fun)
.syne
.desc
The structure types
.codn frame ,
.code catch-frame
and
.code handle-frame
are used by the
.code get-frames
and
.code find-frame
functions to represent information about the currently established
exception catches (see the
.code catch
macro) and handlers
(see
.code handler-bind
and
.codn handler ).
The
.code frame
type serves as the common base for
.code catch-frame
and
.codn handle-frame .
Modifying any of the slots of these structures has no effect on the
actual frame from which they are derived; the frame structures are only
representation which provides information about frames. They are not
the actual frames themselves.
Both
.code catch-frame
and
.code handle-frame
have a
.code types
slot. This holds the list of exception type symbols which are matched
by the catch or handler.
The
.code desc
slot of a
.code catch-frame
holds a list of the descriptions produced by the
.code catch**
macro. If there are no descriptions, then this member is
.codn nil ,
otherwise it is a list whose elements are in correspondence
with the list in the
.code types
slot.
The
.code jump
slot of a
.code catch-frame
is an opaque
.code cptr
("C pointer")
object which is related to the stack address of the catch
frame. If it is altered, the catch frame object becomes invalid
for the purposes of
.codn invoke-catch .
The
.code fun
slot of a
.code handle-frame
is the registered handler function. Note that all the clauses of a
.code handler
macro are compiled to a single function, which is established via
.codn handler-bind ,
so an instance of the
.code handler
macro corresponds to a single
.codn handle-frame .
.coNP Function @ get-frames
.synb
.mets (get-frames)
.syne
.desc
The
.code get-frames
function inquires the current dynamic environment in order to retrieve
information about established exception catch and handler frames.
The function returns a list, ordered from the innermost nesting
level to the outermost nesting, of structure objects derived from the
.code frame
structure type. The list contains two kinds of objects: structures
of type
.code catch-frame
and of type
.codn handle-frame .
These objects are not the frames themselves, but only provide information
about frames. Modifying the slots in these structures has no effect on
the original frames. Also, these structures have their own lifetime and
can endure after the original frames have disappeared. This has implications
for the use of the
.code invoke-catch
function.
The
.code handle-frame
structures have a
.code fun
slot, which holds a function. It may be invoked directly.
A
.code catch-frame
structure may be passed as an argument to the
.code invoke-catch
function.
.coNP Functions @ find-frame and @ find-frames
.synb
.mets (find-frame >> [ exception-symbol <> [ frame-type ]])
.mets (find-frames >> [ exception-symbol <> [ frame-type ]])
.syne
.desc
The
.code find-frame
function locates the first (innermost) instance of a specific kind of
exception frame (a catch frame or a handler frame) which is eligible
for processing an exception of a specific type. If such a frame
is found, it is returned. The returned frame object is of the same kind as the
objects which comprise the list returned by the function
.codn get-frames .
If such a frame is not found,
.code nil
is returned.
The
.meta exception-symbol
argument specifies a match by exception type: the candidate frame
must specify in its list of matches at least one type which is an exception
supertype of
.metn exception-symbol .
If this argument is omitted, it defaults to
.code nil
which finds any handler that matches at least one type. There is no way to
search for handlers which match an empty set of types; the
.code find-frame
function skips such frames.
The
.meta frame-type
argument specifies which frame type to find. Useful values for this
argument are the structure type names
.code catch-frame
and
.code handle-frame
or the actual structure type objects which these type names denote.
If any other value is specified, the function returns
.codn nil .
If the argument is omitted, it defaults to the type of the
.code catch-frame
structure. That is to say, by default, the function looks for catch
frames.
Thus, if
.code find-frame
is called with no arguments at all it finds the innermost catch frame,
if any exists, or else returns
.codn nil .
The
.code find-frames
function is similar to
.code find-frame
except that it returns all matching frames, ordered from the innermost nesting
level to the outermost nesting. If called with no arguments, it returns a
list of the catch frames.
.coNP Function @ invoke-catch
.synb
.mets (invoke-catch < catch-frame < symbol << argument *)
.syne
.desc
The
.code invoke-catch
function abandons the current evaluation context to perform
a nonlocal control transfer directly to the catch
described by the
.meta catch-frame
argument, which must be a structure of type
.code catch-frame
obtained using any of the functions
.codn get-frames ,
.code find-frames
or
.codn find-frame .
The control transfer is possible only if the catch
frame represented by
.meta catch-frame
structure is still established, and if the structure
hasn't been tampered with.
If a given
.code catch-frame
structure is usable with
.codn invoke-catch ,
then a copy of that structure made with
.code copy-struct
is also usable, denoting the same catch frame.
The
.meta symbol
argument should be an exception symbol. It is passed to the
exception frame, as if it had appeared as the first argument of the
.code throw
function. Similarly, the
.metn argument s
are passed to the catch frame as if they were the trailing arguments
of a
.codn throw .
The difference between
.code invoke-catch
and
.code throw
is that
.code invoke-catch
targets a specific catch frame as its exit point, rather than searching for a
matching catch or handler frame. That specific frame receives the control.
The frame receives control even if it it is not otherwise eligible for
catching the exception type denoted by
.metn symbol .
.coNP Macro @ assert
.synb
.mets (assert < expr >> [ format-string << format-arg *])
.syne
.desc
The
.code assert
macro evaluates
.metn expr .
If
.meta expr
yields any true value, then
.code assert
terminates normally, and that value is returned.
If instead
.meta expr
yields
.codn nil ,
then
.code assert
throws an exception of type
.codn assert .
The exception carries an informative character string that contains
a diagnostic detailing the expression which yielded
.codn nil ,
and the source location of that expression, if available.
If the
.meta format-string
and possibly additional format arguments are given to
.code assert
then those arguments are used to format additional text which is appended to
the diagnostic message after a separating character such as a colon.
.SS* Static Error Diagnosis
This section describes a number of features related to the diagnosis
of errors during the static processing of program code prior to evaluation.
The material is of interest to developers of macros intended for broad
reuse.
.NP* Error Exceptions
\*(TL uses exceptions of type
.code eval-error
to identify erroneous situations during both transformation of code
and its evaluation. These exceptions have one argument, which is a
character string. If not handled by program code,
.code eval-error
exceptions are specially recognized and treated by the built-in handling logic.
The message is incorporated into diagnostic output which includes
more information which is deduced.
.NP* Warning Exceptions
\*(TL uses exceptions of type
.code warning
to identify certain situations of interest. Ordinary non-deferrable
warnings have a structure identical to errors, except for the exception
symbol. \*(TX's provides built-in "auto continue" handling for warnings. If a warning
exception is not intercepted by a catch or an accepting handler, then a
diagnostic is issued on the
.code *stderr*
stream, after which a
.code continue
exception is thrown with no arguments. If that
.code continue
exception is not handled, then control returns normally to the point that
exception to resume the computation which generated the warning.
Callers which invoke code that may generate warning exceptions are therefore
not required to handle them. However, callers which do handle warning
exceptions expect to be able to throw a
.code continue
exception in order to resume the computation that triggered the warning,
without allowing other handlers to see the exception.
The generation of a warning should thus conform to the following pattern:
.verb
(catch
(throw 'warning "message")
(continue ()))
.brev
.NP* Deferrable Warnings
\*(TX supports a form of diagnostic known as a
.IR "deferrable warning" .
A deferrable warning is distinguished in two ways. Firstly, it is
either of the type
.code defr-warning
or subtyped from that type. The
.code defr-warning
type itself is a direct subtype of
.codn warning .
Secondly, a deferrable warning carries an additional tag argument after the
exception message. A deferrable exception is thrown according to
this pattern:
.verb
(catch
(throw 'defr-warning "message" . tag)
(continue ()))
.brev
\*(TX's built-in exception handling logic reacts specially to the
presence of the tag material in the exception. First, the global
.I "tentative definition list"
is searched for the presence of the tag, using
.code equal
equality. If the tag is found, then the warning is discarded.
If the tag is not found, then the exception argument list is added
to the global
.IR "deferred warning list" .
In either case,
the
.code continue
exception is thrown to resume the computation which threw the warning,
as in the case of an ordinary non-deferrable warning.
The purpose of this mechanism is to suppress warnings which
become superfluous when more of the program code is examined.
For instance, a warning about a call to an undefined function is
superfluous if a definition of that function is supplied later,
yet before that function call is executed.
Deferred warnings accumulate in the deferred warning list
from which they can be removed. The list is purged at various
times such as when a top-level load completes, and the
deferred warnings are released, as if by a call to the
.code release-deferred-warnings
function.
.coNP Functions @ compile-error and @ compile-warning
.synb
.mets (compile-error < context-obj < fmt-string << fmt-arg *)
.mets (compile-warning < context-obj < fmt-string << fmt-arg *)
.syne
.desc
The functions
.code compile-error
and
.code compile-warning
provide a convenient and uniform way for code transforming
functions such as macro-expanders to generate diagnostics.
The
.code compile-error
function throws an exception of type
.codn eval-error .
The
.code compile-warning
function throws an exception of type
.code warning
and internally provides a
.code catch
for the
.code continue
exception which allow a warning handler to resume execution
after the warning. If a handler throws a
.code continue
exception which is caught by
.codn compile-warning ,
then
.code compile-warning
returns
.codn nil .
Because
.code compile-warning
throws a non-error exception, it returns
.code nil
in the event that no catch is found for the exception, and no handler which
accepts it.
The argument conventions are the same for both functions.
The
.meta context-obj
is typically a compound form to which the diagnostic
applies.
The functions produce a diagnostic message which
incorporates the location information and symbol
obtained from
.meta context-obj
and the
.codn format -style
arguments
.meta fmt-string
and its
.metn fmt-arg s.
.coNP Function @ compile-defr-warning
.synb
.mets (compile-defr-warning < context-obj < tag
.mets \ \ < fmt-string << fmt-arg *)
.syne
.desc
The
.code compile-defr-warning
function throws an exception of type
.code defr-warning
and internally provides a
.code catch
for the
.code continue
exception needed to resume after the warning.
The function produces a diagnostic message which
incorporates the location information and symbol
obtained from
.meta context-obj
and the
.codn format -style
arguments
.meta fmt-string
and its
.metn fmt-arg s.
This diagnostic message constitutes the first
argument of the exception. The
.meta tag
argument is taken as the second argument.
If the exception isn't intercepted by a catch or by
an accepting handler,
.code compile-defr-warning
returns
.codn nil .
In also returns
.code nil
if it catches a
.code continue
exception.
.coNP Function @ purge-deferred-warning
.synb
.mets (purge-deferred-warning << tag )
.syne
.desc
The
.code purge-deferred-warning
removes all warnings marked with
.meta tag
from the deferred list. It also removes all tags
matching
.meta tag
from the tentative definition list.
Tags are compared using the
.code equal
function.
.coNP Function @ register-tentative-def
.synb
.mets (register-tentative-def << tag )
.syne
.desc
The
.code register-tentative-def
function adds
.meta tag
to the list of tentative definitions which are
used to suppress deferrable warnings.
The idea is that a definition of some construct has been seen,
but not yet executed. Thus the construct is not defined, but
it can reasonably be expected that it will be defined;
hence, warnings about its nonexistence can be suppressed.
For example, in the following code, when the expression
.code "(foo)"
is being expanded and transformed, the
.code foo
function does not exist:
.verb
(progn (defun foo ()) (foo))
.brev
The function won't be defined until the
.code progn
is evaluated. Thus a warning is generated that
.code "(foo)"
refers to an undefined function.
However, this warning is discarded, because the
expander for
.code defun
registers a tentative definition tag for
.codn foo .
When the definition of
.code foo
takes place, the
.code defun
operator will call
.code purge-deferred-warning
which will remove not only all accumulated warnings
related to the undefinedness of
.code foo
but also remove the tentative definition.
Note: this mechanism isn't perfect because it will
still suppresses the warning in situations like
.verb
(progn (if nil (defun foo ())) (foo))
.brev
.coNP Function @ tentative-def-exists
.synb
.mets (tentative-def-exists << tag )
.syne
.desc
The
.code tentative-def-exists
function checks whether
.meta tag
has been registered via
.code register-tentative-def
and not yet purged by
.codn purge-deferred-warning .
.coNP Function @ defer-warning
.synb
.mets (defer-warning << args )
.syne
.desc
The
.code defer-warning
function attempts to register a deferred warning. The
.meta args
argument corresponds to the arguments which are passed to the
.code throw
function in order to generate a warning exception, not including the exception
symbol.
Args is expected to have at least two elements, the second of which
is a deferred warning tag.
The
.code defer-warning
function returns
.codn nil .
Note: this function is intended for use in exception handlers. The
following example shows a handler which intercepts warnings. It defers
deferrable warnings, and prints ordinary warnings:
.verb
(handle
(some-form ..) ;; some code which might generate warnings
(defr-warning (msg tag) ;; catch deferrable and defer
(defer-warning (cons msg tag))
(throw 'continue)) ;; warning processed: resume execution
(warning (msg)
(put-line `warning: @msg`) ;; print non-deferrable
(throw 'continue))) ;; warning processed: resume execution
.brev
.coNP Function @ release-deferred-warnings
.synb
.mets (release-deferred-warnings)
.syne
.desc
The
.code release-deferred-warnings
removes all warnings from the deferred list.
Then, it issues each deferred warning as an ordinary warning.
Note: there is normally no need for user programs to use this
function since deferred warnings are issued automatically.
.coNP Function @ dump-deferred-warnings
.synb
.mets (dump-deferred-warnings << stream )
.syne
.desc
The
.code dump-deferred-warnings
empties the list of deferred warnings, and converts each one
into a diagnostic message sent to
sent to
.metn stream .
After the diagnostics are printed, the list of pending warnings
is cleared.
Note: there is normally no need for user programs to use this
function since deferred warnings are issued automatically.
.SS* Delimited Continuations
\*(TL supports delimited continuations, which are integrated with the
.code block
feature. Any named or anonymous block, including the implicit blocks
created around function bodies, can be used as the delimiting
.I prompt
for the capture of a continuation.
A delimited continuation is section of a possible future of the
computation, up to a delimiting prompt,
.I reified
as a first class function.
.TP* Example:
.verb
(defun receive (cont)
(format t "cont returned ~a\en" (call cont 3)))
(defun function ()
(sys:capture-cont 'abcd (fun receive)))
(block abcd
(format t "function returned ~a\en" (function))
4)
Output:
function returned 3
cont returned 4
function returned t
.brev
.PP
Evaluation begins with the
.code block
form. This form calls
.code function
which uses
.code sys:capture-cont
to capture a continuation up to the
.code abcd
prompt. The continuation is passed to the
.code receive
function as an argument.
This captured object represents the continuation of computation
up to that prompt. It appears as a one-argument function which, when called,
resumes the captured computation. Its argument emerges out of the
.code sys:capture-cont
call as a return value. When the computation eventually returns all
the way to the delimiting prompt, the return value of that prompt
will then appear as the return value of the continuation function.
In this example, the function
.code receive
immediately invokes the continuation function which it receives, passing
it the argument value
.codn 3 .
And so,
evaluation now continues in the resumed future represented by the
continuation. Inside the continuation,
.code sys:capture-cont
appears to return, yielding the value
.codn 3 .
This bubbles up through
.code function
up to the
.code "block abcd"
where a message is printed:
.strn "function returned 3" .
The
.code block
terminates, yielding the value 4. Thereby, the continuation ends, since
it is delimited up to that block. Control now returns to the
.code receive
function which invoked the continuation, where the function call form
.code "(call cont)"
terminates, yielding the value
.code 4
that was returned by the continuation's delimiting
.code block
form. The message
.str "cont returned 4"
is printed. The
.code receive
function returns normally, returning the value
.code t
which emerged from the
.code format
call. Control is now back in
.code function
where the
.code sys:capture-cont
form terminates and returns the
.codn t .
This bubbles up to
.code block
which prints
.strn "function returned t" .
In summary, a continuation represents, as a function, the subsequent
computation that is to take place starting at some point, up to some
recently established, dynamically enclosing delimiting prompt. When
the continuation is captured, that future doesn't have to take place;
an alternative future can carry out in which that continuation is
available as a function. That alternative future can invoke the continuation at
will. Invocations (resumptions) of the continuation appear as additional
returns from the capture operator. A resumption of a continuation terminates
when the delimiting prompt terminates, and the continuation yields the
value which emerges from the prompt.
Delimited continuations are implemented by capturing a segment of the
evaluation stack between the prompt and the capture point. When
a continuation is resumed, this saved copy of a stack segment is inserted on
top of the current stack and the procedure context is resumed such
that evaluation appears to emerge from the capture operator.
As the continuation runs to completion, it simply pops these inserted
stack frames naturally. Eventually it pops out of the delimiting prompt,
at which point control ends up at the point which invoked the continuation
function.
The low-level operator for capturing a continuation is
.codn sys:capture-cont .
More expressive and convenient programming with continuations is
provided by the macros
.codn obtain ,
.codn obtain-block ,
.code yield-from
and
.codn yield ,
which create an abstraction which models the continuation as a suspended
procedure supporting two-way communication of data.
A
.code suspend
operator is provided, which is more general. It is identical to the
.code shift
operator described in various computer science literature about
delimited continuations, except that it refers to a specific delimiting
prompt by name.
Continuations raise the issue of what to do about unwinding.
The language Scheme provides the much criticized
.code dynamic-wind
operator which can execute initialization and clean-up code as
a continuation is entered and abandoned. \*(TX takes a simpler,
albeit risky approach. It provides a non-unwinding escape operator
.code sys:abscond-from
for use with continuations. Code which has captured a continuation
can use this operator to escape from the delimiting block without
triggering any unwinding among the frames between the capture point and the
delimiter. When the continuation is restarted, it will then do so
with all of the resources associated with it frames intact.
When the continuation executes normal returns within its context,
the unwinding takes place then. Thus tidy, "thread-like" use
of continuations is possible with a small measure of coding discipline.
Unfortunately, the absconding operator is dangerous: its use
breaks the language guarantee that clean-up associated with a form is done no
matter how a form terminates.
.NP* Comparison with Lexical Closures
Delimited continuations resemble lexical closures in some ways. Both
constructs provide a way to return to some context whose evaluation
has already been abandoned, and to access some aspects of that context.
However, lexical closures are statically scoped. Closures capture the lexically
apparent scope at a given point, and produce a function whose body has access
to that scope, as well as to some arbitrary arguments. Thus, a lexical scope
is reified as a first-class function. By contrast, a delimited continuation
is dynamic. It captures an an entire segment of a program activation chain,
up to the delimiting prompt. This segment includes scopes which are not
lexically visible at the capture point: the scopes of parent functions.
Moreover, the segment includes not only scopes, but also other aspects of
the evaluation context, such as the possibility of returning to callers,
and the (captured portion of) the original dynamic environment, such as
exception handlers. That is to say, a lexical closure's body cannot return to
the surrounding code or see any of its original dynamic environment; it can
only inspect the environment, and then return to its own caller. Whereas a
restarted delimited continuation can continue evaluation of the surrounding
code, return to surrounding forms and parent functions, and access the dynamic
environment. The continuation function returns to its caller when that entire
restarted context terminates, whereas a closure returns to its caller as soon
as the closure body terminates.
.NP* Differences in Compiled vs. Interpreted Behavior
Delimited continuations in \*(TX expose a behavioral difference between
compiled and interpreted code which mutates the values of lexical variables.
When a continuation is captured in compiled code, it captures not only the
bindings of lexical variables, but also potentially their current values
at the time of capture. What this means is that whenever the continuation
is resumed, those variables will appear to have the captured values,
regardless of any mutations that have taken place since. In other words,
the captured future includes those specific values. This is because in
compiled code, variables are allocated on the stack, which is copied as part of
creating a continuation. Those variables are effectively newly instantiated in
each resumption of the continuation, when the captured stack segment
is reinstated into the stack, and take on those original values.
In contrast, interpretation of code only maintains an
environment pointer on the stack; the lexical environment is a dynamically
allocated object whose contents aren't included in the continuation's
stack segment capture. If the captured variables are modified after the
capture, the continuation will see the updated values: all resumptions of the
continuation share the same instance of the captured environment among
themselves, and with the original context where the capture took place.
An additional complication is that when compiled code captures lexical
closures, captured variables are moved into dynamic storage and then
they become shared: the semantics of the mutation of those variables
is then similar to the situation in interpreted code. Therefore, the
above described non-sharing capture behavior of compiled code is not required
to hold.
In continuation-based code which relies on mutation of lexical variables
created with
.code let
or
.codn let* ,
the macros
.code hlet
and
.code hlet*
can be used instead. These macros create variable bindings whose storage is
always outside of the stack, and therefore the variables will exhibit
consistent interpreted and compiled semantics under continuations.
All contexts which capture the same lexical binding of a given
.cod3 hlet / hlet*
variable share a single instance. The most recent assignment
to the variable taking place in any context establishes its value,
as seen by any other context. The resumption of a continuation will not restore
such a variable to a previous value.
If the affected variables are other kinds of bindings such as
function parameters or variables created with specialized binding
constructs such as
.codn with-stream ,
additional coding changes may be required to get interpreted code
working under compilation.
.coNP Function @ sys:capture-cont
.synb
.mets (sys:capture-cont < name < receive-fun <> [ context-form ])
.syne
.desc
The
.code sys:capture-cont
function captures a continuation, and also serves as the resume point
for the resulting continuation. Which of these two situations is the
case (capture or resumption) is distinguished by the use of the
.meta receive-fun
argument, which must be a function capable of being called with one
argument.
A block named
.meta name
must be visible; the continuation is delimited by the closest
enclosing block of this name.
The optional
.meta context-form
argument should be a compound form. If
.code sys:capture-cont
reports an error, it reports it against this form,
and uses the form's operator symbol as the name of the function which
encountered the error. If the argument is omitted,
.code sys:capture-cont
uses its own name.
The
.code sys:capture-cont
function captures a continuation, represented as a function.
It immediately calls
.metn receive-fun ,
passing it it the continuation function as an argument.
If
.meta receive-fun
returns normally, then
.code sys:capture-cont
returns whatever value
.meta receive-fun
returns.
Resuming a continuation is done by invoking the continuation function.
When this happens, the entire continuation context is restored by recreating
its captured evaluation frames on top of the current stack. Inside the
continuation, the
.code sys:capture-cont
function call which captured the continuation now appears to return,
and yields a value. That value is precisely the value which was just
passed to the continuation function moments ago.
The resumed continuation can terminate in one of three ways. Firstly, it can
simply keep executing until it discards all of its evaluation frames below the
delimiting block, and then allows that block to terminate naturally by
evaluating the last form contained in the block. Secondly, can use
.code return-from
against its delimiting block to explicitly abandon all evaluations in between
and terminate that block. Or it may perform
a nonlocal control transfer past the delimited block somewhere into the
evaluation frames of the caller. In the first two cases, the termination
of the block turns into an ordinary return from the continuation function, and
the result value of the terminated block becomes the return value of that
function call. In the last case, the call of the continuation function is
abandoned and unwinding continues through the caller.
If the symbol
.code sys:cont-poison
is passed to the continuation function, the continuation will be
resumed in a different manner: its context will be restored as in the
ordinary resume case, whereupon it will be immediately abandoned by
a nonlocal exit, causing unwinding to take place across all of the
continuation's evaluation frames. The function then returns
.codn nil .
If the symbol
.code sys:cont-free
is passed to the continuation function, the continuation isn't
be resumed at all; rather, the buffer which holds the saved context
of the continuation is released. Thereafter, an attempt to resume
the continuation results in an error exception being thrown.
After releasing the buffer, the function returns
.codn nil .
.TP* Notes:
The continuation function may be used any time after it is produced, and may be
called more than once, regardless of whether the originally captured dynamic
context is still executing. The continuation object may be communicated into
the resumed continuation, which can then use it to call itself, resulting
in multiple nested resumptions of the same continuation. A delimited
continuation is effectively a first class function.
The underlying continuation object produced by
.code sys:capture-cont
stores a copy of the captured dynamic context. Whenever the continuation
function is invoked, a copy of the captured is reinstated as if it were a new
context. Thus each apparent return from the
.code sys:capture-cont
inside a resumed continuation is not actually made in the original context, but
in a copy of that context. That context can be resumed multiple times
sequentially or recursively.
Just like lexical closures, continuations do not copy lexical environments;
they capture lexical environments by reference. If a continuation modifies
the values of captured lexical variables, those modifications are visible to
other resumptions of the same continuation, to other continuations which
capture the same environment, to lexical closures which capture the same
environment and to the original context which created that environment, if it
is still active.
Unlike lexical closures, continuations do capture the local bindings
of special variables. That is to say, if
.code *var*
is a special variable, then a lexical closure created inside a
.code "(let ((*var* 42)) ...)"
form will not capture the local rebinding of
.code *var*
which holds 42. When the closure is invoked and accesses
.codn *var* ,
it accesses whatever value of
.code *var*
is dynamically current, as dictated by the environment which calls the
closure, rather than the capturing environment.
With continuations, the behavior is different. If a continuation
is captured inside a
.code "(let ((*var* 42)) ...)"
form then it does capture the local binding. This is regardless whether
the delimited prompt of the capture is enclosed in this form, or
outside of the form.
The special variable has a binding in a dynamic environment. There is always a
reference to a current dynamic environment associated with every evaluation
context, and a continuation captures that reference. Because it is a
reference, it means that the binding is shared. That is to say, all
invocations of all continuations which capture the same dynamic environment in
which that
.code "(let ((*var* 42)) ...)"
binding was made share the same binding; if
.code *var*
is modified by assignment, the modification is visible to all those views.
Inside a resumed continuation, a form which binds a special variable such as
.code "(let ((*var* 42)) ...)"
may terminate. As expected, this causes the binding to be removed,
revealing either another local binding of
.code *var*
or the global binding. However, this unbinding only affects only that
that executing continuation; it has no effect inside other instances of the
same continuation or other continuations which capture the same variable.
Unbinding isn't a mutation of the dynamic environment, but may be understood
as merely the restoration of an earlier dynamic environment reference.
.TP* "Example:"
The following example shows an implementation of the
.code suspend
operator.
.verb
(defmacro suspend (:form form name var . body)
^(sys:capture-cont ',name (lambda (,var)
(sys:abscond-from ,name ,*body))
',form))
.brev
.coNP Operator @ sys:abscond-from
.synb
.mets (sys:abscond-from < name <> [ value ])
.syne
.desc
The
.code sys:abscond-from
operator closely resembles
.code return-from
and performs the same function: it causes an enclosing block
.meta name
to terminate with
.meta value
which defaults to
.codn nil .
However, unlike
.codn return-from ,
.code sys:abscond-from
does not perform any unwinding.
This operator should never be used for any purpose other than
implementing primitives for the use of delimited continuations.
It is used by the
.code yield-from
and
.code yield
operators to escape out of a block in which a continuation has
been captured. Neglecting to unwind is valid due to the expectation
that control will return into a restarted copy of that context.
.coNP Function @ sys:abscond*
.synb
.mets (sys:abscond* < name <> [ value ])
.syne
.desc
The
.code sys:abscond*
function is similar to the
.code sys:abscond-from
operator, except that
.code name
is an ordinary function parameter, and so when
.code return*
is used, an argument expression must be specified which evaluates
to a symbol. Thus
.code sys:abscond*
allows the target block of a return to be dynamically computed.
The following equivalence holds between the operator and function:
.verb
(sys:abscond-from a b) <--> (sys:abscond* 'a b)
.brev
Expressions used as
.meta name
arguments to
.code abscond*
which do not simply quote a symbol have no equivalent in
.codn abscond-from .
.coNP Macros @ obtain and @ yield-from
.synb
.mets (obtain << forms *)
.mets (yield-from < name <> [ form ])
.syne
.desc
The
.code obtain
and
.code yield-from
macros closely interoperate.
The
.code obtain
macro treats zero or more
.metn form s
as a suspendable execution context called the
.IR "obtain block" .
It is expected that
.metn form s
establish a block named
.meta name
and return its result value to
.codn obtain .
Without evaluating any of the forms in the obtain block,
.code obtain
returns a function, which takes one optional argument.
This argument, called the
.IR "resume value" ,
defaults to
.code nil
if it is omitted.
The function represents the suspended execution context.
The context is resumed whenever the function is called, and executes
until the next
.code yield-from
statement which references the block named
.metn name .
The function's reply argument is noted.
If the
.code yield-from
specifies a
.meta form
argument, then the execution context suspends, and the resume function
terminates and returns the value of that form. When the function is called
again to resume the context, the
.code yield-from
returns the previously noted resume value (and the new resume
value just passed is noted in its place).
If the
.code yield-from
specifies no
.meta form
argument, then it briefly suspends the execution context only
to retrieve the resume value, without producing an item. Since
no item is produced, the resume function does not return.
The execution context implicitly resumes.
When execution reaches the last form in the obtain block, the
resume value is discarded. The execution context terminates, and
the most recent call to the resume function returns the value of
that last form.
.TP* Notes:
The
.code obtain
macro registers a finalizer against the returned resume function.
The finalizer invokes the function, passing it the symbol
.codn sys:cont-poison ,
thereby triggering unwinding in the most recently captured
continuation. Thus, abandoned
.code obtain
blocks are subject to unwinding when they become garbage.
The
.code yield-from
macro works by capturing a continuation and performing a nonlocal
exit to the nearest block called
.metn name .
It passes a special yield object to that block. The
.code obtain
macro generates code which knows what to do with this special yield
object.
.TP* Examples:
The following example shows a function which recursively
traverses a
.code cons
cell structure, yielding all the
.cod2 non- nil
atoms it encounters. Finally, it returns the object
.codn nil .
The function is invoked on a list,
and the invocation is wrapped in an
.code obtain
block to convert it to a generating function.
The generating function is then called six times
to retrieve the five atoms from the list,
and the final
.code nil
value. These are collected into a list.
This example demonstrates the power of delimited
continuations to suspend and resume a recursive
procedure.
.verb
(defun yflatten (obj)
(labels ((flatten-rec (obj)
(cond
((null obj))
((atom obj) (yield-from yflatten obj))
(t (flatten-rec (car obj))
(flatten-rec (cdr obj))))))
(flatten-rec obj)
nil))
(let ((f (obtain (yflatten '(a (b (c . d)) e)))))
(list [f] [f] [f] [f] [f] [f]))
--> (a b c d e nil)
.brev
The following interactive session log exemplifies two-way communication between
the main code and a suspending function.
Here,
.code mappend
is invoked on a list of symbols representing fruit and vegetable names.
The objective is to return a list containing only fruits.
The
.code lambda
function suspends execution and yields a question out of the
.code map
block. It then classifies
the item as a fruit or not according to the reply it receives. The reply
emerges as the result value of the
.code yield-from
call.
The
.code obtain
macro converts the block to a generating function. The first call to the
function is made with no argument, because the argument would be ignored
anyway. The function returns a question, asking whether the first item
in the list, the potato, is a fruit.
To answer positively or negatively, the user calls the function again,
passing in
.code t
or
.codn nil ,
respectively.
The function returns the next question, which is answered in the
same manner.
When the question for the last item is answered, the function
call yields the final item: the ordinary result of the block, which is the list
of fruit names.
.verb
1> (obtain
(block map
(mappend (lambda (item)
(if (yield-from map `is @item a fruit?`)
(list item)))
'(potato apple banana lettuce orange carrot))))
#<interpreted fun: lambda (: reply)>
2> (call *1)
"is potato a fruit?"
3> (call *1 nil)
"is apple a fruit?"
4> (call *1 t)
"is banana a fruit?"
5> (call *1 t)
"is lettuce a fruit?"
6> (call *1 nil)
"is orange a fruit?"
7> (call *1 t)
"is carrot a fruit?"
8> (call *1 nil)
(apple banana orange)
.brev
The following example demonstrates an accumulator. Values passed to the
resume function are added to a counter which is initially zero.
Each call to the function returns the updated value of the accumulator.
Note the use of
.code "(yield-from acc)"
with no arguments to receive the value passed to the first
call to the resume function, without yielding an item.
The first return value
.code 1
is produced by the
.code "(yield-from acc sum)"
form, not by
.codn "(yield-from acc)" .
The latter only obtains the initial value
.code 1
and uses it to establish the seed value of the accumulator. Without causing
the resume function to terminate and return, control passes into the loop,
which yields the first item, causing the resume function call
.code "(call *1 1)"
to return
.codn 1 :
.verb
1> (obtain
(block acc
(let ((sum (yield-from acc)))
(while t (inc sum (yield-from acc sum))))))
#<interpreted fun: lambda (: resume-val)>
2> (call *1 1)
1
3> (call *1 2)
3
4> (call *1 3)
6
5> (call *1 4)
10
.brev
.coNP Macro @ obtain-block
.synb
.mets (obtain-block < name << forms *)
.syne
.desc
The
.code obtain-block
macro combines
.code block
and
.code obtain
into a single expression.
The
.metn form s
are evaluated in a block named
.codn name .
That is to say, the following equivalence holds:
.verb
(obtain-block n f ...) <--> (obtain (block n f ...))
.brev
.coNP Macro @ yield
.synb
.mets (yield <> [ form ])
.syne
.desc
The
.code yield
macro is to
.code yield-from
as
.code return
is to
.codn return-from :
it yields from an anonymous block.
It is equivalent to calling
.code yield-from
using
.code nil
as the block name.
In other words, the following equivalence holds:
.verb
(yield x) <--> (yield-from nil x)
.brev
.TP* Example:
.verb
;; Yield the integers 0 to 4 from a for loop, taking
;; advantage of its implicit anonymous block:
(defvarl f (obtain (for ((i 0)) ((< i 5)) ((inc i))
(yield i))))
[f] -> 0
[f] -> 1
[f] -> 2
[f] -> 3
[f] -> 4
[f] -> nil
[f] -> nil
.brev
.coNP Macros @ obtain* and @ obtain*-block
.synb
.mets (obtain* << forms *)
.mets (obtain*-block < name << forms *)
.syne
.desc
The
.code obtain*
and
.code obtain*-block
macros implement a useful variation of
.code obtain
and
.codn obtain-block .
The
.code obtain*
macro differs from
.code obtain
in exactly one regard: prior to returning the function, it invokes
it one time, with the argument value
.codn nil .
Thus, the following equivalence holds
.verb
(obtain* forms ...) <--> (let ((f (obtain forms ...)))
(call f)
f)
.brev
In other words, the suspended block is immediately resumed, so that it executes
either to completion (in which case its value is discarded), or to its first
.code yield
or
.code yield-from
call (in which case the yielded value is discarded).
Note: the
.code obtain*
macro is useful in creating suspensions which accept data rather than
produce data.
The
.code obtain*-block
macro combines
.code obtain*
and
.code block
in the same manner that
.code obtain-block
combines
.code obtain
and
.codn block .
.TP* Example:
.verb
;; Pass three values into suspended block,
;; which get accumulated into list.
(let ((f (obtain*-block nil
(list (yield nil) (yield nil) (yield nil)))))
(call f 1)
(call f 2)
(call f 3)) -> (1 2 3)
;; Under obtain, extra call is required:
(let ((f (obtain-block nil
(list (yield nil) (yield nil) (yield nil)))))
(call f nil) ;; execute block to first yield
(call f 1) ;; resume first yield with 1
(call f 2)
(call f 3)) -> (1 2 3)
.brev
.coNP Macro @ suspend
.synb
.mets (suspend < block-name < var-name << body-form *)
.syne
.desc
The
.code suspend
operator captures a continuation up to the prompt given by the
symbol
.meta block-name
and binds it to the variable name given by
.metn var-name ,
which must be a symbol suitable for binding variables with
.codn let .
Each
.meta body-form
is then evaluated in the scope of the variable
.metn var-name .
When the last
.meta body-form
is evaluated, a nonlocal exit takes place to the block
named by
.meta block-name
(using the
.code sys:abscond-from
operator, so that unwinding isn't performed).
When the continuation bound to
.meta var-name
is invoked, a copy of the entire block
.meta block-name
is restarted, and in that copy, the
.code suspend
call appears to return normally, yielding the value which had been
passed to the continuation.
.TP* Example
Define John McCarthy's
.code amb
function using
.code block
and
.codn suspend :
.verb
(defmacro amb-scope (. forms)
^(block amb-scope ,*forms))
(defun amb (. args)
(suspend amb-scope cont
(each ((a args))
(if a
(iflet ((r (call cont a)))
(return-from amb-scope r))))))
.brev
Use
.code amb
to bind the
.code x
and
.code y
which satisfy the predicate
.mono
.meti (eql (* x y) 8)
.onom
nondeterministically:
.verb
(amb-scope
(let ((x (amb 1 2 3))
(y (amb 4 5 6)))
(amb (eql (* x y) 8))
(list x y)))
-> (2 4)
.brev
.coNP Macros @ hlet and @ hlet*
.synb
.mets (hlet >> ({ sym | >> ( sym << init-form )}*) << body-form *)
.mets (hlet* >> ({ sym | >> ( sym << init-form )}*) << body-form *)
.syne
.desc
The
.code hlet
and
.code hlet*
macros behave exactly like
.code let
and
.codn let* ,
respectively except that they guarantee that the variable bindings are
allocated in storage which isn't captured by delimited continuations.
The
.code h
in the names stands for "heap", serving as a mnemonic based on the
implementation concept of these bindings being "heap-allocated".
.SS* Regular-Expression Library
\*(TX provides a "pure" regular-expression implementation based on automata
theory, which equates regular expressions, finite automata and sets of strings.
A regular expression determines whether or not a string of input characters
belongs to a set. \*(TX regular expressions do not support features such
as "anchoring" a match to the start or end of a string, or capturing
parenthesized subexpression matches into registers. Parenthesis syntax
denotes only grouping, with no additional meaning.
The semantics of whether a regular expression is used for a substring
search, prefix match, suffix match, string splitting and so forth comes from
the functions which use regular expressions to perform these operations.
.NP* Regular Expressions as Functions
.synb
.mets >> [ regex >> [ start <> [ from-end ]] << string ]
.syne
.desc
A regular expression is callable as a function in \*(TL.
When used this way, it requires a string argument. It searches
the string for the leftmost match for itself, and returns
the matching substring, which could be empty. If no match is
found, it returns
.codn nil .
A regex takes one, two, or three arguments. The required
.meta string
is always the rightmost argument. This allows for convenient
partial application over optional arguments using
macros in the
.code op
family, and macros in which the
.code op
syntax is implicit.
The optional arguments
.meta start
and
.meta from-end
are treated exactly as their like-named counterparts in the
.code search-regst
function.
.TP* Example:
Keep those elements from a list of strings which match
the regular expression
.codn #/a.*b/ :
.verb
(keep-if #/a.*b/ '#"abracadabra zebra hat adlib adobe deer")
--> ("abracadabra" "adlib" "adobe")
.brev
.coNP Functions @, search-regex @ range-regex and @ search-regst
.synb
.mets (search-regex < string < regex >> [ start <> [ from-end ]])
.mets (range-regex < string < regex >> [ start <> [ from-end ]])
.mets (search-regst < string < regex >> [ start <> [ from-end ]])
.syne
.desc
The
.code search-regex
function searches through
.meta string
starting
at position
.meta start
for a match for
.metn regex .
If
.meta start
is omitted, the search starts at position 0. If
.meta from-end
is specified and has a
.cod2 non- nil
value, the search
proceeds in reverse, from the position just beyond the last character of
.metn string ,
toward
.metn start .
If
.meta start
exceeds the length of the string, then
.code search-regex
returns
.codn nil .
If
.meta start
is negative then it indicates positions from the end of the string,
such that -1 is the last character, -2 the second last and so forth.
If the value is so negative that it refers beyond the start of
the string, then the starting position is deemed to be zero.
If
.meta start
is equal to the length of
.metn string ,
and thus refers to the position one character past its
length, then a match occurs at that position if
.meta regex
admits such a match.
The
.code search-regex
function returns
.code nil
if no match is found, otherwise it returns
a cons, whose
.code car
indicates the position of the match, and whose
.code cdr
indicates the length of the match.
If
.meta regex
is capable of matching empty strings, and no other kind of match
is found within
.metn string ,
then search regex reports a zero length match. If
.meta from-end
is false, then this match is reported at
.metn start ,
otherwise it is reported at the position one character beyond
the end of the string.
The
.code range-regex
function is similar to
.codn search-regex ,
except that when
a match is found, it returns a position range, rather than a position
and length. A range object is returned whose
.code from
field indicates the position
of the match, and whose
.code to
indicates the position one element past the
last character of the match. If the match is empty, the two integers
are equal.
Also see the
.code rr
function, which provides an alternative argument syntax for
the semantics of
.codn range-regex .
The
.code search-regst
differs from
.code search-regex
in the representation of the return value in the matching case.
Rather than returning the position and length of the match,
it returns the matching substring of
.metn string .
.coNP Functions @ match-regex and @ match-regst
.synb
.mets (match-regex < string < regex <> [ position ])
.mets (match-regst < string < regex <> [ position ])
.syne
.desc
The
.code match-regex
function tests whether
.meta regex
matches at
.meta position
in
.metn string .
If
.meta position
is not specified, it is taken to be zero. Negative values
of
.meta position
index from the right end of the string such that -1
refers to the last character. Excessively negative
values which index before the first character cause
.code nil
to be returned.
If the regex matches, then the length of the match is returned.
If it does not match, then
.code nil
is returned.
The
.code match-regst
differs from
.code match-regex
in the representation of the return value in the matching case.
Rather than returning the length of the match, it returns
matching substring of
.metn string .
.coNP Functions @ match-regex-right and @ match-regst-right
.synb
.mets (match-regex-right < string < regex <> [ end-position ])
.mets (match-regst-right < string < regex <> [ end-position ])
.syne
.desc
The
.code match-regex-right
function tests whether some substring of
.meta string
which terminates at the character position just before
.meta end-position
matches
.metn regex .
If
.meta end-position
is not specified, it defaults to the length of the string, and the function
performs a right-anchored regex match.
The
.meta end-position
argument can be a negative integer, in which case it denotes
positions from the end of the string, such that -1 refers
to the last character. If the value is excessively negative such
that the position immediately before it is before the start
of the string, then
.code nil
is returned.
If
.meta end-position
is a positive value beyond the length of
.metn string ,
then, likewise,
.code nil
is returned.
If a match is found, then the length of the match is returned.
A more precise way of articulating the role of
.meta end-position
is that for the purposes of matching,
.code string
is considered to terminate just before
.metn end-position :
in other words, that
.meta end-position
is the length of the string. The match is then anchored to the
end of this effective string.
The
.code match-regst-right
differs from
.code match-regst-right
in the representation of the return value in the matching case.
Rather than returning the length of the match, it returns
the matching substring of
.metn string .
.TP* Examples:
.verb
;; Return matching portion rather than length thereof.
(defun match-regex-right-substring (str reg : end-pos)
(set end-pos (or end-pos (length str)))
(let ((len (match-regex-right str reg end-pos)))
(if len
[str (- end-pos len)..end-pos]
nil)))
(match-regex-right-substring "abc" #/c/) -> ""
(match-regex-right-substring "acc" #/c*/) -> "cc"
;; Regex matches starting at multiple positions, but all
;; the matches extend past the limit.
(match-regex-right-substring "acc" #/c*/ 2) -> nil
;; If the above behavior is not wanted, then
;; we can extract the string up to the limiting
;; position and do the match on that.
(match-regex-right-substring ["acc" 0..2] #/c*/) -> "c"
;; Equivalent of above call
(match-regex-right-substring "ac" #/c*/) -> "c"
.brev
.coNP Function @ regex-prefix-match
.synb
.mets (regex-prefix-match < regex < string <> [ position ])
.syne
.desc
The
.code regex-prefix-match
determines whether the input string
might be the prefix of a string which matches regular expression
.metn regex .
The result is true if the input string matches
.meta regex
exactly. However, it is also true in situations in which
the input string doesn't match
.metn regex ,
yet can be extended with one or more additional characters beyond the end such
that the extended string
.B does
match.
The
.meta string
argument must be a character string. The function takes the input string to be
the suffix of
.meta string
which starts at the character position indicated by the
.meta position
argument. If that argument is omitted, then
.meta string
is taken as the input in its entirety. Negative values index backwards from
the end of
.meta string
according to the usual conventions elsewhere in the library.
Note: this function is not to be confused for the semantics
of a regex matching a prefix of a string: that capability is
provided by the functions
.codn match-regex ,
.codn m^ ,
.codn r^ ,
.code f^
and
.codn fr^ .
.TP* Examples:
.verb
;; The empty string is not a viable prefix match for
;; a regex that matches no strings at all:
(regex-prefix-match #/~.*/ "") -> nil
(regex-prefix-match #/[]/ "") -> nil
;; The empty string is a viable prefix of any regex
;; which matches at least one string:
(regex-prefix-match #// "") -> t
(regex-prefix-match #/abc/ "") -> t
;; This string doesn't match the regex because
;; it doesn't end in b, but is a viable prefix:
(regex-prefix-match #/a*b/ "aa") -> t
(regex-prefix-match #/a*b/ "ab") -> t
(regex-prefix-match #/a*b/ "ac") -> nil
(regex-prefix-match #/a*b/ "abc") -> nil
.brev
.coNP Function @ regsub
.synb
.mets (regsub >> { regex | << function } < replacement << string )
.syne
.desc
The
.code regsub
function operates in two modes, depending on whether
the first argument is a regular expression,
or function.
If the first argument is a regular expression it searches
.meta string
for multiple occurrences of non-overlapping matches for that
.metn regex .
A new string is constructed
similar to
.meta string
but in which each matching region is replaced
with using
.meta replacement
as follows.
The
.meta replacement
object may be a character or a string, in which
case it is simply taken to be the replacement for each match
of the regular expression.
The
.meta replacement
object may be a function of one argument, in
which case for every match which is found, this function is invoked,
with the matching piece of text as an argument. The function's
return value is then taken to be the replacement text.
If the first argument is a function, then it is called, with
.meta string
as its argument. The return value must be either a range
object (see the
.code rcons
function) which indicates the extent of
.meta string
to be replaced, or else
.code nil
which indicates that no replacement is to take place.
.TP* Examples:
.verb
;; match every lowercase e or o, and replace by filtering
;; through the upcase-str function:
[regsub #/[eo]/ upcase-str "Hello world!"] -> "HEllO wOrld!"
;; Replace Hello with Goodbye:
(regsub #/Hello/ "Goodbye" "Hello world!") -> "Goodbye world!"
;; Left-anchored replacement with r^ function:
(regsub (fr^ #/H/) "J" "Hello, hello!") -> "Jello, hello!"
.brev
.coNP Function @ regexp
.synb
.mets (regexp << obj )
.syne
.desc
The
.code regexp
function returns
.code t
if
.meta obj
is a compiled regular-expression
object. For any other object type, it returns
.codn nil .
.coNP Functions @ trim-left and @ trim-right
.synb
.mets (trim-left >> { regex | << prefix } << string )
.mets (trim-right >> { regex | << suffix } << string )
.syne
.desc
The
.code trim-left
and
.code trim-right
functions return a new string, equivalent to
.meta string
with a leading or trailing portion removed.
If the first argument is a regular expression
.metn regex ,
then, respectively,
.code trim-left
and
.code trim-right
find a prefix or suffix of
.meta string
which matches the regular expression.
If there is no match, or if the match is empty, then
.meta string
is returned. Otherwise, a copy of
.meta string
is returned in which the matching characters are removed.
If
.meta regex
matches all of
.meta string
then the empty string is returned.
If the first argument is a character string, then it is treated
as if it were a regular-expression match for that literal
sequence of characters. Thus,
.code trim-left
interprets that string as a
.meta prefix
to be removed, and
.code trim-right
as a
.metn suffix .
If
.meta string
starts with
.metn prefix ,
then
.code trim-left
returns a copy of
.meta string
with
.meta prefix
removed. Otherwise,
.meta string
is returned.
Likewise, if
.meta string
ends with
.metn suffix ,
then
.code trim-right
returns a copy of
.meta string
with
.meta suffix
removed. Otherwise,
.meta string
is returned.
.coNP Function @ regex-compile
.synb
.mets (regex-compile < form-or-string <> [ error-stream ])
.syne
.desc
The
.code regex-compile
function takes the source code of a regular expression,
expressed as a Lisp data structure representing an abstract syntax tree, or
else a regular expression specified as a character string, and compiles it to a
regular-expression object.
If
.meta form-or-string
is a character string, it is parsed to an
abstract syntax tree first, if by the
.code regex-parse
function.
If the parse is successful (the result is not
.codn nil )
then
the resulting tree structure is compiled by a recursive call to
.codn regex-compile .
The optional
.meta error-stream
argument is passed down to
.code regex-parse
as well as in the recursive call to
.codn regex-compile ,
if that call takes place.
If
.meta error-stream
is specified, it must be a stream. Any error diagnostics are sent to that
stream.
.TP* Examples:
.verb
;; the equivalent of #/[a-zA-Z0-9_]/
(regex-compile '(set (#\ea . #\ez) (#\eA . #\eZ)
(#\e0 . #\e9) #\e_))
;; the equivalent of #/.*/ and #/.+/
(regex-compile '(0+ wild))
(regex-compile '(1+ wild))
;; #/a|b|c/
(regex-compile '(or (or #\ea #\eb) #\ec))
;; string
(regex-compile "a|b|c")
.brev
.coNP Function @ regex-source
.synb
.mets (regex-source << regex )
.syne
.desc
The
.code regex-source
function returns the source code of compiled regular expression
.metn regex .
The source code isn't the textual notation, but the Lisp
data structure representing the abstract syntax tree: the
same representation as what is returned by
.codn regex-parse .
.coNP Function @ regex-parse
.synb
.mets (regex-parse < string <> [ error-stream ])
.syne
.desc
The
.code regex-parse
function parses a character string which contains a regular expression and
turns it into a Lisp data structure (the abstract syntax tree representation of
the regular expression).
The regular-expression syntax
.code #/RE/
produces the same structure, but as a
literal which is processed at the time \*(TX source code is read; the
.code regex-parse
function performs this parsing at run time.
If there are parse errors, the function returns
.codn nil .
The optional
.meta error-stream
argument specifies a stream to which error messages
are sent from the parser. By default, diagnostic output goes to the
.code *stdnull*
stream, which discards it. If
.meta error-stream
is specified as
.codn t ,
then the diagnostic output goes to the
.code *stdout*
stream.
If
.code regex-parse
returns a
.cod2 non- nil
value, that structure is then something
which is suitable as input to
.codn regex-compile .
There is a small difference in the syntax accepted by
.code regex-parse
and the syntax of regular-expression literals. Any
.code /
(slash) characters occurring in any position within
.meta string
are treated as ordinary characters, not as regular-expression delimiters.
The call
.mono
(regex-parse "/a/")
.onom
matches three characters: a slash, followed by the letter "a", followed
by another slash. Note that the slashes are not escaped.
Note: if a
.code regex-parse
call is written using a string literal as the
.meta string
argument, then note that any backslashes which are to be processed
by the regular expression must be doubled up, otherwise they belong
to the string literal:
.verb
(regex-parse "\e*") ;; error, invalid string literal escape
(regex-parse "\e\e*") ;; correct: the \e* literal match for *
.brev
The double backslash in the string literal produces a single backslash
in the resulting string object that is processed by
.codn regex-parse .
.coNP Function @ regex-optimize
.synb
.mets (regex-optimize << regex-tree-syntax )
.syne
.desc
The
.code regex-compile
function accepts the source code of a regular expression,
expressed as a Lisp data structure representing an abstract syntax tree,
and calculates an equivalent structure in which certain simplifications
have been performed, or in some cases substitutions which eliminate the
dependence on derivative-based processing.
The
.meta regex-tree-syntax
argument is assumed to be correct, as if it were produced by the
.code regex-parse
or
.code regex-from-trie
functions. Incorrect syntax produces unspecified results: an exception may be
thrown, or some object may appear to be successfully returned.
Note: it is unnecessary to call this function to prepare the input for
.code regex-compile
because that function optimizes internally. However, the source code attached
to a compiled regular-expression object is the original unoptimized syntax
tree, and that is used for rendering the
.code #/.../
notation when the object is printed. If the syntax is passed through
.code regex-optimize
before
.codn regex-compile ,
the resulting object will have the optimized code attached to it, and
subsequently render that way in printed form.
.TP* Examples:
.verb
;; a|b|c -> [abc]
(regex-optimize '(or #\ea (or #\eb #\ec))) -> (set #\ea #\eb #\ec)
;; (a|) -> a?
(regex-optimize '(or #\ea nil)) -> (? #\ea)
.brev
.coNP Function @ read-until-match
.synb
.mets (read-until-match < regex >> [ stream <> [ include-match ]])
.syne
.desc
The
.code read-until-match
function reads characters from
.metn stream ,
accumulating them into a string, which is returned.
If an argument is not specified for
.metn stream ,
then the
.code *stdin*
stream is used.
The
.meta include-match
argument is Boolean, indicating whether the delimiting text
matched by
.meta regex
is included in the returned string. It defaults to
.codn nil .
The accumulation of characters is terminated by a match on
.metn regex ,
the end of the stream, or an error.
This means that characters are read from the stream and accumulated while the
stream has more characters available, and while its prefix does not match
.metn regex .
If
.meta regex
matches the stream before any characters are accumulated,
then an empty string is returned.
If the stream ends or an non-exception-throwing error occurs before any
characters are accumulated, the function returns
.codn nil .
When the accumulation of characters terminates by a match on
.metn regex ,
the longest possible matching sequence of characters is
removed from the stream. If
.meta include-match
is true, that matching text is included in
the returned string. Otherwise, it is discarded.
The next available character in the stream is the first
nonmatching character following the matched text.
However, the next available character, as well as some number of
subsequent characters, may originate from the stream's push-back buffer,
rather than from the underlying operating system object,
due to this function's internal use of the
.code unget-char
function. Therefore, the stream position, as would be reported by
.codn seek-stream ,
is unspecified.
.coNP Functions @ scan-until-match and @ count-until-match
.synb
.mets (scan-until-match < regex <> [ stream ])
.mets (count-until-match < regex <> [ stream ])
.syne
.desc
The functions
.code scan-until-match
and
.code count-until-match
read characters from
.meta stream
until a match occurs in the stream for regular expression
.metn regex ,
the stream runs out of characters, or an error occurs.
If the stream runs out of characters, or a non-exception-throwing error
occurs, before a match for
.meta regex
is identified, these functions return
.codn nil .
If a match for
.meta regex
occurs in
.metn stream ,
then
.code count-until-match
returns the number of characters that were read and discarded prior to
encountering the first matching character.
In the same situation, the
.code scan-until-match
function returns a
.code cons
cell whose
.code car
holds the count of discarded characters, that being the same value as what
would be returned by
.codn count-until-match ,
and whose
.code cdr
holds a character string that comprises the text matched by
.metn regex .
The text matched by
.meta regex
is as long as possible, and is removed from the stream.
The next available character in the stream is the first
nonmatching character following the matched text.
However, the next available character, as well as some number of
subsequent characters, may originate from the stream's push-back buffer,
rather than from the underlying operating system object,
due to these functions' internal use of the
.code unget-char
function. Therefore, the stream position, as would be reported by
.codn seek-stream ,
is unspecified.
.coNP Functions @, m^$ @ m^ and @ m$
.synb
.mets (m^$ < regex <> [ position ] << string )
.mets (m^ < regex <> [ position ] << string )
.mets (m$ < regex <> [ end-position ] << string )
.syne
.desc
These functions provide functionality similar to the
.code match-regst
and
.code match-regst-right
functions, but under alternative interfaces which are more
convenient.
The
.code ^
and
.code $
notation used in their names are an allusion to the
regular-expression search-anchoring operators found in
familiar POSIX utilities such as
.codn grep .
The
.meta position
argument, if omitted,
defaults to zero, so that the
entire
.meta string
is operated upon.
The
.meta end-position
argument defaults to the length of
.metn string ,
so that the end position coincides with the end of the
string.
If the
.meta position
or
.meta end-position
arguments are negative, they index backwards
from the length of
.meta string
so that -1 denotes the last character.
A value in either parameter which is excessively
negative or positive, such that it indexes before
the start of the string or exceeds its length
results in a failed match and consequently
.code nil
being returned.
The
.code m^$
function tests whether the entire portion of
.meta string
starting at
.meta position
through to the end of the string is in the set of strings
matched by
.metn regex .
If this is true, then that portion of the string is
returned. Otherwise
.code nil
is returned.
The
.code m^
function tests whether the portion of the
.meta string
starting at
.meta position
has a prefix which matches
.metn regex .
If so, then this matching prefix is returned.
Otherwise
.code nil
is returned.
The
.code m$
function tests whether the portion of
.meta string
ending just before
.meta end-position
has a suffix which matches
.metn regex .
If so, then this matching suffix is returned.
Otherwise
.code nil
is returned.
.coNP Functions @, r^$ @, r^ @ r$ and @ rr
.synb
.mets (r^$ < regex <> [ position ] << string )
.mets (r^ < regex <> [ position ] << string )
.mets (r$ < regex <> [ end-position ] << string )
.mets (rr < regex >> [ position <> [ from-end ]] << string )
.syne
.desc
The first three of these functions perform the same operations as,
respectively,
.codn m^$ ,
.code m^
and
.codn m$ ,
with the same argument conventions. They differ
in return value. When a match is found, they
return a range value indicating the extent of
the matching substring within
.meta string
rather than the matching substring itself.
The
.code rr
function performs the same operation as
.code range-regex
with different conventions with regard to argument
order, harmonizing with those of the other three functions above.
The
.meta position
argument, if omitted,
defaults to zero, so that the
entire
.meta string
is operated upon.
The
.meta end-position
argument defaults to the length of
.metn string ,
so that the end position coincides with the end of the
string.
With one exception, a value in either parameter which is excessively negative
or positive, such that it indexes before the start of the string or exceeds its
length results in a failed match and consequently
.code nil
being returned. The exception is that the
.code rr
function permits a negative
.meta position
value which refers before the start of the string; this is effectively
treated as zero.
The
.meta from-end
argument defaults to
.codn nil .
The
.code r^$
function tests whether the entire portion of
.meta string
starting at
.meta position
through to the end of the string is in the set of strings
matched by
.metn regex .
If this is true, then the matching range is returned,
as a range object.
The
.code r^
function tests whether the portion of the
.meta string
starting at
.meta position
has a prefix which matches
.metn regex .
If so, then the matching range is returned, as a range object.
Otherwise
.code nil
is returned.
The
.code r$
function tests whether the portion of
.meta string
ending just before
.meta end-position
has a suffix which matches
.metn regex .
If so, then the matching range is returned.
Otherwise
.code nil
is returned.
The
.code rr
function searches
.meta string
starting at
.meta position
for a match for
.codn regex .
If
.meta from-end
is specified and true, the rightmost
match is reported.
If a match is found, it is reported
as a range.
A regular expression which matches empty
strings matches at the start position,
and every other position, including
the position just after the last
character, coinciding with the length of
.metn string .
Except for the different argument order such that
.meta string
is always the rightmost argument, the
.code rr
function is equivalent to the
.code range-regex
function, such that correspondingly named
arguments have the same semantics.
.coNP Function @ rra
.synb
.mets (rra < regex >> [ start <> [ end ]] << string )
.syne
.desc
The
.code rra
function searches
.meta string
between the
.meta start
and
.meta end
position for matches for the regular expression
.metn regex .
The matches are returned as a list of range objects.
The
.meta start
argument defaults to zero, and
.meta end
defaults to the length of the string (the position one
past the last character).
Negative values of
.meta start
and
.meta end
indicate positions from the end of the string, such that -1
denotes the last character, -2 the second-to-last and so forth.
If
.meta start
is so negative that it refers before the start of
.metn string ,
it is treated as zero. If this situation is true of the
.meta end
argument, then the function returns
.codn nil .
If
.meta start
refers to a character position beyond the length of
.meta string
(two characters or more beyond the end of the string),
then the function returns
.codn nil .
If this situation is true of
.metn end ,
then
.meta end
is curtailed to the string length.
The
.code rra
function returns all non-overlapping matches, including
zero length matches. Zero length matches may occur before
the first character of the string, or after the last
character. If so, these are included.
.coNP Functions @, f^$ @ f^ and @ f$
.synb
.mets (f^$ < regex <> [ position ])
.mets (f^ < regex <> [ position ])
.mets (f$ < regex <> [ end-position ])
.syne
.desc
These regular-expression functions do not directly
perform regex operations. Rather, they each return
a function of one argument which performs a regex
operation.
The returned functions perform the same operations as,
respectively,
.codn m^$ ,
.code m^
and
.codn m$ .
The following equivalences nearly hold, except that the functions
on the right side produced by
.code op
can accept two arguments when only
.code r
is curried, whereas the functions on the left take only
one argument:
.verb
[f^$ r] <--> (op m^$ r)
[f^$ r p] <--> (op m^$ r p)
[f^ r] <--> (op m^ r)
[f^ r p] <--> (op m^ r p)
[f$ r] <--> (op m$ r)
[f$ r p] <--> (op m$ r p)
.brev
That is to say,
.code f^$
returns a function which binds
.meta regex
and possibly the optional
.metn position .
When this function is invoked, it must be given an argument
which is a string. It performs the same operation as
.code m^$
being called on
.meta regex
and possibly
.metn position .
The same holds between
.code f^
and
.codn m^ ,
and between
.code f$
and
.codn m$ .
.TP* Examples:
.verb
;; produce list which contains only strings
;; beginning with "cat":
(keep-if (f^ #/cat/) '#"dog catalog cat fox catapult")
--> ("catalog" "cat" "catapult")
;; map all strings in a list to just their trailing
;; digits.
(mapcar (f$ #/\ed*/) '#"a123 4 z bc465")
--> ("123" "4" "" "465")
;; check that all strings consist of digits after
;; the third position.
(all '#"ABC123 DFE45 12379" (f^$ #/\ed*/ 3))
--> "79" ; i.e. true
(all '#"ABC123 DFE45 12379A" (f^$ #/\ed*/ 3))
--> nil
.brev
.coNP Functions @, fr^$ @, fr^ @ fr$ and @ frr
.synb
.mets (fr^$ < regex <> [ position ])
.mets (fr^ < regex <> [ position ])
.mets (fr$ < regex <> [ end-position ])
.mets (frr < regex <> [[ start-position ] << from-end ])
.syne
.desc
These regular-expression functions do not directly
perform regex operations. Rather, they each return
a function of one argument which performs a regex
operation.
The returned functions perform the same operations as,
respectively,
.codn r^$ ,
.codn r^ ,
.code r$
and
.codn rr .
The following equivalences nearly hold, except that some of the
functions on the right side produced by op
.code op
can accept additional arguments after the input string,
whereas the functions on the left produced by
.code f^$
et al.
accept only one parameter: the input string.
.verb
[fr^$ r] <--> (op m^$ r)
[fr^$ r p] <--> (op m^$ r p)
[fr^ r] <--> (op m^ r)
[fr^ r p] <--> (op m^ r p)
[fr$ r] <--> (op m$ r)
[fr$ r p] <--> (op m$ r p)
[frr r] <--> (op m$ r)
[frr r s] <--> (op m$ r s)
[frr r s fe] <--> (op m$ r s fe)
.brev
That is to say,
.code fr^$
returns a function which binds
.meta regex
and possibly the optional
.metn position .
When this function is invoked, it must be given an argument
which is a string. It performs the same operation as
.code r^$
being called on
.meta regex
and possibly
.metn position ,
and the string.
The same holds between
.code fr^
and
.codn r^ ,
between
.code fr$
and
.codn r$ ,
and between
.code frr
and
.codn rr .
.TP* Examples:
.verb
;; Remove leading digits from "123A456",
;; other than first digit:
(regsub (fr^ #/\ed+/ 1) "" "123A456")
--> "1A456"
.brev
.SS* Hashing Library
A hash table is an object which retains an association between pairs of
objects. Each pair consists of a key and a value. Given an object which is
similar to a key in the hash table, it is possible to retrieve the
corresponding value. Entries in a hash table are not ordered in any way, and
lookup is facilitated by hashing: quickly mapping a key object to a numeric
value which is then used to index into one of many buckets where the matching
key will be found (if such a key is present in the hash table).
In addition to keys and values, a hash table contains a storage location
which allows it to be associated with user data.
Important to the operation of a hash table is the criterion by which keys are
considered same. By default, this similarity follows the
.code eql
function.
A hash table will search for a stored key which is
.code eql
to the given search key.
A hash table constructed with the
.codn equal -based
property compares keys using
the
.code equal
function instead.
In addition to storing key-value pairs, a hash table can have a piece of
information associated with it, called the user data.
\*(TX hash tables contain a seed value which permutes the hashing operation,
at least for keys of certain types. This feature, if the seed is randomized,
helps to prevent software from being susceptible to hash collision
denial-of-service attacks. However, by default, the seed is not randomized.
Newly created hash tables for which a seed value is not specified take their
seed value from the
.code *hash-seed*
special variable, which is initialized to zero. That includes hash tables
created by parsing hash literal syntax.
Security-sensitive programs
requiring protection against collision attacks may use
.code gen-hash-seed
to create a randomized hash seed, and, depending on their specific need, either
store that value in
.codn *hash-seed* ,
or pass the value to hash-table constructors like
.codn make-hash ,
or both.
Note: randomization of hash seeding isn't a default behavior because it affects
program reproducibility. The seed value affects the order in which keys are
traversed, which can change the output of programs whose inputs have not
changed, and whose logic is is otherwise deterministic.
A hash table can be traversed to visit all of the keys and data. The order of
traversal bears no relation to the order of insertion, or to any properties of
the key type.
During an open traversal, new keys can be inserted into a hash table or deleted
from it while a a traversal is in progress. Insertion of a new key during
traversal will not cause any existing key to be visited twice or to be skipped;
however, it is not specified whether the new key will be traversed. Similarly,
if a key is deleted during traversal, and that key has not yet been visited, it
is not specified whether it will be visited during the remainder of the
traversal. These remarks apply not only to deletion via
.code remhash
or the
.code del
operator, but also to wholesale deletion of all keys via
.codn clearhash .
The garbage collection of hash tables supports weak keys and weak values.
If a hash table has weak keys, this means that from the point of view
of garbage collection, that table holds only weak references to the keys
stored in it. Similarly, if a hash table has weak values, it means that it
holds a weak reference to each value stored. A weak reference is one
which does not prevent the reclamation of an object by the garbage
collector. That is to say, when the garbage collector discovers that the only
references to some object are weak references, then that object is considered
garbage, just as if it had no references to it. The object is reclaimed, and
the weak references "lapse" in some way, which depends on what kind they are.
Hash-table weak references lapse by entry removal. When an object used
as a key in one or more weak-key hash tables becomes unreachable, those hash
entries disappear. This happens even if the values are themselves reachable.
Vice versa, when an object appearing as a value in one or more weak-value hash
tables becomes unreachable, those entries disappear, even if the keys are
reachable. When a hash table has both weak keys and weak values, then an
the behavior is one of two possible semantics. Under the
.codn or -semantics,
the hash table entry is removed if either the key or the value is unreachable.
Under the
.codn and -semantics,
the entry is removed only if both the key and value are unreachable.
If the keys of a weak-key hash table are reachable from the values, or if the
values of a weak-key hash table are reachable from the keys, then the weak
semantics is defeated for the affected entries: the hash table retains those
entries as if it were an ordinary table. A hash table with both weak keys and
values does not have this issue, regardless of its semantics.
An open traversal of a hash table is performed by the
.code maphash
function and the
.code dohash
operator. The traversal is open because code supplied by the program
is evaluated for each entry.
The functions
.codn hash-keys ,
.codn hash-values ,
.codn hash-pairs ,
and
.code hash-alist
also perform an open traversal, because they return
lazy lists. The traversal isn't complete until the returned lazy list
is fully instantiated. In the meanwhile, the
\*(TX program can mutate the hash table from which the lazy list
is being generated.
Certain hash operations expose access to the internal key-value association
entries of a hash table, which are represented as ordinary
.code cons
cells. Modifying the
.code car
field of such a cell potentially violates the integrity of the hash table;
the behavior of subsequent lookup and insertion operations becomes unspecified.
Similarly, if an object is used as a key in an
.codn equal -based
hash table, and that object is mutated in such a way that its equality to
other objects under the
.code equal
function is affected or its hash value under
.code hash-equal
is altered, the behavior of subsequent lookup and insertion operations on the
becomes unspecified.
.coNP Functions @ make-hash and @ hash
.synb
.mets (make-hash < weak-keys < weak-vals
.mets \ \ \ \ \ \ \ \ \ \ < equal-based <> [ hash-seed ])
.mets (hash {:weak-keys | :weak-vals | :weak-or | :weak-and
.mets \ \ \ \ \ \ :eql-based | :equal-based |
.mets \ \ \ \ \ \ :eq-based | :userdata << obj }*)
.syne
.desc
These functions construct a new hash table.
.code make-hash
takes three mandatory Boolean arguments. The Boolean
.meta weak-keys
argument specifies whether the hash table shall have weak keys. The
.meta weak-vals
argument specifies whether it shall have weak values, and
.meta equal-based
specifies whether it is
.codn equal -based.
If the
.meta weak-keys
argument is one of the keywords
.code :weak-and
or
.code :weak-or
then the hash table shall have both weak keys and weak values, with the
semantics implied by the keyword:
.code :weak-and
specifies
.codn and -semantics
and
.code :weak-or
specifies
code or -semantics.
The
.meta weak-vals
argument is then ignored.
If both
.meta weak-keys
and
.meta weak-vals
are true, and
.meta weak-keys
is not one of the keywords
.code :weak-and
or
.codn :weak-or ,
then the hash table has
.codn or -semantics.
The
.code hash
function defaults all three of these properties to false,
and allows them to be overridden to true
by the presence of keyword arguments.
The optional
.meta hash-seed
parameter must be an integer, if specified. Its value perturbs the hashing
function of the hash table, which affects
.code :equal-based
hash tables, when character strings and buffers are used as keys.
If
.meta hash-seed
is omitted, then the value of the
.code *hash-seed*
variable is used as the seed.
It is an error to attempt to construct an
.codn equal -based
hash table which has weak keys.
The
.code hash
function provides an alternative interface. It accepts optional
keyword arguments. The supported keyword symbols are:
.codn :weak-keys ,
.codn :weak-vals ,
.codn :weak-and ,
.codn :weak-or ,
.codn :equal-based ,
.code :eql-based
.code :eq-based
and
.code :userdata
which can be specified in any order to turn on the corresponding properties in
the newly constructed hash table.
Only one of
.codn :equal-based ,
.code :eql-based
and
.code :eq-based
may be specified. If specified, then the hash table uses
.codn equal ,
.code eql
or
.code eq
equality, respectively, for considering two keys to be the same key.
If none of these is specified, the
.code hash
function produces an
.code :equal-based
hash table by default.
If
.codn :weak-keys ,
.code :weak-and
or
.code :weak-or
is specified, then
.code :equal-based
may not be specified.
At most one of
.code :weak-and
or
.code :weak-or
may be specified. If either of these is specified, then the
.code :weak-keys
and
.code :weak-vals
keywords are redundant and unnecessary.
If
.code :weak-keys
and
.code :weak-vals
are both specified, and
.code :weak-and
isn't specified, the situation is equivalent to
.codn :weak-or .
If
.code :userdata
is present, it must be followed by an argument value; that value
specifies the user data for the hash table, which can be retrieved using the
.code hash-userdata
function.
Note: there doesn't exist a keyword for specifying the seed.
This omission is deliberate. These hash construction keywords may appear in the
hash literal
.code #H
syntax. A seed keyword would allow literals to specify their own seed, which
would allow malicious hash literals to be crafted that perpetrate a hash
collision attack against the parser.
.coNP Functions @, hash-construct @ hash-from-pairs and @ hash-from-alist
.synb
.mets (hash-construct < hash-args << key-val-pairs )
.mets (hash-from-pairs < key-val-pairs << hash-arg *)
.mets (hash-from-alist < alist << hash-arg *)
.syne
.desc
The
.code hash-construct
function constructs a populated hash in one step. The
.meta hash-args
argument specifies a list suitable as an argument list in a call to the
.code hash
function. The
.meta key-val-pairs
is a sequence of pairs, which are two-element
lists representing key-value pairs.
A hash is constructed as if by a call to
.mono
.meti (apply hash << hash-args ),
.onom
then populated
with the specified pairs, and returned.
The
.code hash-from-pairs
function is an alternative interface to the same semantics. The
.meta key-val-pairs
argument is first, and the
.meta hash-args
are passed as trailing variadic arguments, rather than a single list argument.
The
.code hash-from-alist
function is similar to
.codn hash-from-pairs ,
except that the
.meta alist
argument specifies they keys and values as an association list.
The elements of the list are
.code cons
cells, each of whose
.code car
is a key, and whose
.code cdr
is the value.
.coNP Function @ hash-list
.synb
.mets (hash-list < key-list << hash-arg *)
.syne
.desc
The
.code hash-list
function constructs a hash as if by a call to
.mono
.meti [apply hash << hash-args ],
.onom
where
.meta hash-args
is a list of the individual
.meta hash-arg
variadic arguments.
The hash is then populated with keys taken from
.meta key-list
and returned.
The value associated with each key is that key itself.
.coNP Function @ hash-zip
.synb
.mets (hash-zip < key-seq < value-seq << hash-arg *)
.syne
.desc
The
.code hash-zip
function constructs a hash as if by a call to
.mono
.meti (apply hash << hash-args ),
.onom
where
.meta hash-args
is a list of the individual
.meta hash-arg
variadic arguments.
The hash is then populated with keys taken from
.meta key-seq
which are paired with values taken from from
.metn value-seq ,
and returned.
If
.meta key-seq
is longer than
.metn value-seq ,
then the excess keys are ignored, and vice versa.
.coNP Function @ hash-update
.synb
.mets (hash-update < hash << function )
.syne
.desc
The
.code hash-update
function replaces each value in
.metn hash ,
with the value of
.meta function
applied to that value.
The return value is
.metn hash .
.coNP Function @ hash-update-1
.synb
.mets (hash-update-1 < hash < key < function <> [ init ])
.syne
.desc
The
.code hash-update-1
function operates on a single entry in the hash table.
If
.meta key
exists in the hash table, then its corresponding value is passed
into
.metn function ,
and the return value of
.meta function
is then installed
in place of the key's value. The value is then returned.
If
.meta key
does not exist in the hash table, and no
.meta init
argument is given,
then
.code hash-update-1
does nothing and returns
.codn nil .
If
.meta key
does not exist in the hash table, and an
.meta init
argument is given,
then
.meta function
is applied to
.metn init ,
and then
.meta key
is inserted into
.meta hash
with the value returned by
.meta function
as the datum. This value
is also returned.
.coNP Function @ group-by
.synb
.mets (group-by < func < sequence << option *)
.syne
.desc
The
.code group-by
function produces a hash table from
.metn sequence ,
which is a
list or vector. Entries of the hash table are not elements of
.metn sequence ,
but lists of elements of
.metn sequence .
The function
.meta func
is applied to
each element of
.meta sequence
to compute a key. That key is used to determine
which list the item is added to in the hash table.
The trailing arguments
.mono
.meti << option *
.onom
if any, consist of the same keywords that are understood by the
.code hash
function, and determine the properties of the hash.
.TP* Example:
Group the integers from 0 to 10 into three buckets keyed on 0, 1 and 2
according to the modulo 3 congruence:
.verb
(group-by (op mod @1 3) (range 0 10)))
-> #H(() (0 (0 3 6 9)) (1 (1 4 7 10)) (2 (2 5 8)))
.brev
.coNP Function @ group-reduce
.synb
.mets (group-reduce < hash < classify-fun < binary-fun < seq
.mets \ \ >> [ init-value <> [ filter-fun ]])
.syne
.desc
The
.code group-reduce
updates hash table
.meta hash
by grouping and reducing sequence
.metn seq .
The function regards the hash table as being populated with
keys denoting accumulator values. Missing accumulators which
need to be created in the hash table are initialized with
.meta init-value
which defaults to
.codn nil .
The function iterates over
.meta seq
and treats each element according to the following steps:
.RS
.IP 1.
Each element is mapped to a hash key through
.metn classify-fun .
.IP 2.
The value associated with the hash key (the accumulator for that
key) is retrieved. If it doesn't exist,
.meta init-value
is used.
.IP 3.
The function
.meta binary-fun
is invoked with two arguments: the accumulator from step 2, and the
original element from
.metn seq .
.IP 4.
The resulting value from step 3 is stored back into the hash table under the
key from step 2.
.RE
.IP
After the above processing, one more step is performed if the
.meta filter-fun
argument is present. In this case, the hash table is destructively mapped through
.meta filter-fun
before being returned. That is to say, every value in the hash table is
projected through
.meta filter-fun
and stored back in the table under the same key, as if by an invocation the
.mono
.meti (hash-update < hash << filter-fun )
.onom
expression.
.IP
If
.code group-reduce
is invoked on an empty hash table, its net result closely resembles a
.code group-by
operation followed by separately performing a
.code reduce-left
on each value in the hash.
.TP* Examples:
Frequency histogram:
.verb
[group-reduce (hash) identity (do inc @1)
"fourscoreandsevenyearsago" 0]
--> #H(() (#\ea 3) (#\ec 1) (#\ed 1) (#\ee 4) (#\ef 1)
(#\eg 1) (#\en 2) (#\eo 3) (#\er 3) (#\es 3)
(#\eu 1) (#\ev 1) (#\ey 1))
.brev
Separate the integers 1\(en10 into even and odd, and sum these groups:
.verb
[group-reduce (hash) evenp + (range 1 10) 0]
-> #H(() (t 30) (nil 25))
.brev
.coNP Functions @ make-similar-hash and @ copy-hash
.synb
.mets (make-similar-hash << hash )
.mets (copy-hash << hash )
.syne
.desc
The
.code make-similar-hash
and
.code copy-hash
functions create a new hash object based on
the existing
.meta hash
object.
.code make-similar-hash
produces an empty hash table which inherits all of the
attributes of
.metn hash .
It uses the same kind of key equality, the
same configuration of weak keys and values, and has the same user data (see
the
.code set-hash-userdata
function).
The
.code copy-hash
also produces a hash table similar to
.metn hash ,
in the same way as
.codn make-similar-hash .
However, rather than producing producing an empty hash table, it returns a
duplicate table which has all the same elements as
.metn hash :
it contains the same key and value objects.
.coNP Function @ inhash
.synb
.mets (inhash < hash < key <> [ init ])
.syne
.desc
The
.code inhash
function searches hash table
.meta hash
for
.metn key .
If
.meta key
is found, then it return the hash table's cons cell which
represents the association between
.meta hash
and
.metn key .
Otherwise, it returns
.codn nil .
If argument
.meta init
is specified, then the function will create
an entry for
.meta key
in
.meta hash
whose value is that of
.metn init .
The cons cell representing that association is returned.
Note: for as long as the
.meta key
continues to exist inside
.metn hash .
modifying the
.code car
field of the returned cons has ramifications for the logical integrity of
the hash; doing so results in unspecified behavior for subsequent
insertion and lookup operations.
Modifying the
.code cdr
field has the effect of updating the association with a new value.
.coNP Accessor @ gethash
.synb
.mets (gethash < hash < key <> [ alt ])
.mets (set (gethash < hash < key <> [ alt ]) << new-value )
.syne
.desc
The
.code gethash
function searches hash table
.meta hash
for key
.metn key .
If the
key is found then the associated value is returned. Otherwise, if
the
.meta alt
argument was specified, it is returned. If the
.meta alt
argument
was not specified,
.code nil
is returned.
A valid
.code gethash
form serves as a place. It denotes either an existing value in a hash
table or a value that would be created by the evaluation of the form.
The
.meta alt
argument is meaningful when
.code gethash
is used as a place, and, if present, is always evaluated whenever the place is
evaluated.
In place update operations, it provides the initial value, which defaults
to
.code nil
if the argument is not specified. For example
.code "(inc (gethash h k d))"
will increment the value stored under key
.code k
in hash table
.code h
by one. If the key does not exist in the hash table, then
the value
.code "(+ 1 d)"
is inserted into the table under that key.
The expression
.code d
is always evaluated, whether or not its value is needed.
If a
.code gethash
place is subject to a deletion, but doesn't exist, it is not an error.
The operation does nothing, and
.code nil
is considered the prior value of the place yielded
by the deletion.
.coNP Function @ sethash
.synb
.mets (sethash < hash < key << value )
.syne
.desc
The
.code sethash
function places a value into
.meta hash
table under the given
.metn key .
If a similar key already exists in the hash table, then that key's
value is replaced by
.metn value .
Otherwise, the
.meta key
and
.meta value
pair is
newly inserted into
.metn hash .
The
.code sethash
function returns the
.meta value
argument.
.coNP Function @ pushhash
.synb
.mets (pushhash < hash < key << element )
.syne
.desc
The
.code pushhash
function is useful when the values stored in a hash table
are lists. If the given
.meta key
does not already exist in
.metn hash ,
then a list of
length one is made which contains
.metn element ,
and stored in
.meta hash
table under
.metn key .
If the
.meta key
already exists in the hash table, then the corresponding
value must be a list. The
.meta element
value is added to the front of that list,
and the extended list then becomes the new value under
.metn key .
The return value is Boolean. If true, indicates that the hash-table entry was
newly created. If false, it indicates that the push took place on an existing
entry.
.coNP Function @ remhash
.synb
.mets (remhash < hash << key )
.syne
.desc
The
.code remhash
function searches
.meta hash
for a key similar to the
.metn key .
If that key is found, then that key and its corresponding value are
removed from the hash table.
If the key is found and removal takes place, then the associated value
is returned. Otherwise
.code nil
is returned.
.coNP Function @ clearhash
.synb
.mets (clearhash << hash )
.syne
.desc
The
.code clearhash
function removes all key-value pairs from
.metn hash ,
causing it to be empty.
If
.meta hash
is already empty prior to the operation, then
.codn nil ,
is returned.
Otherwise an integer is returned indicating the number of entries
that were purged from
.metn hash .
.coNP Function @ hash-count
.synb
.mets (hash-count << hash )
.syne
.desc
The
.code hash-count
function returns an integer representing the number of
key-value pairs stored in
.metn hash .
.coNP Accessor @ hash-userdata
.synb
.mets (hash-userdata << hash )
.mets (set (hash-userdata << hash ) << new-value )
.syne
.desc
The
.code hash-userdata
function retrieves the user data object associated with
.metn hash .
A hash table can be created with user data using the
.code :userdata
keyword in a hash-table literal or in a call to the
.code hash
function, directly, or via other hash-constructing functions which take the
hash construction keywords, such as
.codn group-by .
If a hash table is created without user data, its user
data is initialized to
.codn nil .
Because
.code hash-userdata
is an accessor, a
.code hash-userdata
form can be used as a place. Assigning a value to this place
causes the user data of
.meta hash
to be replaced with that value.
.coNP Function @ get-hash-userdata
.synb
.mets (get-hash-userdata << hash )
.syne
.desc
The
.code get-hash-userdata
function is a deprecated synonym for
.codn hash-userdata .
.coNP Function @ set-hash-userdata
.synb
.mets (set-hash-userdata < hash << object )
.syne
.desc
The
.code set-hash-userdata
replaces, with the
.metn object ,
the user data object
associated with
.metn hash .
.coNP Function @ hashp
.synb
.mets (hashp << object )
.syne
.desc
The
.code hashp
function returns
.code t
if the
.meta object
is a hash table,
otherwise it returns
.codn nil .
.coNP Function @ maphash
.synb
.mets (maphash < binary-function << hash )
.syne
.desc
The
.code maphash
function successively invokes
.meta binary-function
for each entry stored in
.metn hash .
Each entry's key and value are passed as arguments
to
.metn binary-function .
The function returns
.codn nil .
.coNP Functions @ hash-revget and @ hash-keys-of
.synb
.mets (hash-revget < hash < value >> [ testfun <> [ keyfun ]])
.mets (hash-keys-of < hash < value >> [ testfun <> [ keyfun ]])
.syne
.desc
The
.code hash-revget
function performs a reverse lookup on
.metn hash .
It searches through the entries stored in
.meta hash
for an entry whose value matches
.metn value .
If such an entry is found, that entry's key is returned.
Otherwise
.code nil
is returned.
If multiple matching entries exist, it is not specified which entry's
key is returned.
The
.code hash-keys-of
function has exactly the same argument conventions, and likewise
searches the
.metn hash .
However, it returns a list of all keys whose values match
.metn value .
The
.meta keyfun
function is applied to each value in
.meta hash
and the resulting value is compared with
.metn value .
The default
.meta keyfun
is the
.code identity
function.
The comparison is performed using
.metn testfun .
The default
.meta testfun
is the
.code equal
function.
.coNP Function @ hash-invert
.synb
.mets (hash-invert < hash >> [ joinfun >> [ unitfun << hash-arg *]])
.syne
.desc
The
.code hash-invert
function calculates and returns an inversion of hash table
.metn hash .
The values in
.meta hash
become keys in the returned hash table. Conversely, the values
in the returned hash table are derived from the keys.
The optional
.meta joinfun
and
.meta unitfun
arguments must be functions, if they are given.
These functions determine the behavior of
.code hash-invert
with regard to duplicate values in
.meta hash
which turn into duplicate keys.
The
.meta joinfun
function must be callable with two arguments, and
.meta joinfun
must accept one argument.
If
.meta joinfun
is omitted, it defaults to the
.code identity*
function;
.meta unitfun
defaults to
.codn identity .
The
.code hash-invert
function constructs a hash table as if by a call to the
.code hash
function, passing the
.meta hash-arg
arguments which determine the properties of the newly created hash.
The new hash table is then populated by iterating over the key-value pairs of
.meta hash
and inserting them as follows:
The key from
.meta hash
is turned into a value
.meta v1
by invoking the
.meta unitfun
function on it, and taking the return value.
The value from
.meta hash
is used as a key to perform a lookup in the new hash table.
If no entry exists, then a new entry is created, whose value is
.metn v1 .
Otherwise if the entry already exists, then the value
.meta v0
of that entry is combined with
.meta v1
by calling the
.meta joinfun
on the arguments
.meta v0
and
.metn v1 .
The entry is updated with the resulting value.
The new hash table is then returned.
.TP* Examples:
.verb
;; Invert simple 1 to 1 table:
(hash-invert #H(() (a 1) (b 2) (c 3)))
--> #H(() (1 a) (2 b) (3 c))
;; Invert table such that the keys of duplicate values
;; are accumulated into lists:
[hash-invert #H(() (1 a) (2 a) (3 c) (5 c) (7 d)) append list]
--> #H(() (d (7)) (c (3 5)) (a (1 2)))
;; Invert table such that keys of duplicate values are summed:
[hash-invert #H(() (1 a) (2 a) (3 c) (5 c) (7 d)) +]
--> #H(() (d 7) (c 8) (a 3))
.brev
.coNP Functions @ hash-eql and @ hash-equal
.synb
.mets (hash-eql << object )
.mets (hash-equal < object <> [ hash-seed ])
.syne
.desc
These functions each compute an integer hash value from the internal
representation of
.metn object ,
which satisfies the following properties.
If two objects
.code A
and
.code B
are the same under the
.code eql
function, then
.code "(hash-eql A)"
and
.code "(hash-eql B)"
produce the same integer hash value. Similarly,
if two objects
.code A
and
.code B
are the same under the
.code equal
function, then
.code "(hash-equal A)"
and
.code "(hash-equal B)"
each produce the same integer hash value. In all other
circumstances, the hash values of two distinct objects are unrelated, and
may or may not be the same.
Object of struct type may support custom hashing by way of defining
an equality substitution via an
.code equal
method. See the Equality Substitution section under Structures.
The optional
.meta hash-seed
value perturbs the hashing function used by
.code hash-equal
for strings and buffer objects. This seed value must be a nonnegative integer
no wider than 64 bits: that is, in the range 0 to 18446744073709551615.
If the value isn't specified, it defaults to zero. On systems with 32-bit
addresses, only the low 32 bits of this value may be significant.
Effectively, each possible value of the significant part of the seed specifies
a different hashing function. If two objects
.code A
and
.code B
are the same under the
.code equal
function, then
.code "(hash-equal A S)"
and
.code "(hash-equal B S)"
each produce the same integer hash value for any valid seed value
.codn S .
.coNP Functions @, hash-keys @, hash-values @ hash-pairs and @ hash-alist
.synb
.mets (hash-keys << hash )
.mets (hash-values << hash )
.mets (hash-pairs << hash )
.mets (hash-alist << hash )
.syne
.desc
These functions retrieve the bulk key-value data of hash table
.meta hash
in various ways.
.code hash-keys
retrieves a list of the keys.
.code hash-values
retrieves a list of the values.
.code hash-pairs
retrieves a list of pairs,
which are two-element lists consisting of the key, followed by the value.
Finally,
.code hash-alist
retrieves the key-value pairs as a Lisp association list:
a list of cons cells whose
.code car
fields are keys, and whose
.code cdr
fields are the values. Note that
.code hash-alist
returns the actual entries from the hash table, which are
conses. Modifying the
.code cdr
fields of these conses constitutes modifying the hash values
in the original hash table. Modifying the
.code car
fields interferes with the integrity of the hash table,
resulting in unspecified behavior for subsequent hash insertion
and lookup operations.
These functions all retrieve the keys and values in the
same order. For example, if the keys are retrieved with
.codn hash-keys ,
and the values with
.codn hash-values ,
then the corresponding entries from
each list pairwise correspond to the pairs in
.metn hash .
The list returned by each of these functions is lazy, and hence constitutes
an open traversal of the hash table.
.coNP Operator @ dohash
.synb
.mets (dohash >> ( key-var < value-var < hash-form <> [ result-form ])
.mets \ \ << body-form *)
.syne
.desc
The
.code dohash
operator iterates over a hash table. The
.meta hash-form
expression must
evaluate to an object of hash-table type. The
.meta key-var
and
.meta value-var
arguments must be symbols suitable for use as variable names.
Bindings are established for these variables over the scope of the
.metn body-form s
and the optional
.metn result-form .
For each element in the hash table, the
.meta key-var
and
.meta value-var
variables are set to the key and value of that entry, respectively,
and each
.metn body-form ,
if there are any, is evaluated.
When all of the entries of the table are thus processed, the
.meta result-form
is evaluated, and its return value becomes the return value of the dohash form.
If there is no
.metn result-form ,
the return value is
.codn nil .
The
.meta result-form
and
.metn body-form s
are in the scope of an implicit anonymous
block, which means that it is possible to terminate the execution of
dohash early using
.mono
.meti (return << value )
.onom
or
.codn (return) .
.coNP Functions @, hash-uni @, hash-diff @ hash-symdiff and @ hash-isec
.synb
.mets (hash-uni < hash1 < hash2 >> [ joinfun >> [ map1fun <> [ map2fun ]]])
.mets (hash-diff < hash1 << hash2 )
.mets (hash-symdiff < hash1 << hash2 )
.mets (hash-isec < hash1 < hash2 <> [ joinfun ])
.syne
.desc
These functions perform basic set operations on hash tables in a nondestructive
way, returning a new hash table without altering the inputs. The arguments
.meta hash1
and
.meta hash2
must be compatible hash tables. This means that their keys
must use the same kind of equality.
The resulting hash table inherits attributes from
.metn hash1 ,
as if created by the
.code make-similar-hash
function. If
.meta hash1
has userdata, the resulting hash table
has the same userdata. If
.meta hash1
has weak keys, the resulting table has weak
keys, and so forth.
The
.code hash-uni
function performs a set union. The resulting hash contains all of
the keys from
.meta hash1
and all of the keys from
.metn hash2 ,
and their corresponding
values. If a key occurs both in
.meta hash1
and
.metn hash2 ,
then it occurs only once
in the resulting hash. In this case, if the
.meta joinfun
argument is not given,
the value associated with this key is the one from
.metn hash1 .
If
.meta joinfun
is specified then it is called with two arguments: the respective
data items from
.meta hash1
and
.metn hash2 .
The return value of this function is used
as the value in the union hash.
If
.meta map1fun
is specified it must be a function that can be called with one
argument. All values from
.meta hash1
are projected through this function: the function is applied
to each value, and the function's return value is used
in place of the original value.
Similarly, if
.meta map2fun
is present, specifies a function through which values from
.meta hash2
are projected.
The
.code hash-diff
function performs a set difference. First, a copy of
.meta hash1
is made as if by the
.code copy-hash
function. Then from this copy, all keys which occur
in
.code hash2
are deleted.
The
.code hash-symdiff
function performs a symmetric difference. A new hash is returned which
contains all of the keys from
.meta hash1
that are not in
.meta hash2
and vice versa:
all of the keys from
.meta hash2
that are not in
.metn hash1 .
The keys carry their corresponding values from
.meta hash1
and
.metn hash2 ,
respectively.
The
.code hash-isec
function performs a set intersection. The resulting hash contains
only those keys which occur both in
.meta hash1
and
.metn hash2 .
If
.meta joinfun
is not
specified, the values selected for these common keys are those from
.metn hash1 .
If
.meta joinfun
is specified, then for each key which occurs in both
.meta hash1
and
.metn hash2 ,
it is called with two arguments: the respective data items. The return
value is then used as the data item in the intersection hash.
.coNP Functions @ hash-subset and @ hash-proper-subset
.synb
.mets (hash-subset < hash1 << hash2 )
.mets (hash-proper-subset < hash1 << hash2 )
.syne
.desc
The
.code hash-subset
function returns
.code t
if the keys in
.meta hash1
are a subset of the keys in
.metn hash2 .
The
.code hash-proper-subset
function returns
.code t
if the keys in
.meta hash1
are a proper subset of the keys in
.metn hash2 .
This means that
.meta hash2
has all the keys which are in
.meta hash1
and at least one which isn't.
Note: the return value may not be mathematically meaningful if
.meta hash1
and
.meta hash2
use different equality. In any case, the actual behavior
may be understood as follows. The implementation of
.code hash-subset
tests whether each of the keys in
.meta hash1
occurs in
.meta hash2
using their respective equalities.
The implementation of
.code hash-proper-subset
applies
.code hash-subset
first, as above. If that is true, and the two hashes have the same number of
elements, the result is falsified.
.coNP Functions @, hash-begin @, hash-reset @ hash-next and @ hash-peek
.synb
.mets (hash-begin << hash )
.mets (hash-reset < hash-iter << hash )
.mets (hash-next << hash-iter )
.mets (hash-peek << hash-iter )
.syne
.desc
The
.code hash-begin
function returns a an iterator object capable of retrieving the
entries in stored in
.meta hash
one by one.
The
.code hash-reset
function changes the state of an existing iterator, such that it
becomes prepared to retrieve the entries stored in the newly given
.metn hash ,
which may be the same one as the previously associated hash.
In addition,
.code hash-reset
may be given a
.meta hash
argument of
.codn nil ,
which dissociates it from its hash table.
The
.code hash-next
function's
.meta hash-iter
argument is a hash iterator returned by
.codn hash-begin .
If unvisited entries remain in
.metn hash ,
then
.code hash-next
returns the next one as a cons cell whose
.code car
holds the key and whose
.code cdr
holds the value. That entry is then considered visited by the iterator.
If no more entries remain to be visited,
.code hash-next
returns
.codn nil .
The
.code hash-next
function also returns
.code nil
if the iterator has been dissociated from a hash table by
.codn hash-reset .
The
.code hash-peek
function returns the same value that a subsequent call to
.code hash-next
will return for the same
.metn hash-iter ,
without changing the state of
.metn hash-iter .
That is to say, if a cell representing a hash entry is returned, that entry
remains unvisited by the iterator.
.coNP Macro @ with-hash-iter
.synb
.mets (with-hash-iter >> ( isym < hash-form >> [ ksym <> [ vsym ]])
.mets \ \ << body-form *)
.syne
.desc
The
.code with-hash-iter
macro evaluates
.metn body-form s
in an environment in which a lexically scoped function is visible.
The function is named by
.meta isym
which must be a symbol suitable for naming functions with
.codn flet .
The
.meta hash-form
argument must be a form which evaluates to a hash-table object.
Invocations of the function retrieve successive entries of the hash table
as cons-cell pairs of keys and values. The function returns
.code nil
to indicate no more entries remain.
If either of the
.meta ksym
or
.meta vsym
arguments are present, they must be symbols suitable as variable names. They
are bound as variables visible to
.metn body-form s,
initialized to the value
.codn nil .
If
.meta ksym
is specified, then whenever the function
.meta isym
macro is invoked and retrieves a hash-table entry, the
.meta ksym
variable is set to the key. If the function returns
.code nil
then the value of
.meta ksym
is set to
.codn nil .
Similarly, if
.meta vsym
is specified, then the function stores the retrieved
hash value in that variable, or else sets the variable
to
.code nil
if there is no next value.
.coNP Special variable @ *hash-seed*
.desc
The
.code *hash-seed*
special variable is initialized with a value of zero. Whenever a new
hash table is explicitly or implicitly created, it takes its seed from
the value of the
.code *hash-seed*
variable in the current dynamic environment.
The only situation in which
.code *hash-seed*
is not used when creating a new hash table is when
.code make-hash
is called with an argument given for the optional
.meta hash-seed
argument.
Only
.codn equal -based
hash tables make use of their seed, and only for keys which are strings and
buffers. The purpose of the seed is to scramble the hashing function, to make
a hash table resistant to a type of denial-of-service attack, whereby a
malicious input causes a hash table to be populated with a large number of keys
which all map to the same hash-table chain, causing the performance to severely
degrade.
The value of
.code *hash-seed*
must be a nonnegative integer, no wider than 64 bits.
On systems with 32-bit addresses, only the least significant 32 bits of
this value may be significant.
.coNP Function @ gen-hash-seed
.synb
.mets (gen-hash-seed)
.syne
.desc
The
.code gen-hash-seed
function returns an integer value suitable for the
.code *hash-seed*
variable, or as the
.code hash-seed
argument of the
.code make-hash
and
.code hash-equal
functions.
The value is derived from the host environment, from information such
as the process ID and time of day.
.SS* Search Tree Library
\*(TL provides binary search trees, which are objects of type
.codn tree .
Trees have a printed notation denoted by the
.code #T
prefix. A tree may be constructed by invoking the
.code tree
function.
Binary search trees differ from hashes in that they maintain items in
order. They also differ from hashes in that they store only elements,
not key-value pairs. Every tree is associated with three
.IR "key abstraction functions" :
It has a
.I "key function"
which is applied to the elements to map each one to a key.
It also has a
.I "less function"
and
.I "equal function"
for comparing keys.
If these three functions are not specified, they respectively default to
.codn identity ,
.code less
and
.codn equal ,
which means that the tree uses its elements as keys directly, and
that they are compared using
.code less
and
.codn equal .
Note: these default functions work for simple elements such as character
strings or numbers, and also structures implementing
.IR "equality substitution" .
The elements are stored inside a tree using tree nodes, which are objects
of type
.codn tnode ,
whose printed notation is introduced by the
.code #N
prefix.
Several tree-related functions take
.code tnode
objects as arguments or return
.code tnode
objects.
Trees may store duplicate elements. The
.code #T
literal syntax may freely specify duplicate elements.
The
.code tree
constructor function specifies an initial sequence of elements to
be populated into the newly constructed tree. If this initial
sequence contains duplicate elements, they are preserved if the optional
.meta allow-dupes
argument is true, otherwise only the rightmost member of any duplicate
group appears in the tree.
The insertion functions
.code tree-insert
and
.code tree-insert-node
also overwrite duplicates by default, but optionally allow them.
Duplicates are ordered by insertion: most recently inserted duplicate
is rightmost. However, tree lookup chooses an unspecified duplicate.
.coNP Function @ tnode
.synb
.mets (tnode < key < left << right )
.syne
.desc
The
.code tnode
function allocates, initializes and returns a single tree node.
A tree node has three fields
.metn key ,
.meta left
and
.metn right ,
which are accessed using the functions
.codn key ,
.code left
and
.codn right .
.coNP Function @ tnodep
.synb
.mets (tnodep << value )
.syne
.desc
The
.code tnodep
function returns
.code t
if
.meta value
is a tree node. Otherwise, it returns
.codn nil .
.coNP Accessors @, key @ left and @ right
.synb
.mets (key << node )
.mets (left << node )
.mets (right << node )
.mets (set (key << node ) << new-key )
.mets (set (left << node ) << new-left )
.mets (set (right << node ) << new-right )
.syne
.desc
The
.codn key ,
.code left
and
.code right
functions retrieve the corresponding fields of the
.meta node
object, which must be of type
.codn tnode .
Forms based on the
.codn key ,
.code left
and
.code right
symbol are defined as syntactic places.
Assigning a value
.code v
to
.code "(key n)"
using the
.code set
operator, as in
.codn "(set (key n) v)" ,
is equivalent to
.code "(set-key n v)"
except that the value of the expression is
.code v
rather than
.codn n .
Similar statements hold true for
.code left
and
.code right
in relation to
.code set-left
and
.codn set-right .
.coNP Functions @, set-key @ set-left and @ set-right
.synb
.mets (set-key < node << new-key )
.mets (set-left < node << new-left )
.mets (set-right < node << new-right )
.syne
.desc
The
.codn set-key ,
.code set-left
and
.code set-right
functions replace the corresponding fields of
.meta node
with new values.
The
.meta node
argument must be of type
.codn tnode .
These functions all return
.metn node .
.coNP Function @ copy-tnode
.synb
.mets (copy-tnode << node )
.syne
.desc
The
.code copy-tnode
function creates a new
.code tnode
object, whose
.codn key ,
.code left
and
.code right
fields are copied from
.codn node .
.coNP Function @ tree
.synb
.mets (tree >> [ elems
.mets \ \ \ \ \ \ >> [ keyfun >> [ lessfun >> [ equalfun <> [ allow-dupes ]]]])
.syne
.desc
The
.code tree
function constructs and returns a new tree object. All arguments are optional.
The
.meta elems
argument specifies a sequence of the elements to be stored in the tree.
If the argument is absent or the sequence is empty, then an empty
tree is created.
The
.meta keyfun
argument specifies the function which is applied to every element
to produce a key. If omitted, the tree object shall behave as if the
.code identity
function were used, taking the elements themselves to be keys.
The
.meta lessfun
argument specifies the function by which two keys are compared for
inequality. If omitted, the
.code less
function is used. A function used as
.meta lessfun
should take two arguments, produce a Boolean result, and have ordering
properties similar to the
.code less
function.
The
.meta equalfun
argument specifies the function by which two keys are compared for
equality. The default value is the
.code equal
function. A function used as
.meta equalfun
should take two arguments, produce a Boolean result, and have the
properties of an equivalence relation.
These three functions are collectively referred to as the tree's
.IR "key abstraction functions" .
The
.meta allow-dupes
argument, which defaults to
.codn nil ,
is relevant if an
.meta elems
sequence is specified containing some elements which which appear to be
duplicates, according to the tree object's
.meta equalfun
function. If
.meta allow-dupes
is true then duplicates are preserved: the tree will have as many nodes as
there are elements in the
.meta elems
sequence. Moreover, the duplicates appear in the same relative order in
the tree as they appear in the original
.meta elems
sequence.
If
.meta allow-dupes
is false, then duplicates are suppressed: if any element appears more
than once in
.metn elements ,
then only the last occurrence of that element appears in the tree.
Note: the
.code tree-insert
and
.code tree-insert-node
functions also has an optional argument indicating whether an duplicate
insertion replaces an existing element.
Note: although the order of duplicate elements is preserved, when the
.code tree-lookup
function is used look up an key which is duplicated, the element
which is retrieved is unspecified, and can change when the tree is
reorganized due to insertions and deletions.
.coNP Function @ treep
.synb
.mets (treep << value )
.syne
.desc
The
.code treep
function returns
.code t
if
.meta value
is a tree. Otherwise, it returns
.codn nil .
.coNP Function @ tree-count
.synb
.mets (tree-count << tree )
.syne
.desc
The
.code tree-count
function returns an integer indicating the number of nodes currently
inserted into
.metn tree ,
which must be a search tree object.
.coNP Function @ tree-insert-node
.synb
.mets (tree-insert-node < tree < node <> [ allow-dupe ])
.syne
.desc
The
.code tree-insert-node
function inserts an existing
.meta node
object into a search tree.
The
.meta tree
object must be of type
.codn tree ,
and
.meta node
must be of type
.codn tnode .
The
.code key
field of the
.meta node
object holds the element that is being inserted. The actual search key
which is associated with this element is determined by applying
.metn tree 's
.meta keyfun
to the
.metn node 's
.code key
value.
The
.meta node
object must not currently be inserted into any existing tree.
The values stored in the
.code left
and
.code right
fields of
.meta node
are overwritten as required by the semantics of the insertion operation.
Their original values are ignored.
The
.meta allow-dupe
argument, defaulting to
.codn nil ,
is concerned with what happens if the tree already contains one or more
nodes having a key equal to the
.metn node 's
key.
.meta allow-dupe
is false, then
.meta node
replaces an unspecified one of those existing nodes: that replaced node is
deleted from the tree. Key equivalence is determined using tree's equality
function (see the
.meta equalfun
argument of the
.code tree
function).
If
.meta allow-dupe
is true, then the new node is inserted without replacing any node, and
appears together with the existing duplicate or duplicates. Among
the duplicates, the newly inserted node is the rightmost node in the
tree order.
The
.code tree-insert-node
function returns the
.meta node
argument.
.coNP Function @ tree-insert
.synb
.mets (tree-insert < tree < elem <> [ allow-dupe ])
.syne
.desc
The
.code tree-insert
function inserts
.meta elem
into
.metn tree .
The
.meta tree
argument must be an object of type
.codn tree .
The
.meta elem
value may be of any type which is semantically compatible with
.metn tree 's
key abstraction functions.
The
.code tree-insert
function allocates a new
.code tnode
as if by invoking
.mono
.meti (tnode < elem nil nil)
.onom
function, and inserts that
.code tnode
as if by using the
.code tree-insert-node
function.
If one or more elements equal to
.meta elem
already exist in the tree, then the behavior is determined by the
.meta allow-dupe
argument, which defaults to
.codn nil .
The semantics of
.meta allow-dupe
is as given in the description of
.codn tree-insert-node .
The
.code tree-insert
function returns the newly inserted
.code tnode
object.
.coNP Function @ tree-lookup-node
.synb
.mets (tree-lookup-node < tree << key )
.syne
.desc
The
.code tree-lookup-node
searches
.meta tree
for an element which matches
.metn key .
The
.meta tree
argument must be an object of type
.codn tree .
The
.meta key
argument may be a value of any type.
An element inside
.meta tree
matches
.meta key
if the tree's
.meta keyfun
applied to that element produces a key value which is equal to
.meta key
under the tree's
.meta equalfun
function.
If such an element is found, then
.code tree-lookup-node
returns the tree node which contains that element as its
.meta key
field.
If no such element is found, then
.code tree-lookup-node
returns
.codn nil .
If multiple nodes exist in the tree which have a matching key,
it is unspecified which one of those nodes is retrieved.
.coNP Function @ tree-lookup
.synb
.mets (tree-lookup < tree << key )
.syne
.desc
The
.code tree-lookup
function finds an element inside
.meta tree
which matches the given
.metn key .
If the element is found, it is returned. Otherwise,
.code nil
is returned.
Note: the semantics of the
.code tree-lookup
function can be understood in terms of
.codn tree-lookup-node .
A possible implementation is this:
.verb
(defun tree-lookup (tree key)
(iflet ((node (tree-lookup-node tree key)))
(key node)))
.brev
If the tree contains multiple elements which match
.metn key ,
it is unspecified which element is retrieved.
.coNP Function @ tree-delete-node
.synb
.mets (tree-delete-node < tree << key )
.syne
.desc
The
.code tree-delete-node
function searches
.meta tree
for an element which matches
.metn key .
The
.meta tree
argument must be an object of type
.codn tree .
The
.meta key
argument may be a value of any type which is semantically compatible with
.metn tree 's
key abstraction functions.
If the matching element is found, then its node is removed from
the tree, and returned.
Otherwise, if a matching element is not found, then
.code nil
is returned.
If more than one element exists inside
.meta tree
which matches
.metn key ,
it is unspecified which node is deleted and returned.
.coNP Function @ tree-delete
.synb
.mets (tree-delete < tree << key )
.syne
.desc
The
.code tree-delete
function tries to removes from
.meta tree
the element which matches
.metn key .
If successful, it returns that element, otherwise it returns
.codn nil .
If more than one element exists inside
.meta tree
which matches
.metn key ,
it is unspecified which one is deleted.
Note: the semantics of the
.code tree-delete
function can be understood in terms of
.codn tree-delete-node .
A possible implementation is this:
.verb
(defun tree-delete (tree key)
(iflet ((node (tree-delete-node tree key)))
(key node)))
.brev
.coNP Function @ tree-delete-specific-node
.synb
.mets (tree-delete-specific-node < tree << node )
.syne
.desc
The
.code tree-delete-specific-node
function searches
.meta tree
to find the specific node given by the
.meta node
argument. If
.meta node
is inserted into the tree, then it is deleted, and returned.
If
.meta node
is not found in the tree, then the tree is unchanged, and
.code nil
is returned.
Note: the search for
.meta node
is informed by
.metn node 's
key, for efficiency. However, if the tree contains duplicates of that key, then
a linear search takes place among the duplicates.
.coNP Functions @ tree-min-node and @ tree-min
.synb
.mets (tree-min-node << tree )
.mets (tree-min << tree )
.syne
.desc
The
.code tree-min-node
function returns the node in
.meta tree
which holds the lowest element. If the tree is empty, it returns
.codn nil .
The
.code tree-min
function returns the lowest element, or else
.code nil
if the tree is empty.
.coNP Functions @ tree-del-min-node and @ tree-del-min
.synb
.mets (tree-del-min-node << tree )
.mets (tree-del-min << tree )
.syne
.desc
The
.code tree-del-min-node
function returns the node in
.meta tree
which has the lowest key, and removes that node from the tree.
If the tree is empty, it returns
.codn nil .
The
.code tree-del-min
function returns the lowest element and removes it from the tree, or else
.code nil
if the tree is empty.
The following equivalence holds:
.verb
(tree-del-min tr) <--> (iflet ((node (tree-del-min-node tr)))
(key node))
.brev
Note:
.code tree-insert
together with
.code tree-del-min
provide the basis for using a tree as a priority queue. Elements are
inserted into the queue using
.code tree-insert
and then removed in priority order using
.codn tree-del-min .
.coNP Function @ tree-root
.synb
.mets (tree-root << tree )
.syne
.desc
The
.code tree-root
function returns the root node of
.metn tree ,
which must be a
.code tree
object.
If
.meta tree
is empty, then
.code nil
is returned.
.coNP Function @ tree-clear
.synb
.mets (tree-clear << tree )
.syne
.desc
The
.code tree-clear
function deletes all elements from
.metn tree ,
which must be a
.code tree
object.
If
.meta tree
is already empty, then the function returns
.codn nil ,
otherwise it returns an integer which gives the count of the number
of deleted nodes.
.coNP Function @ copy-search-tree
.synb
.mets (copy-search-tree << tree )
.syne
.desc
The
.code copy-search-tree
returns a new tree object which is a copy of
.metn tree .
The
.meta tree
argument must be an object of type
.codn tree .
The returned object has the same key abstraction functions as
.meta tree
and contains the same elements.
The nodes held inside the new tree are freshly allocated,
but their key objects are shared with the original tree.
.coNP Function @ make-similar-tree
.synb
.mets (make-similar-tree << tree )
.syne
.desc
The
.code copy-search-tree
returns a new, empty search tree object.
The
.meta tree
argument must be an object of type
.codn tree .
The returned object has the same key abstraction functions as
.metn tree .
.coNP Function @ tree-begin
.synb
.mets (tree-begin < tree >> [ low-key <> [ high-key ]])
.syne
.desc
The
.code tree-begin
function returns a new object of type
.code tree-iter
which provides in-order traversal of nodes stored in
.metn tree .
The
.meta tree
argument must be an object of type
.codn tree .
If the
.meta low-key
argument is specified, then nodes with keys lesser than
.meta low-key
are omitted from the traversal.
If the
.meta high-key
argument is specified, then nodes with keys equal to
or greater than
.meta high-key
are omitted from the traversal.
The nodes are traversed by applying the
.code tree-next
function to the returned
.code tree-iter
object.
A
.code tree-iter
object is iterable.
.TP* Example:
.verb
(collect-each ((el (tree-begin #T(() 1 2 3 4 5)
2 5)))
(* 10 el))
--> (20 30 40)
.brev
.coNP Function @ tree-reset
.synb
.mets (tree-reset < iter < tree >> [ low-key <> [ high-key ]])
.syne
.desc
The
.code tree-reset
functions is closely analogous to
.codn tree-begin .
The
.meta iter
argument must be an existing
.code tree-iter
object, previously returned by a call to
.codn tree-begin .
Regardless of its current state, the
.meta iter
object is re-initialized to traverse the specified
.meta tree
with the specified parameters, and is then returned.
The
.code tree-reset
function prepares
.meta iter
to traverse in the same manner as would new iterator returned by
.code tree-begin
for the specified
.metn tree ,
.meta low-key
and
.meta high-key
arguments.
.coNP Functions @ tree-next and @ tree-peek
.synb
.mets (tree-next << iter )
.mets (tree-peek << iter )
.syne
.desc
The
.code tree-next
and
.code tree-peek
function returns the next node in sequence from the tree iterator
.metn iter .
The iterator must be an object of type
.codn tree-iter ,
returned by the
.code tree-begin
function.
If there are no more nodes to be visited, these functions
.codn nil .
If, during the traversal of a tree, nodes are inserted or deleted,
the behavior of
.code tree-next
and
.code tree-peek
on
.code tree-iter
objects that were obtained prior to the insertion or deletion is
not specified. An attempt to complete the iteration may not successfully
visit all keys that should be visited.
The
.code tree-next
function changes the state of the iterator. If
.code tree-next
is invoked repeatedly on the same iterator, it returns successive
nodes of the tree.
If
.code tree-peek
is invoked more than once on the same iterator without any intervening calls to
.codn tree-next ,
it returns the same node; it does not appear to change the state of
the iterator and therefore does not advance through successive nodes.
.coNP Function @ sub-tree
.synb
.mets (sub-tree < tree >> [ from-key <> [ to-key ]])
.syne
.desc
The
.code sub-tree
function selects elements from
.metn tree ,
which must be a search tree.
If
.meta from-key
is specified, then elements lesser than
.meta from-key
are omitted from the selection.
If
.meta to-key
is specified, the elements greater than or equal to
.meta to-key
are omitted from the selection.
A list of the selected elements is returned, in which the elements appear in
the same order as they do in
.metn tree .
.coNP Function @ copy-tree-iter
.synb
.mets (copy-tree-iter << iter )
.syne
.desc
The
.code copy-tree-iter
function creates and returns a duplicate of the
.meta iter
object, which must be a tree iterator returned by
.codn tree-begin .
The returned object has the same state as the original; it references the same
traversal position in the same tree. However, it is independent of the original.
Calls to
.code tree-next
on the original have no effect on the duplicate and vice versa.
.coNP Function @ replace-tree-iter
.synb
.mets (replace-tree-iter < dest-iter << src-iter )
.syne
.desc
The
.code replace-tree-iter
function causes the tree iterator
.meta dest-iter
to be in the same state as
.metn src-iter .
Both
.meta dest-iter
and
.meta src-iter
must be tree iterator objects returned by
.codn tree-begin .
The contents of
.meta dest-iter
are updated such that it now references the same tree as
.metn src-iter ,
at the same position.
The
.meta dest-iter
argument is returned.
.coNP Special variable @ *tree-fun-whitelist*
.desc
The
.code *tree-fun-whitelist*
variable holds a list of function names
that may be used in the
.code #T
tree literal syntax as the
.metn keyfun ,
.meta lessfun
or
.meta equalfun
operations of a tree. The initial value of this variable is a list which
holds at least the following three symbols:
.codn identity ,
.code less
and
.codn equal .
The application may change the value of this variable, or dynamically
bind it, in order to allow
.code #T
literals to be processed which specify functions other than these three.
.SS* Partial Evaluation and Combinators
.coNP Macros @ op and @ do
.synb
.mets (op << form +)
.mets (do < oper << form *)
.syne
.desc
Like the
.code lambda
operator, the
.code op
macro denotes an anonymous function.
Unlike
.codn lambda ,
the arguments of the function are implicit, or
optionally specified within the expression, rather than as a formal
parameter list which precedes a body.
The
.meta form
arguments of
.code op
are implicitly turned into a DWIM expression,
which means that argument evaluation follows Lisp-1 rules. (See the
.code dwim
operator).
The argument forms of
.code op
are arbitrary expressions, within which special
conventions are permitted regarding the use of certain implicit variables:
.RS
.meIP >> @ num
A number preceded by a
.code @
is, syntactically, a metanumber. If it appears inside
.code op
as an expression, it behaves as a positional argument, whose
existence it implies. For instance
.code @2
means that the function shall have at least two arguments,
the second argument of which is be substituted in place of the
.codn @2 .
.code op
generates a function which has a number of required arguments equal to the
highest value of
.meta num
appearing in a
.mono
.meti >> @ num
.onom
construct in the body. For instance
.code "(op car @3)"
generates a three-argument function (which passes its third
argument to
.codn car ,
returning the result, and ignores its first two arguments).
There is no way to use
.code op
to generate functions which have optional arguments. The positional
arguments are mutable; they may be assigned.
.coIP @rest
If the meta-symbol
.code @rest
appears in the
.code op
syntax as an expression, it explicitly denotes and evaluates to the list of
trailing arguments. Like the metanumber positional arguments, it
may be assigned.
.coIP @rec
If the meta-symbol
.code @rec
appears in the
.code op
syntax as an expression, it denotes a mutable variable which is bound to the
function itself which is generated by that
.code op
expression.
.coIP "@(rec ...)"
If this syntax appears inside
.codn op ,
it specifies a recursive call to the function.
.RE
.IP
Functions generated by
.code op
are always variadic; they always take additional arguments after
any required ones, whether or not the
.code @rest
syntax is used.
If the body does not contain
any
.mono
.meti >> @ num
.onom
or
.code @rest
syntax, then
.code @rest
is implicitly inserted. What this means is that, for example, since
the form
.code "(op foo)"
does not contain any implicit positional arguments like
.codn @1 ,
and does not contain
.codn @rest ,
it is actually a shorthand for
.codn "(op foo . @rest)" :
a function which applies
.code foo
to all of its arguments.
If the body does contain at least one
.mono
.meti >> @ num
.onom
or
.codn @rest ,
then
.code @rest
isn't implicitly inserted. The notation
.code "(op foo @1)"
denotes a function which takes any number of arguments, and ignores
all but the first one, which is passed to
.codn foo .
The
.code do
operator is similar to
.codn op ,
with the following three differences:
.RS
.IP 1.
The first argument of
.codn do ,
namely
.metn oper ,
is an operator. This argument is not processed for the presence of
implicit variables. Thus for instance
.code "(do @1 ...)"
is invalid. By contrast,
.code "(op @1 ...)"
is possible, and makes sense under the right circumstances.
The
.meta oper
argument may be the name of a macro or special operator, whereas
.code op
doesn't support the invocation of macros or special operators.
For instance
.code "(do let ((x @1)) (+ x 1))"
is possible.
.IP 2.
The
.meta form
arguments of
.code do
are not implicitly treated as DWIM expressions,
but as ordinary expressions.
.IP 3.
When
.code do
syntax doesn't contain any references to implicit variables (metanumbers or
.codn @rest )
then a variadic function is generated which requires one argument.
That argument is added to the form. Thus for instance
.code "(do set x)"
effectively serves as a shorthand for
.codn "(do set x @1)" .
The corresponding defaulting behavior in
.code op
is that a variadic function is generated which requires no arguments.
All of the available arguments are applied. Thus
.code "(op f x)"
is effectively a shorthand for
.codn "(op f x . @rest)" .
.RE
.IP
Because it accepts operators,
.code do
can be used with imperative constructs
which are not functions, like
.codn set .
For example,
.code "(do set x)"
produces an anonymous function which, if called with one argument, stores that
argument into
.metn x .
The actions of
.code op
and
.code do
can be understood by the following examples,
which convey how the syntax is rewritten to lambda.
However, note that the real translator
uses generated symbols for the arguments, which are not equal to any
symbols in the program.
.verb
(op) -> invalid
(op +) -> (lambda rest [+ . rest])
(op + foo) -> (lambda rest [+ foo . rest])
(op @1 . @rest) -> (lambda (arg1 . rest) [arg1 . @rest])
(op @1 @rest) -> (lambda (arg1 . rest) [arg1 @rest])
(op @1 @2) -> (lambda (arg1 arg2 . rest) [arg1 arg2])
(op foo @1 (@2) (bar @3)) -> (lambda (arg1 arg2 arg3 . rest)
[foo arg1 (arg2) (bar arg3)])
(op foo @rest @1) -> (lambda (arg1 . rest) [foo rest arg1])
(do + foo) -> (lambda (arg1 . rest) (+ foo arg1))
(do @1 @2) -> (lambda (arg1 arg2 . rest) (@1 arg2)) ;; invalid!
(do foo @rest @1) -> (lambda (arg1 . rest) (foo rest arg1))
.brev
Note that if argument
.mono
.meti >> @ n
.onom
appears in the syntax, it is not necessary
for arguments
.code @1
through
.mono
.meti >> @ n-1
.onom
to appear. The function will have
.code n
arguments:
.verb
(op @3) -> (lambda (arg1 arg2 arg3 . rest) [arg3])
.brev
The
.code op
and
.code do
operators can be nested, in any combination. This raises the
question: if an expression like
.codn @1 ,
.code @rest
or
.code @rec
occurs in an
.code op
that is nested
within an
.codn op ,
what is the meaning?
An expression with a single
.code @
always belongs with the innermost
.code op
or
.code do
operator. So for instance
.code "(op (op @1))"
means that an
.code "(op @1)"
expression is nested
within an outer
.code op
expression that contains no references to its implicit variables.
The
.code @1
belongs to the inner
.codn op .
There is a way for an inner
.code op
to refer to the implicit variables of an outer one. This is
expressed by adding an extra
.code @
prefix for every level of escape. For example in
.code "(op (op @@1))"
the
.code @@1
belongs to the outer
.codn op :
it is the same as
.code @1
appearing in the outer
.codn op .
That is to say,
in the expression
.codn "(op @1 (op @@1))" ,
the
.code @1
and
.code @@1
are the same thing:
both are parameter 1 of the lambda function generated by the outer
.codn op .
By contrast, in the expression
.code "(op @1 (op @1))"
there are two different parameters:
the first
.code @1
is argument of the outer function, and the second
.code @1
is the first argument of the inner function. If there
are three levels of nesting, then three
.code @
meta-prefixes are needed to insert
a parameter from the outermost
.code op
into the innermost
.codn op .
Note that the implicit variables belonging to an
.code op
can be used in the dot position of a function call, such as:
.verb
[(op list 1 . @1) 2] -> (1 . 2)
.brev
This is a consequence of the special transformations described
in the paragraph
.B "Dot Position in Function Calls"
in the subsection
.B "Additional Syntax"
of the
.BR "TXR Lisp"
section.
The
.code op
syntax works in conjunction with quasiliterals which are nested within it.
The metanumber notation as well as
.code @rest
are recognized without requiring an additional
.code @
escape, which is effectively optional:
.verb
(apply (op list `@1-@rest`) '(1 2 3)) -> "1-2 3"
(apply (op list `@@1-@@rest`) '(1 2 3)) -> "1-2 3"
.brev
Though they produce the same result, the above two examples differ in that
.code @rest
embeds a metasymbol into the quasiliteral structure, whereas
.code @@rest
embeds the Lisp expression
.code @rest
into the quasiliteral. Either way, in the scope of
.codn op ,
.code @rest
undergoes the macro-expansion which renames it to the machine-generated
function argument symbol of the implicit function denoted by the
.code op
macro form.
This convenient omission of the
.code @
character isn't supported for reaching the arguments of an outer
.code op
from a quasiliteral within a nested
.codn op :
.verb
;; To reach @@1, @@@1 must be written.
;; @@1 Lisp expression introduced by @.
(op ... (op ... `@@@1`))
.brev
Because the
.code do
macro may be applied to operators, it is possible to apply it to itself,
as well as to
.codn op ,
as in the following example:
.verb
[[[[(do do do op list) 1] 2] 3] 4] -> (1 2 3 4)
.brev
The chained application associates right-to-left: the rightmost
.code do
is applied to
.codn op ;
the second rightmost
.code do
is applied to the rightmost one and so on. The effect is that partial
application has been achieved. The value
.code 1
is passed to the resulting function, which returns another function
which takes the next argument. Finally, all these chained argument
values are passed to
.codn list .
Each
.cod3 do / op
level is processed independently. The following examples show how the list may
be permuted into several different orders by referring to an implicit argument
at various levels of nesting, making it the first argument of
.codn list .
The unmentioned arguments implicitly follow, in order. This works because
mentioning the argument explicitly means that its corresponding
.code do
operator no longer inserts its argument implicitly into body of the
function which it generates:
.verb
[[[[(do do do op list @1) 1] 2] 3] 4] -> (4 1 2 3)
[[[[(do do do op list @@1) 1] 2] 3] 4] ->(3 1 2 4)
[[[[(do do do op list @@@1) 1] 2] 3] 4] -> (2 1 3 4)
[[[[(do do do op list @@@@1) 1] 2] 3] 4] -> (1 2 3 4))
.brev
The following example mentions all arguments at every
.cod3 do / op
nesting level, thereby explicitly establishing the order
in which they are passed to
.codn list :
.verb
[[[[(do do do op list @1 @@1 @@@1 @@@@1) 1] 2] 3] 4] -> (4 3 2 1)
.brev
.TP* Examples:
.verb
(let ((c 0))
(mapcar (op cons (inc c)) '(a b c)))
--> ((1 . a) (2 . b) (3 . c))
(reduce-left (op + (* 10 @1) @2) '(1 2 3)) --> 123
.brev
.coNP Macro @ lop
.synb
.mets (lop << form +)
.syne
.desc
The
.code lop
macro is variant of
.code op
with special semantics.
The
.meta form
arguments support the same notation as those of the
.code op
operator.
If only one
.meta form
is given then
.code lop
is equivalent to
.codn op .
If two or more
.meta form
arguments are present, then
.code lop
generates a variadic function which inserts all of its trailing
arguments between the first and second
.metn form s.
That is to say, trailing arguments coming into the anonymous function
become the left arguments of the function or function-like object
denoted by the first
.meta form
and the remaining
.metn form s
give additional arguments. Hence the name
.codn lop ,
which stands for \(dqleft-inserting
.codn op \(dq.
This left insertion of the trailing arguments takes place regardless of whether
.code @rest
occurs in any
.metn form .
The
.meta form
syntax determines the number of required arguments of the
generated function, according to the highest-valued meta-number. The trailing
arguments which are inserted into the left position are any arguments in excess
of the required arguments.
The
.code lop
macro's expansion can be understood via the following equivalences,
except that in the real implementation, the symbols
.code rest
and
.code arg1
through
.code arg3
are replaced with hygienic, unique symbols.
.verb
(lop f) <--> (op f) <--> (lambda (. rest) [f . rest])
(lop f x y) <--> (lambda (. rest)
[apply f (append rest [list x y])])
(lop f x @3 y) <--> (lambda (arg1 arg2 arg3 . rest)
[apply f
(append rest
[list x arg3 y])])
.brev
.TP* Examples:
.verb
(mapcar (lop list 3) '(a b c)) --> ((a 3) (b 3) (c 3))
(mapcar (lop list @1) '(a b c)) --> ((a) (b) (c))
(mapcar (lop list @1) '(a b c) '(d e f))
--> ((d a) (e b) (f c))
.brev
.coNP Macro @ ldo
.synb
.mets (ldo < oper << form *)
.syne
.desc
The
.code ldo
macro provides a shorthand notation for uses of the
.code do
macro which inserts the first argument of the anonymous function
as the leftmost argument of the specified operator.
The
.code ldo
syntax can be understood in terms of these equivalences:
.verb
(ldo f) <--> (do f @1)
(ldo f x) <--> (do f @1 x)
(ldo f x y) <--> (do f @1 x y)
(ldo f x @2 y) <--> (do f @1 x @2 y)
.brev
The implicit argument
.code @1
is always inserted as the leftmost argument of the operator
specified by the first form.
.TP* Example:
.verb
;; push elements of l1 onto l2.
(let ((l1 '(a b c)) l2)
(mapdo (ldo push l2) l1)
l2)
--> (c b a)
.brev
.coNP Macros @, ap @, ip @ ado and @ ido
.synb
.mets (ap << form +)
.mets (ip << form +)
.mets (ado << form +)
.mets (ido << form +)
.syne
.desc
The
.code ap
macro is based on the
.code op
macro and has identical argument
conventions.
The
.code ap
macro analyzes its arguments and produces a function
.metn f ,
in exactly the same same way as the
.code op
macro. However, instead of returning
.metn f ,
directly, it returns a different function
.metn g ,
which is a one-argument function that accepts a list.
The list specifies arguments to which
.meta g
applies
.metn f ,
and then returns the resulting value.
In other words, the following equivalence holds:
.verb
(ap form ...) <--> (apf (op form ...))
.brev
The
.code ap
macro nests properly with
.code op
and
.codn do ,
in any combination, in regard to the
.meta ...@@n
notation.
The
.code ip
macro is similar to the
.code ap
macro, except that it is based
on the semantics of the function
.code iapply
rather than
.codn apply ,
according
to the following equivalence:
.verb
(ip form ...) <--> (ipf (op form ...))
.brev
The
.code ado
and
.code ido
macros are related to do macro in the same way that
.code ap
and
.code ip
are related to
.codn op .
They produce a one-argument function which works
as if by applying the function generated by do to its
its own arguments, according to the following equivalence:
.verb
(ado form ...) <--> (apf (do form ...))
(ido form ...) <--> (ipf (do form ...))
.brev
See also: the
.code apf
and
.code ipf
functions.
.TP* Example:
.verb
;; Take a list of pairs and produce a list in which those pairs
;; are reversed.
(mapcar (ap list @2 @1) '((1 2) (a b))) -> ((2 1) (b a))
.brev
.coNP Macros @ opip and @ oand
.synb
.mets (opip << clause *)
.mets (oand << clause *)
.syne
.desc
The
.code opip
and
.code oand
operators make it possible to chain together functions which are expressed
using the
.code op
syntax. (See the
.code op
operator for more information).
Both macros perform the same transformation except that
.code opip
translates its arguments to a call to the
.code chain
function, whereas
.code oand
translates its arguments in the same way to a call to the
.code chand
function.
More precisely, these macros perform the following rewrites:
.verb
(opip arg1 arg2 ... argn) -> [chain {arg1} {arg2} ... {argn}]
(oand arg1 arg2 ... argn) -> [chand {arg1} {arg2} ... {argn}]
.brev
where the above
.code {arg}
notation denotes the following transformation applied to each argument:
.verb
(function ...) -> (op function ...)
(operator ...) -> (do operator ...)
(macro ...) -> (do macro ...)
(dwim ...) -> (dwim ...)
[...] -> [...]
(qref ...) -> (qref ...)
(uref ...) -> (uref ...)
.slot -> .slot
.(method ...) -> .(method ...)
atom -> atom
.brev
In other words, compound forms whose leftmost symbol is a macro or operator
are translated to the
.code do
notation. Compound forms denoting function calls are translated to the
.code op
notation. Compound forms which are
.code dwim
invocations, either explicit or via the DWIM brackets notation, are
used without transformation. Used without transformation also are forms
denoting struct slot access, either explicitly using
.code uref
or
.code qref
or the respective dot notations, as well as any atom forms.
Note: the
.code opip
and
.code oand
macros use their macro environment in determining whether a form is a
macro call, thereby respecting lexical scoping.
.TP* Example:
Take each element from the list
.code "(1 2 3 4)"
and multiply it by three, then add 1.
If the result is odd, collect that into the resulting list:
.mono
(mappend (opip (* 3)
(+ 1)
[iff oddp list])
(range 1 4))
.onom
The above is equivalent to:
.mono
(mappend (chain (op * 3)
(op + 1)
[iff oddp list])
(range 1 4))
.onom
The
.code "(* 3)"
and
.code "(+ 1)"
terms are rewritten to
.code "(op * 3)"
and
.codn "(op + 1)" ,
respectively, whereas
.code "[iff oddp list]"
is passed through untransformed.
.coNP Macro @ flow
.synb
.mets (flow < form << opip-arg *)
.syne
.desc
The
.code flow
macro passes the value of
.meta form
through the processing stages described by the
.meta opip-arg
arguments, yielding the resulting value.
The
.meta opip-arg
arguments follow the semantics of the
.code opip
macro.
The following equivalence holds:
.verb
(flow x ...) <--> [(opip ...) x]
.brev
That is to say,
.code flow
is equivalent to the application of an
.codn opip -generated
function to the value of
.metn form .
.TP* Examples:
.verb
(flow 1 (+ 2) (* 3) (cons 0)) -> (0 . 9)
(flow "abc" (upcase-str) (regsub #/B/ "ZTE")) -> "AZTEC"
.brev
.coNP Macro @ ret
.synb
.mets (ret << form )
.syne
.desc
The
.code ret
macro's
.meta form
argument is treated similarly to the second and subsequent arguments of the
.code op
operator.
The
.code ret
macro produces a function which takes any number of arguments,
and returns the value specified by
.metn form .
.meta form
can contain
.code op
meta syntax like
.code @n
and
.codn @rest .
The following equivalence holds:
.verb
(ret x) <--> (op identity* x))
.brev
Thus the expression
.code "(ret @2)"
returns a function similar to
.codn "(lambda (x y . z) y)" ,
and the expression
.code "(ret 42)"
returns a function similar to
.codn "(lambda (. rest) 42)" .
.coNP Macro @ aret
.synb
.mets (aret << form )
.syne
.desc
The
.code aret
macro's
.meta form
argument is treated similarly to the second and subsequent arguments of the
.code op
operator.
The
.code aret
macro produces a function which takes any number of arguments,
and returns the value specified by
.metn form .
.meta form
can contain
.code ap
meta syntax like
.mono
.meti >> @ n
.onom
and
.codn @rest .
The following equivalence holds:
.verb
(aret x) <--> (ap identity* x))
.brev
Thus the expression
.code "(aret @2)"
returns a function similar to
.codn "(lambda (. rest) (second rest))" ,
and the expression
.code "(aret 42)"
returns a function similar to
.codn "(lambda (. rest) 42)" .
.coNP Function @ dup
.synb
.mets (dup << func )
.syne
.desc
The
.code dup
function returns a one-argument function which calls the two-argument
function
.meta func
by duplicating its argument.
.TP* Example:
.verb
;; square the elements of a list
(mapcar [dup *] '(1 2 3)) -> (1 4 9)
.brev
.coNP Function @ flipargs
.synb
.mets (flipargs << func )
.syne
.desc
The
.code flipargs
function returns a two-argument function which calls the two-argument
function
.meta func
with reversed arguments.
.coNP Functions @ chain and @ chand
.synb
.mets (chain << func *)
.mets (chand << func *)
.syne
.desc
The
.code chain
function accepts zero or more functions as arguments, and returns
a single function, called the chained function, which represents the chained
application of those functions, in left-to-right order.
If
.code chain
is given no arguments, then it returns a variadic function which
ignores all of its arguments and returns
.codn nil .
Otherwise, the first function may accept any number of arguments. The second
and subsequent functions, if any, must accept one argument.
The chained function can be called with an argument list which is acceptable
to the first function. Those arguments are in fact passed to the first
function. The return value of that call is then passed to the second
function, and the return value of that call is passed to the third function
and so on. The final return value is returned to the caller.
The
.code chand
function is similar, except that it combines the functionality of
.code andf
into chaining. The difference between
.code chain
and
.code chand
is that
.code chand
immediately terminates and returns
.code nil
whenever any of the functions returns
.codn nil ,
without calling the remaining functions.
.TP* Example:
.verb
(call [chain + (op * 2)] 3 4) -> 14
.brev
In this example, a two-element chain is formed from the
.code +
function
and the function produced by
.code "(op * 2)"
which is a one-argument
function that returns the value of its argument multiplied by two.
(See the definition of the
.code op
operator).
The chained function is invoked using the
.code call
function, with the arguments
.code 3
and
.codn 4 .
The chained evaluation begins by passing
.code 3
and
.code 4
to
.codn + ,
which yields
.codn 7 .
This
.code 7
is then passed to the
.code "(op * 2)"
doubling function, resulting in
.codn 14 .
A way to write the above example without the use of the DWIM brackets and the
op operator is this:
.verb
(call (chain (fun +) (lambda (x) (* 2 x))) 3 4)
.brev
.coNP Function @ juxt
.synb
.mets (juxt << func *)
.syne
.desc
The
.code juxt
function accepts a variable number of arguments which are functions. It
combines these into a single function which, when invoked, passes its arguments
to each of these functions, and collects the results into a list.
Note: the
.code juxt
function can be understood in terms of the following reference implementation:
.verb
(defun juxt (funcs)
(lambda (. args)
(mapcar (lambda (fun)
(apply fun args))
funcs)))
.brev
The
.code callf
function generalizes
.code juxt
by allowing the combining function to be specified.
.TP* Example:
.verb
;; separate list (1 2 3 4 5 6) into lists of evens and odds,
;; which end up juxtaposed in the output list:
[(op [juxt keep-if remove-if] evenp)
'(1 2 3 4 5 6)] -> ((2 4 6) (1 3 5))
;; call several functions on 1, collecting their results:
[[juxt (op + 1) (op - 1) evenp sin cos] 1]'
-> (2 0 nil 0.841470984807897 0.54030230586814)
.brev
.coNP Functions @ andf and @ orf
.synb
.mets (andf << func *)
.mets (orf << func *)
.syne
.desc
The
.code andf
and
.code orf
functions are the functional equivalent of the
.code and
and
.code or
operators. These functions accept multiple functions and return a new function
which represents the logical combination of those functions.
The input functions should have the same arity. Failing that, there should
exist some common argument arity with which each of these can be invoked. The
resulting combined function is then callable with that many arguments.
The
.code andf
function returns a function which combines the input functions with
a short-circuiting logical conjunction.
The resulting function passes its arguments to the input functions
successively,
in left-to-right order.
As soon as any of the functions returns
.codn nil ,
then
.code nil
is returned and the remaining functions are not called.
If none of the functions return
.codn nil ,
then the value returned by the last function is returned.
If the list of functions is empty, then
.code t
is returned.
That is,
.code (andf)
returns a function which accepts any arguments and returns
.codn t .
The
.code orf
function returns a function which combines the input functions with
a short-circuiting logical disjunction.
The resulting function passes its arguments to the input functions
successively,
in left-to-right order.
As soon as any of the functions returns a
.cod2 non- nil
value, that value is returned and the remaining functions are not called.
If all of the functions return
.codn nil ,
then
.code nil
is returned.
If the list of functions is empty, then
.code nil
is returned.
That is,
.code (orf)
returns a function which accepts any arguments and returns
.codn nil .
.coNP Function @ notf
.synb
.mets (notf << function )
.syne
.desc
The
.code notf
function returns a function which is the Boolean negation
of
.metn function .
The returned function takes a variable number of arguments. When
invoked, it passes all of these arguments to
.meta function
and then inverts the result as if by application of
.codn not .
.coNP Functions @ nandf and @ norf
.synb
.mets (nandf << func *)
.mets (norf << func *)
.syne
.desc
The
.code nandf
and
.code norf
functions are the logical negation of the
.code andf
and
.code orf
functions.
They are related according to the following equivalences:
.verb
[nandf f0 f1 f2 ...] <--> (notf [andf f0 f1 f2 ...])
[norf f0 f1 f2 ...] <--> (notf [orf f0 f1 f2 ...])
.brev
.coNP Functions @ iff and @ iffi
.synb
.mets (iff < condfun >> [ thenfun <> [ elsefun ]])
.mets (iffi < condfun < thenfun <> [ elsefun ])
.syne
.desc
The
.code iff
function is the functional equivalent of the
.code if
operator. It accepts
functional arguments and returns a function.
The resulting function takes its arguments, if any, and applies them to
.metn condfun .
If
.meta condfun
yields true, then the arguments are passed to
.meta thenfun
and the
resulting value is returned. Otherwise the arguments are passed to
.meta elsefun
and the resulting value is returned.
If
.meta thenfun
is omitted then
.code identity
is used as default. This omission is not permitted by
.codn iffi ,
only
.codn iff .
If
.meta elsefun
needs to be called, but is omitted, then
.code nil
is returned.
The
.code iffi
function differs from
.code iff
only in the defaulting behavior with respect
to the
.meta elsefun
argument. If
.meta elsefun
is omitted in a call to
.code iffi
then the default function is
.codn identity .
This is useful in situations when one value is to be
replaced with another one when the condition is true, otherwise
preserved.
The following equivalences hold between
.code iffi
and
.codn iff :
.verb
(iffi a b c) <--> (iff a b c)
(iffi a b) <--> (iff a b identity)
[iffi a b nilf] <--> [iff a b]
[iffi a identity nilf] <--> [iff a]
.brev
The following equivalences illustrate
.code iff
with both optional arguments omitted:
.verb
[iff a] <---> [iff a identity nilf] <---> a
.brev
.coNP Functions @ tf and @ nilf
.synb
.mets (tf << arg *)
.mets (nilf << arg *)
.syne
.desc
The
.code tf
and
.code nilf
functions take zero or more arguments, and ignore them.
The
.code tf
function returns
.codn t ,
and the
.code nilf
function returns
.codn nil .
Note: the following equivalences hold between these functions and the
.code ret
operator, and
.code retf
function.
.verb
(fun tf) <--> (ret t) <--> (retf t)
(fun nilf) <--> (ret nil) <--> (ret) <--> (retf nil)
.brev
In Lisp-1-style code,
.code tf
and
.code nilf
behave like constants which can replace uses of
.code "(ret t)"
and
.codn "(ret nil)" :
.verb
[mapcar (ret nil) list] <--> [mapcar nilf list]
.brev
.TP* Example:
.verb
;; tf and nilf are useful when functions are chained together.
;; test whether (trunc n 2) is odd.
(defun trunc-n-2-odd (n)
[[chain (op trunc @1 2) [iff oddp tf nilf]] n])
.brev
In this example, two functions are chained together, and
.code n
is passed
through the chain such that it is first divided by two via the
function denoted by
.code "(op trunc @1 2)"
and then the result is passed into the
function denoted by
.codn "[iff oddp tf nilf]" .
The
.code iff
function passes its argument into
.codn oddp ,
and if
.code oddp
yields true, it passes the same argument to
.codn tf .
Here
.code tf
proves its utility by ignoring that value and returning
.codn t .
If the argument (the divided value) passed into
.code iff
is even, then iff passes it into the
.code nilf
function, which ignores the value and returns
.codn nil .
.coNP Function @ retf
.synb
.mets (retf << value )
.syne
.desc
The
.code retf
function returns a function. That function can take zero or
more arguments. When called, it ignores its arguments and returns
.metn value .
See also: the
.code ret
macro.
.TP* Example:
.verb
;; the function returned by (retf 42)
;; ignores 1 2 3 and returns 42.
(call (retf 42) 1 2 3) -> 42
.brev
.coNP Functions @ apf and @ ipf
.synb
.mets (apf < function << arg *)
.mets (ipf < function << arg *)
.syne
.desc
The
.code apf
function returns a one-argument function whose argument conventions
are similar to those of the
.code apply
function: it accepts one or more arguments, the last of which should
be a list. When that function is called, it applies
.meta function
to these arguments to as if by
.codn apply .
It then returns whatever
.meta function
returns.
If one or more additional
.metn arg s
are passed to
.codn apf ,
then these are stored in the function which is returned.
When the function is invoked, it prepends all of these stored
arguments to those that it is being given, and the resulting combined
arguments are applied. Thus the
.metn arg s
become the leftmost arguments of
.metn function .
The
.code ipf
function is similar to
.codn apf ,
except that the argument conventions and application semantics of the function
returned by
.code ipf
are based on
.code iapply
rather than
.codn apply .
See also: the
.code ap
macro.
.TP* Example:
.verb
;; Function returned by [apf +] accepts the
;; (1 2 3) list and applies it to +, as
;; if (+ 1 2 3) were called.
(call [apf +] '(1 2 3)) -> 6
.brev
.coNP Function @ callf
.synb
.mets (callf < main-function << arg-function *)
.syne
.desc
The
.code callf
function returns a function which applies each
.meta arg-function
to its arguments, juxtaposing the return values of these calls to form
arguments to which
.meta main-function
is then applied.
The return value of
.meta main-function
is returned.
The following equivalence holds, except for the order of evaluation of
arguments:
.verb
(callf fm f0 f1 f2 ...) <--> (chain (juxt f0 f1 f2 ...)
(apf fm))
.brev
.TP* Example:
.verb
;; Keep those pairs which are two of a kind
(keep-if [callf eql first second] '((1 1) (2 3) (4 4) (5 6)))
-> ((1 1) (4 4))
.brev
The following equivalence holds between
.code juxt
and
.codn callf :
.verb
[juxt f0 f1 f2 ...] <--> [callf list f0 f1 f2 ...]:w
.brev
Thus,
.code juxt
may be regarded as a specialization of
.code callf
in which the main function is implicitly
.codn list .
.coNP Function @ mapf
.synb
.mets (mapf < main-function << arg-function *)
.syne
.desc
The
.code mapf
function returns a function which distributes its arguments
into the
.metn arg-function s.
That is to say, each successive argument of the returned
function is associated with a successive
.metn arg-function .
Each
.meta arg-function
is called, passed the corresponding argument. The return
values of these functions are then passed as arguments
to
.meta main-function
and the resulting value is returned.
If the returned function is called with fewer arguments than there
are
.metn arg-function s,
then only that many functions are used. Conversely, if the function is
called with more arguments than there are
.metn arg-function s,
then those arguments are ignored.
The following equivalence holds:
.verb
(mapf fm f0 f1 ...) <--> (lambda (. rest)
[apply fm [mapcar call
(list f0 f1 ...)
rest]])
.brev
.TP* Example:
.verb
;; Add the squares of 2 and 3
[[mapf + [dup *] [dup *]] 2 3] -> 13
.brev
.SS* Input and Output (Streams)
\*(TL supports input and output streams of various kinds, with
generic operations that work across the stream types.
In general, I/O errors are usually turned into exceptions. When the description
of error reporting is omitted from the description of a function, it can be
assumed that it throws an error.
.coNP Special variables @, *stdout* @, *stddebug* @, *stdin* @ *stderr* and @ *stdnull*
.desc
These variables hold predefined stream objects. The
.codn *stdin* ,
.code *stdout*
and
.code *stderr*
streams closely correspond to the underlying operating system streams.
Various I/O functions require stream objects as arguments.
The
.code *stddebug*
stream goes to the same destination as
.codn *stdout* ,
but is a separate object which can be redirected independently, allowing
debugging output to be separated from normal output.
The
.code *stdnull*
stream is a special kind of stream called a null stream. To read operations,
the stream appears empty, like a stream open on an empty file. To write
operations, it appears as a data sink of infinite capacity which consumes data
and discards it. This stream is similar to
the
.code /dev/null
device on Unix, and in fact has a relationship to it. If an attempt is made
to obtain the underlying file descriptor of
.code *stdnull*
using the
.code fileno
function, then the
.code /dev/null
device is open, if the host platform supports it. The resulting file
descriptor number is returned, and also retained in the
.code *stdnull*
device. When
.code close-stream
is invoked on
.codn *stdnull* ,
that descriptor is closed. This feature of
.code *stdnull*
allows it to be useful for establishing redirections around the
execution of external utilities.
.TP* Example:
.verb
;; redirect output of ls *.txt command to /dev/null
(let ((*stderr *stdnull*))
(sh "ls *.txt"))
.brev
.coNP Special variables @ *print-flo-format* and @ *pprint-flo-format*
.desc
The
.code *print-flo-format*
variable determines the conversion format which is applied when
a floating-point value is converted to decimal text by the
functions
.codn print ,
.codn prinl ,
and
.codn tostring .
The default value is
.codn "~s" .
The related variable
.code *pprint-flo-format*
similarly determines the conversion format applied to floating-point
values by the functions
.codn pprint ,
.codn pprinl ,
and
.codn tostringp .
The default value is
.codn "~a" .
The format string in either variable must specify the consumption of
exactly one
.code format
argument.
The conversion string may use embedded width and precision values:
for instance,
.code "~3,4f"
is a valid value for
.code *print-flo-format*
or
.codn *pprint-flo-format* .
.coNP Special variable @ *print-flo-precision*
.desc
The
.code *print-flo-precision*
special variable specifies the default floating-point printing
precision which is used when the
.code ~a
or
.code ~s
conversion specifier of the
.code format
function is used for printing a floating-point value, and no precision
is specified.
Note that since the default value of the variable
.code *print-flo-format*
is the string
.codn "~s" ,
the
.code *printf-flo-precision*
variable, by default, also determines the precision which applies when
floating-point values are converted to decimal text by the functions
.codn print ,
.codn pprint ,
.codn prinl ,
.codn pprinl ,
.code tostring
and
.codn tostringp .
The default value of
.code *print-flo-precision*
is that of the
.code flo-dig
variable.
Note: to print floating-point values in such a way that their values
can be precisely recovered from the printed representation, it is
recommended to override
.code *print-flo-precision*
to the value of the
.code flo-max-dig
variable.
.coNP Special variable @ *print-flo-digits*
.desc
The
.code *print-flo-precision*
special variable specifies the default floating-point printing
precision which is used when the
.code ~f
or
.code ~e
conversion specifier of the
.code format
function is used for printing a floating-point value, and no precision
is specified.
Its default value is
.codn 3 .
.coNP Special variable @ *print-base*
.desc
The
.code *print-base*
variable controls the base (radix) used for printing integer values.
It applies when the functions
.codn print ,
.codn pprint ,
.codn prinl ,
.codn pprinl ,
.code tostring
and
.code tostringp
process an integer value.
It also applies when the
.code ~a
and
.code ~s
conversion specifiers of the
.code format
function are used for printing an integer value.
The default value of the variable is
.codn 10 .
Meaningful values are:
.codn 2 ,
.codn 8 ,
.code 10
and
.codn 16 .
When base 16 is selected, hexadecimal digits are printed as uppercase
characters.
.coNP Special variable @ *print-circle*
.desc
The
.code *print-circle*
variable is a Boolean which controls whether the circle notation is
in effect for printing aggregate objects: conses, ranges, vectors, hash tables
and structs. The initial value of this variable is
.codn nil :
circle notation printing is disabled.
The circle notation works for structs also, including structs which have
user-defined
.code print
methods. When a
.code print
method calls functions which print objects, such as
.codn print ,
.code pprinl
or
.code format
on the same stream, the detection of circularity and substructure sharing
continues in these recursive invocations.
However, there are limitations in the degree of support for circle notation
printing across
.code print
methods. Namely, a
.code print
method of a struct
.meta S
must not procure and submit for printing objects which are not part of the
ordinary structure that is reachable from the (static or instance) slots of
.metn S ,
if those objects have already been printed prior to invoking the
.code print
method, and have been printed without a
.code #=
circle notation label. The "ordinary structure that is reachable from the
slots" denotes structure that is directly reachable by traversing conses,
ranges, vectors, hashes and struct slots: all printable aggregate objects.
.coNP Special variable @ *read-unknown-structs*
.desc
The
.code *read-unknown-structs*
variable controls the behavior of the parser upon encountering
structure literal
.code #S
syntax which specifies an unknown structure type.
If this variable's value is
.code nil
then such a literal is erroneous; an exception is thrown. Otherwise, such
a structure is converted not into a structure object, which is impossible,
but into a list object whose first element is the symbol
.codn sys:struct-lit .
The remaining elements are taken from the
.code #S
syntax.
.coNP Function @ format
.synb
.mets (format < stream-designator < format-string << format-arg *)
.syne
.desc
The
.code format
function performs output to a stream given by
.metn stream-designator ,
by interpreting the actions implicit in a
.metn format-string ,
incorporating material pulled from additional arguments given by
.mono
.meti << format-arg *.
.onom
Though the function is simple to invoke, there is complexity in format string
language, which is documented below.
The
.meta stream-designator
argument can be a stream object, or one of the values
.code t
or
.codn nil .
The value
.code t
serves as a shorthand for
.codn *stdout* .
The value
.code nil
means that the function will send output into a newly instantiated string
output stream, and then return the resulting string.
.TP* "Format string syntax:"
Within
.metn format-string ,
most characters represent themselves. Those
characters are simply output. The character
.code ~
(tilde) introduces formatting
directives, which are denoted by a single character, usually a letter.
The special sequence
.code ~~
(tilde-tilde) encodes a single tilde. Nothing is
permitted between the two tildes.
The syntax of a directive is generally as follows:
.mono
.mets <> ~[ width ] <> [, precision ] < letter
.onom
In other words, the
.code ~
(tilde) character, followed by a
.meta width
specifier, a
.meta precision
specifier introduced by a comma,
and a
.metn letter ,
such that
.meta width
and
.meta precision
are independently optional: either or both may be omitted.
No whitespace is allowed between these elements.
The
.meta letter
is a single alphabetic character which determines the
general action of the directive. The optional width and precision
are specified as follows:
.RS
.meIP < width
The width specifier consists of an optional
.code <
(left angle bracket) character or
.code ^
(caret)
character followed by an optional width specification.
If the leading
.code <
character is present, then the printing will be left-adjusted within
this field. If the
.code ^
character is present, the printing will be centered within the field.
Otherwise it will be right-adjusted by default.
The width can be specified as a decimal integer with an optional leading
minus sign, or as the character
.codn * .
The
.code *
notation means that instead of digits, the value of the next argument is
consumed, and expected to be an integer which specifies the width. If the
width, specified either way, is negative, then the field will be left-adjusted.
If the value is positive, but either the
.code <
or
.code ^
prefix character is present in the width
specifier, then the field is adjusted according to that character.
The padding calculations for alignment and centering take into account
character display width, as defined by the
.code display-width
function. For instance, a character string containing four Chinese
characters (kanji) has a display width of 8, not 4.
The width specification does not restrict the printed portion of a datum.
Rather, for some kinds of conversions, it is the precision specification that
performs such truncation. A datum's display width (or that of its printed
portion, after such truncation is applied) can equal or exceed the specified
field width. In this situation it overflows the field: the printed portion is
rendered in its entirety without any padding applied on either side for
alignment or centering.
.meIP < precision
The precision specifier is introduced by a leading comma. If this comma appears
immediately after the directive's
.code ~
character, then it means that
.meta width
is being omitted; there is only a precision field.
The precision specifier may begin with these optional characters, whose effect
.RS
.coIP 0
the "leading zero option": pad with leading zeros;
.coIP +
print a sign for positive values;
.coIP -
print a single leading zero in place of a positive sign; and
.IP space
print a space in place of a positive sign.
.RE
The precision options apply only when the value being printed is a number;
otherwise they are ignored.
If the
.codn + ,
.code -
or
space are multiply specified, the rightmost one takes precedence.
The precision specifier itself follows: it must be either a decimal integer
or the
.code *
character indicating that the precision value comes from an integer argument.
The leading zero option is only active if accompanied by a precision
value, either coming from additional digits in the formatting directive,
or from an argument indicated by
.codn * .
If no precision specifier is present, then the leading zero option
is interpreted as a specifier indicating a precision value of zero, rather
than requesting leading zeros. To request zero padding together with zero
precision, either two or more zero digits are required, or else the leading
zero indicator must be given together with the
.code *
specifier.
For non-numeric values, the precision specifies the maximum number of
print positions to occupy, taking into account the display width of each
character of the printed representation of the object, as according
to the
.code display-width
function. The object's printed representation is truncated, if necessary, to
the maximum number of characters which will not exceed the specified number of
print positions.
A numeric argument is formatted into the field in two distinct steps, both of
which involve the precision value in a different role. The details of the first
of these steps, and the role payed by precision, depends on which conversion
directive is used. The second step works in a generic way, and is described
below.
The second step, namely setting a the printed representation of the number
into the text field, occurs in the following way.
First, the precision that was specified, or else the default precision that was
used by the first stage in the absence of the precision being specified, is
clamped to one less than the field width, or else zero if the field width
is zero. The resulting value is called the effective precision.
Next, the length of the printed representation of the number, not including
its sign, is calculated. If this part of the number is shorter than the
effective precision, then it is padded on the left with spaces or leading zeros
so that the resulting string is equal to the precision.
Next, if the number is negative, or else if adding a positive sign has been
requested, then the sign is added. It is added to the left of the padding
zeros, or else to the right of padding spaces, whichever the case may be.
At this stage, if the number is not yet adorned with a sign, and either the
.code -
or space precision option had been given, then the appropriate character,
the digit
.code 0
or a space, is added in the place where the sign would go. This is done
only if the result will not overflow the field width, but without regard
for whether the character will overflow the effective precision.
Finally, the resulting number is rendered into the field, using the requested
left, right or center adjustment, as if it were a character string. If it
overflows the field, it is reproduced in its entirety without any adjustment
being performed.
.RE
.TP* "Format directives:"
.RS
Format directives are case sensitive, so that for example
.code ~x
and
.code ~X
have a
different effect, and
.code ~A
doesn't exist whereas
.code ~a
does. They are:
.coIP a
Prints any object in an aesthetic way, as if by the
.code pprint
function.
The aesthetic notation violates read-print consistency: this notation
is not necessarily readable if it is implanted in \*(TX source code.
The field width specifier is honored, including the left-right adjustment
semantics.
When the
.code a
specifier is used for numbers, the formatting is performed in two
distinct steps: the printed representation of the number is calculated
first, and then that representation is set into the field. The precision
parameter plays two different roles in these two steps.
In the first step, the rendering of a floating-point number to its printed
representation, the precision specifies the maximum number of total significant
figures, which do not include any digits in the exponent, if one is printed.
Numbers are printed in E notation if their magnitude is small, or
else if their exponent exceeds their precision. If the precision is not
specified, then it is obtained from the
.code *print-flo-precision*
special variable, whose default value is the same as that of the
.code flo-dig
variable.
Floating point values which are integers are
printed without a trailing
.code .0
(point zero).
The
.code +
flag in the precision is honored for rendering an explicit
.code +
sign on nonnegative values.
If a leading zero is specified in the precision, and a nonzero width is
specified, then the printed value's integer part will be padded with leading
zeros up to one less than the field width. These zeros are placed before the
sign. A precision value of zero imposed on floating-point values is
equivalent to a value of one; it is not possible to request zero significant
figures.
Integers are not affected by the precision value in the conversion to
text; all of the digits of the integer are taken into the second step.
.coIP s
Prints any object in a standard way, as if by the
.code print
function. Objects for
which read-print consistency is possible are printed in a way such that
if their notation is implanted in \*(TX source, they are readable.
The field width specifier is honored, including the left-right adjustment
semantics. The precision field is treated similarly to the
.code ~a
format directive, except that non-exponentiated floating point numbers that
would be mistaken for integers include a trailing
.code .0
for the sake of read-print
consistency. Objects truncated by precision may not have read-print
consistency. For instance, if a string object is truncated, it loses its
trailing closing quote, so that the resulting representation is no longer
a properly formed string object. For integer objects, the
.code *print-base*
variable is honored. Effectively, an integer is printed by the
.code s
directive as if by the
.codn b ,
.codn o ,
.codn d ,
or
.code x
directive, depending on the value of the variable.
.coIP d
Requires an argument of integer or character type type. The integer
value or character code is printed in decimal.
Width and precision semantics are as described for the
.code a
format directive, for integers.
.coIP x
Requires an argument of character, integer or buffer type. The integer value,
character code, or buffer contents are printed in hexadecimal, using lowercase
letters for the digits
.code a
through
.codn f .
Width and precision semantics are as described for the
.code a
format directive, for integers.
.coIP X
Like the
.code x
directive, but the hexadecimal digits
.code a
through
.code f
are rendered in uppercase.
.coIP o
Like the
.code x
directive, but octal is used instead of hexadecimal.
.coIP b
Like the
.code x
directive, but binary is used instead of hexadecimal.
.coIP f
The
.code f
directive prints numbers in a fixed point decimal notation, with
a fixed number of digits after the decimal point. It requires a numeric
argument. (Unlike
.codn x ,
.code X
and
.codn o ,
it does not allow an argument of character type).
The formatting performed by
.code f
is performed in two distinct steps: the printed representation of the number is
calculated first, and then that representation is set into the field. The
precision parameter coming from the directive is only involved in
the first step.
In the first step, the precision specifier gives the number of digits past the
decimal point. The number is rounded off to the specified precision, if
necessary. Furthermore, that many digits are always printed, regardless of the
actual precision of the number or its type. If it is omitted, then the value
is obtained from the special variable
.codn *print-flo-digits* ,
whose default value is three: three digits past the decimal point. A precision
of zero means no digits past the decimal point, and in this case the decimal
point is suppressed (regardless of whether the numeric argument is
floating-point or integer).
No limit is placed on the number of significant figures in the number by
either the precision or width value.
When the resulting textual number passes to the second formatting step, the
precision value, for the purposes of that step, is calculated by taking one
less than the field width, or else zero if the field width is zero.
This value is not related to the precision that had been used to determine
the number of places past the decimal point.
.coIP e
The
.code e
directive prints numbers in E notation. It requires
a numeric argument. (Unlike
.codn x ,
.code X
and
.codn o ,
it does not allow an argument of character type).
The formatting performed by
.code e
is performed in two distinct steps: the printed representation of the number is
calculated first, and then that representation is set into the field. The
precision parameter coming from the directive is only involved in
the first step.
In the first step, the precision specifier gives the number of digits past the
decimal point printed in the E notation, not counting the digits in
the exponent. Exactly that many digits are printed, regardless of the
precision of the number. If the precision is omitted, then the number of
digits after the decimal point is obtained from the value of the special
variable
.codn *print-flo-digits* ,
whose default value is three. If the precision is zero, then a decimal portion
is truncated off entirely, including the decimal point.
When the resulting textual number passes to the second formatting step, the
precision value, for the purposes of that step, is calculated by taking one
less than the field width, or else zero if the field width is zero.
This value is not related to the precision that had been used to determine
the number of places past the decimal point.
.coIP p
The
.code p
directive prints a numeric representation in hexadecimal of the bit pattern
of the object, which is meaningful to someone familiar with the internals
of \*(TX. If the object is a pointer to heaped data, that value
has a correspondence to its address.
.coIP !
The
.code !
directive establishes hanging indentation, and turns on the stream's
indentation mode. Subsequent lines printed within the execution of the
same
.code format
call will be automatically indented. If no width is specified, then
the directive sets the hanging indentation to the current printing
column position. If a width is specified, then it represents an offset
(positive or negative). If the
.code <
prefix character is present, the hanging indentation is set to the
specified offset relative to the current printing column.
If the
.code <
prefix is present on the width field, then the offset is applied
relative to the indentation which was saved on entry into the
.code format
function.
The indentation mode and indentation column are automatically restored to their
previous values when
.code format
function terminates, naturally or via an exception or nonlocal jump.
The effect of a precision field (even if zero) combined with the
.code !
directive is currently not specified, and reserved for future extension.
The precision field is processed syntactically, and no error occurs, however.
.RE
.coNP Function @ fmt
.synb
.mets (fmt < format-string << format-arg *)
.syne
.desc
The
.code fmt
function provides a shorthand for formatting to a string, according
to the following equivalence which holds between
.code fmt
and
.codn format :
.verb
(fmt s arg ...) <--> (format nil s arg ...)
.brev
.coNP Macro @ pic
.synb
.mets (pic < format-string << format-arg *)
.syne
.desc
The
.code pic
macro ("picture based formatting") provides a notation for constructing a
character string under the control of
.meta format-string
which indicates the insertion of zero or more
.meta format-arg
argument values.
Like the
.code fmt
function or quasiliteral syntax, the
.code pic
macro returns a character string.
The
.code pic
macro's
.meta format-string
notation is different from quasiliterals or from
.codn fmt .
The
.code pic
.meta format-string
argument isn't an evaluated expression, but syntax. It must be either a string
literal or else a string quasiliteral. No other syntax is permitted.
If
.meta pic
is a string, is scanned left to right in search of
.IR "pic patterns" .
Any characters not belonging to a pic pattern are copied into the output
string verbatim. When a pic pattern is found, it is removed from
.meta format-string
and applied to the next successive
.meta format-arg
to perform a conversion and formatting of that value to text. The resulting
text is appended to the output string, and the process continues in search of the next pic pattern.
When the
.meta format-string
is exhausted, the constructed string is returned.
If
.meta format-string
is a quasiliteral, then all of the text strings embedded within the
quasiliteral are examined in the same way, in left to right order. Each such
string is transformed into an expression which produces a character string
according to the semantics of the pic patterns it contains, and the resulting
expressions are substituted into the original quasiliteral to produce a
transformed quasiliteral.
There must be exactly as many
.meta format-arg
arguments as there are pic patterns in
.metn format-string .
The
.code pic
macro arranges for the left-to-right evaluation of the
.meta format-arg
expressions. If
.meta format-string
is a quasiliteral, the evaluation of these expressions is interleaved
into the quasiliterals expressions and variables, in the order implied
by the placement of the corresponding pic patterns relative to the
quasiliteral elements. For instance, if
.meta format-string
is
.code `@(abc)<<<@(xyz)`
then the function
.code abc
is called first, then the
.meta format-argument
is evaluated which produces a value for the
.code <<<
pic pattern, after which the
.code xyz
function is called.
There are two kinds of pic patterns: alignment patterns, numeric patterns and
escape patterns.
Escape patterns consist of a two-character sequence introduced by the
.code ~
(tilde)
character, which is followed by one of the characters that are special in
pic pattern syntax:
.verb
< > | + - 0 # . ! ~ , ( )
.brev
An escape pattern produces the second character as its output. For instance
.code ~~
encoded a single
.code ~
character, and
.code ~#
encodes a literal
.code #
character that is not part of any pattern.
Alignment patterns are described next.
.RS
.coIP <<...<<
A sequence of one or more
.code <
(less than)
characters specifies that the corresponding argument is rendered left-aligned
in a field whose width is given by the number of
.code <
characters. If the argument's textual representation doesn't fit into the field,
it overflows.
.coIP >>...>>
A sequence of one or more
.code >
(greater than)
characters specifies that the corresponding argument is rendered right-aligned
in a field whose width is given by the number of
.code >
characters. If the argument's textual representation doesn't fit into the field,
it overflows.
.coIP ||...||
A sequence of one or more
.code |
(pipe) characters specifies that the corresponding argument is centered
in a field whose width is given by the number of
.code |
characters. If the argument's textual representation doesn't fit into the field,
it overflows. If the argument cannot be precisely centered, because the
even-odd parity of its character count is different from the parity of the
field width, it is centered slightly to the left: one less space appears on its
left side in respect to its right side.
.RE
.IP
The numeric patterns, by means of their visual pattern and several optional
prefix codes, specify the parameters for the conversion of a numeric
argument, which is rendered right-aligned in a fixed-width field. Numeric
patterns that do not contain any commas conform this simple rule:
.mono
.mets <> [ sign ] [0] {#}+ >> [ point {#}+ | !]
.onom
or else if they contain commas, the placement of these commas is governed
by the more complicated rule:
.mono
.mets <> [ sign ] [0 [,]] {#}+ {,{#}+}* >> [ point {#}+ {,{#}+}* | !]
.onom
Commas may be placed anywhere within the pattern of hash characters, except at
the beginning or end, or adjacent to the decimal point. If the leading zero is
present, a comma may appear immediately after it, before the first hash.
A second form of both of the above patterns is supported, for specifying
that negative numbers be shown in parentheses. Instead of the sign, an
opening parenthesis may appear, which must be matched by a closing parenthesis
which follows a valid pattern interior:
.mono
.mets ( [0] {#}+ >> [ point {#}+ | !] )
.onom
With embedded commas:
.mono
.mets ( [0 [,]] {#}+ {,{#}+}* >> [ point {#}+ {,{#}+}* | !] )
.onom
The pattern consists of an optional
.meta sign
which is one of the characters
.code +
(plus) or
.code -
(minus), or else it may optionally begin with an opening parenthesis,
indicating one of the two alternative forms.
This is followed by an optional leading zero.
After this comes a sequence of one or more
.code #
(hash) characters, which may contain exactly one
.meta point
element, which is defined as one of the characters
.code .
(period)
or
.code !
(exclamation mark).
This
.meta point
element may appear at most once, and must not be the first or
last character, unless it is the exclamation mark,
in which case it may appear last.
Except if ending in the exclamation mark, a numeric pattern specifies a field
width which is equal to the number of characters occurring in the pattern
itself.
For instance, the patterns
.codn #### ,
.code +###
and
.code 0#.#
all specify a field width of four. If the numeric pattern ends in an exclamation
mark, that character is not counted toward the field width that it specifies.
Thus the pattern
.code ###!
specifies a field width of three.
If the leading sign is present, it has the following meanings:
.RS
.coIP +
If the corresponding numeric argument is nonnegative, the
.code +
character shall appear before first digit. Otherwise the minus character
will appear.
.coIP -
Like
.code +
except that when the numeric argument is nonnegative, instead of a
.code +
character, a space appears before the first digit. This space counts
toward the field width and therefore contributes to overflow.
.RE
.IP
If a leading sign is not present, then no extra character appears before
the first digit of a positive value, which means that an extra character
of field width is available for representing nonnegative values.
If the leading zero is present, it specifies that the number is
padded with zeros on the left. In combination with the
.code -
sign, this shall not cause the leading space before a positive value to
be overwritten with a zero; leading zeros, if any, begin after that space.
The remainder of the pattern specifies the number of digits of the fractional
part which is indicated by number of
.code #
characters after the
.metn point .
The number is rounded to that many fractional digits, which are all rendered,
even if there are trailing zeros.
If no
.meta point
is not specified, then the number of fractional digits is zero. The same is
true if
.meta point
is specified as
.code !
as the last character. In both cases, the numeric argument is rounded to
integer, and rendered without any decimal point or fractional part.
There is a difference between
.meta point
being specified using the ordinary decimal point character
.code .
versus the
.code !
character. The
.code !
character specifies that if the conversion of the numeric argument overflows
the field, then instead of showing any digits, the field is filled with a
pattern consisting of
.code #
(hash) characters, and possibly an embedded decimal point. In contrast, the
.code .
character permits the field's width to increase to accommodate overflowing
output. If overflow takes place and the
.code !
character appears other than as the rightmost character of the pattern,
then the decimal point character
.code .
character appears at the position indicated by that
.code !
character. If the
.code !
character is the rightmost character of the pattern, then, just as
in the case of normal, non-overflowing output, it doesn't contribute to the
width of the hash fill, and only hash characters appear.
If commas appear in the numeric pattern according to the more complex syntactic
rule, they count toward the field width and specify the insertion of
digit-separating commas at the indicated locations. Digit separators may be
specified on either side of the decimal point, but not adjacent to it. In the
output, a digit separating comma shall not appear if it would be immediately
preceded by a
.code +
or
.code -
sign or space. In this situations, the sign character or space appears
in place of the digit separator. A digit separator that appears in a position
occupied by a space is also suppressed in favor of the space. Digit separators
are included among leading zeros. It is not logically possible for a digit
separator to appear as the first character of a pattern's output, because it
may not be the first character of a pattern. However, if a numeric pattern is
preceded or followed by a comma, those commas are ordinary characters which are
copied to the output.
When, due to the presence of
.codn ! ,
an overflowing field is handled by the generation of a the hash character fill,
the hash characters are treated as digits for the purpose of digit separation.
When the pattern uses parentheses to specify that negative numbers are
to be shown with parentheses, the parentheses count toward the field width.
The field portion between the parentheses is called the inner field.
The parentheses appear in the output when the number is negative, and are
placed immediately outside of the inner field, so that if leading zeros are not
requested, there may be one or more spaces between the opening parenthesis and
the first digit. If the number is nonnegative, then each parenthesis is
replaced by one space, flanking the inner field in the same manner as
parentheses.
.TP* Examples:
.verb
;; numeric formatting
(pic "######" 1234.1) -> " 1234"
(pic "######.#" 1234.1) -> " 1234.1"
(pic "#######.##" 1234.1) -> " 1234.10"
(pic "#######.##" -1234.1) -> " -1234.10"
(pic "0######.##" 1234.1) -> "0001234.10"
(pic "+######.##" 1234.1) -> " +1234.10"
(pic "-######.##" 1234.1) -> " 1234.10"
(pic "+0#####.##" 1234.1) -> "+001234.10"
(pic "-0#####.##" 1234.1) -> " 001234.10"
(pic "#######.##" -1234.1) -> " -1234.10"
;; digit separation
(pic "0,###,###.##" 1234.1) -> "0,000,123.10"
(pic "#,###,###.##" 1234.1) -> " 123.10"
;; overflow with !
(pic "#!#" 1234) -> "###"
(pic "#!#" 123) -> "###"
(pic "-##!#" -123) -> "#####"
(pic "+##!#" 123) -> "#####"
(pic "###!" 1234) -> "###"
;; negative parentheses
(pic "(#,###.##) 1234.56) -> " 1,234.56 "
(pic "(#,###.##) -234.56) -> "( 234.56)"
;; alignment, multiple arguments
(pic "<<<<<< 0#.# >>>>>>>" "foo" (+ 2 2) "bar")
--> "foo 04.0 bar"
;; quasiliteral
(let ((a 2) (b "###") (c 13.5))
(pic `abc@(+ a a)###.##@b>>>>` c "x"))
--> "abc4 13.50### x"
;; filename generation
(mapcar (do pic "foo~-0##.jpg") (rlist 0..5 8 12))
--> ("foo-000.jpg" "foo-001.jpg" "foo-002.jpg" "foo-003.jpg"
"foo-004.jpg" "foo-005.jpg" "foo-008.jpg" "foo-012.jpg")
.brev
.coNP Functions @, print @, pprint @, prinl @, pprinl @ tostring and @ tostringp
.synb
.mets (print < obj >> [ stream <> [ pretty-p ]])
.mets (pprint < obj <> [ stream ])
.mets (prinl < obj <> [ stream ])
.mets (pprinl < obj <> [ stream ])
.mets (tostring << obj )
.mets (tostringp << obj )
.syne
.desc
The
.code print
and
.code pprint
functions render a printed character representation of the
.meta obj
argument into
.metn stream .
If the
.meta stream
argument is not supplied, then
the destination is the stream currently stored in the
.code *stdout*
variable.
If Boolean argument
.meta pretty-p
is not supplied or is explicitly specified as
.codn nil ,
then the
.code print
function renders in a way which strives for read-print
consistency: an object is printed in a notation which is recognized as
a similar object of the same kind when it appears in \*(TX source code.
Floating-point objects are printed as if using the
.code format
function, with formatting controlled by the
.code *print-flo-format*
variable.
If
.meta pretty-p
is true, then
.code print
does not strive for read-print consistency.
Strings are printed by sending their characters to the output
stream, as if by the
.code put-string
function, rather than being rendered in the string literal notation
consisting of double quotes, and escape sequences for control
characters. Likewise, character objects are printed via
.code put-char
rather than the
.code #\e
notation. Buffer objects are printed by sending their bytes to the
output stream using
.code put-byte
rather than being rendered in the
.code #b
notation.
Symbols are printed without their package prefix, except that
symbols from the keyword package are still printed with the leading colon.
Floating-point objects are printed as if using the
.code format
function, with formatting controlled by the
.code *pprint-flo-format*
variable.
When aggregate objects like conses, ranges and vectors are printed,
the notations of these objects themselves are unaffected by the
.code pretty-p
flag; however, that flag is distributed to the elements.
The
.code print
function returns
.metn obj .
The
.code pprint
("pretty print") function is equivalent to
.codn print ,
with the
.meta pretty-p
argument hardcoded true.
The
.code prinl
function ("print and new line") behaves like a call to
.code print
with
.meta pretty-p
defaulting to
.codn nil ,
followed by issuing a newline characters to the stream.
The
.code pprinl
function ("pretty print and new line") behaves like
.code pprint
followed by issuing a newline to the stream.
The
.code tostring
and
.code tostringp
functions are like
.code print
and
.codn pprint ,
but they do not accept a stream argument. Instead they print to a freshly
instantiated string stream, and return the resulting string.
The following equivalences hold between calls to the
.code format
function and calls to the above functions:
.verb
(format stream "~s" obj) <--> (print obj stream)
(format t "~s" obj) <--> (print obj)
(format t "~s\en" obj) <--> (prinl obj)
(format nil "~s" obj) <--> (tostring obj)
.brev
For
.codn pprint ,
.code tostringp
and
.codn pprinl ,
the equivalence is produced by using
.code ~a
in format rather than
.codn ~s .
.TP* Notes:
For floating-point numbers, the above description of the behavior in
terms of the format specifiers
.code ~s
and
.code ~a
only applies with respect to the default values of the variables
.code *print-flo-format*
and
.codn *pprint-flo-format* .
For characters, the print function behaves as follows: most control
characters in the Unicode
.code C0
and
.code C1
range are rendered using the
.code #\ex
notation,
using two hex digits. Codes in the range
.code D800
to
.codn DFFF ,
and the codes
.code FFFE
and
.code FFFF
are printed in the
.code #\exNNNN
with four hexadecimal digits, and
character above this range are printed using the same notation, but with six
hexadecimal digits. Certain characters in the
.code C0
range are printed using
their names such as
.code #\enul
and
.codn #\ereturn ,
which are documented
in the Character Literals section.
The
.code DC00
character is printed as
.codn #\epnul .
All other characters are printed as
.mono
.meti >> #\e char
.onom
where
.meta char
is the actual character.
Caution: read-print consistency is affected by trailing material. If additional
digits are printed immediately after a number without intervening whitespace,
they extend that number. If hex digits are printed after the character
.codn x ,
which is rendered as
.codn #\ex ,
they look like a hex character code.
.coNP Function @ tprint
.synb
.mets (tprint < obj <> [ stream ])
.syne
.desc
The
.code tprint
function prints a representation of
.meta obj
on
.metn stream .
If the stream argument is not supplied, then
the destination is the stream currently stored in the
.code *stdout*
variable.
For all object types except lists and vectors,
.code tprint
behaves like
.codn pprinl .
If
.code obj
is a list or vector, then
.code tprint
recurses: the
.code tprint
function is applied to each element. An empty list or vector
results in no output at all. This effectively means that an arbitrarily nested
structure of lists and vectors is printed flattened, with one element on each
line.
.coNP Function @ display-width
.synb
.mets (display-width << char )
.mets (display-width << string )
.syne
.desc
The
.code display-width
function calculates the number of places occupied by the printed representation
of
.meta char
or
.meta string
on a monospace display which renders certain characters, such as the East Asian
kanji and other characters, using two places.
For a
.meta string
argument, this value is the sum of the individual display width of the
string's constituent characters. The display width of an empty string is zero.
Control characters are assigned a display width of zero, regardless of
their display control semantics, if any.
Characters marked by Unicode as being wide or full width, have a display
width of two. Other characters have a display width of one.
.coNP Function @ streamp
.synb
.mets (streamp << obj )
.syne
.desc
The
.code streamp
function returns
.code t
if
.meta obj
is any type of stream. Otherwise it returns
.codn nil .
.coNP Function @ real-time-stream-p
.synb
.mets (real-time-stream-p << obj )
.syne
.desc
The
.code real-time-streamp-p
function returns
.code t
if
.meta obj
is a stream marked as
"real-time". If
.meta obj
is not a stream, or not a stream marked as "real-time",
then it returns
.codn nil .
Only certain kinds of streams accept the real-time attribute: file streams and
tail streams. This attribute controls the semantics of the application of
.code lazy-stream-cons
to the stream. For a real-time stream,
.code lazy-stream-cons
returns a stream with "naive" semantics which returns data as soon as it is
available, at the cost of generating spurious
.code nil
item when the stream
terminates. The application has to recognize and discard that
.code nil
item.
The ordinary lazy streams read ahead by one line and suppress this extra
item, so their representation is more accurate.
When \*(TX starts up, it automatically marks the
.code *stdin*
stream as real-time, if it is connected to a TTY device (a device for which
the POSIX function
.code isatty
reports true). This is only supported on platforms that have this function.
The behavior is overridden by the
.code -n
command-line option.
.coNP Function @ open-file
.synb
.mets (open-file < path <> [ mode-string ])
.syne
.desc
The
.code open-file
function creates a stream connected to the file
which is located at the given
.metn path ,
which is a string.
The
.meta mode-string
argument is a string which uses the same
conventions as the mode argument of the C language
.code fopen
function, with greater permissiveness, and some extensions.
The syntax of
.meta mode-string
is described by the following grammar.
Note that it permits no whitespace characters:
.mono
.mets < mode-string := [ < mode ] [ < options ]
.mets < mode := { < selector [ + ] | + }
.mets < selector := { r | w | a | m }
.mets < options := { b | x | l | u | i | n | < digit | < redirection }
.mets < digit := { 0 | 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 }
.onom
If the
.meta mode-string
argument is omitted, the behavior is the same as an empty
mode string.
The
.meta mode
part of the mode string generates the following possibilities:
.RS
.meIP empty
If the
.meta mode
is missing, then a default mode is implied. The default
is specific to the particular stream-opening function. In
the case of
.codn open-file ,
the default mode is
.codn r .
.coIP +
A
.meta mode
consisting of just the
.code +
character is equivalent to
.codn r+ .
.coIP r
This
.meta mode
means that the file is opened for reading.
.coIP r+
The file is opened for reading and writing. It is not created
if it doesn't exist.
.coIP w
The file is opened for writing. If it exists, it is truncated
to zero length. If it doesn't exist, it is created.
.coIP w+
The file is opened for reading and writing. If it exists, it
is truncated to zero length. If it doesn't exist, it is created.
.coIP m
The file is opened for modification. This is the same as
.code w
except that the file is not truncated if it exists.
.coIP m+
The file is opened for reading and modification. This is the same as
.code w+
except that the file is not truncated if it exists.
.coIP a
The file is opened for writing. If it doesn't exist, it is
created. If it exists, the current position is advanced to
one byte past the end of the file, so that newly written data
are appended.
.coIP a+
The file is opened for reading and writing. If it doesn't exist,
it is created. The read position is at the beginning of the file,
but writes are appended to the end regardless of the position.
.RE
.IP
The meanings of the option characters are:
.RS
.coIP b
The file is opened in binary mode: no line-ending translation takes place.
In the absence of this option, files are opened in text mode, in which newline
characters in the stream are an abstract indication of the end of a line,
translate to a system-specific way of terminating lines in text files.
.coIP x
The file is created and opened only if it does not already exist.
Otherwise, a
.code file-error
exception is thrown.
This option is allowed only with the
.code w
and
.code w+
modes.
.coIP l
Specifies that the stream will be line buffered. This means that an implicit
flush operation takes place whenever the newline character is output.
.coIP u
Specifies that the stream will be unbuffered. It is erroneous for both
.code l
and
.code u
to be specified.
.coIP i
Specifies that the stream will have the real-time
property set. For a description of the semantics, see the
.code real-time-stream-p
function. Briefly, this property affects the semantics of lazy lists which draw
input from the stream.
In addition, for a stream opened for writing or reading and writing, the
.code i
mode letter specifies that the stream will be line buffered, unless
specified as unbuffered with
.codn u .
.coIP n
Specifies that the operation shall not block.
.meIP digit
A decimal digit specifies the stream buffer size
as binary exponential buffer size order, such that
.code 0
specifies 1024 bytes,
.code 1
specifies 2048 and so forth up to
.code 9
specifying 524288 bytes. If no such digit is specified, then the
stream uses a default buffer size. It is erroneous for the
size order digit to be present together with the option
.codn u .
.meIP redirection
This option refers to a special syntax that only has an effect
in mode strings that are passed to the
.code open-process
function; the syntax performs I/O redirections in the child process
created by that function, and is described in that function's
documentation.
.RE
.coNP Function @ open-tail
.synb
.mets (open-tail < path >> [ mode-string <> [ seek-to-end-p ]])
.syne
.desc
The
.code open-tail
function creates a tail stream connected to the file which is
located at the given
.metn path .
The
.meta mode-string
argument is a string which uses
the same conventions as the mode argument of the C language
.code fopen
function. If this argument is omitted, then
.str r
is used.
See the
.code open-file
function for a discussion of modes.
The
.code seek-to-end-p
argument is a Boolean which determines whether the initial
read/write position is at the start of the file, or just past the end.
It defaults to
.codn nil .
This argument only makes a difference if the file exists
at the time
.code open-tail
is called. If the file does not exist, and is later
created, then the tail stream will follow that file from the beginning. In
other words,
.meta seek-to-end-p
controls whether the tail stream reads all the
existing data in the file, if any, or whether it reads only newly added data
from approximately the time the stream is created.
A tail stream has special semantics with regard to reading at the end
of file. A tail stream never reports an end-of-file condition; instead
it polls the file until more data is added. Furthermore, if the file
is truncated, or replaced with a smaller file, the tail stream follows
this change: it automatically opens the smaller file and starts reading from
the beginning (the
.meta seek-to-end-p
flag only applies to the initial open).
In this manner, a tail stream can dynamically grow rotating log files.
Caveat: since a tail stream can reopen a new file which has the same
name as the original file, it behave incorrectly if the program
changes the current working directory, and the pathname is relative.
.coNP Function @ open-directory
.synb
.mets (open-directory << path )
.syne
.desc
The
.code open-directory
function tries to create a stream which reads the
directory given by the string argument
.metn path .
If a filesystem object exists
under the path, is accessible, and is a directory, then the function
returns a stream. Otherwise, a file error exception is thrown.
The resulting stream supports the
.code get-line
operation. Each call to the
.code get-line
operation retrieves a string representing the next directory
entry. The value
.code nil
is returned when there are no more directory entries.
The
.code .
and
.code ..
entries in Unix filesystems are not skipped.
.coNP Function @ tmpfile
.synb
.mets (tmpfile)
.syne
.desc
The
.code tmpfile
function creates a new temporary binary file which is different from any
existing file. It opens a stream for that file and returns the stream. The
stream is created with the
.code open-file
mode
.strn w+b .
When the stream is closed, or the \*(TX image terminates, the file is deleted.
Note: the
.code tmpfile
function is implemented using the same-named ISO C and POSIX library function.
On POSIX systems of sufficient quality,
.code tmpfile
deletes the file before returning the open stream, such that the file object
continues to exist while the stream is open, but is not known by any name
in the file system. POSIX (IEEE Std 1003.1-2017) notes that in some
implementations, "a permanent file may be left behind if the process calling
tmpfile() is killed while it is processing a call to tmpfile".
Notes: if a unique file is required which exists in the file system under a
known name until explicitly deleted, the
.code mkstemp
function may be used. If a unique directory needs to be created, the
.code mkdtemp
function may be used. These two functions are described in the Unix Filesystem
Complex Operations section of the manual.
.coNP Function @ make-string-input-stream
.synb
.mets (make-string-input-stream << string )
.syne
.desc
The
.code make-string-input-stream
function produces an input stream object. Character read operations on the
stream object read successive characters from
.metn string .
Output operations and byte operations are not supported.
.coNP Function @ make-string-byte-input-stream
.synb
.mets (make-string-byte-input-stream << string )
.syne
.desc
The
.code make-string-byte-input-stream
function produces an input stream object. Byte read operations on
this stream object read successive byte values obtained by encoding
.meta string
into UTF-8. Character read operations are not supported, and neither
are output operations.
.coNP Function @ make-strlist-input-stream
.synb
.mets (make-strlist-input-stream << list )
.syne
.desc
The
.code make-strlist-input-stream
function produces an input stream object based on a list of strings.
Through the character read operations invoked on this stream,
the list of strings appears as a list of newline-terminated lines.
Output operations and byte operations are not supported.
.coNP Function @ make-string-output-stream
.synb
.mets (make-string-output-stream)
.syne
.desc
The
.code make-string-output-stream
function, which takes no arguments, creates a string output stream.
Data sent to this stream is accumulated into a string object.
String output streams support both character and byte output operations.
Bytes are assumed to represent a UTF-8 encoding, and are decoded in order
to form characters which are stored into the string.
If an incomplete UTF-8 code is output, and a character output operation then
takes place, that code is assumed to be terminated and is decoded as invalid
bytes. The UTF-8 decoding machine is reset and ready for the start of a new
code.
The
.code get-string-from-stream
function is used to retrieve the accumulated string.
If the null character is written to a string output stream, the behavior
is unspecified. \*(TX strings cannot contain null bytes. The pseudo-null
character
.codn #\exDC00 ,
also notated
.codn #\epnul ,
will produce a null byte when converted to UTF-8 and thus serves as an
effective internal representation of the null character in external data.
.coNP Function @ get-string-from-stream
.synb
.mets (get-string-from-stream << stream )
.syne
.desc
The
.meta stream
argument must be a string output stream. This function finalizes
the data sent to the stream and retrieves the accumulated character string.
If a partial UTF-8 code has been written to
.metn stream ,
and then this
function is called, the byte stream is considered complete and the partial
code is decoded as invalid bytes.
After this function is called, further output on the stream is not possible.
.coNP Function @ make-strlist-output-stream
.synb
.mets (make-strlist-output-stream)
.syne
.desc
The
.code make-strlist-output-stream
function is similar to
.codn make-string-output-stream .
However, the stream object produced by this function does not produce a string,
but a list of strings. The data is broken into multiple strings by newline
characters written to the stream. Newline characters do not appear in the
string list. Also, byte output operations are not supported.
.coNP Function @ get-list-from-stream
.synb
.mets (get-list-from-stream << stream )
.syne
.desc
The
.code get-list-from-stream
function returns the string list which has accumulated inside
a string output stream given by
.metn stream .
The string output stream is
finalized, so that further output is no longer possible.
.coNP Macro @ with-in-string-stream
.synb
.mets (with-in-string-stream >> ( stream-var << string )
.mets \ \ << body-form *)
.syne
.desc
The
.code with-in-string-stream
macro binds the symbol
.meta stream-var
as a variable, initializing it with a newly created
string input stream. The string input stream is
constructed from
.meta string
as if by the
.mono
.meti (make-string-input-stream << string )
.onom
expression.
Then it evaluates the
.metn body-form s
in the scope of the variable.
The value of the last
.meta body-form
is returned, or else
.code nil
if no forms are present.
The
.meta stream-var
argument must be a bindable symbol,
as defined by the
.code bindable
function.
The
.meta string
argument must be a form
which evaluates to a character string value.
.coNP Macro @ with-in-string-byte-stream
.synb
.mets (with-in-string-byte-stream >> ( stream-var << string )
.mets \ \ << body-form *)
.syne
.desc
The
.code with-in-string-byte-stream
macro binds the symbol
.meta stream-var
as a variable, initializing it with a newly created
string byte input stream. The string input stream is
constructed from
.meta string
as if by the
.mono
.meti (make-string-byte-input-stream << string )
.onom
expression.
Then it evaluates the
.metn body-form s
in the scope of the variable.
The value of the last
.meta body-form
is returned, or else
.code nil
if no forms are present.
The
.meta string
argument must be a form
which evaluates to a character string value.
.coNP Macro @ with-out-string-stream
.synb
.mets (with-out-string-stream <> ( stream-var ) << body-form *)
.syne
.desc
The
.code with-out-string-stream
macro binds the symbol specified
by the
.meta stream-var
argument as a variable, initializing it
with a newly created string output stream. The output
stream is created as if by the
.code make-string-output-stream
function.
Then it evaluates
.metn body-form s
in the scope of that variable.
After these forms are evaluated, the string is extracted
from the string output stream, as if by the
.code get-string-from-stream
function, and returned as the result value
of the form.
.coNP Macro @ with-out-strlist-stream
.synb
.mets (with-out-strlist-stream <> ( stream-var ) << body-form *)
.syne
.desc
The
.code with-out-strlist-stream
macro binds the symbol specified
by the
.meta stream-var
argument as a variable, initializing it
with a newly created string list output stream. The output
stream is created as if by the
.code make-strlist-output-stream
function.
Then it evaluates
.metn body-form s
in the scope of that variable.
After these forms are evaluated, the string list is extracted
from the string output stream, as if by the
.code get-strlist-from-stream
function, and returned as the result value
of the form.
.coNP Function @ make-byte-input-stream
.synb
.mets (make-byte-input-stream << obj )
.syne
.desc
The
.code make-byte-input-stream
creates a stream which supports the
.code get-byte
operation for traversing a byte-wise representation of
.metn obj .
The function serves as a generic interface for calling one of
several other stream constructing functions based on the
type of the
.meta obj
argument.
The
.meta obj
argument must be either a buffer, in which case
.code make-byte-input-stream
behaves like
.codn make-buf-stream ,
or else a string, in which case the function behaves like
.codn make-string-byte-input-stream .
Note: the repertoire of types handled by
.code make-byte-input-stream
may expand in future language versions.
.coNP Function @ close-stream
.synb
.mets (close-stream < stream <> [ throw-on-error-p ])
.syne
.desc
The
.code close-stream
function performs a close operation on
.metn stream ,
whose meaning is depends on the type of the stream. For some types of streams,
such as string streams, it does nothing. For streams which are connected
to operating system files or devices, will perform a close of the underlying
file descriptor, and dissociate that descriptor from the stream. Any buffered
data is flushed first.
.code close-stream
returns a Boolean true value if the close has occurred without
errors, otherwise
.codn nil .
For most streams, "without errors" means that any buffered output data is
flushed successfully.
For command and process pipes (see
.code open-command
and
.codn open-process ),
success also
means that the process terminates normally, with a successful error code, or an
unsuccessful one. An abnormal termination is considered an error,
as is the inability to retrieve the termination status, as well as the situation
that the process continues running in spite of the close attempt.
Detecting these situations is platform specific.
If the
.meta throw-on-error-p
argument is specified, and isn't
.codn nil ,
then the
function throws an exception if an error occurs during the close operation
instead of returning
.codn nil .
If
.code close-stream
is called in such a way that it returns a value, without throwing an exception,
that value is retained. Additional calls to the function with the same
.meta stream
object return that same value without having any effect on the stream.
These additional calls ignore the
.meta throw-on-error-p
argument.
.coNP Macro @ with-stream
.synb
.mets (with-stream >> ( stream-var << init-form )
.mets \ \ << body-form *)
.syne
.desc
The
.code with-stream
macro binds the variable whose name is given by the
.meta stream-var
argument, and macro arranges for the evaluation of
.metn body-form s
in the scope of that variable.
The variable is initialized with the value produced
by the evaluation of
.meta init-form
which must be an expression which evaluates to a stream.
After each
.meta body-form
is evaluated, the stream is closed, as if by the
.mono
.meti (close-stream << stream-var )
.onom
expression.
The value of the last
.meta body-form
then becomes the result value of the form,
or else
.code nil
if these forms are absent.
If the evaluation of the
.metn body-form s
is abandoned, the stream is still closed. That is to say,
the closure of the stream is a protected action, as if by
the
.code unwind-protect
operator.
.coNP Functions @, get-error @ get-error-str and @ clear-error
.synb
.mets (get-error << stream )
.mets (get-error-str << stream )
.mets (clear-error << stream )
.syne
.desc
When a stream operation fails, the
.code get-error
and
.code get-error-str
functions may be used to inquire about a more detailed cause of the error.
Not all streams support these functions to the same extent. For instance,
string input streams have no persistent state. The only error which occurs
is the condition when the string has no more data.
The
.code get-error
inquires
.meta stream
about its error condition.
The function returns
.code nil
to indicate there is no error condition,
.code t
to indicate an end-of-data condition,
or else a value which is specific to the stream type indicating the
specific error type.
Note: for some streams, it is possible for the
.code t
value to be returned even though no operation has failed; that is to say, the
streams "know" they are at the end of the data even though no read operation
has failed. Code which depends on this will not work with streams which
do not thus indicate the end-of-data
.IR "a priori" ,
but by means of a read operation which fails.
The
.code get-error-str
function returns a text representation of the error code. The
.code nil
error code is represented as the string
.codn "no error" ;
the
.code t
error code as
.code "eof"
and other codes have a stream-specific representation.
The
.code clear-error
function removes the error situation from a stream. On some streams, it does
nothing. If an error has occurred on a stream, this function should be called
prior to retrying any I/O or positioning operations.
The return value is the previous error code, or
.code nil
if there was no error, or the operation is not supported on the stream.
.coNP Functions @, get-line @ get-char and @ get-byte
.synb
.mets (get-line <> [ stream ])
.mets (get-char <> [ stream ])
.mets (get-byte <> [ stream ])
.syne
.desc
These fundamental stream functions perform input. The
.meta stream
argument
is optional. If it is specified, it should be an input stream which supports
the given operation. If it is not specified, then the
.code *stdin*
stream is used.
The
.code get-char
function pulls a character from a stream which supports character
input. Streams which support character input also support the
.code get-line
function which extracts a line of text delimited by the end of the stream or a
newline character and returns it as a string. (The newline character does not
appear in the string which is returned).
Character input from streams based on bytes requires UTF-8 decoding, so that
.code get-char
may actually read several bytes from the underlying low-level
operating system stream.
The
.code get-byte
function bypasses UTF-8 decoding and reads raw bytes from
any stream which supports byte input. Bytes are represented as integer
values in the range 0 to 255.
Note that if a stream supports both byte input and character input, then mixing
the two operations will interfere with the UTF-8 decoding.
These functions return
.code nil
when the end of data is reached. Errors are
represented as exceptions.
See also:
.code get-lines
.coNP Function @ get-string
.synb
.mets (get-string >> [ stream >> [ count <> [ close-after-p ]]])
.syne
.desc
The
.code get-string
function reads characters from a stream, and assembles them into
a string, which is returned. If the
.meta stream
argument is omitted, then the
.code *stdin*
stream is used.
The stream is closed after extracting the data, unless
.meta close-after-p
is specified as
.codn nil .
The default value of this argument is
.codn t .
If the
.meta count
argument is missing, then all of the characters from the
stream are read and assembled into a string.
If present, the
.meta count
argument should be a positive integer indicating
a limit on how many characters to read. The returned string will be no
longer than
.metn count ,
but may be shorter.
.coNP Functions @ unget-char and @ unget-byte
.synb
.mets (unget-char < char <> [ stream ])
.mets (unget-byte < byte <> [ stream ])
.syne
.desc
These functions put back, into a stream, a character or byte which was
previously read. The character or byte must match the one which was most
recently read. If the
.meta stream
argument is omitted, then the
.code *stdin*
stream is used.
If the operation succeeds, the byte or character value is returned.
A
.code nil
return indicates that the operation is unsupported.
Some streams do not support these operations; some support
only one of them. In general, if a stream supports
.codn get-char ,
it supports
.codn unget-char ,
and likewise for
.code get-byte
and
.codn unget-byte .
Streams may require a pushed back byte or character to match
the character which was previously read from that stream
position, and may not allow a byte or character to be pushed
back beyond the beginning of the stream.
Space may be available for only one byte of pushback under the
.code unget-byte
operation.
The number of characters that may be pushed back by
.code unget-char
is not limited.
Pushing both a byte and a character, in either order, is also unsupported.
Pushing a byte and then reading a character, or pushing a character and
reading a byte, are unsupported mixtures of operations.
If the stream is binary, then pushing back a byte decrements its position,
except if the position is already zero. At that point, the position becomes
indeterminate.
The behavior of pushing back immediately after a
.code seek-stream
positioning operation is unspecified.
.coNP Functions @, put-string @, put-line @ put-char and @ put-byte
.synb
.mets (put-string < string <> [ stream ])
.mets (put-line >> [ string <> [ stream ]])
.mets (put-char < char <> [ stream ])
.mets (put-byte < byte <> [ stream ])
.syne
.desc
These functions perform output on an output stream. The
.meta stream
argument
must be an output stream which supports the given operation. If it is omitted,
then
.code *stdout*
is used.
The
.code put-char
function writes a character given by
.code char
to a stream. If the
stream is based on bytes, then the character is encoded into UTF-8 and multiple
bytes are written. Streams which support
.code put-char
also support
.code put-line
and
.codn put-string .
The
.code put-string
function writes the characters of a string out to
the stream as if by multiple calls to
.codn put-char .
The
.meta string
argument
may be a symbol, in which case its name is used as the string.
The
.code put-line
function is like
.codn put-string ,
but also writes an additional newline
character. The string is optional in
.codn put-line ,
and defaults to the empty string.
The
.code put-byte
function writes a raw byte given by the
.meta byte
argument
to
.metn stream ,
if
.meta stream
supports a byte write operation. The byte
value is specified as an integer value in the range 0 to 255.
All these functions return
.codn t .
On failure, they do not return, but throw exceptions of type
.codn file-error .
.coNP Functions @ put-strings and @ put-lines
.synb
.mets (put-strings < sequence <> [ stream ]])
.mets (put-lines < sequence <> [ stream ]])
.syne
.desc
These functions assume
.meta sequence
to be a sequence of strings, or of
symbols, or a mixture thereof. These strings are sent to the stream. The
.meta stream
argument must be an output stream. If it is omitted, then
.code *stdout*
is used.
The
.code put-strings
function iterates over
.meta sequence
and writes each element
to the stream as if using the
.code put-string
function.
The
.code put-lines
function iterates over
.code sequence
and writes each element
to the stream as if using the
.code put-line
function.
Both functions return
.codn t .
.coNP Function @ flush-stream
.synb
.mets (flush-stream <> [ stream ])
.syne
.desc
The
.code flush-stream
function is meaningful for output streams which accumulate data
which is passed on to the operating system in larger transfer units.
Calling
.code flush-stream
causes all accumulated data inside
.meta stream
to be passed
to the operating system. If called on streams for which this function is not
meaningful, it does nothing, and returns
.codn nil .
If
.meta stream
is omitted, the current value of
.code *stdout*
is used.
.coNP Function @ seek-stream
.synb
.mets (seek-stream < stream < offset << whence )
.syne
.desc
The
.code seek-stream
function is meaningful for file streams. It changes the
current read/write position within
.metn stream .
It can also be used to determine the current position: see the notes about the
return value below.
The
.meta offset
argument is a positive or negative integer which gives a
displacement that is measured from the point identified by the
.meta whence
argument.
Note that for text files, there isn't necessarily a 1:1 correspondence between
characters and positions due to line-ending conversions and conversions
to and from UTF-8.
The
.meta whence
argument is one of three keywords:
.codn :from-start ,
.code :from-current
and
.codn :from-end .
These denote the start of the file, the current position in the file
and the end of the file.
If
.meta offset
is zero, and
.meta whence
is
.codn :from-current ,
then
.code seek-stream
returns the current absolute position within the
stream, if it can successfully obtain it. Otherwise, it
returns
.code t
if it is successful.
If a character has been successfully put back into a text stream with
.code unget-char
and is still pending, then the position value is unspecified. If a
byte has been put back into a binary stream with
.codn unget-byte ,
and the previous position wasn't zero, then the position is decremented by one.
On failure, it throws an exception of type
.codn stream-error .
.coNP Function @ truncate-stream
.synb
.mets (truncate-stream < stream <> [ length ])
.syne
.desc
The
.code truncate-stream
causes the length of the underlying file associated with
.meta stream
to be set to
.meta length
bytes.
The stream must be a file stream, and must be open for writing.
If
.meta length
is omitted, then it defaults to the current position, retrieved
as if by invoking the
.code seek-stream
with an
.meta offset
argument of zero and
.meta whence
argument of
.codn :from-current .
Hence, after the
.code truncate-stream
operation, that position is one byte past the end of the file.
.coNP Functions @ stream-get-prop and @ stream-set-prop
.synb
.mets (stream-get-prop < stream << indicator )
.mets (stream-set-prop < stream < indicator << value )
.syne
.desc
These functions get and set properties on a stream. Only certain properties
are meaningful with certain kinds of streams, and the meaning depends on
the stream. If two or more stream types support a property of the same name, it
is expected that the property has the same or similar meaning for both
streams to the maximum extent that similarity is possible.
The
.code stream-set-prop
function sets a property on a stream. The
.meta indicator
argument is a symbol, usually a keyword symbol, denoting the property,
and
.meta value
is the property value. If the stream understands and accepts the
property, the function returns
.codn t .
Otherwise it returns
.codn nil .
The
.code stream-get-prop
function inquires about the value of a property on a
stream. If the stream understands the property, then it returns its current
value. If the stream does not understand a property, nil is returned, which is
also returned if the property exists, but its value happens to be
.codn nil .
The
.code :name
property is widely supported by streams of various types. It associates
the stream with a name. This property is not always modifiable.
File, process and stream socket I/O streams have a
.code :fd
property which can be accessed, but not modified. It retrieves
the same value as the
.code fileno
function.
The "real time"
property supported by these streams, connected with the
.code real-time-stream-p
function, also appears as the
.code :real-time
property.
I/O streams also have a property called
.code :byte-oriented
which, if set, suppresses the decoding of UTF-8 on character input. Rather,
each byte of the file corresponds directly to one character. Bytes
in the range 1 to 255 correspond to the character code points U+0001
to U+00FF. Byte value 0 is mapped to the code point U+DC00.
The logging priority of the
.code *stdlog*
syslog stream is controlled by the
.code :prio
property.
If
.meta stream
is a catenated stream (see the function
.codn make-catenated-stream )
then these functions transparently operate on the current head stream of the
catenation.
.coNP Functions @ make-catenated-stream and @ cat-streams
.synb
.mets (make-catenated-stream << stream *)
.mets (cat-streams << stream-list )
.syne
.desc
The
.code make-catenated-stream
function takes zero or more arguments which
are input streams of the same type, and combines
them into a single virtual stream called a catenated stream.
The
.code cat-streams
function takes a single list of input streams of
the same type, and similarly combines them into a catenated stream.
A catenated stream does not support seeking operations or output,
regardless of the capabilities of the streams in the list.
If the stream list is not empty, then the leftmost element of the
list is called the head stream.
The
.codn get-char ,
.codn get-byte ,
.codn get-line ,
.code unget-char
and
.code unget-byte
functions delegate
to the corresponding operations on the head stream, if it exists.
If the stream list is empty, they return
.code nil
to the caller.
If the
.codn get-char ,
.code get-byte
or
.code get-line
operation on the head stream yields
.codn nil ,
and there are more lists in the stream, then the stream is closed, removed from
the list, and the next stream, if any, becomes the head list. The operation is
then tried again. If any of these operations fail on the last list, it is not
removed from the list, so that a stream remains in place which can take the
.code unget-char
or
.code unget-byte
operations.
In this manner, the catenated streams appear to be a single stream.
Note that the operations can fail due to being unsupported. It is
the caller's responsibility to make sure all of the streams in the list
are compatible with the intended operations.
If the stream list is empty then an empty catenated stream is produced.
Input operations on this stream yield
.codn nil ,
and the
.code unget-char
and
.code unget-byte
operations throw an exception.
.coNP Function @ catenated-stream-p
.synb
.mets (catenated-stream-p << obj )
.syne
.desc
The
.code catenated-stream-p
function returns
.code t
if
.meta obj
is a catenated stream. Otherwise it returns
.codn nil .
.coNP Function @ catenated-stream-push
.synb
.mets (catenated-stream-push < new-stream << cat-stream )
.syne
.desc
The
.code catenated-stream-push
function pushes
.meta new-stream
to the front of the stream list inside
.metn cat-stream .
If an
.code unget-byte
or
.code unget-char
operation was successfully performed on
.meta cat-stream
previously to a call to
.codn catenated-stream-push ,
those operations were forwarded to the front stream.
If those bytes or characters are still pending,
they are pending inside that stream, and thus
are logically preceded by the contents
of
.metn new-stream .
.coNP Functions @ open-files and @ open-files*
.synb
.mets (open-files < path-list >> [ alternative-stream <> [ mode-string ]])
.mets (open-files* < path-list >> [ alternative-stream <> [ mode-string ]])
.syne
.desc
The
.code open-files
and
.code open-files*
functions create a list of streams by invoking
the
.code open-file
function on each element of
.metn path-list .
By default, the mode string
.str r
is passed to
.codn open-file ;
if the
.meta mode-string
argument specified, it overrides this default. In that situation,
the specified mode should permit reading.
These streams are turned
into a catenated stream as if they were the arguments of a call to
.codn make-catenated-stream .
The effect is that multiple files appear to be catenated together into a single
input stream.
If the optional
.meta alternative-stream
argument is supplied, then if
.meta path-list
is empty,
.meta alternative-stream
is returned instead of an empty catenated stream.
The difference between
.code open-files
and
.code open-files*
is that
.code open-files
creates all of the
streams up-front. So if any of the paths cannot be opened, the operation throws.
The
.code open-files*
variant is lazy: it creates a lazy list of streams out of the
path list. The streams are opened as needed: before the second stream is opened,
the program has to read the first stream to the end, and so on.
.TP* Example:
Collect lines from all files that are given as arguments on the command line. If
there are no files, then read from standard input:
.verb
@(next (open-files *args* *stdin*))
@(collect)
@line
@(end)
.brev
.coNP Function @ path-equal
.synb
.mets (path-equal < left-path << right-path )
.syne
.desc
The
.code path-equal
function determines whether the two paths
.meta left-path
and
.meta right-path
are equal under a certain definition of equivalence, whose requirements are given below.
The function returns
.code t
if the paths are equal, otherwise
.codn nil .
If
.meta left-path
and
.meta right-path
are strings which are identical under the
.code equal
function, then they are considered equal paths.
Otherwise, the two paths are equal if the relative path from
.meta left-path
to
.meta right-path
is
.str .
(dot), as would be determined by the
.code path-rel
function, if it were applied to
.meta left-path
and
.meta right-path
as its arguments. If
.code path-rel
would return the dot path, then the two paths are equal. If
.code path-rel
would return any other value, or throw an exception, then the paths are unequal.
.TP* Examples:
.verb
;; simple case
(path-equal "a" "a") -> t
(path-equal "a" "b") -> nil
;; trailing slashes don't matter
(path-equal "a" "a/") -> t
(path-equal "a/" "a/") -> t
;; .. components resolved:
(path-equal "a/b/../c" "a/c") -> t
;; . components resolved:
(path-equal "a" "a/././.") -> t
(path-equal "a/." "a/././.") -> t
;; (On Microsoft Windows)
;; different drive:
(path-equal "c:/a" "d:/b/../a") -> nil
;; same drive:
(path-equal "c:/a" "c:/b/../a") -> t
.brev
.coNP Functions @ abs-path-p and @ portable-abs-path-p
.synb
.mets (abs-path-p << path )
.mets (portable-abs-path-p << path )
.syne
.desc
The
.code abs-path-p
and
.code portable-abs-path-p
functions test whether the argument
.meta path
is an absolute path, returning a
.code t
or
.code nil
indication.
The
.code portable-abs-path-p
function behaves in the same manner on all platforms, implementing
a platform-agnostic definition of
.IR "absolute path" ,
as follows.
An absolute path is a string which either begins with a slash or backslash
character, or which begins with an alphanumeric word, followed by a colon,
followed by a slash or backslash.
The empty string isn't an absolute path.
Examples of absolute paths under
.codn portable-abs-path-p :
.verb
/etc
c:/tmp
ftp://user@server
disk0:/home
Z:\eUsers
.brev
Examples of strings which are not absolute paths:
.verb
.
abc
foo:bar/x
$:\eabc
.brev
The
.code abs-path-p
is similar to
.code portable-abs-path-p
except that it reports false for paths which are not absolute paths
according to the host platform. The following paths are not absolute
on POSIX platforms:
.verb
c:/tmp
ftp://user@server
disk0:/home
Z:\eUsers
.brev
.coNP Function @ pure-rel-path-p
.synb
.mets (pure-rel-path-p << path )
.syne
.desc
The
.code pure-rel-path-p
function tests whether the string
.meta path
represents a
.IR "pure relative path" ,
which is defined as a path which isn't absolute according to
.codn abs-path-p ,
which isn't the string
.str .
(single period),
which doesn't begin with a period followed by a slash or backslash,
and which doesn't begin with an alphanumeric word
terminated by a colon.
The empty string is a pure relative path.
Other examples of pure relative paths:
.verb
abc.d
.tmp/bar
1234
x
$:/xyz
.brev
Examples of strings which are not pure relative paths:
.verb
.
/
/etc
./abc
.\e
foo:
$:\eabc
.brev
.coNP Functions @ dir-name and @ base-name
.synb
.mets (dir-name << path )
.mets (base-name < path <> [ suffix ])
.syne
.desc
The
.code dir-name
and
.code base-name
functions calculate, respective, the directory part and
base name part of a pathname.
The calculation is performed in a platform-dependent way, using the
characters in the variable
.code path-sep-chars
as path component separators.
Both functions first remove from any further consideration all superfluous
trailing occurrences of the directory separator characters from
.codn path .
Thus input such as
.str "a////"
is reduced to just
.strn "a" ,
and
.str "///"
is reduced to
.strn "/" .
The resulting trimmed path is the
.I "effective path" .
If the effective path is an empty string, then
.code dir-name
returns
.str "."
and
.code base-name
returns the empty string.
If the effective path is not empty, and contains no path separator
characters, then
.code dir-name
returns
.str "."
and
.code base-name
returns the effective path.
Otherwise, the effective path is divided into two parts: the
.I "raw directory prefix"
and the remainder.
The raw directory prefix is the maximally long prefix of the effective
path which ends in a separator character.
The
.code dir-name
function returns the raw directory prefix, if that prefix consists of
nothing but a single directory separator character. Otherwise it
returns the raw directory prefix, with the trailing path separator
removed.
The
.code base-name
function returns the remaining part of the effective path, after
the raw directory prefix.
If the
.meta suffix
argument is given to
.codn base-name ,
it specifies a proper suffix to be removed from the returned base name.
First, the base name is calculated according to the foregoing rules.
Then, if
.meta suffix
matches a trailing portion of the base name, but not the entire base name,
it is removed from the base name.
The
.meta suffix
parameter may be given a
.codn nil ,
argument, which is treated exactly as if it were absent.
Note: this requirement allows for the following idiom
to work correctly even in cases when
.code p
has no suffix:
.verb
;; calculate base name of p with short suffix removed
(base-name p (short-suffix p))
;; calculate base name of p with long suffix removed
(base-name p (long-suffix p))
.brev
.TP* Examples:
.verb
(base-name "") -> ""
(base-name "/") -> "/"
(base-name ".") -> "."
(base-name "./") -> "."
(base-name "a") -> "a"
(base-name "/a") -> "a"
(base-name "/a/") -> "a"
(base-name "/a/b") -> "b"
(base-name "/a/b/") -> "b"
(base-name "/a/b///") -> "b"
;; with suffix
(base-name "" "") -> ""
(base-name "/" "/") -> "/"
(base-name "/" "") -> "/"
(base-name "." ".") -> "."
(base-name "." "") -> "."
(base-name "./" "/") -> "."
(base-name "a" "a") -> "a"
(base-name "a" "") -> "a"
(base-name "a.b" ".b") -> "a"
(base-name "a.b/" ".b") -> "a"
(base-name "a.b/" ".b/") -> "a.b"
(base-name "a.b/" "a.b") -> "a.b"
.brev
.coNP Functions @ long-suffix and @ short-suffix
.synb
.mets (long-suffix < path <> [ alt ])
.mets (short-suffix < path <> [ alt ])
.syne
.desc
The
.code long-suffix
and
.code short-suffix
functions calculate the
.I "long suffix"
and
.I "short suffix"
of
.metn path ,
which must be a string.
If
.meta path
does not contain any occurrences of the character
.code .
(period) in the role of a suffix delimiter, then
.meta path
does not have a suffix. In this situation, both
functions return the
.meta alt
argument, which defaults to
.code nil
if it is omitted.
What it means for
.meta path
to have a suffix delimiter is that the
.code .
character occurs somewhere in the last component of
.metn path ,
other than as the first character of that component.
What constitutes the last component is specified
in more detail below.
If a suffix delimiter is present, then the long or short suffix is the
substring of
.meta path
which includes the delimiting period and all characters which follow,
except that if
.meta path
ends in a sequence of one or more path separator characters,
those characters are omitted from the returned suffix.
If multiple periods occur in the last component of the path,
the delimiter for the long suffix is the leftmost period and
the delimiter for the short suffix is the rightmost period.
If the delimiting period is the rightmost character of
.metn path ,
or occurs immediately before a trailing path separator,
then the suffix delimited by that period is the period itself.
If
.meta path
contains only one suffix delimiter, then its long and short suffix coincide.
For the purpose of identifying the last component of
.metn path ,
if
.meta path
ends a sequence of one or more path-separator characters, then those
characters are removed from consideration.
If the remaining string contains path-separator characters, then
the last component consists of that portion of it which follows
the rightmost path-separator character. Otherwise, the last component
is the entire string. The suffix, if present, is identified and
extracted from this last component.
.TP* Examples:
.verb
(short-suffix "") -> nil
(short-suffix ".") -> nil
(short-suffix "abc") -> nil
(short-suffix ".abc") -> nil
(short-suffix "/.abc") -> nil
(short-suffix "abc" "") -> ""
(short-suffix "abc.") -> "."
(short-suffix "abc.tar") -> ".tar"
(short-suffix "abc.tar///") -> ".tar"
(short-suffix "abc.tar.gz") -> ".gz"
(short-suffix "abc.tar.gz/") -> ".gz"
(short-suffix "x.y.z/abc.tar.gz/") -> ".gz"
(short-suffix "x.y.z/abc.tar.gz//") -> nil
(long-suffix "") -> nil
(long-suffix ".") -> nil
(long-suffix "abc") -> nil
(long-suffix ".abc") -> nil
(long-suffix "/.abc") -> nil
(long-suffix "abc.") -> "."
(long-suffix "abc.tar") -> ".tar"
(long-suffix "abc.tar///") -> ".tar"
(long-suffix "abc.tar.gz") -> ".tar.gz"
(long-suffix "abc.tar.gz/") -> ".tar.gz"
(long-suffix "x.y.z/abc.tar.gz/") -> ".tar.gz"
.brev
.coNP Functions @ trim-long-suffix and @ trim-short-suffix
.synb
.mets (trim-long-suffix << path )
.mets (trim-short-suffix << path )
.syne
.desc
The
.code trim-long-suffix
and
.code trim-short-suffix
functions calculate the portion of
.meta path
.I "long suffix"
and
.I "short suffix"
of the string argument
.metn path ,
and return a path with the suffix removed.
Respectively,
.code trim-long-suffix
and
.code trim-short-suffix
calculate the suffix in exactly the same manner as
.code long-suffix
and
.codn short-suffix .
If
.meta path
is found not to contain a suffix, then it is returned.
If
.meta path
contains a suffix, then a new string is returned from which
the suffix is deleted. If the suffix is followed by one or more path separator
characters, these are preserved in the return value.
.TP* Examples:
.verb
(trim-short-suffix "") -> ""
(trim-short-suffix "a") -> "a"
(trim-short-suffix ".") -> "."
(trim-short-suffix ".a") -> ".a"
(trim-short-suffix "a.") -> "a"
(trim-short-suffix "a.b") -> "a"
(trim-short-suffix "a.b.c") -> "a.b"
(trim-short-suffix "a./") -> "a/"
(trim-short-suffix "a.b/") -> "a/"
(trim-short-suffix "a.b.c/") -> "a.b/"
(trim-long-suffix "a.b.c") -> "a"
(trim-long-suffix "a.b.c/") -> "a/"
(trim-long-suffix "a.b.c///") -> "a///"
.brev
.coNP Function @ add-suffix
.synb
.mets (add-suffix < path << suffix )
.syne
.desc
The
.code add-suffix
function combines the string arguments
.meta path
and
.meta suffix
in a way which harmonizes with the
.code long-suffix
and
.code short-suffix
functions.
If
.meta path
does not end in a path separator character, that category being defined by the
.code path-sep-chars
variable, then
.code add-suffix
returns the trivial string catenation of
.meta path
and
.metn suffix .
Otherwise,
.code add-suffix
returns a string formed by inserting
.meta suffix
into
.meta path
just prior to the sequence of trailing path separator characters.
The returned string is a catenation of that portion of
.meta path
which excludes the sequence of trailing path separators,
followed by
.metn suffix ,
followed by the sequence of trailing path separators.
A path separator which occurs as a part of syntax that indicates an absolute
pathname is not considered a trailing separator. A path which begins with a
separator is absolute. Other platform-specific path patterns may constitute
an absolute pathname.
Note: in cases when
.meta suffix
does not begin with a period, or is inserted in such a way
that it is the start of a path component, then the functions
.code long-suffix
and
.code short-suffix
will not recognize
.meta suffix
in the resulting path.
.TP* Examples:
.verb
(add-suffix "" "") -> ""
(add-suffix "" "a") -> "a"
(add-suffix "." "a") -> ".a"
(add-suffix "." ".a") -> "..a"
(add-suffix "/" ".b") -> "/.b"
(add-suffix "//" ".b") -> "/.b/"
(add-suffix "//" "b") -> "/b/"
(add-suffix "a" "") -> "a"
(add-suffix "a" ".b") -> "a.b"
(add-suffix "a/" ".b") -> "a.b/"
(add-suffix "a//" ".b") -> "a.b//"
;; On MS Windows
(add-suffix "c://" "x") -> "c:/x/"
(add-suffix "host://" "x") -> "host://x"
(add-suffix "host:///" "x") -> "host://x/"
.brev
.coNP Function @ path-cat
.synb
.mets (path-cat >> [ dir-path <> { rel-path }*])
.syne
.desc
The
.code path-cat
function joins together zero or more paths, returning the combined path.
All arguments are strings.
The following description defines the behavior when
.code path-cat
is given exactly two arguments, which are interpreted as
.meta dir-path
and
.metn rel-path .
A description of the variable-argument semantics follows.
Firstly, the two-argument
.code path-cat
is related to the functions
.code dir-name
and
.code base-name
in the following way: if
.meta p
is some path denoting an object in the file system, then
.code "(path-cat (dir-name p) (base-name p))"
produces a path
.meta p*
which denotes the same object. The paths
.meta p
and
.meta p*
might not be equivalent strings.
The
.code path-cat
function ensures that paths are joined without superfluous
path-separator characters, regardless of whether
.meta dir-path
ends in a separator.
If a separator must be added, the character
.code /
(forward slash) is always used, even on platforms where
.code \e
(backslash) is also a pathname separator, and even if either argument includes
backslashes.
The
.code path-cat
function eliminates trivial occurrences of the
.code .
(dot) path component. It preserves trailing separators in the following
way: if
.meta rel-path
ends in a path-separator character, then the returned string shall
end in that character; and if
.meta rel-path
vanishes entirely because it is equivalent to the dot, then the returned
string is
.meta dir-name
itself.
If
.meta dir-path
is an empty string, then
.code rel-path
is returned, and vice versa.
The variadic semantics of
.code path-cat
are as follows.
If
.code path-cat
is called with no arguments at all, it returns the path
.str .
(period) denoting the relative path of the current directory.
If
.code path-cat
is called with one argument, that argument is returned.
If
.code path-cat
is called with three or more arguments, a left-associative reduction
takes place using the two-argument semantics. The first two arguments
are catenated into a single path, which is then catenated
with the third argument, and so on.
The above semantics imply that the following equivalence holds:
.verb
[reduce-left path-cat list] <--> [apply path-cat list]
.brev
.TP* Examples:
.verb
(path-cat "" "") --> ""
(path-cat "" ".") --> "."
(path-cat "." "") --> "."
(path-cat "." ".") --> "."
(path-cat "abc" ".") --> "abc"
(path-cat "." "abc") --> "abc"
(path-cat "./" ".") --> "./"
(path-cat "." "./") --> "./"
(path-cat "abc/" ".") --> "abc/"
(path-cat "./" "abc") --> "abc"
(path-cat "/" ".") --> "/"
(path-cat "/" "abc") --> "/abc"
(path-cat "ab/cd" "ef") --> "ab/cd/ef"
(path-cat "a" "b" "c") --> "a/b/c"
(path-cat "a" "b" "" "c" "/") --> "a/b/c/"
.brev
.coNP Function @ rel-path
.synb
.mets (rel-path < from-path << to-path )
.syne
.desc
The
.code rel-path
function calculates the relative path between two file system locations
indicated by string arguments
.meta from-path
and
.metn to-path .
The
.meta from-path
is assumed to be a directory. The return value is a relative path
which could be used to access an object named by
.meta to-path
if
.meta from-path
were the current working directory.
The calculation performed by
.code rel-path
is a pure calculation; it has no interaction with the host operating system.
No component of either input path has to exist. Symbolic links are not
resolved. This can lead to incorrect results, as noted below.
Either both the inputs must be absolute paths, or must both be relative,
otherwise an error exception is thrown.
On the MS Windows platform, if one input specifies a drive letter prefix, the
other input must specify the same prefix, or else an error exception is thrown;
there is no relative path between locations on different drives.
The behavior is unspecified if the arguments are two UNC paths indicating
different hosts.
The
.code rel-path
function first splits both paths into components according to the
platform-specific pathname separators indicated by the
.code path-sep-chars
variable.
Next, it eliminates all empty components,
.code .
(dot) components and
.code ..
(dotdot)
components from both separated paths. All dot components are removed,
and any component which is neither dot nor dotdot is removed if it is
followed by dotdot.
Then, a common prefix is determined between the two component sequences,
and a relative component sequence is calculated from them as follows:
If the component sequence corresponding to
.meta from-path
is longer than the common prefix, then the excess part of that
sequence after the common prefix must not contain any
.code ..
(dotdot) components, or else an error exception is thrown.
Otherwise, every component in this excess part of the
.meta from-path
component sequence is converted to
.code ..
in order to express the relative navigation from
.meta from-path
up to the directory indicated by the common prefix.
Next, if the component sequence corresponding to
.meta to-path
has any components in excess of the common prefix, those excess components are
appended to this possibly empty sequence of dotdot components, in
order to express navigation from the common prefix down to the
.meta to-path
object. This excess sequence coming from
.meta to-path
may include
.code ..
components.
Finally, if the resulting sequence is nonempty, it is joined together using the leftmost
path separator character indicated in
.code path-sep-chars
and returned. If it is empty, then the string
.str .
is returned.
Note: because the function doesn't access the file system and in particular
does not resolve symbolic links or other indirection devices, the result
may be incorrect. For example, suppose that the current working directory
contains a symbolic link called
.code up
which expands to
.code ..
(dotdot). The expression
.code "(rel-path \(dqup/a\(dq \(dq../a\(dq)"
is oblivious to this, and calculates
.strn ../../../a .
The correct result in light of
.code up
being an alias for
.code ..
calls for a return value of
.strn . .
The exact problem is that any symbolic links in the excess part of
.meta from-path
after the common prefix are assumed by
.code rel-path
to be simple subdirectory names, which can be navigated in reverse
using a
.code ..
link. This reverse navigation assumption is false for any symbolic link which
which does not act as an alias for a subdirectory in the same location.
In situations where this possibility exists, it is recommended to use
.code realpath
function to canonicalize the input paths.
The following is an example of the algorithm being applied to arguments
.str a/d/../b/x/y/
and
.strn a/b/w ,
where the assumption is that this is on a POSIX platform where the leftmost
character in
.code path-sep-chars
is
.codn / :
Firstly, both inputs are converted to component sequences, those respectively being:
.verb
("a" "d" ".." "b" "x" "y" "")
("a" "b" "w")
.brev
Next the
.code ..
and empty components are removed:
.verb
("a" "b" "x" "y")
("a" "b" "w")
.brev
At this point, the common prefix is identified:
.verb
("a" "b")
.brev
The
.meta from-path
has two components in excess of the prefix:
.verb
("x" "y")
.brev
which are each replaced by
.strn .. .
The
.meta to-path
has one component in excess of the common prefix,
.strn w .
These two sequences are appended together:
.verb
(".." ".." "w")
.brev
The resulting path is then formed by joining these with the separator
character, resulting in the relative path
.strn "../../w" .
.TP* Examples:
.verb
;; mixtures of relative and absolute
(rel-path "/abc" "abc") -> ;; error
(rel-path "abc" "/abc") -> ;; error
;; dotdot in excess part of from path:
(rel-path "../../x" "y") -> ;; error
(rel-path "." ".") -> "."
(rel-path "./abc" "abc") -> "."
(rel-path "abc" "./abc") -> "."
(rel-path "./abc" "./abc") -> "."
(rel-path "abc" "abc") -> "."
(rel-path "." "abc") -> "abc"
(rel-path "abc/def" "abc/ghi") -> "../ghi"
(rel-path "xyz/../abc/def" "abc/ghi") -> "../ghi"
(rel-path "abc" "d/e/f/g/h") -> "../d/e/f/g/h"
(rel-path "abc" "d/e/../g/h") -> "../d/g/h"
(rel-path "d/e/../g/h" ".") -> "../../.."
(rel-path "d/e/../g/h" "a/b") -> "../../../a/b"
(rel-path "x" "../../../y") -> "../../../../y"
(rel-path "x///" "x") -> "."
(rel-path "x" "x///") -> "."
(rel-path "///x" "/x") -> "."
.brev
.coNP Variable @ path-sep-chars
.desc
The
.code path-sep-chars
variable holds a string consisting of the characters which the underlying
operating system recognizes as pathname separators.
If a particular of these characters is considered preferred on
the host platform, that character is placed in the first position of
.codn path-sep-chars .
Altering the value of this variable has no effect on any \*(TL library
function.
.coNP Functions @ read and @ iread
.synb
.mets (read >> [ source
.mets \ \ \ \ \ \ >> [ err-stream >> [ err-retval >> [ name <> [ lineno ]]]]])
.mets (iread >> [ source
.mets \ \ \ \ \ \ \ >> [ err-stream >> [ err-retval >> [ name <> [ lineno ]]]]])
.syne
.desc
The
.code read
function converts text denoting \*(TL structure, into the
corresponding data structure. The
.meta source
argument may be either a character
string, or a stream. If it is omitted, then
.code *stdin*
is used as the stream.
The
.meta source
must provide the text representation of one complete \*(TL object.
If
.meta source
and the function being applied is
.codn read ,
then if the object is followed by any non-whitespace material, the
situation is treated as a syntax error, even if that material is
a syntactically valid additional object.
The
.code iread
function ignores this situation. Other differences between
.code read
and
.code iread
are given below.
Multiple calls to
.code read
on the same stream will extract successive objects from the stream.
To parse successive objects from a string, it is necessary
to convert it to a string stream.
The optional
.meta err-stream
argument can be used to specify a stream to which
diagnostics of parse errors are sent.
If absent, the diagnostics are suppressed.
The optional
.meta name
argument can be used to specify the file name which is used for reporting
errors. If this argument is missing, the name is taken from the name
property of the
.meta source
argument if it is a stream, or else the word
.code string
is used as the name if
.meta source
is a string.
The optional
.code lineno
argument, defaulting to 1, specifies the starting line number. This,
like the
.meta name
argument, is used for reporting errors.
If there are no parse errors, the function returns the parsed data
structure. If there are parse errors, and the
.meta err-retval
parameter is
present, its value is returned. If the
.meta err-retval
parameter
is not present, then an exception of type
.code syntax-error
is thrown.
The
.code iread
function ("interactive read") is similar to
.code read
except that it parses a modified version of the syntax. The modified
syntax does not support the application of the dot and dotdot operators
on a top-level expression. For instance, if the input is
.code a.b
or
.code "a .. b"
then
.code iread
will only read the
.code a
token whereas
.code read
will read the entire expression.
This modified syntax allows
.code iread
to return immediately when an expression is recognized, which is the
expected behavior if the input is being read from an interactive terminal.
By contrast,
.code read
waits for more input after seeing a complete expression, because of the
possibility that the expression will be further extended by means of the dot or
dotdot operators. An explicit end-of-input signal must be given from the
terminal to terminate the expression.
The special variable
.code *rec-source-loc*
controls whether these functions record source location info similarly to
.codn load .
Note: if these functions are used to scan data which is evaluated as Lisp code,
it may be useful to set
.code *rec-source-loc*
true in order to obtain better diagnostics. However, source location recording
incurs a performance and storage penalty.
.coNP Function @ parse-errors
.synb
.mets (parse-errors << stream )
.syne
.desc
The
.code parse-errors
function retrieves information, from a
.metn stream ,
pertaining to the status of the most recent parsing operation performed
on that stream: namely, a previous call to
.codn read ,
.code iread
or
.codn get-json .
If the
.meta stream
object has not been used for parsing, or else the most recent
parsing operation did not encounter errors, then
.code parse-errors
returns
.codn nil .
If the most recent parsing operation on
.meta stream
encountered errors, then
.code parse-errors
function returns a positive integer value indicating the error count.
Otherwise it returns
.codn nil .
If a parse error operation encounters a syntax error before obtaining any token
from the stream, then the error count is zero and
.code parse-errors
returns
.codn nil .
Consequently,
.code parse-errors
may be used after a failed parse operation to distinguish a true
syntax error from an end-of-stream condition.
.coNP Function @ record-adapter
.synb
.mets (record-adapter < regex >> [ stream <> [ include-match ]])
.syne
.desc
The
.code record-adapter
function returns a new stream object which acts as an
.I adapter
to the existing
.metn stream .
If an argument is not specified for
.metn stream ,
then the
.code *std-input*
stream is used.
With the exception of
.metn get-line ,
all operations on the returned adapter transparently delegate to the original
.meta stream
object.
When the
.code get-line
function is used on the adapter, it behaves differently. A string is
extracted from
.metn stream ,
and returned. However, the string isn't a line delimited by a newline
character, but rather a record delimited by
.metn regex .
This record is extracted as if by a call to the
.code read-until-match
function, invoked with the
.metn regex ,
.meta stream
and
.meta include-match
arguments.
All behavior which is built on the
.code get-lines
function is affected by the record-delimiting semantics of a record adapter's
.code get-line
implementation. Notably, the
.code get-lines
and
.code lazy-stream-cons
functions return a lazy list of delimited records rather than of lines.
.SS* Stream Output Indentation
\*(TL streams provide support for establishing hanging indentations
in text output. Each stream which supports output has a built-in state variable
called indentation mode, and another variable indicating the current
indentation amount. When indentation mode is enabled, then prior to the
first character of every line, the stream prepends the indentation: space
characters equal in number to the current indentation value.
This logic is implemented by the
.code put-char
and
.code put-string
functions, and all functions based on these. The
.code put-byte
function does not interact with indentation. The column position tracking
will be incorrect if byte and character output are mixed, affecting
the placement of indentation.
Indentation mode takes on four numeric values, given by the four
variables
.codn indent-off ,
.codn indent-data ,
.code indent-code
and
.codn indent-foff .
As far as stream output is concerned, the code and data modes
represented by
.code indent-code
and
.code indent-data
behave
the same way: both represent the "indentation turned on" state.
The difference between them influences the behavior of the
.code width-check
function. This function isn't used by any lower-level stream output
routines. It is used by the object printing functions like
.code print
and
.code pprint
to break up long lines.
The
.code indent-off
and
.code indent-foff
modes are also treated the same way by lower level stream output,
indicating "indentation turned off". The modes are distinguished
by
.code print
and
.code pprint
in the following way:
.code indent-off
is a "soft" disable which allows these object-printing routines
to temporarily turn on indentation while traversing aggregate objects.
Whereas the
.code indent-foff
("force off") value is a "hard" disable: the object-printing routines will not
enable indentation and will not break up long lines.
.coNP Variables @, indent-off @, indent-data @ indent-code and @ indent-foff
.desc
These variables hold integer values representing output stream
indentation modes. The value of
.code indent-off
is zero.
.coNP Functions @ get-indent-mode and @ set-indent-mode
.synb
.mets (get-indent-mode << stream )
.mets (set-indent-mode < stream << new-mode )
.mets (test-set-indent-mode < stream < compare-mode << new-mode )
.syne
.desc
These functions retrieve and manipulate the stream indent mode.
The
.code get-indent-mode
retrieves the current indent mode of
.metn stream .
The
.code set-indent-mode
function sets the indent mode of
.meta stream
to
.meta new-mode
and returns the previous mode.
Note: it is encouraged to save and restore the indentation mode,
and in a way that is exception safe.
If a block of code sets up indentation on a stream such as
.code *stdout*
and is terminated by an exception, the indentation will remain in
effect and affect subsequent output. The
.code with-resources
macro or
.code unwind-protect
operator may be used.
.coNP Functions @ test-set-indent-mode and @ test-neq-set-indent-mode
.synb
.mets (test-set-indent-mode < stream < compare-mode << new-mode )
.mets (test-neq-set-indent-mode < stream < compare-mode << new-mode )
.syne
.desc
The
.code test-set-indent-mode
function sets the indent mode of
.meta stream
to
.meta new-mode
if and only if its current mode is equal to
.metn compare-mode .
Whether or not it changes the mode, it returns the previous mode.
The
.code test-neq-set-indent-mode
only differs in that it sets
.meta stream
to
.meta new-mode
if and only if the current mode is
.B not
equal to
.metn compare-mode .
.coNP Functions @, get-indent @ set-indent and @ inc-indent
.synb
.mets (get-indent << stream )
.mets (set-indent < stream << new-indent )
.mets (inc-indent < stream << indent-delta )
.syne
.desc
These functions manipulate the indentation value of the stream.
The indentation takes effect the next time a character is output
following a newline character.
The
.code get-indent
function retrieves the current indentation amount.
The
.code set-indent
function sets
.metn stream 's
indentation to the value
.meta new-indent
and returns the previous value.
Negative values are clamped to zero.
The
.code inc-indent
function sets
.metn stream 's
indentation relative to the current printing column position,
and returns the old value.
The indentation is calculated by adding
.meta indent-delta
to the current column position.
If a negative indentation results, it is clamped to zero.
.coNP Function @ width-check
.synb
.mets (width-check < stream << alt-char )
.syne
.desc
The
.code width-check
function examines the state of the stream, taking into consideration
the current printing column position, the indentation state,
the indentation amount and an internal "force break" flag. It makes a decision
either to introduce a line break by printing a newline character, or else to
print the
.meta alt-char
character.
If a decision is made not to emit a line break, but
.meta alt-char
is
.codn nil ,
then the function has no effect at all.
The return value is
.code t
if the function has issued a line break, otherwise
.codn nil .
.coNP Function @ force-break
.synb
.mets (force-break << stream )
.syne
.desc
If the
.code force-break
function is called on a stream, it sets an internal "force break" flag which
affects the future behavior of
.codn width-check .
The
.code width-check
function examines this flag. If the flag is set,
.code width-check
clears it, and issues a line break without considering any other
conditions.
The
.metn stream 's
.code force-break
flag is also cleared whenever a newline character is output.
The
.code force-break
function returns
.codn stream .
Note: the
.code force-break
is involved in line breaking decisions. Whenever a list or list-like syntax is
being printed, whenever an element of that syntax is broken into multiple
lines, a break is forced after that element, in order to avoid output
which resembles the following diagonally-creeping pattern:
.verb
(a b c (d e f
g h i) j (k l
m n) o)
.brev
but instead is rendered in a more horizontally compact pattern:
.verb
(a b c (d e f
g h i)
j (k l
m n)
o)
.brev
When the printer prints
.code "(d e f g h i)"
it uses the
.code width-check
function between the elements; that function issues the
break between the
.code f
and
.codn g .
The printer monitors the return value of
.codn width-check ;
it knows that since one of the calls returned
.codn t ,
the object had been broken into two or more lines. It then calls
.code force-break
after printing the last element
.code i
of that object. Then, due to the force flag, the outer recursion of the
printer which is printing
.code "(a b c ...)"
will experience a break when it calls
.code width-check
before printing
.codn j .
Custom
.code print
methods defined on structure objects can take advantage of
.code width-check
and
.code force-break
in the same way so that user-defined output integrates
with the formatting algorithm.
.SS* Stream Output Limiting
Streams have two properties which are used by the \*(TL object printer to
optionally truncate the output generated by aggregate objects.
A stream can specify a maximum length for aggregate objects via the
.code set-max-length
function. Using the
.code set-max-depth
function, the maximum depth can also be specified.
This feature is
useful when diagnostic output is being produced, and the objects involved are
so large that the diagnostic output overwhelms the output device or the user,
so as to become uninformative. Output limiting also prevents the printer's
non-termination on infinite, lazy structures.
It is recommended that functions which operate on streams passed in as
parameters save and restore these parameters, if they need to manipulate them,
for instance using
.codn with-resources :
.verb
(defun output-function (arg stream)
;; temporarily impose maximum width and depth
(with-resources ((ml (set-max-length stream 42)
(set-max-length stream ml))
(mw (set-max-depth stream 12)
(set-max-depth stream mw)))
(prinl arg stream)
...))
.brev
.coNP Function @ set-max-length
.synb
.mets (set-max-length < stream << value )
.syne
.desc
The
.code set-max-length
function establishes the maximum length for aggregate object printing.
It affects the printing of lists, vectors, hash tables, strings
as well as quasiliterals and quasiword list literals (QLLs).
The default value is 0 and this value means that no limit is imposed.
Otherwise, the value must be a positive integer.
When the list, vector or hash-table object being printed has more
elements than the maximum length, then elements are printed only up to
the maximum count, and then the remaining elements are summarized by
printing the
.code ...
(three dots) character sequence as if it were an additional element.
This sequence is an invalid token; it cannot be read as input.
When a character string is printed, and the maximum length parameter
is nonzero, a maximum character count is determined as follows.
Firstly, if the maximum length value is less than 3, it is taken
to be 3. Then it is multiplied by 8. Thus, a maximum length
of 10 allows 80 characters, whereas a maximum length of 1
allows 24 characters.
If a string which exceeds the maximum number of characters is being printed
with read-print consistency, as by the
.code print
function, then only a prefix of the string is printed, limited
to the maximum number of characters. Then, the literal syntax is
closed using the character sequence
.code \e...\(dq
(backslash, dot, dot, dot, double quote)
whose leading invalid escape sequence
.code \e.
(backslash, dot) ensures that the truncated object is not readable.
If a string which exceeds the maximum number of characters is being printed
without read-print consistency, as by the
.code pprint
function, then only a prefix of the string is printed, limited
to the maximum number of characters. Then the
character sequence
.code ...
is emitted.
Quasiliterals are treated using a combination of behaviors. Elements of a
quasiliteral are literal sequence of text, and embedded variables and
expressions. The maximum length specifies both the maximum number of elements
in the quasiliteral, and the maximum number of characters in any element which
is a sequence of text. When either limit is exceeded, the
quasiliteral is immediately terminated with the sequence
.code \e...`
(escaped dot, dot, dot, backtick). The maximum character limit is applied to
the units of text cumulatively, rather than individually. As in the case of
string literals, the limit is determined by multiplying the length by 8, and
clamping at a minimum value of 24.
When a QLL is printed, the space-separated elements
of the literal are individually subject to the maximum character limit as if
they were independent quasiliterals. Furthermore, the sequence of these
elements is subject to the maximum length. If there are more elements in the
QLL, then the sequence
.code \e...`
(escaped dot, dot, dot, backtick) is emitted and thus the QLL ends.
The
.code set-max-length
function returns the previous value.
.coNP Function @ set-max-depth
.synb
.mets (set-max-depth < stream << value )
.syne
.desc
The
.code set-max-length
function establishes the maximum depth for the printing of nested
objects. It affects the printing of lists, vectors, hash tables
and structures. The default value is 0 and this value means that no limit is
imposed. Otherwise, the value must be a positive integer.
The depth of an object not enclosed in any object is zero. The depth of the
element of an aggregate is one greater than the depth of the aggregate itself.
For instance, given the list
.code "(1 (2 3))"
the list itself has depth 0, the atom
.code 1
has depth 1, as does the sublist
.codn "(2 3)" ,
and the
.code 2
and
.code 3
atoms have depth 2.
When an object is printed whose depth exceeds the maximum depth, then three dot
character sequence
.code ...
is printed instead of that object. This notation is an invalid token; it cannot be
read as input.
Additionally, when a vector, list, hash table or structure is printed which itself
doesn't exceed the maximum depth, but whose elements do exceed, then that object
is summarized, respectively, as
.codn "(...)" ,
.codn "#(...)" ,
.code "H#(...)"
and
.codn "S#(...)" ,
rather than repeating the
.code ...
sequence for each of its elements.
The
.code set-max-depth
function returns the previous value.
.SS* Coprocesses
.coNP Functions @, open-command @ open-process and @ open-subprocess
.synb
.mets (open-command < system-command <> [ mode-string ])
.mets (open-process < program < mode-string <> [ argument-list ])
.mets (open-subprocess < program < mode-string
.mets \ \ >> [ argument-list <> [ function ]])
.syne
.desc
These functions spawn external programs which execute concurrently
with the \*(TX program. They all return a unidirectional stream for
communicating with these programs: either an output stream, or an input
stream, depending on the contents of
.metn mode-string .
In
.codn open-command ,
the
.meta mode-string
argument is optional, defaulting to
the value
.str r
if it is missing. See the
.code open-file
function for a discussion of modes.
The
.code open-command
function is implemented using POSIX
.codn popen .
Those elements of
.meta mode-string
which are applicable to
.code popen
are passed to it, and hence their semantics follows from
their processing in that function.
The
.code open-command
function accepts, via the
.meta system-command
string parameter, a
system command, which is in a system-dependent syntax. On a POSIX system, this
would be in the POSIX Shell Command Language.
The
.code open-process
function specifies a program to invoke via the
.meta command
argument. This is subject to the operating system's search strategy.
On POSIX systems, if it is an absolute or relative path, it is treated as
such, but if it is a simple base name, then it is subject to searching
via the components of the PATH environment variable. If
.code open-process
is not able to find
.metn program ,
or is otherwise unable to execute
the program, the child process will exit, using the value of the C variable
.code errno
as its exit status. This value can be retrieved via
.codn close-stream .
The
.meta argument-list
argument is a list of strings which specifies additional
optional arguments to be passed to the program. The
.meta program
argument
becomes the first argument, and
.meta argument-list
becomes the second and
subsequent arguments. If
.meta argument-list
is omitted, it defaults to empty.
If a coprocess is open for writing
.mono
.meti >> ( mode-string
.onom
is specified as
.strn w ),
then
writing on the returned stream feeds input to that program's standard input
file descriptor. Indicating the end of input is performed by closing the
stream.
If a coprocess is open for reading
.mono
.meti >> ( mode-string
.onom
is specified as
.strn r ),
then
the program's output can be gathered by reading from the returned stream.
When the program finishes output, it will close the stream, which can be
detected as normal end of data.
The standard input and error file descriptors of an input coprocess
are obtained from the streams stored in the
.code *stdin*
and
.code *stderr*
special variables, respectively. Similarly, the standard output and error
file descriptors of an output coprocess are obtained from the
.code *stdout*
and
.code *stderr*
special variables. These variables must contain streams on which the
.code fileno
function is meaningful, otherwise the operation will fail.
What this functionality means is that rebinding the special variables
for standard streams has the effect of redirection. For example,
the following two expressions achieve the same effect of creating
a stream which reads the output of the
.code cat
program, which reads and produces the contents of the file
.codn text-file .
.verb
;; redirect input by rebinding *stdin*
(let ((*stdin* (open-file "text-file")))
(open-command "cat"))
;; redirect input using POSIX shell redirection syntax
(open-command "cat < text-file")
.brev
The following is erroneous:
.verb
;; (let ((*stdin* (make-string-input-stream "abc")))
(open-command "cat"))
.brev
A string input or output stream doesn't have an operating system file
descriptor; it cannot be passed to a coprocess.
The streams
.codn *stdin* ,
.code *stdout*
and
.code *stderr*
are not synchronized with their underlying file descriptors prior to
the execution of a coprocess. It is up to the program to ensure that
previous output to
.code *stdout*
or
.code *stderr*
is flushed, so that the output of the coprocess isn't reordered
with regard to output produced by the program. Similarly,
input buffered in
.code *stdin*
is not available to the coprocess, even though it has not
yet been read by the program. The program is responsible for preventing this
situation also.
If a coprocess terminates abnormally or unsuccessfully, an exception is raised.
On platforms which have the
.code fork
function, the
.meta mode-string
argument of
.code open-process
supports a special
.meta redirection
syntax. This syntax specifies I/O redirections which are done in the
context of the child process, before the specified program is executed.
Instances of the syntax are considered options; if
.meta mode-string
specifies a mode such as
.code r
that mode must precede the redirections. Redirections may be mixed with
other options.
Up to four redirections may be specified using one
of two forms: a short form or the long form. If more than four
redirections are specified, the
.meta mode-string
is considered ill-formed.
The short form of the syntax consists of three characters: the prefix
character
.codn > ,
a single decimal digit indicating the file descriptor to be redirected,
and then a third character which is either another digit, or else one of the
two characters
.code n
or
.codn x .
If the third character is a digit, it indicates the target file descriptor
of the redirection. For instance
.code >21
indicates that file descriptor 2 is to be redirected to 1 (so that material
written to standard error goes to the same destination as that written
to standard output).
If the third character is
.codn n ,
it means that the file descriptor will be redirected to the file
.codn /dev/null .
For instance,
.code >2n
indicates that descriptor 2 (standard error) will be redirected to
the null device. If the third character is
.codn x ,
it indicates that the file descriptor shall be closed. For instance
.code >0x
means to close descriptor 0 (standard input).
The long form of the syntax allows file descriptors that require more
than one decimal digit. It consists of the same prefix character
.code >
which is immediately followed by an open parenthesis
.codn ( .
The parenthesis is immediately followed by one or more digits which
give the to-be-redirected file descriptor. This is followed by
one or more whitespace characters, and then either another multi-digit
decimal file descriptor or one of the two letters
.code n
or
.codn x .
This second element must be immediately followed by the closing
parenthesis
.codn ) .
Thus
.code >21
and
.code >2n
may be written in the long form, respectively, as
.code ">(2 1)"
and
.codn ">(2 n)" ,
while
.code ">(32 47)"
has no short form equivalent.
Multiple redirections may be specified, in any mixture of the long and
short form. For instance
.code "r>21>0n>(27 31)"
specifies a process pipe that is open for reading, capturing the
output of the process. In that process, standard error is redirected
to standard output, standard input is connected to the null device,
and descriptor 27 is redirected to descriptor 31.
Note: on platforms which don't have a
.code fork
function, the implementation of
.code open-process
is simulated via
.code open-command
and therefore does not support the redirection syntax; it is parsed
and ignored.
The
.code open-subprocess
function is a variant of
.code open-process
that is available on platforms which have a
.code fork
function. This function has all the same argument conventions and semantics as
.codn open-process ,
adding the
.meta function
argument. If this argument isn't
.codn nil ,
then it must specify a function which can be called with no arguments.
This function is called in the child process after any redirections are
established, just before the program specified by the
.meta program
argument is executed. Moreover, the
.code open-subprocess
function allows
.meta program
to be specified as
.code nil
in which case
.meta function
must be specified. When
.meta function
returns, the child process terminates as if by a call to
.code exit*
with an argument of zero.
.SS* I/O-Related Convenience Functions
The functions in this group create a stream, perform an I/O operation
on it, and ensure that it is closed, in one convenient operation. They
operate on files or command streams.
Several other functions in this category exist, which operate with buffers.
They are documented in the Buffer Functions subsection under the
FOREIGN FUNCTION INTERFACE section.
.coNP Functions @, file-get @ file-get-string and @ file-get-lines
.synb
.mets (file-get << name )
.mets (file-get-string << name )
.mets (file-get-lines << name )
.syne
.desc
The
.code file-get
function opens a text stream over the file indicated by the string argument
.meta name
for reading, reads the printed representation of a \*(TL object from it,
and returns that object, ensuring that the stream is closed.
The
.code file-get-string
is similar to
.code file-get
except that it reads the entire file as a text stream and returns
its contents in a single character string.
The
.code file-get-lines
function opens a text stream over the file indicated by
.meta name
and returns produces a lazy list of strings representing the lines
of text of that file as if by a call to the
.code get-lines
function, and returns that list. The stream remains open until the
list is consumed to the end, as indicated in the description of
.codn get-lines .
.coNP Functions @, file-put @ file-put-string and @ file-put-lines
.synb
.mets (file-put < name << obj )
.mets (file-put-string < name << string )
.mets (file-put-lines < name << list )
.syne
.desc
The
.codn file-put ,
.code file-put-string
and
.code file-put-lines
functions open a text stream over the file indicated by the string argument
.metn name ,
write the argument object into the file in their specific manner,
and then close the file.
If the file doesn't exist, it is created.
If it exists, it is truncated to zero length and overwritten.
The
.code file-put
function writes a printed representation of
.meta obj
using the
.code prinl
function. The return value is that of
.codn prinl .
The
.code file-put-string
function writes
.meta string
to the stream using the
.code put-string
function. The return value is that of
.codn put-string .
The
.code file-put-lines
function writes
.meta list
to the stream using the
.code put-lines
function. The return value is that of
.codn put-lines .
.coNP Functions @, file-append @ file-append-string and @ file-append-lines
.synb
.mets (file-append < name << obj )
.mets (file-append-string < name << string )
.mets (file-append-lines < name << list )
.syne
.desc
The
.codn file-append ,
.code file-append-string
and
.code file-append-lines
functions open a text stream over the file indicated by the string argument
.metn name ,
write the argument object into the stream in their specific manner,
and then close the stream.
These functions are close counterparts of, respectively,
.codn file-get ,
.code file-append-string
and
.codn file-append-lines .
These functions behave differently when the indicated file
already exists. Rather than being truncated and overwritten,
the file is extended by appending the new data to its end.
.coNP Functions @, command-get @ command-get-string and @ command-get-lines
.synb
.mets (command-get << cmd )
.mets (command-get-string << cmd )
.mets (command-get-lines << cmd )
.syne
.desc
The
.code command-get
function opens text stream over an input command pipe created for
the command string
.metn cmd ,
as if by the
.code open-command
function. It reads the printed representation of a \*(TL object from it, and
returns that object, ensuring that the stream is closed.
The
.code command-get-string
is similar to
.code command-get
except that it reads the entire file as a text stream and returns
its contents in a single character string.
The
.code command-get-lines
function opens a text stream over an input command pipe created for the
command string
.meta cmd
and returns produces a lazy list of strings representing the lines
of text of that file as if by a call to the
.code get-lines
function, and returns that list. The stream remains open until the
list is consumed to the end, as indicated in the description of
.codn get-lines .
.coNP Functions @, command-put @ command-put-string and @ command-put-lines
.synb
.mets (command-put < cmd << obj )
.mets (command-put-string < cmd << string )
.mets (command-put-lines < cmd << list )
.syne
.desc
The
.codn command-put ,
.code command-put-string
and
.code command-put-lines
functions open an output text stream over an output command pipe created
for the command specified in the string argument
.metn cmd ,
as if by the
.code open-command
function.
They write the argument object into the stream in their specific manner,
and then close the stream.
The
.code command-put
function writes a printed representation of
.meta obj
using the
.code prinl
function. The return value is that of
.codn prinl .
The
.code command-put-string
function writes
.meta string
to the stream using the
.code put-string
function. The return value is that of
.codn put-string .
The
.code command-put-lines
function writes
.meta list
to the stream using the
.code put-lines
function. The return value is that of
.codn put-lines .
.SS* Buffer streams
A stream type exists which allows
.code buf
objects to be manipulated through the stream interface.
A buffer stream is created using the
.code make-buf-stream
function, which can either attach the stream to an existing buffer,
or create a new buffer that can later be retrieved from the stream
using
.codn get-buf-from-stream .
Operations on the buffer stream treat the underlying buffer much like if it
were a memory-based file. Unless the underlying buffer is a "borrowed buffer"
referencing the storage belonging to another object
(such as the buffer object produced by the
.code buf-d
FFI type's get semantics) the stream operations can change the buffer's size.
Seeking beyond the end of the buffer an then writing one or more bytes
extends the buffer's length, filling the newly allocated area with zero bytes.
The
.code truncate-stream
function is supported also.
Buffer streams also support the
.code :byte-oriented
property.
Macros
.code with-out-buf-stream
and
.code with-in-buf-stream
are provided to simplify the steps involved in using buffer streams
in some common scenarios. Note that in spite of the naming of these
macros there is only one buffer stream type, which supports bidirectional I/O.
.coNP Function @ make-buf-stream
.synb
.mets (make-buf-stream <> [ buf ])
.syne
.desc
The
.code make-buf-stream
function return a new buffer stream. If the
.meta buf
argument is supplied, it must be a
.code buf
object. The stream is then associated with this object.
If the argument is omitted, a buffer of length zero is created and associated
with the stream.
.coNP Function @ get-buf-from-stream
.synb
.mets (get-buf-from-stream << buf-stream )
.syne
.desc
The
.code get-buf-from-stream
returns the buffer object associated with
.meta buf-stream
which must be a buffer stream.
.coNP Macros @ with-out-buf-stream and @ with-in-buf-stream
.synb
.mets (with-out-buf-stream >> ( var <> [ buf-expr ])
.mets \ \ << body-form *)
.mets (with-in-buf-stream >> ( var << buf-expr )
.mets \ \ << body-form *)
.syne
.desc
The
.code with-out-buf-stream
and
.code with-in-buf-stream
macros both bind variable
.meta var
to an implicitly created buffer stream, and evaluate zero or more
.metn body-form s
in the environment where the variable is visible.
The
.meta buf-expr
argument, which may be omitted in the use of the
.code with-out-buf-stream
macro, must be an expression which evaluates to a
.code buf
object.
The
.meta var
argument must be a symbol suitable for naming a variable.
The implicitly allocated buffer stream is connected
to the buffer specified by
.meta buf-expr
or, when
.meta buf-expr
is omitted, to a newly allocated buffer.
The code generated by the
.code with-out-buf-stream
macro, if it terminates normally, yields the buffer object
as its result value.
The
.code with-in-buf-stream
returns the value of the last
.metn body-form ,
or else
.code nil
if no forms are specified.
.TP* Examples:
.verb
(with-out-buf-stream (*stdout* (make-buf 24))
(put-string "Hello, world!"))
-> #b'48656c6c6f2c2077 6f726c6421000000 0000000000000000'
(with-out-buf-stream (*stdout*) (put-string "Hello, world!"))
-> #b'48656c6c6f2c2077 6f726c6421'
.brev
.SS* Foreign Pointers
.coNP The @ cptr type
Objects of type
.code cptr
are Lisp values which contain a foreign pointer ("C pointer"). This data type
is used by the
.code dlopen
function and is generally useful in conjunction with the Foreign Function
Interface (FFI). An arbitrary pointer emanating from a foreign function
can be captured as a
.code cptr
value, which can be passed back into foreign code. For this purpose, there
exits also a matching FFI type called
.codn cptr .
The
.code cptr
type supports a symbolic type tag, which defaults to
.codn nil .
The type tag plays a role in FFI. The FFI
.code cptr
type supports a tag attribute. When a
.code cptr
object is converted to a foreign pointer under the control of the FFI
type, and that FFI type has a tag other than
.codn nil ,
the object's tag must exactly match that of the FFI type, or the conversion
throws an error. In the reverse direction, when a foreign pointer is
converted to a
.code cptr
object under control of the FFI
.code cptr
type, the object inherits the type tag from the FFI type.
Although
.code cptr
objects are conceptually non-aggregate values, corresponding to pointers,
they are
.I "de facto"
aggregates due to their implementation as references to heap objects.
When a
.code cptr
object is passed to a foreign function by pointer, for
instance using a parameter of type
.codn "(ptr cptr)" ,
its internal pointer is potentially updated to the new value coming from the
function.
.coNP Function @ cptr-int
.synb
.mets (cptr-int < integer <> [ type-symbol ])
.syne
.desc
The
.code cptr-int
function converts
.meta integer
into a pointer in a system-specific way
which is consistent with the system's addressing structure. Then it returns
that pointer contained in a
.code cptr
object.
The
.meta integer
argument must be an integer which is in range for a pointer value.
Note: this range is wider than the
.code fixnum
range; a portion of the range of
.code bignum
integers can denote pointers.
An extended range of values is accepted. The entire addressable space may be
expressed by non-negative values. A range of negative values also expresses a
portion of the address space, in accordance with the platform's concept of a
signed integer.
For instance, on a system with 32-bit addresses, the values 0 to 4294967295
express all of the addresses as a pure binary value. Furthermore, the values
-2147483648 to -1 also express the upper part of this range, corresponding,
respectively, to the addresses 2147483648 to 4294967295. On that platform,
values of
.meta integer
outside of the range -2147483648 to 4294967295 are invalid.
The
.meta type-symbol
argument should be a symbol. If omitted, it defaults to
.codn nil .
This symbol becomes the
.code cptr
object's type tag.
.coNP Function @ cptr-obj
.synb
.mets (cptr-obj < object <> [ type-symbol ])
.syne
.desc
The
.code cptr-obj
function converts
.meta object
object directly to a
.codn cptr .
The
.meta object
argument may be of any type.
The raw representation of
.meta object
is simply stored in a new instance of
.code cptr
and returned.
The
.meta type-symbol
argument should be a symbol. If omitted, it defaults to
.codn nil .
This symbol becomes the
.code cptr
object's type tag.
The lifetime of the returned
.code cptr
object is independent from that of
.metn object .
If the lifetime of
.meta object
reaches its end before that of the
.codn cptr ,
the pointer stored inside the
.code cptr
becomes invalid.
.coNP Function @ int-cptr
.synb
.mets (int-cptr << cptr )
.syne
.desc
The
.code int-cptr
function retrieves the pointer value of the
.meta cptr
object as an integer.
If an integer
.meta n
is in a range convertible to
.code cptr
type, then the expression
.mono
.meti (int-cptr (cptr-int << n ))
.onom
reproduces
.metn n .
.coNP Function @ cptr-buf
.synb
.mets (cptr-buf < buf <> [ type-symbol ])
.syne
.desc
The
.code cptr-buf
function returns a
.code cptr
object which holds a pointer to a buffer object's storage
area. The
.meta buf
argument must be of type
.codn buf .
The
.meta type-symbol
argument should be a symbol. If omitted, it defaults to
.codn nil .
This symbol becomes the
.code cptr
object's type tag.
The lifetime of the returned
.code cptr
object is independent from that of
.metn buf .
If the lifetime of
.meta buf
reaches its end before that of the
.codn cptr ,
the pointer stored inside the
.code cptr
becomes invalid.
.coNP Function @ cptr-cast
.synb
.mets (cptr-cast < type-symbol << cptr )
.syne
.desc
The
.code cptr-cast
function produces a new
.code cptr
object which has the same pointer as
.meta cptr
but whose type is given by
.metn type-symbol .
Casting
.meta cptr
objects with
.code cptr-cast
circumvents the safety mechanism which
.code cptr
type tagging provides.
.coNP Function @ copy-cptr
.synb
.mets (cptr-copy << cptr )
.syne
.desc
The
.code copy-cptr
function creates a new
.code cptr
object similar to
.metn cptr ,
which has the same address and type symbol as
.metn cptr .
.coNP Function @ cptr-zap
.synb
.mets (cptr-zap << cptr )
.syne
.desc
The
.code cptr-zap
function changes the pointer value of the
.meta cptr
object to the null pointer.
The
.meta cptr
argument must be of
.code cptr
type.
The return value is
.meta cptr
itself.
Note: it is recommended to use
.code cptr-zap
when the program has taken some action which invalidates the pointer value
stored in a
.code cptr
object, where a risk exists that the value may be subsequently misused.
.coNP Function @ cptr-free
.synb
.mets (cptr-free << cptr )
.syne
.desc
The
.code cptr-free
function passes the
.meta cptr
object's pointer to the C library
.code free
function. After this action, it behaves exactly like
.codn cptr-zap .
The
.meta cptr
argument must be of
.code cptr
type.
The return value is
.meta cptr
itself.
Note: this function is unsafe. If the pointer didn't originate from the
.code malloc
family of memory allocation functions, or has already been freed, or
copies of the pointer exist which are still in use, the consequences are
likely catastrophic.
.coNP Function @ cptrp
.synb
.mets (cptrp << value )
.syne
.desc
The
.code cptrp
function tests whether
.meta value
is a
.codn cptr .
It returns
.code t
if this is the
case,
.code nil
otherwise.
.coNP Function @ cptr-type
.synb
.mets (cptr-type << cptr )
.syne
.desc
The
.code cptr-type
function retrieves the
.meta cptr
object's type tag.
.coNP Function @ cptr-get
.synb
.mets (cptr-get < cptr <> [ type ])
.syne
.desc
The
.code cptr-get
function extracts a Lisp value by converting a C object
at the memory location denoted by
.metn cptr ,
according to the FFI type
.metn type .
The external representation at the specified memory location is
is scanned according to the
.meta type
and converted to a Lisp value which is returned.
If the
.meta type
argument is specified, it must be a FFI type object.
If omitted, then the
.code cptr
object's type tag is interpreted as a FFI type symbol and resolved to
a type; the resulting type, if one is found is substituted for
.metn type .
If the lookup fails an error exception is thrown.
The
.meta cptr
object must be of type
.code cptr
and point to a memory area suitably aligned for, and large
enough to hold a foreign representation of
.metn type ,
at the byte offset indicated by the
.meta offset
argument.
If
.meta cptr
is a null pointer, an exception is thrown.
The
.code cptr-get
operation is similar to the "get semantics" performed by FFI
in order to extract the return value of foreign function
calls, and by the FFI callback mechanism to extract the
arguments coming into a callback.
The
.meta type
argument may not be a variable length type, such as an array of
unspecified size.
Note: the functions
.code cptr-get
and
.code cptr-out
are useful in simplifying the interaction with "semi-opaque" foreign objects:
objects which serve as API handles that are treated as opaque pointers in API
argument calls, but which expose some internal members that the application
must access directly. The
.code cptr
objects pass through the foreign API without undergoing conversion,
as usual. The application uses these two functions to perform conversion as
necessary. Under this technique, the description of the foreign object need not
be complete. Structure members which occur after the last member that the
application is interested in need not be described in the FFI type.
.coNP Function @ cptr-out
.synb
.mets (cptr-out < cptr < obj <> [ type ])
.syne
.desc
The
.code cptr-out
function converts a Lisp value into a C representation,
which is stored at the memory location denoted by
.metn cptr ,
according to the FFI type
.metn type .
The function's return value is
.metn obj .
If the
.meta type
argument is specified, it must be a FFI type object.
If omitted, then the
.code cptr
object's type tag is interpreted as a FFI type symbol and resolved to
a type; the resulting type, if one is found is substituted for
.metn type .
If the lookup fails an error exception is thrown.
The
.meta obj
argument must be an object compatible with the conversions
implied by
.metn type .
The
.meta cptr
object must be of type
.code cptr
and point to a memory area suitably aligned for, and large
enough to hold a foreign representation of
.metn type ,
at the byte offset indicated by the
.meta offset
argument.
If
.meta cptr
is a null pointer, an exception is thrown.
It is assumed that
.meta obj
is an object which was returned by an earlier call to
.codn cptr-get ,
and that the
.meta cptr
and
.meta type
arguments are the same objects that were used in that call.
The
.code cptr-out
function performs the "out semantics" encoding action, similar
to the treatment applied to the arguments of a callback prior to
returning to foreign code.
.coNP Variable @ cptr-null
.desc
The
.code cptr-null
variable holds a null pointer as a
.code cptr
instance.
Two
.code cptr
objects may be compared for equality using the
.code equal
function, which tests whether their pointers are equal.
The
.code cptr-null
variable compares
.code equal
to values which have been subject to
.code cptr-zap
or
.codn cptr-free .
A null
.code cptr
may be produced by the expression
.codn "(cptr-obj nil)" ;
however, this creates a freshly allocated object on each evaluation.
The expression
.code "(cptr-int 0)"
also produces a null pointer on all platforms where \*(TX is found.
.coNP Function @ cptr-size-hint
.synb
.mets (cptr-size-hint < cptr << bytes )
.syne
.desc
The
.code cptr-size-hint
function indicates to the garbage collector that the given
.meta cptr
object is associated with
.meta bytes
of foreign memory that are otherwise invisible to the garbage collector.
Note: this function should be used if the foreign memory is indirectly
managed by the
.meta cptr
object in cooperation with the garbage collector. Specifically,
.meta cptr
should have a finalizer registered against it which will liberate the
foreign memory.
.SS* User-Defined Streams
In \*(TL, stream objects aren't structure types, and therefore lie outside of
the object-oriented programming system. However, \*(TL supports a delegation
mechanism which allows a structure which provides certain methods to be used as
a stream.
The function
.code make-struct-delegate-stream
takes as an argument the instance of a structure, which is
referred to as the
.IR "stream interface object" .
The function returns a stream object such that when
stream operations are invoked on this stream, it delegates these
operations to methods of the stream interface object.
A structure type called
.code stream-wrap
is provided, whose instances can serve as stream interface objects.
This structure has a slot called
.meta stream
which holds a stream, and it provides all of the methods required for
the delegation mechanism used by
.codn make-struct-delegate-stream .
This
.code stream-wrap
operations simply invoke the ordinary stream operations on the
.meta stream
slot. The
.code stream-wrap
type can be used as a base class for a derived class which intercepts
certain operations on a stream (by defining the corresponding methods) while
allowing other operations to transparently pass to the stream (via the
base methods inherited from
.codn stream-wrap ).
.coNP Function @ make-struct-delegate-stream
.synb
.mets (make-struct-delegate-stream << object )
.syne
.desc
The
.code make-struct-delegate-stream
function returns a stream whose operations depend on the
.metn object ,
a stream interface object.
The
.meta object
argument must be a structure which implements certain
subsets of, or all of, the following methods:
.codn put-string ,
.codn put-char ,
.codn put-byte ,
.codn get-line ,
.codn get-char ,
.codn get-byte ,
.codn unget-char ,
.codn unget-byte ,
.codn put-buf ,
.codn fill-buf ,
.codn close ,
.codn flush ,
.codn seek ,
.codn truncate ,
.codn get-prop ,
.codn set-prop ,
.codn get-error ,
.codn get-error-str ,
.code clear-error
and
.codn get-fd .
Implementing
.code get-prop
is mandatory, and that method must support the
.code :name
property.
Failure to implement some of the other methods will impair the use of certain
stream operations on the object.
.coNP Method @ put-string
.synb
.mets << stream .(put-string str)
.syne
.desc
The
.code put-string
method is implemented on a stream interface object.
It should behave in a manner consistent with the
description of the
.code put-string
stream I/O function.
.coNP Method @ put-char
.synb
.mets << stream .(put-char chr)
.syne
.desc
The
.code put-char
method is implemented on a stream interface object.
It should behave in a manner consistent with the
description of the
.code put-char
stream I/O function.
.coNP Method @ put-byte
.synb
.mets << stream .(put-byte byte)
.syne
.desc
The
.code put-byte
method is implemented on a stream interface object.
It should behave in a manner consistent with the
description of the
.code put-byte
stream I/O function.
.coNP Method @ get-line
.synb
.mets << stream .(get-line)
.syne
.desc
The
.code get-line
method is implemented on a stream interface object.
It should behave in a manner consistent with the
description of the
.code get-line
stream I/O function.
.coNP Method @ get-char
.synb
.mets << stream .(get-char)
.syne
.desc
The
.code get-char
method is implemented on a stream interface object.
It should behave in a manner consistent with the
description of the
.code get-char
stream I/O function.
.coNP Method @ get-byte
.synb
.mets << stream .(get-byte)
.syne
.desc
The
.code get-byte
method is implemented on a stream interface object.
It should behave in a manner consistent with the
description of the
.code get-byte
stream I/O function.
.coNP Method @ unget-char
.synb
.mets << stream .(unget-char chr)
.syne
.desc
The
.code unget-char
method is implemented on a stream interface object.
It should behave in a manner consistent with the
description of the
.code unget-char
stream I/O function.
.coNP Method @ unget-byte
.synb
.mets << stream .(unget-byte byte)
.syne
.desc
The
.code unget-byte
method is implemented on a stream interface object.
It should behave in a manner consistent with the
description of the
.code unget-byte
stream I/O function.
.coNP Method @ put-buf
.synb
.mets << stream .(put-buf buf pos)
.syne
.desc
The
.code put-buf
method is implemented on a stream interface object.
It should behave in a manner consistent with the
description of the
.code put-buf
stream I/O function.
Note: there is a severe restriction on the use of the
.meta buf
argument. The buffer object denoted by the
.meta buf
argument may be specially allocated and have a lifetime
which is scoped to the method invocation. The
.code put-buf
method shall not permit the
.meta buf
object to be used beyond the duration of the method
invocation.
.coNP Method @ fill-buf
.synb
.mets << stream .(fill-buf buf pos)
.syne
.desc
The
.code fill-buf
method is implemented on a stream interface object.
It should behave in a manner consistent with the
description of the
.code fill-buf
stream I/O function.
Note: there is a severe restriction on the use of the
.meta buf
argument. The buffer object denoted by the
.meta buf
argument may be specially allocated and have a lifetime
which is scoped to the method invocation. The
.code fill-buf
method shall not permit the
.meta buf
object to be used beyond the duration of the method
invocation.
.coNP Method @ close
.synb
.mets << stream .(close throw-on-error-p)
.syne
.desc
The
.code close
method is implemented on a stream interface object.
It should behave in a manner consistent with the
description of the
.code close-stream
stream I/O function.
.coNP Method @ flush
.synb
.mets << stream .(flush offs whence)
.syne
.desc
The
.code flush
method is implemented on a stream interface object.
It should behave in a manner consistent with the
description of the
.code flush-stream
stream I/O function.
.coNP Method @ seek
.synb
.mets << stream .(seek offs whence)
.syne
.desc
The
.code seek
method is implemented on a stream interface object.
It should behave in a manner consistent with the
description of the
.code seek-stream
stream I/O function.
.coNP Method @ truncate
.synb
.mets << stream .(truncate len)
.syne
.desc
The
.code truncate
method is implemented on a stream interface object.
It should behave in a manner consistent with the
description of the
.code truncate-stream
stream I/O function.
.coNP Method @ get-prop
.synb
.mets << stream .(get-prop sym)
.syne
.desc
The
.code get-prop
method is implemented on a stream interface object.
It should behave in a manner consistent with the
description of the
.code get-prop
stream I/O function.
.coNP Method @ set-prop
.synb
.mets << stream .(set-prop sym nval)
.syne
.desc
The
.code set-prop
method is implemented on a stream interface object.
It should behave in a manner consistent with the
description of the
.code set-prop
stream I/O function.
.coNP Method @ get-error
.synb
.mets << stream .(get-error)
.syne
.desc
The
.code get-error
method is implemented on a stream interface object.
It should behave in a manner consistent with the
description of the
.code get-error
stream I/O function.
.coNP Method @ get-error-str
.synb
.mets << stream .(get-error-str)
.syne
.desc
The
.code get-error-str
method is implemented on a stream interface object.
It should behave in a manner consistent with the
description of the
.code get-error-str
stream I/O function.
.coNP Method @ clear-error
.synb
.mets << stream .(clear-error)
.syne
.desc
The
.code clear-error
method is implemented on a stream interface object.
It should behave in a manner consistent with the
description of the
.code clear-error
stream I/O function.
.coNP Method @ get-fd
.synb
.mets << stream .(get-fd)
.syne
.desc
The
.code get-fd
method is implemented on a stream interface object.
It should behave in a manner consistent with the
description of the
.code fileno
stream I/O function.
.coNP Structure @ stream-wrap
.synb
.mets (defstruct stream-wrap nil
.mets \ \ stream
.mets \ \ (:method put-string (me str)
.mets \ \ \ \ (put-string str me.stream))
.mets \ \ (:method put-char (me chr)
.mets \ \ \ \ (put-char chr me.stream))
.mets \ \ (:method put-byte (me byte)
.mets \ \ \ \ (put-byte byte me.stream))
.mets \ \ (:method get-line (me)
.mets \ \ \ \ (get-line me.stream))
.mets \ \ (:method get-char (me)
.mets \ \ \ \ (get-char me.stream))
.mets \ \ (:method get-byte (me)
.mets \ \ \ \ (get-byte me.stream))
.mets \ \ (:method unget-char (me chr)
.mets \ \ \ \ (unget-char chr me.stream))
.mets \ \ (:method unget-byte (me byte)
.mets \ \ \ \ (unget-byte byte me.stream))
.mets \ \ (:method put-buf (me buf pos)
.mets \ \ \ \ (put-buf buf pos me.stream))
.mets \ \ (:method fill-buf (me buf pos)
.mets \ \ \ \ (fill-buf buf pos me.stream))
.mets \ \ (:method close (me throw-on-error)
.mets \ \ \ \ (close-stream me.stream throw-on-error))
.mets \ \ (:method flush (me)
.mets \ \ \ \ (flush-stream me.stream))
.mets \ \ (:method seek (me offs whence)
.mets \ \ \ \ (seek-stream me.stream offs whence))
.mets \ \ (:method truncate (me len)
.mets \ \ \ \ (truncate-stream me.stream len))
.mets \ \ (:method get-prop (me sym)
.mets \ \ \ \ (stream-get-prop me.stream sym))
.mets \ \ (:method set-prop (me sym nval)
.mets \ \ \ \ (stream-set-prop me.stream sym nval))
.mets \ \ (:method get-error (me)
.mets \ \ \ \ (get-error me.stream))
.mets \ \ (:method get-error-str (me)
.mets \ \ \ \ (get-error-str me.stream))
.mets \ \ (:method clear-error (me)
.mets \ \ \ \ (clear-error me.stream))
.mets \ \ (:method get-fd (me)
.mets \ \ \ \ (fileno me.stream)))
.syne
.desc
The
.code stream-wrap
class provides a trivial implementation of a stream interface.
It has a single slot,
.code stream
which should be initialized with a stream object.
Each methods of
.metn stream-wrap ,
shown in its entirety in the above Syntax section, simply
invoke the corresponding stream I/O library functions, passing
the method arguments, and the value of the
.code stream
slot to that function, and consequently returning whatever that function
returns.
Note: the
.code stream-wrap
function is intended to useful as an inheritance base. A user-defined
structure can inherit from
.code stream-wrap
and provide its own versions of some of the methods, thereby intercepting
those operations to customize the behavior.
For instance, a function equivalent to the
.code record-adapter
function could be implemented by constructing an object derived from
.code stream-wrap
which overrides the behavior of the
.code get-line
method, and then using the
.code make-struct-delegate-stream
to return a stream based on this object.
.TP* Example:
.verb
;;; Implementation of my-record-adapter,
;;; a function resembling
;;; the record-adapter implementation
(defstruct rec-input stream-wrap
regex
include-match-p
;; get-line overridden to use regex-based
;; extraction using read-until-match
(:method get-line (me)
(read-until-match me.regex me.stream
me.include-match-p))))
(defun my-record-adapter (regex stream include-match-p)
(let ((recin (new rec-input
stream stream
regex regex
include-match-p include-match-p)))
(make-struct-delegate-stream recin)))
.brev
.SS* Symbols and Packages
\*(TL has a package system inspired by the salient features of ANSI Common
Lisp, but substantially simpler.
Each symbol has a name, which is a string.
A package is an object which serves as a container of symbols; the package
associates the name strings with symbols.
A symbol which exists inside a package is said to be interned in that package.
A symbol can be interned in more than one package.
A symbol may also have a home package. A symbol which has a home package
is always interned in that package.
A symbol which has a home package is called an
.IR "interned symbol" .
A symbol which is interned in one or more packages, but has no home package,
is a
.IR "quasi-interned symbol" .
When a quasi-interned symbol is printed, if it is not interned in
the package currently held in the
.code *package*
variable, it will appear in uninterned notation denoted by a
.code #:
prefix, even though it is interned in one or more packages.
This is because in any situation when a symbol is printed with a package
prefix, that prefix corresponds to the name of its home package.
The reverse isn't true: when a symbol token is read bearing a package
prefix, the token denotes any interned symbol in the indicated package,
whether or not the package is the home package of that symbol.
Packages are held in a global list which can be used to search for a package by
name. The
.code find-package
function performs this lookup. A package may be
deleted from the list with the
.code delete-package
function, but it continues
to exist until the program loses the last reference to that package.
When a package is deleted with
.codn delete-package ,
its symbols are uninterned from all other packages.
An existing symbol can be brought into a package via the
.code use-sym
function, causing it to be interned in that package. A symbol which thus exists
inside a package which is not its home package is called a
.IR "foreign symbol" ,
relative to that package.
The contrasting term with
.I "foreign symbol"
is
.IR "local symbol" ,
which refers to a symbol, relative to a package, which is interned in that
package and that package is also its home. Every symbol interned in
a package is either foreign or local.
If a foreign symbol is introduced into a package, and has the same name
as an existing local symbol, the local symbol continues to exist, but
is hidden: it is not accessible via a name lookup on that package.
While hidden, a symbol loses its home package and is thus
degraded to either quasi-interned or uninterned status, depending
on whether that symbol is interned in other packages.
When a foreign symbol is removed from a package via
.codn unuse-sym ,
then if a hidden symbol exists in that package of the same name,
that hidden symbol is reinterned in that package and reacquires
that package as its home package, becoming an interned symbol again.
Finally, packages have a
.IR "fallback package list" :
a list of associated packages, which may be empty. The fallback package
list is manipulated with the functions
.code package-fallback-list
and
.codn set-package-fallback-list ,
and with the
.code :fallback
clause of the
.code defpackage
macro. The fallback package list plays a role only in three situations:
one in the \*(TL parser, one in the printer, and one in the interactive
listener. Besides that, two library functions refer to it:
.code intern-fb
and
.codn find-symbol-fb .
The parser situation involving the fallback list occurs when the
\*(TL parser resolves an unqualified symbol token: a symbol token not carrying
a package prefix. Such a symbol name is resolved against the current package
(the package currently stored in the
.code *package*
special variable). If a symbol matching the token
is not found in the current package, then the packages in its fallback
package list are searched for the symbol. The first matching symbol which is
found in the fallback list is returned. If no matching symbol is found in the
fallback list, then the token is interned as a new symbol is interned in the
current package. The packages in the current package's fallback list may
themselves have fallback lists. Those fallback lists are not involved; no such
recursion takes place.
The printer situation involving the fallback list is as follows.
If a symbol is being printed in a machine-readable way (not "pretty"),
has a home package and is not a keyword symbol, then a search takes place
through the current package first and then its fallback list. If the symbol is
found anywhere in that sequence of locations, and is not occluded by a
same-named symbol occurring earlier in that sequence, then the symbol is
printed without a package prefix.
The listener situation involving the fallback list is a follows.
When tab completion is used on a symbol without a package
prefix, the listener searches for completions not only in the current
package, but in the fallback list also.
.TP* "Dialect Notes:"
The \*(TL package system doesn't support the ANSI Common Lisp
concept of package use, replacing that concept with fallback packages.
Though the
.code use-package
and
.code unuse-package
functions exist and are similar to the ones in ANSI CL,
they actually operate on individual foreign symbols, bringing
them in or removing them, respectively. These functions effectively
iterate over the local symbols of the used or unused package, and invoke
.code use-sym
or
.codn unuse-sym ,
respectively.
The \*(TL package system consequently doesn't support the concept
of shadowing symbols, and conflicts do not exist. When a foreign symbol is
introduced into a package which already has a symbol by that name, that symbol
is silently removed from that package if it is itself foreign, or else hidden
if it is local.
The \*(TL package system also doesn't feature the concept of
internal and external symbols. The rationale is that this distinction
divides symbols into subsets in a redundant way. Packages are already
subsets of symbols. A module can use two packages to simulate private
symbols. An example of this is given in the Package Examples section below.
The \*(TL fallback package list mechanism resembles ANSI CL package use,
and satisfies similar use scenarios. However, this mechanism does not cause
a symbol to be considered visible in a package. If a package
.code foo
contains no symbol
.codn bar ,
but one of the packages in
.codn foo 's
fallback list does contain
.codn bar ,
that symbol is nevertheless not considered visible in
.codn foo .
The syntax
.code foo:bar
will not resolve.
The fallback mechanism only comes into play when a package is installed
as the current package in the
.code *package*
variable. It then allows unqualified symbol references to refer across
the fallback list.
.NP* Package Examples
The following example illustrates a simple scenario of a module
whose identifies are in a package, and which also has private identifiers
in a private package.
.verb
;; Define three packages.
(defpackage mod-priv
(:fallback usr))
(defpackage mod)
(defpackage client
(:fallback mod usr)
(:use-from mod-priv other-priv))
;; Switch to mod-priv package
(in-package mod-priv)
(defun priv-fun (arg)
(list arg))
;; Another function with a name in the mod-priv package.
(defun other-priv (arg)
(cons arg arg))
;; Define a function in mod; a public function.
;; Note that we don't have to change to the mod package,
;; to define functions with names in that package.
;; We rely on interning being allowed for the qualified
;; mod:public-fun syntax.
(defun mod:public-fun (arg)
(priv-fun arg)) ;; priv-fun here is mod-priv:priv-fun
;; Switch to client package
(in-package client)
(priv-fun) ;; ERROR: refers to client:priv-fun, not defined
(mod:priv-fun) ;; ERROR: mod-priv:priv-fun not used in mod
(mod-priv:priv-fun 3) ;; OK: direct reference via qualifier
(public-fun 3) ;; OK: mod:public-fun symbol via fallback
(other-priv 3) ;; OK: foreign symbol mod-priv:other-priv
;; present in client due to :use-from
.brev
The following example shows how to create a package called
.code custom
in which the
.code +
symbol from the
.code usr
package is replaced with a local symbol. A function is
then defined using the local symbol, which allows strings
to be catenated with
.codn + :
.verb
(defpackage custom
(:fallback usr)
(:local + - * /))
(defmacro outside-macro (x) ^(+ ,x 42))
(in-package custom)
(defun binary-+ (: (left 0) (right 0))
(if (and (numberp left) (numberp right))
(usr:+ left right)
`@left@right`))
(defun + (. args)
[reduce-left binary-+ args])
(+) -> 0
(+ 1) -> 1
(+ 1 "a") -> "1a"
(+ 1 2) -> 3
(+ "a") -> "a"
(+ "a" "b" "c") -> "abc"
;; macro expansions using usr:+ are not affected
(outside-macro "a") -> ;; error: + invalid operands "a" 42
.brev
.NP* Packages and the Extraction Language
The \*(TX extraction language has a syntax in which certain Lisp symbolic
expressions denoting directives
.code "@(collect ...)"
or
.code "@(end)"
behave as if they were the tokens of a phrase structure. As a matter of
implementation, these are processed specially in the parser and lexical
analyzer, and are not read in the same way as ordinary Lisp forms.
On the other hand, some directives are not this way. For instance the
.codn "@(bind ...)" ,
syntax is processed as a true Lisp expression, in which the
.code bind
token is subject to the usual rules for interning a symbol, sensitive to
.code *package*
in the usual way.
The following notes describe the treatment of "special" directives that are
involved in phrase structure syntax. It applies to all directives which head
off a block that must be terminated by
.codn "@(end)" ,
all "punctuation" directives like
.code "@(and)"
or
.code "@(end)"
and all subphrase indicators like
.code "@(last)"
or
.codn "@(elif)" .
Firstly, each such directive may have a package prefix on its main symbol, yet
is still recognized as the same token. That is to say,
.code "@(foo:collect)"
is still treated by the tokenizer and parser as the
.code "@(collect)"
token, regardless of the package prefix, and regardless of whether
.code foo:end
is the same symbol as the
.code usr:end
symbol.
However, this doesn't mean that any
.code foo:collect
is allowed to denote the
.code collect
directive.
A qualified symbol such as
.code foo:collect
must correspond to (be the same object as) precisely one of two symbols:
either the same-named symbol in the
.code usr
package, or else the same-named symbol in the
.code keyword
package. If this condition isn't satisfied, the situation is a syntax
error. Note that this check uses the original
.code usr
and
.code keyword
packages, not the packages which are currently named
.str "usr"
or
.str "keyword"
in the current
.codn *package-alist* .
A check is also performed for an unqualified symbol.
An unqualified symbol like
.code collect
must also resolve, in the context of the current value of the
.code *package*
variable, to the same named-symbol in either the original
.code usr
or
.code keyword
package. Thus if the current package isn't
.codn usr ,
and
.code "@(collect)"
is being processed, the current package must be such that
.code collect
resolves to
.codn usr:collect .
either because that symbol is present in the current pack via
import, or else visible via the fallback list.
These rules are designed to approximate what the behavior would be
if these directives were actually scanned as Lisp forms in the usual
way and then recognized as phrase structure tokens according to
the identity of their leading symbol. The additional restriction is added that
that the directive symbol names are treated as reserved. If there exists a
user-defined pattern function called
.code mypackage:end
it may not be invoked using the syntax
.codn "@(mypackage:end)" ,
which is erroneous; though it is invocable indirectly via the
.code "@(call)"
directive.
.NP* Package Library Conventions
Various functions in the package and symbol area of the library have a
.meta package
parameter. When the argument is optional, it defaults to the current
value of the
.code *package*
special variable.
If specified, the argument may be a character string, which is taken as the
name of a package. It may also be a symbol, in which case the symbol's name,
which is a character string, is used. Thus the objects
.codn :sys ,
.codn usr:sys ,
.code abc:sys
and
.str sys
all refer to the same package, the system package which is named
.strn sys .
A
.code package
parameter may also simply be a package object.
Some functions, like
.code use-package
and
.code unuse-package
functions accept a list of packages as their first argument.
This may be a list of objects which follow the above conventions:
strings, symbols or package objects.
Also, instead of a list, an atom may be passed: a string, symbol
or package object. It is treated as a singleton list consisting
of that object.
.coNP Variables @, user-package @ keyword-package and @ system-package
.desc
These variables hold predefined packages. The
.code user-package
contains all of the public symbols in the \*(TL library.
The
.code keyword-package
holds keyword symbols, which are printed with
a leading colon. The
.code system-package
is for internal symbols, helping
the implementation avoid name clashes with user code in some situations.
These variables shouldn't be modified. If they are modified, the consequences
are unspecified.
The names of these packages, respectively, are
.strn usr ,
.strn sys ,
and
.strn keyword .
.coNP Special variable @ *package*
.desc
This variable holds the current package. The global value of this variable
is initialized to a package called
.strn pub .
The
.code pub
package has the
.code usr
package in its fallback list; thus when
.code pub
is current, all of the
.code usr
symbols, comprising the content of the \*(TL library, are visible.
All forms read and evaluated from the \*(TX command line, in the interactive listener,
from files via
.code load
or
.code compile-file
or from the \*(TX pattern language are processed in this default
.code pub
package, unless arrangement are made to change to a different package.
The current package is used as the default package for interning symbol tokens
which do not carry the colon-delimited package prefix.
The current package also affects printing. When a symbol is printed whose
home package matches the current package, it is printed without a package
prefix. (Keyword symbols are always printed with the colon prefix, even if the
keyword package is current.)
.coNP Function @ make-sym
.synb
.mets (make-sym << name )
.syne
.desc
The
.code make-sym
function creates and returns a new symbol object. The argument
.metn name ,
which must be a string, specifies the name of the symbol. The symbol
does not belong to any package (it is said to be "uninterned").
Note: an uninterned symbol can be interned into a package with the
.code rehome-sym
function. Also see the
.code intern
function.
.coNP Function @ gensym
.synb
.mets (gensym <> [ prefix ])
.syne
.desc
The
.code gensym
function is similar to
.codn make-sym .
It creates and returns a new symbol object. If the
.meta prefix
argument is omitted, it defaults to
.strn g .
The difference between
.code gensym
and
.code make-sym
is that
.code gensym
creates the symbol's name
by combining the
.meta prefix
with a numeric suffix. The suffix is obtained by incrementing the
.code *gensym-counter*
and taking the new value.
The name string then calculated from the prefix and the counter value
as if by evaluating a form similar to
.codn "(fmt \(dq~a~,04d\(dq prefix counter)" .
From this it can be inferred that
.meta prefix
can be an object of any kind.
Note: the generated symbol's name, though varying thanks to the incrementing
counter, is not the basis of its uniqueness. The basis of the symbol's
uniqueness is that it is a freshly created object, distinct from any other
object. The related function
.code make-sym
still returns unique symbols even if repeatedly called with the same
string argument.
.coNP Special variable @ *gensym-counter*
.desc
This variable is initialized to 0. Each time the
.code gensym
function is called,
it is incremented. The incremented value forms the basis of the numeric
suffix which
.code gensym
uses to form the name of the new symbol.
.coNP Function @ make-package
.synb
.mets (make-package < name <> [ weak ])
.syne
.desc
The
.code make-package
function creates and returns a package named
.metn name ,
where
.meta name
is a string. It is an error if a package by that name exists already.
Note: ordinary creation of packages for everyday program modularization
should be performed with the
.code defpackage
macro rather than by direct use of
.codn make-package .
If the
.meta weak
parameter is given an argument which is a Boolean true, then the resulting
package holds symbols weakly, from a garbage collection point of view. If the
only reference to a symbol is that which occurs inside the weak package, then
that symbol may be removed from the package and reclaimed by the garbage
collector.
Note: weak packages address the following problem. The application creates a
package for the purpose of reading Lisp data. Symbols occurring in that data
therefore are interned into the package. Subsequently, the application retains
references to some of the symbols, discarding the others. If the package isn't
weak, then because the application is retaining some of the symbols, and those
symbols hold a reference to the package, and the package holds a reference to
all symbols that were interned in it, all of the symbols are retained. If a
weak package is used, then the uninterested symbols are eligible for garbage
collection.
.coNP Function @ delete-package
.synb
.mets (delete-package << package )
.syne
.desc
The
.code delete-package
breaks the association between a package and its name.
After
.codn delete-package ,
the
.meta package
object continues to exist, but cannot be found using
.codn find-package .
Furthermore,
.code delete-package
iterates over all remaining packages. For each remaining package
.metn p ,
it performs the semantic action of the
.mono
.meti (unuse-package < package << p )
.onom
expression. That is to say, all of the remaining packages
are scrubbed of any foreign symbols which are the local symbols
of the deleted
.metn package .
.coNP Function @ merge-delete-package
.synb
.mets (merge-delete-package dst-package <> [ src-package ])
.syne
.desc
The
.code merge-delete-package
iterates over all of the local symbols of
.meta src-package
and rehomes each symbol into
.metn dst-package .
Then, it deletes
.metn src-package .
Note: the local symbols are identified as if using
.codn package-local-symbols ,
rehoming is performed as if using
.codn rehome-sym ,
and deleting
.meta src-package
is performed as if using
.codn delete-package .
.coNP Function @ packagep
.synb
.mets (packagep << obj )
.syne
.desc
The
.code packagep
function returns
.code t
if
.meta obj
is a package, otherwise it returns
.codn nil .
.coNP Function @ find-package
.synb
.mets (find-package << name )
.syne
.desc
The argument
.meta name
should be a string. If a package called
.meta name
exists,
then it is returned. Otherwise
.code nil
is returned.
.coNP Special variable @ *package-alist*
.desc
The
.code *package-alist*
variable contains the master association list
which contains an entry about each existing
package.
Each element of the list is a cons cell
whose
.code car
field is the name of a package and whose
.code cdr
is a package object.
Note: the \*(TL application can overwrite or rebind this
variable to manipulate the active package list. This is
useful for
.IR sandboxing :
safely evaluating code that is obtained as an input
from an untrusted source, or calculated from such an input.
The contents of
.code *package-alist*
have security implications because textual source code
can refer to any symbol in any package by invoking
a package prefix. For instance, even if the
.code open
function's name is not available in the current package
(established by the
.code *package*
variable) that symbol can easily be obtained using the
syntax
.codn usr:open .
However, the entire
.code usr
package itself can be removed from
.codn *package-alist* .
In that situation, the syntax
.code usr:open
is no longer valid.
At the same time, selected symbols from the original
.code usr
can be nevertheless made available via some intermediate
package, which is present in
.code *package-alist*
and which contains a subset of the
.code usr
symbols that has been curated for safety. That curated package may even
be called
.codn usr ,
so that if for instance
.code cons
is present in that package, it may be referred to as
.code usr:cons
in the usual way.
.coNP Function @ package-alist
.synb
.mets (package-alist)
.syne
.desc
The
.code package-alist
function retrieves the value of
.codn *package-alist* .
Note: this function is obsolescent. There is no reason to use it
in new code instead of just accessing
.code *package-alist*
directly.
.coNP Function @ package-name
.synb
.mets (package-name << package )
.syne
.desc
The
.code package-name
function retrieves the name of a package.
.coNP Function @ package-symbols
.synb
.mets (package-symbols <> [ package ])
.syne
.desc
The
.code package-symbols
function returns a list of all the symbols
which are interned in
.metn package .
.coNP Functions @ package-local-symbols and @ package-foreign-symbols
.synb
.mets (package-local-symbols <> [ package ])
.mets (package-foreign-symbols <> [ package ])
.syne
.desc
The
.code package-local-symbols
function returns a list of all the symbols
which are interned in
.metn package ,
and whose home package is that package.
The
.code package-foreign-symbols
function returns a list of all the symbols which
are interned in
.metn package ,
which do not have that package as their home package,
or do not have a home package at all.
The union of the local and foreign symbols contains exactly
the same elements as the list returned by
.codn package-symbols :
the symbols interned in a package are partitioned into
local and foreign.
.coNP Functions @ package-fallback-list and @ set-package-fallback-list
.synb
.mets (package-fallback-list << package )
.mets (set-package-fallback-list < package << package-list )
.syne
.desc
The
.code package-fallback-list
returns the current
.I "fallback package list"
associated with
.metn package .
The
.code set-package-fallback-list
replaces the fallback package list of
.meta package
with
.metn package-list .
The
.meta package-list
argument must be a list which is a mixture of symbols, strings or
package objects. Strings are taken to be package names, which must
resolve to existing packages. Symbols are reduced to strings via
.codn symbol-name .
.coNP Functions @ intern and @ intern-fb
.synb
.mets (intern < name <> [ package ])
.mets (intern-fb < name <> [ package ])
.syne
.desc
The argument
.meta name
must be a string. The optional argument
.meta package
must be a package. If
.meta package
is not supplied, then the value
taken is that of
.codn *package* .
The
.code intern
function searches
.meta package
for a symbol called
.metn name .
If that symbol is found, it is returned. If that symbol is not found,
then a new symbol called
.meta name
is created and inserted into
.metn package ,
and that symbol is returned. In this case, the package becomes the
symbol's home package.
The
.code intern-fb
function is very similar to
.code intern
except that if the symbol is not found in
.meta package
then the packages listed in the fallback list of
.meta package
are searched, in order. Only these packages themselves are searched,
not their own fallback lists. If a symbol called
.meta name
is found, the search terminates and that symbol is returned.
Only if nothing is found in the fallback list will
.code intern-fb
create a new symbol and insert it into
.metn package ,
exactly like
.codn intern .
.coNP Function @ unintern
.synb
.mets (unintern < symbol <> [ package ])
.syne
.desc
The
.code unintern
function removes
.meta symbol
from
.metn package .
The
.code nil
symbol may not be removed from the
.code usr
package; an error exception is thrown in this case.
If
.code symbol
isn't
.codn nil ,
then
.meta package
is searched to determine whether it contains
.meta symbol
as an interned symbol (either local or foreign), or a hidden symbol.
If
.meta symbol
is a hidden symbol, then it is removed from the hidden symbol store.
Thereafter, even if a same-named foreign symbol is removed from the
package via
.code unuse-sym
or
.codn unuse-package ,
those operations will no longer restore the hidden symbol to interned
status. In this case,
.meta unintern
returns the hidden symbol that was removed from the hidden store.
If
.meta symbol
is a foreign symbol, then it is removed from the package. If the package
has a hidden symbol of the same name, that hidden symbol is reinterned
in the package, and the package once again becomes its home package.
In this case,
.meta symbol
is returned.
If
.meta symbol
is a local symbol, the symbol is removed from the package.
In this case also,
.meta symbol
is returned.
If
.meta symbol
is not found in the package as either an interned or hidden
symbol, then the function has no effect and returns
.codn nil .
.coNP Functions @ find-symbol and @ find-symbol-fb
.synb
.mets (find-symbol < name >> [ package <> [ notfound-val ]])
.mets (find-symbol-fb < name >> [ package <> [ notfound-val ]])
.syne
.desc
The
.code find-symbol
and
.code find-symbol-fb
functions search
.meta package
for a symbol called
.metn name .
That argument must be a character string.
If the
.meta package
argument is omitted, the parameter defaults to the
current value of
.codn *package* .
If the symbol is found in
.meta package
then it is returned.
If the symbol is not found in
.metn package ,
then the function
.code find-symbol-fb
also searches the packages listed in the fallback list of
.meta package
are searched, in order. Only these packages themselves are searched,
not their own fallback lists. If a symbol called
.meta name
is found, the search terminates and that symbol is returned.
The function
.code find-symbol
only searches
.metn package ,
ignoring its fallback list.
If a symbol called
.meta name
isn't found, then these functions return
.meta notfound-val
is returned, which defaults to
.codn nil .
Note: an ambiguous situation exists when
.meta notfound-val
is a symbol, such as its default value
.codn nil ,
because if that symbol is successfully found,
it is indistinguishable from
.metn notfound-val .
.coNP Function @ rehome-sym
.synb
.mets (rehome-sym < symbol <> [ package ])
.syne
.desc
The arguments
.meta symbol
and
.meta package
must be a symbol and package object,
respectively, and
.meta symbol
must not be the symbol
.codn nil .
The
.code rehome-sym
function moves
.meta symbol
into
.metn package .
If
.meta symbol
is already interned in a package, it is first removed from that package.
If a symbol of the same name exists in
.metn package ,
that symbol is first removed
from
.metn package .
Also, if a symbol of the same name exists in the hidden symbol store of
.metn package ,
that hidden symbol is removed.
Then
.code symbol
is interned into
.metn package ,
and
.meta package
becomes its home package, making it a local symbol of
.metn package .
Note: if
.code symbol
is currently the hidden symbol of some package, it is not removed
from the hidden symbol store of that package. This is a degenerate
case. The implication is that if that hidden symbol is ever
restored in that package, it will once again have that package as
its home package, and consequently it will turn into a foreign
symbol of
.metn package .
.coNP Function @ symbolp
.synb
.mets (symbolp << obj )
.syne
.desc
The
.code symbolp
function returns
.code t
if
.meta obj
is a symbol, otherwise it returns
.codn nil .
.coNP Function @ symbol-name
.synb
.mets (symbol-name << symbol )
.syne
.desc
The
.code symbol-name
function returns the name of
.metn symbol .
.coNP Function @ symbol-package
.synb
.mets (symbol-package << symbol )
.syne
.desc
The
.code symbol-package
function returns the home package of
.metn symbol .
If
.meta symbol
has no home package, it returns
.codn nil .
.coNP Function @ keywordp
.synb
.mets (keywordp << obj )
.syne
.desc
The
.code keywordp
function returns
.code t
if
.meta obj
is a keyword symbol, otherwise it
returns
.codn nil .
.coNP Function @ bindable
.synb
.mets (bindable << obj )
.syne
.desc
The
.code bindable
function returns
.code t
if
.meta obj
is a bindable symbol, otherwise it returns
.codn nil .
All symbols are bindable, except for keyword symbols, and the
special symbols
.code t
and
.codn nil .
.coNP Function @ use-sym
.synb
.mets (use-sym < symbol <> [ package ])
.syne
.desc
The
.code use-sym
function brings an existing
.code symbol
into
.metn package .
In all cases, the function returns
.codn symbol .
If
.meta symbol
is already interned in
.metn package ,
then the function has no effect.
Otherwise
.meta symbol
is interned in
.metn package .
If a symbol having the same name as
.meta symbol
already exists in
.metn package ,
then it is replaced.
If that replaced symbol is a local symbol of
.metn package ,
then the replaced symbol turns into a hidden symbol associated
with the package. It is placed into a special hidden symbol store
associated with
.meta package
and is stripped of its home package, becoming quasi-interned or uninterned.
An odd case is possible whereby
.meta symbol
is already a hidden symbol of
.metn package .
In this case, the hidden symbol replaces some foreign symbol and
is interned in
.metn package .
Thus it simultaneously exists as both an interned
foreign symbol and as a hidden symbol of
.metn package .
.coNP Function @ unuse-sym
.synb
.mets (unuse-sym < symbol <> [ package ])
.syne
.desc
The
.code unuse-sym
function removes
.meta symbol
from
.metn package .
If
.meta symbol
is not interned in
.metn package ,
the function does nothing and returns
.codn nil .
If
.meta symbol
is a local symbol of
.metn package ,
an error is thrown: a package cannot "unuse" its own symbol. Removing
a symbol from its own home package requires the
.code unintern
function.
Otherwise
.meta symbol
is a foreign symbol interned in
.meta package
and is removed.
If the package has a hidden symbol of the same name as
.metn symbol ,
that symbol is reinterned into
.meta package
as a local symbol. In this case, that previously hidden symbol is
returned.
If the package has no hidden symbol matching the removed
.metn symbol ,
then
.meta symbol
itself is returned.
.coNP Functions @ use-package and @ unuse-package
.synb
.mets (use-package < package-list <> [ package ])
.mets (unuse-package < package-list <> [ package ])
.syne
.desc
The
.meta use-package
and
.meta unuse-package
are convenience functions which perform a mass import of symbols from one
package to another, or a mass removal, respectively.
The
.code use-package
function iterates over all of the local symbols of the packages in
.metn package-list .
For each symbol
.metn s ,
it performs the semantic action implied by the
.mono
.meti (use-sym < s << package )
.onom
expression.
Similarly
.code unuse-package
iterates
.meta package-list
in the same way, performing, effectively, the semantic action
of the
.mono
.meti (unuse-sym < s << package )
.onom
expression.
The
.meta package-list
argument must be a list which is a mixture of symbols, strings or
package objects. Strings are taken to be package names, which must
resolve to existing packages. Symbols are reduced to strings via
.codn symbol-name .
.coNP Macro @ defpackage
.synb
.mets (defpackage < name << clause *)
.syne
.desc
The
.code defpackage
macro provides a convenient means to create a package and establish its
properties in a single construct. It is intended for the ordinary situations
in which packages support the organization of programs into modules.
The
.code name
argument, giving the package name, may be a symbol or a character string.
If it is a symbol, then the symbol's name is taken to be name for the
package.
If a package called
.code name
already exists, then
.code defpackage
selects that package for further operations. Otherwise, a new,
empty package is created. In either case, this package is referred
to as the
.I "present package"
in the following descriptions.
The
.code name
may be optionally followed by one or more clauses, which are processed
in the order that they appear. Each clause is a compound form headed
by a keyword.
The supported clauses are as follows:
.RS
.meIP (:fallback << package-name *)
The
.code :fallback
clause specifies the packages to comprise the fallback list of
the present package. If this clause is omitted, or if it is present
with no
.meta package-name
arguments, then the present package has an empty fallback list.
Each
.meta package-name
may be a string or symbol naming an existing package. It is permitted
for the present package itself to appear in its own fallback list.
This is useful for creating a package with a nonempty fallback list
which doesn't actually provide access to any other package.
.meIP (:use << package-name *)
The
.code :use
clause specifies packages whose local symbols are to be interned
into the present package as foreign symbols. Each
.meta package-name
may be a string or symbol naming an existing package.
The list of package names is processed as if by a call to
.codn use-package .
.meIP (:use-syms << symbol *)
The
.code :use-syms
clause specifies individual symbols to be interned in the present package.
The arguments are symbols.
.meIP (:use-from < package-name << symbol-name *)
The
.code :use-from
clause specifies the names of local symbols in a package denoted by
.meta package-name
to be used in the present package. All arguments of
.code :use-from
are either strings or symbols which are reduced to strings by mapping
to their names. Each
.meta symbol-name
is interned in the package identified by
.metn package-name ,
which may have the effect of creating that symbol.
This symbol is expected to be a local symbol of that package. If
that is so, the symbol is brought into the present package via
.codn use-sym .
Otherwise if the symbol is foreign to package identified by
.metn package-name ,
then an error exception is thrown.
.meIP (:local << symbol-name *)
The
.code :local
clause specifies the names of symbols to be interned in the new package
as local symbols. Each
.meta symbol-name
argument must be either a character string or a symbol. If it is a symbol, its
name is taken, thereby reducing the argument to a character string.
The arguments are processed in the order in which they appear. Each name is
first interned in the newly created package using the
.code intern
function. Then, if the resulting symbol is foreign to the package, it is
removed with
.code unuse-sym
and the name is interned again.
.RE
.coNP Macro @ in-package
.synb
.mets (in-package << name )
.syne
.desc
The
.code in-package
macro causes the
.code *package*
special variable to take on the package denoted by
.metn name .
The macro checks, at expansion time, that
.meta name
is either a string or symbol. An error is thrown if
this isn't the case.
The
.meta name
argument expression isn't evaluated, and so must not be quoted.
The code generated by the macro performs a search for the
package. If the package is not found at the time when
the macro's expansion is evaluated, an error is thrown.
.SS* Pseudorandom Numbers
.coNP Special variable @ *random-state*
.desc
The
.code *random-state*
variable holds an object which encapsulates the state
of a pseudorandom number generator. This variable is the default argument
value for the
.code random-fixnum
and
.codn "random functions" ,
for the convenience of writing programs which are not concerned about the
management of random state.
On the other hand, programs can create and manage random states, making it
possible to obtain repeatable sequences of pseudorandom numbers which do not
interfere with each other. For instance objects or modules in a program can
have their own independent streams of random numbers which are repeatable,
independently of other modules making calls to the random number functions.
When \*(TX starts up, the
.code *random-state*
variable is initialized with
a newly created random state object, which is produced as if by
the call
.codn "(make-random-state 42)" .
.coNP Special variable @ *random-warmup*
.desc
The
.code *random-warmup*
special variable specifies the value which is used by
.code make-random-state
in place of a missing
.meta warmup-period
argument.
To "warm up" a pseudorandom number generator (PRNG) means to obtain some
values from it which are discarded, prior to use. The number of values
discarded is the
.IR "warm-up period" .
The WELL512a PRNG used in \*(TX produces 32-bit values, natively. Thus each
warm-up iteration retrieves and discards a 32-bit value. The PRNG has
a state space consisting of a vector of sixteen 32-bit words, making
the state space 4096 bits wide.
Warm up is required because PRNG-s, in particular PRNG-s with large state
spaces and long periods, produce fairly predictable sequences of values in the
beginning, before transitioning into chaotic behavior. This problem is worse
for low complexity seeds, such as small integer values.
The sequences are predictable in two ways. Firstly, some initial values
extracted from the PRNG may exhibit patterns ("problem 1"). Secondly, the initial values
from sequences produced from similar seeds (for instance consecutive integers)
may be similar or identical ("problem 2").
.TP* Notes:
The default value of
.code *random-warmup*
is only 8. This is insufficient to
ensure good initial PRNG behavior for seeds even as large as 64 bits or more.
That is to say, even if as many as eight bytes' worth of true random bits are
used as the seed, the PRNG will exhibit predictable behaviors, and a poor
distribution of values.
Applications which critically depend on good PRNG behavior should choose
large warm-up periods into the hundreds or thousands of iterations.
If a small warm-up period is used, it is recommended to use larger seeds
which initialize more of the 4096-bit state space.
\*(TX's PRNG implementation addresses "problem 1" by padding the
unseeded portions of the state space with random values (from a static table
that doesn't change). For instance, if the integer 1 is used to seed the space,
then one 32-bit word of the space is set to the value 1. The remaining 15 are
populated from the random table. This helps to ensure that a good PRNG sequence
is obtained immediately. However, it doesn't address "problem 2": that
similar seed values generate similar sequences, when the warm-up period is
small. For instance, if 65536 different random state objects are created, from
each of the 16-bit seeds in the range [0, 65536), and then a random 16-bit
value is extracted from each state, only 1024 unique values result.
.coNP Function @ make-random-state
.synb
.mets (make-random-state >> [ seed <> [ warmup-period ])
.syne
.desc
The
.code make-random-state
function creates and returns a new random state,
an object of the same kind as what is stored in the
.code *random-state*
variable.
The seed, if specified, must be an integer value, a buffer,
an existing random state object, or else a vector returned from a call to the
function
.codn random-state-get-vec .
Note that the sign of the seed is ignored, so that negative seed
values are equivalent to their additive inverses.
If seed is not specified, then
.code make-random-state
produces a seed based
on some information in the process environment, such as current
time of day. It is not guaranteed that two calls to
.code make-random-state
that are separated by less than some minimum increment of real time produce
different seeds. The minimum time increment depends on the platform.
On a platform with a millisecond-resolution real-time clock, the minimum
time increment is a millisecond. Calls to
.code make-random-state
less than a millisecond apart may predictably produce the same seed.
If an integer or buffer seed is specified, then the integer value is mapped to
a pseudorandom sequence, in a platform-independent way.
If an existing random state is specified as a seed, then it is duplicated. The
returned random state object is a distinct object which is in the same
state as the input object. It will produce the same remaining pseudorandom
number sequence, as will the input object.
If a vector is specified as a seed, then a random state is constructed
which duplicates the random state object which was captured in that vector
representation by the
.code random-state-get-vec
function.
The
.meta warm-up-period
argument specifies the number of values which are immediately obtained and
discarded from the newly-seeded generator before it is returned. This
procedure is referred to as PRNG
.IR warm-up .
Warm-up is not performed if
.meta seed
is a vector or random state object. In this situation, if the
.meta warm-up-period
is present, it may still be required to be an integer, even though it is ignored.
If warm-up is performed, but the
.meta warm-up-period
argument is missing, then the value of the
.code *random-warmup*
special variable is used. Note: this variable has a default value which may be too
small for some applications of pseudorandom numbers; see the Notes under
.codn *random-warmup* .
.coNP Function @ random-state-p
.synb
.mets (random-state-p << obj )
.syne
.desc
The
.code random-state-p
function returns
.code t
if
.meta obj
is a random state, otherwise it
returns
.codn nil .
.coNP Function @ random-state-get-vec
.synb
.mets (random-state-get-vec <> [ random-state ])
.syne
.desc
The
.code random-state-get-vec
function converts a random state into a vector of integer values.
If the
.meta random-state
argument, which must be a random state object, is omitted,
then the value of the
.code *random-state*
is used.
.coNP Functions @, random-fixnum @ random and @ rand
.synb
.mets (random-fixnum <> [ random-state ])
.mets (random < random-state << modulus )
.mets (rand < modulus <> [ random-state ])
.syne
.desc
All three functions produce pseudorandom numbers, which are positive integers.
The numbers are obtained from a WELL512a PRNG, whose state is stored in the
random state object.
The
.code random-fixnum
function produces a random fixnum integer: a reduced range
integer which fits into a value that does not have to be heap-allocated.
The
.code random
and
.code rand
functions produce a value in the range [0,
.metn modulus ).
They differ only in the order of arguments. In the
.code rand
function, the random state
object is the second argument and is optional. If it is omitted, the global
.code *random-state*
object is used.
The
.meta modulus
argument must be a positive integer. If
.meta modulus
is 1, then the function returns zero without altering the state of the
pseudorandom number generator.
.coNP Functions @ random-float and @ random-float-incl
.synb
.mets (random-float <> [ random-state ])
.mets (random-float-incl <> [ random-state ])l
.syne
.desc
The
.code random-float
function produces a pseudorandom floating-point value in the range [0.0, 1.0).
The
.code random-float-incl
produces a pseudorandom floating-point value in the range [0.0, 1.0], thus
differing from
.code random-float
by including the 1.0 limit value.
The numbers are obtained from a WELL512a PRNG, whose state is stored in the
random state object given by the argument to the optional
.meta random-state
parameter, which defaults to the value of
.codn *random-state* .
Because the floating-point type does not provide a representation of every real
value in the range 0.0 to 1.0, it is not possible to impose the requirement
that every value shall occur with equal likelihood.
Rather, these functions are intended to produce an a uniform distribution of
values according to the following pragmatic requirements. A subset
.I S
of the real values in the
specified range, [0.0, 1.0) or [0.0, 1.0] is identified whose elements are representable
in the floating-point type and which are uniformly spaced along the interval. Then,
a random element is chosen from
.I S
and returned, such that every element is equally likely to be selected.
Note that these requirements do not correspond to the more mathematically ideal
concept of uniformly choosing actual real numbers in the [0, 1] interval of the
real number line, and then finding the closest floating-point representation.
Such a requirement would mean that the boundary values 0.0 and 1.0 appear in
the output half as frequently as all the interior values, because each of these
two floating-point values is a representations of a range of numbers, half of
which lies outside of the [0, 1] interval.
.coNP Function @ random-buf
.synb
.mets (random-buf < size <> [ random-state ])
.syne
.desc
The
.code random-buf
function creates a
.code buf
object of the specified
.meta size
fills it with pseudorandom bytes, and returns it.
The bytes are obtained from the random state object given by the optional
.meta random-state
parameter, which defaults to the value of
.codn *random-state* .
See the section
.B Buffers
for a description of
.code buf
objects.
.coNP Function @ random-sample
.synb
.mets (random-sample < size < seq <> [ random-state ])
.syne
.desc
The
.code random-sample
function returns a vector of
.meta size
randomly selected elements from the sequence
.metn seq ,
using reservoir sampling.
If the number of elements in
.meta seq
is equal to or smaller than
.metn size ,
then the function returns a vector of all the elements of
.meta seq
in their original order.
In other cases, the selected elements are not required to appear
in their original order.
No element of sequence
.meta seq
is selected more than once; duplicate values can appear
in the output only if
.meta seq
itself contains duplicates.
.SS* Time
.coNP Functions @, time @ time-usec and @ time-nsec
.synb
.mets (time)
.mets (time-usec)
.mets (time-nsec)
.syne
.desc
The
.code time
function returns the number of seconds that have elapsed since
midnight, January 1, 1970, in the UTC timezone: a point in
time called
.IR "the epoch" .
The
.code time-usec
function returns a cons cell whose
.code car
field holds the seconds measured in the same way, and whose
.code cdr
field extends the precision by giving
number of microseconds as an integer value between 0 and 999999.
The
.code time-nsec
function is similar to
.code time-usec
except that the returned cons cell's
.code cdr
field gives a number of nanoseconds as an integer value
between 0 and 999999999.
Note: on hosts where obtaining nanosecond precision is not available, the
.code time-nsec
function obtains a microseconds value instead, and multiplies
it by 1000.
.coNP Functions @ time-string-local and @ time-string-utc
.synb
.mets (time-string-local < time << format )
.mets (time-string-utc < time << format )
.syne
.desc
These functions take the numeric time returned by the
.code time
function, and convert it to a textual representation in a flexible way,
according to the contents of the
.meta format
string.
The
.code time-string-local
function converts the time to the local timezone of
the host system. The
.code time-string-utc
function produces time in UTC.
The
.meta format
argument is a string, and follows exactly the same conventions as
the format string of the C library function
.codn strftime .
The
.meta time
argument is an integer representing seconds obtained from the
time function or from the
.code car
field of the cons returned by the
.code time-usec
function.
.coNP Functions @ time-fields-local and @ time-fields-utc
.synb
.mets (time-fields-local << time )
.mets (time-fields-utc << time )
.syne
.desc
These functions take the numeric time returned by the time function,
and convert it to a list of seven fields.
The
.code time-string-local
function converts the time to the local timezone of
the host system. The
.code time-string-utc
function produces time in UTC.
The fields returned as a list consist of six integers, and a Boolean value.
The six integers represent the year, month, day, hour, minute and second.
The Boolean value indicates whether daylight savings time is in effect
(always
.code nil
in the case of
.codn time-fields-utc ).
The
.meta time
argument is an integer representing seconds obtained from the
.code time
function or from the
.code time-usec
function.
.coNP Structure @ time
.synb
.mets (defstruct time nil
.mets \ \ year month day hour min sec dst
.mets \ \ gmtoff zone)
.syne
.desc
The
.code time
structure represents a time broken down into individual fields.
The structure almost directly corresponds to the
.code "struct tm"
type in the ISO C language. There are differences.
Whereas the
.code "struct tm"
member
.code tm_year
represents a year since 1900, the
.code year
slot of the
.code time
structure represents the absolute year, not relative to 1900.
Furthermore, the
.code month
slot represents a one-based numeric month, such that 1 represents
January, whereas the C member
.code tm_mon
uses a zero-based month. The
.code dst
slot is a \*(TL Boolean value. The slots
.codn hour ,
.codn min ,
and
.code sec
correspond directly to
.codn tm_hour ,
.codn tm_min ,
and
.codn tm_sec .
The slot
.code gmtoff
represents the number of seconds east of UTC, and
.code zone
holds a string giving the abbreviated time zone name.
On platforms where the C type
.code "struct tm"
has fields corresponding to these slots, values for
these slots are calculated and stored into them by the
.code time-struct-local
and
.code time-struct-utc
functions, and also the related
.code time-local
and
.code time-utc
methods. On platforms where the corresponding fields are not
present in the C language
.codn "struct tm" ,
these slots are unaffected by those functions,
retaining the default initial value
.code nil
or a previously stored value, if applicable.
Lastly, the values of
.code gmtoff
and
.code zone
are not ignored by functions which accept a
.code time
structure as a source of input values.
.coNP Functions @ time-struct-local and @ time-struct-utc
.synb
.mets (time-struct-local << time )
.mets (time-struct-utc << time )
.syne
.desc
These functions take the numeric time returned by the time function,
and convert it to an instance of the
.code time
structure.
The
.code time-struct-local
function converts the time to the local timezone of
the host system. The
.code time-struct-utc
function produces time in UTC.
The
.meta time
argument is an integer representing seconds obtained from the
.code time
function or from the
.code time-usec
function.
.coNP Functions @, time-parse @ time-parse-local and @ time-parse-utc
.synb
.mets (time-parse < format << string )
.mets (time-parse-local < format << string )
.mets (time-parse-utc < format << string )
.syne
.desc
The
.code time-parse
function scans a time description in
.meta string
according to the specification given in the
.meta format
string. If the scan is successful, a structure
of type
.code time
is returned, otherwise
.codn nil .
The
.meta format
argument follows the same conventions as the POSIX
C library function
.codn strptime .
Prior to obtaining the time from
.meta format
and
.meta string
the returned structure is created and initialized
with a time which represents time 0 ("the epoch")
if interpreted in the UTC timezone as by the
.meta time-utc
method.
The
.code time-parse-local
and
.code time-parse-utc
functions return an integer time value: the same value
that would be returned by the
.code time-local
and
.code time-utc
methods, respectively, when applied to the structure
object returned by
.codn time-parse .
Thus, these equivalences hold:
.verb
(time-parse-local f s) <--> (time-parse f s).(time-local)
(time-parse-utc f s) <--> (time-parse f s).(time-utc)
.brev
Note: the availability of these three functions
depends on the availability of
.codn strptime .
.coNP Methods @ time-local and @ time-utc
.synb
.mets << time-struct .(time-local)
.mets << time-struct .(time-utc)
.syne
.desc
The
.code time
structure has two methods called
.code time-local
and
.codn time-utc .
The
.code time-local
function considers the slots of the
.code time
structure instance
.meta time-struct
to be local time, and returns its integer representation
as the number of seconds since the epoch.
The
.code time-utc
function is similar, except it considers
the slots of
.meta time-struct
to be in the UTC time zone.
Note: these functions work by converting the slots into arguments
which are applied to
.code make-time
or
.codn make-time-utc .
.coNP Method @ time-string
.synb
.mets << time-struct .(time-string << format )
.syne
.desc
The
.code time
structure has a method called
.codn time-string .
This method accepts a
.meta format
string argument, which it uses to convert
the fields to a character string representation
which is returned.
The
.meta format
argument is a string, and follows exactly the same conventions as
the format string of the C library function
.codn strftime .
.coNP Method @ time-parse
.synb
.mets << time-struct .(time-parse < format << string )
.syne
.desc
The
.code time-parse
method scans a time description in
.meta string
according to the specification given in the
.meta format
string.
If the scan is successful, the structure
is updated with the parsed information, and
the remaining unmatched portion of
.meta string
is returned. If all of
.meta string
is matched, then an empty string is returned.
Slots of
.meta time-struct
which are originally
.code nil
are replaced with zero, even if these
zero values are not actually parsed from
.metn string .
If the scan is unsuccessful, then
.code nil
is returned and the structure is not
altered.
The
.meta format
argument follows the same conventions as the POSIX
C library function
.codn strptime .
Note: the
.code time-parse
method may be unavailable if the host system does not
provide the
.code strptime
function. In this case, the
.code time-parse
static slot of the
.code time
struct is
.codn nil .
.coNP Functions @ make-time and @ make-time-utc
.synb
.mets (make-time < year < month < day
.mets \ \ \ \ \ \ \ \ \ \ < hour < minute < second << dst-advice )
.mets (make-time-utc < year < month < day
.mets \ \ \ \ \ \ \ \ \ \ \ \ \ \ < hour < minute < second << dst-advice )
.syne
.desc
The
.code make-time
function returns a time value, similar to the one returned by the
.code time
function. The
.code time
value is constructed not from the system clock, but
from a date and time specified as arguments. The
.meta year
argument is a calendar year, like 2014.
The
.meta month
argument ranges from 1 to 12.
The
.meta hour
argument is a 24-hour time, ranging from 0 to 23.
These arguments represent a local time, in the current time zone.
The
.meta dst-advice
argument specifies whether the time is expressed in
daylight savings time (DST). It takes on three possible values:
.codn nil ,
the keyword
.codn :auto ,
or else the symbol
.codn t .
Any other value has the same interpretation as
.codn t .
If
.meta dst-advice
is
.codn t ,
then the time is assumed to be expressed in DST.
If the argument is
.codn nil ,
then the time is assumed not to be in DST.
If
.meta dst-advice
is
.codn :auto ,
then the function tries to determine whether
DST is in effect in the current time zone for the specified date and time.
The
.code make-time-utc
function is similar to
.codn make-time ,
except that
it treats the time as UTC rather than in the local time zone.
The
.meta dst-advice
argument is supported by
.code make-time-utc
for function
call compatibility with
.codn make-time .
It may or may not have any effect
on the output (since the UTC zone by definition doesn't have daylight
savings time).
.SS* Data Integrity
.coNP Function @ crc32-stream
.synb
.mets (crc32-stream < stream >> [ nbytes <> [ crc-prev ]])
.syne
.desc
The
.code crc32-stream
calculates the CRC-32 sum over the bytes read from
.metn stream ,
starting at the stream's current position.
If the
.meta nbytes
argument is specified, it should be a nonnegative
integer. It gives the number of bytes which should be read
and included in the sum. If the argument is omitted, then bytes are read
until the end of the stream.
The optional
.meta crc-prev
argument defaults to zero. It is fully documented under the
.code crc32
function.
The
.code crc32-stream
functions returns the calculated CRC-32 as a nonnegative integer.
.coNP Function @ crc32
.synb
.mets (crc32 < obj <> [ crc-prev ])
.syne
.desc
The
.code crc32
function calculates the CRC-32 sum over
.metn obj ,
which may be a character string or a buffer.
If
.meta obj
is a buffer, then the sum is calculated over all of the bytes contained
in that buffer, according to its current length.
If
.meta obj
is a character string, then the sum is calculated over the bytes
which constitute its UTF-8 representation.
The optional
.meta crc-prev
argument defaults to zero. If specified, it should be a nonnegative integer in
the 32-bit range. This argument is useful when a single CRC-32 must be
calculated in multiple operations over several objects. The first call should
specify a value of zero, or omit the argument. To continue the checksum,
each subsequent call to the function should pass as the
.meta crc-prev
argument the CRC-32 obtained from the previous call.
The
.code crc32
function returns the calculated CRC-32 as a nonnegative integer.
.TP* Examples:
.mono
;; Single operation
(crc32 "ABCD") --> 3675725989
;; In two steps, demonstrating crc-prev argument:
(crc32 "CD" (crc32 "AB")) -> 3675725989
.onom
.coNP Functions @ sha256-stream and @ md5-stream
.synb
.mets (sha256-stream < stream >> [ nbytes <> [ buf ]])
.mets (md5-stream < stream >> [ nbytes <> [ buf ]])
.syne
.desc
The
.code sha256-stream
calculates the NIST SHA-256 digest over the bytes read from
.metn stream ,
starting at the stream's current position.
The
.code md5-stream
function calculates the MD5 digest, using the
RSA Data Security, Inc. MD5 Message-Digest Algorithm.
If the
.meta nbytes
argument is specified, it should be a nonnegative
integer. It gives the number of bytes which should be read
and included in the digest. If the argument is omitted, then bytes are read
until the end of the stream.
If the
.meta buf
argument is omitted, the digest value is returned as a new,
buffer object. This buffer is 32 bytes long in the case of SHA-256,
holding a 256-bit digest, and 16 bytes long in the case of MD5,
holding a 128-bit digest.
If the
.meta buf
argument is specified, it must be a buffer that is at least 16 bytes long
in the case of MD5, and at least 32 bytes long in the case of SHA-256.
The hash is placed into that buffer, which is then returned.
.coNP Functions @ sha256 and @ md5
.synb
.mets (sha256 < obj <> [ buf ])
.mets (md5 < obj <> [ buf ])
.syne
.desc
The
.code sha256
function calculates the NIST SHA-256 digest over
.metn obj ,
which may be a character string or a buffer.
Similarly, the
.code md5
functions calculates the MD5 digest over
.metn obj ,
using the RSA Data Security, Inc. MD5 Message-Digest Algorithm.
If
.meta obj
is a buffer, then the digest is calculated over all of the bytes contained
in that buffer, according to its current length.
If
.meta obj
is a character string, then the digest is calculated over the bytes
which constitute its UTF-8 representation.
If the
.meta buf
argument is omitted, the digest value is returned as a new,
buffer object. This buffer is 32 bytes long in the case of SHA-256,
holding a 256-bit digest, and 16 bytes long in the case of MD5,
holding a 128-bit digest.
If the
.meta buf
argument is specified, it must be a buffer that is at least 16 bytes long
in the case of MD5, and at least 32 bytes long in the case of SHA-256.
The hash is placed into that buffer, which is then returned.
.coNP Functions @, sha256-begin @ sha256-hash and @ sha256-end
.synb
.mets (sha256-begin)
.mets (sha256-hash < ctx << obj )
.mets (sha256-end < ctx <> [ buf ])
.syne
.desc
The three functions
.codn sha256-begin ,
.code sha256-hash
and
.code sha256-end
implement a stateful computation of SHA256 digest which allows multiple input
sources to contribute to the result. Furthermore, the context object may be
serially reused for calculating multiple digests.
The
.code sha256-begin
function, which takes no arguments, returns a new SHA256 digest-producing
context object.
The
.code sha256-hash
updates the state of the SHA256 digest object
.meta ctx
by including
.meta obj
into the digest calculation. The
.meta obj
argument may be: a character or character string, whose UTF-8 representation is
digested; a buffer object, whose contents are digested; or an integer,
representing a byte value in the range 0 to 255 included in the digest.
The
.code sha256-hash
function may be called multiple times to include any mixture of
strings and buffers into the digest calculation.
The
.code sha256-end
function finalizes the digest calculation and returns the digest in
a buffer. If the
.meta buf
argument is omitted, then a new 32-byte buffer is created for this
purpose. Otherwise,
.meta buf
must specify a
.code buf
object that is at least 32 bytes long. The digest is stored into this
buffer and that the buffer is returned.
The
.code sha256-end
function additionally resets the
.meta ctx
object into the initial state of a newly created context object, so
that it may be used for another digest session.
.coNP Functions @, md5-begin @ md5-hash and @ md5-end
.synb
.mets (md5-begin)
.mets (md5-hash < ctx << obj )
.mets (md5-end < ctx <> [ buf ])
.syne
.desc
The three functions
.codn md5-begin ,
.code md5-hash
and
.code md5-end
implement a stateful computation of MD5 digest which allows multiple input
sources to contribute to the result. Furthermore, the context object may be
serially reused for calculating multiple digests.
The
.code md5-begin
function, which takes no arguments, returns a new MD5 digest-producing
context object.
The
.code md5-hash
updates the state of the MD5 digest object
.meta ctx
by including
.meta obj
into the digest calculation. The
.meta obj
argument may be: a character or character string, whose UTF-8 representation is
digested; a buffer object, whose contents are digested; or an integer,
representing a byte value in the range 0 to 255 included in the digest.
The
.code md5-hash
function may be called multiple times to include any mixture of
strings and buffers into the digest calculation.
The
.code md5-end
function finalizes the digest calculation and returns the digest in
a buffer. If the
.meta buf
argument is omitted, then a new 16-byte buffer is created for this
purpose. Otherwise,
.meta buf
must specify a
.code buf
object that is at least 16 bytes long. The digest is stored into this
buffer and that the buffer is returned.
The
.code md5-end
function additionally resets the
.meta ctx
object into the initial state of a newly created context object, so
that it may be used for another digest session.
.SS* The Awk Utility
The \*(TL library provides a macro called
.code awk
which is inspired by the Unix utility Awk. The macro implements
a processing paradigm similar to that of the utility: it scans
one or more input streams, which are divided into records or fields,
under the control of user-settable regular-expression-based delimiters.
The records and fields are matched against a sequence of programmer-defined
conditions (called "patterns" in the original Awk), which have associated
actions. Like in Awk, the default action is to print the current record.
Unlike Awk, the
.code awk
macro is a robust, self-contained language feature which can be used
anywhere where a \*(TL expression is called for, cleanly nests
with itself and can produce a return value when done. By contrast,
a function in the Awk language, or an action body, cannot instantiate
a local Awk processing machine.
The
.code awk
macro implements some of the most important Awk
conventions and semantics, in Lisp syntax, while eschewing others.
It does not implement the Awk convention that
variables become defined upon first mention; variables must be
defined to be used. It doesn't implement Awk's weak type system.
A character string which looks like a number isn't a number,
and an empty string or undefined variable doesn't serve as zero
in arithmetic expressions enclosed in the macro.
All expression evaluation within
.code awk
is the usual \*(TL evaluation.
The
.code awk
macro also does not provide a library of functions corresponding to
those in the Awk library, nor does it provide counterparts various
global variables in Awk such as the
.code ENVIRON
and
.code PROCINFO
arrays, or
.code RSTART
and
.codn RLENGTH .
Such features of Awk are extraneous to its central paradigm.
.coNP Macro @ awk
.synb
.mets (awk >> {( condition << action *)}*)
.syne
.desc
The
.code awk
macro processes one or more input sources, which may be streams or
files. Each input source is scanned into records, and each record
is broken into fields. For each record, the sequence of condition-action
clauses (except for certain special clauses) is processed. Every
.meta condition
is evaluated, and if it yields true, the corresponding
.metn action s
are evaluated.
The
.meta condition
and
.meta action
forms are understood to be in a scope in which certain local
identifiers exist in the variable namespace as well as in the function
namespace. These are called
.I "awk functions"
and
.IR "awk macros" .
If
.meta condition
is one of the following keyword symbols, then it is a special clause,
with special semantics:
.codn :name ,
.codn :let ,
.codn :inputs ,
.codn :output ,
.codn :begin ,
.codn :set ,
.codn :end ,
.codn :begin-file ,
.code :set-file
and
.codn :end-file .
These clause types are explained below.
In such a clause, the
.meta action
expressions are not necessarily forms to be evaluated; the treatment
of these expressions depends on the clause. Otherwise, if
.meta condition
is not one of the above keyword symbols, the clause is an ordinary
condition-action clause, and
.meta condition
is a \*(TL expression, evaluated to determine a Boolean value
which controls whether the
.meta action
forms are evaluated. In every ordinary condition-action clause which
contains no
.meta action
forms, the
.code awk
macro substitutes the single action equivalent to the form
.codn "(prn)" :
a call to the local
.code awk
function
.codn prn .
The behavior of this macro, when called with no arguments, as above,
is to print the current
record (contents of the variable
.codn rec )
followed by the output record terminator from the variable
.codn ors .
While the processing loop in
.code awk
scans an input source, it also binds the special variable
.code *stdin*
to the open stream associated with that source. This binding is
in effect across all ordinary clauses, as well as across the
special clauses
.code :begin-file
and
.codn :end-file .
The following is a description of the special clauses:
.RS
.meIP (:name << obj )
The
.code :name
clause establishes the name of the implicit block contained
within the expansion of the
.code awk
macro to be the object
.metn obj ,
usually a symbol.
Forms enclosed in the macro can use
.code return-from
to abandon the
.code awk
form, specifying the same object as the argument.
If the
.code :name
form is omitted, the implicit block is named
.codn awk .
It is an error for two or more
.code :name
forms to appear.
Note: in \*(TX 255 and older, the
.code :name
clause must have an argument which is a symbol.
The symbol
.code nil
is not permitted.
.meIP (:let >> { sym | >> ( sym << init-form )}*)
Regardless of what order they appear in relation to
other clauses in the same
.code awk
macro,
.code :let
clauses are evaluated first before the macro takes any other action. The
argument forms of this clause are variables or variable-init forms. They are
treated the same way as analogous forms in the
.code let*
special form. Note that these are not enclosed in an extra list
as they are in the that form. The bindings established by the
.code :let
clause have a scope which extends over all the other clauses in the
.code awk
macro.
If multiple
.code :let
clauses are present, they are effectively consolidated into
a single clause, in the order they appear.
Note that the lexical variables, functions and macros established by the
.code awk
macro
(called, respectively,
.IR "awk macros" ,
.I "awk functions"
and
.IR "awk variables" )
are in an inner scope relative to
.code :let
bindings. For instance if
.code :let
creates a binding for a variable called
.codn fs ,
that variable will be visible only to subsequent forms appearing
in the same
.code :let
clause or later
.code :let
clauses, and also visible in
.code :inputs
and
.code :output
clauses.
In
.codn :begin ,
.codn :set ,
.codn :end ,
and ordinary clauses, it will be shadowed by the
.code awk
variable
.codn fs ,
which holds the field-separator regular expression or string.
.meIP (:inputs << source-form *)
The
.code :inputs
clause is evaluated by the
.code awk
macro after processing the
.code :let
clauses. Each
.meta source-form
is evaluated and the values of these forms are gathered into a list.
This list then comprises the list of input sources for the
.code awk
processing task.
Each input source must be one of three kinds of objects.
It may be a stream object, which must be capable of character
input. It may be a list of strings, which
.code awk
will convert to an input stream as if by the
.code make-strlist-input-stream
function.
Or else it must be a character
string, which denotes a filesystem pathname which
.code awk
will open for reading.
If the
.code :inputs
clause is omitted, then a defaulting behavior occurs for obtaining
the list of input sources. If the special variable
.code *args*
isn't the empty list, then
.code *args*
is taken as the input sources. Otherwise, the
.code *stdin*
stream is taken as the one and only input source.
If the
.code awk
macro uses
.code *args*
via the above defaulting behavior, it copies
.code *args*
and sets that variable to
.codn nil .
This is done in order that if
.code awk
is used from the \*(TX command line, for example using the
.code -e
command-line option, after
.code awk
terminates, \*(TX will not try to open the next argument
as a script file or treat it as an option.
Note: programs which want
.code awk
not to modify
.code *args*
can explicitly specify
.code *args*
as the argument to the
.code :inputs
keyword, rather than allow
.code *args*
to be used through the defaulting behavior. Only the
defaulting behavior consumes the arguments by overwriting
.code *args*
with
.codn nil .
It is an error to specify more than one
.code :inputs
clause.
.meIP (:output << output-form )
The
.code :output
clause is processed just after the
.code :inputs
clause. It must have exactly one argument, which is an expression
that evaluates to a string, or else to an output stream.
If it evaluates to a string, then that string is used as the name
of a file to open for writing, and the resulting stream
is taken in place of that string.
The
.code :output
clause, if present, has the effect of creating a local binding for the
.code *stdout*
special variable.
This new value of
.code *stdout*
is visible to all forms within the macro.
If a
.code :let
clause is present, it establishes bindings
in a scope which is nested within the scope established
by
.codn :output .
Therefore,
.metn init-form s
in the
.code :let
may refer to the new value of
.code *stdout*
established by
.codn :output .
Furthermore,
.code :let
can rebind
.codn *stdout* ,
causing the definition provided by
.code :output
to be shadowed.
In the case when the
.code :output
argument is a string such that a new stream is opened
on the file, the
.code awk
macro will close that stream when it finishes executing.
Moreover, that stream is treated uniformly as a member of
the set of streams that are implicitly managed by the
redirection macros in the same
.code awk
macro invocation. In brief, the implication is that if
.code :output
creates a stream for the file pathname
.str "out.txt"
and somewhere in the same
.code awk
macro, there is a redirection of the form, or equivalent to
.mono
(-> "out.txt")
.onom
then this redirection shall refer to the same stream
that was established by
.codn :output .
Note also that in this example situation, the expression
.mono
(-> "out.txt" :close)
.onom
has the effect of closing the
.code :output
stream.
.meIP (:begin << form *)
All
.code :begin
clauses are processed in the order in which they appear, before
input processing begins.
Each
.code form
is evaluated. These forms have in their scope the local
.code awk
variables and macros.
.meIP (:set >> { place << new-value }*)
The
.code :set
clause provides a shorthand which allows the frequently occurring pattern
.code "(:begin (set ...))"
to be condensed to
.codn "(:set ...)" .
.meIP (:end << form *)
All
.code :end
clauses are processed, in the order in which they appear,
when the input processing loop terminates.
This termination occurs when all records
from all input sources are either processed or skipped, or else
by an explicit termination such
as a dynamic nonlocal transfer, such as
.codn return-from ,
or the throwing of an exception.
Upon termination, the end clauses are processed in the order they appear. Each
.code form
is evaluated, left to right.
In the normal termination case, the value of the last
.meta form
of the last end clause appears as the return value of the
.code awk
macro.
Note that only termination of the
.code awk
macro initiated from condition-action clauses,
.code :begin-file
clauses, or
.code :end-file
clauses triggers
.code :end
clause processing.
If termination of the
.code awk
macro is initiated from within a
.codn :let ,
.codn :inputs ,
.code :output
or
.code :begin
clause, then end
clauses are not processed.
If an
.code :end
clause performs a nonlocal transfer, the remaining
.code :end
forms in that clause and
.code :end
clauses which follow are not evaluated.
.meIP (:begin-file << form *)
All
.code :begin-file
clauses are processed in the order in which they appear, before
.code awk
switches to each new input.
If both
.code :begin
and
.code :begin-file
forms are specified, then before the first input is processed,
.code :begin
clauses are processed first, then the
.code :begin-file
clauses.
.meIP (:set-file >> { place << new-value }*)
The
.code :set-file
clause is a shorthand which translates
.code "(:set-file ...)"
to
.codn "(:begin-file (set ...))" .
.meIP (:end-file << form *)
All
.code :end-file
clauses are processed after the processing of an input
source finishes.
If both
.code :end
and
.code :end-file
forms are specified, then before after the last input is processed,
.code :end-file
clauses are processed first, then the
.code :end
clauses.
The
.code :end-file
clauses are processed unconditionally, no matter how
the processing of an input source terminates, whether terminated
naturally by running out of records, prematurely by invocation of the
.code next-file
macro, or via a dynamic nonlocal control transfer such as a block
return or exception throw.
If a
.code :begin-file
clause performs a nonlocal transfer,
.code :end-file
processing is not triggered, because the processing of the input
source is deemed not to have taken place.
.meIP (:fields >> { sym | >> ( sym <> [ fun ]) | -}*)
The
.code :fields
clause may be specified in order to give symbolic names to fields,
and optionally specify conversions for them.
Every argument must be one of three expressions. It may be
a bindable symbol other than
.code -
(minus). It may be a list whose first element is
a symbol other than
.code -
optionally followed the name of a function.
Or else it may be the
.code -
symbol, which has a special meaning.
Symbols other than
.code -
may not be repeated, and the
.code :fields
clause may appear at most once in a given instance of the
.code awk
macro.
Each argument is understood to correspond to a field expression for a successive field,
starting with the leftmost
.meta sym
corresponding with the first field,
.codn "[f 0]" .
Each
.meta sym
other than
.code -
becomes the name of a symbol macro which denotes its corresponding
field expression, expanded over the scope of the
.code awk
macro. The
.code -
symbol is a placeholder which doesn't bind a symbol macro to the
corresponding field.
Additionally, every two-element entry which associates the field symbol
.meta sym
with a function name
.meta fun
specifies a field conversion. After each record is read and divided into
fields, those fields for which
.meta fun
is specified are updated by passing their value to this function
and replacing them by the returned value.
The
.meta fun
symbol may also be one of the short-hand symbols available in the
.code fconv
macro, such as
.codn i ,
.code x
and others.
If at least one such conversion is specified in a
.code :fields
clause, then the value of
.code rec
is updated from the converted fields in the usual manner, as if
the fields had been assigned.
Furthermore, it is ensured that every field for which a
.code :fields
clause specifies a conversion exists. Fields with an empty string
value are automatically added so that a field exists for the
rightmost conversion, and the value of
.code nf
is updated to include these fields.
.meIP >> ( condition << action *)
Clauses which do not have one of the specially recognized keywords
in the first position are ordinary condition-action clauses. After
processing the
.code :begin
clauses,
.code awk
enters a loop in which it extracts successive records
from the input sources according to the
.code rs
(record separator) variable. Each record is divided into fields according
to the
.code fs
(field separator)
variable, and various
.code awk
variables are updated. Then, the condition-action clauses are processed, in the order
in which they appear. Each
.meta condition
is evaluated. If the resulting value is a regular expression
or a function, then this regular expression or function is invoked on the value
stored in the record variable
.codn rec ,
and the result is taken to be the truth value of
.metn condition .
Otherwise, if the resulting value of
.meta condition
is other than a function or regular expression, it is taken directly
to be the truth value.
If the condition is true, then its associated
.meta action
forms are evaluated. Either way, processing passes to the next condition
clause (unless an explicit step is taken in one of the
.metn action s
to prevent this, for instance by invoking the
.code next
and
.code next-file
macros).
When an input source runs out of records,
.code awk
switches to the next input source. When there are no more input sources,
the macro terminates.
.RE
.coNP Variables @ rec and @ orec
.desc
The
.code awk
variable
.code rec
holds the current record. It is automatically updated prior to the
processing of the condition-pattern clauses. Prior to the extraction
of the first record, its value is
.codn nil .
It is possible to assign to
.codn rec .
The value assigned to
.code rec
must be a character string. Immediately upon the assignment, the character
string is delimited into fields according to the field separator
.code awk
variable
.codn fs ,
and these fields are assigned to the field list
.codn f .
At the same time, the
.code nf
variable is updated to reflect the new number of fields.
Likewise, modification of these variables causes
.code rec
to be reconstructed by a catenation of the textual representation
of the fields in
.code f
separated by copies of the output field separator
.codn ofs .
The
.code orec
variable ("original record") also holds the current record. It is automatically
updated prior to the processing of the condition-clauses at the same time as
.code rec
with the same contents. Like
.codn rec ,
it is initially
.code nil
before the first record is read. The
.code orec
variable is unaffected by modification of
the variables
.codn rec ,
.code f
and
.codn nf .
It may be assigned. Doing so has no effect on any other
variable.
.coNP Variable @ f
.desc
The
.code awk
variable
.code f
holds the list of fields. Prior to the first record being read,
its value is
.codn nil .
Whenever a new record is read, it is divided into fields according
to the field separator variable
.codn fs ,
and these fields are stored in
.code f
as a list of character strings.
If the variable
.code f
is assigned, the new value must be a sequence. The variable
.code nf
is automatically updated to reflect the length of this sequence.
Furthermore, the
.code rec
variable is updated by catenating a string representation of the
elements of this sequence, separated by the contents of the
.code ofs
(output field separator)
.code awk
variable.
Note that assigning to a DWIM bracket form which indexes
.codn f ,
such as for instance
.code "[f 0]"
constitutes an implicit modification of
.codn f ,
and triggers the recalculation of
.codn rec .
Modifications of the
.code f
list which do not involve an implicit or explicit assignment to the variable
.code f
itself do not have this recalculating effect.
Unlike in Awk, assigning to the nonexistent field
.mono
.meti [f << m ]
.onom
where
.meta m
>=
.code nf
is erroneous.
.coNP Variable @ nf
.desc
The
.code awk
variable
.code nf
holds the current number of fields in the sequence
.codn f .
Prior to the first record being read, it is initially zero.
If
.code nf
is assigned, then
.code f
is modified to reflect the new number of fields. Fields are deleted from
.code f
if the new value of
.code nf
is smaller. If the new value of
.code nf
is larger, then fields are added. The added fields are empty strings,
which means that
.code f
must be a sequence of a type capable of holding elements which are
strings.
If
.code nf
is assigned, then
.code rec
is also recalculated, in the same way as described in the documentation for the
.code f
variable.
.coNP Variable @ nr
.desc
The
.code awk
variable
.code nr
holds the current absolute record number. Record numbers start at 1.
Absolute means that this value does not reset to 1 when
.code awk
switches to a new input source; it keeps incrementing for each record.
See the
.code fnr
variable.
Prior to the first record being read, the value of
.code nr
is zero.
.coNP Variable @ fnr
.desc
The
.code awk
variable
.code fnr
holds the current record number within the file. The first record is 1.
Prior to the first record being read from the first input source,
the value of
.code fnr
is zero. Thereafter, it resets to 1 for the first record of each input
source and increments for the remaining records of the same input
source.
.coNP Variable @ arg
.desc
The
.code awk
variable
.code arg
is an integer which indicates what input source is being processed.
Prior to input processing, it holds the value zero. When the first
record is extracted from the first input source, it is set to 1.
Thereafter, it is incremented whenever
.code awk
switches to a new input source.
.coNP Variable @ fname
.desc
The
.code awk
variable
.code fname
provides access to a character string which, if the current input is
a file stream, is the name of the underlying file. Assigning to this
variable changes its value, but has no effect on the input stream.
Whenever a new input source is used by
.codn awk ,
this variable is set from the file name on which it is opening
a stream. When using an existing stream rather than opening a file,
.code awk
sets this variable from the
.code :name
property of the stream.
Note that the redirection macros
.code <-
and
.code <!
have no effect on this variable. Within their scope,
.code fname
retains its value.
.coNP Variable @ rs
.desc
The
.code awk
variable
.code rs
specifies a string or regular expression which is used for
delimiting characters read from the inputs into pieces called records.
Note: the record extraction is internally implemented using record streams
instantiated by the
.code record-adapter
function.
The regular-expression pattern stored in
.code rs
is used to matches substrings in the input which separate or terminate records.
Unless the
.code krs
variable is set true, the substrings which match
.code rs
are discarded and the records consist of the nonmatching extents between
them.
The initial value of
.code rs
is
.strn "\en" :
the newline character. This means that, by default, records are lines.
If
.code rs
is changed to the value
.codn nil ,
then record separation operates in
.IR "paragraph mode" ,
which is described below.
If a match for the record separator occurs at the end of the stream,
it is not considered to delimit an empty record, but acts as the
terminator for the previous record.
When a new value is assigned to
.codn rs ,
it has no effect on the most recently scanned and delimited record which is
still current, or previous records. The new value applies to the next, not yet
read record.
In paragraph mode, records are separated by a newline character followed by one
or more blank lines (empty lines or lines containing only a mixture of
tabs and spaces). This means that, effectively, the record-separating
sequences match the regular expression
.codn "/\en[ \en\et]*\en/" .
There are two differences between paragraph mode and simply using the above
regular expression as
.codn rs .
The first difference is that if the first record which is read upon entering
paragraph mode is empty (because the input begins with a match for the
separator regex), then that record is thrown away, and the next record is read.
The second difference is that, if field separation based on the
.code fs
variable is in effect, then regardless of the value of
.codn fs ,
newline characters separate fields. Therefore, the programmer-defined
.code fs
doesn't have to include a match for newline. Moreover, if it is a simple
fixed string, it need not be converted to a regular expression which also
matches a newline.
.coNP Variable @ krs
.desc
The
.code awk
variable
.code krs
stands for "keep record separator". It is a Boolean variable, initialized to
.codn nil .
If it is set to a true value, then the separating text matched
by the pattern in the
.code rs
variable is retained as part of the preceding record rather than removed.
When a new value is assigned to
.codn krs ,
it has no effect on the most recently scanned and delimited record which is
still current, or previous records. The new value applies to the next, not yet
read record.
.coNP Variables @ fs and @ ft
.desc
The
.code awk
variable
.code fs
and
.code ft
each specify a string or regular expression which is used for each
record that is stored in the
.code rec
variable into fields.
Both variables are initialized to
.codn nil ,
in which case a default behavior is in effect, described below.
Use of these variable is mutually exclusive; it is an error for both of these
variables to simultaneously have a value other than
.codn nil .
The value stored in either variable must be
.codn nil ,
a character string or a regular expression. If it contains a string or
regex, it is said to contain a pattern. A string value effectively behaves
as a fixed regular expression which matches the sequence of characters
in the string verbatim, without treating any of them as regex operators.
The splitting of
.code rec
into fields is influenced by the Boolean
.code kfs
("keep field separators")
variable, whose effect is discussed in its description.
If
.code kfs
is false, the splitting is carried out as follows.
If
.code fs
contains a pattern, then
.code rec
is treated specially when it is the empty string: in that case,
the pattern in
.code fs
is ignored, and no fields are produced: the field list
.code f
is the empty list, and
.code nf
is zero. A nonempty record is split by searching it for matches for the
.code fs
pattern. If a match does not occur, then the entire record is a field.
If one match occurs, then the record is split into two fields, either of which,
or both, might be empty. If two matches occur, the record is split into
three fields, and so on. If
.code fs
finds only an empty string match in the record, then it is considered
to match each of the empty strings between two consecutive characters of the
record. Consequently, the record is split into its individual characters, each
one becoming a field. Note: all of these behaviors, except for the special
treatment of the empty record, are accomplished by a call to the
.code split-str
function.
If the variable
.code ft
("field tokenize") contains a pattern, that pattern is used to positively
recognize tokens within the input record, rather than to match separating
material between them. Those matching tokens then constitute the fields.
The tokenizing is performed using the
.code tok-str
function.
If
.code fs
and
.code ft
are both
.codn nil ,
as is initially the case, then the splitting into fields is performed
as if the
.code ft
variable held the regular expression
.codn "/[^\en\et ]+/" .
This means that, by default, fields are sequences of consecutive characters
which are not spaces, tabs or newlines.
Newlines are excluded from fields (and thus separate them) because they can
occur in a record when the value of the record separator
.code rs
is customized.
.coNP Variable @ kfs
.desc
The
.code awk
variable
.code kfs
is a Boolean flag which is initialized to
.codn nil .
If it is set to any other value, it indicates a request to retain
the pieces of the record which separate the fields (even when they are
empty strings). The retained pieces appear as fields, interspersed
among the regular fields so that all of the fields appear in the order
in which they were extracted from the record.
When
.code kfs
is set, and tokenization-style delimiting is in effect due to
.code ft
being set, there is always at least one field, even if the record is empty.
If the record doesn't match the tokenizing regular expression in
.code ft
then a single field is generated, then the entire record is
taken as one field, denoting the nonmatching space, even
if the record is the empty string.
If the record matches one or more tokens, then the first and
last field will always contain the nonmatching material before
the first and last token, respectively. This is true even if
the material is empty. Thus
.code "[f 0]"
always has the material before the first token, whether or not
the first token is matched immediately at the first character
position in the record. This behavior follows from the semantics
of the
.code keep-sep
parameter of the
.code tok-str
function.
Similarly, when splitting based on
.code fs
is in effect and
.code kfs
is set, there is always at least one field, even if the record
is empty. If
.code fs
finds no match in the record, then the entire record,
even if empty, is taken as one field. In that case, there
are no separator to retain. When
.code fs
finds one or more
matches, then these are included as fields. Separators are
always between the fields. If a separator finds a nonempty
match at the beginning of a record, that causes an empty field
to be split off: the separator is understood as intervening
between an empty string before the first character of the
record, and subsequent material which follows the text
matched by the separator. Thus the first field is an empty
field, and the second is the matched text which is
included due to
.code kfs
being set. An analogous situation occurs at the end of the record: if
.code fs
matches a nonempty string at the tail of the record, it splits off an empty
last field, preceded by a field holding the matched separator portion.
Empty matches are only permitted to occur between the characters
of the record, not before the first character of after the last.
If
.code fs
matches the entire record, then there will be three fields:
the first and last of these three will be empty strings,
and the middle field, the separator, will be a copy of the record.
Under
.codn kfs ,
empty matches cause empty string to be included among the
fields. All of this follows from the semantics of the
.code keep-sep
parameter of the
.code split-str
function.
.coNP Variable @ fw
.desc
The
.code awk
variable
.code fw
controls the fixed-width-based delimiting of records into fields.
The variable is initialized to
.codn nil .
In that state, it has no effect.
When this variable holds a
.cod2 non- nil
value, it is expected to be a list of integers.
The use of the
.code fs
or
.code ft
variables is suppressed, and fields are extracted according
to the widths indicated by the list. The fields are consecutive,
such that if the list is
.code "(5 3)"
then the first five characters of the record are identified
as field
.code "[f 0]"
and the next three characters after that as
.codn "[f 1]" .
Only complete fields are extracted from the record. If, after
the extraction of the maximum possible complete fields, more characters
remain, those characters are assigned to an extra field.
An empty record produces an empty list of fields regardless
of the integers stored in fw.
A zero width extracts a zero length field, except when
no more characters remain in the record.
If
.code nil
is stored into
.code fw
then control over field separation is relinquished to the
.code fs
or
.code ft
variables, according to their current values.
If
.code fw
holds a value other than
.code nil
or else a list of nonnegative integers, the behavior is unspecified.
.TP* Examples
The following table shows how various combinations of the
value the input record
.code rec
and field widths in the variable
.code fw
give rise to field values
.codn f :
.verb
rec fw f
---------------------------------
"abc" (0) ("" "abc")
"abc" (2) ("ab" "c")
"abc" (1 2) ("a" "bc")
"abc" (1 3) ("a" "bc")
"abc" (1 1) ("a" "b" "c")
"abc" (3) ("abc")
"abc" (4) ("abc")
"" (4) nil
"" (0) nil
.brev
.coNP Variable @ ofs
.desc
The
.code awk
variable
.code ofs
hold the output field separator. Its initial value is a string
consisting of a single space character.
When the
.code prn
function prints two or more arguments, or fields,
the value of
.code ofs
is used to separate them.
Whenever
.code rec
is implicitly updated due to a change in the variable
.code f
or
.codn nf ,
.code ofs
is used to separate the fields, as they appear in
.codn rec .
.coNP Variable @ ors
.desc
The
.code awk
variable
.codn ors ,
though it stands for "output record separator" holds what
is in fact the output record terminator. It is named after the
.code ORS
variable in Awk.
Each call to the
.code prn
function terminates its output by emitting the value of
.codn ors .
The initial value of
.code ors
is a character string consisting of a single newline,
and so the
.code prn
function prints lines.
.coNP Function @ prn
.synb
.mets (prn << form *)
.syne
.desc
The
.code awk
function
.code prn
performs output into the
.code *stdout*
stream. The
.code :output
clause affects the destination by rebinding
.codn *stdout* .
If called with no arguments,
.code prn
prints
.code rec
followed by
.codn ors .
Otherwise, it prints the values of the arguments, separated by
.codn ofs ,
followed by
.codn ors .
When a condition-action clause specifies no action forms,
then a call to
.code prn
with no arguments is the default action.
Each argument
.meta form
is printed by conversion to a string, as if by the expression
.code `@val`
where
.code val
is some variable which holds the value produced by the
evaluation of
.metn form .
Thus if the value is
.codn nil ,
the output for that argument is an empty string, rather than the text
.strn nil .
.coNP Macro @ next
.synb
.mets (next)
.syne
.desc
The
.code awk
macro
.code next
may be invoked in a condition-pattern clause. It terminates
the processing of that clause, and all subsequent clauses,
causing
.code awk
to process the next record, if there is one. If there is no next
record,
.code awk
terminates.
.coNP Macro @ again
.synb
.mets (again)
.syne
.desc
The
.code awk
macro
.code again
may be invoked in a condition-pattern clause. It terminates the
processing of that clause, and all subsequent clauses.
Then, the current value of the record, namely the datum stored
in the
.code awk
variable
.codn rec ,
is delimited into fields, and all of the condition-pattern clauses
are processed again.
No other state is modified. In particular, the record number
.code nr
and the
.code orec
variable holding the original record both retain their current values.
Note: this is an original feature in the \*(TL
.code awk
macro, which has no counterpart in POSIX or GNU Awk.
.coNP Macro @ next-file
.synb
.mets (next-file)
.syne
.desc
The
.code awk
macro
.code next-file
may be invoked in a condition-pattern clause. It terminates
the processing of that clause and all subsequent clauses.
Then
.code awk
abandons the current input source and moves to the next one.
If there is no next input source,
.code awk
terminates.
.coNP Macros @, rng @, -rng @, rng- @, -rng- @, --rng @, --rng- @, rng+ @ -rng+ and @ --rng+
.synb
.mets (rng < from-condition << to-condition )
.mets (-rng < from-condition << to-condition )
.mets (rng- < from-condition << to-condition )
.mets (-rng- < from-condition << to-condition )
.mets (--rng < from-condition << to-condition )
.mets (--rng- < from-condition << to-condition )
.mets (rng+ < from-condition << to-condition )
.mets (-rng+ < from-condition << to-condition )
.mets (--rng+ < from-condition << to-condition )
.syne
.desc
The nine
.code awk
macros in the
.code rng
family may be used anywhere within an ordinary condition-pattern
.code awk
clause.
Each provides a Boolean test which is true if the current record lands within
a range of records delimited by conditions. Each provides its own
distinct, useful nuance, which is identified by the mnemonic characters
prefixed or suffixed to the name.
The basic
.code rng
macro inclusively matches ranges of records. Each such range begins with a record
for which
.meta from-condition
yields true, and ends on the record for which
.meta to-condition
is true. What it means to match is that the
.code rng
expression yields a Boolean true value when it is evaluated in the context
of processing any of the records which are included in the range.
The table below summarizes the semantic variations of these nine
range macro operators. The leftmost column labeled
.code DATA
represents the stream of records being processed. Each entry in this column gives
the literal piece of text which comprises the content of one record in the stream.
The remaining nine columns, labeled with the nine range operators, inform about
the behavior of these operators with respect to these records. In each of these
columns the letter
.code X
marks those records for which the column's range operator yields true,
if it is invoked with the arguments
.code #/H/
and
.code #/T/
as its
.meta from-condition
and
.metn to-condition ,
respectively.
For example, the
.code rng
column the values of the
.code "(rng #/H/ #/T/)"
expression, indicating that the expression starts being true when the
.code H1
record is seen, stays true for the
.code T1
record, and then reverts to false:
.verb
DATA rng -rng rng- -rng- --rng --rng- rng+ -rng+ --rng+
----------------------------------------------------------
PROLOG
H1 X X X
H2 X X X X X X
H3 X X X X X X
B1 X X X X X X X X X
B2 X X X X X X X X
T1 X X X X X X
T2 X X X
T3 X X X
EPILOG
.brev
The prefix and suffix characters of the operator names are intended
to be mnemonic. A single
.code -
(dash) indicates the exclusion of one record. A double
.code --
(dash dash)
indicates the exclusion of all leading records which match
.metn from-condition ;
this appears on the left side only.
The
.code +
character, appearing on the right only, indicates that
all consecutive records which match
.meta to-condition
are included in the range, not only the first one.
Ranges are oblivious to the division between successive sources of input; a
range can start in one file of records and terminate in another.
To prevent a range from spanning input transitions, additional complexity
is required in the expression.
Ranges expressed using the
.code rng
family macros may combine with other expressions, including
other ranges, and allow arbitrary nesting: the
.meta from-condition
or
.meta to-condition
can be a range, or an expression containing ranges.
The expressions
.meta from-condition
and
.meta to-condition
are ordinary expressions which are evaluated. However, their
evaluation is unusual in two ways.
Firstly, if either expression
produces, as its result, a function or regular-expression object,
then that function or regular-expression object is applied to
the current record (value of the
.code rec
variable), and the result of that application is then taken
as the result of the condition. This allows for expressions like
.code "(rng (f^ #/start/) #/end/)"
which denotes a range which begins with a record which
begins with the prefix
.str start
and ends with a record which contains
.str end
as a substring.
Secondly, the conditions are evaluated
out of order with respect to the surrounding expression
in which they occur. Ranges and their constituent
.meta from-condition
and
.meta to-condition
are evaluated just prior to the processing of the condition-action clauses.
Each
.code rng
expression is reduced to a Boolean value.
Then, when the condition-action clauses are processed and their
.meta condition
and
.meta action
forms are evaluated, each occurrence of a
.code rng
expression simply denotes its previously evaluated Boolean value.
Therefore, it is not possible for expressions to short circuit
the evaluation of ranges. Ranges cannot "miss" their starting or
terminating conditions; every range occurring anywhere in the condition-action
clauses is tested against every record that is processed.
Because of this perturbed evaluation order, code which happens to place side
effects into ranges may produce surprising results.
For instance, the expression
.code "(if nil (rng (prinl 'hello) (prinl 'world)))"
will produce output even though the
.code if
condition is
.codn nil ,
and, moreover, this output will happen before the clauses are processed in
which this
.code if
expression appears. At the time when the
.code if
itself is evaluated, the
.code rng
expression merely fetches a previously computed Boolean value which indicates
whether the range is active for this record.
Also, the behavior is unspecified if range expressions attempt to modify
any of the special
.code awk
variables
.codn rec ,
.codn f ,
.codn fs ,
.code ft
and
.codn kfs .
It is not recommended to place any side effects into range expressions.
A more detailed description of the range operators follows.
.RS
.meIP (rng < from << to )
This type of range becomes active when a record is encountered for which the
.meta from
expression yields true. While the range is active, the expression evaluates
true. If, when the range is active, a record is encountered for which the
.meta to
expression yields true, the range remains active for that record and is
deactivated after the completion of processing for that record. If
the range is inactive and a record is encountered or which both
.meta from
and
.meta to
are true, then the range is activated for that record and then deactivated
when that record is processed.
Records for which
.meta from
and
.meta to
are not true do not affect the range's activation state.
.meIP (-rng < from << to )
This type of range is active under the same conditions as the
.code rng
type. However, the expression yields a Boolean false value for the
first record which begins a range. That is to say, when the range is
inactive, and a record is scanned for which
.meta from
is true, the range activates, but the range expression yields
.codn nil .
This is true regardless of whether the
.meta to
expression yields true for that record. If there are additional records
in the range, the expression yields a true value for those records.
.meIP (rng- < from << to )
This type of range is active under the same conditions as the
.code rng
type. However, the range expression yields
.code nil
for the record for which
.code to
yields true which terminates the range. This occurs even if that is
the same record which activated the range by triggering the
.meta from
condition. Note that if a range terminates abruptly due to no more records
being available, the range expression still yields true for the last record.
.meIP (-rng- < from << to )
This type of range is active under the same conditions as the
.code rng
type. However, the range expression yields
.code nil
for the first record which activates the range, and for the last
record which deactivates the range by activating the
.code to
condition. If the range is active over fewer than three records, then
the expression never yields true for that range. If the range terminates
abruptly due to no more records being available, and if the last record
processed isn't the one which activated the range due to triggering the
.code from
condition, the expression yields true for that record.
.meIP (--rng < from << to )
This type of range is active under the same conditions as
.codn rng .
However, the range expression yields
.code nil
for the entire leading sequence of consecutive records for which
.meta from
is true. If
.meta from
is true of the
.meta to
record which terminates the range,
.code nil
is returned for that record also.
.meIP (--rng- < from << to )
This type of range is active under the same conditions as
.codn rng .
However, the range expression yields
.code nil
for the entire leading sequence of consecutive records for which
.meta from
is true, and also yields nil for the last record which triggers the
.meta to
condition.
.meIP (rng+ < from << to )
This range is active under different conditions compared to
.codn rng .
Though it becomes active in the same way, when the
.meta from
expression yields true, the deactivation logic is different.
The range is deactivated when a record for which
.meta to
is true is followed by a record for which
.meta to
is not true. That record is excluded from the range; if the
.meta from
expression happens to be true for that record, a new range begins
at that record. Thus, effectively, the range is terminated not
by single record which triggers
.meta to
but by a sequence of one or more such consecutive records.
.meIP (-rng+ < from << to )
This range is active under the same conditions as
.codn rng+ .
However, the range expression yields
.code nil
for the first record in the range. If the range contains only one record, then
it returns
.code nil
for that record.
.meIP (--rng+ < from << to )
This range is active under the same conditions as
.codn rng+ .
However, the range expression yields
.code nil
for the entire leading sequence of consecutive records for which
.meta from
is true. This is the case even for those for which the
.meta to
expression is true.
.RE
.coNP Macro @ ff
.synb
.mets (ff < opip-arg *)
.syne
.desc
The
.code awk
macro
.code ff
(filter fields)
provides a shorthand for filtering the field list
.code f
trough a pipeline of chained functions expressed using
.code opip
argument syntax.
The following equivalence holds, except that
.code f
refers to the
.code awk
variable even if the
.code ff
invocation occurs in code which establishes
a binding which shadows
.codn f .
.verb
(ff a b c ...) <--> (set f [(opip a b c ...) f])
.brev
.TP* Example:
.verb
;; convert all fields from string to floating-point
(ff (mapcar flo-str))
.brev
.coNP Macro @ mf
.synb
.mets (mf < opip-arg *)
.syne
.desc
The
.code awk
macro
.code mf
(map fields)
provides a shorthand for mapping each field
individually trough a pipeline of chained functions expressed using
.code opip
argument syntax.
The following equivalence holds, except that
.code f
refers to the
.code awk
variable even if the
.code mf
invocation occurs in code which establishes
a binding which shadows
.codn f .
.verb
(mf a b c ...) <--> (set f (mapcar (opip a b c ...) f))
.brev
.TP* Example:
.verb
;; convert all fields from string to floating-point
(mf flo-str)
.brev
.coNP Macro @ fconv
.synb
.mets (fconv >> { clause | : | - }*)
.syne
.desc
The
.code awk
macro
.code fconv
provides a succinct way to request conversions of the textual fields.
Conversions are expressed by clauses which correspond with fields.
Each
.meta clause
is an expression which must evaluate to a function. The clause is evaluated
in the same manner as the argument a
.code dwim
operator, using Lisp-1-style name lookup. Thus, functions may be
specified simply by using their name as a
.metn clause .
Furthermore, several local functions exist in the scope of each
.metn clause ,
providing a shorthand notation. These are described below.
Conversion proceeds by applying the function produced by
a clause to the field to which that clause corresponds, positionally.
The return value of the function applied to the field replaces
the field.
When a clause is specified as the symbol
.code -
(minus)
it has a special meaning: this minus clause occupies a field
position and corresponds to a field, but performs no conversion
on its field.
The
.code :
(colon)
keyword symbol isn't a clause and does not correspond to a field position.
Rather, it acts as a separator among clauses. It need not appear at
all. If it appears, it may appear at most twice. Thus, the
clauses may be separated into up to three sequences.
If the colon does not appear, then all the clauses are
.IR "prefix clauses" .
Prefix clauses line up with fields from left to right. If there are fewer
fields than prefix clauses, the values of the excess clauses are evaluated, but
ignored.
Vice versa, if there are fewer prefix clauses than fields, then the excess
fields are not subject to conversions.
If the colon appears once, then the clauses before the colon, if any, are
prefix clauses, as described in the previous paragraph. Clauses after the
colon, if any, are
.IR "interior clauses" .
Interior clauses apply to any fields which are left unconverted by the prefix
clauses. All interior clauses are evaluated. If there are fewer fields than
interior clauses, then the values of the excess interior clauses are ignored.
If there are more fields than clauses, then the clause values are cycled:
reused from the beginning against the excess fields, enough times to convert
all the fields.
If the colon appears twice, then the clauses before the first colon, if any,
are prefix clauses, the clauses between the two colons are interior clauses,
and those after the second colon are
.IR "suffix clauses" .
The presence of suffix clauses change the behavior relative to the one-colon
case as follows. After the conversions are performed according to the prefix
clauses, the remaining fields are counted. If there are are only as many
fields as there are suffix clauses, or fewer, then the interior clauses are
evaluated, but ignored. The remaining fields are processed against the suffix
clauses. If after processing the prefix clauses there are more fields
remaining than suffix clauses, then a number of rightmost fields equal to the
number of suffix clauses is reserved for those clauses. The interior fields
are applied only to the unreserved middle fields which precede these reserved
rightmost fields, using the same repeating behavior as in the one-colon case.
Finally, the previously reserved rightmost fields are processed using
the suffix clauses.
The following special convenience functions are in scope of the clauses,
effectively providing a shorthand for commonly-needed conversions:
.RS
.coIP i
Provides conversion to integer. It is identical to the
.code toint
function, with the default radix.
.coIP o
Converts a string value holding an octal representation
to the integer which it denotes. It is equivalent to
.code toint
with a
.meta radix
argument of 8.
.coIP x
Converts a string value holding a hexadecimal representation
to the integer which it denotes. It is equivalent to
.code toint
is equivalent to
with a
.meta radix
argument of 16.
.coIP b
Converts a string value holding a binary (base two) representation
to the integer which it denotes. It is equivalent to
.code toint
with a
.meta radix
argument of 2.
.coIP c
Converts a string value holding a C-language-style representation
to the integer which it denotes, meaning that the
.code 0x
prefix denotes a hexadecimal value, a leading zero octal, otherwise
decimal. These prefixes follow the
.code +
or
.code -
sign, if present.
The
.code c
function is equivalent to
.code toint
invoked with a
.meta radix
argument of
.codn #\ec .
.coIP r
Converts a string holding a floating-point representation to
the floating-point value which it denotes. It is equivalent to
.codn tofloat .
.ccIP @, iz @, oz @, xz @, bz @ cz and @ rz
Conversion similar to
.codn i ,
.codn o ,
.codn x ,
.codn b ,
.code c
and
.codn r ,
but equivalent to using the functions
.code tointz
and
.codn tofloatz .
Thus fields which are non-numeric strings or the object
.code nil
get converted to 0, or 0.0 in the case of
.codn rz .
.coIP -
Performs no conversion: the corresponding field is taken as-is.
.RE
.IP
Because
.code fconv
macro destructively operates on the elements of the field list
.codn f ,
it has the same effect as an assignment to the fields:
the value of
.code rec
is updated.
The return value of
.code fconv
is
.codn f .
Note: because
.code f
is
.code nil
when no fields have been extracted, a
.code fconv
expression can be used as the condition in an
.code awk
clause which triggers the action if one or more fields have been
extracted, and performs conversions on them.
Note: although
.code fconv
is intended for converting textual fields, and the semantic descriptions below
consequently make references to string inputs, the behavior of
.code fconv
with respect to non-string fields can be inferred. For instance if a field
actually holds the floating-point value 3.14, and the
.code i
conversion is applied to it, it will produce 3, because it works by
means of the
.code toint
function.
Note: a somewhat less flexible mechanism for converting fields, related to
.codn fconv ,
is present in the
.code :fields
clause of the
.code awk
macro, which can specify names for the positional fields, along with
conversion functions. The
.code :fields
clause has different syntax, and doesn't support the
.code :
(colon) separator, instead assuming a fixed number of fields
enumerated from the left.
.TP* Examples:
.verb
;; convert up to first three fields to integer:
(awk ((fconv i i i)))
;; convert all fields to floating-point
(awk ((fconv : r :)))
;; convert first and second fields to integer
;; from hexadecimal;
;; convert last field to integer from octal;
;; process pairs of fields in between
;; these by leaving the first element of
;; each pair unconverted and converting second
;; to floating-point;
(awk ((fconv x x : - r : o)))
;; convert all fields, except the first,
;; from integer, turning empty strings
;; and non-integer junk as zero;
;; leave first field unconverted:
(awk ((fconv - : iz)))
.brev
.coNP Macros @, -> @, ->> @, <- @ !> and @ <!
.synb
.mets (-> < path << form *)
.mets (->> < path << form *)
.mets (<- < path << form *)
.mets (!> < command << form *)
.mets (<! < command << form *)
.syne
.desc
These
.code awk
macros provide convenient redirection of output and input to and from
files and commands.
When at least one
.meta form
argument is present, the functions
.codn -> ,
.code ->>
and
.code !>
evaluate each
.meta form
in a dynamic environment in which the
.code *stdout*
variable is bound to a file output stream, for the first two
functions, or output command pipe in the case of the last one.
Similarly, when at least
.meta form
argument is present, the remaining functions
.code <-
and
.code <!
evaluate each
.meta form
in a dynamic environment in which
.code *stdin*
is bound to a file input stream or input command pipe, respectively.
The
.meta path
and
.meta command
arguments are treated as forms, and evaluated.
They should evaluate to strings.
The first evaluation of one of these macros for a given
.meta path
or
.meta command
being used in a particular direction (input or output) and type (file or
command) creates a stream. That stream is then associated with the given
.meta path
or
.meta command
string, together with the direction and type. Upon a subsequent evaluation
of one of these macros for the same
.meta path
or
.meta command
string, direction and type, a new stream is not opened; rather, the
previously associated stream is used.
The
.code ->
macro indicates that the file named
.meta path
is to be opened for writing and overwritten, or created if it doesn't exist.
The
.code ->>
macro indicates that the file named by
.meta path
is to be opened in append mode, created if necessary.
The
.code <-
macro indicates that the file given by
.meta path
is to be opened for reading.
The
.code !>
macro indicates that
.meta command
is to be opened as an output command pipe. The
.code <!
macro indicates that
.meta command
is to be opened as an input command pipe.
If any of these macros is invoked without any
.meta form
arguments, then it yields the stream object associated with
.meta path
or
.meta command
argument, direction and type. If the association doesn't exist,
the stream is first created.
If
.meta form
arguments are present, then the value of the last one is yielded
as a value, except in the case when the last form yields the
.code :close
keyword symbol.
If the last
.meta form
yields the
.code :close
keyword symbol, the association between the
.meta path
or
.metn command ,
direction and type and the stream is removed, and the stream
is closed. In this case, the result value of the macro isn't the
.code :close
symbol, but rather the return value of the
.meta close-stream
call that is implicitly applied to the stream.
Even if there is only one
.meta form
which yields
.codn :close ,
the stream is created, if it doesn't exist prior to the macro
invocation.
In each invocation of these macros, after every
.meta form
is evaluated, the stream is implicitly flushed, if it is an output stream.
The association between the
.meta pipe
or
.meta command
strings, direction and type is scoped to the innermost enclosing
.code awk
macro. An inner
.code awk
macro cannot refer to the associations established in an outer
.code awk
macro. An outer
.code awk
macro can obtain an association's stream object and communicate
that stream to the nested macro where it can be used.
When the surrounding
.code awk
macro terminates, all of the streams opened by these
redirection macros are closed, without breaking those associations.
If lexical closures are captured inside the macro, and then invoked after the
macro has terminated, and inside those closures the redirection macros are
used, those macro instances will with closed stream objects, and so
attempts to perform I/O will fail.
.coNP Examples of @ awk Macro Usage
The following examples are
.code awk
macro equivalents of the examples of the POSIX
.code awk
utility given in IEEE Std 1003.1, 2013 Edition.
.RS
.IP 1.
Print lines for which field 3 is greater than 5:
.verb
;; print lines with fields separated by ofs,
;; and [f 2] converted to integer:
(awk ((and [f 2] (fconv - - iz) (> [f 2] 5))))
;; print strictly original lines from orec
(awk ((and [f 2] (fconv - - iz) (> [f 2] 5))
(prn orec)))
.brev
.IP 2.
Print every tenth line:
.verb
(awk ((zerop (mod nr 10))))
.brev
.IP 3.
Print any line with a substring matching a regex:
.verb
(awk (#/(G|D)(2\ed[\ew]*)/))
.brev
Note the subtle flaw here: the
.code [\ew]*
portion of the regular expression contributes nothing
to what lines are matched. The following example
has a similar flaw.
.IP 4.
Print any line with a substring beginning with a
.code G
or
.code D
followed by a sequence of digits and characters:
.verb
(awk (#/(G|D)([\ed\ew]*)/))
.brev
.IP 5.
Print lines where the second field matches a regex,
while the fourth one doesn't:
.verb
(awk (:let (r #/xyz/))
((and [f 3] [r [f 1]] (not [r [f 3]]))))
.brev
.IP 6.
Print lines containing a backslash in the second field:
.verb
(awk ((find #\e\e [f 1])))
.brev
.IP 7.
Print lines containing a backslash using a regex constructed
from a string. Note that backslash escapes are interpreted
twice: once in the string literal, and once in the parsing
of the regex, requiring four backslashes to encode one:
.verb
(awk (:let (r (regex-compile "\e\e\e\e")))
((and [f 1] [r [f 1]])))
.brev
.IP 8.
Print penultimate and ultimate field in each record,
separating then by a colon:
.verb
;; original: {OFS=":";print $(NF-1), $NF}
;;
(awk (t (set ofs ":") (prn [f -2] [f -1])))
.brev
.IP
Note that the above behaves
more correctly than the original Awk example because in the
when there is only one field,
.code $(NF-1)
reduces to
.code $0
which refers to the entire record, not to the field.
This sort of bug is why the \*(TL
.code awk
does not imitate the design decision to make the record
the first numbered field.
.IP 9.
Output the line number and number of fields separated by colon,
by producing a single string first:
.verb
(awk (t (prn `@nr:@nf`)))
.brev
.IP 10.
Print lines longer than 72 characters:
.verb
(awk ((> (len rec) 72)))
.brev
.IP 11.
Print first two fields in reverse order, separated by
.codn ofs :
.verb
(awk (t (prn [f 1] [f 0])))
.brev
.IP 12.
Same as 11, but with field separation consisting of a
comma, or spaces and tabs, or both in sequence:
.verb
(awk (:set fs #/,[ \et]*|[ \et]+/)
(t (prn [f 1] [f 0])))
.brev
.IP 13.
Add the values in the first column, then print sum and
average:
.verb
;; original:
;; {s += $1}
;; END {print "sum is ", s, " average is", s/NR}
;;
(awk (:let (s 0) (n 0))
([f 0] (fconv r) (inc s [f 0]) (inc n))
(:end (prn `sum is @s average is @(/ s n)`)))
.brev
Note that the original is not robust against blank lines
in the input. Blank lines are treated as if they had a
first column field of zero, and are counted toward the
denominator in the calculation of the average.
.IP 14.
Print fields in reverse order, one per line:
.verb
(awk (t (tprint (reverse f))))
.brev
.IP 15.
Print all lines between occurrences of
.code start
and
.codn stop :
.verb
(awk ((rng #/start/ #/stop/)))
.brev
.IP 16.
Print lines whose first field is different from
the corresponding field in the previous line:
.verb
(awk (:let prev)
((nequal [f 0] prev) (prn) (set prev [f 0])))
.brev
.IP 17.
Simulate the
.code echo
utility:
.verb
(awk (:begin (prn `@{*args* " "}`)))
.brev
Note: if this is evaluated in the command line, for instance with the
.code -e
option, an explicit exit is required to prevent the arguments from being
processed by
\*(TX after
.code awk
completes:
.verb
(awk (:begin (prn `@{*args* " "}`) (exit 0)))
.brev
.IP 18.
Print the components of the
.code PATH
environment variable, one per line:
.verb
;; Process variable as if it were a file:
(awk (:inputs (make-string-input-stream
(getenv "PATH")))
(:set fs ":")
(t (tprint f)))
;; Just get, split and print; awk macro is irrelevant
(awk (:begin (tprint (split-str (getenv "PATH") ":"))))
.brev
.IP 19.
Given a file called
.code input
which contains page headers of the format
.str "Page #"
and a \*(TL file called
.code prog.tl
which contains:
.verb
(awk (:let (n (toint n)))
(#/Page/ (set [f 1] (pinc n)))
(t))
.brev
the command line:
.verb
txr -Dn=5 prog.tl input
.brev
prints the file, filling in page numbers starting at 5.
.RE
.SS* Environment Variables and Command Line
Note that environment variable names, their values, and command-line
arguments are all regarded as being externally encoded in UTF-8. \*(TX performs
the encoding and decoding automatically.
.coNP Special variables @, *args-full* @ *args-eff* and @ *args*
.desc
The
.code *args-full*
variable holds the original, complete list of arguments passed
from the operating system, including the program executable
name.
During command-line-option processing, \*(TX may transform the
argument list. The hash-bang mechanism, and the
.code --args
and
.code --eargs
options can inject new command-line arguments, as can code
which is executed during argument processing via the
.code -e
options and others.
The
.code *args-eff*
variable holds the list of
.I "effective arguments" ,
which is the argument list after these transformations are applied.
This variable is established and set to the same value as
.code *args-full*
prior to command-line processing, but is not updated with its final
value until after command-line processing.
The
.code *args*
variable holds a list of strings representing the remaining
arguments which follow any options processed by the \*(TX executable,
and the script name. This list is a suffix of
.codn *args-eff* .
Thus, the arguments before
.code *args*
can be calculated using the expression
.codn "(ldiff *args-eff* *args*)" .
The
.code *args*
variable is available to \*(TL expressions invoked from the
command line via the
.codn -p ,
.code -e
and other such options. During these evaluations,
.code *args*
holds all the remaining options, after the invoking option and its
argument expression. In other words, code executed from the command line
has access to the remaining arguments which follow it.
Furthermore, this code may modify the value of
.codn *args* .
Such a modification is visible to the option processing code.
That is to say code executed from the command line can rewrite the remaining
list of arguments, and that list takes effect.
.coNP Function @ env
.synb
.mets (env)
.syne
.desc
The
.code env
function retrieves the list of environment variables. Each
variable is represented by a single entry in the list: a string which
contains an
.code =
(equal) character somewhere, separating the variable name
from its value.
Multiple calls to
.code env
may return the same list, or lists which share structure.
If a list returned by
.code env
is modified, the behavior is unspecified.
See also: the
.code env-hash
function.
.coNP Function @ env-hash
.synb
.mets (env-hash)
.syne
.desc
The
.code env-hash
function returns an
.code :equal-based
hash whose keys and values are strings. The hash table is populated
with the environment variables, represented as key-value character string
pairs.
The
.code env-hash
function allocates the hash table when it is first invoked; thereafter,
it returns the same hash table.
The hash table is updated by the functions
.codn setenv ,
.code unsetenv
and
.codn getenv .
Note: calls to the underlying C library functions
.code setenv
and
.codn getenv ,
and other direct manipulations of the environment, will not update the hash table.
.coNP Functions @, getenv @ setenv and @ unsetenv
.synb
.mets (getenv << name )
.mets (setenv < name < value <> [ overwrite-p ])
.mets (unsetenv << name )
.syne
.desc
These functions provide access to, as well as manipulation of, environment
variables. Of these three,
.code setenv
and
.code unsetenv
might not be available on some platforms, or
.code unsetenv
might be be present in a simulated form which sets the variable
.meta name
to the empty string rather than deleting it.
The
.code getenv
function searches the environment for the environment variable whose name
is
.metn name .
If the variable is found, its value is returned. Otherwise
.code nil
is returned.
The
.code setenv
function creates or modifies the environment variable indicated by
.metn name .
The
.meta value
string argument specifies the new value for the variable.
If
.meta value
is
.codn nil ,
then
.code setenv
behaves like
.codn unsetenv ,
except that it observes the
.meta overwrite-p
argument. That is to say, the meaning of a null
.meta value
is that the variable is to be removed.
If the
.meta overwrite-p
argument is specified, and is true,
then the variable is overwritten if it already exists.
If the argument is false, then the variable is not modified if it
already exists. If the argument is not specified, it defaults
to the value
.metn t ,
effectively giving rise to a two-argument form of
.code setenv
which creates or overwrites environment variables.
A variable removal is deemed to be an overwrite.
Thus if both
.meta value
and
.meta overwrite-p
are
.codn nil ,
then
.code setenv
does nothing.
The
.code setenv
function unconditionally returns
.meta value
regardless of whether or not it overwrites or removes an existing variable.
The
.code unsetenv
function removes the environment variable
specified by
.metn name ,
if it exists. On some platforms, it instead sets the environment variable
to the empty string.
Note: supporting removal semantics in
.code setenv
allows for the following simple save/modify/restore pattern:
.verb
(let* ((old-val (getenv "SOME-VAR")))
(unwind-protect
(progn (setenv "SOME-VAR" new-val)
...)
(setenv "SOME-VAR" old-val)))
.brev
This works in the case when
.code SOME-VAR
exists, as well as in the case that it doesn't exist.
In both cases, its previous value or, respectively, non-existence,
is restored by the
.code unwind-protect
cleanup form.
These functions interact with the list returned by the
.code env
function and with the hash table returned by the
.code env-hash
function as follows.
A previously returned list returned by
.code env
is not modified. The
.code setenv
and
.code unsetenv
functions may cause a subsequent call to
.code env
to return a different list. The
.code getenv
function has no effect on the list.
The hash table previously returned by
.code env-hash
is modified by
.code setenv
in the manner consistent with its semantics. A new entry is created in the table,
if required, and an existing entry is overwritten only if the
.code overwrite-p
flag is specified. Likewise, if
.code setenv
is invoked in a way that causes the environment variable to be deleted, it
is removed from the hash also.
The
.code unsetenv
function causes the variable to be removed from the hash table also.
The
.code getenv
function accesses the underlying environment and updates the hash
table with the name-value pair which is retrieved.
.coNP Function @ replace-env
.synb
.mets (replace-env << env-list )
.syne
.desc
The
.code replace-env
function replaces the environment with the environment variables specified in
.metn env-list .
The argument is a list of character strings, in the same format
as the list returned by the
.code env
function: each element of the list describes an environment variable
as a single character string in which the name is separated by the
value by the
.code =
character. As a special concession, if this character is missing, the
.code replace-env
function treats that entry as being a name with an empty value.
The
.code replace-env
first empties the existing environment, rendering it devoid of environment
variables. Then it installs the entries specified in
.metn env-list .
The return value is
.metn env-list .
Note:
.code replace-env
may be used to specify an exact environment to child programs executed
by functions like
.codn open-process ,
.code sh
or
.codn run .
Note: the previous environment may be saved by calling
.code env
and retaining the returned list. Then after modifying the environment,
the original environment can be restored by passing that retained
list to
.codn replace-env .
.coNP Special variable @ *child-env*
.desc
The
.code *child-env*
variable specifies the list of command-line variables established for programs
executed via the functions
.codn exec ,
.codn run ,
.codn sh ,
.code open-command
and
.codn open-process .
The initial top-level value of this variable is the symbol
.code t
which indicates that
.code *child-env*
is to be ignored, such that the executed program
inherits the current set of environment variables.
If
.code *child-env*
has any other value, it must be a possibly empty list of environment
variables, in the same format as what is returned by
.code env
function and accepted by
.codn replace-env .
That value completely specifies the environment that executed programs
shall receive.
.TP* Example:
.verb
(let ((*child-env* '("a=b")))
;; /usr/bin/env sees only "a" environment variable
(get-lines (open-process "/usr/bin/env" "r")))
-> ("a=b")
.brev
.SS* Command-Line-Option Processing
\*(TL provides a support for recognizing, extracting and validating
the POSIX-style options from a list of command-line arguments.
The supported options can be defined as a list of option descriptor
objects each of which is constructed by a call to the
.code opt
function. Each option can have a long name, a short name,
a type, and a description.
The
.code getopts
function takes a list of option descriptors, and a list of arguments,
producing a parse, or else throwing an exception of type
.code opt-error
if an error is detected. The returned object, an instance of struct type
.codn opts ,
can then be queried for specific option values, or for the remaining non-option
arguments.
The
.code opthelp
function takes a list of option descriptors and an output stream,
and generates help text on that stream. A program supporting a
.code --help
option can use this to generate that portion of its help text which
describes the available options. Also provided are functions
.code opthelp-conventions
and
.codn opthelp-types ,
which have the same interface as
.code opthelp
and print additional information. These may be used together with
.code opthelp
to provide more detailed help under a single
.code --help
option, or under separate options like
.codn --extra-help .
The
.code define-option-struct
macro provides a more streamlined, declarative mechanism built on the
same facility. The options are declared in a more condensed way, and
using symbols instead of strings. Furthermore, the parsed option values
become slot values of an object, named by the same symbols.
.NP* Command-Line-Option Conventions
A command-line option can have a short or long name. A short name is always
one character long, and treated specially in the command-line syntax. Long
options have names two or more characters long. An option can have both a long
and short name. Options may not begin with the
.code -
(ASCII dash) character. A long option may not contain the
.code =
character.
Short options are invoked by specifying an argument with a single leading
.code -
followed by the option character. Multiple short options which take
no argument can be "clumped": combined into a single argument consisting of
a single
.code -
followed by multiple short option characters.
An option can take an argument, in which case the argument is required.
An option which takes no argument is Boolean, and a Boolean option
never takes an argument: "takes no argument" and "Boolean" effectively
mean the same thing.
Long options are invoked as an argument which begins with a
.code --
(double dash)
immediately followed by the name. When a long option takes an argument,
it is mandatory. It must be specified in the same argument, separated
from the name by the
.code =
character. If that is omitted, then the next command-line argument
is taken as the argument. That argument is removed, and not recognized as
an option, even if it looks like one.
A Boolean long option can be explicitly specified as false using the
.code --no-
prefix rather than the
.code --
prefix.
Short options may be invoked using long name syntax; if
.code a
is a short option, then it may be referenced on the command line as
.code --a
and treated as a long option in all other ways, including the use
of
.code --no-
to explicitly specify false for a Boolean option.
If a short option takes an argument, it may not clump with other
short option. The following command-line argument is taken as the
options argument. That argument is removed and is not recognized as
an option even if it looks like one.
If the command-line argument
.code --
occurs in the command line where an option would otherwise be recognized,
it signifies the end of the options. The subsequent arguments are the
non-option arguments, even if they resemble options.
.NP* Command-Line Processing Examples
The following example illustrates a complete \*(TL program which
parses command-line options:
.verb
(defvarl options
(list (opt "v" "verbose" :dec
"Verbosity level. Higher values produce more chatter.")
(opt nil "help" :bool
"List this help text.")
(opt "x" nil :hex
"The X factor: a number with a mysterious\e \e
interpretation, affecting the program\e \e
behavior in strange ways.")
(opt "z" nil) ;; undocumented option
(opt nil "cee" :cint
"C style integer.")
(opt "g" "gravity" :float
"Gravitational constant. This gives\e \e
the gravitational field\e \e
strength at the Earth's surface.")
(opt "l" "lit" :str
"A character string given in TXR Lisp notation.")
(opt "c" nil 'upcase-str
"Custom treatment: ARG is converted to uppercase.")
(opt "b" "bool" :bool
"A flag you can flip true.")))
(defvarl prog-name *load-path*)
(let ((o (getopts options *args*)))
(when [o "help"]
(put-line "Usage:\en")
(put-line ` @{prog-name} [options] arg*`)
(opthelp options)
(exit 0))
(put-line `args after opts are: @{o.out-args ", "}`))
.brev
The next example is equivalent to the previous, but using the
.code define-option-struct
macro:
.verb
(define-option-struct prog-opts nil
(v verbose :dec
"Verbosity level. Higher values produce more chatter.")
(nil help :bool
"List this help text.")
(nil extra-help
:bool
"List help text with more detailed information.")
(x nil :hex
"The X factor: a number with a mysterious\e \e
interpretation, affecting the program\e \e
behavior in strange ways.")
;; undocumented Boolean:
(z nil)
(nil cee :cint
"C style integer.")
(g gravity :float
"Gravitational constant. This gives\e \e
the gravitational field\e \e
strength at the Earth's surface.")
(l lit :str
"A character string given in TXR Lisp notation.")
(c nil upcase-str
"Custom treatment: ARG is converted to uppercase.")
(b bool :bool
"A flag you can flip true."))
(defvarl prog-name *load-path*)
(let ((o (new prog-opts)))
o.(getopts *args*)
(when (or o.help o.extra-help)
(put-line "Usage:\en")
(put-line ` @{prog-name} [options] arg*`)
o.(opthelp)
(when o.extra-help
o.(opthelp-types)
o.(opthelp-conventions))
(exit -1))
(put-line `args after opts are: @{o.out-args ", "}`))
.brev
.coNP Structure @ opt-desc
.synb
.mets (defstruct opt-desc
.mets \ \ short long helptext type
.mets \ \ ... < unspecified << slots )
.syne
.desc
The
.code opt-desc
structure describes a single command-line option.
The
.code short
and
.code long
slots are either
.code nil
or else hold strings.
The
.code short
slot gives the option's short name: a one-character-long
string which may not be the ASCII dash character
.codn - .
The
.code long
slot gives the option's long name: a string two or more
characters long which doesn't begin with a dash.
An option must have at least one of these names.
The
.code helptext
slot provides a descriptive string. This string may be long. The
.code opthelp
function displays this text, formatting into multiple lines as necessary.
If
.code helptext
is
.codn nil ,
the option is considered undocumented.
The
.code type
slot may be a symbol naming a global function which takes one argument,
or it may be such a function object. Otherwise it must be one of the
following keyword symbols:
.RS
.coIP :bool
This indicates that the type of the option is Boolean. Such
an option doesn't take any argument. Its value is
.code t
or
.codn nil .
.coIP :dec
This indicates that the option requires an argument, which is a
decimal integer with an optional positive or negative sign.
This argument is converted to an integer object.
.coIP :hex
This type indicates that the option requires an argument consisting
of a hexadecimal integer with an optional positive or negative sign.
This is converted to an integer object.
.coIP :oct
This type indicates that the option requires an argument consisting
of a octal integer with an optional positive or negative sign.
This is converted to an integer object.
.coIP :cint
This type indicates that the option requires an integer argument
whose format conforms to one of three C language conventions in most respects,
other than that this integer may have an arbitrary range.
All forms may carry an optional positive or negative leading sign
at the very beginning.
The first convention consists of decimal digits, which must not have
a superfluous leading zero. The second convention consists of octal
digits which are introduced by an extra leading zero.
The third convention consists of hexadecimal digits introduced by the
.code 0x
prefix.
.coIP :float
This type indicates a decimal floating-point argument, which is converted to
a floating-point number. Its basic form is: an optional leading plus or
minus sign, followed by a sequence of one or more digits which may contain
a single decimal point anywhere, including the very beginning of the
sequence or at the end, optionally followed by the letter
.code e
or
.code E
followed by a decimal integer which may have a leading positive or negative
sign, and include leading zeros.
.coIP :text
This type indicates a simple textual argument. The argument is taken as
verbatim UTF-8 text, converted to a string without interpreting
the characters in any special way.
.coIP :str
This type indicates that the argument consists of the interior notation of
a TXR Lisp character string. It is processed by adding a double quote
at the beginning or end, and parsed as a string literal. This parsing must
successfully yield a string object, otherwise the argument is ill-formed.
.meIP (list << type )
If the type is specified as a compound form headed by the
.code list
symbol, it indicates that the command-line option's argument is a list
of elements. The argument appears on the command line as a single string
contained within one argument. It may contain commas, and is split into pieces
using the comma character as a separator. The pieces are then individually
treated as of type
.meta type
and converted accordingly. The option's argument is then a list object
whose elements are the converted pieces. For instance
.code "(list :dec)"
will convert a list of comma-separated decimal integer tokens into
a list of integer objects. The
.code list
option type does not nest.
.meIP (cumul << type )
If the type is specified as a compound form headed by the
.code cumul
symbol, it indicates that if the option is specified multiple times,
the values coming from the multiple occurrences are accumulated into a list.
The
.meta type
argument may be a
.code list
type, exemplified by
.code "(cumul (list :dec))"
or a basic type, such as
.codn "(cumul :str)" .
However, this type specifier does not nest. Combinations such as
.code "(cumul (cumul ...)"
and
.code "(list (cumul ...))"
are invalid.
The option values are accumulated in reverse order, so that the rightmost
repetition becomes the first item in the list. For instance, if the
.code -x
option has type
.codn "(cumul :dec)" ,
and the arguments presented for parsing are
.codn "(\(dq-x\(dq \(dq1\(dq \(dq-x\(dq \(dq2\(dq)" ,
then the option's value will be
.codn "(2 1)" .
If a
.codn list -typed
option is cumulative, then the option value will be a list of lists.
Each repetition of the option produces a list, and the lists are accumulated.
.RE
.IP
If
.code type
is a function, then the option requires an argument. The argument string
is passed to the function, and the value is whatever the function returns.
The
.code opt-desc
structure may have additional slots which are not specified.
The
.code opt
convenience function is provided for constructing
.code opt-desc
objects.
.coNP Function @ opt
.synb
.mets (opt < short < long >> [ type <> [ helptext ]])
.syne
.desc
The
.code opt
function provides a slightly condensed syntax for constructing
an object of type
.codn opt-desc .
The required arguments
.meta short
and
.meta long
are strings, corresponding to
.code opt-desc
slots of the same name.
The optional parameter
.meta type
corresponds to the same-named slot and defaults to
.codn :bool .
The optional parameter
.meta helptext
corresponds to the same-named slot and defaults to
.code nil
(no help text provided for the option).
The
.code opt
function follows this equivalence:
.verb
(opt a b c d) <--> (new opt-desc short a long b
type c helptext d)
.brev
.coNP Structure @ opts
.synb
.mets (defstruct opts nil
.mets \ \ in-args out-args
.mets \ \ ... < unspecified << slots )
.syne
.desc
The
.code opts
structure represents a parsed command line, containing decoded
information obtained from the options and an indication of where
the non-option arguments start.
The
.code opts
structure supports direct indexing for option retrieval.
That is the only documented interface for accessing the parsed
options; the implementation of the information set describing
the parsed options is unspecified.
The
.code in-args
slot holds the original argument list.
The
.code out-args
slot holds the tail of the argument list consisting of the non-option
arguments.
The mechanism by means of which
.code out-args
is calculated, and by means of which the information about the
options is populated, is unspecified. The only interface to that
mechanism is the
.code getopts
function.
The
.code opts
object supports indexing, including indexed assignment.
If
.code o
is an instance of
.code opts
returned by
.codn getopts ,
then the expression
.code "[o \(dqv\(dq]"
tests whether the option
.str v
is available in
.codn o ;
that is, whether it has been specified in the command line.
If so, then its associated value is returned, otherwise
.code nil
is returned. This
.code nil
is ambiguous: for a Boolean option it indicates that either
the option was not specified, or that it was explicitly
specified as false. For a Boolean option that was specified
(positively), the value
.code t
is returned.
The expression
.code "[o \(dqv\(dq dfl]"
yields the value of option
.str v
if that option has been specified. If the option hasn't
been specified, then the expression yields the value
.codn dfl .
Assigning to
.code "[o \(dqv\(dq]"
is possible. This replaces the value associated with option
.strn v .
The assignment is erroneous if no such option was parsed
from the command line, even if it is a valid option.
If an option is defined with both a long form and a short form,
and either form of that option occurs in the command line being
processed, then the option appears under both names in the index.
For instance if option
.str --verbose
has the short form
.strn -v ,
and either option occurs, then both the keys
.str "v"
and
.str "verbose"
will exist in the
.code opts
structure returned by
.codn getopts .
Note that this behavior is different from that of the structure produced
.code define-option-struct
macro. Under that approach, if an option is defined with a long and short name,
the structure will have only a single slot for that option, named after the
long name.
.coNP Function @ getopts
.synb
.mets (getopts < option-desc-list << arg-list )
.syne
.desc
The
.code getopts
function takes a list of
.code opt-desc
structures and a list of strings
.meta arg-list
representing command-line arguments.
The
.meta arg-list
is parsed. If the parse is unsuccessful, an exception of type
.code opt-error
is thrown, derived from
.codn error .
If there are problems in
.code option-desc-list
itself, then an exception of type
.code error
is thrown.
If the parse is successful,
.code getopts
returns an instance of the
.code opts
structure describing the parsed options and listing the non-option
arguments.
.coNP Functions @, opthelp @ opthelp-types and @ opthelp-conventions
.synb
.mets (opthelp < opt-desc-list <> [ stream ])
.mets (opthelp-types < opt-desc-list <> [ stream ])
.mets (opthelp-conventions < opt-desc-list <> [ stream ])
.syne
.desc
The
.code opthelp
function processes the list of
.code opt-desc
structures
.meta opt-desc-list
and compiles a customized body of help describing all of the
options which have help text. These are presented in alphabetical
order. Options which do not have help text, if any, are simply
listed together under a heading which indicates their undocumented status.
The text is formatted to fit within 79 columns, and begins and ends with a
blank line. Its format consists of headings which begin in the first column,
and paragraphs and tables which feature a two space left margin.
A blank line follows each section heading. The heading begins with a capital
letter. Its remaining words are uncapitalized, and it ends with a colon.
The text is sent to
.metn stream ,
if specified. This argument defaults to
.codn *stdout* .
If there are problems in
.code option-desc-list
itself, then an exception of type
.code error
is thrown.
The
.code opthelp-types
supplementary help function processes the
.metn opt-desc-list ,
considering only those options which are documented. If any of them have typed
arguments, then a legend is printed explaining the types. The legend includes
only information about those option argument types which appear in
.metn opt-desc-list .
The
.code opthelp-conventions
supplementary help function processes
.metn opt-desc-list ,
considering only those options which are documented. It prints a guide
to the use of options, which includes information only about the kinds
of options actually present in
.metn opt-desc-list .
.coNP Macro @ define-option-struct
.synb
.mets (define-option-struct < name < super << opt-specifier *)
.syne
.desc
The
.code define-option-struct
macro defines a struct type, instances of which provide command-line option
parsing.
The
.meta name
and
.meta super
parameters are subject to the same requirements and have the same
semantics as the same-named parameters of
.codn defstruct .
The
.meta opt-specifier
arguments are lists of between two and four elements:
.mono
.meti >> ( short-symbol < long-symbol >> [ type <> [ help-text ]]).
.onom
The
.meta short-symbol
and
.meta long-symbol
must be symbols suitable for use as slot names. One of them may be
specified as
.code nil
indicating that the option has no long form, or no short form.
If a
.meta opt-specifier
specifies both a
.meta short-symbol
and a
.meta long-symbol
then only a slot named by
.meta long-symbol
shall exist in the structure.
The struct type defined by
.code define-option-struct
has four methods:
.codn getopts ,
.codn opthelp ,
.code opthelp-types
and
.codn opthelp-conventions .
It also has two slots:
.code in-args
and
.codn out-args ,
which function in a manner identical to their same-named
counterparts in the
.code opts
class.
The
.code getopts
method takes a single argument: the argument list to be processed.
When the argument list is successfully processed.
The
.codn opthelp ,
.code opthelp-types
and
.code opthelp-conventions
methods take an optional stream argument.
Note: to encode the option names
.str "t"
or
.strn "nil" ,
or option names which clash with the slot names
.code in-args
and
.code out-args
or the method names such as
.code getopts
or
.codn opthelp ,
symbols with these names from a package other than
.code usr
must be used.
.SS* System Programming
.coNP Accessor @ errno
.synb
.mets (errno <> [ new-errno ])
.mets (set (errno) << new-value )
.syne
.desc
The
.code errno
function retrieves the current value of the C library error variable
.codn errno .
If the argument
.meta new-errno
is present and is not
.codn nil ,
then it
specifies a value which is stored into
.codn errno .
The value returned is the prior value.
The place form of
.code errno
does not take an argument.
.coNP Function @ strerror
.synb
.mets (strerror << errno-value )
.syne
.desc
The
.code strerror
returns a character string which provides the host platform's description
of the integer
.meta errno-value
obtained from the
.code errno
function.
If the host platform fails to provide a description, the function returns
.codn nil .
.coNP Function @ exit
.synb
.mets (exit <> [ status ])
.syne
.desc
The
.code exit
function terminates the entire process (running \*(TX image), specifying
the termination status to the operating system. Values of the optional
.meta status
parameter may be
.codn nil ,
.codn t ,
or an integer value. The value
.code nil
indicates an unsuccessful termination status, whereas
.code t
indicates a successful termination status.
An absence of the
.meta status
argument also specifies a successful termination status.
If
.meta status
is an integer value, it specifies a successful termination if it is
.codn 0 ,
otherwise the interpretation of the value is platform-specific.
.coNP Variables @, e2big @, eacces @, eaddrinuse @, eaddrnotavail @, eafnosupport @, eagain @, ealready @, ebadf @, ebadmsg @, ebusy @, ecanceled @, echild @, econnaborted @, econnrefused @, econnreset @, edeadlk @, edestaddrreq @, edom @, edquot @, eexist @, efault @, efbig @, ehostunreach @, eidrm @, eilseq @, einprogress @, eintr @, einval @, eio @, eisconn @, eisdir @, eloop @, emfile @, emlink @, emsgsize @, emultihop @, enametoolong @, enetdown @, enetreset @, enetunreach @, enfile @, enobufs @, enodata @, enodev @, enoent @, enoexec @, enolck @, enolink @, enomem @, enomsg @, enoprotoopt @, enospc @, enosr @, enostr @, enosys @, enotconn @, enotdir @, enotempty @, enotrecoverable @, enotsock @, enotsup @, enotty @, enxio @, eopnotsupp @, eoverflow @, eownerdead @, eperm @, epipe @, eproto @, eprotonosupport @, eprototype @, erange @, erofs @, espipe @, esrch @, estale @, etime @, etimedout @, etxtbsy @ ewouldblock and @ exdev
.desc
These variables correspond to the POSIX
.cod2 \(dq errno
constants\(dq, namely
.codn E2BIG ,
.codn EACCES ,
.code EADDRINUSE
and so forth.
Variables corresponding to all of the
.code "<errno.h>"
constants from the Issue 6 2004 edition of POSIX are included.
The variables
.code eownerdead
and
.code enotrecoverable
from Issue 7 2018 are subject to the availability of the corresponding constants
in the host platform.
.coNP Function @ abort
.synb
.mets (abort)
.syne
.desc
The
.code abort
function terminates the entire process (running \*(TX image), specifying
an abnormal termination status to the process.
Note:
.code abort
calls the C library function
.code abort
which works by raising the
.code SIG_ABRT
signal, known in \*(TX as the
.code sig-abrt
variable. Abnormal termination of the process is this signal's
default action.
.coNP Functions @ at-exit-call and @ at-exit-do-not-call
.synb
.mets (at-exit-call << function )
.mets (at-exit-do-not-call << function )
.syne
.desc
The
.code at-exit-call
function registers
.meta function
to be called when the process terminates normally.
Multiple functions can be registered, and the same function
can be registered more than once. The registered
functions are called in reverse order of their
registrations.
The
.code at-exit-do-not-call
function removes all previous
.code at-exit-call
registrations of
.metn function .
The
.code at-exit-call
function returns
.metn function .
The
.code at-exit-do-not-call
function returns
.code t
if it removed anything,
.code nil
if no registrations of
.meta function
were found.
.coNP Function @ usleep
.synb
.mets (usleep << usec )
.syne
.desc
The
.code usleep
function suspends the execution of the program for at least
.meta usec
microseconds.
The return value is
.code t
if the sleep was successfully executed. A
.code nil
value indicates premature wakeup or complete failure.
Note: the actual sleep resolution is not guaranteed, and depends on granularity
of the system timer. Actual sleep times may be rounded up to the nearest 10
millisecond multiple on a system where timed suspensions are triggered by a 100
Hz tick.
.coNP Functions @ mkdir and @ ensure-dir
.synb
.mets (mkdir < path <> [ mode ])
.mets (ensure-dir < path <> [ mode ])
.syne
.desc
.code mkdir
tries to create the directory named
.meta path
using the POSIX
.code mkdir
function.
An exception of type
.code file-error
is thrown if the function fails. Returns
.code t
on success.
The
.meta mode
argument specifies the request numeric permissions
for the newly created directory. If omitted, the requested permissions are
.code #o777
(511): readable and writable to everyone. The requested permissions
are subject to the system
.codn umask .
The function
.code ensure-dir
also creates a directory named
.metn path .
Unlike
.codn mkdir ,
it also attempt to create all the necessary parent directories,
and does not throw an error if
.meta path
refers to an existing object, if that object is a directory or a symbolic
link to a directory. Rather, in that case it returns
.code nil
instead of
.codn t .
.coNP Function @ chdir
.synb
.mets (chdir << path )
.syne
.desc
.code chdir
changes the current working directory to
.metn path ,
and returns
.metn t ,
or else throws an exception of type
.codn file-error .
.coNP Function @ pwd
.synb
.mets (pwd)
.syne
.desc
The
.code pwd
function retrieves the current working directory.
If the underlying
.code getcwd
C library function fails with an
.code errno
other than
.codn ERANGE ,
an exception will be thrown.
.coNP Function @ rmdir
.synb
.mets (rmdir << path )
.syne
.desc
The
.code rmdir
function removes the directory named by
.codn path .
If successful, it returns
.metn t ,
otherwise it throws an exception of type
.codn file-error .
Note:
.code rmdir
calls the same-named POSIX function, which requires
.code path
to be the name of an empty directory.
.coNP Function @ remove-path
.synb
.mets (remove-path < path <> [ throw-on-error-p ])
.syne
.desc
The
.code remove-path
function tries to remove the filesystem object named
by
.metn path ,
which may be a file, directory or something else.
If successful, it returns
.codn t .
The optional Boolean parameter
.metn throw-on-error-p ,
which defaults to
.codn nil .
A failure to remove the object results in an exception of type
.code file-error
being thrown, unless the failure reason is that the object indicated by
.meta path
doesn't exist. In this non-existence case, the behavior is controlled by the
.meta throw-on-error
argument. If that argument is true, the exception is thrown. Otherwise,
the function returns normally, producing the value
.code nil
to indicate that it didn't perform a removal.
.coNP Function @ rename-path
.synb
.mets (rename-path < from-path << to-path )
.syne
.desc
The
.code rename-path
function tries to rename filesystem path
.metn from-path ,
which may refer to a file, directory or something else, to the path
.metn to-path .
If successful, it returns
.codn t .
A failure results in an exception of type
.codn file-error .
.coNP Functions @ sh and @ run
.synb
.mets (sh << system-command )
.mets (run < program <> [ argument-list ])
.syne
.desc
The
.code sh
function executes
.meta system-command
using the system command interpreter.
The run function spawns a
.metn program ,
searching for it using the
system PATH. Using either method, the executed process receives environment
variables from the parent.
\*(TX blocks until the process finishes executing. If the program terminates
normally, then its integer exit status is returned. The value zero indicates
successful termination.
The return value
.code nil
indicates an abnormal termination, or the inability
to run the process at all.
In the case of the
.code run
function, if the child process is created successfully
but the program cannot be executed, then the exit status will be an
.code errno
value from the failed
.code exec
attempt.
The standard input, output and error file descriptors of an executed
command are obtained from the streams stored in the
.codn *stdin* ,
.code *stdout*
and
.code *stderr*
special variables, respectively. For a detailed description of the
behavior and restrictions, see the
.code open-command
function, whose description of this mechanism applies to the
.code run
and
.code sh
function also.
Note: as of \*(TX 120, the
.code sh
function is implemented using
.code run
and not by means of the
.code system
C library function, as previously. The
.code run
function is used to invoke the system interpreter by name. On Unix-like
systems, the string
.code /bin/sh
is assumed to denote the system interpreter, which is expected to
support a pair of arguments
.mono
.meti -c < command
.onom
to specify the command to be executed. On MS Windows, the interpreter
is assumed to be the relative pathname
.code cmd.exe
and expected to support
.mono
.meti /C < command
.onom
as a way of specifying a command to execute.
.SS* Unix Filesystem Manipulation
.coNP Structure @ stat
.synb
.mets (defstruct stat nil
.mets \ \ dev ino mod nlink uid gid
.mets \ \ rdev size blksize blocks
.mets \ \ atime atime-nsec mtime mtime-nsec
.mets \ \ ctime ctime-nsec path)
.syne
.desc
The
.code stat
structure defines the type of object which is returned
by the
.code stat
and
.code lstat
functions. Except for
.codn path ,
.codn atime-nsec ,
.code ctime-nsec
and
.codn mtime-nsec ,
the slots are the direct counterparts of the
members of POSIX C structure
.codn "struct stat" .
For instance the slot
.code dev
corresponds to
.codn st_dev .
The
.code path
slot is set by the functions
.code stat
and
.codn lstat .
Its value is
.code nil
when the path is not available.
The
.codn atime-nsec ,
.code ctime-nsec
and
.code mtime-nsec
fields give the fractional parts of
.codn atime ,
.code ctime
and
.codn mtime ,
respectively. They are derived from the newer style information
in which the POSIX function provides the timestamps in
.code "struct timespec"
format. If that is not available from the platform, these
fields take on values of zero.
.coNP Functions @, stat @ lstat and @ fstat
.synb
.mets (stat >> { path | < stream | << fd } <> [ struct ])
.mets (lstat << path )
.mets (fstat >> { path | < stream | << fd } <> [ struct ])
.syne
.desc
The
.code stat
function retrieves information about a filesystem object whose pathname
is given by the string argument
.metn path ,
or else about a system object associated with the open stream
.metn stream ,
or one associated with the integer file descriptor
.metn fd .
If a
.meta stream
is specified, that stream must be of a kind from which the
.code fileno
function can retrieve a file descriptor, otherwise an exception of type
.code file-error
is thrown.
If the object is not found or cannot be
accessed, an exception is thrown.
Otherwise, if the
.meta struct
argument is missing, information is retrieved and returned, in the form of a
new structure of type
.codn stat .
If the
.meta struct
argument is present, it must be either: an instance of the
.code struct
structure type, or of a type derived from that type by inheritance, or
else structure type which has all the same slots as the
.code struct
type. The retrieved information is stored into
.meta struct
and that object is returned rather than a new object.
If
.meta path
refers to a symbolic link, the
.code stat
function retrieves information about the target of the link, if it exists,
or else throws an exception of type
.codn file-error .
The
.code lstat
function behaves the same as
.code stat
on objects which are not symbolic links. For a symbolic link, it retrieves
information about the link itself, rather than its target.
The
.code path
slot of the returned structure
holds a copy of their
.meta path
argument value.
When information is retrieved using a
.meta stream
or
.meta fd
argument, this slot is
.codn nil .
The
.code fstat
function is an alias for
.codn stat .
Note: until \*(TX 231,
.code stat
and
.code fstat
were distinct functions:
.code stat
accepted only
.meta path
arguments, whereas
.code fstat
function accepted only
.meta stream
or
.meta fd
arguments.
.coNP Variables @, s-ifmt @, s-iflnk @, s-ifreg @, s-ifblk ..., @ s-ixoth
.desc
The following variables exist, having integer values. These are bitmasks
which can be applied against the value given by the
.code mode
slot of the
.code stat
structure returned by the function
.codn stat :
.codn s-ifmt ,
.codn s-ifsock ,
.codn s-iflnk ,
.codn s-ifreg ,
.codn s-ifblk ,
.codn s-ifdir ,
.codn s-ifchr ,
.codn s-ififo ,
.codn s-isuid ,
.codn s-isgid ,
.codn s-isvtx ,
.codn s-irwxu ,
.codn s-irusr ,
.codn s-iwusr ,
.codn s-ixusr ,
.codn s-irwxg ,
.codn s-irgrp ,
.codn s-iwgrp ,
.codn s-ixgrp ,
.codn s-irwxo ,
.codn s-iroth ,
.code s-iwoth
and
.codn s-ixoth .
These variables correspond to the C language constants from POSIX:
.codn S_IFMT ,
.codn S_IFLNK ,
.code S_IFREG
and so forth.
The
.code logtest
function can be used to test these against values of mode.
For example
.code "(logtest mode s-irgrp)"
tests for the group read permission.
.coNP Function @ umask
.synb
.mets (umask <> [ mask ])
.syne
.desc
The
.code umask
function provides access to the Unix C library function of the same name,
which controls which permissions are denied
when files are newly created.
If
.code umask
is called with no argument, it returns the current value of the mask.
If the
.meta mask
argument is present, it must be an integer specifying the new mask to be
installed. The previous mask is returned.
If
.meta mask
is absent, then
.code umask
returns the previous mask.
Note: the value of the
.meta mask
argument may be calculated as a bitwise or of the following constants:
.codn s-irwxu ,
.codn s-irusr ,
.codn s-iwusr ,
.codn s-ixusr ,
.codn s-irwxg ,
.codn s-irgrp ,
.codn s-iwgrp ,
.codn s-ixgrp ,
.codn s-irwxo ,
.codn s-iroth ,
.code s-iwoth
and
.codn s-ixoth ,
which correspond to the POSIX C constants
.codn S_IRWXU ,
.codn S_IRUSR ,
.codn S_IWUSR ,
.codn S_IXUSR ,
.codn S_IRWXG ,
.codn S_IRGRP ,
.codn S_IWGRP ,
.codn S_IXGRP ,
.codn S_IRWXO ,
.codn S_IROTH ,
.code S_IWOTH
and
.codn S_IXOTH .
Implementation note: since the
.code umask
C library function provides no way to retrieve the current mask without
overwriting with a new one, the \*(TX
.code umask
function, when given no argument, simulates the pure retrieval of the mask
by calling the C function with an argument of
.code #o777
to temporarily install the maximally safe mask. The value returned is then
reinstated as the mask by another call to
.codn umask ,
and that value is also returned.
.coNP Functions @, makedev @ minor and @ major
.synb
.mets (makedev < minor << major )
.mets (minor << dev )
.mets (major << dev )
.syne
.desc
The parameters
.metn minor ,
.meta major
and
.meta dev
are all integers. The
.code makedev
function constructs a combined device number from a minor and major pair (by
calling the Unix
.code makedev
function). This device number is suitable as an
argument to the
.code mknod
function (see below). Device numbers also appear as values of the
.code dev
slot of the
.code stat
structure.
The
.code minor
and
.code major
functions extract the minor and major device number
from a combined device number.
.coNP Function @ chmod
.synb
.mets (chmod < target << mode )
.syne
.desc
The
.code chmod
function changes the permissions of the filesystem object
specified by
.metn target .
It is implemented in terms of the POSIX functions
.code chmod
and
.codn fchmod .
If
.meta mode
is a character string representing a symbolic mode, then the function
also makes use of
.code stat
or
.code fstat
and
.codn umask .
The permissions are specified by
.metn mode ,
which must be an integer or a string.
An integer
.meta mode
is a bitwise combination of permission mode bits. The value is passed
directly to the POSIX
.code chmod
or
.code fchmod
function.
Note: to construct a mode value, applications may use
.code logior
to combine the values
of the variables like
.code s-irusr
or
.code s-ixoth
or take advantage of the well-known numeric structure of POSIX
permissions to express them octal in octal notation. For instance the mode
.code #o750
denotes that the owner has read, write and execute permissions,
the group owner has read and execute, others have no permission.
This value may also be calculated using
.codn "(logior s-irwxu s-irgrp s-ixgrp)" .
If the argument to
.meta mode
is a string, it is interpreted according to the symbolic syntax
of the POSIX
.code chmod
utility. For instance, a
.meta mode
value of
.str a+w,-s
means to give all users (owner, group and others) write permission,
and remove the setuid and setgid bits.
The full syntax and semantics of symbolic
.meta mode
strings is given in the POSIX standard IEEE 1003.1.
The function throws a
.code file-error
exception if an error occurs, otherwise it returns
.codn t .
The
.meta target
argument may be a character string, in which case it specifies a pathname in
the filesystem. In this case, the POSIX function
.code chmod
is invoked.
The
.meta target
argument may also be an integer file descriptor, or a stream. In these two
cases, the POSIX
.code fchmod
function is invoked. For a stream
.metn target ,
the integer file descriptor is retrieved from the stream using
.code fileno
function.
.TP* Example:
.verb
;; Set permissions of foo.txt to "rw-r--r--"
;; (owner can read and write; group owner
;; and other users can only read).
;; numerically:
(chmod "foo.txt" #o644)
;; symbolically:
(chmod "foo.txt" (logior s-irusr s-iwusr
s-irgrp
s-iroth))
.brev
Implementation note: The implementation of the symbolic
.meta mode
processing is based on the descriptions given in IEEE 1003.1-2018,
Issue 7 and also on the
.code chmod
program from from GNU Coreutils 8.28: and experiments with its behavior,
and its documentation.
.coNP Functions @ chown and @ lchown
.synb
.mets (chown < target < id << gid )
.mets (lchown < target < id << gid )
.syne
.desc
The
.code chown
and
.code lchown
functions change the user and group ownership of the filesystem object
specified by
.metn target .
They implemented in terms of the POSIX functions
.codn chown ,
.code fchown
and
.codn lchown .
The ownership attributes are specified by
.meta uid
and
.metn gid ,
both integer arguments.
The existing ownership attributes may be obtained using the
.code stat
function.
These functions throw a
.code file-error
exception if an error occurs, otherwise they returns
.codn t .
The
.meta target
argument may be a character string, in which case it specifies a pathname in
the filesystem. In this case, the same-named POSIX function
.code chown
is invoked by
.codn chown ,
whereas
.code lchown
likewise invokes its respective same-named POSIX counterpart.
The difference is that if
.meta target
is a pathname denoting a symbolic link, then
.code lchown
operates on the symbolic link, whereas
.code chown
dereferences the symbolic link.
The
.meta target
argument may also be an integer file descriptor, or a stream. In these two
cases, the POSIX
.code fchown
function is invoked by either function. For a stream
.metn target ,
the integer file descriptor is retrieved from the stream using
.code fileno
function.
Note: in most POSIX systems, unprivileged processes may not change the user
ownership denoted by
.metn uid .
They may change the group ownership indicated in
.metn gid ,
if that value corresponds to the effective group ID of the calling
process or one of its ancillary group IDs.
To avoid trying to change the user ownership (and therefore failing),
the caller should specify a
.meta uid
value which matches the object's existing owner.
.coNP Functions @ utimes and @ lutimes
.synb
.mets (utimes < target < atime-s < atime-ns < mtime-s << mtime-ns )
.mets (lutimes < target < atime-s < atime-ns < mtime-s << mtime-ns )
.syne
.desc
The functions
.code utimes
and
.code lutimes
change the access and modification timestamps of a file indicated by the
.meta target
argument.
The difference between the two functions is that if
.meta target
is the pathname of a symbolic link, then
.code lutimes
operates on the symbolic link itself, whereas
.code utimes
resolves the symbolic link.
Note: the full, complete functionality of these functions requires the
platform to provide the POSIX functions
.code futimens
and
.code utimensat
functions. If these functions are not available, then other functions are
relied on, with some reductions in functionality, that are documented below.
The
.meta target
argument specifies the file to operate on. It may be an integer file descriptor,
an open stream, or a character string representing a pathname.
The
.meta atime-s
and
.meta mtime-s
parameters specify the whole seconds part of the new access and modification
times, expressed as seconds since the epoch.
The
.meta atime-ns
and
.meta mtime-ns
parameters specify the fractional part of the access and modification
times, expressed in nanoseconds. If an integer argument is given to these
parameters, it must lie in the range 0 to 999999999, or else the symbols
.code nil
or
.code t
may be passed as arguments.
If the symbol
.code nil
is passed as the nanoseconds part of the access or modification time,
then the access or modification time, respectively, shall not be modified
by the operation. The corresponding seconds argument is ignored.
If the symbol
.code t
is passed as the nanoseconds part of the access or modification time,
then the access or modification time, respectively, shall be obtained
from the current system time. The corresponding seconds argument is ignored.
If the
.code utimensat
and
.code futimens
functions are not available from the host system, then the above
.code nil
and
.code t
convention in the nanoseconds arguments is not supported; the function
will fail by throwing an exception if an attempt is made to pass these
arguments.
If the
.code utimensat
and
.code futimens
functions are not available from the host system, then operating on
a symbolic link with
.code lutimes
is only possible if the system provides the
.code lutimes
C library function, otherwise the operation fails by throwing an exception
(if given a path argument for
.metn target ,
even if that path isn't a symbolic link).
If the implementation falls back on the
.codn utimes ,
.codn futimes ,
and
.code lutimes
functions, then the nanoseconds arguments are truncated to microsecond
precision.
If the implementation falls back on
.codn utime ,
then the nanoseconds arguments are ignored; the times are effectively
truncated to whole seconds.
.coNP Function @ mknod
.synb
.mets (mknod < path < mode <> [ dev ])
.syne
.desc
The
.code mknod
function tries to create an entry in the filesystem: a file,
FIFO, or a device special file, under the name
.metn path .
If it is successful,
it returns
.codn t ,
otherwise it throws an exception of type
.codn file-error .
The
.meta mode
argument is a bitwise or combination of the requested permissions,
and the type of object to create: one of the constants
.codn s-ifreg ,
.codn s-ififo ,
.codn s-ifchr ,
.code s-ifblk
or
.codn s-ifsock .
The permissions are subject to the system
.codn umask .
If a block or character special device
.cod2 ( s-ifchr
or
.codn s-ifblk )
is being
created, then the
.meta dev
argument specifies the major and minor numbers
of the device. A suitable value can be constructed from a major and minor
pair using the
.code makedev
function.
.TP* Example:
.verb
;; make a character device (8, 3) called /dev/foo
;; requesting rwx------ permissions
(mknod "dev/foo" (logior #o700 s-ifchr) (makedev 8 3))
.brev
.coNP Function @ mkfifo
.synb
.mets (mkfifo < path << mode )
.syne
.desc
The
.code mkfifo
function creates a POSIX FIFO object.
If it is successful,
it returns
.codn t ,
otherwise it throws an exception of type
.codn file-error .
The
.meta mode
argument is a bitwise or combination of the requested permissions,
and is subject to the system
.codn umask .
Note: the
.code mknod
function can also create FIFOs, specified via the bitwise combination
of the
.code s-ififo
type and the permission mode bits.
.coNP Functions @ symlink and @ link
.synb
.mets (symlink < target << path )
.mets (link < target << path )
.syne
.desc
The
.code symlink
function creates a symbolic link called
.meta path
whose contents
are the absolute or relative path
.metn target .
.meta target
does not actually have to exist.
The link function creates a hard link. The object at
.meta target
is installed
into the filesystem at
.meta path
also.
If these functions succeed, they return
.codn t .
Otherwise they throw an exception
of type
.codn file-error .
.coNP Function @ readlink
.synb
.mets (readlink << path )
.syne
.desc
If
.meta path
names a filesystem object which is a symbolic link, the
.code readlink
function reads the contents of that symbolic link and returns it
as a string. Otherwise, it fails by throwing an exception of type
.codn file-error .
.coNP Function @ realpath
.synb
.mets (realpath << path )
.syne
.desc
The
.code realpath
function provides access to the same-named POSIX function.
It processes the input string
.meta path
by expanding all symbolic links, removes all superfluous
.str ".."
and
.str "."
path components, and extra component-separating slash characters,
to produce a canonical absolute pathname.
If the underlying POSIX function indicates failure, then
.code nil
is returned. In that situation the
.code errno
value is available using the
.code errno
function.
.SS* Unix Filesystem Complex Operations
Functions in this category are complex functionality implemented using
a combination of multiple calls into the host system's POSIX API.
.coNP Functions @ copy-file and @ copy-files
.synb
.mets (copy-file < from-path < to-path >> [ perms-p <> [ times-p ]])
.mets (copy-file < from-list < to-dir >> [ perms-p <> [ times-p ]])
.syne
.desc
The
.code copy-file
function creates a replica of the file
.code from-path
at the destination path
.metn to-path .
Both paths are opened using
.code open-file
in binary mode, as if using
.mono
.meti (open-file < from-path \(dqb\(dq)
.onom
and
.mono
.meti (open-file < to-path \(dqwb\(dq)
.onom
respectively. Then bytes are read from one stream and written to the other,
in blocks which whose size is a power of two at least as large as 16834.
If the optional Boolean parameter
.meta perms-p
is specified, and is true, then the permissions of
.meta from-path
are propagated to
.metn to-path .
If the optional Boolean parameter
.meta times-p
is specified, and is true, then the access and modification timestamps of
.meta from-path
are propagated to
.metn to-path .
The
.code copy-file
function returns
.code nil
if it is successful, and throws an exception derived from
.code file-error
on failure.
The
.code copy-files
function copies multiple files, whose pathnames are given by the list argument
.meta from-list
into the target directory whose path is given by
.metn to-dir .
The target directory must exist.
For source each path in
.metn from-list ,
the
.code copy-files
function forms a target path by combining the base name of the
source path with
.metn target-dir .
(See the
.code base-name
and
.code path-cat
functions).
Then, the source path is copied to the resulting target path, as if by the
.code copy-file
function.
The
.code copy-files
function returns
.code nil
if it is successful, and throws an exception derived from
.code file-error
on failure.
Additionally,
.code copy-files
provides an internal catch for the
.code retry
and
.code skip
restart exceptions. If the caller, using a handler frame established by
.codn handle ,
catches an error emanating from the
.code copy-files
function, it can retry the failed operation by throwing the
.code retry
exception, or continue copying with the next file by throwing the
.code skip
exception.
.TP* Example:
.verb
;; Copy all "/mnt/cdrom/*.jpg" files into "images" directory,
;; preserving their time stamps,
;; continuing the operation in the face of
;; file-error exceptions.
(handle
(copy-files (glob "/mnt/cdrom/*.jpg") "images" nil t)
(file-error (throw 'skip)))
.brev
.coNP Function @ copy-path-rec
.synb
.mets (copy-path-rec < from-path < to-path << option *)
.syne
.desc
The
.code copy-path-rec
function replicates a file system object identified by the pathname
.metn from-path ,
creating a similar object named
.metn to-path .
If
.code from-path
is a directory, it is recursively traversed and its structure and content
is replicated under
.codn to-path .
The
.meta option
arguments are keywords, which may be the following:
.RS
.IP :perms
Propagate the permissions of all objects under
.meta from-path
onto their
.meta to-path
counterparts. In the absence of this option, the copied objects
receive permissions with are calculated by applying the
.code umask
of the calling process to the maximally liberal.
.IP :times
Propagate the modification and access time stamps of all objects under
.meta from-path
onto their
.meta to-path
counterparts.
.IP :symlinks
Copy symbolic links literally rather than dereferencing them.
Symbolic links are not altered in any way; their exact content
is preserved. Thus, relative symlinks which point outside of the
.meta from-path
tree may turn into dangling symlinks in the
.meta to-path
tree.
.IP :owner
Propagate the ownership of all objects under
.meta from-path
to their
.meta to-path
counterparts. Ownership refers to the owner user ID and group ID.
Without this option, the ownership of the copied objects is derived
from the effective user ID and group ID of the calling process.
Note that it is assumed that the host system may requires superuser
privileges to set both ownerships IDs of an object, and to set them to an
arbitrary value. An unprivileged process may not change the user ID of a file,
and may only change the group ID of a file which they own, to one of the groups
of which that process is a member, either via the effective GID, or the
ancillary list. The
.code copy-path-rec
function tests whether the application is running under superuser privileges;
if not, then it only honors the
.code :owner
option for those objects under
.meta from-path
which are owned by the caller, and owned by a group to
which the caller belongs.
Other objects are copied as if the
.code :owner
option were not in effect, avoiding an attempt to set their ownership
that is likely to fail.
.IP :all
The
.code :all
keyword is a shorthand representing all of the options being applied:
permissions, times, symlinks and ownership are replicated.
.RE
.IP
The
.code copy-path-rec
function creates all necessary pathname components required for
.meta to-path
to come into existence, as if by using the
.code ensure-dir
function.
Whenever an object under
.meta from-path
has a counterpart in
.meta to-path
which already exists, the situation is handled as follows:
.RS
.IP 1.
If a directory object is copied to an existing directory object,
then that existing directory object is accepted as the copy, and
the operation continues recursively within that directory. If any options are
specified, then the requested attributes are propagated to that existing
directory.
.IP 2.
If a non-directory object is copied to a directory object, the
situation throws an exception: the
.code copy-path-rec
function refuses to delete an entire directory or subdirectory in order
to make way for a file, symbolic link, special device or any other kind
of non-directory object.
.IP 3.
If any object is copied to an existing non-directory object,
that target object is removed first, then the copy operation proceeds.
.RE
.IP
Copying of files takes place similarly as what is described for the
.code copy-file
function.
Special objects such as FIFOs, character devices, block devices and sockets
are copied by creating a new, similar objects at the destination path.
In the case of devices, the major and minor numbers of the copy are
derived from the original, so that the copy refers to the same device.
However, the copy of a socket or a FIFO is effectively a new, different
endpoint because these objects are identified by their pathname.
Processes using the copy of a socket or a FIFO will not connect to
processes which are working with the original.
The
.code copy-path-rec
function returns
.code nil
if it is successful. It throws an exception derived from
.code file-error
when encountering failures.
Additionally
.code copy-path-rec
provides an internal catch for the
.code retry
and
.code skip
restart exceptions. If the caller, using a handler frame established by
.codn handle ,
catches an error emanating from the
.code copy-files
function, it can retry the failed operation by throwing the
.code retry
exception, or continue copying with the next object by throwing the
.code skip
exception.
.coNP Function @ remove-path-rec
.synb
.mets (remove-path-rec << path )
.syne
.desc
The
.code remove-path-rec
function attempts to remove the filesystem object named by
.metn path .
If
.meta path
refers to a directory, that directory is recursively traversed
to remove all of its contents, and is then removed.
The
.code remove-path-rec
function returns
.code nil
if it is successful. It throws an exception derived from
.code file-error
when encountering failures.
Additionally
.code remove-path-rec
provides an internal catch for the
.code retry
and
.code skip
restart exceptions. If the caller, using a handler frame established by
.codn handle ,
catches an error emanating from the
.code copy-files
function, it can retry the failed operation by throwing the
.code retry
exception, or continue removing other objects by throwing the
.code skip
exception. Skipping a failed remove operation may cause subsequent
operations to fail. Notably, the failure to remove an item inside
a directory means that removal of that directory itself will fail,
and ultimately,
.meta path
will still exist when
.code remove-path-rec
completes and returns.
.coNP Functions @ chmod-rec and @ chown-rec
.synb
.mets (chmod-rec < path << mode )
.mets (chown-rec < path < uid << gid )
.syne
.desc
The
.code chmod-rec
and
.code chown-rec
functions are recursive counterparts of
.code chmod
and
.codn lchown .
The filesystem object given by
.meta path
is recursively traversed, and each of its constituent objects
is subject to a permission change in the case of
.codn chown-rec ,
or an ownership change in the case of
.codn chown-rec .
The
.code chmod-rec
function alters the permission of each object that is not a symbolic link
using the
.code chmod
function, and
.meta mode
is interpreted accordingly: it may be an integer or string.
Each object which is a symbolic link is ignored.
The
.code chown-rec
function alters the permission of each object encountered, including
symbolic links, using the
.code lchown
function.
These functions establish restart catches, similarly to
.code remove-path-rec
and
.codn copy-path-rec ,
allowing the caller to retry individual failed operations or skip the objects
on which operations have failed.
.coNP Function @ touch
.synb
.mets (touch < path <> [ ref-path ])
.syne
.desc
The
.code touch
function updates the modification timestamp of the filesystem object
named by
.metn path .
If the object doesn't exist, it is created as a regular file.
If
.meta ref-path
is specified, then the modification timestamp of the object denoted by
.meta path
is updated to be equivalent to the modification timestamp of
the object denoted by
.metn ref-path .
Otherwise
.meta ref-path
being absent, the modification timestamp of
.meta path
is set to the current time.
If
.meta path
is a symbolic link, it is dereferenced;
.code touch
operates on the target of the link.
.coNP Function @ mkdtemp
.synb
.mets (mkdtemp << prefix )
.syne
.desc
The
.code mkdtemp
function combines the
.metn prefix ,
which is a string, with a generated suffix to create a unique directory
name. The directory is created, and the name is returned.
If the
.code prefix
argument ends in with a sequence of one or more
.code X
characters, the behavior is unspecified.
Note: this function is implemented using the same-named POSIX function.
Whereas the POSIX function requires the template to end in a sequence of
at least six
.code X
characters, which are replaced by the generated suffix, the \*(TL function
handles this detail internally, requiring only the prefix part without those
characters.
.coNP Function @ mkstemp
.synb
.mets (mkstemp < prefix <> [ suffix ])
.syne
.desc
The
.code mkstemp
function create a unique file name by adding a generated infix between the
.meta prefix
and
.meta suffix
strings.
The file is created, and a stream open in
.str w+b
mode for the file is returned.
If either the
.meta prefix
or
.meta suffix
contain
.code X
characters, the behavior is unspecified.
If
.meta suffix
is omitted, it defaults to the empty string.
The name of the file is available by interrogating the returned stream's
.code :name
property using the function
.codn stream-get-prop .
Notes: this function is implemented using the POSIX function
.code mkstemp
or, if available, using the
.code mkstemps
function which is not standardized, but appears in the GNU C Library
and some other systems. If
.code mkstemps
is unavailable, then the suffix functionality is not available: the
.meta suffix
argument must either be omitted, or must be an empty string.
Whereas the C library functions require the template to contain a sequence
at least six
.code X
characters, which are replaced by the generated portion, the \*(TL function
handles this detail internally, requiring no such characters in any of its
inputs.
.SS* Unix Filesystem Object Existence, Type and Access Tests
Functions in this category perform various tests on the attributes of
filesystem objects.
The functions all have a
.meta path
parameter, which accepts three types of arguments. If a character
string is specified, it denotes a filesystem path to
be probed for properties such as ownership and permissions.
The object is probed using the
.code stat
function except in the case of
.code path-symlink-p
which uses
.codn lstat .
If instead a stream is specified as
.metn path ,
then the associated filesystem descriptor is probed for these properties.
If an integer value is specified, it is treated as a POSIX
open file descriptor that is to be probed.
Otherwise, a
.code stat
structure, for example one returned by the
.code stat
or
.code lstat
function may be specified, in which case no system object
is probed. The properties to be tested are those given in the
.code stat
object.
Note: in a situation when it is necessary to use any of these functions to
probe the properties of a symbolic link itself (other than the function
.code path-symlink-p
which does so implicitly) it is necessary to first invoke
.code lstat
on the symlink's path, and then pass the resulting
.code stat
structure to that function instead of the path.
Some of the accessibility tests (functions which determine whether the
calling process has certain access rights) may not be perfectly accurate, since
they are based strictly on portable information available via
.codn stat ,
together with the basic, portable POSIX APIs for inquiring about
security credentials, such as
.codn getuid .
They ignoring any special permissions which may exist such as operating system
and file system specific extended attributes (for example, file immutability
connected to a "secure level" and such) and special process capabilities
not reflected in the basic credentials.
The accessibility tests use the real credentials of the caller, rather than the
effective credentials. Thus, in a setuid process, where the real and effective
privileges are different, the access tests inquire about whether the real user
has the given access, not the effective user.
.coNP Function @ path-exists-p
.synb
.mets (path-exists-p << path )
.syne
.desc
The
.code path-exists-p
function returns
.code t
if
.meta path
is a string which resolves to a filesystem object.
Otherwise it returns
.codn nil .
If the
.meta path
names a dangling symbolic link, it is considered nonexistent.
If
.meta path
is an object returned by
.code stat
or
.codn lstat ,
.code path-exists-p
unconditionally returns
.codn t .
.coNP Functions @, path-file-p @, path-dir-p @, path-symlink-p @, path-blkdev-p @, path-chrdev-p @ path-sock-p and @ path-pipe-p
.synb
.mets (path-file-p << path )
.mets (path-dir-p << path )
.mets (path-symlink-p << path )
.mets (path-blkdev-p << path )
.mets (path-chrdev-p << path )
.mets (path-sock-p << path )
.mets (path-pipe-p << path )
.syne
.desc
.code path-file-p
tests whether
.meta path
exists and is a regular file.
.code path-dir-p
tests whether
.meta path
exists and is a directory.
.code path-symlink-p
tests whether
.meta path
exists and is a symbolic link.
Similarly,
.code path-blkdev-p
tests for a block device,
.code path-chrdev-p
for a character device,
.code path-sock-p
for a socket and
.code path-pipe-p
for a named pipe.
.coNP Function @ path-dir-empty
.synb
.mets (path-dir-empty << path )
.syne
.desc
The
.code path-dir-empty
function returns
.code t
if
.meta path
is an empty directory.
Implementation note: this function performs a test similar to
.codn path-dir-p ;
then, if it is confirmed that
.meta path
is a directory, a directory stream is opened and entries are read.
If an entry is seen which has a name other than
.str .
or
.str ..
then it is concluded that the directory is not empty and
.code nil
is returned. If no such entry is seen, then the directory is deemed empty and
.code t
is returned.
.coNP Functions @, path-setgid-p @ path-setuid-p and @ path-sticky-p
.synb
.mets (path-setgid-p << path )
.mets (path-setuid-p << path )
.mets (path-sticky-p << path )
.syne
.desc
.code path-setgid-p
tests whether
.meta path
exists and has the set-group-ID permission set.
.code path-setuid-p
tests whether
.meta path
exists and has the set-user-ID permission set.
.code path-sticky-p
tests whether
.meta path
exists and has the "sticky" permission bit set.
.coNP Functions @ path-mine-p and @ path-my-group-p
.synb
.mets (path-mine-p << path )
.mets (path-my-group-p << path )
.syne
.desc
.code path-mine-p
tests whether
.meta path
exists, and is effectively owned by the calling process; that is,
it has a user ID equal to the real user ID of the process.
.code path-my-group-p
tests whether
.meta path
exists, and is effectively owned by a group to which the calling process
belongs. This means that the group owner is either the same as the
real group ID of the calling process, or else is among the
supplementary group IDs of the calling process.
.coNP Function @ path-readable-to-me-p
.synb
.mets (path-readable-to-me-p << path )
.syne
.desc
.code path-readable-to-me-p
tests whether the calling process can read the
object named by
.metn path .
If necessary, this test examines the real user ID of the
calling process, the real group ID, and the list of supplementary groups.
.coNP Function @ path-writable-to-me-p
.synb
.mets (path-writable-to-me-p << path )
.syne
.desc
.code path-writable-to-me-p
tests whether the calling process can write the
object named by
.metn path .
If necessary, this test examines the real user ID of the
calling process, the real group ID, and the list of supplementary groups.
.coNP Function @ path-read-writable-to-me-p
.synb
.mets (path-read-writable-to-me-p << path )
.syne
.desc
.code path-readable-to-me-p
tests whether the calling process can both read and write the
object named by
.metn path .
If necessary, this test examines the real user ID of the
calling process, the real group ID, and the list of supplementary groups.
.coNP Function @ path-executable-to-me-p
.synb
.mets (path-executable-to-me-p << path )
.syne
.desc
.code path-executable-to-me-p
tests whether the calling process can execute the
object named by
.metn path ,
or perform a search (name lookup, not implying sequential readability) on it,
if it is a directory.
If necessary, this test examines the real user ID of the
calling process, the real group ID, and the list of supplementary groups.
.coNP Functions @ path-private-to-me-p and @ path-strictly-private-to-me-p
.synb
.mets (path-private-to-me-p << path )
.mets (path-strictly-private-to-me-p << path )
.syne
.desc
The
.code path-private-to-me-p
and
.code path-strictly-private-to-me-p
functions report whether the calling process can rely on the
object indicated by
.code path
to be, respectively, private or strictly private to the security context
implied by its real user ID.
"Private" means that beside the real user ID of the calling process and
the superuser, no other user ID has write access to the object, and thus its
contents may be trusted to be be free from tampering by any other user.
"Strictly private" means that not only is the object private, as above,
but users other than the real user ID of the calling process
and superuser also not not have read access.
The rules which the function applies are as follows:
A file to be examined is initially assumed to be strictly private.
If the file is not owned by the real user ID of the caller, or
else by the superuser, then it is not private.
If the file grants write permission to "others", then it is not private.
If the file grants read permission to "others", then it is not strictly
private.
If the file grants write permission to the group owner, then it is not
private if the group contains names other than that of the file owner or the
superuser.
If the file grants read permission to the group owner, then it is not
strictly private if the group contains names other than that of the file owner
or the superuser.
Note that this interpretation of "private" and "strictly private" is vulnerable
to the following time-of-check to time-of-use race condition with regard to the
group check. At the time of the check, the group might be empty or contain
only the caller as a member. But by the time the file is subsequently accessed,
the group might have been innocently extended by the system administrator to
include additional users, who can maliciously modify the file.
Also note that the function is vulnerable to a time-of-check to time-of-use
race if
.meta path
is a string rather than a
.code stat
structure. If any components of the
.meta path
are symbolic links or directories that can be manipulated by other
users, then the object named by
.meta path
file can pass the check, but can later
.meta path
can be subverted to refer to a different object.
One way to guard against this race is to open the file, then use
.code fstat
on the stream to obtain a
.code stat
structure which is then used as an argument to
.code path-private-to-me-p
or
.codn path-strictly-private-to-me-p .
.coNP Functions @ path-newer and @ path-older
.synb
.mets (path-newer < left-path << right-path )
.mets (path-older < left-path << right-path )
.syne
.desc
The
.code path-newer
function compares two paths or stat results by modification time.
It returns
.code t
if
.meta left-path
exists, and either
.meta right-path
does not exist, or has a modification time stamp in the past
relative to
.metn left-path .
The
.code path-older
function is equivalent to
.code path-newer
with the arguments reversed.
Note:
.code path-newer
takes advantage of subsecond timestamp resolution information,
if available. The implementation is based on using the
.code mtime-nsec
field of the
.code stat
structure, if it isn't
.codn nil .
.coNP Function @ path-same-object
.synb
.mets (path-same-object < left-path << right-path )
.syne
.desc
The
.code path-same-object
function returns
.code t
if
.meta left-path
and
.meta right-path
resolve to the same filesystem object: the same inode number on the same
device.
.coNP Function @ path-search
.synb
.mets (path-search < name <> [ search-path ])
.syne
.desc
The
.code path-search
function searches for the existence of a filesystem object named by
.meta name
in the directories specified
.metn search-path .
If
.meta name
is the empty string or one of the two strings
.str .
(dot)
or
.str ..
(dotdot),
then
.code nil
is returned. If
.meta name
contains any path separator characters (any of the set of characters
found in the
.code path-sep-chars
string) then the function returns
.meta name
without performing any search. In all these trivial cases, the
.meta search-path
argument is ignored.
The
.meta search-path
argument, if present, may be a string or a list of strings.
If omitted, then it takes on the value of the
.code PATH
environment variable if that variable exists, or else takes on
the value
.code nil
indicating an empty search path.
If
.meta search-path
is a string, it is converted to a list of directories by splitting on the
separator character, which may be
.code :
(colon)
or
.code ;
(semicolon) depending on the system. Then, for each directory in the list,
.code path-search
affixes the
.meta name
to that component, as if using the
.code path-cat
function, and tests whether the resulting path refers to an existing filesystem
object.
If so, then the search terminates and that resulting path is returned.
If the entire list is traversed without finding a filesystem object, then
.code nil
is returned.
If any error whatsoever occurs while determining whether the resulting path
exists, the situation is treated as nonexistence, and the search continues.
Note: subtle discrepancies may exist between
.code path-search
and the host platform's mechanisms for searching for an executable program.
For instance, since
.code path-search
is interested in existence only, it may return a path which exists, but is
not executable. Whereas a path searching implementation which tests for
executability will in that case continue searching, and not return that
path.
.SS* Unix Credentials
.coNP Functions @, getuid @, geteuid @ getgid and @ getegid
.synb
.mets (getuid)
.mets (geteuid)
.mets (getgid)
.mets (getegid)
.syne
.desc
These functions directly correspond to the POSIX C library functions
of the same name. They retrieve the real user ID, effective user ID,
real group ID and effective group ID, respectively, of the calling
process.
.coNP Functions @, setuid @, seteuid @ setgid and @ setegid
.synb
.mets (setuid << uid )
.mets (seteuid << uid )
.mets (setgid << gid )
.mets (setegid << gid )
.syne
.desc
These functions directly correspond to the POSIX C library functions
of the same name. They set the real user ID, effective user ID,
real group ID and effective group ID, respectively, of the calling
process.
On success, they return
.codn t .
On failure, they throw an exception of type
.codn system-error .
.coNP Function @ getgroups
.synb
.mets (getgroups)
.syne
.desc
The
.code getgroups
function retrieves the list of supplementary group IDs of the calling
process by calling the same-named POSIX C library function.
Whether or not the effective group ID retrieved by
.code getegid
is included in this list is system-dependent. Programs should not
depend on its presence or absence.
.coNP Function @ setgroups
.synb
.mets (setgroups << gid-list )
.syne
.desc
The
.code setgroups
function corresponds to a C library function found in some Unix
operating systems, complementary to the
.code getgroups
function. The argument to
.meta gid-list
must be a list of numeric group IDs.
If the function is successful, this list is installed as the list of
supplementary group IDs of the calling process, and the value
.code t
is returned.
On failure, it throws an exception of type
.codn system-error .
.coNP Functions @ getresuid and @ getresgid
.synb
.mets (getresuid)
.mets (getresgid)
.syne
.desc
These functions directly correspond to the POSIX C library functions
of the same names available in some Unix operating systems.
Each function retrieves a three element list of numeric IDs.
The
.code getresuid
function retrieves the real, effective and saved user ID of
the calling process.
The
.code getresgid
function retrieves the real, effective and saved group ID of
the calling process.
.coNP Functions @ setresuid and @ setresgid
.synb
.mets (setresuid < real-uid < effective-uid << saved-uid )
.mets (setresgid < real-gid < effective-gid << saved-gid )
.syne
.desc
These functions directly correspond to the POSIX C library functions of the
same names available in some Unix operating systems. They change the real,
effective and saved user ID or group ID, respectively, of the calling process.
A value of -1 for any of the IDs specifies that the ID is not to be changed.
Only privileged processes may arbitrarily change IDs to different values.
Unprivileged processes are restricted in the following way:
each of the new IDs that is replaced must have a new value which is equal to
one of the existing three IDs.
.SS* Unix Password Database
.coNP Structure @ passwd
.synb
.mets (defstruct passwd nil
.mets \ \ name passwd uid gid
.mets \ \ gecos dir shell)
.syne
.desc
The
.code passwd
structure corresponds to the C type
.codn "struct passwd" .
Objects of this struct are produced by the password database
query functions
.codn getpwent ,
.codn getpwuid ,
and
.codn getpwnam .
.coNP Functions @, getpwent @ setpwent and @ endpwent
.synb
.mets (getpwent)
.mets (setpwent)
.mets (endpwent)
.syne
.desc
The first time
.code getpwent
function is called, it returns the first password database entry.
On subsequent calls it returns successive entries.
Entries are returned as instances of the
.code passwd
structure. If the function cannot retrieve an entry for any reason,
it returns
.codn nil .
The
.code setpwent
function rewinds the database scan.
The
.code endpwent
function releases the resources associated with the scan.
.coNP Function @ getpwuid
.synb
.mets (getpwuid << uid )
.syne
.desc
The
.code getpwuid
searches the password database for an entry whose user ID field
is equal to the numeric
.metn uid .
If the search is successful, then a
.code passwd
structure representing the database entry is returned.
If the search fails,
.code nil
is returned.
.coNP Function @ getpwnam
.synb
.mets (getpwnam << name )
.syne
.desc
The
.code getpwnam
searches the password database for an entry whose user name
is equal to
.metn name .
If the search is successful, then a
.code passwd
structure representing the database entry is returned.
If the search fails,
.code nil
is returned.
.SS* Unix Group Database
.coNP Structure @ group
.synb
.mets (defstruct group nil
.mets \ \ name passwd gid mem)
.syne
.desc
The
.code group
structure corresponds to the C type
.codn "struct group" .
Objects of this struct are produced by the password database
query functions
.codn getgrent ,
.codn getgrgid ,
and
.codn getgrnam .
.coNP Functions @, getgrent @ setgrent and @ endgrent
.synb
.mets (getgrent)
.mets (setgrent)
.mets (endgrent)
.syne
.desc
The first time
.code getgrent
function is called, it returns the first group database entry.
On subsequent calls it returns successive entries.
Entries are returned as instances of the
.code passwd
structure. If the function cannot retrieve an entry for any reason,
it returns
.codn nil .
The
.code setgrent
function rewinds the database scan.
The
.code endgrent
function releases the resources associated with the scan.
.coNP Function @ getgrgid
.synb
.mets (getgrgid << gid )
.syne
.desc
The
.code getgrgid
searches the group database for an entry whose group ID field
is equal to the numeric
.metn gid .
If the search is successful, then a
.code group
structure representing the database entry is returned.
If the search fails,
.code nil
is returned.
.coNP Function @ getgrnam
.synb
.mets (getgrnam << name )
.syne
.desc
The
.code getgrnam
searches the group database for an entry whose group name
is equal to
.metn name .
If the search is successful, then a
.code group
structure representing the database entry is returned.
If the search fails,
.code nil
is returned.
.SS* Unix Password Hashing
.coNP Function @ crypt
.synb
.mets (crypt < key << salt )
.syne
.desc
The
.code crypt
function is a wrapper for the Unix C library function of the same name.
It calculates a hash over the
.meta key
and
.meta salt
arguments, which are strings. The hash is returned as a string.
The
.meta key
and
.meta salt
arguments are converted into UTF-8 prior to being passed into the underlying
platform function. The hash value is assumed to be UTF-8 and converted to
Unicode characters, though it is not expected to contain anything but 7
bit ASCII characters.
Note: the underlying C library function uses a static buffer for its return
value. The return value of the \*(TL function is a copy of that buffer.
.SS* Unix Signal Handling
On platforms where certain advanced features of POSIX signal handling are
available at the C API level, \*(TX exposes signal-handling functionality.
A \*(TX program can install a \*(TL function (such as an anonymous
.codn lambda ,
or the function object associated with a named function) as the handler for
a signal.
When that signal is delivered, \*(TX will intercept it with its own safe,
internal handler, mark the signal as deferred (in a \*(TX sense) and then
dispatch the registered function at a convenient time.
Handlers currently are not permitted to interrupt the execution of most
\*(TX internal code. Immediate, asynchronous execution of handlers is
currently enabled only while \*(TX is blocked on I/O operations or sleeping.
Additionally, the
.code sig-check
function can be used to dispatch and clear deferred
signals. These handlers are then safely called if they were subroutines of
.codn sig-check ,
and not asynchronous interrupts.
.coNP Variables @, sig-hup @, sig-int @, sig-quit @, sig-ill @, sig-trap @, sig-abrt @, sig-bus @, sig-fpe @, sig-kill @, sig-usr1 @, sig-segv @, sig-usr2 @, sig-pipe @, sig-alrm @, sig-term @, sig-chld @, sig-cont @, sig-stop @, sig-tstp @, sig-ttin @, sig-ttou @, sig-urg @, sig-xcpu @, sig-xfsz @, sig-vtalrm @, sig-prof @, sig-poll @, sig-sys @, sig-winch @, sig-iot @, sig-stkflt @, sig-io @ sig-lost and @ sig-pwr
.desc
These variables correspond to the C signal constants
.codn SIGHUP ,
.code SIGINT
and so forth.
The variables
.codn sig-winch ,
.codn sig-iot ,
.codn sig-stkflt ,
.codn sig-io ,
.code sig-lost
and
.code sig-pwr
may not be available since a system may lack the corresponding signal
constants. See notes for the function
.codn log-authpriv .
The highest signal number is 31.
.coNP Functions @ set-sig-handler and @ get-sig-handler
.synb
.mets (set-sig-handler < signal-number << handling-spec )
.mets (get-sig-handler << signal-number )
.syne
.desc
The
.code set-sig-handler
function is used to specify the handling for a signal, such
as the installation of a handler function. It updates the signal handling for
a signal whose number is
.meta signal-number
(usually one of the constants like
.codn sig-hup ,
.code sig-int
and so forth), and returns the previous value. The
.code get-sig-handler
function returns the current value.
The
.meta signal-number
must be an integer the range 1 to 31.
Initially, all 31 signal handling specifications are set to the value
.codn t .
The
.meta handling-spec
parameter may be a function. If a function is specified,
then the signal is enabled and connected to that function until another
call to
.code set-sig-handler
changes the handling for that signal.
If
.meta handling-spec
is the symbol
.codn nil ,
then the function previously associated
with the signal, if any, is removed, and the signal is disabled. For a signal
to be disabled means that the signal is set to the
.code SIG_IGN
disposition (refer to the C API).
If
.meta handling-spec
is the symbol
.codn t ,
then the function previously associated
with the signal, if any, is removed, and the signal is set to its default
disposition. This means that it is set to
.code SIG_DFL
(refer to the C API).
Some signals terminate the process if they are generated while the
handling is configured to the default disposition.
Note that the certain signals like
.code sig-quit
and
.code sig-kill
cannot be ignored or handled.
Please observe the signal documentation in the IEEE POSIX standard, and your
platform.
A signal handling function must take two arguments. It is of the form:
.mono
.mets (lambda >> ( signal << async-p ) ...)
.onom
The
.meta signal
argument is an integer indicating the signal number for which the
handler is being invoked. The
.meta asyncp-p
argument is a Boolean value.
If it is
.codn t ,
it indicates that the handler is being invoked
asynchronously\(emdirectly in a signal handling context. If it is
.codn nil ,
then it
is a deferred call. Handlers may do more things in a deferred call, such
as terminate by throwing exceptions, and perform I/O.
The return value of a handler is normally ignored. However if it invoked
asynchronously (the
.meta async-p
argument is true), then if the handler returns
a
.cod2 non- nil
value, it is understood that the handler
requesting that it be deferred. This means that the signal will be marked
as deferred, and the handler will be called again at some later
time in a deferred context, whereby
.meta async-p
is
.codn nil .
This is not guaranteed, however;
it's possible that another signal will arrive before that happens,
possibly resulting in another async call, so the handler must
be prepared to deal with an async call at any time.
If a handler is invoked synchronously, then its return value is ignored.
In the current implementation, signals do not queue. If a signal is delivered
to the process again, while it is marked as deferred, it simply stays deferred;
there is no counter associated with a signal, only a Boolean flag.
.coNP Function @ sig-check
.synb
.mets (sig-check)
.syne
.desc
The
.code sig-check
function tests whether any signals are deferred, and for each
deferred signal in turn, it executes the corresponding handler. For a signal to
be deferred means that the signal was caught by an internal handler in
\*(TX and the event was recorded by a flag. If a handler function is removed
while a signal is deferred, the deferred flag is cleared for that signal.
Calls to the
.code sig-check
function may be inserted into CPU-intensive code that
has no opportunity to be interrupted by signals, because it doesn't invoke any
I/O functions.
.coNP Function @ raise
.synb
.mets (raise << signal )
.syne
.desc
The
.code raise
function sends
.meta signal
to the process.
It is a wrapper for the C function of the same name.
The return value is
.code t
if the function succeeds, otherwise
.codn nil .
.coNP Function @ kill
.synb
.mets (kill < process-id <> [ signal ])
.syne
.desc
The
.code kill
function is used for sending a signal to a process group or process.
It is a wrapper for the POSIX
.code kill
function.
If the
.meta signal
argument is omitted, it defaults to the same value as
.codn sig-term .
The return value is
.code t
if the function succeeds, otherwise
.codn nil .
.coNP Function @ strsignal
.synb
.mets (strsignal << signal )
.syne
.desc
The
.code strsignal
function returns a character string describing the specified signal number.
It is based on the same-named POSIX C library function.
.SS* Unix Processes
.coNP Functions @ fork and @ wait
.synb
.mets (fork)
.mets (wait >> [ pid <> [ flags ]])
.syne
.desc
The
.code fork
and
.code wait
functions are interfaces to the Unix functions
.code fork
and
.codn waitpid .
The
.code fork
function creates a child process which is a replica of the parent. Both
processes return from the function. In the child process, the return value is
zero. In the parent, it is an integer representing the process ID of the child.
If the function fails to create a child, it returns
.code nil
rather than an integer. In this case, the
.code errno
function can be used to inquire about the cause.
The
.code wait
function, if successful, returns a cons cell consisting of a pair of integers.
The
.code car
of the cons is the process ID of the process or group which was successfully
waited on, and the
.code cdr
is the status. If
.code wait
fails, it returns
.codn nil .
The
.code errno
function can be used to inquire about the cause.
The
.meta process-id
argument, if not supplied, defaults to -1, which means that
.code wait
waits for any process, rather than a specific process. Certain other
values have special meaning, as documented in the POSIX standard
for the
.code waitpid
function.
The
.meta flags
argument defaults to zero. If it is specified as nonzero, it should be
a bitwise combination (via the
.code logior
function) of the variables
.codn w-nohang ,
.code w-untraced
and
.codn w-continued .
If
.code w-nohang
is used, then
.code wait
returns a cons cell whose
.code car
specifies a process ID value of zero in the situation that at least
one of the processes designated by
.code process-id
exist and are children of the calling process, but have not changed state.
In this case, the status value in the
.code cdr
is unspecified.
Status values may be inspected with the functions
.codn w-ifexited ,
.codn w-exitstatus ,
.codn w-ifsignaled ,
.codn w-termsig ,
.codn w-coredump ,
.codn w-ifstopped ,
.code w-stopsig
and
.codn w-ifcontinued .
.coNP Functions @, w-ifexited @, w-exitstatus @, w-ifsignaled @, w-termsig @, w-coredump @ w-ifstopped and @ w-stopsig
.synb
.mets (w-ifexited << status )
.mets (w-exitstatus << status )
.mets (w-ifsignaled << status )
.mets (w-termsig << status )
.mets (w-coredump << status )
.mets (w-ifstopped << status )
.mets (w-stopsig << status )
.mets (w-ifcontinued << status )
.syne
.desc
These functions analyze process exit values produced by the
.code wait
function.
They are closely based on the
POSIX macros
.codn WIFEXITED ,
.codn WEXITSTATUS ,
and so on.
The
.meta status
value is either an integer, or a cons cell. In this case, the cons
cell is expected to have an integer in its
.code cdr
which is used as the status.
The
.codn w-ifexited ,
.codn w-ifsignaled ,
.codn w-coredump ,
.code w-ifstopped
and
.code w-ifcontinued
functions have Lisp Boolean return semantics, unlike their C language
counterparts: they return
.code t
or
.codn nil ,
rather than zero or nonzero. The others return integer values.
.coNP Function @ exec
.synb
.mets (exec < file <> [ args ])
.syne
.desc
The exec function replaces the process image with the executable specified
by string argument
.metn file .
The executable is found by searching the system path.
The
.meta file
argument becomes the first argument of the executable, argument zero.
If
.meta args
is specified, it is a list of strings. These are passed as the additional
arguments of the executable.
If
.code exec
fails, an exception of type
.code file-error
is thrown.
.coNP Function @ exit*
.synb
.mets (exit* << status )
.syne
.desc
The
.code exit*
function terminates the entire process (running \*(TX image), specifying
the termination status to the operating system. The
.meta status
argument is treated exactly like that of the
.code exit
function. Unlike that function, this one exits the process immediately,
cleaning up only low-level operating system resources such as closing file
descriptors and releasing memory mappings, without performing userspace
cleanup.
.code exit*
is implemented using a call to the POSIX function
.codn _exit .
.coNP Functions @ getpid and @ getppid
.synb
.mets (getpid)
.mets (getppid)
.syne
.desc
These functions retrieve the current process ID and the parent process ID
respectively. They are wrappers for the POSIX functions
.code getpid
and
.codn getppid .
.coNP Function @ daemon
.synb
.mets (daemon < nochdir << noclose )
.syne
.desc
This is a wrapper for the function
.code daemon
which originated in BSD Unix.
It returns
.code t
if successful,
.code nil
otherwise, and the
.code errno
variable is set in that case.
Unlike in the underlying same-named platform function, the
.meta nochdir
and
.meta noclose
arguments are Boolean, rather than integer values.
.SS* Unix File Descriptors
.coNP Function @ open-fileno
.synb
.mets (open-fileno < file-descriptor <> [ mode-string ])
.syne
.desc
The
.code open-fileno
function creates a \*(TX stream over a file descriptor. The
.meta file-descriptor
argument must be an integer denoting a valid file descriptor.
For a description of
.metn mode-string ,
see the
.code open-file
function.
.coNP Function @ fileno
.synb
.mets (fileno << stream )
.syne
.desc
The
.code fileno
function returns the underlying file descriptor of
.metn stream ,
if it has one. Otherwise, it returns
.codn nil .
This is equivalent to querying the stream using
.code stream-get-prop
for the
.code :fd
property.
.coNP Function @ dupfd
.synb
.mets (dupfd < old-fileno <> [ new-fileno ])
.syne
.desc
The
.code dupfd
function provides an interface to the POSIX functions
.code dup
or
.codn dup2 ,
when called with one or two arguments, respectively.
.coNP Function @ pipe
.synb
.mets (pipe)
.syne
.desc
The
.code pipe
function, if successful, returns a pair of integer file descriptors
as a cons-cell pair. The descriptor in the
.code car
field of the pair is the read end of the pipe.
The
.code cdr
holds the write end.
If the function fails, it throws an exception of type
.codn file-error .
.coNP Function @ close
.synb
.mets (close < fileno <> [ throw-on-error-p ])
.syne
.desc
The
.code close
function passes the integer descriptor
.meta fileno
to the POSIX
.code close
function. If the operation is successful, then
.code t
is returned. Otherwise an exception of type
.code file-error
is thrown, unless the
.meta throw-on-error-p
argument is present, with a true value. In that case,
.code close
indicates failure by returning
.codn nil .
.coNP Function @ poll
.synb
.mets (poll < poll-list <> [ timeout ])
.syne
.desc
The
.code poll
function suspends execution while monitoring one or more file descriptors
for specified events. It is a wrapper for the same-named POSIX function.
The
.meta poll-list
argument is a sequence of
.code cons
pairs. The
.code car
of each pair is either an integer file descriptor, or else a stream
object which has a file descriptor (the
.code fileno
function can be applied to that stream to retrieve a descriptor).
The
.code cdr
of each pair is an integer bit mask specifying the events, whose
occurrence the file descriptor is to be monitored for. The variables
.codn poll-in ,
.codn poll-out ,
.code poll-err
and several others are available which hold bitmask values corresponding
to the constants
.codn POLLIN ,
.codn POLLOUT ,
.code POLLERR
used with the C language
.code poll
function.
The
.meta timeout
argument, if absent, defaults to the value -1, which specifies an indefinite
wait. A nonnegative value specifies a wait with a timeout, measured in
milliseconds.
The function returns a list of pairs representing the descriptors or streams
which were successfully polled. If the function times out, it returns an
empty list. If an error occurs, an exception is thrown.
The returned list is similar in structure to the input list. However, it holds
only entries which polled positive. The
.code cdr
of every pair now holds a bitmask of the events which were to have occurred.
.coNP Function @ isatty
.synb
.mets (isatty << stream )
.mets (isatty << fileno )
.syne
.desc
The
.code isatty
function provides access to the underlying POSIX function of the same name.
If the argument is a
.meta stream
object which has a
.code :fd
property, then the file descriptor number is retrieved. The behavior is
then as if that descriptor number were passed as the
.meta fileno
argument.
If the argument is not a
.metn stream ,
it must be a
.metn fileno :
an integer in the representation range of the C type
.codn int .
The POSIX
.code isatty
is invoked on this integer. If it that returns 1, then
.code t
is returned, otherwise
.codn nil .
.SS* Unix File Control
.coNP Variables @, o-accmode @, o-rdonly @, o-wronly @, o-rdwr @, o-creat @, o-noctty @, o-trunc @, o-append @, o-nonblock @, o-sync @, o-async @, o-directory @, o-nofollow @, o-cloexec @, o-direct @ o-noatime and @ o-path
.desc
These variables correspond to the POSIX file mode constants
.codn O_ACCMODE ,
.codn O_RDONLY ,
.codn O_WRONLY ,
.codn O_RDWR ,
.codn O_CREAT ,
.codn O_NOCTTY ,
and so forth.
The availability of the variables
.codn o-async ,
.codn o-directory ,
.codn o-nofollow ,
.codn o-cloexec ,
.codn o-direct ,
.code o-noatime
and
.code o-path
depends on the host platform.
Some of these flags may be set or cleared on an existing file descriptor
using the
.code f-setfl
command of the
.code fcntl
function, in accordance with POSIX and the host platform documentation.
.coNP Variables @, seek-set @ seek-cur and @ seek-end
.desc
These variables correspond to the ISO C constants
.codn SEEK_SET ,
.code SEEK_CUR
and
.codn SEEK_END .
These values, usually associated with the ISO C
.code fseek
function, are also used in the
.code fcntl
file locking interface as values of the
.code whence
member of the
.code flock
structure.
.coNP Variables @, f-dupfd @, f-dupfd-cloexec @, f-getfd @, f-setfd @, f-getfl @, f-setfl @, f-getlk @ f-setlk and @ f-setlkw
.desc
These variables correspond to the POSIX
.code fcntl
command constants
.codn F_DUPFD ,
.codn F_GETFD ,
.codn F_SETFD ,
and so forth. Availability of the
.code f-dupfd-cloexec
depends on the host platform.
.coNP Variable @ fd-cloexec
.desc
The
.code fd-cloexec
variable corresponds to the POSIX
.code FD_CLOEXEC
constant. It denotes the flag which may be set by the
.code fd-setfd
command of the
.code fcntl
function.
.coNP Variables @, f-rdlck @ f-wrlck and @ f-unlck
.desc
These variables correspond to the POSIX lock type constants
.codn F_RDLCK ,
.code F_WRLCK
and
.codn F_UNLCK .
They specify the possible values of the
.code type
field of the
.code flock
structure.
.coNP Structure @ flock
.synb
.mets (defstruct flock nil
.mets \ \ type whence
.mets \ \ start len
.mets \ \ pid)
.syne
.desc
The
.code flock
structure corresponds to the POSIX structure of the same name.
An instance of this structure must be specified as the third
argument of the
.code fcntl
function when the
.meta command
argument is one of the values
.codn f-getlk ,
.code f-setlk
or
.codn f-setlkw .
All slots must be initialized with appropriate values before
calling
.code fcntl
with the exception that the
.code f-getlk
command does not access the existing value of the
.code pid
slot.
.coNP Function @ fcntl
.synb
.mets (fcntl < fileno < command <> [ arg ])
.syne
.desc
The
.code fcntl
function corresponds to the same-named POSIX function.
The
.meta fileno
and
.meta command
arguments must be integers.
The \*(TL
.code fileno
restricts the
.meta command
argument to the supported values for which symbolic variable names are provided.
Other integer
.meta command
values are rejected by returning -1 and setting the
.code errno
variable to
.codn EINVAL .
Whether the third argument is required, and what type it must be, depends on the
.meta command
value. Commands not requiring the third argument ignore it if it is passed.
.code fcntl
commands for which POSIX requires an argument of type
.code long
require
.meta arg
to be an integer.
The file locking commands
.codn f-getlk ,
.code f-setlk
and
.code f-setlkw
require
.meta arg
to be a
.code flock
structure.
The
.code fcntl
function doesn't throw an error if the underlying POSIX function indicates
failure; the underlying function's return value is converted to a Lisp integer
and returned.
.SS* Unix Itimers
Itimers ("interval timers") can be used in combination with signal handling to
execute asynchronous actions. Itimers deliver delayed, one-time signals,
and also periodically recurring signals. For more information, consult the
POSIX specification.
.coNP Variables @, itimer-real @ itimer-virtual and @ itimer-prof
.desc
These variables correspond to the POSIX constants
.codn ITIMER_REAL ,
.code ITIMER_VIRTUAL
and
.codn ITIMER_PROF .
Their values are suitable as the
.meta timer
argument of the
.code getitimer
and
.code setitimer
functions.
.coNP Functions @ getitimer and @ setitimer
.synb
.mets (getitimer << timer )
.mets (setitimer < timer < interval << value )
.syne
.desc
The
.code getitimer
function returns the current value of the specified timer,
which must be
.codn itimer-real ,
.code itimer-virtual
or
.codn itimer-prof .
The current value consists of a list of two integer values, which
represents microseconds. The first value is the timer interval,
and the second value is the timer's current value.
Like
.codn getitimer ,
the
.code setitimer
function also retrieves the specified timer.
In addition, it stores a new value in the timer,
which is given by the two arguments, expressed in microseconds.
.SS* Unix Syslog
On platforms where a Unix-like syslog API is available, \*(TX exports this
interface. \*(TX programs can configure logging via the
.code openlog
function,
control the logging mask via
.code setlogmask
and generate logs via
.codn syslog ,
or using special syslog streams.
.coNP Variables @, log-pid @, log-cons @, log-ndelay @, log-odelay @ log-nowait and @ log-perror
.desc
These variables take on the values of the corresponding C preprocessor
constants from the
.code <syslog.h>
header:
.codn LOG_PID ,
.codn LOG_CONS ,
etc.
These integer values represent logging options used in the
.meta options
argument to the
.code openlog
function.
Note:
.code LOG_PERROR
is not in POSIX, and so
.code log-perror
might not be available.
See notes about
.code LOG_AUTHPRIV
in the documentation for
.codn log-authpriv .
.coNP Special variables @, log-user @, log-daemon @ log-auth and @ log-authpriv
.desc
These variables take on the values of the corresponding C preprocessor
constants from the
.code <syslog.h>
header:
.codn LOG_USER ,
.codn LOG_DAEMON ,
.code LOG_AUTH
and
.codn LOG_AUTHPRIV .
These are the integer facility codes specified in the
.code openlog
function.
Note:
.code LOG_AUTHPRIV
is not in POSIX, and so
.code log-authpriv
might not be available.
For portability use code like
.code "(or (symbol-value 'log-authpriv) 0)"
to evaluate to 0 if
.code log-authpriv
doesn't exist, or else check for its existence
using
.codn "(boundp 'log-authpriv)" .
.coNP Variables @, log-emerg @, log-alert @, log-crit @, log-err @, log-warning @, log-notice @ log-info and @ log-debug
.desc
These variables take on the values of the corresponding C preprocessor
constants from the
.code <syslog.h>
header:
.codn LOG_EMERG ,
.codn LOG_ALERT ,
etc.
These are the integer priority codes specified in the
.code syslog
function.
.coNP Special variable @ *stdlog*
.desc
The
.code *stdlog*
variable holds a special kind of stream: a syslog stream. Each
newline-terminated line of text sent to this stream becomes a log message.
The stream internally maintains a priority value that is applied
when it generates messages. By default, this value is that of
.codn log-info .
The stream holds the priority as the value of the
.code :prio
stream property, which may be changed with the
.code stream-set-prop
function.
The latest priority value which has been configured on the stream is used
at the time the newline character is processed and the log message
is generated, not necessarily the value which was in effect at the time the
accumulation of a line began to take place.
Messages sent to
.code *stdlog*
are delimited by newline characters. That is to say, each line of
text written to the stream is a new log.
.coNP Function @ openlog
.synb
.mets (openlog < id-string >> [ options <> [ facility ]])
.syne
.desc
The
.code openlog
function is a wrapper for the
.code openlog
C function, and the
arguments have the same semantics. It is not necessary to use
.code openlog
in order
to call the
.code syslog
function or to write data to
.codn *stdlog* .
The call is necessary in order to override the default identifying string, to
set options, such as having the PID (process ID) recorded in log messages, and
to specify the facility.
The
.meta id-string
argument is mandatory.
The
.meta options
argument is a bitwise mask (see the logior function) of option
values such as
.code log-pid
and
.codn log-cons .
If it is missing, then a value of 0 is
used, specifying the absence of any options.
The
.meta facility
argument is one of the values
.codn log-user ,
.code log-daemon
or
.codn log-auth .
If it is missing, then
.code log-user
is assumed.
.coNP Function @ closelog
.synb
.mets (closelog)
.syne
.desc
The
.code closelog
function is a wrapper for the C function
.codn closelog .
.coNP Function @ setlogmask
.synb
.mets (setlogmask << bitmask-integer )
.syne
.desc
The
.code setlogmask
function interfaces to the corresponding C function, and has the
same argument and return value semantics. The
.meta bitmask-integer
argument is a mask of priority
values to enable. The return value is the prior value. Note that if the
argument is zero, then the function doesn't set the mask to zero; it only
returns the current value of the mask.
Note that the priority values like
.code log-emerg
and
.code log-debug
are integer
enumerations, not bitmasks. These values cannot be combined directly to create
a bitmask. Rather, the
.code mask
function should be used on these values.
.TP* Example:
.verb
;; Enable LOG_EMERG and LOG_ALERT messages,
;; suppressing all others
(setlogmask (mask log-emerg log-alert))
.brev
.coNP Function @ syslog
.synb
.mets (syslog < priority < format << format-arg *)
.syne
.desc
The
.code syslog
function is the interface to the
.code syslog
C function. The
.code printf
formatting capabilities of the function are not used;
the
.meta format
argument follows the conventions of the \*(TL
.code format
function instead. Note in particular that
the
.code %m
convention for interpolating the value of strerror(errno) which is
available in some versions of the
.code syslog
C function is currently not supported.
Note that syslog messages are not newline-terminated.
.SS* Unix Path Globbing
On platforms where the POSIX
.code glob
function is available \*(TX provides this functionality in
the form of a like-named function, and some numeric constants.
\*(TX also provides access the
.code fnmatch
function, where available.
.coNP Variables @, glob-err @, glob-mark @, glob-nosort @, glob-nocheck @, glob-noescape @, glob-period @, glob-altdirfunc @, glob-brace @, glob-nomagic @, glob-tilde @ glob-tilde-check and @ glob-onlydir
.desc
These variables take on the values of the corresponding C preprocessor
constants from the
.code <glob.h>
header:
.codn GLOB_ERR ,
.codn GLOB_MARK ,
.codn GLOB_NOSORT ,
etc.
These values are passed as the optional second argument of the
.code glob
function. They are bitmasks and so multiple values can be combined
using the
.code logior
function.
Note that the
.codn glob-period ,
.codn glob-altdirfunc ,
.codn glob-brace ,
.codn glob-nomagic ,
.codn glob-tilde ,
.code glob-tilde-check
and
.code glob-onlydir
variables may not be available. They are extensions in the GNU C library
implementation of
.codn glob .
The standard
.code GLOB_APPEND
flag is not represented as a \*(TX variable. The
.code glob
function uses it internally when calling the C library function
multiple times, due to having been given multiple patterns.
.coNP Function @ glob
.synb
.mets (glob >> { pattern | << patterns } >> [ flags <> [ errfun ]])
.syne
.desc
The
.code glob
function is a interface to the Unix function of the same name.
The first argument must either be a single
.metn pattern ,
which is a string, or else sequence of strings specifying multiple
.metn patterns ,
which are strings.
Each string is a glob pattern: a pattern which
matches zero or more pathnames, similar to a regular expression.
The function tries to expand the patterns and return a list of strings
representing the matching pathnames in the file system.
If there are no matches, then an empty list is returned.
The optional
.meta flags
argument defaults to zero. If given, it may be a bitwise combination of the
values of the variables
.codn glob-err ,
.codn glob-mark ,
.code glob-nosort
and others. The
.code glob-append
If the
.meta errfun
argument is specified, it gives a callback function which is invoked
when
.code glob
encounters errors accessing paths. The function takes two arguments:
the pathname and the
.code errno
value which occurred for that pathname. The function's return value is
Boolean. If the function returns true, then
.code glob
will terminate.
The
.meta errfun
may terminate the traversal by a nonlocal exit, such as by throwing
an exception or performing a block return.
The
.meta errfun
may not reenter the
.code glob
function. This situation is detected and diagnosed by an exception.
The
.meta errfun
may not capture a continuation across the error boundary. That is to say,
code invoked from the error may not capture a continuation up to a prompt
which surrounds the
.code glob
call. Such an attempt is detected and diagnosed by an exception.
If a sequence of
.meta patterns
is specified instead of a single pattern,
.code glob
makes multiple calls to the underlying C library function. The
second and subsequent calls specify the
.code GLOB_APPEND
flag to add the matches to the result. The following equivalence applies:
.verb
(glob (list p0 p1 ...) f e) <--> (append (glob p0 f e)
(glob p1 f e)
...)
.brev
Details of the semantics of the
.code glob
function, and the meaning of all the
.meta flags
arguments are given in the documentation for the C function.
.coNP Variables @, fnm-pathname @, fnm-noescape @, fnm-period @, fnm-leading-dir @ fnm-casefold and @ fnm-extmatch
.desc
These variables take on the values of the corresponding C preprocessor
constants from the
.code <fnmatch.h>
header:
.codn FNM_PATHNAME ,
.codn FNM_NOESCAPE ,
.codn FNM_PERIOD ,
etc.
These values are bit masks which may be combined with the
.code logior
function to form the optional third
.meta flags
argument of the
.code fnmatch
function.
Note that the
.codn fnm-leading-dir ,
.code fnm-casefold
and
.code fnm-extmatch
functions may not be available.
They are GNU extensions, found in the GNU C library.
.coNP Function @ fnmatch
.synb
.mets (fnmatch < pattern < string <> [ flags ]])
.syne
.desc
The
.code fnmatch
function, if available, provides access
to the like-named POSIX C library function.
The
.meta pattern
argument specifies a POSIX-shell-style filename-pattern-matching expression.
Its exact features and dialect are controlled by
.metn flags .
If
.meta string
matches
.meta pattern
then
.code t
is returned. If there is no match, then
.code nil
is returned. If the C function indicates that an error has occurred,
an exception is thrown.
.SS* Unix Filesystem Traversal
On platforms where the POSIX
.code nftw
function is available \*(TX provides this functionality in
the form of the analogous Lisp function
.codn ftw ,
accompanied by some numeric constants.
Likewise, on platforms where the POSIX functions
.code opendir
and
.code readdir
are available, \*(TX provides the functionality in the form of same-named
Lisp functions, a structure type named
.code dirent
and some accompanying numeric constants.
.coNP Variables @, ftw-phys @, ftw-mount @, ftw-chdir @ ftw-depth and @ ftw-actionretval
.desc
These variables hold numeric values that may be combined into a single
bitmask bitmask value using the
.code logior
function. This value is suitable as the
.meta flags
argument of the
.code ftw
function.
These variables correspond to the C constants
.codn FTW_PHYS ,
.codn FTW_MOUNT ,
etc.
Note that
.code ftw-actionretval
is a GNU extension that is not available on all platforms. If the platform's
.code nftw
function doesn't have this feature, then this variable is not defined.
.coNP Variables @, ftw-f @, ftw-d @, ftw-dnr @, ftw-ns @, ftw-sl @ ftw-dp and @ ftw-sln
.desc
These variables provide symbolic names for the integer values that are
passed as the
.code type
argument of the callback function called by
.codn ftw .
This argument classifies the kind of file system node visited, or
error condition encountered.
These variables correspond to the C constants
.codn FTW_F ,
.codn FTW_D ,
etc.
Not all of them are present. If the underlying platform doesn't have
a given constant, then the corresponding variable doesn't exist in \*(TX.
.coNP Variables @, ftw-continue @, ftw-stop @ ftw-skip-subtree and @ ftw-skip-siblings
.desc
These variables are defined if the variable
.code ftw-actionretval
is defined.
If the value of
.code ftw-actionretval
is included in the
.meta flags
argument of
.codn ftw ,
then the callback function can use the values of these variables
as return codes. Ordinarily, the callback returns zero to continue
the search and nonzero to stop.
These variables correspond to the C constants
.codn FTW_CONTINUE ,
.codn FTW_STOP ,
etc.
.coNP Function @ ftw
.synb
.mets (ftw < path-or-list < callbackfun >> [ flags <> [ nopenfd ]])
.mets >> [ callbackfun < path < type < stat-struct < level << base ]
.syne
.desc
The
.code ftw
function provides access to the
.code nftw
POSIX C library function.
Note that the
.meta flags
and
.meta nopenfd
arguments are reversed with respect to the C language interface.
They are both optional;
.meta flags
defaults to zero, and
.meta nopenfd
defaults to 20.
The
.meta path-or-list
argument may be a string specifying the top-level pathname that
.code ftw
shall visit. Or else,
.meta path-or-list
may be a list. If it is a list, then
.code ftw
recursively invokes itself over each of the elements, taking
that element as the
.meta path-or-name
argument of the recursive call, passing down all other argument
values as-is.
The traversal stops when any recursive invocation of
.code ftw
returns a value other than
.code t
or
.codn nil ,
and that value is returned. If
.code t
or
.code nil
is returned, the traversal continues with the
application of
.code ftw
to the next list element, if any.
If the list is completely traversed, and some recursive
invocations of
.code ftw
return
.codn t ,
then the return value is
.codn t .
If all recursive invocations return
.code nil
then
.code nil
is returned.
If the list is empty,
.code t
is returned.
The
.code ftw
function walks the filesystem, as directed by the
.meta path-or-list
argument and
.meta flags
bitmask arguments.
For each visited entry, it calls the supplied
.meta callbackfun
function, which receives five arguments. If this function returns
normally, it must return either
.codn nil ,
.codn t ,
or an integer value in the range of the C type
.codn int .
The
.code ftw
function can continue the traversal by returning any non-integer value,
or the integer value zero.
If
.code ftw-actionretval
is included in the
.meta flags
bitmask, then the only integer code which continues the traversal without
any special semantics is
.code ftw-continue
and only
.code ftw-stop
stops the traversal. (Non-integer return values behave like
.codn ftw-continue ).
The
.meta path
argument of
.meta callbackfun
gives the path of the
visited filesystem object.
The
.meta type
argument is an integer code which indicates the kind of
object that is visited, or an error situation in visiting
that filesystem entry. See the documentation for
.code ftw-f
and
.code ftw-d
for possible values.
The
.meta stat-struct
argument provides information about the filesystem object
as a
.code stat
structure, the same kind of object as what is returned by the
.code stat
function.
The
.meta level
argument is an integer value representing the directory level
depth. This value is obtained from the C structure
.code FTW
in the
.code nftw
C API.
The
.meta base
argument indicates the length of the directory part of the
.code path
argument. Characters in excess of this length are thus the base name of the
visited object, and the expression
.mono
.meti >> [ path << base ..:]
.onom
calculates the base name.
The
.code ftw
function returns either
.code t
upon successful completion, or an integer value returned by
.metn callbackfun ,
as described below.
On failure it throws an exception derived from
.codn file-error ,
whose specific type is based on analyzing the POSIX
.code errno
value.
The
.meta callbackfun
may return a value of any type. If it returns a value that is not of integer
type, then zero is returned to the
.code nftw
function and traversal continues. Similarly, traversal continues
if the function returns an integer zero.
If
.meta callbackfun
returns an integer value, that value must be in the range of the C type
.codn int .
That
.code int
value is returned to
.codn nftw .
If the value is not zero, and is not -1, then
.code nftw
will terminate, and return that value, which
.code ftw
then returns. If the value is -1, then
.code nftw
is deemed to have failed, and
.code ftw
will thrown an exception of type
.codn file-error ,
whose specific type is based on analyzing the POSIX
.code errno
value. If the value is zero, then the traversal continues.
The
.meta callbackfun
may also terminate the traversal by a nonlocal exit, such as by throwing
an exception or performing a block return.
The
.meta callbackfun
may not reenter the
.code ftw
function. This situation is detected and diagnosed by an exception.
The
.meta callbackfun
may not capture a continuation across the callback boundary. That is to say,
code invoked from the callback may not capture a continuation up to a prompt
which surrounds the
.code ftw
call. Such an attempt is detected and diagnosed by an exception.
.coNP Structure @ dirent
.synb
.mets (defstruct dirent nil
.mets \ \ name ino type)
.syne
.desc
Objects of the
.code dirent
structure type are returned by the
.code readdir
function.
The
.code name
slot is a character string giving the name of the directory entry.
If the
.code opendir
function's
.meta prefix-p
argument is specified as true,
then
.code readdir
operations produce
.code dirent
structures whose
.code name
slot is a path formed by combining the directory path with the directory
entry name.
The
.code ino
slot is an integer giving the inode number of the object named by the
directory entry.
The
.code type
slot indicates the type of the object, which is an integer code. Support for
this member is platform-dependent. If the directory traversal doesn't provide
the information, then this slot takes on the
.code nil
value. In this situation, the
.code dirstat
function may be used to backfill the missing information.
.coNP Variables @, dt-blk @, dt-chr @, dt-dir @, dt-fifo @, dt-lnk @, dt-reg @ dt-sock and @ dt-unknown
.desc
These variables give the possible type code values exhibited by the
.code type
slot of the
.code dirent
structure.
If the underlying host platform does not feature a
.code d_type
field in the
.code dirent
C structure, then almost all these variables are defined anyway using the values that they
have on GNU/Linux.
These definitions are useful in conjunction with the
.code dirstat
function below.
If the host platform does does not feature a
.code d_type
field in the
.code dirent
structure, then the variable
.code dt-unknown
is not defined. Note: the application can take advantage of this this to detect
the situation, in order to conditionally define code in such a way that some
run-time checking is avoided.
.coNP Function @ opendir
.synb
.mets (opendir < dir-path <> [ prefix-p ])
.syne
.desc
The
.code opendir
function initiates a traversal of the directory object named by the
string argument
.metn dir-path ,
which must be the name of a directory. If
.code opendir
is not able to open the directory traversal, it throws an exception of type
.codn system-error .
Otherwise an object of type
.code dir
is returned, which is a directory traversal handle suitable as an argument
for the
.code readdir
function.
If the
.meta prefix-p
argument is specified and has a true value, then it indicates that
the subsequent
.code readdir
operations should produce the value of the
.code name
slot of the
.code dirent
structure by combining
.meta dir-path
with the directory entry name using the
.code path-cat
function.
.coNP Function @ readdir
.synb
.mets (readdir < dir-handle <> [ dirent-struct ])
.syne
.desc
The
.code readdir
function returns the next available directory entry from the directory
traversal controlled by
.metn dir-handle ,
which must be a
.code dir
object returned by
.codn opendir .
If no more directory entries remain, then
.code readdir
returns
.codn nil .
In this situation, the
.meta dir-handle
is also closed, as if by a call to
.codn closedir .
Otherwise, the next available directory entry is returned as a
structure object of type
.codn dirent .
The
.code readdir
function internally skips and does not report the
.str .
(dot)
and
.str ..
(dotdot) directory entries.
If the
.meta dirent-struct
argument is specified, then it must be a
.code dirent
structure, or one which has all of the required slots.
In this case,
.code readdir
stores values in that structure and returns it. If
.meta dirent-struct
is absent, then
.code readdir
allocates a fresh
.code dirent
structure.
.coNP Function @ closedir
.synb
.mets (opendir << dir-handle )
.syne
.desc
The
.code closedir
function terminates the directory traversal managed by
.metn dir-handle ,
releasing its resources.
If this has already been done before,
.code closedir
returns
.codn nil ,
otherwise it returns
.codn t .
Further
.code readdir
calls on the same
.meta dir-handle
return
.codn nil .
Note: the
.code readdir
function implicitly closes
.meta dir-handle
when the handle indicates that no more directory entries remain to be traversed.
.coNP Function @ dirstat
.synb
.mets (dirstat < dirent-struct >> [ dir-path <> [ struct ]])
.syne
.desc
The
.code dirstat
function invokes
.code lstat
on the object represented by the
.code dirent
structure
.metn dirent-struct ,
sets the
.code type
slot of the
.meta dirent-struct
accordingly, and then returns the value that
.code lstat
returned.
If the
.meta struct
argument is specified, it is passed to
.codn lstat .
The
.meta dir-path
parameter must be specified, if the
.code name
slot of
.meta dirent-struct
is a simple directory entry name, rather than the full path to the object.
In that case, the slot's value gives the effective path.
If the
.code name
slot is already a path (due to, for instance, a true value of
.meta prefix-p
having been passed to
.codn opendir )
then
.meta dir-path
must not be specified.
If
.meta dir-path
is specified, then its value is combined with the
.meta name
slot of
.meta dirent-struct
using
.code path-cat
to form the effective path.
The
.code lstat
function is invoked on the effective path, and if it succeeds,
then type information is obtained from the resulting
structure to set the value of the
.code type
slot of
.metn dirent-struct .
The same structure that was returned by
.code lstat
is then returned.
.SS* Unix Sockets
On platforms where the underlying system interface is available, \*(TX provides
a sockets library for communicating over Internet networks, or over Unix
sockets.
Stream as well as datagram sockets are supported.
The classic Version 4 of the Internet protocol is supported, as well
as IP Version 6.
Sockets are mapped to \*(TX streams. The
.code open-socket
function creates a socket of a specified type, in a particular address family.
This socket is actually a stream (always, even if it isn't used for
data transfer, but only as a passive contact point).
The functions
.codn sock-connect ,
.codn sock-bind ,
.codn sock-listen ,
.code sock-accept
and
.code sock-shutdown
are used for enacting socket communication scenarios.
Stream sockets use ordinary streams, reusing the same underlying framework
that is used for file I/O and process types.
Datagram socket streams are implemented using special datagram socket streams.
Datagram socket streams eliminate the need for operations analogous to the
.code sendto
and
.code recvfrom
socket API functions, even in server programs which handle multiple
clients. An overview of datagrams is treated in the following section,
Datagram Socket Streams.
The
.code getaddrinfo
function is provided for resolving host names and services to IPv4 and IPv6
addresses.
Several structure types are provided for representing socket addresses,
and options for
.codn getaddrinfo .
Various numeric constants are also provided:
.codn af-unix ,
.codn af-inet ,
.codn af-inet6 ,
.codn sock-stream ,
.code sock-dgram
and others.
.NP* Datagram Socket Streams
Datagram socket streams are a new paradigm unique to \*(TX which
attempts to unify the programming model of stream and datagram
sockets.
A datagram socket stream is created by the
.code open-socket
function, when the
.code sock-dgram
socket type is specified. Another way in which a datagram socket
is created is when
.code sock-accept
is invoked on a datagram socket, and returns a new socket.
I/O is performed on datagram sockets using the regular I/O functions.
None of the functions take or return peer addresses. There are no I/O
functions which are analogous to the C library
.code recvfrom
and
.code sendto
functions which are usually used for datagram programming.
Datagram I/O assumes that the datagram datagram socket is connected to a
specific remote peer, and that peer is implicitly used for all I/O.
Datagram streams solve the message framing problem by
considering a single datagram to be an entire stream. On input, a datagram
stream holds an entire datagram in a buffer. The stream ends
(experiences the EOF condition) after the last byte of this buffer
is removed by an input operation. Another datagram will be received and
buffered if the EOF condition is first explicitly cleared with the
.code clear-error
function, and then another input operation is attempted.
On output, a datagram stream gathers data into an ever-growing output buffer
which isn't subject to any automatic flushing. An explicit
.code flush-stream
operation sends the buffer contents to the connected peer as a new
datagram, and empties the buffer. Subsequent output operations prepare
data for a new datagram. The
.code close-stream
function implicitly flushes the stream in the same way, and thus also
potentially generates a datagram.
A client-style datagram stream can be explicitly connected to a peer with the
.code sock-connect
function. This is equivalent to connecting a
datagram socket using the C library
.code connect
function. Writes on the stream will be transmitted using the C library function
.codn send .
A client-style datagram stream can also be "soft-connected" to a
peer using the
.code sock-set-peer
function. Writes on the stream will transmit data using the C library function
.code sendto
to the peer address.
A datagram server program which needs
to communicate using multiple peers is implemented by means of the
.code sock-accept
function which, unlike the C library
.code accept
function, works with datagram sockets as well as stream sockets.
The server creates a datagram socket, and uses
.code sock-bind
to bind it to a local address. Optionally, it may also call
.code sock-listen
which is a no-op on datagram sockets. Supporting this function on datagram
sockets allows program code to be more easily changed between datagram and
stream operation.
The server then uses
.code sock-accept
to accept new clients. Note that this is not possible with the C
library function
.codn accept ,
which only works with stream sockets.
The
.code sock-accept
function receives a datagram from a client, and creates a new datagram
socket stream which is connected to that client, and whose input buffer
contains the received datagram. Input operations on this stream consume
the datagram. Note that clearing the EOF condition and trying to receive
another datagram is not recommended on datagram streams returned
by
.codn sock-accept ,
since they share the same underlying operating system socket, which is
not actually connected to a specific peer. The receive operation could
receive a datagram from any peer, without any indication which peer that is.
Datagram servers should issue a new
.code sock-accept
call for each client datagram, treating it as a new stream.
Datagram sockets ignore almost all aspects of the
.meta mode-string
passed in
.code open-socket
and
.codn sock-accept .
The only attribute not ignored is the buffer size specified
with a decimal digit character; however, it cannot be the
only item in the mode string. The string must be syntactically
valid, as described under the
.code open-file
function. The buffer size attribute controls the size used by
the datagram socket for receiving a datagram: the capture size.
A datagram socket has obtains a default capture size if one isn't
specified by the
.metn mode-string .
The default capture size is 65536 bytes for a datagram socket created by
.codn open-socket .
If a size is not passed to
.code sock-accept
via its
.meta mode-string
argument when it is invoked on a datagram socket,
that socket's size is used as the capture size of the
newly created datagram socket which is returned.
.coNP Structure @ sockaddr
.synb
.mets (defstruct sockaddr nil
.mets \ canonname
.mets \ (:static family nil))
.syne
.desc
The
.code sockaddr
structure represents the abstract base class for socket addresses, from which
several other types are derived:
.codn sockaddr-in ,
.code sockaddr-in6
and
.codn sockaddr-un .
It has a single static slot named
.code family
and a single instance slot
.codn canonname ,
both initialized to
.codn nil .
Note: the
.code canonname
slot is optionally set by the
.code getaddrinfo
function on address structures that it returns, if requested via the
.code ai-canonname
flag. The slot only provides information to the application, playing no
semantic role in addressing.
.coNP Structure @ sockaddr-in
.synb
.mets (defstruct sockaddr-in sockaddr
.mets \ (addr 0) (port 0) (prefix 32)
.mets \ (:static family af-inet))
.syne
.desc
The
.code sockaddr-in
address represents a socket address used in the context of networking over
IP Version 4. It may be used with sockets in the
.code af-inet
address family.
The
.code addr
slot holds an integer denoting an abstract IPv4 address. For instance the hexadecimal
integer literal constant
.code #x7F000001
or its decimal equivalent
.code 2130706433
represents the loopback address, whose familiar "dot notation" is
.codn 127.0.0.1 .
Conversion of the abstract IP address to four bytes in network order, as
required, is handled internally.
The
.code port
slot holds the TCP or UDP port number, whose value ranges from 0 to 65535.
Zero isn't a valid port; the value is used for requesting an ephemeral port number
in active connections. Zero also appears in situations when the port number isn't required:
for instance, when the
.code getaddrinfo
function is used with the aim of looking up the address of a host, without
caring about the port number.
The
.code prefix
field is set by the function
.codn inaddr-str ,
when it recognizes and parses a prefix field in the textual representation.
The
.code family
static slot holds the value
.codn af-inet .
.coNP Structure @ sockaddr-in6
.synb
.mets (defstruct sockaddr-in6 sockaddr
.mets \ (addr 0) (port 0) (flow-info 0) (scope-id 0)
.mets \ (prefix 128)
.mets \ (:static family af-inet6))
.syne
.desc
The
.code sockaddr-in6
address represents a socket address used in the context of networking over
IP Version 6. It may be used with sockets in the
.code af-inet6
address family.
The
.code addr
slot holds an integer denoting an abstract IPv6 address. IPv6 addresses are
pure binary integers up to 128 bits wide.
The
.code port
slot holds the TCP or UDP port number, whose value ranges from 0 to 65535.
In IPv6, the port number functions similarly to IPv6; see
.codn sockaddr-in .
The
.code flow-info
and
.code scope-id
are special IPv6 parameters corresponding to the
.code sin6_flowinfo
and
.code sin6_scope_id
slots of the
.code sockaddr_in6
C language structure. Their meaning and use are beyond the scope of this document.
The
.code prefix
field is set by the function
.codn in6addr-str ,
when it recognizes and parses a prefix field in the textual representation.
The
.code family
static slot holds the value
.codn af-inet6 .
.coNP Structure @ sockaddr-un
.synb
.mets (defstruct sockaddr-un sockaddr
.mets \ path
.mets \ (:static family af-unix))
.syne
.desc
The
.code sockaddr-un
address represents a socket address used for interprocess communication
within a single operating system node, using the "Unix domain" sockets
of the
.code af-unix
address family.
This structure has only one slot,
.code path
which holds the rendezvous name for connecting pairs of socket endpoints.
This name appears in the filesystem.
When the
.code sockaddr-un
structure is converted to the C structure
.codn "struct sockaddr_un" ,
the
.code path
slot undergoes conversion to UTF-8. The resulting bytes are stored in the
.code sun_path
member of the C structure. If the resulting UTF-8 byte string
is larger than the
.code sun_path
array, it is silently truncated.
Note: Linux systems have support for "abstract" names which do not appear in
the filesystem. These abstract names are distinguished by starting with a null
byte. For more information, consult Linux documentation.
This convention is supported in the
.code path
slot of the
.code sockaddr-un
structure. If
.code path
contains occurrences of the pseudo-null character U+DC00, these translate
to null bytes in the
.code sun_path
member of the corresponding C structure
.codn "struct sockaddr_un" .
For example, the path
.str "\exDC00;foo"
is valid and represents an abstract address consisting of the three bytes
.str "foo"
followed by null padding bytes.
The
.code family
static slot holds the value
.codn af-unix .
.coNP Structure @ addrinfo
.synb
.mets (defstruct addrinfo nil
.mets \ \ (flags 0) (family 0) (socktype 0))
.syne
.desc
The
.code addrinfo
structure is used in conjunction with the
.code getaddrinfo
function. If that function's
.meta hints
argument is specified, it is of this type.
The purpose of the argument is to narrow down
or possibly alter the selection of addresses which
are returned.
The
.code flags
slot holds a bitwise or combination (see the
.code logior
function) of
.code getaddrinfo
flags: values given by the variables
.codn ai-passive ,
.codn ai-numerichost ,
.codn ai-v4mapped ,
.codn ai-canonname ,
.codn ai-all ,
.code ai-addrconfig
and
.codn ai-numericserv .
These correspond to the C constants
.codn AI_PASSIVE ,
.code AI_NUMERICHOST
and so forth.
If
.code ai-canonname
is specified, then every returned address structure will have its
.code canonname
member set to a string value rather than
.codn nil .
This string is a copy of the canonical name reported by the underlying
C library function, which that function places only into the first
returned address structure.
The
.code family
slot holds an address family, which may be the value of
.codn af-unspec ,
.codn af-unix ,
.code af-inet
or
.codn af-inet6 .
The
.code socktype
slot holds, a socket type. Socket types are given
by the variables
.code sock-dgram
and
.codn sock-stream .
.coNP Function @ getaddrinfo
.synb
.mets (getaddrinfo >> [ node >> [ service <> [ hints ]]])
.syne
.desc
The
.code getaddrinfo
returns a list of socket addresses based on search criteria expressed
in its arguments.
That is to say, the returned list, unless empty, contains objects of type
.code sockaddr-in
and
.codn sockaddr-in6 .
The function is implemented directly in terms of the like-named C library
function. All parameters are optional. Omitting any argument causes a null
pointer to be passed for the corresponding parameter of the C library function.
The
.meta node
and
.meta service
parameters may be character strings which specify a host name, and service.
The contents of these strings may be symbolic, like
.str www.example.com
and
.str ssh
or numeric, like
.str 10.13.1.5
and
.strn 80 .
If an argument is given for the
.code hints
parameter, it must be of type
.codn addrinfo .
The
.meta node
and
.meta service
parameters may also be given integer arguments.
An integer argument value in either of these parameters is converted to a null
pointer when calling the C
.code getaddrinfo
function. The integer values are then simply installed into every returned
address as the IP address or port number, respectively. However, if both
arguments are numeric, then no addresses are returned, since the C library
function is then called with a null node and service.
.coNP Variables @, af-unix @ af-inet and @ af-inet6
.desc
These variables hold integers which give the values of address
families. They correspond to the C constants
.codn AF_UNIX ,
.code AF_INET
and
.codn AF_INET6 .
Address family values are used in the
.meta hints
argument of the
.code getaddrinfo
function, and in the
.code open-socket
function.
Note that unlike the C language socket addressing structures,
the \*(TX socket addresses do not contain an address family slot.
That is because they indicate their family via their type.
That is to say, an object of type
.code sockaddr-in
is an address which is implicitly associated with the
.code af-inet
family via its type.
.coNP Variables @ sock-stream and @ sock-dgram
.desc
These variables hold integers which give the values of address
families. They correspond to the C constants
.code SOCK_STREAM
and
.codn SOCK_DGRAM .
.coNP Variables @, ai-passive @, ai-numerichost @, ai-v4mapped @, ai-all @ ai-addrconfig and @ ai-numericserv
.desc
These variables hold integers which are bitmasks that combine
together via bitwise or, to express the
.code flags
slot of the
.code addrinfo
structure. They correspond to the C constants
.codn AI_PASSIVE ,
.codn AI_NUMERICHOST ,
.code AI_V4MAPPED
and so forth. They influence the behavior of the
.code getaddrinfo
function.
.coNP Variables @, inaddr-any @, inaddr-loopback @ in6addr-any and @ in6addr-loopback
.desc
These integer-valued variables provide constants for commonly used IPv4
and IPv6 address values.
The value of
.code inaddr-any
and
.code in6addr-any
is zero. This address is used in binding a passive socket to all of the
external interfaces of a host, so that it can accept connections or datagrams
from all attached networks.
The
.code inaddr-loopback
variable is IPv4 loopback address, the same integer as the hexadecimal
constant
.code #x7F000001.
The
.code in6addr-loopback
is the IPv6 loopback address. Its value is 1.
.TP* Example:
.verb
;; Construct an IPv6 socket address suitable for binding
;; a socket to the loopback network, port 1234:
(new sockaddr-in6 addr in6addr-loopback port 1234)
;; Mistake: IPv4 address used with IPv6 sockaddr.
(new sockaddr-in6 addr inaddr-loopback)
.brev
.coNP Function @ open-socket
.synb
.mets (open-socket < family < type <> [ mode-string ])
.syne
.desc
The
.code open-socket
function creates a socket, which is a kind of stream.
The
.meta family
parameter specifies the address family of the socket. One of the
values
.codn af-unix ,
.code af-inet
or
.code af-inet6
should be used to create a Unix domain, Internet IPv4 or Internet IPv6
socket, respectively.
The
.meta type
parameter specifies the socket type, either
.code sock-stream
(stream socket) or
.code sock-dgram
(datagram socket).
The
.meta mode-string
specifies several properties of the stream; for a description of
.meta mode-string
parameters, refer to the
.code open-file
function. Note that the defaulting behavior for an omitted
.meta mode-string
argument is different under
.code open-socket
from other functions. Because sockets are almost always used for bidirectional
data flow, the default mode string is
.str r+b
rather than the usual
.strn r .
The rationale for including the
.str b
flag in the default mode string is that network protocols are usually defined
in a way that is independent of machine and operating system, down to the byte
level, even when they are textual. It doesn't make sense for the same \*(TX
program to see a network stream differently based on what platform it is
running on. Line-ending conversion has to do with how a platform locally stores
text files, whereas network streams are almost always external formats.
Like other stream types, stream sockets are buffered and marked as
non-real-time streams. Specifying the
.str i
mode in
.meta mode-string
marks a socket as a real-time stream, and, if it is opened for writing
or reading and writing, changes it to use line buffering.
.coNP Function @ open-socket-pair
.synb
.mets (open-socket-pair < family < type <> [ mode-string ])
.syne
.desc
The
.code open-socket-pair
function provides an interface to the functionality of the
.code socketpair
C library function.
If successful, it creates and returns a list of two stream objects,
which are sockets that are connected together.
Note: the Internet address families
.code af-inet
and
.code af-inet6
are not supported.
The
.code mode-string
is applied to each stream. For a description, see
.code open-socket
and
.codn open-file .
.coNP Functions @ sock-family and @ sock-type
.synb
.mets (sock-family << socket )
.mets (sock-type << socket )
.syne
.desc
These functions retrieve the integer values representing the address family
and type of a socket. The argument to the
.meta socket
parameter must be a socket stream or a file or process stream. For a file stream,
both functions return
.codn nil .
An exception of type
.code type-error
is thrown for other stream types.
.coNP Accessor @ sock-peer
.synb
.mets (sock-peer << socket )
.mets (set (sock-peer << socket ) << address )
.syne
.desc
The
.code sock-peer
function retrieves the peer address
has most recently been assigned to
.metn socket .
Sockets which are not connected initially
have a peer address value of
.codn nil .
A socket which is connected to a remote peer
receives that peer's address as its
.codn sock-peer .
If a socket is connected to a remote peer via
a successful use of the
.code sock-connect
function, then its
.code sock-peer
address is set to match that of the peer.
Sockets returned by the
.code sock-accept
function are connected, and have the remote endpoint address as their
.code sock-peer
address.
Assigning an address to a
.code sock-peer
form is equivalent to using
.code sock-set-peer
to set the address.
Implementation note: the
.code sock-peer
function does not use the
.code getpeername
C library function; the association between a stream and
.code sockaddr
struct is maintained by \*(TX.
.coNP Function @ sock-set-peer
.synb
.mets (sock-set-peer < socket << address )
.syne
.desc
The
.code sock-set-peer
function stores
.meta address
into
.meta socket
as that socket's peer.
Subsequently, the
.code sock-peer
function will retrieve that address.
If
.meta address
is not an appropriate address object in the address family of
.metn socket ,
the behavior is unspecified.
.coNP Function @ sock-connect
.synb
.mets (sock-connect < socket < address <> [ timeout-usec ])
.syne
.desc
The
.code sock-connect
function connects a socket stream to a peer address.
The
.meta address
argument must be a
.code sockaddr
object of type matching the address family of the socket.
If the operation fails, an exception of type
.code socket-error
is thrown. Otherwise, the function returns
.metn socket .
If the
.meta timeout-usec
argument is specified, it must be a fixnum integer.
It denotes a connection timeout period in microseconds.
If the connection doesn't succeed within the specified timeout,
an exception of type
.code timeout-error
is thrown.
.coNP Function @ sock-bind
.synb
.mets (sock-bind < socket << address )
.syne
.desc
The
.code sock-bind
function binds a socket stream to a local address
after enabling the socket stream's
.code so-reuseaddr
option.
The
.meta address
argument must be a
.code sockaddr
object of type matching the address family of the socket.
If the operation fails, an exception of type
.code socket-error
is thrown. Otherwise, the function returns
.codn t .
.coNP Function @ sock-listen
.synb
.mets (sock-listen < socket <> [ backlog ])
.syne
.desc
The
.code sock-listen
function prepares
.meta socket
for listening for connections. The
.meta backlog
parameter, if specified, requires an integer
argument. The default value is 16.
.coNP Function @ sock-accept
.synb
.mets (sock-accept < socket >> [ mode-string <> [ timeout-usec ]])
.syne
.desc
The
.code sock-accept
function waits for a client connection on
.metn socket ,
which must have been prepared for listening for
connections using
.code sock-bind
and
.codn sock-listen .
If the operation fails, an exception of type
.code socket-error
is thrown. Otherwise, the function returns a new socket
which is connected to the remote peer.
The peer's address may be retrieved from this socket using
.codn sock-peer .
The
.code mode-string
parameter is applied to the new socket just like the
similar argument in
.codn open-socket .
It defaults to
.strn r+b .
If the
.meta timeout-usec
argument is specified, it must be a fixnum integer.
It denotes a timeout period in microseconds.
If no peer connects for the specified timeout,
.code sock-accept
throws an exception of type
.codn timeout-error .
.coNP Variables @, shut-rd @ shut-wr and @ shut-rdwr
.desc
The values of these variables are useful as the second argument to the
.code sock-shutdown
function.
.coNP Function @ sock-shutdown
.synb
.mets (sock-shutdown < sock <> [ direction ])
.syne
.desc
The
.code sock-shutdown
function indicates that no further communication is to take place on
.meta socket
in the specified direction(s).
If the operation fails, an exception of type
.code socket-error
is thrown. Otherwise, the function returns
.codn t .
The
.code direction
parameter is one of the values given by the variables
.codn shut-rd ,
.code shut-wr
or
.codn shut-rdwr .
These values shut down communication in the read direction, write direction,
or both directions, respectively.
If the argument is omitted,
.code sock-shutdown
defaults to closing the write direction.
Notes: shutting down is most commonly requested in the write direction, to perform
a "half close". The communicating process thereby indicates that it has written
all the data which it intends to write. When the shutdown action is processed on the
remote end, that end is unblocked from waiting on any further data, and
effectively experiences an "end of stream" condition on its own socket or
socket-like endpoint, while continuing to be able to transmit data.
Shutting down in the reading direction is potentially abrupt. If it is executed
before an "end of stream" indication is received from a peer, it results in an
abortive close.
.coNP Functions @ sock-recv-timeout and @ sock-send-timeout
.synb
.mets (sock-recv-timeout < sock << usec )
.mets (sock-send-timeout < sock << usec )
.syne
.desc
The
.code sock-recv-timeout
and
.code sock-send-timeout
functions configure, respectively, receive and send timeouts on socket
.metn sock .
The
.meta usec
parameter specifies the value, in microseconds. It must be a
.code fixnum
integer.
When a receive timeout is configured on a socket, then an
exception of type
.code timeout-error
is thrown when an input operation waits for at least
.code usec
microseconds without receiving input.
Similarly, when a send timeout is configured, then an
exception of type
.code timeout-error
is thrown when an output operation waits for at least
.code usec
microseconds for the availability of buffer space in the socket.
.coNP Variables @, sol-socket @, ipproto-ip @, ipproto-ipv6 @ ipproto-tcp and @ ipproto-udp
.desc
These variables represent the protocol levels of socket options and are
suitable for use as the
.meta level
argument of the
.code sock-opt
and
.code sock-set-opt
functions.
The variables correspond to the POSIX C constants
.codn SOL_SOCKET ,
.codn IPPROTO_IP ,
.codn IPPROTO_IPV6 ,
.code IPPROTO_TCP
and
.codn IPPROTO_UDP .
.coNP Variables @, so-acceptconn @, so-broadcast @, so-debug @, so-dontroute @, so-error @, so-keepalive @, so-linger @, so-oobinline @, so-rcvbuf @, so-rcvlowat @, so-rcvtimeo @, so-reuseaddr @, so-sndbuf @, so-sndlowat @ so-sndtimeo and @ so-type
.desc
These variables represent socket options at the
.code sol-socket
protocol level and are suitable for use as the
.meta option
argument of the
.code sock-opt
and
.code sock-set-opt
functions.
The variables correspond to the POSIX C constants
.codn SO_ACCEPTCONN ,
.codn SO_BROADCAST ,
.codn SO_DEBUG ,
etc.
Note that the
.code sock-recv-timeout
and
.code sock-send-timeout
are a more convenient interface for setting the value of the
.code so-rcvtimeo
and
.code so-sndtimeo
socket options.
.coNP Variables @, ipv6-join-group @, ipv6-leave-group @, ipv6-multicast-hops @, ipv6-multicast-if @, ipv6-multicast-loop @ ipv6-unicast-hops and @ ipv6-v6only
.desc
These variables represent socket options at the
.code ipproto-ipv6
protocol level and are suitable for use as the
.meta option
argument of the
.code sock-opt
and
.code sock-set-opt
functions.
The variables correspond to the POSIX C constants
.codn IPV6_JOIN_GROUP ,
.codn IPV6_LEAVE_GROUP ,
.codn IPV6_MULTICAST_HOPS ,
etc.
.coNP Variable @ tcp-nodelay
.desc
This variable represents a socket option at the
.code ipproto-tcp
protocol level and is suitable for use as the
.meta option
argument of the
.code sock-opt
and
.code sock-set-opt
functions.
The variable corresponds to the POSIX C constant
.codn TCP_NODELAY .
.coNP Accessor @ sock-opt
.synb
.mets (sock-opt < socket < level < option <> [ ffi-type ])
.mets (set (sock-opt < socket < level < option <> [ ffi-type ]) << value )
.syne
.desc
The
.code sock-opt
function retrieves the value of the specified socket option,
at the specified protocol level,
associated with
.codn socket ,
which must be a socket stream.
The
.code level
argument should be one of the protocol levels
.codn sol-socket ,
.codn ipproto-ip ,
.codn ipproto-ipv6 ,
.code ipproto-tcp
and
.codn ipproto-udp .
The
.code option
argument should be one of the socket options
.codn so-acceptconn ,
.codn so-broadcast ,
.codn so-debug ,
\&...,
.codn ipv6-join-group ,
\&...,
.code ipv6-v6only
and
.codn tcp-nodelay .
The
.meta ffi-type
argument, which must be a compiled FFI type,
specifies the type of the socket option's value.
The type is most commonly
.code int
or
.codn uint ,
but it can be any other fixed-size type, including
.codn struct s.
(Variable-size types, such as C
.code char
arrays, are unsupported.)
The
.meta ffi-type
argument defaults to
.codn "(ffi int)" .
Assigning a value to a
.code sock-opt
place is equivalent to calling
.code sock-set-opt
with that value.
Note: the
.code sock-opt
and
.code sock-set-opt
functions call the POSIX C
.code getsockopt
and
.code setsockopt
functions, respectively.
Consult the POSIX specification for more information about these
functions and in particular the various socket options
(and the types they require).
.coNP Function @ sock-set-opt
.synb
.mets (sock-set-opt < socket < level < option < value <> [ ffi-type ])
.syne
.desc
The
.code sock-set-opt
function sets the value of the specified socket option,
at the specified protocol level,
associated with
.codn socket ,
which must be a socket stream.
See the documentation of the
.code sock-opt
function for a description of the
.metn level ,
.meta option
and
.meta ffi-type
arguments.
Like the
.code sock-opt
function,
.codn sock-set-opt 's
.meta ffi-type
argument defaults to
.codn "(ffi int)" .
.coNP Functions @ str-inaddr and @ str-in6addr
.synb
.mets (str-inaddr address <> [ port ])
.mets (str-in6addr address <> [ port ])
.syne
.desc
The
.code str-inaddr
and
.code str-in6addr
functions convert an IPv4 and IPv6 address, respectively, to textual
notation which is returned as a character string.
The conversion is done in conformance with RFC 5952, section 4.
IPv6 addresses representing IPv6-mapped IPv4 addresses are printed
in the hybrid notation exemplified by
.codn ::ffff:192.168.1.1 .
The
.meta address
parameter must be a nonnegative integer in the appropriate range
for the address type.
If the
.meta port
number argument is supplied, it is included in the returned character string,
according to the requirements in section 6 of RFC 5952 pertaining to IPv6
addresses (including IPv6-mapped IPv6 addresses) and section 3.2.3 of RFC 3986
for IPv4 addresses. In brief, IPv6 addresses with ports are expressed as
.code [address]:port
and IPv6 addresses follow the traditional
.code address:port
pattern.
.coNP Functions @ str-inaddr-net and @ str-in6addr-net
.synb
.mets (str-inaddr-net < address <> [ width ])
.mets (str-in6addr-net < address <> [ width ])
.syne
.desc
The functions
.code str-inaddr-net
and
.code str-in6addr-net
convert, respectively, IPv4 and IPv6 network prefix addresses to
the "slash notation". For IPv6 addresses, the requirements of section
2.3 of RFC 4291 are implemented. For IPv4, section 3.1 of RFC 4632 is followed.
The condensed portion of the IP address is always determined by measuring
the contiguous extent of all-zero bits in the least significant position
of the address. For instance an IPv4 address which has at least 24 zero bits
in the least significant position, so that the only nonzero bits are in the
highest octet, is always condensed to a single decimal number: the value of
the first octet.
If the
.meta width
parameter is specified, then its value is incorporated into the returned
textual notation as the width. No check is made whether this width
large enough to span all of the nonzero bits in the address.
If
.meta width
is omitted, then it is calculated as the number of bits in the address,
excluding the contiguous all-zero bits in the least significant position:
how many times the address can be shifted to the right before a 1 appears
in the least significant bit.
.coNP Functions @ inaddr-str and @ in6addr-str
.synb
.mets (inaddr-str << string )
.mets (in6addr-str << string )
.syne
.desc
The
.code inaddr-str
and
.code in6addr-str
functions recover an IPv4 or IPv6 address from a textual representation.
If the parse is successful, the address is returned as, respectively, a
.code sockaddr-in
or
.code sockaddr-in6
structure.
If
.meta string
is a malformed address, due to any issue such as invalid syntax or
a numeric value being out of range, an exception is thrown.
The
.code inaddr-str
function recognizes the dot notation consisting of four decimal numbers
separated by period characters. The numbers must be in the range 0 to 255.
Note: superfluous leading zeros are permitted, though this is a nonstandard
extension; not all implementations of this notations support this.
A prefix may be specified in the notation as a slash followed by a decimal
number, in the range 0 to 32. In this case, the integer value of the
prefix appears as the
.code prefix
member of the returned
.code sockaddr-in
structure. Furthermore, the address is masked, so that any bits not
included in the prefix are zero. For instance, the address
.str 255.255.255.255/1
is equivalent to
.strn 128.0.0.0 ,
except that the
.code prefix
if the returned structure is 1 rather than 32.
When a prefix is not specified, the
.code prefix
member of the structure retains its default value of 32.
When the prefix is specified, the address part need not contain all four
octets; it may contain between one and four octets. Thus,
.str 192.168/16
is a valid address, equivalent to
.strn 192.168.0.0/16 .
A port number may be specified in the notation as a colon, followed by a
decimal number in the range 0 to 65535. The integer value of this port
number appears as the
.code port
member of the returned structure. An example of this notation is
.strn 127.0.0.1:23 .
A prefix and port number may both be specified; in this case the prefix must
appear first, followed by the port number. For example,
.strn "127/8:23" .
The
.code in6addr-str
function recognizes the IPv6 notation consisting of 16-bit hexadecimal pieces
separated by colons. If the operation is successful, it returns a
.code sockaddr-in6
structure. Each piece must be a value in the range 0 to FFFF.
The hexadecimal digits may be any mixture of uppercase and lowercase. Leading
zeros are permitted.
Up to eight such pieces must be specified. If fewer pieces are specified,
then the token
.code ::
(double colon)
must appear in the address exactly once. That token denotes the condensation of
a sufficient number of zero-valued pieces to make eight pieces.
The token must be in one of three positions: it may be the leftmost element of
the address, immediately followed by a hexadecimal piece; it may be the rightmost element
of the address, preceded by a hexadecimal piece; or else, it may be in the
middle of the address, flanked on both sides by hexadecimal pieces.
The
.code in6addr-str
also recognizes the special notation for IPv6-mapped IPv4 addresses. This
notation consists of the address string
.str ::FFFF
which may appear in any uppercase/lowercase mixture, possibly with leading
zeros, followed by an IPv4 address given in the four-octet dot notation.
For example,
.strn ::FFFF:127.0.0.1 .
A prefix may be specified using a slash, followed by a decimal number in the
range 0 to 128. The handling of the prefix is similar to that of
.code inaddr-str
except that pieces of the address may not be omitted. Condensing the
pieces of the IPv6 address is always done by means of the
.code ::
token, whether or not a prefix is present. Furthermore, the octets specified in
the IPv6-mapped IPv4 notation must all be present, regardless of the prefix.
A port number may be specified in the notation as follows: the entire address,
including any slash-separated prefix, must appear surrounded in square
brackets. The closing square bracket must be followed by a colon and one or
more digits denoting a decimal number in the range 0 to 65535. For instance
.strn "[1:2:3::4/64]:1234".
.SS* Unix Terminal Control
\*(TX provides access to the terminal control "termios" interfaces defined by
POSIX, and some of the extensions to it in Linux. By using termios, programs
can control serial devices, consoles and virtual terminals. Terminal control
in POSIX revolves around a C language structure called
.codn "struct termios" .
This is mirrored in a \*(TL structure also called
.codn termios .
Like-named \*(TL functions are provided which correspond to the C functions
.codn tcgetattr ,
.codn tcsetattr ,
.codn tcsendbreak ,
.codn tcdrain ,
.code tcflush
and
.codn tcflow .
These have somewhat different argument conventions. The TTY device is specified last,
so that it can conveniently default to the
.code *stdin*
stream. A TTY device can be specified as either a stream object or a numeric
file descriptor.
The functions
.codn cfgetispeed ,
.codn cfgetospeed ,
.code cfsetispeed
and
.code cfsetospeed
are not provided, because they are unnecessary. Device speed (informally, "baud rate")
is specified directly as a integer value in the
.code termios
structure. The \*(TL termios functions automatically convert between integer
values and the speed constants (like
.codn B38400 )
used by the C API.
All of the various termios-related constants are provided, including some nonstandard
ones. They appear in lowercase. For instance
.code IGNBRK
and
.code PARENB
are simply known as the predefined \*(TL variables
.code ignbrk
and
.codn parenb .
.coNP Structure @ termios
.synb
.mets (defstruct termios nil
.mets \ \ iflag oflag cflag lflag
.mets \ \ cc ispeed ospeed)
.syne
.desc
The
.code termios
structure represents the kernel level terminal device configuration.
It holds hardware related setting such as serial line speed, parity
and handshaking. It also holds software settings like translations,
and settings affecting input behaviors. The structure closely corresponds
to the C language
.code termios
structure which exists in the POSIX API.
The
.codn iflag ,
.codn oflag ,
.code cflag
and
.code lflag
slots correspond to the
.codn c_iflag ,
.codn c_oflag ,
.code c_cflag
and
.code c_lflag
members of the C structure. They hold integer values representing
bitfields.
The
.code cc
slot corresponds to the
.code c_cc
member of the C structure. Whereas the
C structure's
.code c_cc
member is an array of the C type
.codn cc_t ,
the
.code cc
slot is a vector of integers, whose values must have the same range as the
.code cc_t
type.
.coNP Variables @, ignbrk @, brkint @, ignpar @, parmrk @, inpck @, istrip @, inlcr @, igncr @, icrnl @, iuclc @, ixon @, ixany @, ixoff @ imaxbel and @ iutf8
.desc
These variables specify bitmask values for the
.code iflag
slot of the
.code termios
structure. They correspond to the C language preprocessor symbols
.codn IGNBRK ,
.codn BRKINT ,
.code IGNPAR
and so forth.
The
.code imaxbel
and
.code iutf8
variables are specific to Linux and may not be present.
Portable code should test for their presence with
.codn boundp .
The
.code iuclc
variable is a legacy feature not found on all systems.
Note: the
.code termios
methods
.code set-iflags
and
.code clear-iflags
provide a convenient means for setting and clearing combinations of
these flags.
.coNP Variables @, opost @, olcuc @, onlcr @, ocrnl @, onocr @, onlret @, ofill @, ofdel @, vtdly @, vt0 @, vt1 @, nldly @, nl0 @, nl1 @, crdly @, cr0 @, cr1 @, cr2 @, cr3 @, tabdly @, tab0 @, tab1 @, tab2 @, tab3 @, bsdly @, bs0 @, bs1 @, ffdly @ ff0 and @ ff1
.desc
These variables specify bitmask values for the
.code oflag
slot of the
.code termios
structure. They correspond to the C language preprocessor symbols
.codn OPOST ,
.codn OLCUC ,
.code ONLCR
and so forth.
The variable
.code ofdel
is Linux-specific. Portable programs should test for its presence using
.codn boundp .
The
.code olcuc
variable is a legacy feature not found on all systems.
Likewise, whether the following groups of symbols are present is
platform-specific:
.codn nldly ,
.code nl0
and
.codn nl1 ;
.codn crdly ,
.codn cr0 ,
.codn cr1 ,
.code cr2
and
.codn cr3 ;
.codn tabdly ,
.codn tab0 ,
.codn tab1 ,
.code tab2
and
.codn tab3 ;
.codn bsdly ,
.code bs0
and
.codn bs1 ;
and
.codn ffdly ,
.code ff0
and
.codn ff1 .
Note: the
.code termios
methods
.code set-oflags
and
.code clear-oflags
provide a convenient means for setting and clearing combinations of
these flags.
.coNP Variables @, csize @, cs5 @, cs6 @, cs7 @, cs8 @, cstopb @, cread @, parenb @, parodd @, hupcl @, clocal @, cbaud @, cbaudex @ cmspar and @ crtscts
.desc
These variables specify bitmask values for the
.code cflag
slot of the
.code termios
structure. They correspond to the C language preprocessor symbols
.codn CSIZE ,
.codn CS5 ,
.code CS6
and so forth.
The following are present on Linux, and may not be available
on other platforms. Portable code should test for them using
.codn boundp :
.codn cbaud ,
.codn cbaudex ,
.code cmspar
and
.codn crtscts .
Note: the
.code termios
methods
.code set-cflags
and
.code clear-cflags
provide a convenient means for setting and clearing combinations of
these flags.
.coNP Variables @, isig @, icanon @, echo @, echoe @, echok @, echonl @, noflsh @, tostop @, iexten @, xcase @, echoctl @, echoprt @, echoke @, flusho @ pendin and @ extproc
.desc
These variables specify bitmask values for the
.code lflag
slot of the
.code termios
structure. They correspond to the C language preprocessor symbols
.codn ISIG ,
.codn ICANON ,
.code ECHO
and so forth.
The following are present on Linux, and may not be available
on other platforms. Portable code should test for them using
.codn boundp :
.codn iexten ,
.codn xcase ,
.codn echoctl ,
.codn echoprt ,
.codn echoke ,
.codn flusho ,
.code pendin
and
.codn extproc .
Note: the
.code termios
methods
.code set-lflags
and
.code clear-lflags
provide a convenient means for setting and clearing combinations of
these flags.
.coNP Variables @, vintr @, vquit @, verase @, vkill @, veof @, vtime @, vmin @, vswtc @, vstart @, vstop @, vsusp @, veol @, vreprint @, vdiscard @, vwerase @ vlnext and @ veol2
.desc
These variables specify integer offsets into the vector stored in the
.code cc
slot of the
.code termios
structure. They correspond to the C language preprocessor symbols
.codn VINTR ,
.codn VQUIT ,
.code VERASE
and so forth.
The following are present on Linux, and may not be available
on other platforms. Portable code should test for them using
.codn boundp :
.codn vswtc ,
.codn vreprint ,
.codn vdiscard ,
.code vlnext
and
.codn veol2 .
.coNP Variables @, tcooff @, tcoon @ tcioff and @ tcion
.desc
These variables hold integer values suitable as the
.meta action
argument of the
.code tcflow
function. They correspond to the C language preprocessor symbols
.codn TCOOFF ,
.codn TCOON ,
.code TCIOFF
and
.codn TCION .
.coNP Variables @, tciflush @ tcoflush and @ tcioflush
.desc
These variables hold integer values suitable as the
.meta queue
argument of the
.code tcflush
function. They correspond to the C language preprocessor symbols
.codn TCIFLUSH ,
.code TCOFLUSH
and
.codn TCIOFLUSH .
.coNP Variables @, tcsanow @ tcsadrain and @ tcsaflush
.desc
These variables hold integer values suitable as the
.meta actions
argument of the
.code tcsetattr
function. They correspond to the C language preprocessor symbols
.codn TCSANOW ,
.code TCSADRAIN
and
.codn TCSAFLUSH .
.coNP Functions @ tcgetattr and @ tcsetattr
.synb
.mets (tcgetattr <> [ device ])
.mets (tcsetattr < termios >> [ actions <> [ device ]])
.syne
.desc
The
.code tcgetattr
and
.code tcsetattr
functions, respectively, retrieve and install the configuration
of the terminal driver associated with the specified device.
These functions are wrappers for the like-named POSIX C library functions,
but with different argument conventions, and operating using
a \*(TL structure.
The
.code tcgetattr
function, if successful, returns a new instance of the
.code termios
structure.
The
.code tcsetattr
function requires an instance of a
.code termios
structure as an argument to its
.meta termios
parameter.
A program may alter the settings of a terminal device by
retrieving them using
.codn tcgetattr ,
manipulating the structure returned by this function, and
then using
.code tcsetattr
to install the modified structure into the device.
The
.meta actions
argument of
.code tcsetattr
may be given as the value of one of the variables
.codn tcsanow ,
.code tcsadrain
or
.codn tcsaflush .
If it is omitted, the default is
.codn tcsadrain .
If an argument is given for
.meta device
it must be either a stream, or an integer file descriptor.
In either case, it is expected to be associated with a
terminal (TTY) device.
If the argument is omitted, it defaults to the stream currently
stored in the
.code *stdin*
stream special variable, expected to be associated with
a terminal device.
.TP* Notes:
The C
.code termios
structure usually does not have members for representing the input
and output speed. \*(TL does not use such members, in any case, even
if they are present. The speeds are encoded in the
.code cc_iflag
and
.code cc_lflag
bitmasks. When retrieving the settings, the
.code tcgetattr
function uses the POSIX functions
.code cfgetispeed
and
.code cfgetospeed
to retrieve the speed values from the C structure. These values
are installed as the
.code ispeed
and
.code ospeed
slots of the Lisp structure. A reverse conversion takes place
when setting are installed using
.codn tcsetattr :
the speed values are taken from the slots, and installed into
the C structure using
.code cfsetispeed
and
.code cfsetospeed
before the structure is passed to the C
.code tcsetattr
function.
On Linux, TTY devices do not have a separate input and output speed.
The C
.code termios
structure stores only one speed which is taken as both the input
and output speed, with a special exception. The input speed may be
programmed as zero. In that case, it is independently represented.
speed may be programmed as zero.
.coNP Function @ tcsendbreak
.synb
.mets (tcsendbreak >> [ duration <> [ device ]])
.syne
.desc
The
.code tcsendbreak
function generates a break signal on serial devices. The
.meta duration
parameter specifies the length of the break signal in milliseconds.
If the argument is omitted, the value 500 is used.
The
.meta device
parameter is exactly the same as that of the
.code tcsetattr
function.
.coNP Function @ tcdrain
.synb
.mets (tcdrain <> [ device ])
.syne
.desc
The
.code tcdrain
function waits until all queued output on a terminal
device has been transmitted. It is a direct wrapper
for the like-named POSIX C function.
The
.meta device
parameter is exactly the same as that of the
.code tcsetattr
function.
.coNP Function @ tcflush
.synb
.mets (tcflush < queue <> [ device ])
.syne
.desc
The
.code tcflush
function discards either untransmitted output data,
or received and yet unread input data, depending on the
value of the
.meta queue
argument. It is a direct wrapper for the like-named
POSIX C function.
The
.meta queue
argument should be the value of one of the variables
.codn tciflush ,
.code tcoflush
and
.codn tcioflush ,
which specify the flushing of input data, output
data or both.
The
.meta device
parameter is exactly the same as that of the
.code tcsetattr
function.
.coNP Function @ tcflow
.synb
.mets (tcflow < action <> [ device ])
.syne
.desc
The
.code tcflow
function provides bidirectional flow control on the
specified terminal device. It is a direct wrapper
for the like-named POSIX C function.
The
.meta action
argument should be the value of one of the
variables
.codn tcooff ,
.codn tcoon ,
.code tcioff
and
.codn tcion .
The
.meta device
parameter is exactly the same as that of the
.code tcsetattr
function.
.coNP Methods @, set-iflags @, set-oflags @, set-cflags @, set-lflags @, clear-iflags @, clear-oflags @ clear-cflags and @ clear-lflags
.synb
.mets << termios .(set-iflags << flags *)
.mets << termios .(set-oflags << flags *)
.mets << termios .(set-cflags << flags *)
.mets << termios .(set-lflags << flags *)
.mets << termios .(clear-iflags << flags *)
.mets << termios .(clear-oflags << flags *)
.mets << termios .(clear-cflags << flags *)
.mets << termios .(clear-lflags << flags *)
.syne
.desc
These methods of the
.code termios
structure set or clear multiple flags of the four bitmask flag fields.
The
.meta flags
arguments specify zero or more integer values. These values
are combined together bitwise, as if by the
.code logior
function to form a single effective mask.
If there are no
.meta flags
arguments, then the effective mask is zero.
The
.code set-iflags
method sets, in the
.code iflag
slot of the
.meta termios
structure, all of the bits which
are set in the effective mask. That is to say,
the effective mask is combined with the value in
.code iflag
by a
.code logior
operation, and the result is stored back into
.codn iflag .
Similarly, the
.codn set-oflags ,
.code set-cflags
and
.code set-lflags
methods operate on the
.codn oflag ,
.code cflag
and
.code lflag
slots of the structure.
The
.code clear-iflags
method clears, in the
.code iflag
slot of the
.meta termios
structure, all of the bits which are
set in the effective mask. That is to say,
the effective mask is bitwise inverted as if
by the
.code lognot
function, and then combined with the
existing value of the
.code iflag
slot using
.codn logand .
The resulting value is stored back into the
.code iflag
slot.
Similarly, the
.codn clear-oflags ,
.code clear-cflags
and
.code clear-lflags
methods operate on the
.codn oflag ,
.code cflag
and
.code lflag
slots of the structure.
Note: the methods
.codn go-raw ,
.code go-cbreak
and
.code go-canon
are provided for changing the settings to raw, "cbreak" and canonical mode.
These methods should be preferred to directly manipulating the flag and
.code cc
slots.
.TP* Example
In this example,
.code tio
is assumed to be a variable holding an instance of a
.code termios
struct:
.verb
;; clear the ignbrk, brkint, and various other flags:
tio.(clear-iflags ignbrk brkint parmrk istrip
inlcr igncr icrnl ixon)
;; set the csize and parenb flags:
tio.(set-cflags csize parenb)
.brev
.coNP Methods @ go-raw and @ go-cbreak
.synb
.mets << termios .(go-raw)
.mets << termios .(go-cbreak)
.syne
.desc
The
.code go-raw
and
.code go-cbreak
methods of the
.code termios
structure manipulate the flag slots, as well as certain elements
of the
.code cc
slot, in order to prepare the terminal settings for, respectively,
"raw" and "cbreak" mode, described below.
Note that manipulating the
.code termios
structure doesn't actually put these settings into effect in
the terminal device; the settings represented by the structure must
be installed into the device using
.codn tcsetattr .
There is no way to reverse the effect of these methods.
To precisely restore the previous terminal settings, the program
should retain a copy of the original
.code termios
structure.
"Raw" mode refers to a configuration of the terminal device driver in which input
and output is passed transparently and without accumulation, conversion or
interpretation. Input isn't buffered into lines; as soon as a single byte is
received, it is available to the program. No special input characters such as
commands for generating an interrupt or process suspend request are processed
by the terminal driver; all characters are treated as input data. Input isn't
echoed; the only output which takes place is that generated by program
output requests to the device.
"Cbreak" mode is named after a concept and function in the "curses" terminal
control library. It refers to a configuration of the terminal device driver
which is less transparent than "raw" mode. Input isn't buffered into lines,
and line editing commands are ordinary input characters, allowing
character-at-a-time input. However, most input translations are preserved,
except that the conversion of CR characters to NL is disabled. The
signal-generating characters are processed in this mode. This latter feature of
the configuration is the likely inspiration for the word "cbreak". Unless
otherwise configured, the interrupt character corresponds to the
.key Ctrl-C
key, and "break" is another term for an interactive interruption.
.coNP Methods @ string-encode and @ string-decode
.synb
.mets << termios .(string-encode)
.mets << termios .(string-decode << string )
.syne
.desc
The
.code string-encode
method converts the terminal state stored in a
.code termios
structure into a textual format, returning that representation
as a character string.
The
.code string-decode
method parses the character representation produced by
.code string-encode
and populates the
.meta termios
structure with the settings are encoded in that string.
If a string is passed to
.code string-decode
which wasn't produced by
.codn string-encode ,
the behavior is unspecified. An exception may or may not be
thrown, and the contents of
.meta termios
may or may not be affected.
Note: the textual representation produced by
.code string-encode
is intended to be identical to that produced by the
.code -g
option of the GNU Coreutils version of the
.code stty
utility, on the same platform. That is to say, the output of
.code "stty -g"
may be used as input into
.codn string-decode ,
and the output of
.code string-encode
may be used as an argument to
.codn stty .
.SS* Unix System Identification
.coNP Structure @ utsname
.synb
.mets (defstruct utsname nil
.mets \ \ sysname nodename release
.mets \ \ version machine domainname)
.syne
.desc
The
.code utsname
structure corresponds to the POSIX structure of the same name.
An instance of this structure is returned by the
.code uname
function.
.coNP Function @ uname
.synb
.mets (uname)
.syne
.desc
The
.code uname
function corresponds to the POSIX
function of the same name. It returns an instance of the
.code utsname
structure. Each slot of the returned structure is
initialized with a character string that identifies the corresponding attribute
of the host system.
The host system might not support the reporting of the
NIS domain name. In this case, the
.code domainname
slot of the returned
.code utsname
structure will have the value
.codn nil .
.SS* Unix Resource Limits
.coNP Structure @ rlim
.synb
.mets (defstruct rlim nil
.mets \ \ cur max)
.syne
.desc
The
.code rlim
structure is required by the functions
.code getrlimit
and
.codn setrlimit .
It is analogous to the C structure by the same name described in POSIX.
.coNP Variables @, rlim-saved-max @ rlim-saved-cur and @ rlim-infinity
.desc
These variables correspond to the POSIX constants
.codn RLIM_SAVED_MAX ,
.code RLIM_SAVED_CUR
and
.codn RLIM_INFINITY .
They have the same values, and are suitable as slot values of the
.code rlim
structure.
Variables @, rlimit-core @, rlimit-cpu @, rlimit-data @, rlimit-fsize @, rlimit-nofile @ rlimit-stack and @ rlimit-as
.desc
These variables correspond to the POSIX constants
.codn RLIMIT_CORE ,
.codn RLIMIT_CPU ,
.code RLIMIT_DATA
and so forth.
.coNP Functions @ getrlimit and @ setrlimit
.synb
.mets (getrlimit < resource <> [ rlim ])
.mets (setrlimit < resource << rlim )
.syne
.desc
The
.code getrlimit
function retrieves information about the limits imposed for a particular parameter indicated by the
.meta resource
integer.
The
.code setrlimit
function changes the limit information for a resource parameter.
The
.meta resource
parameter is the value of one of the variables
.codn rlimit-core ,
.codn rlimit-cpu ,
.code rlimit-data
and so forth.
The
.meta rlim
argument is a structure of type
.codn rlim .
If this argument is given to the
.code getrlimit
function, then it fills in that structure with the retrieved
parameters. Otherwise it allocates a new structure and fills
that one. In either situation, the filled structure is returned,
if the underlying call to the host operating system is successful.
In the case of
.codn setrlimit ,
the
.code rlim
object must have non-negative integer values which are in the range
of the platform's
.code rlim_t
type.
If the underlying system call fails, then these functions throw
an exception. In the successful case, the
.code getrlimit
function returns the
.code rlim
structure, and
.code setrlimit
returns
.codn t .
Further information about resource limits is available in the POSIX standard
and platform documentation.
.SS* Unix Memory Mapping
The \*(TL interface to the POSIX
.code mmap
family of functions is based around the
.code carray
type. The
.code mmap
function returns a special variant of a
.code carray
object which keeps track of the memory mapping. When such an object
becomes unreachable and is reclaimed by garbage collection, the mapping
is automatically unmapped.
In addition to
.codn mmap ,
the functions
.codn munmap ,
.codn mprotect ,
.code madvise
and
.code msync
are provided, all taking a
.code carray
as their leftmost argument.
The \*(TL functions do not strictly follow the argument conventions of the
same-named, corresponding POSIX functions. Adjustments which are likely to
be defaulted are moved to the right.
For instance, the
.code msync
operation is often applied to the entire memory mapping. Therefore,
the first argument is the
.code carray
object which keeps track of the mapping. The second argument specifies
the flags to be applied, which constitute the last argument of the
underlying POSIX function.
The remaining two arguments are the size and offset. If these are omitted,
then
.code msync
applies to the entire region, whose address and size are known to the
.code carray
object.
Cautionary note: misuse of
.code mmap
and related functions can easily cause the \*(TX image to receive
a fatal signal due to a bad memory access. Care must be taken to prevent
such a situation, or else to catch such signals and recover.
.coNP Function @ mmap
.synb
.mets (mmap < ffi-type < length < prot < flags
.mets \ \ \ \ \ >> [ source >> [ offset <> [ addr ]]])
.syne
.desc
The
.code mmap
function provides access to the same-named POSIX platform function
for creating memory mappings. The POSIX function can be used for creating
virtual memory views of files and special devices. Views can be read-only,
and they can be mutable. They can be in such a way that changes appear
only in the mapping itself, or in such a way that the changes are actually
propagated to the mapped object itself. Mappings can be shared among
processes, providing a shared memory mechanism: for instance, if
.code fork
is called, any
.code map-shared
mappings created by the parent are shared with the child: the child
process does not get a copy of a shared mapping, but a reference to it.
The function can also be used simply for allocating memory: on some
platforms, the POSIX
.code mmap
function is used as the basis for the
.code malloc
function. It behaves as a pure allocator when asked to create a mapping which
is private, and anonymous (not backed by an object).
The \*(TL
.code mmap
function is integrated with the
.code carray
type and the FFI type system. A mapping returned by
.code mmap
is represented by a
.code carray
object.
The required
.meta ffi-type
argument specifies the element type of the array; it must be a compiled
FFI type. Note: this may be produced by the
.code ffi
macro. For instance, the type
.code int
may be specified using the expression
.codn "(ffi int)" .
The type must be a complete type suitable as the element type of an array;
a type with a zero size such as
.code void
is invalid.
The
.meta length
argument specifies the requested length of the mapping in bytes. Note that
.code mmap
allocates or configures virtual memory pages, not bytes. Internally
to the system, the
.meta length
argument is converted to a number of pages. If it specifies a fractional
number of pages, it is rounded up. For instance, if the page size is 4096
bytes, and
.meta length
is specified as 5000, it will be internally rounded up to 8192.
The returned \*(TL
.code carray
object is oblivious to this padding: it works with the given 5000-byte size.
Note: the
.code page-size
variable holds the system's page size. However, by the use of
.code mmap
extensions, it is possible for individual mappings to have their own page size.
Mixed page size virtual memory systems exist.
The
.code mmap
function determines the number of elements in the array by dividing the
.meta length
by the size of
.metn type ,
using a division that truncates toward zero. The returned
.code carray
shall have that many elements. If the division is inexact, it means that
some bytes from the underlying memory mapping are unused, even if
.code length
is a multiple of the page size.
The required
.meta prot
argument must be some bitwise combination of the portable values
.codn prot-read ,
.code prot-write
and
.codn prot-exec .
Additional system-specific
.code prot-
values may be available also for specifying additional properties. If
.meta prot
is specified as zero, then the mapping, if successfully created, may be
inaccessible:
.code prot-read
must be present to ensure read access, and
.code prot-write
to ensure write access.
The
.meta flags
argument is a bitwise combination of values given by various
.code map-
variables. At the very least, it must contain exactly one of
.code map-shared
or
.codn map-private ,
to request a shared or private mapping, respectively.
If a mapping is requested which is neither shared nor private,
the underlying POSIX function will likely fail.
If a
.meta source
is specified, indicating a filesystem object to be mapped, the
.code map-anon
flag must be omitted. Vice versa, if
.meta source
is not specified, this means that the mapping will be anonymous.
In this situation, the
.code map-anon
flag must be present.
Note: in the context of
.codn mmap ,
"anonymous" means "not associated with a filesystem object referenced by a
descriptor". It does not mean "without a name", but refers to a pure memory
allocation from virtual memory. Memory maps do not have a name, whether
anonymous or not. Moreover, the filesystem object associated with a memory map
itself does not necessarily have a name. An open file that has been deleted
from the directory structure is anonymous, yet a memory mapping can be created
using its descriptor, and that mapping is not "anonymous".
The
.meta offset
argument is used with a non-anonymous mapping. It specifies that the mapping
doesn't begin at the start of the file or file-like object, but rather at
the specified offset. The offset may not be an arbitrary integer; it must be
a multiple of the page size. Unless certain nonportable
.meta flags
are used to specify an alternative page size, the value of the
.code page-size
variable may be relied upon to indicate the page size. If an
.meta offset
is specified for an anonymous mapping, with a nonzero value, the
underlying POSIX function may indicate failure.
If the
.meta length
and
.meta offset
values cause one or more pages to be mapped which are beyond the end of the
file, then accessing those pages may produce a signal which is fatal if
not handled.
The
.meta addr
argument is used for specifying the address in conjunction with the
.code map-fixed
flag. Possibly, certain nonportable values in the
.meta flags
argument may similarly require
.metn addr .
If no bit is present in
.meta flags
which requires
.metn addr ,
then
.meta addr
should either not be specified, or specified as zero.
A nonzero value of
.meta addr
must be a multiple of the page size.
The
.code mmap
function returns a
.code carray
object if successful. Upon failure, an exception derived from
.code error
is thrown.
Note: when a
.code carray
object returned by
.code mmap
is identified by the garbage collector as unreachable, and reclaimed,
the memory mapping is unmapped. The
.code munmap
function can be invoked on the
.code carray
to release the mapping before the object becomes garbage. The
.code carray-free
function cannot be used on a mapped
.codn carray .
.coNP Function @ munmap
.synb
.mets (munmap << carray )
.syne
.desc
The
.code munmap
function releases the memory mapping tracked by
.metn carray ,
which must be an object previously returned by
.codn mmap .
An exception is thrown if the object is any other kind of
.codn carray .
Note: the memory mapping is released by means of the same-named POSIX function.
No provision is made for selectively unmapping the pages of a mapping;
the entire mapping associated with a
.meta carray
is removed.
When the memory mapping is released,
.code munmap
returns
.codn t .
Thereafter, the
.metn carray 's
contents may no longer be accessed, subject to
.code error
exceptions being thrown.
If
.code munmap
is called again on a
.code carray
on which it had previously been successfully called, the additional calls
return
.codn nil .
.coNP Functions @, mprotect @ madvise and @ msync
.synb
.mets (mprotect < carray < prot >> [ offset <> [ size ]])
.mets (madvise < carray < advice >> [ offset <> [ size ]])
.mets (msync < carray < flags >> [ offset <> [ size ]])
.syne
.desc
The functions
.codn mprotect ,
.code madvise
and
.code msync
perform various operations and adjustments on a memory mapping, using the
same-named, corresponding POSIX functions.
All functions follow the same argument conventions with regard to the
.meta carray
argument and the optional
.meta offset
and
.meta size
arguments. The respective second arguments
.metn prot ,
.meta advice
and
.meta flags
are all integers. Of these,
.meta prot
and
.meta flags
are bitmapped flags, whereas
.meta advice
specifies an enumerated command.
The
.meta prot
argument is a bitwise combination of
.code prot-
values such as
.codn prot-read ,
.code prot-write
and
.codn prot-exec .
The
.code mprotect
function adjusts the protection bits of the mapping accordingly.
The
.meta advice
argument of
.code madvise
should specify one of the following portable values, or else some
system-specific nonportable
.code madv-
value:
.codn madv-normal ,
.codn madv-random ,
.codn madv-sequential ,
.code madv-willneed
or
.codn madv-dontneed .
The
.meta flags
argument of
.code msync
should specify exactly one of the values
.code ms-async
and
.codn ms-sync .
Additional
.code ms-
values such as
.code ms-invalidate
may be combined in.
If
.meta offset
and
.meta size
are omitted, they default to zero and the size of the entire mapping, respectively,
so the operation applies to the entire mapping.
If only
.meta size
is specified, it must not exceed the mapping size, or an error exception is
thrown. The
.meta offset
argument defaults to zero.
If only
.meta offset
is specified, it must not exceed the length of the mapping, or else
an error exception is thrown. The size is calculated as the difference between
the offset and the length. It may be zero.
If both
.meta offset
and
.meta size
are specified, they must not specify a region any portion of which lies outside
the mapping. If
.meta size
is zero,
.meta offset
may be equal to the length of the mapping.
The
.meta offset
must be a multiple of the page size, or else the operation will fail,
since these functions work with virtual memory pages, and not individual
bytes. The
.meta length
is adjusted by the system to a multiple of the applicable page size,
as noted in the description of
.codn mmap .
When any of these three functions succeeds, it returns
.codn t .
Otherwise, it throws an exception.
.coNP Variables @, map-shared @, map-private @ map-anon and @ map-fixed
.desc
The integer values of these variables are bitmasks, intended to be combined with
.code logior
to prepare a value for the
.meta flags
argument of
.codn mmap .
Additional nonportable, system-dependent
.code map-
variables may be available. Their names are derived by taking the
.codn MAP_ -prefixed
symbol from the platform header file, converting it to lowercase and
replacing underscores by hyphen characters.
Any such variable which exists, but has a value of zero, is
present only for compatibility with other systems. For instance
.code map-huge-shift
may be present in non-Linux ports of \*(TX, but with a zero value; it has
a nonzero value on Linux systems to which it is specific. Applications critically
relying on certain flags should test the corresponding variables for nonzero to
make sure they are actually available.
.coNP Variables @, prot-none @, prot-read @ prot-write and @ prot-exec
.desc
The integer values of these variable are bitmasks, intended to be combined with
.code logior
to prepare a value for the
.meta prot
argument of
.code mmap
and
.codn mprotect .
Additional nonportable, system-dependent
.code prot-
variables may be available. Their names are derived by taking the
.codn PROT_ -prefixed
symbol from the platform header file, converting it to lowercase and
replacing underscores by hyphen characters.
Any such variable which exists, but has a value of zero, is
present only for compatibility with other systems.
.coNP Variables @, madv-normal @, madv-random @, madv-sequential @ madv-willneed and @ madv-dontneed
.desc
The integer values of these variable are bitmasks, intended to be combined with
.code logior
to prepare a value for the
.meta advice
argument of the
.code madvise
function.
Additional nonportable, system-dependent
.code madv-
variables may be available. Their names are derived by taking the
.codn MADV_ -prefixed
symbol from the platform header file, converting it to lower case and
replacing underscores by hyphen characters.
Any such variable which exists, but has a value of zero, is
present only for compatibility with another system.
.coNP Variables @, ms-async @ ms-sync and @ ms-invalidate
.desc
The integer values of these variable are bitmasks, intended to be combined with
.code logior
to prepare a value for the
.meta advice
argument of the
.code msync
function.
As described under
.codn msync ,
exactly one of
.code ms-async
and
.code ms-sync
should be present;
.code ms-invalidate
is optional.
.SS* Web Programming Support
.coNP Functions @ url-encode and @ url-decode
.synb
.mets (url-encode < string <> [ space-plus-p ])
.mets (url-decode < string <> [ space-plus-p ])
.syne
.desc
These functions convert character strings to and from a form which is suitable
for embedding into the request portions of URL syntax.
Encoding a string for URL use means identifying in it certain characters that
might have a special meaning in the URL syntax and representing it using
"percent encoding": the percent character, followed by the ASCII value of the
character. Spaces and control characters are also encoded, as are all byte
values greater than or equal to 127 (7F hex). The printable ASCII characters
which are percent-encoded consist of this set:
.verb
:/?#[]@!$&'()*+,;=%
.brev
More generally, strings can consists of Unicode characters, but the URL
encoding consists only of printable ASCII characters. Unicode characters in the
original string are encoded by expanding into UTF-8, and applying
percent-encoding the UTF-8 bytes, which are all in the range
.codn \ex80-\exFF .
Decoding is the reverse process: reconstituting the UTF-8 byte sequence
specified by the URL-encoding, and then decoding the UTF-8 sequence into the
string of Unicode characters.
There is an additional complication: whether or not to encode spaces as plus,
and to decode plus characters to spaces. In encoding, if spaces are not encoded
to the plus character, then they are encoded as
.codn %20 ,
since spaces are reserved
characters that must be encoded. In decoding, if plus characters are not
decoded to spaces, then they are left alone: they become plus characters in the
decoded string.
The
.code url-encode
function performs the encoding process. If the
.code space-plus-p
argument is omitted or specified as
.codn nil ,
then spaces are encoded as
.codn %20 .
If the argument is a value other than
.codn nil ,
then spaces are encoded as the
character
.code +
(plus).
The
.code url-decode
function performs the decoding process. If the
.code space-plus-p
argument is omitted or specified as
.codn nil ,
then
.code +
(plus)
characters in the
encoded data are retained as
.code +
characters in the decoded strings. Otherwise,
plus characters are converted to spaces.
.coNP Functions @, html-encode @ html-encode* and @ html-decode
.synb
.mets (html-encode << text-string )
.mets (html-decode << html-string )
.syne
.desc
The
.code html-encode
and
.code html-decode
functions convert between an HTML and raw
representation of text.
The
.code html-encode
function returns a string which is based on the content of
.metn text-string ,
but in which all characters which have special meaning in HTML
have been replaced by HTML codes for representing those characters literally.
The returned string is the HTML-encoded verbatim representation of
.metn text-string .
The
.code html-decode
function converts
.metn html-string ,
which may contain HTML
character encodings, into a string which contains the actual characters
represented by those encodings.
The function composition
.code "(html-decode (html-encode text))"
returns a string which is equal to
.codn text .
The reverse composition
.code "(html-encode (html-decode html))"
does not necessarily return a string equal to
.codn html .
For instance if html is the string
.strn "<p>Hello, world!</p>" ,
then
.code html-decode
produces
.strn "<p>Hello, world!</p>" .
From this,
.code html-encode
produces
.strn "<p>Hello, world!</p>" .
The
.code html-encode*
function is similar to
.code html-encode
except that it does not encode the single and double quote characters
(ASCII 39 and 34, respectively). Text prepared by this function may not
be suitable for insertion into a HTML template, depending on the
context of its insertion. It is suitable as text placed between
tags but not necessarily as tag attribute material.
.coNP Functions @, base64-encode @ base64-decode and @ base64-decode-buf
.synb
.mets (base64-encode >> [ string | << buf ] <> [ column-width ])
.mets (base64-decode < string)
.mets (base64-decode-buf < string)
.syne
.desc
The
.code base64-encode
function converts the UTF-8 representation of
.metn string ,
or the contents of
.metn buf ,
to Base64 and returns that representation as a string.
The Base64 encoding is described in RFC 4648, section 5.
The second argument must either be a character string, or
a buffer object.
The
.code base64-decode
functions performs the opposite conversion; it extracts the
bytes encoded in a Base64 string, and decodes them as UTF-8
to return a character string.
The
.code base64-decode-buf
extracts the bytes encoded in a Base64 string, and returns
a new buffer object containing these bytes.
The Base64 encoding divides the UTF-8 representation of
.meta string
or the bytes contained in
.meta buf
into groups of six bits, each representing the values 0 to 63. Each value is
then mapped to the characters
.code A
to
.codn Z ,
.code a
to
.codn z ,
the digits
.code 0
to
.code 9
and the characters
.code +
and
.codn / .
One or two consecutive occurrences of the character
.code =
are added as padding so that the number of
non-whitespace characters is divisible by four. These characters map to
the code 0, but are understood not to contribute to the length of the
encoded message. The
.code base64-encode
function enforces this convention, but
.code base64-decode
doesn't require these padding characters.
Base64-encoding an empty string or zero-length buffer results in an empty
string.
If the
.meta column-width
argument is passed to
.codn base64-encode ,
then the Base64 encoded string, unless empty, contains newline
characters, which divide it into lines which are
.meta column-width
long, except possibly for the last line.
.coNP Functions @ base64-stream-enc and @ base64-stream-dec
.synb
.mets (base64-stream-enc < out < in >> [ nbytes <> [ column-width ]])
.mets (base64-stream-dec < out << in )
.syne
.desc
The
.code base64-stream-enc
and
.code base64-stream-dec
perform, respectively, bulk Base64 encoding and decoding between streams.
This format is described in RFC 4648, section 5.
The
.meta in
and
.meta out
arguments must be stream objects.
The
.meta out
stream must support output. In the decode operation, it must support
byte output.
The
.meta in
stream must support input. In the encode operation it must support
byte input.
The
.code base64-stream-enc
function reads a sequence of bytes from the
.meta in
stream and writes characters to the
.meta out
stream comprising the Base64 encoding of that sequence. If the
.meta nbytes
argument is specified, it must be a nonnegative integer. At most
.meta nbytes
bytes will be read from the
.meta in
stream. If
.meta nbytes
is omitted, then the operation will read from the
.meta in
stream without limit, until that stream indicates that no more bytes
are available.
The optional
.meta column-with
argument influences the formatting of Base64 output, in the same manner
as documented for the
.code base64-encode
function.
The
.code base64-stream-dec
function reads the characters of a Base64 encoding from the
.meta in
stream and writes the corresponding byte sequence to the
.meta out
stream. It keeps reading and decoding until it encounters the end of the
stream, or a character not used in Base64: a character that is not whitespace
according to
.codn chr-isspace ,
isn't any of the Base64 coding characters (not an alphanumeric character,
and not one of the characters
.codn + ,
.code /
or
.codn = .
If the function stops due to a non-Base64 character, that character is
pushed back into the
.meta in
stream.
The
.code base64-stream-enc
function returns the number of bytes encoded;
the
.code base64-stream-dec
function returns the number of bytes decoded.
.coNP Functions @, base64url-encode @ base64url-decode and @ base64url-decode-buf
.synb
.mets (base64url-encode >> [ string | << buf ] <> [ column-width ])
.mets (base64url-decode < string)
.mets (base64url-decode-buf < string)
.syne
.desc
The
.codn base64url-encode ,
.code base64url-decode
and
.code base64url-decode-buf
functions conform, in nearly every respect, to the descriptions of,
respectively,
.codn base64-encode ,
.code base64-decode
and
.codn base64-decode-buf .
The difference is that these functions use the encoding described in
section 6 of RFC 4648, rather than section 5. This means that, in the
encoding alphabet, instead of the symbols
.code +
(plus)
and
.code /
(slash)
the symbols
.code -
(minus)
and
.code _
(underline) are used.
.coNP Functions @ base64url-stream-enc and @ base64url-stream-dec
.synb
.mets (base64url-stream-enc < out < in >> [ nbytes <> [ column-width ]])
.mets (base64url-stream-dec < out << in )
.syne
.desc
The
.code base64url-stream-enc
and
.code base64url-stream-dec
functions conform, in nearly every respect, to the descriptions of,
respectively,
.code base64-stream-enc
and
.codn base64-stream-dec .
The difference is that these functions use the encoding described in
section 6 of RFC 4648, rather than section 5. This means that, in the
encoding alphabet, instead of the symbols
.code +
(plus)
and
.code /
(slash)
the symbols
.code -
(minus)
and
.code _
(underline) are used.
.SS* Filter Module
The filter module provides a trie (pronounced "try") data structure,
which is suitable for representing dictionaries for efficient filtering.
Dictionaries are unordered collections of keys, which are strings, which
have associated values, which are also strings. A trie can be used to filter
text, such that keys appearing in the text are replaced by the corresponding
values. A trie supports this filtering operation by providing an efficient
prefix-based lookup method which only looks at each input character once, and
which does not require knowledge of the length of the key in advance.
.coNP Function @ make-trie
.synb
.mets (make-trie)
.syne
.desc
The
.code make-trie
function creates an empty trie. There is no special data type for
a trie; a trie is some existing type such as a hash table.
.coNP Function @ trie-add
.synb
.mets (trie-add < trie < key << value )
.syne
.desc
The
.code trie-add
function adds the string
.meta key
to the trie, associating
it with
.metn value .
If
.meta key
already exists in
.metn trie ,
then the value is updated with
.metn value .
The
.meta trie
must not have been compressed with
.metn trie-compress .
A trie can contain keys which are prefixes of other keys. For instance
it can contain
.str dog
and
.strn dogma .
When a trie is used for matching
and substitution, the longest match is used. If the input presents
the text
.strn doggy ,
then the match is
.strn dog .
If the input is
.strn dogmatic ,
then
.str dogma
matches.
.coNP Function @ trie-compress
.synb
.mets (trie-compress << trie )
.syne
.desc
The
.code trie-compress
function changes the representation of
.meta trie
to a representation which occupies less space and supports faster lookups.
The new representation is returned.
The compressed representation of a trie does not support the
.code trie-add
function.
The
.code trie-compress
function destructively manipulates
.metn trie ,
and may return an object
that is the same object as
.codn trie ,
or it may return a different object,
while at the same time still modifying the internals of
.metn trie .
Consequently, the program should not retain the input object
.codn trie ,
but use the returned object in its place.
.coNP Function @ trie-lookup-begin
.synb
.mets (trie-lookup-begin << trie )
.syne
.desc
The
.code trie-lookup-begin
function returns a context object for performing
an open-coded lookup traversal of a trie. The
.meta tri
argument
is expected to be a trie that was created by the
.code make-trie
function.
.coNP Function @ trie-lookup-feed-char
.synb
.mets (trie-lookup-feed-char < trie-context << char )
.syne
.desc
The
.code trie-lookup-feed-char
function performs a one character step in a trie
lookup. The
.meta trie-context
argument must be a trie context returned
by
.metn trie-lookup-begin ,
or by some previous call to
.codn trie-lookup-feed-char .
The
.meta char
argument is the next character to match.
If the lookup is successful (the match through the trie can continue
with the given character) then a new trie context object is returned.
The old trie context remains valid.
If the lookup is unsuccessful,
.code nil
is returned.
Note: determining whether a given string is stored in a trie can be
performed looking up every character of the string successively
with
.codn trie-lookup-feed-char ,
using the newly returned context
for each successive operation. If every character is found, it means
that either that exact string is found in the trie, or a prefix.
The ambiguity can be resolved by testing whether the trie has a value
at the last node using
.codn trie-value-at .
For instance, if
.str catalog
is inserted into an empty trie with value
.strn foo ,
then
.str cat
will look up successfully, being a prefix of
.strn catalog ;
however, the value at
.str cat
is
.codn nil ,
indicating that
.str cat
is only a prefix of one or more entries in the trie.
.coNP Function @ trie-value-at
.synb
.mets (trie-value-at << trie-context )
.syne
.desc
The
.code trie-value-at
function returns the value stored at the node in
in the trie given by
.metn trie-context .
Nodes which have not been given
a value hold the value
.codn nil .
.coNP Function @ filter-string-tree
.synb
.mets (filter-string-tree < filter << obj )
.syne
.desc
The
.code filter-string-tree
function returns a tree structure similar to
.meta obj
in which all of the
string atoms have been filtered through
.metn filter .
The
.meta obj
argument is a string-tree structure: either the symbol
.codn nil ,
denoting an empty structure; a string; or a list of tree structures. If
.meta obj
is
.codn nil ,
then
.code filter-string-tree
returns
.codn nil .
The
.meta filter
argument is a filter: it is either a trie, a function, or nil.
If
.meta filter
is
.codn nil ,
then
.code filter-string-trie
just returns
.metn obj .
If
.meta filter
is a function, it must be a function that can be called
with one argument. The strings of the string tree are filtered by passing
each one into the function and substituting the return value into the
corresponding place in the returned structure.
Otherwise if
.meta filter
is a trie, then this trie is used for filtering,
the string elements similarly to a function. For each string, a new
string is returned in which occurrences of the keys in the trie are
replaced by the values in the trie.
.coNP Function @ filter-equal
.synb
.mets (filter-equal < filter-1 < filter-2 < obj-1 << obj-2 )
.syne
.desc
The
.code filter-equal
function tests whether two string trees are equal
under the given filters.
The precise semantics can be given by this expression:
.mono
.mets (equal (filter-string-tree < filter-1 << obj-1 )
.mets \ \ \ \ \ \ (filter-string-tree < filter-2 << obj-2 ))
.onom
The string tree
.meta obj-1
is filtered through
.metn filter-1 ,
as if by the
.code filter-string-tree
function, and similarly,
.meta obj-2
is
filtered through
.metn filter-2 .
The resulting structures are compared
using
.codn equal ,
and the result of that is returned.
.coNP Function @ regex-from-trie
.synb
.mets (regex-from-trie << trie )
.syne
.desc
The
.code regex-from-trie
function returns a representation of
.meta trie
as regular-expression abstract syntax, suitable for
processing by the
.code regex-compile
function.
The values stored in the trie nodes are not represented in
the regular expression.
The
.meta trie
may be one that has been compressed via
.codn trie-compress ;
in fact, a compressed
.meta trie
results in more compact syntax.
Note: this function is useful for creating a compact, prefix-compressed
regular expression which matches a list of strings.
.coNP Special variable @ *filters*
.desc
The
.code *filters*
special variable holds a hash table which associates symbols with
filters. This hash table defines the named filters used in the
\*(TX pattern language. The names are the hash-table keys, and filter
objects are the values. Filter objects are one of three representations.
The value
.code nil
represents a null filter, which performs no filtering, passing the input
string through. A filter object may be a raw or compressed trie.
It may also be a Lisp function, which must be callable with one argument
of string type, and must return a string.
The application may define new filters by associating symbolic keys in
.code *filters*
with values which conform to the above representation of filters.
The behavior is unspecified if any of the predefined filters
are removed or redefined, and are subsequently used, or if the
.code *filters*
variable is replaced or rebound with a hash-table value which omits
those keys, or associates them with different values.
Note that functions
.codn html-encode ,
.code html-encode*
and
.code html-decode
use, respectively, the HTML-related
.codn :tohtml ,
.code :tohtml*
and
.codn :fromhtml .
.SS* Access To TXR Pattern Language From Lisp
It is useful to be able to invoke the abilities of the \*(TX pattern Language
from \*(TL. An interface for doing this provided in the form of the
.code match-fun
function, which is used for invoking a \*(TX pattern function.
The
.code match-fun
function has a cumbersome interface which requires the \*(TL program to
explicitly deal with the variable bindings emerging from the pattern match
in the form of an association list.
To make it the interface easier to use, \*(TX provides
the macros
.codn txr-if ,
.code txr-when
and
.codn txr-case .
.coNP Function @ match-fun
.synb
.mets (match-fun < name < args >> [ input <> [ files ]])
.syne
.desc
The
.code match-fun
function invokes a \*(TX pattern function whose name is
given by
.metn name ,
which must be a symbol.
The
.meta args
argument is a list of expressions. The expressions may be symbols
which will be interpreted as pattern variables, and may be bound or unbound.
If they are not symbols, then they are treated as expressions (of the
pattern language, not \*(TL) and evaluated accordingly.
The optional
.meta input
argument is an object of one of several types. It may be a stream,
character string or list of strings. If it is a string, then
it is converted to a list containing that string.
A list of strings represents zero or more lines of text to be
processed. If the
.meta input
argument is omitted, then it defaults to
.codn nil ,
interpreted as an empty list of lines.
The
.meta files
argument is a list of filename specifications, which follow
the same conventions as files given on the \*(TX command line. If the pattern
function uses the
.code @(next)
directive, it can process these additional files. If this argument is
omitted, it defaults to
.codn nil .
The
.code match-fun
function's return value falls into three cases. If there is a
match failure, it returns
.codn nil .
Otherwise it returns a cons cell. The
.code car
field
of the cons cell holds the list of captured bindings. The
.code cdr
of the cons cell is one of two values. If the entire input was processed, the
cdr field holds the symbol
.codn t .
Otherwise it holds another cons cell whose
.code car
is the remainder of the list of lines which were not matched, and whose
.code cdr
is the line number.
.TP* Example:
.verb
@(define foo (x y))
@x:@y
@line
@(end)
@(do
(format t "~s\en"
(match-fun 'foo '(a b)
'("alpha:beta" "gamma" "omega") nil)))
Output:
(((a . "alpha") (b . "beta")) ("omega") . 3)
.brev
In the above example, the pattern function
.code foo
is called with arguments
.codn "(a b)" .
These are unbound variables, so they correspond to parameters
.code x
and
.code y
of the function. If
.code x
and
.code y
get bound, those values propagate to
.code a
and
.codn b .
The data being matched consists of the lines
.strn alpha:beta ,
.str gamma
and
.strn omega .
Inside
.codn foo ,
.code x
and
.code y
bind to
.str alpha
and
.strn beta ,
and then the line variable binds to
.strn gamma .
The input stream is left with
.strn omega .
Hence, the return value consists of the bindings of
.code x
and
.code y
transferred to
.code a
and
.codn b ,
and the second cons cell which gives information about the rest of the
stream: it is the part starting at
.strn omega ,
which is line 3. Note that the binding for the
.code line
variable does not propagate
out of the pattern function
.codn foo ;
it is local inside it.
.coNP Function @ match-fboundp
.synb
.mets (match-fboundp << symbol )
.syne
.desc
The
.code match-fboundp
function returns
.code t
or
.code nil
if, respectively,
.meta symbol
is the name of an existing pattern function.
.coNP Macro @ txr-if
.synb
.mets (txr-if < name <> ( argument *) < input
.mets \ \ \ \ \ \ \ < then-expr <> [ else-expr ])
.syne
.desc
The
.code txr-if
macro invokes the \*(TX pattern-matching function
.meta name
on some input given by the
.meta input
parameter, whose semantics are the same as the
.meta input
argument of the
.code match-fun
function.
If
.meta name
succeeds, then
.meta then-expr
is evaluated, and if it fails,
.meta else-expr
is evaluated instead.
In the successful case,
.meta then-expr
is evaluated in a scope in which the bindings emerging from the
.meta name
function are turned into \*(TL variables.
The result of
.code txr-if
is that of
.metn then-expr .
In the failed case,
.meta else-expr
is evaluated in a scope which does not have any new bindings.
The result of
.code txr-if
is that of
.metn else-expr .
If
.meta else-expr
is missing, the result is
.codn nil .
The
.meta argument
forms supply arguments to the pattern function
.metn name .
There must be as many of these arguments as the function
has parameters.
Any argument which is a symbol is treated, for the purposes
of calling the pattern function, as an unbound pattern variable.
The function may or may not produce a binding for that variable.
Also, every argument which is a symbol also denotes a local variable
that is established around
.meta then-expr
if the function succeeds. For any such pattern variable for which the function
produces a binding, the corresponding local variable will be initialized
with the value of that pattern variable. For any such pattern variable
which is left unbound by the function, the corresponding local variable
will be set to
.codn nil .
Any
.meta argument
can be a form other than a symbol. In this situation, the argument is
evaluated, and will be passed to the pattern function as the value of
the binding for the corresponding argument.
.TP* Example:
.verb
@(define date (year month day))
@{year /\ed\ed\ed\ed/}-@{month /\ed\ed/}-@{day /\ed\ed/}
@(end)
@(do
(each ((date '("09-10-20" "2009-10-20"
"July-15-2014" "foo")))
(txr-if date (y m d) date
(put-line `match: year @y, month @m, day @d`)
(put-line `no match for @date`))))
Output:
no match for 09-10-20
match: year 2009, month 10, day 20
no match for July-15-2014
no match for foo
.brev
.coNP Macro @ txr-when
.synb
.mets (txr-when < name <> ( argument *) < input << form *)
.syne
.desc
The
.code txr-when
macro is based on
.codn txr-if .
It is equivalent to
.mono
.meti \ \ (txr-if < name <> ( argument *) < input (progn << form *))
.onom
If the pattern function
.meta name
produces a match, then each
.meta form
is evaluated in the scope of the variables established by the
.meta argument
expressions. The result of the
.code txr-when
form is that of the last
.metn form .
If the pattern function fails then the forms are not evaluated,
and the result value is
.codn nil .
.coNP Macro @ txr-case
.synb
.mets (txr-case < input-form
.mets \ \ >> {( name <> ( argument *) << form *)}*
.mets \ \ >> [( t << form *)])
.syne
.desc
The
.code txr-case
macro evaluates
.meta input-form
and then uses the value as an input to zero or more test clauses.
Each test clause invokes the pattern function named by that clause's
.meta name
argument.
If the function succeeds, then each
.meta form
is evaluated, and the value of the last
.meta form
is taken to be the result value of
.codn txr-case ,
which terminates. If there are no forms, then
.code txr-case
terminates with a
.code nil
result.
The forms are evaluated in an environment in which variables are bound
based on the
.meta argument
forms, with values depending on the result of the
invocation of the
.meta name
pattern function, in the same manner as documented in detail for the
.code txr-if
macro.
If the function fails, then the forms are not evaluated, and control passes to
the next clause.
A clause which begins with the symbol
.code t
executes unconditionally and causes
.code txr-case
to terminate. If it has no forms, then
.code txr-case
yields
.codn nil ,
otherwise the forms are evaluated in order and the value of the last
one specifies the result of
.codn txr-case .
The value of the input
.meta input-form
is expected to be one of the same kinds of objects as given by the
requirements for the
.meta input
argument of the
.code match-fun
functions.
If
.meta input-form
evaluates to a stream object according to the
.code streamp
function, then the stream is converted to a lazy list of lines,
as if by invoking the
.code get-lines
function on that stream; that list then serves as input to the clauses.
.coNP Function @ txr-parse
.synb
.mets (txr-parse >> [ source >> [ error-stream
.mets \ \ \ \ \ \ \ \ \ \ \ >> [ error-retval <> [ name ]]]])
.syne
.desc
The
.code txr-parse
function converts textual \*(TX query syntax into a Lisp data
structure representation.
The
.meta source
argument may be either a character
string, or a stream. If it is omitted, then
.code *stdin*
is used as the stream.
The
.meta source
must provide the text representation of one complete \*(TX query.
The optional
.meta error-stream
argument can be used to specify a stream to which
diagnostics of parse errors are sent.
If absent, the diagnostics are suppressed.
The optional
.meta name
argument can be used to specify the file name which is used for reporting
errors. If this argument is missing, the name is taken from the name
property of the
.meta source
argument if it is a stream, or else the word
.code string
is used as the name if
.meta source
is a string.
If there are no parse errors, the function returns the parsed data
structure. If there are parse errors, and the
.meta error-retval
parameter is
present, its value is returned. If the
.meta error-retval
parameter
is not present, then an exception of type
.code syntax-error
is thrown.
.SS* Debugging Functions
.coNP Functions @ source-loc and @ source-loc-str
.synb
.mets (source-loc << form )
.mets (source-loc-str < form <> [ alternative ])
.syne
.desc
These functions map an expression in a \*(TX program to the file name and
line number of the source code where that form came from.
The
.code source-loc
function returns the raw information as a cons cell
whose
.cod3 car / cdr
consist of the line number, and file name.
The
.code source-loc-str
function formats the information as a string.
Forms which were parsed from a file have source location info
tracking to their origin in that file. Forms which are the result
of macro-expansion are traced to the form whose evaluation produced
them. That is to say, they inherit that form's source location info.
More precisely, when a form is produced by macro-expansion,
it usually consists of material which was passed to the macro as arguments,
plus some original material allocated by the macro, and possibly
literal structure material which is part of the macro code.
After the expansion is produced, any of its constituent material
which already has source location info keeps that info. Those nodes
which are newly allocated by the macro-expansion process inherit
their source location info from the form which yields the expansion.
If
.meta form
is not a piece of the program source code that was constructed by the
\*(TX parser or by a macro, and thus it was neither attributed with
source location info, nor has it inherited such info, then
.code source-loc
returns
.codn nil .
In the same situation, and if its
.meta alternative
argument is missing, the
.code source-loc-str
returns a string whose text conveys that the source location is not
available. If the
.meta alternative
argument is present, it is returned.
.coNP Functions @ rlcp and @ rlcp-tree
.synb
.mets (rlcp < dest-form << source-form )
.mets (rlcp < dest-tree << source-form )
.syne
.desc
The
.code rlcp
function copies the source code location info ("rl" means "read location")
from the
.meta source-form
object to the
.meta dest-form
object. These objects
are pieces of list-based syntax.
If
.meta dest-form
already has source code location info, then no copying
takes place.
The
.code rlcp-tree
function copies the source code location info from
.code rlcp
into every cons cell in the
.meta dest-tree
tree structure which doesn't already have location info.
It may be regarded as a recursive application of
.code rlcp
via
.cod3 car / cdr
recursion on the tree structure. However, the traversal performed by
.code rlcp-tree
gracefully handles circular structures.
Note: these functions are intended to be used in certain kinds of macros. If a
macro transforms
.meta source-form
to
.metn dest-form ,
this function can be used to propagate the
source code location info also, so that when the \*(TL evaluator
encounters errors in transformed code, it can give diagnostics which refer
to the original untransformed source code.
The macro expander already performs this transfer. If a macro call form
has location info, the expander propagates that info to that form's
expansion. In some situations, it is useful for a macro or other code
transformer to perform this action explicitly.
.coNP Special variable @ *rec-source-loc*
.desc
The Boolean special variable
.code *rec-source-loc*
controls whether the
.code read
and
.code iread
functions record source location info. The variable is
.code nil
by default, so that these functions do not record source location info.
If it is true, then these functions record source location info.
Regardless of the value of this variable, source location info is
recorded for Lisp forms which are read from files or streams under the
.code load
function or specified on the \*(TX command line. Source location
info is also always recorded when reading the \*(TX pattern language syntax.
Note: recording and propagating location info incurs a memory and performance
penalty. The individual cons cells and certain other literal objects in the
structure which emerges from the parser are associated with source location
info via a global weak hash table.
.coNP Function @ macro-ancestor
.synb
.mets (macro-ancestor << form )
.syne
.desc
The
.code macro-ancestor
function returns information about the macro-expansion ancestor of
.metn form .
The ancestor is the original form whose expansion produced
.metn form .
If
.meta form
is not the result of macro-expansion, or the ancestor information
is unavailable, the function returns
.codn nil .
.SS* Profiling
.coNP Operator @ prof
.synb
.mets (prof << form *)
.syne
.desc
The
.code prof
operator evaluates the enclosed forms from left to right similarly
to
.codn progn ,
while determining the memory allocation requests and time
consumed by the evaluation of the forms.
If there are no forms, the prof operator measures the smallest measurable
operation of evaluating nothing and producing
.codn nil .
If the evaluation terminates normally (not abruptly by a nonlocal
control transfer), then
.code prof
yields a list consisting of:
.mono
.mets >> ( value < malloc-bytes < gc-bytes << milliseconds )
.onom
where
.meta value
is the value returned by the rightmost
.metn form ,
or
.code nil
if there are no forms,
.meta malloc-bytes
is the total number of bytes of all memory allocation
requests (or at least those known to the \*(TX runtime, such as those of all
internal objects),
.meta gc-bytes
is the total number of bytes drawn from the
garbage-collected heaps, and
.meta milliseconds
is the total processor time
consumed over the execution of those forms.
Notes:
The bytes allocated by the garbage collector from the C function
.code malloc
to create
heap areas are not counted as
.metn malloc-bytes .
.meta malloc-bytes
includes storage
such as the space used for dynamic strings, vectors and bignums (in addition to
their gc-heap-allocated nodes), and the various structures used by the
.code cobj
type objects such as streams and hashes. Objects in external libraries that use
uninstrumented allocators are not counted: for instance the C
.code "FILE *"
streams.
.coNP Macro @ pprof
.synb
.mets (pprof << form *)
.syne
.desc
The
.code pprof
(pretty-printing
.codn prof )
macro is similar to
.codn progn .
It evaluates
.metn form s,
and returns the rightmost one, or
.code nil
if there are no forms.
Over the evaluation of
.metn form s,
it counts memory allocations, and measures
CPU time. If
.metn form s
terminate normally, then just prior to returning,
.code pprof
prints these statistics in a concise report on the
.codn *stdout* .
The
.code pprof
macro relies on the
.code prof
operator.
.SS* Garbage Collection
.coNP Function @ sys:gc
.synb
.mets (sys:gc <> [ full ])
.syne
.desc
The
.code gc
function triggers garbage collection. Garbage collection means
that unreachable objects are identified and reclaimed, so that their
storage can be reused.
The function returns
.code nil
if garbage collection is disabled (and consequently nothing is done), otherwise
.codn t .
The Boolean
.meta full
argument, defaulting to
.codn nil ,
indicates whether a full garbage collection should be requested.
Even if this argument is
.codn nil ,
a full garbage collection may occur due to having been scheduled.
.coNP Function @ sys:gc-set-delta
.synb
.mets (sys:gc-set-delta << bytes )
.syne
.desc
The
.code gc-set-delta
function sets the GC delta parameter.
Note: This function may disappear in a future release of \*(TX or suffer
a backward-incompatible change in its syntax or behavior.
When the amount of new dynamic memory allocated since the last garbage
collection equals or exceeds the GC delta, a garbage collection pass is
triggered. From that point, a new delta begins to be accumulated.
Dynamic memory is used for allocating heaps of small garbage-collected objects
such as cons cells, as well as the satellite data attached to some objects:
like the storage arrays of vectors, strings or bignum integers. Most garbage
collector behaviors are based on counting objects in the heaps.
Sometimes a program works with a small number of objects which are very large,
frequently allocating new, large objects and turning old ones into garbage.
For instance a single large integer could be many megabytes long. In such a
situation, a small number of heap objects therefore control a large amount of
memory. This requires garbage collection to be triggered much more often than
when working with small objects, such as conses, to prevent runaway allocation
of memory. It is for this reason that the garbage collector uses the GC delta.
There is a default GC delta of 64 megabytes. This may be overridden in
special builds of \*(TX for small systems.
.coNP Function @ finalize
.synb
.mets (finalize < object < function <> [ reverse-order-p ])
.syne
.desc
The
.code finalize
function registers
.meta function
to be invoked in the situation when
.meta object
is identified by the garbage collector as unreachable.
A function registered in this way is called a finalizer.
If and when this situation occurs, the finalizer
.meta function
will be called with
.meta object
as its only argument.
Multiple finalizer functions can be registered for the same object,
up to an internal limit which is not required to be greater than 255.
If the limit is exceeded,
.code finalize
throws an error exception.
All registered finalizers are called when the object becomes unreachable.
Finalizers registered against an object may also be invoked
and removed using the
.code call-finalizers
function.
If the
.meta reverse-order-p
argument isn't specified, or is
.codn nil ,
then finalizer is registered at the end of the list.
If
.meta reverse-order-p
is true, then the finalizer is registered at the front of
the list.
Finalizers which are activated in the same finalization processing phase
are called in the order in which they appear in the
registration list.
After a finalization call takes place, its registration is removed. However,
neither
.meta object
nor
.meta function
are reclaimed immediately; they are treated as if they were reachable objects
until at least the next garbage collection pass.
It is therefore safe for
.meta function
to store somewhere a persistent reference to
.meta object
or to itself, thereby reinstating these objects as reachable.
A finalizer is itself permitted to call
.code finalize
to register the original
.code object
or any other object for finalization. Finalization processing can be
understood as taking place in one or more rounds. At the start of each round,
finalizers are identified that are to be called, arranged in order, and removed
from the registration list. If this identification stage produces no
finalizers, then finalization ends. Otherwise, those finalizers are processed,
and then another round is initiated, to look for eligible finalizers that may have been
registered during the previous round.
Note: it is possible for the application to create an infinite finalization
loop, if one or more objects have finalizers that register new finalizers,
which register new finalizers and so on.
Note: if a finalizer is invoked by the garbage collector rather than explicit
finalization via
.codn call-finalizers ,
and that finalizer calls
.code finalize
to make a registration, that registration will not be eligible for processing in
the same phase, because the criteria for finalization is unreachability.
.coNP Function @ call-finalizers
.synb
.mets (call-finalizers << object )
.syne
.desc
The
.code call-finalizers
function invokes and removes the finalizers, if any, registered against
.metn object .
If any finalizers are called, it returns
.codn t ,
otherwise
.codn nil .
Finalization performed by
.code call-finalizers
works in the manner described under the specification of the
.code finalize
function.
It is permissible for a finalizer function itself to call
.codn call-finalizers .
Such a call can happen in two possible contexts: finalization
initiated by by garbage collection, or under the scope of a
.code call-finalizers
invocation from application code. Doing so is safe, since the finalization
logic may be reentered recursively. When finalizers are being called during a
round of processing, those finalizers have already been removed from the
registration list, and will not be redundantly invoked by a recursive
invocation of finalization.
Under the scope of garbage-collection-driven reclamation, the
order of finalizer calls may not be what the application logic
expects. For instance even though a finalizer registered for some object
.code A
itself invokes
.codn "(call-finalizers B)" ,
it may be the case during GC reclamation that both
.code A
and
.code B
are identified as unreachable objects at the same time, and some or all
finalizers registered against
.code B
have already been called before the given
.code A
finalizer performs the explicit
.code call-finalizers
invocation against
.codn B .
Thus the call either has no effect at all, or only calls some remaining
.code B
finalizers that have not yet been processed, rather than all of them,
as the application expects.
The application must avoid creating a dependency on the order of
finalization calls, to prevent the situation that the finalization actions are
only correct under an explicit
.code call-finalizers
but incorrect under spontaneous reclamation driven by garbage collection.
.SS* Stack-Overflow Protection
\*(TX features a rudimentary mechanism for guarding against stack overflows,
which cause the \*(TX process to crash. This capability is separate
from and exists in addition to the possibility of catching a
.code sig-segv
(segmentation violation) signal upon stack overflow using
.codn set-sig-handler .
The stack-overflow guard mechanism is based on \*(TX, at certain key places
in the execution, checking the current position of the stack relative to
a predetermined limit. If the position exceeds the limit, then an exception
of type
.codn stack-overflow ,
derived from
.codn error ,
is thrown.
The stack-overflow guard mechanism is configured on startup.
On platforms where it is possible to
inquire the system's actual stack limit, and where the stack limit is
at least 512 kilobytes, \*(TX sets the limit to within a
certain percentage of the actual value. If it is not possible to determine the
system's stack limit, or if the system indicates that the stack size is
unlimited, then a default limit is imposed. If the system's limit is
configured below a certain small value, then that small value is used
as the stack limit.
The
.code get-stack-limit
and
.code set-stack-limit
functions are provided to manipulate the stack limit.
The mechanism cannot contain absolutely all sources of stack-overflow threat
under all conditions. External functions are not protected, and not all
internal functions are monitored. If \*(TX is close to the limit, but
a function is called whose stack growth is not monitored, such as
an external function or unmonitored internal function, it is possible that
the stack may overflow anyway.
.coNP Functions @ get-stack-limit and @ set-stack-limit
.synb
.mets (get-stack-limit)
.mets (set-stack-limit << value )
.syne
.desc
The
.code get-stack-limit
returns the current value of the stack limit. If the guard mechanism is
not enabled, it returns
.codn nil ,
otherwise it returns a positive integer, which is measured in bytes.
The
.code set-stack-limit
configures the stack limit according to
.metn value ,
possibly enabling or disabling the guard mechanism, and returns the previous
stack limit in exactly the same manner as
.codn get-stack-limit .
The
.meta value
must be a non-negative integer or else the symbol
.codn nil .
The values zero or
.code nil
disable the guard mechanism. Positive integer values set the limit.
The value may be truncated to a multiple of some denomination or
otherwise adjusted, so that a subsequent call to
.code get-stack-limit
need not retrieve that exact value.
If
.meta value
is too close to the system's stack limit or beyond, the effectiveness
of the stack-overflow detection mechanism is compromised.
Likewise, if
.meta value
is too low, the operation of \*(TX shall become unreliable. Values
smaller than 32767 bytes are strongly discouraged.
.SS* Modularization
.coNP Variable @ self-path
.desc
This variable holds the invocation pathname of the \*(TX program.
The value of
.code self-path
when \*(TL expressions are being evaluated in command-line arguments
is the string
.strn cmdline-expr .
The value of
.code self-path
when a \*(TX query is supplied on the command line via the
.code -c
command-line option is the string
.strn cmdline .
Note that for programs read from a file,
.code self-path
holds the resolved name, and not the invocation name. For instance if
.code foo.tl
is invoked using the name
.codn foo ,
whereby \*(TX infers the suffix, then
.code self-path
holds the suffixed name.
.coNP Variable @ stdlib
.desc
The
.code stdlib
variable expands to the directory where the \*(TX standard library
is installed. It includes the trailing slash.
Note: there is no need to use the value of this variable to load library
modules. Library modules are keyed to specific symbols, and lazily loaded. When
a \*(TL library function, macro or variable is referenced for the first time,
the library module which defines it is loaded. This includes references
which occur during the code expansion phase, at "macro time", so it works for
macros. In the middle of processing a syntax tree, the expander may encounter a
symbol that is registered for autoloading, and trigger the load. When the load
completes, the symbol might now be defined as a macro, which the expander
can immediately use to expand the given form that is being traversed.
.coNP Function @ load
.synb
.mets (load << target )
.syne
.desc
The
.code load
function causes a file containing \*(TL or \*(TX code to be read and processed.
The
.meta target
argument is a string. The function can load \*(TL source files as well
as compiled files.
Firstly, the value in
.meta target
is converted to a
.I "tentative pathname"
as follows.
If
.meta target
specifies a pure relative pathname, as defined by the
.code pure-rel-path-p
function, then a special behavior applies.
If an existing load operation is in progress, then the special variable
.code *load-path*
has a binding. In this case,
.code load
will assume that the relative pathname is a reference relative to the
directory portion of that pathname.
If
.code *load-path*
has the value
.codn nil ,
then a pure relative
.meta target
pathname is used as-is, and thus resolved relative to the current working
directory.
Once the tentative pathname is determined,
.code load
determines whether the name is suffixed. The name is suffixed if it
ends in any of these four suffixes:
.codn .tlo ,
.codn .tl ,
.code .txr
or
.codn .txr_profile .
Depending on whether the tentative pathname is suffixed,
.code load
tries to make one or more attempts to open several variations of that name.
These variations are called
.I "actual paths" .
If any attempt fails due to an error other than non-existence,
such as a permission error, then no further attempts are made; the
error exception propagates to
.codn load 's
caller.
If the tentative pathname is suffixed, then
.code load
tries to open a file by that actual pathname. If that attempt
fails, no other names are tried.
If the tentative pathname is unsuffixed, then first the suffix
.code .tlo
is appended to the name, and an attempt is made to open a file
with this actual path. If that file is not found, then the suffix
.code .tl
is similarly tried. If that file is not found, then the unsuffixed
name is tried.
If an unsuffixed file is opened, its contents are treated as interpreted Lisp.
Files ending in
.code .txr_profile
are also treated as interpreted Lisp. Files ending in
.code .tlo
are treated as compiled Lisp, and those ending in
.code .txr
are treated as the \*(TX Pattern Language.
If the file is treated as \*(TL, then Lisp forms are read from it in
succession. Each form is evaluated as if by the
.code eval
function, before the next form is read.
If a syntax error is encountered, an exception of type
.code eval-error
is thrown.
If a file is treated as a compiled \*(TL object file, then the compiled images
of top-level forms are read from it, converted into compiled objects, and
executed.
If the file treated as \*(TX Pattern Language code,
then its contents are parsed in their entirety.
If the parse is successful, the query is executed.
Previous \*(TX pattern variable and function bindings are in
effect. If the query binds new variables and functions,
these emerge from the
.code load
and take effect. If the parse is unsuccessful, an exception of type
.code query-error
is thrown.
Parser error messages are directed to the
.code *stderr*
stream.
Over the evaluation of either a \*(TL, compiled file, or \*(TX file,
.code load
establishes a new dynamic binding for several special
variables. The variable
.code *load-path*
is given a new binding containing the actual pathname.
The
.code *package*
variable is also given a new dynamic binding, whose value is the
same as the existing binding. Thus if the processing of the
loaded file has the effect of altering the value of
.codn *package* ,
that effect will be undone when the binding is removed
after the load completes.
When the
.code load
function terminates normally after processing a file,
it returns
.codn nil .
If the file contains a \*(TX pattern query which is
processed to completion, the matching success or failure
of that query has no bearing on the return value of
.codn load .
Note that this behavior is different from the
.code @(load)
directive which itself fails if the loaded query
fails, causing subsequent directives not to be processed.
A \*(TX pattern language file loaded with the Lisp
.code load
function does not have the usual implicit access to the
command-line arguments, unlike a top-level \*(TX query.
If the directives in the file try to match input, they
work against the
.code *stdin*
stream. The
.code @(next)
directive behaves as it does when no more arguments
are available.
If the source or compiled file begins with the characters
.codn #! ,
usually indicating a hash-bang script,
.code load
reads reads the first line of the file and discards it.
Processing of the file then begins with the first byte
following that line.
.coNP Special variable @ *load-path*
.desc
The
.code *load-path*
special variable has a top-level value which is
.codn nil .
When a file is being loaded, it is dynamically bound to the
pathname of that file. This value is visible to the forms
are evaluated in that file during the loading process.
The
.code *load-path*
variable is bound when a file is loaded from the command
line.
If the
.code -i
command-line option is used to enter the interactive listener,
and a file to be loaded is also specified, then the
.code *load-path*
variable remains bound to the name of that file inside the
listener.
The
.code load
function establishes a binding for
.code *load-path*
prior to processing and evaluating all the top-level forms
in the target file. When the forms are evaluated, the binding
is discarded and
.code load
returns.
The
.code compile-file
function also establishes a binding for
.codn *load-path* .
The
.code @(load)
directive, also similarly establishes a binding around the
parsing and processing of a loaded \*(TX source file.
Also, during the processing of the profile file (see Interactive Profile File),
the variable is bound to the name of that file.
.coNP Special variable @ *load-hooks*
.desc
The
.code *load-hooks*
variable is at the centre of a mechanism which associates the deferred
execution of actions, associated with a loaded module or program termination.
The application may push values onto this list which are expected to be
functions, or objects that may be called as functions. These objects must
be capable of being called with no arguments.
In the situations specified below, the list of functions is processed as follows.
First
.code *load-hooks*
is examined, the list which it holds is remembered. Then the variable
is reset to
.codn nil ,
following which the remembered list is traversed in order. Each of the
functions in the list is invoked, with no arguments.
The
.code *load-hooks*
list is processed, as described above, whenever the
.code load
function terminates, whether normally or by throwing an exception. In this
situation, the
.code *load-hooks*
variable which is accessed is that binding which was established by that
invocation of
.codn load .
The execution of the functions from the
.code *load-hooks*
list takes place in the dynamic environment of the
.codn load :
all of the dynamic variable bindings established by that
.code load
are still visible, including that of
.codn *load-hooks* .
The
.code *load-hooks*
list is also processed after processing a \*(TX or \*(TL file that
is specified on the command line. If the interactive listener is
also being entered, this processing of
.code *load-hooks*
occurs prior to entering the listener. This situation occurs in the
context of the top-level dynamic environment, and so the global value of
.code *load-hooks*
is referenced.
Lastly,
.code *load-hooks*
is also processed if the \*(TX process terminates normally, regardless
of its exit status. This processing executes in whatever dynamic
environment is current at the point of exit, using its value of the
.code *load-hooks*
variable is used. It is unspecified whether, at exit time, the
.code *load-hooks*
functions are executed first, or whether the functions registered by
.code at-exit-call
are executed first. However, their executions do not interleave.
Note that
.code *load-hooks*
is not processed after the listener reads the
.code .txr_profile
file. Hooks installed by the profile file will activate when the process
exits.
.coNP Macros @ push-after-load and @ pop-after-load
.synb
.mets (push-after-load << form *)
.mets (pop-after-load)
.syne
.desc
The
.code push-after-load
and
.code pop-after-load
macros work with the
.code *load-hooks*
list.
The
.code push-after-load
macro's arguments are zero or more
.metn form s.
These forms are converted into the body of an anonymous function,
which is pushed onto the
.code *load-hooks*
list. The return value is the new value of
.codn *load-hooks* .
The
.code pop-after-macro
removes the first item from
.codn *load-hooks* .
The return value is the new value of
.codn *load-hooks* .
The following equivalences hold:
.verb
(push-after-load ...) <--> (push (lambda () ...) *load-hooks*)
(pop-after-load) <--> (set *load-hooks* (cdr *load-hooks*))
.brev
.coNP Macro @ load-for
.synb
.mets (load-for >> {( kind < sym << target )}*)
.syne
.desc
The
.code load-for
macro takes multiple arguments, each of which is a three-element
clause. Each clause specifies that a given
.meta target
file is to be conditionally loaded based on whether a symbol
.meta sym
has a certain kind of binding.
Each argument clause has the syntax
.mono
.meti >> ( kind < sym << target )
.onom
where
.meta kind
is one of the five symbols
.codn var ,
.codn fun ,
.codn macro ,
.code struct
or
.codn pkg .
The
.meta sym
element is a symbol suitable for use as a variable, function
or structure name, and
.meta target
is an expression which is evaluated to produce a value that is suitable
as an argument to the
.code load
function.
First, all
.code target
expressions in all clauses are unconditionally evaluated in left-to-right
order. Then the clauses are processed in that order. If the
.meta kind
symbol of a clause is
.codn var ,
then
.code load-for
tests whether
.meta sym
has a binding in the variable namespace using the
.code boundp
function. If a binding does not exist, then the value of the
.meta target
expression is passed to the
.code load
function. Otherwise,
.code load
is not called.
Similarly, if
.meta kind
is the symbol
.codn fun ,
then
.meta sym
is instead tested using
.codn fboundp ,
if
.meta kind
is
.codn macro ,
then
.meta sym
is tested using
.codn mboundp ,
if
.meta kind
is
.codn struct ,
then
.meta sym
is tested using
.codn find-struct-type ,
and if
.meta kind
is
.codn pkg ,
then
.meta sym
is tested using
.codn find-package .
When
.code load-for
invokes the
.code load
function, it confirms whether loading file has had the expected effect of
providing a definition of
.meta sym
of the right
.metn kind .
If this isn't the case, an error is thrown.
The
.code load-for
function returns
.codn nil .
.coNP Variable @ txr-exe-path
.desc
This variable holds the absolute pathname of the executable file
of the running \*(TX instance.
.SS* Function Tracing
.coNP Special variable @ *trace-output*
.desc
The
.code *trace-output*
special variable holds a stream to which all trace output
is sent. Trace output consists of diagnostics enabled by the
.code trace
macro.
.coNP Macros @ trace and @ untrace
.synb
.mets (trace << function-name *)
.mets (untrace << function-name *)
.syne
.desc
The
.code trace
and
.code untrace
macros control function tracing.
When
.code trace
is called with one or more arguments, it considers each
argument to be the name of a global function. For each
function, it turns on tracing, if it is not already turned on.
If an argument denotes a nonexistent function, or is invalid
function name syntax,
.code trace
terminates by throwing an exception, without processing the
subsequent arguments, or undoing the effects already applied
due to processing the previous arguments.
When
.code trace
is called with no arguments, it lists the names of functions
for which tracing is currently enabled. In other cases it
returns
.codn nil .
When
.code untrace
is called with one or more arguments, it considers each
argument to be the name of a global function. For each
function, it turns off tracing, if tracing is enabled.
When
.code untrace
is called with no arguments, it disables tracing for all
functions.
The
.code untrace
macro always returns
.code nil
and silently tolerates arguments which are not names of functions
currently being traced.
Tracing a function consists of printing a message prior to entry into the
function indicating its name and arguments, and another message upon leaving
the function indicating its return value, which is syntactically correlated
with the entry message, using a combination of matching and indentation.
These messages are posted to the
.code *trace-output*
stream.
When traced functions call each other or recurse, these trace messages
nest. The nesting is detected and translated into indentation levels.
Tracing works by replacing a function definition with a trace hook function, and
retaining the previous definition. The trace hook calls the previous definition
and produces the diagnostics around it. When
.code untrace
is used to disable tracing, the previous definition is restored.
Methods can be traced; their names are given using
.mono
.meti (meth < struct << slot )
.onom
syntax: see the
.code func-get-name
function.
Macros can be traced; their names are given using
.mono
.meti (macro << name )
.onom
syntax. Note that
.code trace
will not show the destructured internal macro arguments, but only the
two arguments passed to the expander function: the whole form, and the
environment.
The
.code trace
and
.code untrace
functions return
.codn nil .
.SS* Dynamic Library Access
.coNP Function @ dlopen
.synb
.mets (dlopen >> [{ lib-name | nil} <> [ flags ])
.syne
.desc
The
.code dlopen
function provides access to the POSIX C library function of the
same name.
The argument to the optional
.meta lib-name
parameter may be a character string, or
.codn nil .
If it is
.codn nil ,
then the POSIX function is called with a null pointer for
its name argument, returning the handle for the main program,
if possible.
The
.meta flags
argument should be expressed as some bitwise combination of the values
of the variables
.codn rtld-lazy ,
.codn rtld-now ,
or other
.code rtld-
variables which give names to the
.codn dlopen -related
flags. If the
.meta flags
argument is omitted, the default value used is
.codn rtld-lazy .
If the function succeeds, it returns an object of type
.code cptr
which represents the open library handle ("dlhandle").
Otherwise it throws an exception, whose message incorporates, if possible,
error text retrieved from the
.code dlerror
POSIX function.
The
.code cptr
handle returned by
.code dlopen
will automatically be subject to
.code dlclose
when reclaimed by the garbage collector.
.coNP Function @ dlclose
.synb
.mets (dlclose << dlhandle )
.syne
.desc
The
.code dlclose
closes the library indicated by
.metn dlhandle ,
which must be a
.code cptr
object previously returned by
.codn dlopen .
The handle is closed by passing the stored pointer to the POSIX
.code dlclose
function. The internal pointer contained in the
.code cptr
object is then reset to null.
It is permissible to invoke
.code dlclose
more than once on a
.code cptr
object which was created by
.codn dlopen .
The first invocation resets the
.code cptr
object's pointer to null; the subsequent invocations
do nothing.
The
.code dlclose
function returns
.code t
if the POSIX function reports a successful result (zero), otherwise
it returns
.codn nil .
It also returns
.code nil
if invoked on a previously closed, and hence nulled-out
.code cptr
handle.
.coNP Functions @ dlsym and @ dlvsym
.synb
.mets (dlsym < dlhandle << sym-name )
.mets (dlvsym < dlhandle < sym-name << ver-name )
.syne
.desc
The
.code dlsym
function provides access to the same-named POSIX function. The
.code dlvsym
function provides access to the same-named GNU C Library function,
if available.
The
.meta dlhandle
argument must be a
.code cptr
handle previously returned by
.code dlopen
and not subsequently closed by
.code dlclose
or altered in any way.
The
.meta sym-name
and
.meta ver-name
arguments are character strings.
If these functions succeed, they return a
.code cptr
value which holds the address of the symbol which was found
in the library.
If they fail, they return a
.code cptr
object containing a null pointer.
.coNP Functions @ dlsym-checked and @ dlvsym-checked
.synb
.mets (dlsym-checked < dlhandle << sym-name )
.mets (dlvsym-checked < dlhandle < sym-name << ver-name )
.syne
.desc
The
.code dlsym-checked
and
.code dlvsym-checked
functions are alternatives to
.code dlsym
and
.codn dlvsym ,
respectively. Instead of returning a null
.code cptr
on failure, these functions throw an exception.
.coNP Variables @, rtld-lazy @, rtld-now @, rtld-global @, rtld-local @, rtld-nodelete @ rtld-noload and @ rtld-deepbind
.desc
These variables provide the same values as constants in the POSIX C library
header
.code "<dlfcn.h>"
named
.codn RTLD_LAZY ,
.codn RTLD_NOW ,
.codn RTLD_LOCAL ,
etc.
.SS* Data Interchange Support
.coNP Macro @ json
.synb
.mets (json [quote | sys:qquote] << object )
.syne
.desc
The
.code json
macro exists in supports of the JSON literal and quasiliteral
.mono
.meti >> #J json-syntax
.onom
and
.mono
.meti >> #J^ json-syntax
.onom
notations, which use the macro as their target abstract syntax.
The macro transforms itself by deleting the
.code json
symbol, producing either the
.mono
.meti (quote << object )
.onom
quote syntax, or else the
.mono
.meti (sys:qquote << object )
.onom
quasiquote syntax, depending on which quoting symbol is present.
If the application produces and expands a
.code json
macro form which does not conform to this syntax, or does not
specify one of the above two quoting symbols, the behavior is unspecified.
.coNP Functions @ put-json and @ put-jsonl
.synb
.mets (put-json < obj >> [ stream <> [ flat-p ]])
.mets (put-jsonl < obj >> [ stream <> [ flat-p ]])
.syne
.desc
The
.code put-json
function converts
.meta obj
into JSON notation, and writes that notation into
.meta stream
as a sequence of characters.
If
.meta stream
is an external stream such as a file stream, then the JSON is
rendered by conversion of the characters into UTF-8, in the usual
manner characteristic of those streams.
The behavior is unspecified if
.meta obj
or any component of
.meta obj
is an object incompatible with the JSON representation conventions.
An exception may be thrown.
An object conforms to the JSON representation conventions if it is:
.RS
.IP 1.
one of the symbols
.codn nil ,
.code t
or
.codn null ,
which map to the JSON keywords
.codn false ,
.code true
and
.codn null ,
respectively.
.IP 2.
a floating-point number.
.IP 3.
a character string.
.IP 4.
a vector of JSON-conforming objects.
.IP 5.
a hash table whose keys and values are JSON-conforming objects.
.RE
.IP
Note that unless the keys in a hash table are all strings, nonstandard JSON
is produced, since RFC 8259 requires JSON object keys to be strings.
If the
.code flat-p
argument is present and has a true value, then the JSON is generated
without any line breaks or indentation. Otherwise, the JSON output is subject
to such formatting.
The difference between
.code put-json
and
.code put-jsonl
is that the latter emits a newline character after the JSON output.
When a string object is output as JSON string syntax, the following rules
.RS
.IP 1.
The characters
.code \e
(backslash, reverse solidus) and
.code \(dq
(double quote)
are preceded by a backslash escape.
.IP 2.
The characters U+0008 (BS), U+0009 (TAB), U+000A (LF), U+000C (FF) and
U+000D (CR) are rendered as, respectively,
.codn \eb ,
.codn \et ,
.codn \en ,
.code \ef
and
.codn \er .
.IP 3.
If the character sequence
.code "</script"
occurs in a string, then in the JSON representation the slash is escaped, such
that the sequence is rendered as
.codn "<\e/script" .
Instances of
.code /
(forward slash, solidus) in other situations are unescaped. Rationale: this is
a feature of JSON which allows for safer embedding of the resulting
JSON into HTML
.code script
tags.
.IP 4
If the character sequence
.code <!--
occurs in a string, then in the JSON representation, the sequence is
rendered as
.codn <\eu0021-- .
Instances of
.code !
(exclamation mark) in other situations are not encoded. Rationale: safe
embedding in HTML
.code script
tags.
.IP 5
If the character sequence
.code -->
occurs in a string, then in the JSON representation, the sequence is
rendered as
.codn -\eu002D> .
Instances of
.code -
(hyphen) in other situations are not encoded. Rationale: safe
embedding in HTML
.code script
tags.
.IP 6.
The code point U+DC00 (\*(TX's pseudo-null character) is translated into the
.code "\eu0000"
escape syntax.
.IP 7.
The code points U+DC01 through U+DCFF are send to the stream as-is.
If the stream performs UTF-8 encoding, these characters turn into individual
bytes in the range 0 to 255.
.IP 8.
Control characters in the U+0001 to U+001F other than the ones subject
to rule 1 above are rendered as
.code \eu
escape sequences. Likewise, code points in the range U+007F to U+00BF,
the range U+D800 to U+DBFF, U+DD00 to U+DFFF, and the code points
U+FFFE and U+FFFF are also encoded as
.code \eu
escape sequences.
.IP 9.
A character outside of the BMP (Basic Multilingual Plane) in the range
U+10000 to U+10FFFF is encoded using as a pair of consecutive
.code \eu
escape sequences, specifying the code points of a UTF-16 surrogate pair
encoding that character. This representation is described in RFC 8259.
.RE
The
.code put-json
and
.code put-jsonl
functions return
.metn t .
.coNP Function @ tojson
.synb
.mets (tojson < obj <> [ flat-p ])
.syne
.desc
The
.code tojson
function converts
.meta obj
into JSON notation, returned as a character string.
The function can be understood as constructing a string output stream,
calling the
.code put-json
function to write the object into that stream,
and then retrieving and returning the constructed string.
The
.meta flat-p
argument is passed to
.codn put-json .
.coNP Function @ get-json
.synb
.mets (get-json >> [ source
.mets \ \ \ \ \ \ \ \ \ \ >> [ err-stream
.mets \ \ \ \ \ \ \ \ \ \ \ >> [ err-retval >> [ name <> [ lineno ]]]]])
.syne
.desc
The
.code get-json
function closely resembles the
.code read
function, and follows the same argument and error reporting conventions.
Rather than reading a Lisp object from the input source, it reads a JSON
object, with support for \*(TX's JSON extensions.
If an object is successfully read, its Lisp representation is returned.
JSON numbers produce floating-point number objects.
JSON strings produce string objects. The keywords
.codn true ,
.code false
and
.code null
map to the Lisp symbols
.codn t ,
.codn nil ,
and
.codn null ,
respectively.
JSON objects map to hash tables, and JSON arrays to vectors.
.coNP Function @ put-jsons
.synb
.mets (put-jsons < seq >> [ stream <> [ flat-p ]])
.syne
.desc
The
.code put-jsons
function writes multiple JSON representations into
.metn stream .
The objects are specified by the
.meta seq
argument, which must be an iterable object. The
.code put-jsons
function iterates over
.meta seq
and writes each element to the stream as if by using the
.code put-jsonl
function. Consequently, a newline character is written after each object.
If the
.meta stream
argument is not specified, the parameter takes on the value of
.metn *stdout* .
The
.meta flat-p
argument has the same meaning as in
.code put-json
with regard to the individual elements. If it is specified and true,
then exactly as many lines of text are written to
.meta stream
as there are elements in
.metn seq .
The
.code put-jsons
function returns
.metn t .
.coNP Function @ get-jsons
.synb
.mets (get-jsons <> [ source ])
.syne
.desc
The
.meta get-jsons
function reads zero or more JSON representations from
.meta source
until an end-of-stream or error condition is encountered.
If
.meta source
is a character string, then the input takes place from a stream
created from the character string using
.codn make-string-byte-input-stream .
Otherwise, if
.meta source
is specified, it must be an input stream supporting byte input;
input takes place from that stream. If the
.meta source
argument is omitted, it defaults to
.codn *stdin* .
The objects are read as if by calls to
.code get-json
and accumulated into a list.
If the end-of-stream condition is read, then the list of accumulated objects is
returned. If an error occurs, then an exception is thrown and the list of
accumulated objects is not available.
If an end-of-stream condition occurs before any character is seen other than
JSON whitespace, then the empty list
.code nil
is returned.
.coNP Functions @ file-get-json and @ file-get-jsons
.synb
.mets (file-get-json << name )
.mets (file-get-jsons << name )
.syne
.desc
The
.code file-get-json
and
.code file-get-jsons
function open a text stream over the file indicated by the string argument
.meta name
for reading. The functions ensure that the stream is closed when
they terminate.
The
.code file-get-json
function invokes
.code get-json
to read a single JSON object, which is returned if that function
returns normally.
The
.code file-get-jsons
function invokes
.code get-jsons
to retrieve a list of JSON objects from the stream, which is returned
if that function returns normally.
.coNP Functions @ file-put-json and @ file-put-jsons
.synb
.mets (file-put-json < name < obj <> [ flat-p ])
.mets (file-put-jsons < name < seq <> [ flat-p ])
.syne
.desc
The
.code file-put-json
and
.code file-put-jsons
functions open a text stream over the file indicated by the string argument
.metn name ,
using the function
.code open-file
with a
.meta mode-string
argument of
.strn w ,
write the argument object into the stream in their specific manner,
and then close the stream.
The
.code file-put-json
function writes a JSON representation of
.meta obj
using the
.code put-json
function. The
.meta flat-p
argument is passed to that function, defaulting to
.codn nil .
The value returned is that of
.codn put-json .
The
.code file-put-jsons
function writes zero or more JSON representations of objects from
.metn seq ,
which must be an iterable object, using the
.code put-jsons
function. The
.meta flat-p
argument is passed to that function, defaulting to
.codn nil .
The value returned is that of
.codn put-jsons .
.coNP Functions @ file-put-json and @ file-put-jsons
.synb
.mets (file-append-json < name < obj <> [ flat-p ])
.mets (file-append-jsons < name < seq <> [ flat-p ])
.syne
.desc
The
.code file-append-json
and
.code file-append-jsons
are identical in almost all requirements to the functions
.code file-put-json
and
.codn file-put-jsons .
The only difference is that when these functions open
a text stream using
.codn open-file ,
they specify a
.meta mode-string
argument of
.str a
rather than
.strn w ,
in order to append data to the target file rather than overwrite it.
.coNP Functions @ command-get-json and @ command-get-jsons
.synb
.mets (command-get-json << cmd )
.mets (command-get-jsons << cmd )
.syne
.desc
The
.code command-get-json
and
.code command-get-jsons
functions opens text stream over an input command pipe created for
the command string
.metn cmd ,
as if by the
.code open-command
function. They ensure that the stream is closed when they terminate.
The
.code command-get-json
function calls
.code get-json
on the stream, and returns the value returned by that function.
Similarly,
.code command-get-jsons
function calls
.code get-jsons
on the stream, and returns the value returned by that function.
.coNP Functions @ command-put-json and @ command-put-jsons
.synb
.mets (command-put-json < cmd < obj <> [ flat-p ])
.mets (command-put-jsons < cmd < seq <> [ flat-p ])
.syne
.desc
The
.code command-put-json
and
.code command-put-jsons
functions open an output text stream over an output command pipe created
for the command specified in the string argument
.metn cmd ,
using the function
.code open-command
function, write the argument object into the stream, in their specific manner,
and then close the stream.
The
.code command-put-json
function writes a JSON representation of
.meta obj
using the
.code put-json
function. The
.meta flat-p
argument is passed to that function, defaulting to
.codn nil .
The value returned is that of
.codn put-json .
The
.code command-put-jsons
function writes zero or more JSON representations of objects from
.metn seq ,
which must be an iterable object, using the
.code put-jsons
function. The
.meta flat-p
argument is passed to that function, defaulting to
.codn nil .
The value returned is that of
.codn put-jsons .
.coNP Variable @ *read-bad-json*
.desc
This dynamic variable, initialized to a value of
.codn nil ,
controls whether the parser is tolerant to certain non-conformances in the
syntax of JSON data, which are ordinarily syntax errors.
If the value of this variable is true, then the last element in a JSON
array or the last element pair in a JSON object may be followed by spurious
trailing comma, which is ignored.
Note: in the future, the variable may be extended to enable other instances of
tolerance in the area of JSON parsing.
.TP* Example:
.verb
(get-json "{ 3:4, }") -> ;; syntax error
(let ((*read-bad-json* t))
(get-json "{ 3:4, }"))
--> #H(() (3.0 4.0))
.brev
.SH* FOREIGN FUNCTION INTERFACE
On platforms where it is supported, \*(TX provides a feature called the
.IR "foreign function interface" ,
or FFI. This refers to the ability to interoperate with programming
interfaces which are defined by the binary data type representations
and calling conventions of the platform's principal C language compiler.
\*(TX's FFI module provides a succinct Lisp-based type notation for expressing C
data types, together with memory-management semantics pertinent to the transfer
of data between software components. The notation is used to describe the
arguments and return values of functions in external libraries, and of Lisp
callback functions that can be called from those libraries. Driven by the
compiled representation of the type notation, the FFI module performs
transparent conversions between Lisp data types and C data types, and
automatically manages memory around foreign calls and incoming callbacks,
for many common interfacing conventions.
The FFI module consists of a library of functions which provide all of its
semantics. On top of these functions, the FFI module provides a number of
macros which comprise an expressive, convenient language for defining
foreign interfaces.
The FFI module supports passing and returning both structures and arrays
by value. Passing arrays by value isn't a feature of the C language syntax;
from the C point of view, these by-value array objects in the \*(TX FFI
type system are equivalent to C arrays encapsulated in
.codn struct s.
A
.code carray
type is provided for situations when foreign code generates arrays of
undeclared, dynamic length, other than strings, and returns these arrays
by the usual convention of pointer to the first element. The handling of
.code carray
requires more responsibility from the application.
.SS* Cautionary Notes
The FFI feature is inherently unsafe. If the FFI type language is used to write
incorrect type definitions which do not match the actual binary interface of a
foreign function, undefined behavior results. Improper use of FFI can corrupt
memory, creating instability and security problems. It can also
cause memory leaks and/or use-after-free errors due to inappropriate
deallocation of memory.
The implicit memory management behaviors encoded in the FFI type system
are convenient, but risky. A minor declarative detail such as writing
.code str
instead of
.code str-d
in the middle of some nested type can make the difference between correct code
and code which causes a memory leak, or instability by freeing memory which is
in use.
FFI developers are encouraged to unit test their FFI definitions carefully
and use tools such as Valgrind to detect memory misuses and leaks.
.SS* Key Concepts
.NP* The \fIput\fP operation
When a function call takes place from the \*(TL arena into a foreign
library function, argument values must be prepared into the foreign
representation. This takes place by converting Lisp objects into
stack-allocated temporary buffers representing C objects. For aggregate objects
containing pointers, additional buffers are allocated dynamically. For
instance, suppose a structure contains a string and is passed by value. The
structure will be converted to a stack-allocated equivalent C structure, in
which the string will appear as a pointer. That pointer may use dynamically
allocated (via
.codn malloc )
string data. The operation which prepares argument material before a foreign
function call is the
.I put
operation. In FFI callback dispatch, the operation which propagates the
callback return value to the foreign caller is also the put operation.
.NP* The \fIin\fP operation
After a foreign function call returns from a foreign library back to the \*(TL
arena, the arguments have to be examined one more time, because two-way
communication is possible, and because some of the material has temporary,
dynamically allocated buffers associated with it which must be released. For
instance a structure passed by pointer may be updated by the foreign function.
FFI needs to propagate the changes which the foreign function performed to the
C version of the structure, back to the original Lisp structure. Furthermore,
a structure passed by pointer uses a dynamically allocated buffer. This buffer
must be freed. The operation which handles the responsibility for propagating
argument data back into \*(TL objects, and frees any temporary memory that had
been arranged by the
.I put
operation is the
.I in
operation.
The in operation has two nuances: the by-value nuance and the
by-pointer nuance.
Data passed into a function by value such as function arguments or via
.code ptr-in
are subject to the by-value nuance. Updates to the foreign representation
of these objects does not propagate back to the Lisp representation;
however, those objects may contain pointers requiring
the by-pointer nuance of the in operation of those pointers to be invoked.
.NP* The \fIget\fP operation
After a foreign call completes, it is also necessary to retrieve the call's
return value, convert it to a Lisp object, and free any dynamic memory.
This is performed by the
.I get
operation.
The
.I get
operation is also used by a Lisp callback function, called from a foreign
library, to convert the arguments to Lisp objects.
.NP* The \fIout\fP operation
When a Lisp callback invoked by a foreign library completes, it must
provide a return value, and also update any argument objects with new
values. The return value is propagated using the put operation. Updates
to arguments are performed by the
.code out
operation. This operation is like the reverse of the in operation. Like
that operation, it has a by-value and by-pointer nuance.
For instance, if a callback receives a structure by value, upon return, there
is no use in reconstructing a new version of the structure from the updated
Lisp structure; the caller will not receive the change. However, if the
structure contains pointers to data that
was updated, by the callback, those changes must materialize. This is achieved
by triggering the by-value nuance of the structure type's out operation, which
will recursively invoke the out operation of embedded pointers, which will
in turn invoke the by-pointer nuance.
.SS* The FFI Type System
The FFI type system consists of a notation built using Lisp syntax. Basic,
unparametrized types are denoted by symbolic atoms. Similarly to a concept
in the C language,
.code typedef
names can be globally defined, using the
.code ffi-typedef
function, or the
.code typedef
macro.
Like in the C language,
.code typedef
names are aliases for an existing type, and not distinct types. However,
this is of no consequence, since the FFI doesn't perform any type checking
between two foreign types, and thus never takes into consideration whether two
such types are equal. The main concern in FFI is correspondence between Lisp
values and foreign types. For instance, a Lisp string argument will not convert
to a foreign function parameter of type
.codn int .
Compound expressions denote the construction of derived types, or types which
are instantiated with parameters. Each such expression has a type constructor
symbol in the operator position, from a limited, fixed vocabulary, which cannot
be extended.
Some constituents of compound type syntax are expressions which evaluate to
integer values: the dimension value for array types, the size for buffers,
the width for bitfields and the value expressions for enumeration constants are
such expressions. These expressions allow full use of \*(TL. They are
evaluated without visibility into any apparent surrounding lexical scope.
Some predefined types which are provided are in fact typedef names.
For instance, the
.code size-t
type is a typedef name for some other integral type, defined in a
platform-specific way. Which type that is may be determined by passing
the syntax to the type compiler function using the expression
.codn "(ffi-type-compile 'size-t)" .
The type compiler converts the
.code size-t
syntax to the compiled type object, resolving the typedef name to
the type which it denotes. The printed representation of that object
reveals the identity of the type. For instance, it might be
.codn "#<ffi-type uint>" ,
indicating that
.code size-t
is an alias for the
.code uint
basic type, which corresponds to the C type
.codn "unsigned int" .
.SS* Simple FFI Types
.coNP FFI types @, char @, zchar @ uchar and @ bchar
These first two of these types,
.code char
and
.code zchar
correspond to the C character type
.codn char .
The
.code uchar
and
.code bchar
types correspond to
.codn "unsigned char" .
Both Lisp integers and character values
convert to these representations if they are in their numeric range.
Out-of-range values produce an exception.
A foreign
.codn char ,
.codn zchar ,
and
.code bchar
value converts to a Lisp character, whereas a
.code uchar
value converts to an integer.
If these types are used for representing individual scalar values,
there is no difference among
.codn char ,
.code zchar
and
.codn bchar .
What is different among these three types is that the
.code array
and
.code zarray
type constructors treat them specially. Arrays of these types are
subject to conversion to and from Lisp strings. The variation among
these types expresses different conversion semantics. That is to say,
an array of
.code bchar
converts between the foreign and native Lisp representation differently
from an array of
.codn zchar ,
which in turn converts differently from an array of
.codn char .
Note: it is recommended to avoid using the types
.code bchar
and
.code zchar
other than for expressing the element type of an
.code array
or
.codn zarray .
.coNP FFI types @, short @, ushort @, int @, uint @ long and @ ulong
These types correspond to the C integer types
.codn short ,
.codn "unsigned short" ,
.codn int ,
.codn "unsigned int" ,
.code long
and
.codn "unsigned long" .
Lisp characters and integers convert to these foreign representations, if they
are in their numeric range. Foreign values of these types convert
to Lisp integers.
.coNP FFI types @ longlong and @ ulonglong
These types are
.code typedef
names for integer types whose representation corresponds to the C types
.code "long long"
and
.codn "unsigned long long" .
.coNP FFI types @ int8 and @ uint8
These types correspond to 8-bit signed and unsigned integers.
They convert like integer types: both Lisp integers and characters
convert to these types, if in a suitable range; and under
the reverse conversion, the foreign values become Lisp integers.
.coNP FFI types @, int16 @, uint16 @, int32 @, uint32 @ int64 and @ uint64
These types correspond denote precisely sized C integer types.
They convert like integer types: both Lisp integers and characters
convert to these types, if in a suitable range; and under
the reverse conversion, the foreign values become Lisp integers.
.coNP FFI types @ float and @ double
These types correspond to the same-named C types. They convert
Lisp integers, characters and floating-point numbers to these C types.
Because the \*(TL
.code float
is represented as a C
.code double
it converts directly to
.code double
without the possibility of range error or loss of precision.
A conversion to type
.code float
is subject to a range check; an exception is thrown if the Lisp
floating-point value is out of range of this type. Even when the
conversion is possible, it alters the value, results in a loss of precision.
In the reverse direction, values of both types convert to the one and
only \*(TL
.code float
type.
.coNP FFI type @ bool
The type
.code bool
is a typedef name for the
.code uchar
instance of the parametrized
.code bool
type, which is to say,
.codn "(bool uchar)" .
.coNP FFI type @ val
The FFI
.code val
type denotes the machine representation of a Lisp value cell, which is
corresponds to a C pointer. Not all cell values are actually pointers, but
values that are heap objects, such as vectors and conses, are.
The
.code val
type transparently converts any Lisp object to a foreign pointer value
with no representation change at all; and performs the reverse conversion
from pointer to Lisp value.
Note: this is utterly dangerous. Lisp values that aren't pointers must not
be dereferenced by foreign code. Foreign code must not generate Lisp pointer
values that aren't objects which came from a Lisp heap.
Interpreting a Lisp value in foreign code requires a correct decoding of
its type tag, and, if necessary, stripping the tag bits to recover a heap
pointer and interpreting the type code stored in the heap object.
The conversion from foreign bit pattern to Lisp value is subject to a
validity checks; an exception will be thrown if the bit pattern isn't a valid
Lisp object. Nevertheless, the checks has cases which report as false
positives: admit some invalid objects may be admitted into the Lisp realm,
possibly with catastrophic results.
.coNP FFI type @ cptr
This type corresponds to a foreign pointer of any type, including a pointer to a function.
The
.code cptr
type converts between a foreign pointer and a Lisp object of type
.codn cptr .
Lisp objects of type
.code cptr
are tagged with a symbolic tag, which may be
.codn nil .
The unparametrized
.code cptr
converts foreign pointers to
.code cptr
objects which are tagged with
.codn nil .
In the reverse direction, it converts
.code cptr
Lisp objects of type
.code cptr
to foreign pointer, without regard for their type tag.
There is a parametrized version of the
.code cptr
FFI type, which provides a measure of type safety.
Note: the
.code cptr
type, in the context of FFI, is particularly useful for representing
C pointers that are used in C library interfaces as "opaque" handles.
For instance a FFI binding for the C functions
.code fopen
and
.code fclose
may use the
.code cptr
to represent the
.code "FILE *"
type. That is to say,
.code cptr
can be specified as the return type for
.codn fopen ,
thereby capturing the stream handle in a
.code cptr
object when that function is invoked through FFI. Then, the captured
.code cptr
object can be passed as the argument of
.code fclose
to close the stream.
.coNP FFI types @, str @, bstr @ str-d and @ bstr-d
These FFI types correspond to the C pointer type
.codn "char *" ,
providing automatic conversion between Lisp strings and null-terminated
C strings. The
.code str
and
.code str-d
types use UTF-8 encoding. The
.code bstr
and
.code bstr-d
types do not use UTF-8: only Lisp strings which contain strictly
code points in the range U+0000 to U+00FF may convert to these types;
out-of-range characters trigger an error exception.
The
.code -d
suffixed types differ from the unsuffixed variants
in that they denote the transfer of ownership of dynamically allocated memory,
and thus the responsibility for freeing that memory.
The
.code str
type behaves as follows. The put operation allocates, using
.codn malloc ,
a buffer large enough to hold the UTF-8 encoded version of the Lisp
string, encodes the string into that buffer, and then stores the
.code "char *"
pointer into the argument space. The in operation deallocates the
buffer. If
.code str
is passed by pointer, the in operation also takes the current value of the
.code "char *"
pointer, which may have been replaced by a different pointer, and creates a new
Lisp string by decoding UTF-8 from that buffer. The get operation retrieves the
C pointer and duplicates a new string by decoding the UTF-8 contents. The type
has no out operation: a string is not expected to be modified in-place.
The type
.code str-d
type differs in behavior from
.code str
as follows. Firstly, it has no in operation. Consequently,
.code str-d
doesn't deallocate the buffer that had been allocated by put.
Under the get operation, the
.code str-d
type assumes that ownership over the C pointer has been granted, and
after duplicating a new string from the decoded UTF-8 data in the C string,
it deallocates that C string by invoking the C library function
.code free
on it.
The type
.code bstr-d
behaves like
.code str-d
with regard to memory management; it differs from
.code str-d
in the same way that
.code str
differs from
.codn bstr :
it doesn't perform UTF-8 encoding or decoding.
Like other types, the string types combine with the
.code ptr
type family. Because the
.code ptr
family has memory management semantics, as does the string family,
it is important to understand the memory management implications
of the combination of the two.
The types
.code "(ptr str-d)"
and
.code "(ptr str)"
are effectively equivalent. They denote a string passed by pointer,
with in-out semantics. The effect is that the string is dynamic in both
directions. What that means is that the foreign function either must not
free the pointer it was given, or else it must replace it with one which the
caller can also free (or with a null pointer). The two are equivalent
because
.code str-d
has no in operation, so its get operation is used instead; but that operation
is similar to the in operation of the
.code str
type:
both decode the string currently referenced by the
.code "char *"
pointer, and then pass that pointer to the C
.code free
function.
To receive a string pointer by pointer from a foreign
function, one of the types
.code "(ptr-out str)"
or
.code "(ptr-out str-d)"
should be used, which have different semantics. In either situation, FFI will
prepare a pointer-sized uninitialized buffer, which the called function fills
with a
.code "char *"
pointer. In the
.code str
case, FFI will duplicate that string to a Lisp string. In the
.code str-d
case, FFI will also free the string received from the foreign function.
The type combination
.code "(ptr-in str-d)"
refers to a string pointer passed to a foreign function by pointer,
whereby the foreign function will retain and free the pointer. The type
combination
.code "(ptr-in str)"
passes the string pointer in the same way, but the foreign module mustn't
use the pointer after returning. FFI will free the pointer that had been
passed.
.coNP FFI types @ wstr and @ wstr-d
The FFI type
.code wstr
corresponds to the C type
.code "wchar_t *"
pointing to the first character of a null terminated wide string.
It converts between Lisp strings and symbols, and C strings.
The memory management is similar to the
.code str
and
.code str-d
types, except that no UTF-8 conversion takes place.
.coNP FFI types @ buf and @ buf-d
The
.code buf
type creates a correspondence between the \*(TL
.code buf
type and a C pointer to a block of arbitrary data. Note that there also
exists a parametrized version of the
.code buf
and
.code buf-d
type syntax which specifies a size.
Under the
.code buf
type's put operation, no memory allocation takes place. The pointer to the
buffer object's data is written into the argument space, so the foreign
function can manipulate the buffer directly. If the object isn't a buffer
but rather the symbol
.codn nil ,
then a null pointer is written.
The
.code buf
in operation has semantics
as follows. In the pass-by-pointer nuance, the buffer pointer currently in the
argument space is compared to the original one which had been written there
from the buffer object. If they are identical, then the in operation
yields the original buffer object. Otherwise, if the altered
pointer is non-null,
it allocates a new buffer equal in size to the original one and copies in the
new data from the new pointer that was placed into the argument space by the
foreign function. If the altered pointer is null, then instead of allocating a
new buffer, the object
.code nil
is returned.
The by-value nuance of the in operation does nothing.
The get operation is not meaningful for an unsized
.codn buf :
it yields a zero length
.code buf
object. For this reason, parametrized
.code buf
type should be used for retrieving a buffer with a specific fixed size.
The
.code buf-d
type has different memory management from
.codn buf .
The put operation of
.code buf-d
allocates a copy of the buffer and writes into the argument space
a pointer to the copy. It is assumed that the foreign function takes
ownership of the copy.
The in operation of
.code buf-d
is also different. The by-value nuance of the in operation
is a no-op, like that of
.codn buf .
The by-pointer nuance doesn't attempt to compare the previously written
pointer to the current value. Rather, it assumes that if there is any non-null
pointer value in the argument space, then it should take ownership of
that object and return it as a new buffer. Thus if two-way dynamic buffer
passing is requested using
.code "(buf buf-d)"
it means that the foreign function must replace the pointer with a null
to indicate that it has consumed the buffer. Any non-null value in the
argument space indicates that the foreign function has either rejected
the pointer (not taken ownership), or has replaced it with a new object,
whose ownership is being passed.
Unidirectional by-pointer passing of a
.code buf-d
can be performed using the types
.code "(ptr-out buf-d)"
or
.codn "(ptr-int buf-d)" .
The former type will not invoke
.codn buf-d 's
put operation. It will only allocate a pointer-sized space, without
initializing it. After the foreign call, the by-pointer semantics of the
in operation will be triggered If the foreign function places a non-null
pointer into the space, its ownership will be seized by a newly instantiated
buffer object. Otherwise the function must place a null pointer, which
results in a
.code nil
value emerging from the in operation as documented above. The latter type will
achieve a transfer of ownership in the other direction, by invoking the
.code buf-d
put operation, which places a copy of the buffer into the pointer-sized
location prepared in the argument space. After
the call, it will invoke the by-value in semantics of
.codn buf-d ,
which is a no-op: thus no attempt is made to extract a buffer, even if
the foreign function alters the pointer.
.coNP FFI type @ closure
The
.code closure
type converts three kinds of Lisp objects to a C pointer: the object
.codn nil ,
the
.code cptr
type, or the special
.code ffi-closure
type.
When the
.code nil
symbol is converted to a
.code closure
type, it becomes a null function pointer.
A
.code cptr
object of any kind converts to a
.codn closure ;
the internal pointer is converted to a function pointer.
Instances of the
.code ffi-closure
type are produced by the
.code ffi-make-closure
function, or by calls to functions defined by the
.code deffi-cb
macro. The
.code closure
type is useful for passing callbacks to foreign functions: Lisp functions
which appear to be C functions to foreign code.
In the reverse direction, when a
.code closure
object is converted from the foreign function pointer representation
to a Lisp object, it becomes a
.code cptr
object whose tag is the
.code closure
symbol.
.coNP FFI type @ void
The
.code void
type is useful for indicating the return type of foreign functions and
callbacks which return no value. It corresponds to a zero-sized object.
It will convert any lisp value into zero bytes, and convert
zero bytes into
.codn nil .
.SS* Parametrized FFI Type Operators
The following following parametrized type operators are available.
.coNP FFI type @ enum
.synb
.mets (enum < name >> {( sym << value ) | << sym }*)
.syne
.desc
The type
.code enum
specifies an enumerated type, which establishes a correspondence between
a set of Lisp symbols and foreign integer values of type
.codn int .
The
.meta name
argument must either be
.code nil
or a symbol for which the
.code bindable
function returns true. It gives the tag name of the enumerated
type. The remaining arguments specify the enumeration constants.
In the enumeration constant syntax, each occurrence of
.meta sym
They must be a bindable symbol according to the
.code bindable
function. The symbols may not repeat within the same enumerated type.
Unlike in the C language, different enumerations may use the same symbols;
they are in separate spaces.
If a
.meta sym
is given, it is associated with an integer value which is one greater
than the integer value associated with the previous symbol.
If there is no previous symbol, then the value is zero. If the previous
symbol has been assigned the highest possible value of the FFI
.code int
type, then an error exception is thrown.
If
.mono
.meti >> ( sym << value )
.onom
is given, then
.meta sym
is given the specified value. The
.meta value
is an expression which must evaluate to an integer value in range of the FFI
.code int
type.
It is evaluated in an environment in which the previous
symbols from the same enumeration appear as variables whose bindings are
their enumeration values, making it possible to use earlier enumerations in the
definition of later enumerations.
The FFI
.code enum
type converts two kinds of Lisp values to the foreign type
.codn int :
symbols which are in the set defined by the type, and integer values
which are in the range which that foreign type can represent.
Out-of-range integer values, symbols not defined in the enumeration, and
objects not of symbol or integer type all trigger an exception.
In the reverse direction, the
.code enum
type extracts from the foreign representation values of FFI type
.codn int ,
and converts them, if possible, to symbols. If an integer value occurs
which is not assigned to any enumeration symbol, then the conversion produces
that integer value itself rather than a symbol. If an integer value occurs
which is assigned to multiple enumeration symbols, it is not specified which
of those symbols is produced.
.coNP FFI type @ enumed
.synb
.mets (enumed < type < name >> {( sym << value ) | << sym }*)
.syne
.desc
The
.code enumed
type operator is a generalization of
.code enum
which allows the base integer type of the enumeration to be specified.
The following equivalence holds:
.verb
(enum n a b c ...) <--> (enumed int n a b c ...)
.brev
Any integer type or
.meta typedef
name may be specified for
.metn type ,
including any one of the endian types. The enumeration inherits its
size, alignment and other foreign representation details from
.metn type .
The values associated with the enumeration symbols must be in
the representation range of
.metn type ,
which is not checked until the conversion of a symbol
through the enumeration is attempted at run time.
The
.code enumed
type is a clone of the underlying type, inheriting most of its properties.
In particular, it is possible to derive an
.code enumed
type from an underlying bitfield type. The resulting type is still a bitfield,
and may only be used as a
.code struct
or
.code union
member. Moreover, because it is a bitfield type, there is a restriction against
creating aliases for it with
.codn typedef .
An
.code enumed
bitfield allows the values of a bit field to be specified symbolically.
.coNP FFI type @ struct
.synb
.mets (struct < name >> {( slot < type <> [ init-form ])}*)
.syne
.desc
The FFI
.code struct
type maps between a Lisp
.code struct
and a C
.codn struct .
The
.meta name
argument of the syntax gives the structure type's name, known as the tag.
If this argument is the symbol
.meta nil
then the structure type is named by a newly generated uninterned
symbol (with
.codn gensym ).
The
.meta name
is entered into a global namespace of tags which is shared by structures
and unions.
The
.meta name
also specifies the Lisp
.code struct
name associated with the FFI type.
The
.meta slot
and
.meta type
pairs specify the structure members. The
.meta slot
elements must be symbols, and the
.meta type
elements must be FFI type expressions.
A
.code struct
definition with no members refers to a previously defined
.code struct
or
.code union
type which has the same
.meta name
in the global
.cod3 struct / union
tag space.
If no prior
.code struct
or
.code union
exists, then a definition with no slots specifies a new, incomplete
structure type.
A
.code struct
definition with no members never causes a Lisp structure type to be created.
A
.code struct
definition that specifies one or more members either defines a new structure
type, or completes an existing one. If an incomplete structure or union
type which has the same
.meta name
exists, then the newly appearing definition is understood to provide
a completion of that type. If the incomplete type is a
.codn union ,
it thereby converted to a
.code struct
type.
If a complete structure type which has the
same
.meta name
already exists, then the newly appearing definition replaces that type
in the tag namespace.
A
.code struct
definition with members is entered into the
.cod3 struct / union
tag space immediately as an incomplete type (if it isn't already), before the
members are processed. Therefore, the member definitions can refer to the
.code struct
type. The type becomes complete when the last member is processed,
except in the special situation when that member causes the type to become
a flexible structure, described several paragraphs below.
A
.code struct
definition that specifies members causes a Lisp
.code struct
having the same
.code name
to exist, if such a type doesn't already exist. If such a type is
created, instance slots are defined for it which correspond to the
member definitions in the FFI
.code struct
definition.
For any
.meta slot
which specifies an
.meta init-form
expression, that expression is evaluated during the processing of the type syntax,
in the global environment. The resulting value then becomes the initial value for
the slot. The semantics of this value is similar to that of a quoted object
appearing as an
.meta init-form
in the
.code defstruct
macro's
.meta slot-specifier
syntax. For example, if the type expression
.codn "(struct s (a int expr))" ,
which specifies a slot
.code a
initialized by
.codn expr ,
generates a Lisp struct type, the manner in which that type is generated
will resemble that of
.code "(defstruct s nil (a (quote [value-of-expr])))"
where
.code [value-of-expr]
denotes the substitution of the value of
.code expr
which had been obtained by evaluation in the global environment.
Note: if more flexible initialization semantics is required, the application must
define the Lisp struct type first with the desired characteristics, before
processing the FFI struct type. The FFI struct type will then related to the
existing Lisp struct type.
Those members whose
.meta slot
name is specified as
.code nil
is ignored; no instance slots are created in the Lisp type.
If a
.meta init-form
is specified for such a slot, there exists is no situation in which that
form will be evaluated.
When a Lisp object is converted to a struct, it must, firstly, be of the struct
type specified by
.metn name .
Secondly, that type must have all of the slots defined in the FFI type.
The slots are pulled from the Lisp structure in the order that they appear
in the FFI
.code struct
definition. They are placed into the target memory area in that order,
with all required padding between the members, and possibly after
the last member, for alignment.
Whenever a member is defined using
.code nil
as the
.meta slot
name, that member represents anonymous padding. The corresponding
.meta type
expression is used only to determine the size of the padding only. Its data
transfer semantics is completely suppressed. When converting from Lisp, the
anonymous padding member simply generates a skip of the number of byte
corresponding to the size of its type, plus any necessary additional padding
for the alignment of the subsequent member.
Structure members may be bitfields, which are described using the
.codn ubit ,
.code sbit
and
.code bit
compound type operators.
A structure member must not be an incomplete or zero-sized array,
unless it is the last member. If the last member of FFI structure is
an incomplete array, then it is a flexible structure.
A structure member must not be a flexible structure, unless it is the
last member; the containing structure is then itself a flexible structure.
Flexible structures correspond to the C concept of a "flexible array member":
the idea that the last member of a structure may be an array of unknown size,
which allows for variable-length data at the end of a structure, provided
that the memory is suitably allocated.
Flexible structures are subject to special restrictions and requirements. See
the section Flexible Structures below. In particular, flexible structures
may not be passed or returned by value.
See also: the
.code make-zstruct
function and the
.code znew
macro.
.coNP FFI type @ union
.synb
.mets (union < name >> {( slot << type )}*)
.syne
.desc
The FFI
.code union
type resembles the
.code struct
type syntactically. It provides handling for foreign objects of C
.code union
type.
The
.meta name
argument specifies the name for the union type, known as a tag.
If this argument is the symbol
.meta nil
then the union type is named by a newly generated uninterned
symbol (with
.codn gensym ).
The
.meta name
is entered into a global namespace of tags which is shared by structures
and unions.
The
.meta slot
and
.code type
pairs specify the union members. The
.meta slot
elements must be symbols, and the
.meta type
elements must be FFI type expressions.
A
.meta union
definition with no member refers to a previously defined
.code struct
or
.code union
type which has the same
.meta name
in the global
.cod3 struct / union
tag space.
If no prior
.code struct
or
.code union
exists, then a definition with no slots specifies a new,
.code union
type that is incomplete.
A
.meta union
definition that specifies one or more members either defines a new structure
type, or completes an existing one. If an incomplete structure type which has
the same
.meta name
exists, then the newly appearing definition is understood to provide
a completion of that type. If the prior incomplete type is a
.codn struct ,
it is converted to
.code union
type. If a complete structure or union type which has the
same
.meta name
already exists, then the newly appearing definition replaces that type
in the tag namespace.
A struct
.code union
definition with members is entered into the
.cod3 struct / union
tag space immediately as an incomplete type (if it isn't already), before the
members are processed. Therefore, the member definitions can refer to the
.code union
type. The type becomes complete when the last member is processed.
Unlike the FFI
.code struct
type, the
.code union
type doesn't provide automatic conversion between C and Lisp data.
This is because the
.code union
is inherently unsafe, due to its placement of multiple types into the
same storage, and lack of any information to discriminate which type
is currently stored. Instead, the FFI
.code union
creates a correspondence between a C union that is regarded as just
a region of memory, and a \*(TL data type called
.codn union .
An instance of the Lisp
.code union
type holds a copy of the C union memory, and also contains type information
about the unions members. Functions are provided to store and retrieve the
members; it is these functions which provide the conversion between the
Lisp types and the foreign representations stored in the C union.
This is done under control of the application, because due to the inherent
lack of safety of the C
.codn union ,
only the application program knows which member of the union may be accessed.
Conversion between the C
.code union
and the Lisp
.code union
consists of just a memory copying operation.
The following functions are provided for manipulating unions:
.code make-union
instantiates a new union object;
.code union-members
retrieves a list of the symbols serving as the union's member names;
.code union-get
retrieves a specified member from the union's storage, converting it
to a Lisp object;
.code union-put
places a Lisp object into a union, using the specified member's type
to convert it to a foreign representation;
.code union-in
performs the "in semantics" on the specified member of a union,
propagating modifications in that member back to a Lisp object; and
.code union-out
performs "out semantics" on the specified member of a union,
propagating modifications done on a previously retrieved Lisp object
back into the union.
.coNP FFI type @ array
.synb
.mets (array < dim << type )
.mets (array << type )
.syne
.desc
The FFI
.code array
type creates a correspondence between Lisp sequences and
"by value" fixed size arrays in C. It converts Lisp sequences to C arrays, and
C arrays to Lisp vectors.
Arrays passed by values do not exist
in the C language syntax. Rather, the C type which corresponds to the
FFI array is a C array that is encapsulated in a
.codn struct .
For instance the type
.code "(array 3 char)"
can be visualized as corresponding to the C type
.codn "struct { char anonymous[3]; }" .
Thus, in the FFI syntax, we can specify arrays as function parameters
passed by value and as return values.
On conversion from Lisp to the foreign type, the FFI
.code array
simply iterates over the Lisp sequence, and performs an element for
element conversion to
.metn type .
If the sequence is shorter than the array, then the remaining elements
are filled with zero bits. If the sequence is longer than the array, then the
excess elements in the sequence are ignored.
Since Lisp arrays and C arrays do not share the same representation,
temporary buffers are automatically created and destroyed by FFI
to manage the conversion.
The
.meta dim
argument is an ordinary Lisp expression expanded and evaluated in the
top-level environment. It must produce a nonnegative integer value.
In addition, several types are treated specially: when
.meta type
is one of
.codn char ,
.codn zchar ,
.code bchar
or
.codn wchar ,
the array type establishes a special correspondence with Lisp strings.
When the C array is decoded, a Lisp string is created or updated in place
to reflect the new contents. This is described in detail below.
The second form, whose syntax omits the
.meta dim
element, it denotes a variable length
array. It corresponds to the concept of an incomplete array
in the C language, except that no implicit array-to-pointer conversion
concept is implemented in the FFI type system. This type may not
be used as an array element or structure member. It also may not
be passed or returned by value, only by pointer.
Since the type has unknown length, it has a trivial get operation which returns
.codn nil .
It is useful for passing a variable amount of data into a foreign
function by pointer.
An array of
.code char
represents non-null-terminated UTF-8 character data, which converts to
and from a Lisp string. Any null bytes in the data correspond to
the pseudo-null character
.code #\exDC00
also notated as
.codn #\epnul .
An array of
.code zchar
represents a field of optionally null-terminated UTF-8 character data.
If a null byte occurs in the data then the text terminates before that
null byte, otherwise the data comprises the entire foreign array.
Thus, null bytes do not occur in the data. A null byte in the array will
not generate a pseudo-null character in the Lisp string.
An array of
.code bchar
values represents 8-bit character data that isn't UTF-8 encoded,
and is not null terminated. Each byte holds a character whose code is
in the range 0 to 255. If a null byte occurs in the data, is interpreted
as a string terminator.
.coNP FFI type @ zarray
.synb
.mets (zarray < dim << type )
.mets (zarray << type )
.syne
.desc
The
.code zarray
type is a variant of
.codn array .
When converting from Lisp to C, it ensures that the array is null-terminated.
This means that if the
.meta zarray
is dimensioned, then the
.mono
.meti >> [ dim - 1]
.onom
element of the C array is written out as all zero bytes,
ignoring the corresponding Lisp value in the Lisp array.
If the
.meta zarray
is undimensioned, then the size of the C array is deemed to be one greater
than the actual length of the Lisp array. The elements in the Lisp array are
converted to the corresponding elements of the C array, and then the
last element of the C array is filled with null bytes.
The
.code zarray
type is useful for handling null terminated character arrays representing
strings, and for null terminated vectors.
Unlike
.codn array ,
.code zarray
allows the Lisp object to be one element short. For instance,
when a
.code "(zarray 5 int)"
passed by pointer a foreign function is converted back to Lisp,
the Lisp object is required to have only four elements. If the Lisp object
has five elements, then the fifth one will be decoded from the C array
in earnest; it is not expected to be null. However, when that Lisp
representation is converted back to C, that extra element will be ignored and
output as a zero bytes.
Lastly, the
.code zarray
further extends the special treatment which the
.code array
type applies to the types
.codn zchar ,
.codn char ,
.code wchar
and
.codn bchar .
The
.code zarray
type assumes, and depends on the incoming data being null-terminated, and
converts it to a Lisp string accordingly. The regular
.code array
type doesn't assume null termination. In particular, this means that whereas
.code "(array 42 char)"
will decode 42 bytes of UTF-8, even if some of them are null, converting
those null bytes to the U+DC00 pseudo-null, in contrast, a
.code zarray
will treat the 42 bytes as a null-terminated string, and decode UTF-8 only
up to the first null.
In the other direction, when converting from Lisp string to foreign array,
.code zarray
ensures null termination.
Note that the type combination
.code zarray
of
.code zchar
behaves in a manner indistinguishable from a
.code zarray
of
.codn char .
The one-argument variant of the
.code zarray
syntax which omits the
.meta dim
argument specifies a null-terminated variant of the variable-length array.
Like that type, it corresponds to the concept of an incomplete
array in the C language. It may not be used as an array element
or structure member, and cannot be passed as an argument or returned
as a value.
Unlike the ordinary variable-length
.codn array ,
the
.code zarray
type supports the get operation, which extracts elements, accumulating them
into a resulting vector, until it encounters an element consisting of all zero
bytes. That element terminates the decoding, and isn't included in the
resulting array.
The variable-length
.code zarray
also has a special in operation. Like the get operation, the in operation
extracts all elements until a terminating null, decoding them to a vector.
Then, the entire original vector is replaced with the new vector,
even if the original vector is longer.
.coNP FFI type @ ptr
.synb
.mets (ptr << type )
.syne
.desc
The
.meta ptr
denotes the passage of a value by pointer. The
.meta type
argument gives the pointer's target type. The
.code ptr
type converts a single Lisp value, to and from the target type,
using a C pointer as the external representation.
When used for passing a value to a foreign function, the
.code ptr
type has in-out semantics: it supports the interfacing concept that
the called function can update the datum which has been passed to it "by pointer",
thereby altering the caller's object. Since a Lisp value requires a conversion
to the FFI external representation, it cannot be directly passed by pointer.
Instead, this semantics is simulated. The put semantics of
.code ptr
allocates a temporary buffer, large enough to hold the representation of
.metn type .
The Lisp value is then encoded into this buffer, recursively relying on
the type's put semantics. After the foreign call,
.code ptr
triggers the in semantics of
.meta type
to update the Lisp object from the temporary buffer, and releases the
buffer.
The get semantics of
.code ptr
is used in retrieving a
.code ptr
return value, or, in a FFI callback, for retrieving the values of
incoming arguments that are of
.code ptr
type. The get semantics assumes that the memory referenced by the C
pointer is owned by foreign code. The Lisp object is merely decoded from the
data area, which is then not touched.
The
.code out
semantics of
.codn ptr ,
used by callbacks for updating the values of arguments
passed by pointer, assumes that the argument space already contains a
valid pointer. The pointer is retrieved from the argument space, and the
Lisp value is encoded into the memory referenced by that pointer.
Note that only Lisp objects with mutable slots can be meaningfully passed by
pointer with in-out semantics. If a Lisp object without immutable slots, such
as an integer, is passed using
.code ptr
the incoming updated value of the external representation will be ignored.
Concretely, if a C function has the argument signature
.code "(int *)"
with in-out semantics such that it updates the
.code int
object which is passed in, this function can be called as a foreign function
using a
.code "(ptr int)"
FFI type for the argument. However, the argument of the foreign call on the
\*(TL side is just an integer value, and that cannot be updated.
On the other hand, if a FFI
.code struct
member is declared as of type
.code "(ptr int)"
then the Lisp
.code struct
is expected to have an integer-valued slot corresponding to that member.
The slot is then subject to a bidirectional transfer. FFI will create an
.codn int -sized
temporary data area, encode the slot into that area and place that area's
pointer into the encoded structure. After the call, the new value of the
.code int
will be extracted from the temporary buffer, which will then be released.
The Lisp structure's slot will be updated with the new integer.
This will happen even if the Lisp structure is being passed as a by-value
argument.
.coNP FFI type @ ptr-in
.synb
.mets (ptr-in << type )
.syne
.desc
.code ptr-in
type is a variation of
.code ptr
which denotes the passing of a value by pointer into a function, but
not out. The put semantics of
.code ptr-in
is the same as that of
.codn ptr ,
but after the completion of the foreign function call, the in semantics
differs. The
.code ptr-in
type only frees the temporary buffer, without decoding from it.
The out semantics of
.code ptr-in
differs also. It effectively treats the object as if it were "by value",
since the reverse data transfer is ruled out. In other words,
.code ptr-in
simply triggers the by-value nuance of
.metn type 's
out semantics.
The get semantics of
.code ptr-in
is the same as that of
.codn ptr .
.coNP FFI type @ ptr-out
.synb
.mets (ptr-out << type )
.syne
.desc
The
.code ptr-out
type is a variant of
.code ptr
which denotes a by pointer data transfer out of a function only, not into.
The put semantics of
.code ptr-out
prepares a data area large enough to hold
.meta type
and stores a pointer to that area into the argument space.
The Lisp value isn't encoded into the data area.
The in semantics is the same as that of
.codn ptr :
the by-pointer nuance of
.metn type 's
in semantics is invoked to decode the external representation to
Lisp data.
.coNP FFI type @ ptr-in-d
.synb
.mets (ptr-in-d << type )
.syne
.desc
The
.code ptr-in-d
type is a variant of
.code ptr-in
which transfers ownership of the allocated buffer to the invoked
function. That is to say, the in semantics of
.code ptr-in-d
doesn't involve the freeing of memory that was allocated by put
semantics.
The
.code ptr-in-d
type is useful when a function expects a pointer to an object that
was allocated by
.code malloc
and expects to take responsibility for freeing that object.
Since the function may free the object even before returning,
the pointer must not be used once the function is called. This is
ensured by the in semantics of
.code ptr-in-d
which is the same as that of
.codn ptr-in .
The
.code ptr-in-d
type also has get semantics which assumes that ownership of the
C object is to be seized. FFI will automatically free the C object
when get semantics is invoked to retrieve a value through a
.codn ptr-in-d .
.coNP FFI type @ ptr-out-d
.synb
.mets (ptr-out-d << type )
.syne
.desc
The
.code ptr-out-d
type is a variant of
.code ptr-out
which is useful for capturing return values or, in a callback
producing return values.
The
.code ptr-out-d
type has empty put semantics. If it put semantics is invoked, it does
nothing: no area is allocated for
.meta type
and no pointer is stored into the argument space.
The in semantics is the same as that of
.codn ptr :
a pointer is retrieved from the argument space, the object is subject to
.metn type 's
in semantics to recover the updated Lisp value, and then the object
is freed.
The get semantics of
.code ptr-out-d
is identical to that of
.codn ptr-in-d .
The out semantics is identical to that of
.codn ptr .
.coNP FFI type @ ptr-out-s
.synb
.mets (ptr-out-s << type )
.syne
.desc
The
.code ptr-out-d
type is a variant of
.code ptr-out
similar to
.codn ptr-out-d ,
which assumes that the C object being received has an indefinite
lifetime, and doesn't need to be freed. The suffix stands for "static".
Like
.codn ptr-out-d ,
the
.code ptr-out-s
has no put semantics.
Its in semantics recovers a Lisp value from the external object whose pointer
has been stored by the foreign function, but doesn't free the external
object.
The get semantics retrieves a Lisp value without freeing.
.coNP FFI type @ bool
.synb
.mets (bool << type )
.syne
.desc
The parametrized type
.code bool
can be derived from any integer or floating-point type. There is also an
unparametrized
.code bool
which is a
.code typedef
for the type
.codn "(bool uchar)" .
The
.code bool
type family represents Boolean values, converting between a Lisp Boolean
and foreign Boolean. A given instance of the
.code bool
type inherits all of its characteristics from
.metn type ,
such as its size, alignment and foreign representation. It alters the
get and put semantics, however. The get semantics converts a foreign zero
value of
.meta type
to the Lisp symbol
.codn nil ,
and all other values to the symbol
.codn t .
The put semantics converts the Lisp symbol
.code nil
to a foreign value of zero. Any other Lisp object converts to the foreign
value one.
The
.code bool
types are not integers, and cannot be used as the basis of bitfields:
syntax like
.code "(bit 3 (bool uint))"
is not permitted. However, Boolean bitfields are possible when this
syntax is turned inside out: the
.code bool
type can be derived from a bitfield type, as exemplified by
.codn "(bool (bit 3 uint))" .
This simply applies the above described Boolean conversion semantics to a
three-bit field. A zero/nonzero value of the field converts to
.cod3 nil / t
and a
.code nil
or
.cod2 non- nil
Lisp value converts to a 0 or 1 field value.
.coNP FFI types @ ubit and @ sbit
.synb
.mets ({ubit | sbit} << width )
.syne
.desc
The
.code ubit
and
.code sbit
types denote C-language-style bitfields. These types can only appear
as members of structures. A bitfield type cannot be the argument or return
value of a foreign function or closure, and cannot be a foreign variable.
Arrays of bitfields and pointers, of any kind, to bitfields are a forbidden
type combination that is rejected by the type system.
The
.code ubit
type denotes a bitfield of type
.codn uint ,
corresponding to an
.code unsigned
bitfield in the C language.
The
.code sbit
type denotes a bitfield of type
.codn int .
Unlike in the C language, it is not implementation-defined whether such
a bitfield represents signed values; it converts between Lisp integers
that may be positive or negative, and a foreign representation which is
two's complement.
Bitfields based on some other types are supported using the more general
.code bit
operator, which is described below.
The
.meta width
parameter of is an expression evaluated in the top-level environment,
indicates the number of bits. It may range from
zero to the number of bits in the
.code uint
type.
In a structure, bitfields produced by
.code sbit
and
.code ubit
are allocated out in storage units which have the
same width and alignment requirements as a
.codn uint .
These storage units themselves can be regarded as anonymous members of the
structure. When a new unit needs to be allocated in a structure to hold
bitfields, it is allocated in the same manner as a named member of type
.code uint
would be at the same position.
A zero-length bitfield is permitted. It may be given a name, but the field
will not perform any conversions to and from the corresponding slot in the
Lisp structure. Note that in situations when the FFI struct definition
causes the corresponding Lisp structure type to come into existence, the
Lisp structure type will have slots for all the zero width named bitfields,
even though those slots don't participate in any conversions in conjunction
with the FFI type.
The presence of a zero-length bitfield ensures that a subsequent
structure member, whether bitfield or not, is placed in a new storage
unit of the size of the bitfield's base type.
Details about the algorithm by which bitfields are allocated within a structure
are given in the paragraph below entitled
.BR "Bitfield Allocation Rules" .
A
.code ubit
field stores values which follow a pure binary enumeration. For instance,
a bitfield of width 4 stores values from 0 to 15. On conversion from
the Lisp structure to the foreign structure, the corresponding member
must be a integer value in this range, or an error exception is thrown.
On conversion from the foreign representation to Lisp, the integer
corresponding to the bit pattern is recovered. Bitfields follow the
bit order of the underlying storage word. That is to say, the most
significant binary digit of the bitfield is the one which is closest
to the most significant bit of the underlying storage unit.
If a four-bit field is placed into an empty storage unit and the value
8 its stored, then on a big-endian machine, this has the effect of
setting to 1 the most significant bit of the underlying storage word.
On a little-endian machine, it has the effect of setting bit 3 of
the word (where bit 0 is the least significant bit).
The
.code sbit
field creates a correspondence between a range of Lisp integers,
and a foreign representation based on the two's complement system.
The most significant bit of the bitfield functions as a sign bit.
Values whose most significant bit is clear are positive, and use
a pure binary representation just like their
.code ubit
counterparts. The representation of negative values is defined
by the "two's complement" operation, which maps each value to
its additive inverse. The operation consists of temporarily treating the
entire bitfield as unsigned, and inverting the logical value of all the
bits, and then adding 1 with "wraparound" to zero if 1 is added to a field
consisting of all 1 bits. (Thus zero maps to zero, as expected.)
An anomaly in the two's complement system is that the most negative
value has no positive counterpart. The two's complement operation
on the most negative value produces that same value itself.
A
.code sbit
field of width 1
can only store two values: -1 and 0, represented by the bit patterns
1 and 0. An attempt to convert any other integer value to a
.code sbit
field of width 1 results in an error.
A
.code sbit
field of width 2 can represent the values -2, -1, 0 and 1, which are
stored as the bit patterns 10, 11, 00 and 01, respectively.
.coNP FFI type @ bit
.synb
.mets (bit < width << type )
.syne
.desc
The
.code bit
operator is more general than
.code ubit
and
.codn sbit .
It allows for bitfields based on integer units smaller than or equal to
.codn uint .
The
.meta type
argument may be any of the types
.codn char ,
.codn short ,
.codn int ,
.codn uchar ,
.codn ushort ,
.codn uint ,
.codn int8 ,
.codn int16 ,
.codn int32 ,
.codn uint8 ,
.code uint16
and
.codn uint32 .
When the character types
.code char
and
.code uchar
are used as the basis of bitfields, they convert integer values, not
characters.
In the case of
.codn char ,
the bitfield is signed.
All remarks about
.code ubit
and
.code sbit
apply to
.code bit
also.
Details about the algorithm by which bitfields are allocated within a structure
are given in the paragraph below entitled
.BR "Bitfield Allocation Rules" .
.coNP FFI types @ buf and @ buf-d
.synb
.mets ({buf | buf-d} << size )
.syne
.desc
The parametrized
.code buf
and
.code buf-d
types are variants of the unparametrized
.code buf
and
.codn buf-d ,
respectively. The
.meta size
argument is an expression which is evaluated in the top-level
environment, and must produce a nonnegative integer.
Because they have a size, these types have useful get
semantics.
The get semantics of
.code buf-d
is that a Lisp object of type
.code buf
is created which takes direct ownership of the memory.
The get semantics of
.code buf
is that a Lisp object is created using a dynamically allocated copy
of the memory.
.coNP FFI type @ carray
.synb
.mets (carray << type )
.syne
.desc
The
.code carray
type corresponds to a C pointer, in connection with the concept
of representing a variable length array that is passed and returned
as a pointer to the base element. On the Lisp side, the
.code carray
FFI type corresponds to the
.code carray
Lisp type. The
.code carray
Lisp type is similar to
.codn cptr ,
but supports array indexing operations, and some other features.
It can be regarded as a semantic cross between
.code cptr
and
.codn buf .
The get semantics of
.code carray
is simply that a pointer is retrieved from memory and converted to
a freshly allocated
.code carray
object which holds that pointer, and is marked as having an unknown
size. No copy is made of the underlying array. When the application
determines the size of the array, it can inform that object by calling the
.code carray-set-length
function.
The put semantics of the
.code carray
FFI type is simply to write, into the argument space, the pointer which the
object holds. The object must be a
.code carray
whose element type matches that of the FFI type.
The
.code carray
type has in semantics. When a
.code carray
is passed to a foreign function as an argument to a
.code ptr
or
.code ptr-out
parameter to either a
.code carray
or
.code cptr
type, what is passed to the function is a pointer to the
.codn carray 's
pointer. The foreign function may update this pointer to a
new value, and this value is stored back into the
.code carray
object. The array's length is reset to zero.
If it is an owned
.codn carray ,
arranged by
.codn carray-own ,
then the current array freed before the new pointer is assigned,
and the object's type is reset to borrowed array. The
.code carray
object must not be memory mapped
.code carray
coming from the
.code mmap
function.
The
.code carray
type lacks out semantics, since Lisp code cannot change its address;
so there is no new pointer to propagate back to a foreign caller
which passes a
.code carray
to a Lisp callback, and no other memory management tasks to perform.
The
.code carray
type is particularly useful in situations when
foreign code generates such an array, and the size of that array
isn't known from the object itself.
It is also useful, instead of a variable-length
.codn zarray ,
for passing a dynamic array to foreign code in situations when the application benefits
from managing the memory for the array. The variable-length
.code zarray
FFI type's disadvantage relative to
.code carray
is that the
.code zarray
converts an entire Lisp sequence to a temporarily allocated
array, which is used only for one call. By contrast, the
.code carray
object holds the C representation which Lisp code can manipulate;
and that representation is passed directly, just like in the case of
.codn buf .
Unlike
.codn buf ,
there is no dynamic variant of
.codn carray .
The transfer of ownership of a
.code carray
requires the use of explicit operations like
.code carray-free
and
.codn carray-own .
It is possible to create a
.code carray
view over a buffer, using
.codn carray-buf .
Lastly, the
.code carray
type is the basis for the \*(TL
.code mmap
function, which is documented in the section
.BR "Unix Memory Mapping" .
.coNP FFI type @ cptr
.synb
.mets (cptr << type-sym )
.syne
.desc
The parametrized
.code cptr
type is similar to the unparametrized
.codn cptr .
It also converts between Lisp objects of type
.code cptr
and foreign pointers. Unlike the unparametrized type, it provides a measure of
type safety, and also supports the conversion of
.code carray
objects.
When a foreign pointer is converted to a Lisp object under control of the
parametrized
.codn cptr ,
the resulting Lisp
.code cptr
object is tagged with the
.meta type-sym
symbol.
In the reverse direction, when a Lisp
.code cptr
object is converted to the parametrized type, its type tag must match
.metn type-sym ,
or else the conversion fails with an error exception.
This rule contains a slight relaxation: a
.code cptr
object with a
.code nil
tag can be converted to a foreign representation using any parametrized type,
if its value is null. In other situations, the
.code cptr-cast
function must be used to coerce the pointer object to the matching type.
Note that if
.meta type-sym
is specified as
.codn nil ,
then this is precisely equivalent to the unparametrized
.code cptr
which doesn't provide the above safety measure.
A
.code carray
object may also be converted to a foreign pointer under the control of
a parametrized
.code cptr
type. The
.code carray
object's internal pointer becomes the foreign pointer value.
The conversion is only permitted if the following two restrictions are not met,
otherwise an error exception is thrown.
Firstly, the
.meta type-sym
of the
.code cptr
type must be the name of an FFI type, at the time when the
.code cptr
type expression is processed, otherwise the
.code cptr
is not associated with a type.
Secondly, the
.code carray
object being converted must have an element type which matches the
FFI type denoted by the
.code cptr
object's
.metn type-sym .
Pointer type safety is useful, because FFI can be used to create bindings
to large application programming interfaces (APIs) in which objects of
many different kinds are referenced using pointer handles. The erroneous
situation can occur that a FFI call passes a handle of one kind to a function
expecting a different kind of handle. If all pointer handles are represented
by a single
.code cptr
type, then such a situation proceeds without diagnosis.
If handles of different types are all mapped to
.code cptr
types with different tags, the situation is intercepted and diagnosed
with an error exception.
.coNP FFI type @ align
.synb
.mets (align < width << type )
.syne
.desc
The FFI type operator
.code align
defines a type which is a copy of
.metn type ,
but with the alignment requirement replaced by the
.metn width .
The
.meta width
argument is an expression which is evaluated in the top-level
environment. It must produce a positive integer which is a power of two.
The
.code align
operator can be used to create a version of
.meta type
with stricter or weaker alignment. Alignment affects the placement of
the type as a structure member, and as an array element.
A type with alignment 1 can be placed at any byte offset. A type with
alignment 2 can be placed only at even addresses and offsets.
Alignment can be applied to all types, including arrays and structs.
It may also be applied to bitfields, but special considerations have
to be observed to obtain the intended effect, described below.
However,
out of the elementary types, only the integer and floating point types are
required to support a weakening of alignment. Whether a type which corresponds
to a pointer, such as a
.code str
or
.codn buf ,
can be written at an offset which doesn't meet that type's default alignment
is machine-dependent.
If a FFI struct type is declared with a weakened alignment, whether or not such
a structure can be read or written at the misaligned offsets depends on whether
the individual members support it. If they are integer or floating-point types,
or aggregates thereof, the usage is supported in a machine-independent manner.
A struct type declared to have a weaker alignment, such as 1, does not
lose any of the padding at its end. That is to say, alignment has no effect
on structure size. It affects the offset at which a structure is placed as
a member of an array or another structure, with its padding intact. To
eliminate the padding at the end of a structure, it is necessary to use
.code align
to manipulate the alignment of individual members.
When
.code align
is applied to the type of a bitfield member of a structure, it has no effect on
placement. The alignment of a non-zero bitfield which begins a new
storage unit is taken into consideration for the purpose of determining
the most strictly alignment member of the structure. The alignment of all
other bitfields is ignored.
.SS* Additional Types
.coNP FFI types @, size-t @, ptrdiff-t @, int-ptr-t @, uint-ptr-t @, wint-t @, sig-atomic-t @ time-t and @ clock-t
These additional FFI types for common C language types are provided as
.code typedef
aliases.
.coNP FFI type @ qref
.synb
.mets (qref < struct-type < member1 >> [ member2 ...])
.syne
.desc
The FFI type operator
.code qref
provides a way to reference the type of a member of a struct or union.
The
.meta struct-type
argument must be a type expression denoting a struct or union.
The
.meta member1
argument and any additional arguments must be symbols.
If
.code S
is a struct or union type, and
.code M
is a member, then
.code "(qref S M)"
is a type expression denoting the type of
.codn M .
Moreover, if
.code M
itself is a struct or union, which has a member named
.code N
then the type of
.code N
can be denoted by the expression
.codn "(qref S M N)" .
Similarly, additional symbols reference through additional struct/union
nestings.
Note: the referencing dot syntax can be used to write
.code qref
expressions.
For instance,
.code "(qref S M N)"
can be written as
.code S.M.N
instead.
.coNP FFI type @ elemtype
.synb
.mets (elemtype << type )
.syne
.desc
The FFI type operator
.code elemtype
denotes the element type of
.metn type ,
which must be a pointer, array or enum.
Note: there is also a macro
.codn elemtype .
The macro expression
.code "(elemtype X)"
is equivalent to the expression
.codn "(ffi (elemtype X))" .
.coNP FFI types @, blkcnt-t @, blksize-t @, clockid-t @, dev-t @, fsblkcnt-t @, fsfilcnt-t @, gid-t @, id-t @, ino-t @, key-t @, loff-t @, mode-t @, nlink-t @, off-t @, pid-t @, ssize-t @ uid-t and @ socklen-t
The additional names of various common POSIX types may also be available,
depending on platform. They are provided as
.code typedef
aliases.
.SS* Endian Types
In addition to the type system described in the previous section.
the FFI type system supports
.IR "endian types" ,
which are useful for dealing with
data formats defined by networking protocols and other kinds of standards,
or data structure definitions from other machines.
There are two kinds of
.IR endianness :
.I "Little endian"
refers to the least-significant byte of a data type being
stored at the lowest address in memory, lowest offset in a buffer, lowest
offset in a file, or earlier byte in a communication stream.
.I "Big endian"
is the opposite: it refers to the most significant byte occurring
at the lowest address, offset or stream position.
For each of the signed integral types
.code int16
through
.codn int64 ,
the corresponding unsigned types
.code uint16
through
.codn uint64 ,
and the two floating-point types
.code float
and
.codn double ,
the FFI type system provides a big-endian and little-endian version,
whose names are derived by prefixing the
.code be-
or
.code le-
prefix to its related type.
Thus, the exhaustive list of the endian types is:
.codn be-int16 ,
.codn be-uint16 ,
.codn be-int32 ,
.codn be-uint32 ,
.codn be-int64 ,
.codn be-uint64 ,
.codn be-float ,
.codn be-double ,
.codn le-int16 ,
.codn le-uint16 ,
.codn le-int32 ,
.codn le-uint32 ,
.codn le-int64 ,
.codn le-uint64 ,
.code le-float
and
.codn le-double .
These types have the same size and alignment as their plain, unprefixed
counterparts. Alignment can be overridden with the
.code align
type construction operator to create versions of these types with alternative
alignment.
Endian types are supported as arguments to functions, return values,
members of structs and elements of arrays.
\*(TL's FFI performs the automatic conversion from the abstract Lisp integer
representation to the foreign representations exhibiting the specified
endianness.
.SS* Incomplete Types
In the \*(TL FFI type system, the following types are
.IR incomplete :
the type
.codn void ,
arrays of unspecified size, and any
.code struct
whose last element is of incomplete type.
An incomplete type cannot used as a function parameter type, or a return
value type. It may not be used as an array element or union member type.
A struct member type may be incomplete only if it is the last member.
An incomplete structure whose last member is an array is a
.IR "flexible structure" .
.SS* Flexible Structures
If a FFI
.code struct
type is defined with an incomplete array (an array of unspecified size) as its
last member, then it specifies an incomplete type known as a
.IR "flexible structure" .
That array is the
.IR "terminating array" .
The terminating array corresponds to a slot in the Lisp structure; that
slot is the
.IR "last slot" .
A structure which has a flexible structure as its last member is also,
effectively, a flexible structure.
When a Lisp structure is being converted to the foreign representation
under the control of a flexible structure FFI type, the number of elements
in the terminating array is determined from the length of the object
stored in the last slot of the Lisp structure. The length includes the
terminating null element for
.code zarray
types. The conversion is consistent with the semantics of an incomplete
arrays that is not a structure member.
In the reverse direction, when a foreign representation is being converted
to a Lisp structure under the control of a flexible structure FFI type,
the size of the array that is accessed and extracted is determined from
the length of the object stored in the last slot, or, if the array type
is a
.code zarray
from detecting null-termination of the foreign array. The conversion of
the array itself is consistent with the semantics of an incomplete
arrays that is not a structure member.
Before the conversion takes place, all of the members of the
structure prior to the terminating array, are extracted and converted to
Lisp representations. The corresponding slots of the Lisp structure are
updated. Then if the Lisp structure type has a
.code length
method, that method is invoked. The return value of the method is used
to perform an adjustment on the object in the last slot.
If the existing object in the last slot is a vector, its length is adjusted to
the value returned by the method. If the existing
object isn't a vector, then it is replaced by a new
.codn nil -filled
vector, whose length is given by the return value of
.codn length .
The conversion of the terminating array to Lisp representation the proceeds
after this adjustment, using the adjusted last slot object.
.SS* Bitfield Allocation Rules
The \*(TL FFI type system follows rules for bitfield allocation which were
experimentally derived from the behavior of the GNU C compiler on several
mainstream architectures.
The allocation algorithm can be imagined to walk through the structure
from the first member to the last, maintaining a byte offset
.I O
which indicates how many whole bytes have been allocated to members so far,
and a bit offset
.I B
which indicates, additionally, how many bits have been allocated in the
byte which follows these
.I O
bytes, between 0 and 7.
When a non-bitfield member is placed, then there are two cases: either
.I B
is zero (only
.I O
bytes have been allocated, with no fractional byte) or else
.I B
is nonzero. In this latter case,
.I B
is reset to zero and
.I O
is incremented by one. In either case,
.I O
is adjusted up to the required alignment boundary for the new member.
The member is placed, and
.I O
is incremented again by the size of that member.
When a bitfield member is placed, the algorithm considers the structure
to be allocated in units of the base type of that bitfield member.
For instance if the bitfield is derived from type
.code uint16
then the structure's layout is considered to have been allocated in
.code uint16
units. The algorithm examines the value of
.I O
and
.I B
to determine the first available unit in which at least
one bit of unallocated space remains.
Then, if the unit at that offset has enough space to hold the new
bitfield, according to the bitfield's width, then the bitfield is
placed into that unit. Otherwise, the bitfield is placed into the
next available unit.
After a bitfield is placed, the values of
.I O
and
.I B
are adjusted so that
.I O
reflects the whole number of bytes which have been allocated to the
structure so far, and
.I B
indicates the 0 to 7 additional bits of any bitfield material protruding
past those whole bytes.
A zero-width bitfield is also considered with regard to the storage
unit size indicated by its type. As in the case of the nonzero-width
bitfield, the offset of the first available unit is found which
has at least one bit of unallocated space. Then, if that unit is
entirely empty, the zero-width bitfield has no effect. If that unit is
partially filled, then
.I O
is adjusted to point to the next unit after that, and
.I B
is reset to zero. Note that according to this semantics, a zero-width bitfield
can have an effect even if placed between non-bitfield members, or appears
as the last member of a structure. Also, a structure containing only a
zero-width bitfield has size zero.
If, after the placement of all structure members,
.I B
has a nonzero value, then the offset
.I O
is incremented by one to cover that byte.
As the last allocation step, the size of the structure is then padded up to a
size which is a multiple of the alignment of the most strictly aligned member.
A named bitfield contributes to the alignment of the structure, according to
its type, the same way as a non-bitfield member of the same type.
An unnamed bitfield doesn't contribute alignment, or else may be regarded as
having the weakest possible alignment, which is byte alignment.
If all of the members of a structure are unnamed bitfield members of any type,
it exhibits byte alignment.
The description isn't complete without a treatment of byte and bit order.
Bitfield allocation follows an imaginary "bit endianness" whose direction
follows the machine's byte order: most-significant bits are allocated first on
big endian, least significant bits first on little endian.
If a one-bit-wide bitfield is allocated into a hitherto empty structure, it
will be placed into the first byte of that structure, regardless of the
machine's endianness, and regardless of the underlying storage unit size for
that bitfield. Within that first byte, it will be placed into the most
significant bit position on a big-endian machine (bit 7); and on a
little-endian machine, it will be placed into the least significant bit
position (bit 0). If another one-bit-wide is allocated, it is placed into
bit 6 on big endian, and bit 1 on little endian.
More generally, whenever a bitfield is allocated for a big-endian machine, and
the storage unit is determined into which that bitfield shall be placed, the
most significant bits of that storage unit are filled first on a big-endian
machine, whereas the least significant bits are filled first on a little-endian
machine. From this it follows that on either type of machine, that field shall
be placed at the lowest-addressed byte or bytes in which unallocated bits
remain.
.SS* FFI Call Descriptors
The FFI mechanism makes use of a type-like representation called the "call
descriptor". A call descriptor is an object which uses FFI types to describe
function arguments and return values. A FFI descriptor is required to call
a foreign function, and to create a FFI closure to use as a callback
function from a foreign function back into \*(TL.
A FFI descriptor object can be constructed from a return value type, and a list
of argument types, and several other pieces of information using the
function
.codn ffi-make-call-desc .
This object can then be passed to
.code ffi-call
to specify the C type signature of a foreign function, or to
.code ffi-make-closure
to specify the C type signature of a FFI closure to bind to a Lisp function.
The FFI macros
.code deffi
and
.code deffi-cb
provide a simplified syntax for expressing FFI call descriptors,
which includes a notation for expressing variadic calls.
A note about variadic foreign functions: although there is support
in the call descriptor mechanism for expressing a variadic function,
it expresses a particular
.B instance
of a variadic function, rather than the variadic function's type
per se.
To call the same variadic function using different variadic arguments,
different call descriptors are required. For instance to perform
the equivalent of the C function call
.mono
printf("hello\en")
.onom
requires a certain descriptor. To perform the equivalent of
.mono
printf("hello, %s\en", name)
.onom
requires a different descriptor.
.SS* Foreign Function Type API
This group of functions comprises the basic interface to the \*(TL's FFI
type system module.
.coNP Function @ ffi-type-compile
.synb
.mets (ffi-type-compile << syntax )
.syne
.desc
The
.code ffi-type-compile
function produces and returns a compiled type object from a
.meta syntax
argument which specifies valid FFI syntax.
If the type syntax is invalid, or specifies a nonexistent
type specifier or operator, an exception is thrown.
Note: whenever a function argument is required to be of FFI type,
what it means is that it must be a compiled object, and not
a Lisp expression denoting FFI syntax.
.TP* Examples:
.verb
(ffi-type-compile 'int) -> #<ffi-type int>
(ffi-type-compile
'(array 3 double)) -> #<ffi-type (array 3 double)>
(ffi-type-compile 'blarg) -> ;; error
.brev
.coNP Function @ ffi-make-call-desc
.synb
.mets (ffi-make-call-desc < ntotal < nfixed < rettype
.mets \ \ < argtypes <> [ name ])
.syne
.desc
The
.code ffi-make-call-desc
function constructs a FFI call descriptor.
The
.meta ntotal
argument must be a nonnegative integer; it indicates the number
of arguments in the call.
If the call denotes a variadic function, the
.meta nfixed
argument must be an integer at least 1 and less than
.metn ntotal ,
denoting the number of fixed arguments.
If the call denotes an ordinary, non-variadic function, then
.meta nfixed
must either be specified specified as
.code nil
or else equal to the
.meta ntotal
argument.
The
.meta rettype
parameter must be an FFI type. It specifies the function
return type. Functions which don't return a value are specified
by the (compiled version of) the return type
.codn void .
The
.meta argtypes
argument must be a list of types, containing at least
.meta ntotal
elements. If the function takes no arguments, this list is empty.
If the function is variadic, then the first
.meta nfixed
elements of this list specify the types of the fixed arguments;
the remaining elements specify the variadic arguments.
The
.meta name
argument gives the name of the function for which this description is intended,
or some other identifying symbol. This symbols is used in diagnostic messages
related to errors in the construction of the descriptor itself or its
subsequent use. If this parameter is omitted, then the involved FFI functions
use their own names in reporting diagnostics.
Note: variadic functions must not be called using a non-variadic
descriptor, and vice versa,
even if the return types and
argument types match.
Note: unlike the
.code deffi
and
.code deffi-cb
macros,
the
.code ffi-make-call-desc
function doesn't perform any special treatment of variadic parameter types.
When any of the types
.codn float ,
.code be-float
or
.code le-float
occur in the variadic portion of
.metn argtypes ,
it is unspecified whether a descriptor is successfully produced and returned
or whether an exception is thrown. If a descriptor is successfully produced,
and then subsequently used for making or accepting calls, the behavior is
undefined.
.TP* Example:
.verb
;;
;; describe a call to the variadic function
;;
;; type void (*)(char *, ...)
;;
;; with these actual arguments
;;
;; (char *, int)
;;
(ffi-make-call-desc
2 ;; two arguments
1 ;; one fixed
(ffi-type-compile 'void) ;; returns nothing
(list (ffi-type-compile 'str) ;; str -> char *
(ffi-type-compile 'int))) ;; int
-->
#<ffi-call-desc #<ffi-type void>
(#<ffi-type str> #<ffi-type int>)>
.brev
.coNP Function @ ffi-type-operator-p
.synb
.mets (ffi-type-operator-p << symbol )
.syne
.desc
The
.code ffi-type-operator-p
function return
.code t
if
.meta symbol
is a type operator symbol: a symbol used in the first position of
a recognized compound type form in the FFI type system.
Otherwise, it returns
.codn nil .
.coNP Function @ ffi-type-p
.synb
.mets (ffi-type-p << symbol )
.syne
.desc
The
.code ffi-type-p
function returns
.code t
if
.meta symbol
denotes a type in the FFI type system: either a built-in type or
an alias type name established by
.codn typedef .
Otherwise, it returns
.codn nil .
.coNP Function @ ffi-make-closure
.synb
.mets (ffi-make-closure < lisp-fun < call-desc
.mets \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ >> [ safe-p <> [ abort-val ]])
.syne
.desc
The
.code ffi-make-closure
function binds a Lisp function
.metn lisp-fun ,
which may be a lexical closure, or any callable object, with a FFI call
descriptor
.meta call-desc
to produce a FFI closure.
A FFI closure is an object of type
.code ffi-closure
which is suitable as an argument for the type denoted by the
.code closure
type specifier keyword in the FFI type language.
This type appears a C function pointer in the foreign code,
and may be called as such. When it is called by foreign
code, it triggers a call to
.metn lisp-fun .
The optional
.meta safe-p
parameter controls whether the closure dispatch is "safe", the meaning of
which is described shortly. The default value is
.code t
so that unsafe closure dispatch must be explicitly requested with a
.code nil
argument for this parameter.
A callback closure which is safely dispatched, firstly, does not permit
the capture of delimited continuations across foreign code. Delimited
continuations can be captured inside a closure dispatched that way, but the
delimiting prompt must be within the callback's local stack frame, without
traversing across the foreign stack frames. Secondly, a callback closure which
is safely dispatched doesn't permit direct nonlocal control transfers across
foreign code, such as exception handling. Such transfers, however, appear to
work anyway (with caveats): this is because they are specially handled. The
closure dispatch mechanism intercepts all dynamic control transfers, converts
them to an ordinary return from the callback to the foreign code, and resumes
the control transfer when the foreign code itself finishes and returns.
If the callback returns a value (its return type is other than
.codn void )
then in this situation, the callback returns an all-zero-bits return
value to the foreign caller. If the
.meta abort-val
parameter is specified and its value is other than
.codn nil ,
then that value will be used as the return value instead of an all-zero
bit pattern.
An unsafely dispatched closure permits the capture of continuations from
the callback across the foreign code and direct dynamic control transfers which
abandon the foreign stack frames.
Unsafe closure dispatch is only compatible with foreign code which is
designed with that usage in mind. For instance foreign code which holds
dynamic resources in stack variables will leak those resources if abandoned
this way. There are also issues with capturing continuations across foreign
code.
Note: the C function pointer is called a "closure" because it carries
environment information. For instance, if
.code lisp-fun
is a lexical closure, invocations of it through the FFI closure
occur in its proper lexical environment, even though its external
representation is a simple C function pointer. This requires a special
trampoline trick: a piece of dynamically constructed machine code with the
closure binding embedded inside it, with the C function pointer pointing
to the machine code.
Note: the same call descriptor can be reused multiple times to create
different closures. The same Lisp function can be involved in multiple
FFI closures.
.TP* Example:
.verb
;; Package the TXR cmp-str function as a string
;; comparison callback compatible with:
;;
;; int (*)(const char *, const char *)
;;
(ffi-make-closure
(fun cmp-str)
(ffi-make-call-desc 2 nil ;; two args, non-variadic
(ffi-type-compile 'int) ;; int return
[mapcar ffi-type-compile '(str str)])) ;; args
.brev
.coNP Function @ ffi-call
.synb
.mets (ffi-call < fun-cptr < call-desc <> { arg }*)
.syne
.desc
The
.code ffi-call
function invokes a foreign function.
The
.meta fun-cptr
argument which must be a
.code cptr
object. It is assumed to point to a foreign function.
The
.meta call-desc
argument must be a FFI call descriptor, produced by
.codn ffi-make-call-desc .
The
.meta call-desc
must correctly describe the foreign function.
The zero or more
.meta arg
arguments are values which are converted into foreign
argument values. There must be exactly as many of these arguments
as are required by
.metn call-desc .
The
.code ffi-call
function converts every
.meta arg
to a corresponding foreign object. If these conversions are
successful, the converted foreign arguments are passed by value
to the foreign function indicated by
.metn fun-cptr .
An unsuccessful conversion throws an error.
When the call returns, the foreign function's return value is
converted to a Lisp object and returned, in accordance with the
return type that is declared inside
.metn call-desc .
.coNP Function @ ffi-typedef
.synb
.mets (ffi-typedef < name << type )
.syne
.desc
The
.code ffi-typedef
function installs the compiled FFI type given by
.meta type
as a typedef name under the symbol given by
.metn name .
After this registration, whenever the type compiler encounters that
symbol being used as a type specifier, it will replace it by the
type object it represents.
The
.code ffi-typedef
function returns
.metn type .
.TP* Example:
.verb
;; define refcount-t as an alias for uint32
(ffi-typedef 'refcount-t (ffi-type-compile 'uint32))
.brev
.coNP Function @ ffi-size
.synb
.mets (ffi-size << type )
.syne
.desc
The
.code ffi-size
function returns an integer which gives the storage size of
the given FFI type: the amount of storage required for the
external representation of that type.
Bitfield types do not have a size; it is an error to apply
this function to a bitfield.
The size is machine:specific.
.TP* Example:
.verb
(ffi-size '(ffi-type-compile 'double)) -> 8
(ffi-size '(ffi-type-compile 'char)) -> 1
(ffi-size '(ffi-type-compile
'(array 42 char))) -> 42
.brev
.coNP Function @ ffi-alignof
.synb
.mets (ffi-alignof << type )
.syne
.desc
The
.code ffi-alignof
function returns an integer which gives the alignment
the given FFI type. When an instance of
.meta type
is placed into a structure as a member, it is placed after the previous member
at the smallest available offset which is divisible by the alignment.
The bytes skipped from the smallest available offset to the smallest
available aligned offset are referred to as
.IR padding .
Bitfield types do not have an alignment; it is an error to apply
this function to a bitfield. Bitfields are allocated in
storage cells, and those cells have alignment which is the
same as that of the type
.codn int .
The alignment is machine-specific. It may be more strict than what the
hardware architecture requires, yet at the same time be smaller than the size
of the type. For instance, the size of the type
.code double
is commonly 8, yet the alignment is often 4, and this is so
even on processors like Intel x86 which can load and store
a double at a misaligned address.
The alignment of an array is the same as that of its element type.
The alignment of a structure is that of its member which has the
most strict (largest-valued) alignment.
It is a property of arrays, derived from requirements governing the C
language, that if the first element of an array is at a correctly aligned
address, then all elements are. To ensure that this property holds for
for arrays of structures, structures sometimes must include
padding at the end. This is because the size of a structure without any
padding might not be multiple of its alignment, which is derived from the most
strictly aligned member. For instance, if we assume an architecture on which
the size and alignment of
.code int
is 4, the size of the structure type
.code "(struct ab (a int) (b char))"
would be 5 if no padding were included. However,
in an array of these structures, the second element's
.code a
member would be placed at offset 5, rendering it misaligned.
To ensure that every
.code a
is placed at an offset which is multiple of 4, the struct type is extended
with anonymous padding so that its size is 8.
.TP* Example:
.verb
(ffi-alignof (ffi double)) -> 4
.brev
.coNP Function @ ffi-offsetof
.synb
.mets (ffi-offsetof < type << member )
.syne
.desc
The
.code ffi-alignof
function calculates the byte offset of
.meta member
within the FFI type
.metn type .
If
.meta type
isn't a FFI struct type, or if
.meta member
isn't a symbol naming a member of that type,
the function throws an exception.
An exception is also thrown if
.meta member
is a bitfield.
.TP* Example:
.verb
(ffi-offsetof (ffi (struct ab (a int) (b char))) 'b) -> 4
.brev
.coNP Function @ ffi-arraysize
.synb
.mets (ffi-arraysize << type )
.syne
.desc
The
.code ffi-arraysize
function reports the number of elements in
.metn type ,
which must be an array type: an
.codn array ,
.code zarray
or
.codn carray .
.TP* Example:
.verb
(ffi-arraysize (ffi (array 5 int))) -> 5
.brev
.coNP Function @ ffi-elemsize
.synb
.mets (ffi-elemsize << type )
.syne
.desc
The
.code ffi-elemsize
function reports the size of the element type of an array,
of the target type of a pointer, or of the base integer type of an enumeration.
The
.meta type
argument must be an array, pointer or enumeration type: a type constructed
by one of the operators
.codn array ,
.codn zarray ,
.codn carray ,
.codn ptr ,
.codn ptr-in ,
.codn ptr-out ,
.code enum
or
.codn enumed .
.TP* Example:
.verb
(ffi-elemsize (ffi (array 5 int))) -> 4 ;; (sizeof int)
.brev
.coNP Function @ ffi-elemtype
.synb
.mets (ffi-elemtype << type )
.syne
.desc
The
.code ffi-elemtype
function retrieves the element type of an array type,
target type of a pointer type, or base integer type of an enumeration.
The
.meta type
argument must be an array, pointer or enumeration type: a type constructed
by one of the operators
.codn array ,
.codn zarray ,
.codn carray ,
.codn ptr ,
.codn ptr-in ,
.codn ptr-out ,
.code enum
or
.codn enumed .
.TP* Example:
.verb
(ffi-elemtype (ffi (ptr int))) -> #<ffi-type int>
.brev
.SS* Foreign Function Macro Language
This group of macros provides a higher-level language for working with
FFI types and defining foreign function bindings. The macros are implemented
using the Foreign Function Type API described in the previous section.
.coNP Macro @ with-dyn-lib
.synb
.mets (with-dyn-lib < lib-expr << body-form *)
.syne
.desc
The
.code with-dyn-lib
macro works in conjunction with the
.codn deffi ,
.code deffi-sym
and
.code deffi-var
macros.
When a
.code deffi
form appears as one of the
.metn body-form s
of the
.code with-dyn-lib
macro, that
.code deffi
form is permitted to use the simplified forms of the
.meta fun-expr
argument, to refer to library functions succinctly, without having
to specify the library. The same remark applies to
.code deffi-sym
and
.codn deffi-var ,
regarding their
.meta var-expr
parameter.
A form invoking the
.code with-dyn-lib
macro should be a top-level form. The macro creates a global variable named
by a symbol generated by
.code gensym
whose initializing expression binds it to a dynamic library handle.
The macro then creates an environment in which the enclosed
.codn deffi ,
.code deffi-var
and
.code deffi-sym
forms can implicitly refer to that library via the global variable.
The
.meta lib-expr
argument can take on three different forms:
.RS
.meIP nil
If
.meta lib-expr
is
.codn nil ,
then
.code with-dyn-lib
arranges for the library to refer to the \*(TX executable itself.
.meIP < string
If
.meta lib-expr
is a literal string, then
.code with-dyn-lib
will arrange for the hidden variable to be initialized with
an expression which opens a handle to the specified library.
.meIP < form
If
.meta lib-expr
is any other form, then it is assumed to denote syntax for
opening the handle to a library. That syntax is used verbatim
as the initializing expression for the generated global variable
which holds the library handle.
.RE
.IP
The result value of a
.code with-dyn-lib
form is the symbol which names the generated variable which
holds the library handle.
.TP* Examples:
.verb
;; refer to malloc and free functions
;; in the executable
(with-dyn-lib nil
(deffi malloc "malloc" cptr (size-t))
(deffi free "free" void (cptr)))
;; refer to "draw" function in fictitious
;; "libgraphics" library:
(with-dyn-lib "libgraphics.so.5"
(deffi draw "draw" int (cptr cptr)))
;; refer to "init_foo" function via specific
;; library handle.
(defvarl foo-lib (dlopen "libfoo.so.1"))
(with-dyn-lib foo-lib
(deffi init-foo "init_foo" void (void)))
.brev
.coNP Macro @ deffi
.synb
.mets (deffi < name < fun-expr < rettype << argtypes )
.syne
.desc
The
.code deffi
macro arranges for a Lisp function to be defined, via
.codn defun ,
which calls a foreign function.
The
.meta name
argument must be a symbol suitable as a function name in a
.code defun
form. This specifies the function's Lisp name.
The
.meta fun-expr
parameter specifies the foreign function which is to be called.
The syntactic variants permitted for its argument are
described below.
The
.meta rettype
argument must specify the return type, using the FFI type syntax,
as an unquoted literal. The macro arranges for the compilation of this
syntax via
.codn ffi-type-compile .
The
.meta argtypes
argument must specify a list of the argument types, as an unquoted
literal list, using FFI type syntax. The macro arranges for these types
to be compiled. Furthermore, a special convention may be used for
specifying a variadic function: if the
.code :
(colon)
keyword symbol appears as one of the elements of
.metn argtypes ,
then the
.code deffi
form specifies a fixed call to a foreign function which is variadic. The
argument types before the colon keyword are the types of the fixed arguments.
The types after the colon, if any, are of the variadic arguments. Special
considerations apply to some variadic argument types, described below.
The following syntactic variants are permitted of the
.meta fun-expr
argument:
.RS
.meIP < name-string
If
.meta fun-expr
is a literal string, then the
.code deffi
form must be enclosed in the
.code with-dyn-lib
macro, appearing as one of that macro's
.metn body-form s.
In this situation the literal character string
.meta name-string
specifies a symbol to be found within the library established by the
.meta with-dyn-lib
macro.
.meIP >> ( name-string << ver-string )
This manner of specifying the
.meta fun-expr
also requires the
.code deffi
form to be enclosed in a
.codn with-dyn-lib .
It selects a particular version of a symbol from the library.
.meIP < form
If
.meta fun-expr
is any other form, then it must specify an expression which evaluates to a
.code cptr
object giving the address of a foreign library symbol. If this form
is used, then the
.code deffi
form need not be surrounded by a call to the
.code with-dyn-lib
macro.
.RE
.IP
When the FFI type
.code float
is used as the type of a variadic parameter,
.code deffi
replaces it by the FFI type
.codn double .
This treatment is necessary because the C variadic argument mechanism promotes
.code float
values to
.codn double .
Note: due to this substitution, it is possible to pass floating-point values
which are out of range of the
.code float
type, without any diagnosis. The behavior of is undefined in the
Lisp-to-C direction, if the C function extracts an out-of-range
.code double
argument as if it were of type
.codn float .
The FFI types
.code be-float
and
.code le-float
cannot be used for specifying the types of a variadic argument. If any of
these occur in that position,
.code deffi
throws an error.
Rationale: these types are related to the C type
.code float
type, which requires promotion in variadic passing. Promotion cannot
be performed on floating-point values whose byte order has been rearranged,
because promotion is a value-preserving conversion.
.IP
The result value of a
.code deffi
form is
.metn name .
.coNP Macros @ deffi-cb and @ deffi-cb-unsafe
.synb
.mets (deffi-cb < name < rettype < argtypes <> [ abort-val ])
.mets (deffi-cb-unsafe < name < rettype << argtypes )
.syne
.desc
The
.code deffi-cb
macro defines, using
.code defun
a Lisp function called
.metn name .
Thus the
.meta name
argument must be a symbol suitable as a function name in a
.code defun
form.
The
.meta rettype
and
.meta argtypes
arguments are processed exactly as in the corresponding arguments in the
.code deffi
macro.
The
.code deffi-cb
macro arranges for
.meta rettype
and
.meta argtypes
to be compiled into a FFI call descriptor.
The generated function called
.meta name
then serves as a combinator which takes a Lisp function as its argument,
and binds it to the FFI call descriptor to produce a FFI closure.
That closure may then be passed to foreign functions as a callback.
The
.code deffi-cb
macro generates a callback which uses safe dispatch, which is explained
in the description of the
.code ffi-make-closure
function. The optional
.meta abort-val
parameter specifies an expression which evaluates to the value
to be returned by the callback in the event that a dynamic control
transfer is intercepted. The purpose of this value is to indicate
to the foreign code that the callback wishes to abort operation;
it is useful in situations when a suitable return value will induce
the foreign code to cooperate and itself return to the Lisp code
which will then continue the dynamic control transfer.
The
.code deffi-cb-unsafe
macro is a variant of
.code deffi-cb
with the same argument conventions. The difference is that it arranges for
.code ffi-make-closure
to be invoked with
.code nil
for the
.meta safe-p
parameter. This macro has no
.meta abort-val
parameter, since unsafe callbacks do not use it.
.TP* Example:
.verb
;; create a closure combinator which binds
;; Lisp functions to a call descriptor has the C type
;; signature void (*)(int).
(deffi-cb void-int-closure void (int))
;; use the combinator
;; some-foreign-function's second arg is
;; of type closure, specifying a callback:
(some-foreign-function
42
(void-int-closure (lambda (x)
(puts `callback! @x`))))
.brev
.coNP Macro @ deffi-var
.synb
.mets (deffi-var < name < var-expr << type )
.syne
.desc
The
.code deffi-var
macro defines a global symbol macro which expands to an expression
accessing a foreign variable, creating the illusion that the
variable is available as a Lisp variable holding a Lisp data type.
The
.meta name
argument gives the name of the symbol macro to be defined.
The
.meta var-expr
argument is one of several permitted syntactic forms
which specify the address of the foreign variable.
They are described below.
The
.meta type
argument expresses the variable type in FFI type syntax.
Once the variable is defined, accessing the macro symbol
.meta name
performs a get operation on the foreign variable, yielding
the conversion of that variable to a Lisp value.
An assignment to the symbol performs a put operation,
converting a Lisp object to a value which overwrites
the object.
Note: FFI memory management is not helpful in the use of
variables. Suppose a string value is
stored in a variable of type
.codn str .
This means that FFI dynamically allocates a buffer which
stores the UTF-8 encoded version of the string, and this
buffer is placed into the foreign variable.
Then suppose another such assignment takes place.
The previous value is simply overwritten without being
freed.
The following syntactic variants are permitted of the
.meta var-expr
argument:
.RS
.meIP < name-string
If
.meta var-expr
is a literal string, then the
.code deffi-var
form must be enclosed in the
.code with-dyn-lib
macro, appearing as one of that macro's
.metn body-form s.
In this situation the literal character string
.meta name-string
specifies a symbol to be found within the library established by the
.meta with-dyn-lib
macro.
.meIP >> ( name-string << ver-string )
This manner of specifying the
.meta fun-expr
also requires the
.code deffi
form to be enclosed in a
.codn with-dyn-lib .
It selects a particular version of a symbol from the library.
.meIP < form
If
.meta var-expr
is any other form, then it must specify an expression which evaluates to a
.code cptr
object giving the address of a foreign library symbol. If this form
is used, then the
.code deffi
form need not be surrounded by a call to the
.code with-dyn-lib
macro.
.RE
.coNP Macro @ deffi-sym
.synb
.mets (deffi-sym < name < var-expr <> [ type-sym ])
.syne
.desc
The
.code deffi-sym
macro defines a global lexical variable called
.code name
whose value is a
.code cptr
object that refers to a symbol in a foreign library.
The
.meta name
argument gives the name for the variable to be defined.
This definition takes place place as if by the
.code defparml
macro.
The
.meta var-expr
is syntax which specifies the foreign pointer, using exactly the same
conventions as described for the
.code deffi-var
macro, allowing for a shorthand notation if this form is
enclosed in a
.code with-dyn-lib
macro invocation.
The optional
.meta type-sym
argument must be a symbol. If it is absent, it defaults to nil.
This argument specifies the type label for the
.code cptr
object which holds the pointer to the foreign symbol.
The result value of
.meta deffi-sym
is the symbol
.metn name .
.coNP Macro @ typedef
.synb
.mets (typedef < name << type-syntax )
.syne
.desc
The
.code typedef
macro provides a convenient way to define type aliases.
The
.meta type-syntax
expression is compiled as FFI syntax, and the
.meta name
symbol is installed as an alias denoting that type.
The
.code typedef
macro yields the compiled version of
.meta type-syntax
as its value.
.coNP Macros @ deffi-struct and @ deffi-union
.synb
.mets (deffi-struct < name >> {( slot < type <> [ init-form ])}*)
.mets (deffi-union < name >> {( slot < type <> [ init-form ])}*)
.syne
.desc
The
.code deffi-struct
and
.code deffi-union
macros provide a more compact notation for defining FFI structure and union
types together with matching
.code typedef
names.
The semantics follows from these equivalences:
.verb
(deffi-struct S ...) <--> (typedef S (struct S ...))
(deffi-union U ...) <--> (typedef U (union U ...))
.brev
.TP* Example:
.verb
(deffi-struct point
(x double)
(y double))
.brev
.coNP Macro @ sizeof
.synb
.mets (sizeof < type-syntax <> [ object-expr ])
.syne
.desc
The macro
.code sizeof
calculates the size of the FFI type denoted by
.codn type-syntax .
The
.meta type-syntax
expression is compiled to a type using
.codn ffi-type-compile .
The
.meta object-expr
expression is evaluated to an object value.
If
.code type-syntax
denotes an incomplete array or structure type, and the
.meta object-expr
argument is present, then a
.I "dynamic size" is computed: the actual number of bytes required to store
that object value as a foreign representation.
The
.code sizeof
macro arranges for the size calculation to be carried out at macro-expansion
time, if possible, so that the
.code sizeof
form is replaced by an integer constant. This is possible when the
.meta object-expr
is omitted, or if it is a constant expression according to the
.code constantp
function.
For the type
.codn void ,
incomplete array types, and bitfield types, the one-argument form of
.code sizeof
reports zero.
For incomplete structure types, the one-argument
.code sizeof
reports a size which is equivalent to the offset of the last member.
The size of an incomplete structure does not include padding
for the most strictly aligned member.
.coNP Macro @ alignof
.synb
.mets (alignof << type-syntax )
.syne
.desc
The macro
.code alignof
calculates the alignment of the FFI type denoted by
.code type-syntax
at macro-expansion time, and produces that
integer value as its expansion, such that there is no
run-time computation. It uses the
.code ffi-alignof
function.
.coNP Macro @ offsetof
.synb
.mets (offsetof < type-syntax << member-name )
.syne
.desc
The macro
.code sizeof
calculates the offset of the structure member indicated by
.metn member-name ,
a symbol, inside the FFI struct type indicated by
.metn type-syntax .
This calculation is performed by a macro-expansion-time call to the
.code ffi-offsetof
function, and produces that
integer value as its expansion, such that there is no
run-time computation.
.coNP Macro @ arraysize
.synb
.mets (arraysize << type-syntax )
.syne
.desc
The macro
.code arraysize
calculates the number of elements of the array type indicated by
.metn type-syntax .
This calculation is performed by a macro-expansion-time call to the
.code ffi-arraysize
function, and produces that
integer value as its expansion, such that there is no
run-time computation.
.coNP Macro @ elemsize
.synb
.mets (elemsize << type-syntax )
.syne
.desc
The macro
.code elemsize
calculates the size of the element type of an array type, or
the size of target type of a pointer type indicated by
.metn type-syntax .
This calculation is performed by a macro-expansion-time call to the
.code ffi-elemsize
function, and produces that
integer value as its expansion, such that there is no
run-time computation.
.coNP Macro @ elemtype
.synb
.mets (elemtype << type-syntax )
.syne
.desc
The macro
.code elemtype
produce the element type of an array type, or
the target type of a pointer type indicated by
.metn type-syntax .
Note: the
.code elemtype
macro may be understood in terms of several possible implementations.
The form
.code "(elemtype X)"
is equivalent to
.codn "(ffi-elemtype (ffi-type-compile X))" .
Since there exists an
.code elemtype
type operator, the expression is also equivalent to
.codn "(ffi-type-compile '(elemtype X))" .
.coNP Macro @ ffi
.synb
.mets (ffi << type-syntax )
.syne
.desc
The
.code ffi
macro provides a shorthand notation for compiling a literal
FFI type expression to the corresponding type object. The
following equivalence holds:
.verb
(ffi expr) <--> (load-time (ffi-type-compile 'expr))
.brev
.SS* Zero-filled Object Support
Communicating with foreign interfaces sometimes requires representations
to be initialized consisting of all zero bits, or mostly zero bits.
\*(TX provides convenient ways to prepare Lisp objects such that when those
objects are converted to a foreign representation, they generate zero-filled
representations.
.coNP Function @ make-zstruct
.synb
.mets (make-zstruct < type >> { slot-sym << init-value }*)
.syne
.desc
The
.code make-zstruct
function provides a convenient means of instantiating a structure
for use in foreign function calls, imitating a pattern of initialization
often seen in the C language. It instantiates a Lisp
.code struct
by conversion of zero-filled memory through FFI, thus creating a Lisp
structure which appears zero-filled when converted to the foreign representation.
This simplifies application code, which is spared from providing individual
slot initializations which have this effect.
The
.meta type
argument must be a compiled FFI
.code struct
type. The remaining arguments must occur pairwise. Each
.meta slot-sym
argument must be a symbol naming a slot in the FFI
.code struct
type. The
.meta init-value
argument which follows it specifies the value for that
slot.
The
.code make-zstruct
function operates as follows. Firstly, the Lisp
.code struct
type is retrieved which corresponds to the FFI type given by
.metn type .
A new instance of the Lisp type is instantiated, as if by
a one-argument call to
.codn make-struct .
Next, each slot indicated by a
.meta slot-sym
argument is set to the corresponding
.metn init-value .
Finally, each slot of the struct which is not initialized via
.meta slot-sym
and
.meta init-value
pair, and which is known to the FFI type, is reinitialized by a conversion
from a foreign object of all-zero bits to a Lisp value.
argument. The
.code struct
object is then returned.
Note: the
.code znew
macro provides a less verbose notation based on
.codn make-zstruct .
Note: slots which are not known to the FFI
.code struct
type may be initialized by
.codn make-zstruct .
Each
.meta slot-sym
must be a slot of the Lisp
.code struct
type; but need not be declared as a member in the FFI
.code struct
type.
.coNP Macro @ znew
.synb
.mets (znew < type-syntax >> { slot-sym << init-value }*)
.syne
.desc
The
.code znew
macro provides a convenient way of using
.codn make-zstruct ,
using syntax which resembles that of the
.code new
macro.
The
.code znew
macro generates a
.code make-zstruct
call, arranging for the
.meta type-syntax
argument to be compiled to a FFI type object, and
applies quoting to every
.meta slot-sym
argument.
The following equivalence holds:
.verb
(znew s a i b j ...) <--> (make-zstruct (ffi s)
'a i 'b j ...)
.brev
.TP* Example
Given the following FFI type definition
.verb
(typedef foo (struct foo (a (cptr bar)) (b uint) (c bool)))
.brev
the following results are observed:
.verb
;; ordinary instantiation
(new foo) -> #S(foo a nil b nil c nil)
;; Under znew, a is null cptr of correct type:
(znew foo) -> #S(foo a #<cptr bar: 0> b 0 c nil)
;; value of b is specified; others come from zeros:
(znew foo b 42) -> #S(foo a #<cptr bar: 0> b 42 c nil)
.brev
.coNP Function @ zero-fill
.synb
.mets (zero-fill < type << obj )
.syne
.desc
The
.code zero-fill
function invokes the by-reference in semantics of FFI type
.meta type
against a zero-filled buffer, and a Lisp object
.metn obj .
This means that if
.meta obj
is an aggregate such as a vector, list or structure,
it is updated as if from an all-zero-bit foreign representation.
In that situation,
.meta obj
is also returned.
An object which has by-value semantics, such as an integer,
is not updated. In this case, nevertheless, the return value
is a Lisp object produced by converting an all-zero-bit buffer to
.metn type .
.SS* Foreign Unions
The following group of functions provides the means for working
with foreign unions, in conjunction with the
.code union
FFI type.
.coNP Function @ make-union
.synb
.mets (make-union < type >> [ initval <> [ member ]])
.syne
.desc
The
.code make-union
function instantiates a new object of type
.codn union ,
based on the FFI type specified by the
.meta type
parameter, which must be compiled FFI
.code union
type.
The object provides storage for the foreign representation of
.codn type ,
and that storage is initialized to all zero bytes.
Additionally, if
.meta initval
is specified, but
.meta member
is not, then
.meta initval
is stored into the union's via the first member, as if by
.codn union-put .
If the union type has no members, an error exception is thrown.
If both
.meta initval
and
.meta member
are specified, then
.meta initval
is stored into the union using the specified member, as if by
.codn union-put .
.coNP Function @ union-members
.synb
.mets (union-members << union )
.syne
.desc
The
.code union-members
function retrieves the list of symbols which name the members of
.metn union .
These are derived from the object's FFI type.
It is unspecified whether the list is freshly allocated on each call,
or whether the same list is returned; applications shouldn't
destructively manipulate this list.
.coNP Function @ union-get
.synb
.mets (union-get < union << member )
.syne
.desc
The
.code union-get
function performs the get semantics (conversion from a foreign
representation to Lisp) on the member of
.meta union
which is specified by the
.meta member
argument. That argument must be a symbol corresponding to one of the member
names.
The
.meta union
object's storage buffer is treated as an object of the foreign
type indicated by that member's type information, and converted
accordingly to a Lisp object that is returned.
.coNP Function @ union-put
.synb
.mets (union-put < union < member << new-value )
.syne
.desc
The
.code union-put
function performs the put semantics (conversion from a Lisp object
to foreign representation) on the member of
.meta union
which is specified by the
.meta member
argument. That argument must be a symbol corresponding to one of the member
names.
The object given as
.meta new-value
is converted to the foreign representation according to the type
information of the indicated member, and that representation is
placed into the
.meta union
object's storage buffer.
The return value is
.metn new-value .
.coNP Functions @ union-in and @ union-out
.synb
.mets (union-in < union < memb << memb-obj )
.mets (union-out < union < memb << memb-obj )
.syne
.desc
The
.code union-in
and
.code union-out
functions perform the FFI in semantics and out semantics, respectively.
These semantics are involved in two-way data transfers between foreign
representations and Lisp objects.
The
.meta union
argument must be a
.code union
object and the
.meta memb
argument a symbol which matches one of that object's member names.
In the case of
.codn union-in ,
.meta memb-obj
is a Lisp object that was previously stored into
.meta union
using the
.code union-put
operation, into the same member that is currently indicated by
.metn member .
In the case of
.codn union-out ,
.meta memb-obj
is a Lisp object that was previously retrieved from
.meta union
using the
.code union-get
operation, from the same member that is currently indicated by
.metn member .
The
.code union-in
performs the by-value nuance of the in semantics on the indicated
member: if the member contains pointers to any objects, those
objects are updated from their counterparts in
.meta memb-obj
using their respective by-reference in semantics, recursively.
Similarly
.code union-out
performs the by-value nuance of the out semantics on the indicated
member: if the member contains pointers to any objects, those
objects are updated with their Lisp counterparts in
.meta memb-obj
using their respective by-reference out semantics, recursively.
Note:
.code union-in
is intended to be used after a FFI call, on a union-typed by-value
argument, or a union-typed object contained in an argument,
in situations when the function is expected to have updated
the contents of the union. The
.code union-out
function is intended to be used in a FFI callback, on a union-typed
callback argument or union-typed object contained in such
an argument, in cases when the callback has updated the Lisp
object corresponding to a union member, and that change needs
to be propagated to the foreign caller.
.SS* FFI-type-driven I/O Functions
These functions provide a way to perform I/O on stream using the foreign
representation of Lisp objects, performing conversion between the Lisp
representations in memory and the foreign representations in a stream.
The
.meta stream
argument used with these functions must be a stream object which,
in the case of input functions, supports
.code get-byte
and, in the case of output, supports
.codn put-byte .
.coNP Function @ put-obj
.synb
.mets (put-obj < object < type <> [ stream ])
.syne
.desc
The
.code put-obj
function encodes
.meta object
into a foreign representation, according to the FFI type
.metn type .
The bytes of the foreign representation are then written to
.metn stream .
If
.meta stream
is omitted, it defaults to
.codn *stdout* .
If the operation successfully writes all bytes of the representation to
.metn stream ,
the value
.code t
is returned. A partial write causes the return value to be
.codn nil .
All other stream error situations throw exceptions.
.coNP Function @ get-obj
.synb
.mets (get-obj < type <> [ stream ])
.syne
.desc
The
.code get-obj
function reads from
.meta stream
the bytes corresponding to a foreign representation according to the FFI type
.metn type .
If
.meta stream
is omitted, it defaults to
.codn *stdin* .
If the read is successful, these bytes are decoded, producing a Lisp
object, which is returned.
If the read is incomplete, the value returned is
.metn nil .
All other stream error situations throw exceptions.
.coNP Function @ fill-obj
.synb
.mets (fill-obj < object < type <> [ stream ])
.syne
.desc
The
.code fill-obj
function reads from
.meta stream
the bytes corresponding to a foreign representation according to the FFI type
.metn type .
If the read is successful, then
.meta object
is updated, if possible, from that representation, using the by-value in
semantics of the FFI type and returned. If a by-value update of
.meta object
isn't possible, then a new object is decoded from the data and returned.
If the read is incomplete, the value returned is
.metn nil .
All other stream error situations throw exceptions.
.SS* Buffer Functions
Functions in this area provide a way to perform conversion between
Lisp objects and foreign representation to and from objects of the
.code buf
type.
.coNP Functions @ ffi-put and @ ffi-put-into
.synb
.mets (ffi-put < obj << type )
.mets (ffi-put-into < dst-buf < obj < type <> [ offset ])
.syne
.desc
The
.code ffi-put
function encodes the Lisp object
.meta obj
according to the FFI type
.meta type
and returns a new buffer object of type
.code buf
which holds the foreign representation.
The
.code ffi-put-into
function is similar, except that it uses an existing buffer
.meta dst-buf
which must be large enough to hold the foreign representation.
The
.meta type
argument must be a compiled FFI type.
If
.meta type
is has a variable length, then the actual size of the foreign representation is
calculated from
.metn obj .
The
.meta obj
argument must be an object compatible with the conversions
implied by
.metn type .
The optional
.meta offset
argument specifies a byte offset from the beginning of the data area of
.meta dst-buf
where the foreign representation of
.meta obj
is stored. The default value is zero.
These functions perform the "put semantics" encoding action similar to
what happens to the arguments of an outgoing foreign function call.
Caution: incorrect use of this function, or its use in isolation
without a matching
.code ffi-in
call, can cause memory leaks, because, depending on
.metn type ,
temporary resources may be allocated, and pointers to those resources
will be stored in the buffer.
.coNP Function @ ffi-out
.synb
.mets (ffi-out < dst-buf < obj < type < copy-p <> [ offset ])
.syne
.desc
The
.code ffi-out
function performs the "out semantics" encoding action, similar
to the treatment applied to the arguments of a callback prior to
returning to foreign code.
It is assumed that
.code obj
is an object that was returned by an earlier call to
.codn ffi-get ,
and that the
.meta dst-buf
and
.meta type
arguments are the same objects that were used in that call.
The
.meta copy-p
argument is a Boolean flag which is true if the buffer represents a datum
that is being passed by pointer. If
.meta copy-p
is true, then
.meta obj
is converted to a foreign representation which is stored into
.metn dst-buf .
If it is false, it indicates that the buffer itself is a pass-by-value object.
This means that the object itself will not be copied, but if it is an aggregate
which contains pointers, the operation will recurse on those objects, invoking
their "out semantics" action with pass-by-pointer semantics. The required
pointers to these indirect objects are obtained from
.metn dst-buf .
The optional
.meta offset
argument specifies a byte offset from the beginning of the data area of
.meta dst-buf
where the foreign representation of
.meta obj
is understood to be stored, and where it is updated if requested by
.metn copy-p .
The default value is zero.
The
.code ffi-out
function returns
.metn dst-buf .
.coNP Function @ ffi-in
.synb
.mets (ffi-in < src-buf < obj < type < copy-p <> [ offset ])
.syne
.desc
The
.code ffi-in
function performs the "in semantics" decoding action, similar to the
treatment applied to the arguments of a foreign function call after
it returns, in order to free temporary resources and recover the new
values of objects that have been modified by the foreign function.
It is assumed that
.meta src-buf
is a buffer that was prepared by a call to
.code ffi-put
or
.codn ffi-put-into ,
and that
.meta type
and
.meta obj
are the same values that were passed as the
corresponding arguments of those functions.
The
.code ffi-in
function releases the temporary memory resources that were allocated by
.code ffi-put
or
.codn ffi-put-into ,
which are obtained from the buffer itself, where they appear as pointers.
The function recursively performs the in semantics across the entire type,
and the entire object graph rooted at the buffer.
The
.meta copy-p
argument is a Boolean flag which is true if the buffer represents a datum
that is being passed by pointer. If it is false, it indicates that the
buffer itself is a pass-by-value object. Under pass-by-pointer semantics,
either a whole new object is extracted from the buffer and returned,
or else the slots of
.meta obj
are updated with new values from the buffer.
Under pass-by-value semantics, no such extraction takes place, and
.meta obj
is returned.
However, regardless of the value of
.codn copy-p ,
if the object is an aggregate which contains pointers, the recursive
treatment through those pointers involves pass-by-pointer semantics.
This is consistent with the idea that we can pass a structure by value,
but that structure can have pointers to objects which are updated by
the called function. Those indirect objects are passed by pointer.
They get updated, but the parent structure cannot.
If
.meta type
is has a variable length, then the actual size of the foreign representation is
calculated from
.metn obj .
The optional
.meta offset
argument specifies a byte offset from the beginning of the data area of
.meta src-buf
from which the foreign representation of
.meta obj
is taken.
The
.code ffi-in
function returns either
.meta obj
or a new object which is understood to have been produced as its
replacement.
.coNP Function @ ffi-get
.synb
.mets (ffi-get < src-buf < type <> [ offset ])
.syne
.desc
The
.code ffi-get
function extracts a Lisp value from buffer
.meta src-buf
according to the FFI type
.metn type .
The
.meta src-buf
argument is an object of type
.meta buf
large enough to hold a foreign representation of
.metn type ,
at the byte offset indicated by the
.meta offset
argument.
The
.meta type
argument is compiled FFI type.
The optional
.meta offset
argument defaults to zero.
The external representation in
.meta src-buf
at the specified offset is scanned according to
.meta type
and converted to a Lisp value which is returned.
The
.code ffi-get
operation is similar to the "get semantics" performed by FFI
in order to extract the return value of foreign function
calls, and by the FFI callback mechanism to extract the
arguments coming into a callback.
The
.meta type
argument may not be a variable length type, such as an array of
unspecified size.
.SS* Foreign Arrays
Functions in this area provide a means for working with
foreign arrays, in connection with the FFI
.code carray
type.
.coNP Functions @ carray-vec and @ carray-list
.synb
.mets (carray-vec < vec < type <> [ null-term-p ])
.mets (carray-list < list < type <> [ null-term-p ])
.syne
.desc
The
.code carray-vec
and
.code carray-list
functions allocate storage for the representation of a foreign array, and
return a
.code carray
object which holds a pointer to that storage.
The argument
.metn type ,
which must be a compiled FFI type,
is retained as the
.code carray
object's element type.
Prior to returning, the functions
initializes the foreign array by converting the elements of
.meta vec
or, respectively,
.meta list
into elements of the foreign array.
The conversion is performed using the put semantics of
.metn type ,
which is a compiled FFI type.
The length of the returned
.code carray
is determined from the length of
.meta vec
or
.meta list
and from the value of the Boolean argument
.metn null-term-p .
If
.meta null-term-p
is
.codn nil ,
then the length of the
.code carray
is the same as that of the input
.meta vec
or
.metn list .
A true value of
.meta null-term-p
indicates null termination.
This causes the length of the
.code carray
to be one greater than that of
.meta vec
or
.metn list ,
and the extra element allocated to the foreign array is filled with zero bytes.
.coNP Function @ carrayp
.synb
.mets (carrayp << object )
.syne
.desc
The
.code carrayp
function returns
.code t
if
.meta object
is a
.codn carray ,
otherwise it returns
.codn nil .
.coNP Function @ carray-blank
.synb
.mets (carray-blank < length << type )
.syne
.desc
The
.code carray-blank
function allocates storage for the representation of a foreign array,
filling that storage with zero bytes, and returns a
.code carray
object which holds a pointer to that storage.
The argument
.metn type ,
which must be a compiled FFI type,
is retained as the
.code carray
object's element type.
The
.meta length
argument must be a nonnegative integer; it specifies the number
of elements in the foreign array and is retained as the
.code carray
object's length.
The size of the foreign array is the product of the size of
.meta type
as reported by the
.code ffi-size
function, and of
.metn length .
.coNP Function @ carray-buf
.synb
.mets (carray-buf < buf < type <> [ offset ])
.syne
.desc
The
.code carray-buf
function creates a
.code carray
object which refers to the storage provided and managed by the buffer object
.metn buf ,
providing a view of that storage, and manipulation thereof, as an array.
The optional
.meta offset
parameter specifies an offset from the start of the buffer to the
location which is interpreted as the start of the
.codn carray ,
which extends from that offset to the end of the buffer.
The default value is zero: the
.code carray
covers the entire buffer.
If a value is specified, it must be in the range zero to the length of
.metn buf .
The
.meta type
argument must be a compiled FFI type whose size is nonzero.
The
.code carray
is overlaid onto the storage of
.meta buf
as follows:
First,
.meta offset
is subtracted from the bytewise length of
.metn buf ,
as reported by
.code length-buf
function to produce the effective length of the storage to be used for the array.
The effective length is divided by the size of
.metn type ,
as reported by
.codn ffi-size .
The resulting quotient represents the length (number of elements) of the
.code carray
object.
Note: the returned
.code carray
object holds a reference to
.metn buf ,
preventing
.meta buf
from being reclaimed by garbage collection, thereby protecting the
underlying storage from becoming invalid. A subsequent invocation of
.code carray-own
operation releases this reference.
Note: the relationship between the
.code carray
object and
.meta buf
is inherently unsafe: if
.meta buf
is subsequently subject to operations which reallocate the storage,
such as
.code buf-set-length
the pointer stored inside the referencing
.code carray
object becomes invalid, and operations involving that pointer
have undefined behavior.
Note: if the length of the buffer is not evenly divisible by the size of the
type, the calculated number of elements is rounded down. The trailing portion
of the buffer corresponding to the division remainder, being insufficient
to constitute a whole array element, is excluded from the array view.
.coNP Function @ carray-buf-sync
.synb
.mets (carray-buf-sync << carray )
.syne
.desc
The
.code carray-buf-sync
function requires
.meta carray
to be a
.code carray
object which refers to a
.code buf
object for its storage. Such objects are created by the function
.codn carray-buf .
The
.code carray-buf-sync
function retrieves and returns the buffer object associated with
.meta carray
and at the same time also updates the internal properties of
.meta carray
using the current information: the pointer to the data, and the
length of
.meta carray
are altered to reflect the current state of the buffer.
.coNP Function @ buf-carray
.synb
.mets (buf-carray << carray )
.syne
.desc
The
.code buf-carray
function duplicates the underlying storage of
.meta carray
and returns that storage represented as an object of
.code buf
type.
The storage size is calculated by multiplying the
.code carray
object's element size by the number of elements.
Only that extent of the storage is duplicated.
.coNP Function @ carray-cptr
.synb
.mets (carray-cptr < cptr < type <> [ length ])
.syne
.desc
The
.code carray-cptr
function creates a
.code carray
object based on a pointer derived from a
.code cptr
object.
The
.meta cptr
argument must be of type
.codn cptr .
The object's
.code cptr
type tag is ignored.
The
.meta type
argument must specify a compiled FFI type, which will become
the element type of the returned
.codn carray .
If
.meta length
is specified as
.codn nil ,
or not specified,
then the returned
.code carray
object will be of unknown length. Otherwise,
.meta length
must be a nonnegative integer which will be taken as the
length of the array.
Note: this conversion is inherently unsafe.
.coNP Function @ cptr-carray
.synb
.mets (cptr-carray < carray <> [ type-symbol ])
.syne
.desc
The
.code cptr-carray
function returns a
.code cptr
object which holds a pointer to a
.code carray
object's storage area. The
.meta carray
argument must be of type
.codn carray .
The
.meta type-symbol
argument should be a symbol. If omitted, it defaults to
.codn nil .
This symbol becomes the
.code cptr
object's type tag.
The lifetime of the returned
.code cptr
object is independent from that of
.metn carray .
If the lifetime of
.meta carray
reaches its end before that of the
.codn cptr ,
the pointer stored inside the
.code cptr
becomes invalid.
.coNP Function @ length-carray
.synb
.mets (length-carray << carray )
.syne
.desc
The
.code length-carry
function returns the length of the
.meta carray
argument, which must be an object of type
.codn carray .
If
.meta carray
has an unknown length, then
.code nil
is returned.
.coNP Function @ copy-carray
.synb
.mets (copy-carray << carray )
.syne
.desc
The
.code copy-carray
function returns a duplicate of
.metn carray .
The duplicate has the same element type and length, but has its own
copy of the underlying storage. This is true whether or not
.meta carray
owns its storage or not. In either case, the duplicate owns
.I its
copy of the storage.
.coNP Function @ carray-set-length
.synb
.mets (carray-set-length < carray << length )
.syne
.desc
The
.code carry-set-length
attempts to change the length of
.metn carray ,
which must be an object of
.code carray
type.
The
.meta length
argument indicates the new length, which must be
a nonnegative integer.
The operation throws an
.code error
exception if
.meta length
is negative.
An
.code error
exception is also thrown if
.meta carray
is an object which owns the underlying storage. There is no provision in the
.code carray
type to change the storage size.
It is permissible to change the length of a
.code carray
object which acts as a view into a buffer (as constructed via the
.code carray-buf
operation).
This creates a potentially unsafe situation in which the length requires
a larger amount of backing storage than is provided by the buffer.
.coNP Accessor @ carray-ref
.synb
.mets (carray-ref < carray << idx )
.mets (set (carray-ref < carray << idx ) << new-val )
.syne
.desc
The
.code carray-ref
function accesses an element of the foreign array
.metn carray ,
converting that element to a Lisp value, which is returned.
The
.meta idx
argument must be a nonnegative integer. If
.meta carray
has a known length,
.meta idx
must be less than the length.
If
.meta carray
has an unknown length, then the access is permitted regardless of how
positive is the value of
.metn idx .
Whether the access has well-defined behavior depends on the actual extent of
the underlying array storage.
The validity of any access to the underlying storage depends on the
validity of the pointer to that storage.
The access to the array storage proceeds as follows. Every
.code carray
object has an element type, which is a compiled FFI type.
A byte offset
address is calculated by multiplying the size of the element type of
.meta carray
by
.metn idx .
Then, the get semantics of the element type is invoked to convert, to
a Lisp object, a region of data starting at calculated byte offset in the array
storage. The resulting object is returned.
Assigning an a value to a
.code carray-ref
form is equivalent to using
.code carray-refset
to store the value.
.coNP Function @ carray-refset
.synb
.mets (carray-refset < carray < idx << new-val )
.syne
.desc
The
.code carray-refset
function accesses an element of the foreign array
.metn carray ,
overwriting that element with a new value obtained
from a conversion of the Lisp value
.metn new-val .
The return value is
.metn new-val .
The
.meta idx
argument must be a nonnegative integer. If
.meta carray
has a known length,
.meta idx
must be less than the length.
If
.meta carray
has an unknown length, then the access is permitted regardless of how
positive is the value of
.metn idx .
Whether the access has well-defined behavior depends on the actual extent of
the underlying array storage.
The validity of any access to the underlying storage depends on the
validity of the pointer to that storage.
The access to the array storage proceeds as follows. Every
.code carray
object has an element type, which is a compiled FFI type.
A byte offset
address is calculated by multiplying the size of the element type of
.meta carray
by
.metn idx .
Then, the put semantics of the element type is invoked to convert
.meta new-val
to a foreign representation, which is written into the array storage
started at the calculated byte offset.
If
.meta new-val
has a type which is not compatible with the element type, or a value which
is out of range or otherwise unsuitable, an exception is thrown.
.coNP Functions @ carray-dup and @ carray-own
.synb
.mets (carray-dup << carray )
.mets (carray-own << carray )
.syne
.desc
The
.code carray-dup
function acts upon a
.code carray
object which doesn't own its underlying array storage.
It allocates a duplicate copy of the array storage referenced by
.metn carray ,
and assigns to
.meta carray
the new copy. Then it marks
.meta carray
as owning that storage. Lastly, if
.meta carray
references another object, that reference is removed;
.meta carray
no longer prevents the other object from being reclaimed by
the garbage collector.
If
.meta carray
already owns its storage, then this function has no effect.
If
.meta carray
has an unknown size, then an error exception is thrown.
A
.code carray
produced by the functions
.code carray-vec
or
.code carray-blank
already owns its storage.
A
.code carray
object does not own its storage if it is produced by
.code carray-buf
or by the conversion of a foreign pointer under the control of the
.code carray
FFI type.
Because
.code carray
objects derived from foreign pointers via FFI have an unknown size,
before using
.codn carray-dup ,
the application must determine the length of the array, and call
.code carray-set-length
to establish that length.
After
.codn carray-dup ,
the length may not be altered.
The
.code carray-dup
function returns
.code t
if it has performed the duplication operation. If it has done
nothing, it returns
.codn nil .
The
.code carray-own
function resembles
.codn carray-dup ,
differing from that function only in two ways.
Instead of allocating a duplicate copy of the underlying array storage,
.code carray-own
causes
.meta carray
to
.B assume
ownership of the existing storage. Secondly, it is an error to use
.code carray-own
on a
.meta carray
which references a buffer object.
The
.meta carray-own
function always returns
.codn nil .
In all other regards, the descriptions of
.code carray-dup
apply to
.codn carray-own .
.coNP Function @ carray-free
.synb
.mets (carray-free << carray )
.syne
.desc
If
.meta carray
is a
.code carray
object which owns the storage to which it refers, then
.code carray-free
function liberates that storage by passing the pointer to the C
library function
.codn free .
It then replaces that pointer with a null pointer, and
changes the size to zero.
If
.meta carray
doesn't own the storage, an exception is thrown.
.coNP Function @ carray-type
.synb
.mets (carray-type << carray )
.syne
.desc
The
.code carray-type
function returns the element type of
.metn carray ,
a compiled FFI type.
.coNP Functions @ vec-carray and @ list-carray
.synb
.mets (vec-carray < carray <> [ null-term-p ])
.mets (list-carray < carray <> [ null-term-p ])
.syne
.desc
The
.code vec-carray
and
.code list-carray
functions convert the array storage of
.meta carray
to a freshly constructed object representation: vector, and list, respectively.
The new vector or list is returned.
The
.meta carray
object must have a known size; an
.code error
exception is thrown if these functions are invoked on a
.code carray
object of unknown size.
The effective length of the new vector or list is derived
from the length of
.metn carray ,
taking into account the value of
.metn null-term-p .
The
.meta null-term-p
Boolean parameter defaults to
.codn nil .
If specified as true, then it has the effect that the
effective length of the returned vector or list is
one less than that of
.metn carray :
in other words, a true value of
.meta null-term-p
indicates that
.meta carray
holds storage which represents a null-terminated array, and the
terminating null element is to be excluded from the conversion.
If
.meta null-term-p
is true, but the length of
.meta carray
is already zero, then it has no effect; the effective length remains zero,
and a zero-length vector or list is returned.
Conversion of the foreign array to the vector or list is performed
by iterating over all of its elements, starting from element zero, up to the
element before the effective length.
.coNP Functions @ carray-get and @ carray-getz
.synb
.mets (carray-get << carray )
.mets (carray-getz << carray )
.syne
.desc
The
.code carray-get
and
.code carray-getz
functions treat the contents of
.meta carray
as a FFI
.code array
and
.code zarray
type, respectively.
They invoke the get semantics to convert the FFI array to a Lisp
object, and return that object.
If the element type is one of
.codn char ,
.code bchar
or
.codn wchar ,
then the expected string conversion semantics applies.
.coNP Functions @ carray-put and @ carray-putz
.synb
.mets (carray-put < carray << new-val )
.mets (carray-putz < carray << new-val )
.syne
.desc
The
.code carray-put
and
.code carray-putz
functions treat the contents of
.meta carray
as a FFI
.code array
and
.code zarray
type, respectively.
They invoke the put semantics to convert the Lisp object
.meta new-val
array to the foreign array representation, which is placed into
the array storage referenced by
.metn carray .
If the element type is one of
.codn char ,
.code bchar
or
.codn wchar ,
then the expected string conversion semantics applies.
Both of these functions return
.metn carray .
.coNP Accessor @ carray-sub
.synb
.mets (carray-sub < carray >> [ from <> [ to ]])
.mets (set (carray-sub < carray >> [ from <> [ to ]]) << new-val )
.syne
.desc
The
.code carray-sub
function extracts a subrange of a
.meta carray
object, returning a new
.code carray
object denoting that subrange.
The semantics of
.meta from
and
.meta to
work exactly like the corresponding arguments of the
.code sub
accessor, following the same conventions.
The returned
.code carray
shares the array has the same element type as the original and
shares the same array storage. If, subsequently, elements of the
original array are modified which lie in the range, then the
modifications will affect the previously returned subrange
.codn carray .
The returned
.code carray
references the original object, to ensure that as long as the returned object
is reachable by the garbage collector, so is the original. This relationship
can be severed by invoking
.code carray-dup
on the returned object, after which the two no longer share storage,
and modifications in the original are not reflected in the subrange.
If
.code carray-sub
is used as a syntactic place, the argument expressions
.metn carray ,
.metn from ,
.meta to
and
.meta new-val
are evaluated just once. The prior value, if required, is accessed by calling
.code carray-sub
and
.meta new-val
is then stored via
.codn carray-replace .
.coNP Function @ carray-replace
.synb
.mets (carray-replace < carray < item-sequence >> [ from <> [ to ]])
.syne
.desc
The
.code carray-replace
function is a specialized version of
.code replace
which works on
.code carray
objects. It replaces a sub-range of
.meta carray
with elements from
.metn item-sequence .
The replacement sequence need not have the same length
as the range which it replaces.
The semantics of
.meta from
and
.meta to
work exactly like the corresponding arguments of the
.code replace
function, following the same conventions.
The semantics of the
.code carray-replace
operation itself differs from the
.code replace
semantics on sequences in one important regard: the
.code carray
object's length always remains the same.
The range indicated by
.meta from
and
.meta to
is deleted from
.meta carray
and replaced by elements of
.metn item-sequence ,
which undergo conversion to the foreign type that defines the
elements of
.metn carray .
If this operation would make the
.code carray
longer, any elements in excess of the object's length are discarded,
whether they are the original elements, or whether they come from
.metn item-sequence .
Under no circumstances does
.code carray-replace
write an element beyond the length of the underlying storage.
If this operation would make the
.meta carray
shorter (the range being replaced is longer than
.metn item-sequence )
then the downward relocation of items above the replacement range
creates a gap at the end of
.meta carray
which is filled with zero bytes.
The return value is
.meta carray
itself.
.coNP Function @ carray-pun
.synb
.mets (carray-pun < carray < type >> [ offset <> [ size-limit ]])
.syne
.desc
The
.code carray-pun
creates a new
.code carray
object which provides an aliased view of the same data that is referenced by
the original
.meta carray
object.
The
.meta type
argument specifies the element type used by the returned aliasing array.
If the
.meta offset
argument is specified, then the aliased view is displaced by that many
bytes from the start of the
.meta carray
object. The
.meta offset
argument must not be larger than the bytewise length of the array,
or an error exception is thrown. The bytewise length of the array
is the product of the number of elements and the element size.
The default value of
.meta offset
is zero: no displacement.
If
.meta size-limit
is specified, it indicates the size, in bytes, of the aliased view.
This limit must not be such that the aliased view would extend beyond the
array, or an error exception is thrown. If omitted,
.meta size-limit
defaults to the entire remainder of the array, after the offset.
The number of elements of the returned array are then calculated from
.metn size-limit .
The
.code carray-pun
function calculates how many elements of
.meta type
fit into
.metn size-limit .
This value becomes the length of the aliasing array which is returned.
Since the returned aliasing array and the original refer to the same
storage, modifications performed in one view are reflected in the other.
The aliasing array holds a reference to the original, so that as long as
it is reachable by the garbage collector, so is the original.
That relationship is severed if
.code carray-dup
is invoked on the aliasing array.
The meaning of the aliasing depends entirely on the bitwise representations of
the types involved.
Note:
.code carray-pun
does not check whether
.meta offset
is a value that is suitably aligned for accessing elements of
.metn type ;
on some platforms that must be ensured.
The
.code carray-pun
function may be invoked on an object that was itself returned by
.codn carray-pun .
.coNP Functions @ carray-uint and @ carray-int
.synb
.mets (carray-uint < number <> [ type ])
.mets (carray-int < number <> [ type ])
.syne
.desc
The
.code carray-uint
and
.code carray-int
functions convert
.metn number ,
an integer, to a binary image, which is then used as
the underlying storage for a
.codn carray .
The
.meta type
argument, a compiled FFI type, determines the element type for the returned
.codn carray .
If it is omitted, it defaults to the
.code uchar
type, so that the array is effectively of bytes.
Regardless of
.metn type ,
these functions first determine the number of bytes required to represent
.meta number
in a big endian format. Then the number of elements is determined for the
array, so that it provides at least as that many bytes of storage. The
representation of
.meta number
is then placed into this storage, such that its least significant byte
coincides with the last byte of that storage. If the number is smaller
than the storage provided by the array, it extended with padding bytes on the
left, near the beginning of the array.
In the case of
.codn carray-uint ,
.meta number
must be a nonnegative integer. An unsigned representation is produced
which carries no sign bit. The representation is as many bytes wide as
are required to cover the number up to its most-significant bit whose
value is 1. If any padding bytes are required due to the array being larger,
they are always zero.
The
.code carray-int
function encodes negative integers also, using a variable-length two's
complement representation. The number of bits required to hold the number
is calculated as the smallest width which can represent the value in two's
complement, including a sign bit. Any unused bits in the most-significant
byte are filled with copies of the sign bit: in other words, sign extension
takes place up to the byte size. The sign extension continues through the
padding bytes if the array is larger than the number of bytes required to represent
.metn number ;
the padding bytes are filled with the value
.code #b11111111
(255) if the number is negative, or else 0 if it is nonnegative.
.coNP Functions @ uint-carray and @ int-carray
.synb
.mets (uint-carray << carray )
.mets (int-carray << carray )
.syne
.desc
The
.code uint-carray
and
.code int-carray
functions treat the storage bytes
.meta carray
object as the representation of an integer.
The
.code uint-carray
function simply treats all of the bytes as a big-endian unsigned integer in
a pure binary representation, and returns that integer, which is necessarily
always nonnegative.
The
.code int-carray
function treats the bytes as a two's complement representation. The returned
number is negative if the first storage byte of
.meta carray
has a 1 in the most significant bit position: in other words, is in the
range
.code #x80
to
.codn #xFF .
In this case, the two's complement of the entire representation is calculated:
all of the bits are inverted, the resulting positive integer is extracted.
Then 1 is added to that integer, and it is negated. Thus, for example, if all
of the bytes are
.codn #xFF ,
the value -1 is returned.
.coNP Functions @ fill-carray and @ put-carray
.synb
.mets (fill-carray < carray >> [ pos <> [ stream ]])
.mets (put-carray < carray >> [ pos <> [ stream ]])
.syne
.desc
The
.code fill-carray
and
.code put-carray
functions perform stream output using the
.code carray
object as a buffer.
The semantics of these functions is as follows.
A temporary buffer is created which aliases the storage of
.meta carray
and this buffer is used as an argument in an invocation of, respectively,
the buffer I/O function
.code fill-buf
or
.codn put-buf .
The value returned by the buffer I/O function is returned.
The
.meta pos
and
.meta stream
arguments are defaulted exactly in the same manner as by
.code fill-buf
and
.codn put-buf ,
and have the same meaning. In particular,
.meta pos
indicates a byte offset into the
.code carray
object's storage, not an array index.
.SH* LISP COMPILATION
.SS* Overview
\*(TX supports two modes of processing of Lisp programs: evaluation and compilation.
Expressions entered into the listener, loaded from source files via
.codn load ,
processed by the
.code eval
function, or embedded into the \*(TX pattern language, are processed by the
.IR evaluator .
The evaluator expands all macros, and then interprets the program
by traversing its raw syntax tree structure. It uses an inefficient
representation of lexical variables consisting of heap-allocated environment
objects which store variable bindings as Lisp association lists. Every time a
variable is accessed, the chain of environments is searched for the binding.
\*(TX also provides a compiler and virtual machine for more efficient execution
of Lisp programs. In this mode of processing, top-level expressions are
translated into the instructions of Lisp-oriented virtual machine. The virtual
machine language is traversed more efficiently compared to the traversal of the
cons cells of the original Lisp syntax tree. Moreover, compiled code uses a
much more efficient representation for lexical variables which doesn't involve
searching through an environment chain. Lexical variables are always allocated
on the stack (the native one established by the operating system). They are
transparently relocated to dynamic storage only when captured by lexical
closures, and without sacrificing access speed.
\*(TX provides the function
.code compile
for compiling individual functions, both anonymous and named. File compilation
is supported via the function
.codn compile-file .
The function
.code compile-toplevel
is provided for compiling expressions in the global environment. This
function is the basis for both
.code compile
and
.codn compile-file .
The
.code disassemble
function is provided to list the compiled code in a more understandable way;
.code disassemble
takes a compiled code object and decodes it into an assembly language
presentation of its virtual-machine code, accompanied by a dump of the various
information tables.
File compilation via
.code compile-file
refers to a processing step whereby a source file containing \*(TL forms
(typically named with a
.code .tl
file name suffix) is translated into an object file (named with a
.code .tlo
suffix) containing a compiled version of those forms.
The compiled object file can then be loaded via the
.code load
function instead of the source file. Usually, loading the compiled file
produces the same effect as if the source file were loaded. However, note that
the behavior of compiled code can differ from interpreted code in a number of
ways. Differences in behavior can be deliberately induced. Certain erroneous
or dubious situations can also cause compiled code to behave differently from
interpreted code.
Compilation not only provides faster execution; compiled files also load much
faster than source files. Moreover, they can be distributed unaccompanied by
the source files, and resist reverse engineering.
.SS* Top-Level Forms
An important concept in file compilation via
.code compile-file
is that of the
.IR "top-level form" ,
and how that term is defined. The file compiler individually processes
top-level forms; for each such form, it emits a translated image.
In the context of file compilation, a top-level form isn't simply any Lisp form
which is not enclosed by another one. Rather, in this specific context, it has
this specific definition, which allows some enclosed forms to still be
considered top-level forms:
.IP 1.
If a form appearing in a \*(TL source file isn't enclosed in another
form, it is a top-level form.
.IP 2.
If a
.code progn
form is top-level form, then each of its constituent forms is also a top-level
form.
.IP 3.
If a
.code compile-only
form is top-level form, then each of its constituent forms is also a top-level
form.
.IP 4.
If an
.code eval-only
form is top-level form, then each of its constituent forms is also a top-level
form.
.IP 5.
If a
.code load-time
form is top-level form, then its argument is a top-level form.
.IP 6.
When a macro form is identified as a top-level form, it is macro-expanded
as if by
.code macroexpand
before considering whether it contains top-level forms under rules 2\(en5.
.IP 7.
Rules 2\(en6 are applied recursively.
.IP 8.
No other forms are top-level forms.
.RE
A top-level form is a
.I primary
top-level form if it doesn't contain any other top-level forms.
This means that it is not a form based on any of the operators
.codn progn ,
.code compile-only
or
.codn eval-only .
Note that the constituent body forms of a
.code macrolet
or
.code symacrolet
top-level form are not individual top-level forms, even if the
expansion of the construct combines the expanded versions of those
forms with
.codn progn .
.SS* File Compilation Model
The file compiler reads each successive forms from a file, performs a partial
expansion on that form, then traverses it to identify all of the
top-level forms which it contains. Each top-level form is subject to
three actions, either of the latter two of which may be omitted: compilation,
execution and emission. Compilation refers to the translation to compiled form.
Execution is the invocation of the compiled form. Emission refers to appending
an externalized representation of the compiled form (its image) to the output
which is written into the compiled file.
By default, all three actions take place for every top-level form. Using the
operators
.code compile-only
or
.codn eval-only ,
execution or emission, or both, may be suppressed. If both are suppressed,
then compilation isn't performed; the forms processed in this mode are
effectively ignored.
When a compiled file is loaded, the images of compiled forms are read from
it and converted back to compiled objects, which are executed in sequence.
Partial expansion means that file compilation doesn't fully expand each
form that is encountered. Rather, an incremental expansion is performed,
similar to the algorithm used by the
.code eval
function:
.RS
.IP 1.
First, if
.meta form
is a macro, it is macro-expanded as if by an application of the function
.codn macroexpand .
.IP 2.
If the resulting expanded form is a
.codn progn ,
.codn compile-only ,
or
.code eval-only
form, then
.code compile-file
iterates over that form's argument expressions, compiling each expression
recursively as if it were a separate expression.
.IP 3.
Otherwise, if the expanded form isn't one of the above three kinds of
expressions, it is subject to a full expansion and compilation.
.RE
.SS* Treatment of Literals
Programs specify not only code, but also data. Data embedded in a program is
called
.IR "literal data" .
There are restrictions on what kinds of object may be used as literal data
in programs subject to file compilation. Programs which stray outside of these
restrictions will produce compiled files which load incorrectly or fail to
load.
Literal objects arise not only from the use of literal such as numbers,
characters and strings, and not only from quoted symbols or lists.
For instance, compiled forms which define or reference free variables or global
functions require the names of these variables or functions to be represented
as literals.
An object used as a literal in file-compiled code must be
.I externalizable
which means that it has a printed representation which can be scanned to
produce a similar object. An object which does not have a readable printed
representation will give rise to a compiled file which triggers an exception.
Literals which are themselves read from program source code naturally meet this
restriction; however, with the use of macros, it is possible to embed arbitrary
objects into program code.
If the same object appears in two or more places in the code specified in a
single file, the file compilation and loading mechanism ensures that the
multiple occurrences of that object in the compiled file become a single object
when the compiled file is loaded. For example, if macros are used in such a way
that the compiled file defines a function which has a name generated by
.codn gensym ,
and there are calls to that function throughout that file, this will work
properly: the multiple occurrences of the gensym will appear as the same symbol.
However: that symbol in the loaded file will not be identical to any other
symbol in the \*(TX image; it will be newly allocated each time the compiled
file is loaded.
Interned symbols are recorded in a compiled file by means of their textual
names and package prefixes. When a compiled file is loaded, the interned
symbols which occur as literals in it are entered into the specified packages
under the specified names. The value of the
.code *package*
special variable has no influence on this.
Circular structures in compiled literals are preserved; on loading, similar
circular structures are reproduced.
.SS* Treatment of the Hash-Bang Line
\*(TX supports the hash-bang mechanism in compiled
.code .tlo
files, thereby allowing compiled scripts to be executable.
When a source file begins with the
.code #!
(hash bang) character sequence, the file compiler propagates that
line (all characters up to and including the terminating newline) to the
compiled file, subject to the following transformation: occurrences of
.str --lisp
which are not followed by a dash are replaced with
.strn --compiled .
Furthermore, certain permissions are propagated from a hash-bang source
file to the target file. If the source file is executable to its owner,
then the target file is made executable as if by using
.code chmod
with the
.code +x
mode: all the executable permissions that are allowed by the current
.code umask
are are enabled on the target file. If the target file is thus being marked
executable, then additional permissions are also treated as follows. If the
target file has the same owner as the source file, and the source file's setuid
permission bit is set, then this is propagated to the target file. Similarly,
if the target file has the same group owner as the source file, and the source
file's group execute bit and setgid permission bit are set, then the setgid
bit is set on the target file.
.SS* Compiled File Compatibility
\*(TX's virtual-machine architecture for executing compiled code
is evolving, and that evolution has implications for the compatibility between
compiled files and the \*(TX executable image.
The basic requirement is that a given version of \*(TX can load and execute
the compiled files which that same version has produced.
Furthermore, these files are architecture-independent, except that their
encoding is in the local byte order ("endianness") of the host machine.
The byte order is explicitly indicated in the files, and the
.code load
function resolves it. Thus a file produced by \*(TX running on a 64-bit
big-endian Power PC can be loaded by \*(TX running on 32-bit x86, which is
little endian.
A given \*(TX version may also be capable of loading files produced by
an older version, or even ones produced by a newer version. Whether this
is possible depends on the versions involved. Furthermore,
there is a general issue at play: code compiled by newer
versions of \*(TX may require functions that are not present
in older versions, preventing that code from running. Newer
\*(TX may support new syntax not recognized by older \*(TX,
and that syntax may end up in compiled files.
Compiled files contain a minor and major version number (which is independent
of the \*(TX version). The
.code load
function examines these numbers and decides whether the file is loadable,
or whether it must be rejected.
The first version of \*(TX which featured the compiler and virtual machine was
191. Older versions therefore cannot load compiled files.
Versions 191 and 192 produce version 1 compiled files, and load only that
version.
Versions 193 through 198 produce version 2 compiled files and load only
that version.
Version 199 produces version 3 files and loads versions 2 and 3.
Versions 200 through 215 produce version 4 files and load
versions 2, 3 and 4.
Versions 216 through 243 produce version 5.0 files and load
versions 2, 3, 4 and 5, regardless of minor version.
Versions 244 through 251 produce version 5.1 files and load
versions 2, 3, 4 and 5, regardless of minor version.
Versions 252 through 259 produce version 6.0 files and load
only version 6, regardless of minor version.
Versions 260 through 273 produce version 7.0 files and load
versions 6 and 7, regardless of minor version.
Version 261 introduces JSON
.code #J
syntax. Compiled code which contains embedded JSON literals
is not loadable by \*(TX 260 and older.
.SS* Semantic Differences between Compilation and Interpretation
The
.code compile-only
and
.code eval-only
operators can be used to deliberately produce code which behaves differently
when compiled and interpreted. In addition, unwanted differences in behavior
can also occur. The situations are summarized below.
.coNP Differences due to @ load-time
Forms evaluated by
.code load-time
are treated differently by the compiler. When a top-level form is compiled,
its embedded
.code load-time
forms are factored out such that the compiled image of the top-level form
will evaluate these forms before other evaluations take place.
The interpreter doesn't perform this factoring; it evaluates a
.code load-time
form when it encounters it for the first time.
.coNP Treatment of literals
The compiler identifies multiple occurrences of equivalent strings and bignum
integers that occur as literals, and condenses each one to a single instance,
within the scope of the compilation. The scope is possibly as wide as a file.
If the literal
.str abc
appears in multiple places in the same file that is processed by
.codn compile-file ,
in the resulting compiled file, there may be just a single
.str abc
object. For instance, if the file contains two functions:
.verb
(defun f1 () "abc")
(defun f2 () "abc")
.brev
when compiled, these will return the same object such that
.verb
(eq (f1) (f2)) -> t
.brev
No such de-duplication is performed for interpreted code.
Consequently, code which depends on multiple occurrences of these objects to be
distinct objects may behave correctly when interpreted, but misbehave when
compiled. Or vice versa.
One example is code which modifies a string literal.
Under compilation, the change will affect all occurrences of that literal
that have been merged into one object. Another example is an
expression like
.codn "(eq \(dqabc\(dq \(dqabc\(dq)" ,
which yields
.code nil
under interpretation because the two strings are distinct object in spite
of appearing side by side in the syntax, but
.code t
when compiled, since they denote the same string object.
In the future, objects other than strings and bignums may be similarly
consolidated, such as lists and vectors, which means that interpreted
code which works today when compiled may misbehave in the future.
Note that objects which are literally notated in source code are not the only
kinds of objects considered to be literals. Objects which are constructed by
macros and inserted into macro-expansions are also literals. Literals are
self-evaluating objects that appear as expressions in the syntax which remains
after macro-expansion, as well as arguments of the
.code quote
operator. If a macro calculates a new string each time it is expanded,
and inserts it into the expansion as a literal, the compiler will identify
and consolidate groups of such strings that are identical.
.coNP Treatment of symbols
A source file may contain unqualified symbol tokens which are interned
in the current package.
In contrast, a compiled file encodes symbols with full package qualification.
When a compiled file is loaded, the current package at that time has no effect
on the symbols in the compiled file, even if those symbols were specified as
unqualified in the original source file.
This difference can lead to surprising behaviors. Suppose a source file
contains references to functions or variables or other entities which do not
exist. Furthermore, suppose the entities were referenced, in that file, using
unqualified symbols which didn't exist, and were expected to come from a
different package from the one where they ended up interned. For instance,
supposed the file is being processed in a package called
.code abc
and is expecting to use a function
.code calc
which should come from the
.code xyz
package. Unfortunately, no such symbol exists. Therefore, the symbol is
interned as
.code abc:calc
and not
.codn xyz:calc .
In that case, it
should be
sufficient to ensure that the
.code xyz:calc
function exists, and then reload the source file. The unqualified symbol token
.code calc
in that file will be correctly resolved to
.code xyz:calc
that time. However, if the file is compiled, reloading will not be sufficient.
Even though the symbol
.code xyz:calc
exists, the file will continue to try to refer a function using the symbol
.code abc:calc
which comes from a fully qualified representation stored in the compiled file.
The file will have to be recompiled to fix the issue.
.coNP Treatment of unbound variables
Unbound variables are treated differently by the compiler. A reference
to an unbound variable is treated as a global lexical access. This means that
if a variable access is compiled first and then a
.code defvar
is processed which introduces the variable as a dynamically scoped ("special")
variable, the compiled code will not treat the variable as special; it
will refer to the global binding of the variable, even when a dynamic binding
for that variable exists. The interpreter treats all variable references
that do not have lexical bindings as referring to dynamic variables.
The compiler treats a variable as dynamic if a
.code defvar
has been processed which marked that variable as special.
.coNP Unbound symbols in @ dwim
Arguments of a
.code dwim
form (or the equivalent bracket notation) which are unbound
symbols are treated differently by the
compiler. The code is compiled under the assumption that all such symbols
refer to global functions. For instance, if neither
.code f
nor
.code x
are defined, then
.code "[f x]"
will be compiled under the assumption that they are functions. If they are
later defined as variables, the compiled code will fail because no function
named
.code x
exists. The interpreter resolves each symbol in a
.code dwim
form at the time the form is being executed. If a symbol is defined
as a variable at that time, it is accessed as a variable. If it defined as a
function, it is accessed as a function.
.coNP Bound symbols in @ dwim
The symbolic arguments of a
.code dwim
form that refer to global bindings are also treated differently by the compiler.
For each such symbol, the compiler determines whether it refers to a
function or variable and, further, whether the variable is global lexical or
special. This treatment of the symbol is then cemented in the compiled code;
the compiled code will treat that symbol that way regardless of the
run-time situation. By contrast, the interpreter performs this classification
each time the arguments of a
.code dwim
form are evaluated. The rules are otherwise the same: if the symbol is bound as
a variable, it is treated as a variable. If it is bound as a function, it is
treated as a function. If it has both bindings, it is treated as a variable.
The difference is that this is resolved at compile time for compiled code,
and at evaluation time for interpreted code.
.coNP File-wide insertion of gensyms
The following degenerate situation occurs, illustrated by example. Suppose the
following definitions are given:
.verb
(defvarl %gensym%)
(defmacro define-secret-fun ((. args) . body)
(set %gensym% (gensym))
^(defun ,%gensym% (,*args) ,*body))
(defmacro call-secret-fun (. args)
^(,%gensym% ,*args))
.brev
The idea is to be able to define a function whose name is an uninterned
symbol and then call it. An example module might use these definitions as
follows:
.verb
(define-secret-fun (a) (put-line `a is @a`))
(call-secret-fun 42)
.brev
The effect is that the second top-level form calls the function, which
prints 42 to standard out. This works both interpreted and compiled with
.codn compile-file .
Each of these two macro calls generates a top-level form into which
the same gensym is inserted. This works under file compilation due to a
deliberate strategy in the layout of compiled files, which allows such
uses. Namely, the file compiler combines multiple top-level forms
into a single object, which is read at once, and which uses the circle
notation to unify gensym references.
However, suppose the following change is introduced:
.verb
(define-secret-fun (a) (put-line `a is @a`))
(defpackage foo) ;; newly inserted form
(call-secret-fun 42)
.brev
This still works when interpreted, and compiles successfully. However, when the
compiled file is loaded, the compiled version of the
.code call-secret-fun
form fails with an error complaining that the
.code #:g0039
(or other gensym name) function is not defined.
This is because for this modified source file, the file compiler is not
able to combine the compiled forms into a single object. It would not
be correct to do so in the presence of the
.code defpackage
form, because the evaluation of that form affects the subsequent interpretation
of symbols. After the package definition is executed, it is possible for
a subsequent top-level form to refer to a symbol in the
.code foo
package such as
.code foo:bar
to occur, which would be erroneous if the package didn't exist.
The file compiler therefore arranges for the compiled forms after the
.code defpackage
to be emitted into a separate object. But that division in the output file
consequently prevents the occurrences of the gensym to resolve to the same
symbol object.
In other words, the strategy for allowing global gensym use is in conflict
with support for forms which have a necessary read-time effect such as
.codn defpackage .
The solution is to rearrange the file to unravel the interference, or
to use interned symbols instead of gensyms.
.coNP Delimited Continuations
There are differences in behavior between compiled and interpreted code
with regard to delimited continuations. This is covered in the
Delimited Continuations section of the manual.
.SS* Compilation Library
.coNP Function @ compile-toplevel
.synb
.mets (compile-toplevel < form << expanded-p )
.syne
.desc
The
.code compile-toplevel
function takes the Lisp form
.meta form
and compiles it. The return value is a
.I "virtual-machine description"
object representing the compiled form. This object isn't of function type, but may be
invoked as if it were a function with no arguments.
Invoking the compiled object is expected to produce the same effect as
evaluating the original
.meta form
using the
.code eval
function.
The
.meta expanded-p
argument indicates that
.meta form
has already been expanded and is to be compiled without further expansion.
If
.meta expanded-p
is
.codn nil ,
then it is subject to a full expansion.
Note: in spite of the name,
.code compile-toplevel
makes no consideration whether or not
.meta form
is a "top-level form" according to the definition of that term
as it applies to
.code compile-file
processing.
Note: a form like
.code "(progn (defmacro foo ()) (foo))"
will not be processed by
.code compile-toplevel
in a manner similar to the processing by
.code eval
or
.codn compile-file .
In this example,
.code defmacro
form will not be evaluated prior to the expansion of
.code "(foo)"
(and in fact not evaluated at all)
and so the latter expression isn't correctly referring to that macro.
The form
.code "(progn (macro-time (defmacro foo ())) (foo))"
can be processed by
.codn compile-toplevel ;
however, the macro definition now takes place during expansion, and isn't
compiled.
The
.code compile-file
function has no such issue when it encounters such a form at the top-level,
because that function will consider a top-level
.code progn
form to consist of multiple top-level forms that are compiled
individually, and also executed immediately after being compiled.
.TP* Example
.verb
;; compile (+ 2 2) form and execute to calculate 4
;;
(defparm comp (compile-toplevel '(+ 2 2)))
(call comp) -> 4
[comp] -> 4
.brev
.coNP Function @ compile
.synb
.mets (compile << function-name )
.mets (compile << lambda-expression )
.mets (compile << function-object )
.syne
.desc
The
.code compile
function compiles functions.
It can compile named functions when
the argument is a
.metn function-name .
A function name is a symbol denoting an existing interpreted function,
or compound syntax such as
.mono
.meti (meth < type << name )
.onom
to refer to methods. The code of the interpreted function is retrieved,
compiled in a manner which produces an anonymous compiled function,
and then that function replaces the original function under the same name.
If the argument is a lambda expression, then that function is
compiled.
If the argument is a function object, and that object is an interpreted
function, then its code and lexical environment are retrieved and compiled.
In all cases, the return value of
.code compile
is the compiled function.
Note: when an interpreted function object is compiled, the compiled environment
does not share bindings with the original interpreted environment.
Modifications to the bindings of either environment have no effect on the
other. However, the objects referenced by the bindings are shared. Shared
bindings may be arranged using the
.code hlet
or
.code hlet*
macros.
.coNP Functions @ compile-file and @ compile-update-file
.synb
.mets (compile-file < input-path <> [ output-path ])
.mets (compile-update-file < input-path <> [ output-path ])
.syne
.desc
The
.code compile-file
function reads forms from an input file, and produces a compiled output file.
First,
.meta input-path
is converted to a
.I "tentative pathname"
as follows.
If
.meta input-path
specifies a pure relative pathname, as defined by the
.code pure-rel-path-p
function, then a special behavior applies.
If an existing load operation is in progress, then the special variable
.code *load-path*
has a binding. In this case,
.code load
will assume that the relative pathname is a reference relative to the
directory portion of that pathname.
If
.code *load-path*
has the value
.codn nil ,
then a pure relative
.meta input-path
pathname is used as-is, and thus resolved relative to the current working
directory.
The tentative pathname is converted to an
.I "actual input pathname"
as follows. Firstly, if the tentative pathname ends with one of the suffixes
.code .tl
or
.code .txr
then it is considered suffixed, otherwise it is considered unsuffixed.
If it is suffixed, then the actual pathname is the same as the tentative pathname.
In the unsuffixed case, two possible actual input pathnames are formed. First,
the suffix
.code .tl
is added to the tentative pathname. If that path exists, it is taken
taken as the actual path. Otherwise, the unmodified tentative path
is taken as the actual input path.
If the actual path ends in the suffix
.code .txr
then the behavior is unspecified.
If the
.meta output-path
parameter is given an argument, then that argument specifies the
output path.
Otherwise the output path is derived from the tentative input path
as follows. If the tentative input path is unsuffixed, then
.code .tlo
is added to it to produce the output path.
Otherwise, the suffix is removed from the tentative input path
and replaced with the
.code .tlo
suffix.
The
.code compile-file
function binds the variables
.code *load-path*
and
.code *package*
similarly to the
.code load
function.
Over the compilation of the input file,
.code compile-file
establishes a new dynamic binding for several special
variables. The variable
.code *load-path*
is given a new binding containing the actual input pathname.
The
.code *package*
variable is also given a new dynamic binding, whose value is the
same as the existing binding. Thus if the compilation of the
file has side the effect of altering the value of
.codn *package* ,
that effect will be undone when the binding is removed
after the compilation completes.
Compilation proceeds according to the File Compilation Model.
If the compilation process fails to produce a successful translation
for each form in the input file, the output file is removed.
The
.code compile-update-file
function differs from
.code compile-file
in the following regard: compilation is performed only if the input
file is newer than the output file, or else if the output file doesn't
exist.
The
.code compile-file
always returns
.code t
if it terminates normally, which occurs if it successfully translates
every form in the input file, depositing the translation into the output
file. If compilation fails,
.code compile-file
terminates by throwing an exception.
The
.code compile-update-file
function returns
.code t
if it successfully compiles, similarly to
.codn compile-file .
If compilation is skipped, the function returns
.codn nil .
Note: the following idiom may be used to load a file, compiling it if
necessary:
.verb
(or (compile-update-file "file")
(load-file "file"))
.brev
However, note that it relies on the effect of compiling a source file being the
same as the effect of loading the compiled file.
This can only be true if the source file contains no
.code compile-only
or
.code eval-only
top-level forms.
.coNP Special variable @ *opt-level*
.desc
The special variable
.code *opt-level*
provides control over compiler optimizations.
The variable takes on integer
values. If the value is
.codn nil ,
it is interpreted as zero. The meaningful range is from 0 to 6.
The initial value of the variable is 6.
The meanings of the values are as follows:
.RS
.IP 0
Almost all optimizations are disabled, except for some strength
reductions of instances of he
.code equal
function, to take advantage of certain conditional instructions.
.IP 1
Constant folding is applied, as well as algebraic reductions to list processing
and arithmetic code. Two-argument calls to several common arithmetic operators
are translated into calls to more efficient two-argument internal functions.
.IP 2
Blocks which can be easily confirmed not to be used as exit points are removed.
Variable frames in which no lexically captured variables are bound, and no
dynamic variables are bound, are eliminated.
.IP 3
Lambda expressions and calls to combinator functions such as
.code chain
and
.code andf
are lifted to load time, if possible.
.IP 4
Control flow optimizations are applied: jump threading and elimination of
unreachable code. Some peephole optimizations are applied to improve
certain instruction patterns.
.IP 5
Data flow optimizations are applied, such as elimination of dead register
moves, or useless propagations of values from one register to another.
More peephole optimizations are applied.
.IP 6
Certain more rarely applicable optimizations are applied which reduce code size
by merging some identical code blocks, or improving some more rarely
occurring instruction patterns.
.RE
.coNP Macro @ with-compilation-unit
.synb
.mets (with-compilation-unit << form *)
.syne
.desc
When a file is processed by
.codn compile-file ,
certain actions, such as the issuance of diagnostics about undefined functions
and variables, are delayed until the file is completely processed.
The
.code with-compilation-unit
macro allows these actions to be collectively deferred until multiple files
are completely processed.
The macro evaluates each enclosed
.meta form
in a single compilation environment. After the last
.meta form
is evaluated, deferred actions of any enclosed
.code compile-file
forms are performed, and then the value of the last
.meta form
is returned.
It is permissible to nest
.code with-compilation-unit
forms, lexically or dynamically. The outermost invocation of
.code with-compilation-unit
dominates; all deferred
.code compile-file
actions are held until the outermost enclosing
.code with-compilation-unit
terminates.
.coNP Operators @ compile-only and @ eval-only
.synb
.mets (compile-only << form *)
.mets (eval-only << form *)
.syne
.desc
These operators take on a special behavior only when they appear as top-level forms
in the context of file compilation.
When a
.code compile-only
or
.code eval-only
form is processed by the evaluator rather than the compiler, or when it is
processed outside of file compilation, or when it is appears as other than a
top-level form even under file compilation, then these operators behave
in a manner identical to
.codn progn .
When a
.code compile-only
form appears as a top-level form under file compilation, it indicates to the
file compiler that the
.metn form s
enclosed in it are not to be evaluated. By default, the file compiler executes
each top-level form after compiling it. The
.code compile-only
operator suppresses this evaluation.
When an
.code eval-only
form appears as a top-level form under file compilation, it indicates to the
file compiler that the
.metn form s
enclosed in it are not to be emitted into the output file. By default, the file
compiler includes the compiled image in the output written to the output file.
The
.code eval-only
operator suppresses this inclusion.
Forms which are surrounded by both an
.code eval-only
form and a
.code compile-only
form are neither executed nor emitted into the output file. In this situation,
the forms are skipped entirely; no compilation takes place.
.TP* Notes:
The
.code compile-file
function not only compiles, but also executes every form for the following
reason: the correct compilation of forms can depend on the execution of earlier
forms. For instance, code may depend on macros. Macros may in turn depend on
functions and variables. All those definitions are required in order to compile
the dependent code. Those dependencies may be in a separate file which is
loaded by a
.code load
form; that
.code load
form must be executed.
Note that execution of a form implies that the
.code load-time
forms that it contains are evaluated (prior to other evaluations). Suppression
of the execution of a form also suppresses the evaluation of
.code load-time
forms.
Situations in which
.code compile-only
is useful are those in which it is desirable to stage the execution of some
top-level form into the compiled file, and not have it happen during
compilation. For instance:
.verb
;; in a main module
(compile-only (start-application))
.brev
It is not desirable to have the file compiler try to start the application
as a side effect of compiling the main module. The right behavior is to
compile the
.code "(start-application)"
top-level form so that this will happen when that module is loaded.
Situation in which
.code eval-only
is useful is for specifying forms which have a compile-time effect only,
but are not propagated into the compiled file.
For example, since the correct treatment of literal symbols occurring in a
compiled file does not depend on the
.code *package*
variable, in many cases, the
.code in-package
invocation in the file can be wrapped with
.codn eval-only :
.verb
(eval-only (in-package app))
.brev
The
.code in-package
form must be evaluated during compilation so that the remaining forms are read
in the correct package. However the loading of the compiled versions of those
forms doesn't require that package to be in effect; thus a compiled image
of the
.code in-package
form need not appear in the compiled file.
Macros definitions may be treated with
.code eval-only
if the intent is only to make the expanded code available in the compiled file,
and not to propagate compiled versions of the macros which produced it.
.coNP Macro @ load-time
.synb
.mets (load-time << form )
.syne
.desc
The
.code load-time
macro makes it possible for a program to evaluate a form, such that,
subsequently, the value of that form is then treated as if it were
a literal object.
Literals are pieces of the program syntax which are not evaluated at all.
On the other hand, the values of expressions are not literals.
From time to time, certain situations benefit from the program being
able to perform an evaluation, and then have the result of that evaluation
treated as a literal.
There is already an operator named
.code macro-time
which makes this possible in its particular manner: that operator
allows one or more expressions to be evaluated during macro expansion.
The result of the
.code macro-time
is then quoted and substituted in place of the expression. That result
then appears as a true quoted literal to the executing code.
The
.code load-time
macro similarly arranges for the single form
.meta form
to be evaluated. However, this evaluation doesn't take place at
expansion time. It is delayed until the program executes.
What exactly "delayed until the program executes" means depends on whether
.code load-time
is used in compiled or interpreted code, and in what situation is
it compiled.
If the
.code load-time
form appears in interpreted code, then the exact time when
.meta form
is evaluated is unspecified. The evaluator may identify all
.code load-time
forms which occur anywhere in a top-level expression, and perform
their evaluations immediately, before evaluating the form itself.
Then, when the
.code load-time
forms are encountered again during the evaluation of the form,
they simply retrieve the previously evaluated values as if
they were literal. Or else, the evaluation may be performed late: when the
.code load-time
form itself is encountered during normal evaluation. In that case,
.meta form
will still be evaluated only once and then its value will be be
inserted as a literal in subsequent reevaluations of that
.code load-time
form, if any.
If a
.code load-time
form appears in a non-top-level expression which is compiled, the
compiler arranges for the compiled version of
.meta form
to be executed when compiled version of the entire expression is
executed. This execution occurs early, before the execution of
forms that are not wrapped in
.codn load-time .
The value produced by
.code form
is entered into the static data vector associated with the
compiled top-level expression, which also holds ordinary literals.
Whenever the value of that
.code load-time
form is required, the compiled code references it from the data
vector as if it were a true literal.
When a
.code load-time
top-level form is processed by
.codn compile-file ,
it has no unusual semantics; the effect is that it is replaced by
its argument
.metn form ,
which is in that case also considered a top-level form.
The implications of the translation scheme may be understood
separately from the perspective of code processed with
.codn compile-toplevel ,
.code compile
and
.codn compile-file .
A
.code load-time
form appearing in a form passed to
.code compile-toplevel
is translated such that its embedded
.meta form
will be executed each time the virtual-machine description returned by
.code compile-toplevel
is executed, and the execution of all such forms is placed ahead
of other code.
A
.code load-time
form appearing in an interpreted function which is processed by
.code compile
is evaluated immediately, and its value becomes a literal
in the compiled version of the function.
A
.code load-time
form appearing as a non-top-level form inside a file that is processed by
.code compile-file
is compiled along with that form and deposited into the object file.
When the object file is loaded, each compiled top-level form is executed.
Each compiled top-level form's
.code load-time
calculations are executed first, and the corresponding
.meta form
values become literals at that point. This execution order is individually
ensured for each top-level form.
Thus, the
.code load-time
forms in a given top-level form may rely on the side-effects of
prior top-level forms having taken place.
Note that, by default,
.code compile-file
also immediately executes each top-level form which it compiles and deposits
into the output file. This execution is equivalent to a load; it causes
.code load-time
forms to be evaluated. The
.code compile-only
operator must be used around
.code load-time
forms which must be evaluated only when the compiled file is loaded,
and not at compile time.
In all situations, the evaluation of
.meta form
takes place in the global environment. Even if the
.code load-time
form is surrounded by constructs which establish lexical bindings,
those lexical bindings aren't visible to
.metn form .
Which dynamic bindings are visible to
.meta form
depends on the exact situation. If a
.code load-time
form occurs in code that had been processed by
.code compile-file
and is now being loaded by
.codn load ,
then the dynamic environment in effect is the one in which the
.code load
occurred, with any modifications to that environment that were performed
by previously executed forms. If a
.code load-time
form occurs in code that had been processed by
.codn compile-toplevel ,
then
.meta form
is evaluated in the dynamic environment of the caller which invokes
the execution of the resulting compiled object.
When a
.code load-time
form occurs in the code of an function being processed by
.codn compile ,
then
.meta form
is evaluated in the dynamic environment of the caller which invokes
.codn compile .
If a
.code load-time
form occurs in a form processed processed by the evaluator, it is unspecified
whether it takes place in the original dynamic environment in which the
evaluator was invoked, or whether it is in the dynamic environment of
the immediately enclosing form which surrounds the
.code load-time
form.
A
.code load-time
form may be nested inside another
.code load-time
form. In this situation, two cases occur.
If the two forms are not embedded in a
.codn lambda ,
or else are embedded in the same
.codn lambda ,
then the inner
.code load-time
form is superfluous due to the presence of the outer
.codn load-time .
That is to say, the inner
.mono
.meti (load-time << form )
.onom
expression is equivalent to
.metn form ,
because the outer form already establishes its evaluation to be in a load-time
context.
If the inner
.code load-time
form occurs in a
.codn lambda ,
but the outer form occurs outside of that
.codn lambda ,
then the semantics of the inner
.code load-time
form is relevant and necessary. This is because expressions occurring in a
.code lambda
are evaluated when the
.code lambda
is called, which may take place from a non-load-time context, even if the
.code lambda
itself was produced in a load-time context.
An expression being embedded in a
.code lambda
means that it appears either in the
.code lambda
body, or else in the parameter list as the initializing
expression for an optional parameter.
.TP* Notes:
When interpreted code containing
.code load-time
is evaluated, a mutating side effect may take place
on the tree structure of that code itself as a result of the
.code load-time
evaluation.
If that previously evaluated code is subsequently compiled, the compiled
translation may be different from compiling the original unevaluated code.
Specifically, the compiler may take advantage of the
.code load-time
evaluation which had already taken place in the interpreter, and simply take
that value, and avoid compiling
.meta form
entirely. This also has implications on the dynamic environment
that is in effect when
.meta form
is evaluated. If
.meta form
is evaluated by the interpreter, then it interacts with the dynamic environment
which as in effect in that situation; then when the compiler later just takes
the result of that evaluation, the compiler's dynamic environment is irrelevant
since
.meta form
isn't being evaluated any more.
If
.metn form ,
when evaluated multiple times, potentially produces a different value on each
evaluation, this has implications for the situation when an object produced by
.code compile-toplevel
is invoked multiple times. Each time such an object is invoked, the
.code load-time
forms are evaluated. If they produce different values, then it appears that
the values of literals are changing. All lexical closures derived from the
same compiled object share the same literal data.
The
.code load
function never evaluates a compiled expression more than once. If the same
compiled file is loaded more than once, a new compiled object instance is
produced from each compiled expression, carrying its own storage area for
literals. The
.code compile
function also never evaluates a compiled expression more than once; it produces
a compiled object, and then executes it once in order to obtain a lexical
closure which is returned. Invoking the closure doesn't cause the
.code load-time
expressions to be evaluated.
The
.code load-time
form is subject to compiler optimizations. A top-level expression is assumed to
be evaluated at load time, so
.code load-time
does nothing in a top-level expression. It becomes active inside forms
embedded in a
.code lambda
expressions. Since
.code load-time
may be used to hoist calculations outside of loops,
.code load-time
is also active in those parts of loops which are repeatedly evaluated.
The use of
.code load-time
is similar to defining a variable and then referring to the variable.
For instance, a file containing this:
.verb
(defvarl a (list 1 2))
(defun f () (cons 0 a))
.brev
is similar to
.verb
(defun f () (cons 0 (load-time (list 1 2))))
.brev
When either file is loaded, in source or compiled form,
.code list
expression is evaluated at load time, and then when
.code f
is invoked, it retrieves the list.
Both approaches have advantages. The variable-based approach gives the value a
name. The semantics of the variable is straightforward. The variable
.code a
can easily be assigned a new value. Using its name, the variable can be
inspected from the interactive listener. The variable can be referenced
from multiple top-level forms directly; it is not a static datum tied to
a table of literal values that is tied to a single top-level form.
Furthermore, the use of
.cod3 defvar / defvarl
versus
.cod3 defparm / defparml
controls whether the variable gets replaced with a new value when the
file is reloaded.
The advantage of
.code load-time
is that it doesn't require a separate top-level form to achieve its load-time
effect: the expression is simply nested at the point where it is needed. The
.code load-time
form can therefore be generated by macros, whose expansions cannot inject
extra top-level forms into the site where they are invoked.
If a macro writer would like some form to be evaluated at load time and
its value accessible in a macro expansion that appears arbitrarily nested
in code, then
.code load-time
may provide the path to a straightforward implementation strategy.
Access to a
.code load-time
value is fast because it doesn't involve referencing through a variable
binding; compiled code accesses the value directly via its fixed position
in the static data table associated with that code. This advantage is
insignificant, however, because access to lexical variables in compiled code is
similarly fast, and a value can easily be propagated from a global variable
to a lexical for the sake of speed. That said,
.code load-time
eliminates that copying step too.
A
.code load-time
is also useful when the value is not required, and instead the form produces
a useful effect, which should be hoisted to load time. For instance, consider
a macro which produces the following expansion:
.verb
(progn (load-time (defvar #:g0025)) (other-logic ... #:g0025))
.brev
no matter where this expansion is inserted,
.code compile-file
and
.code load
will ensure that the
.code defvar
is executed once, when the compiled file is loaded, as if that
.code defvar
appeared on its own as a top-level form. Then the
.code other-logic
form can refer to the variable, without the
.code defvar
being evaluated on each execution of the
.codn progn .
The author of a macro can use
.code load-time
to stage the evaluation of global effects that the macro expansion depends on
simply by bundling these effects into the expansion, wrapped in
.codn load-time .
.TP* "Dialect Note:"
The
.code load-time
macro is similar to the ANSI Common Lisp
.code load-time-value
special operator. It doesn't support the
.meta read-only-p
argument featured in the ANSI CL operator.
The semantics of
.code load-time
is somewhat more precisely specified in terms of concrete
implementation concepts. The ANSI CL
.code load-time-value
may evaluate
.meta form
more than once in interpreted code; effectively, the ANSI CL
implementation may treat
.code "(load-time-value x)"
as
.codn "(progn x)" .
This is not true of \*(TL's
.code load-time
which requires once-only evaluation even in interpreted code.
The name
.code load-time
is used instead of
.code load-time-value
for several reasons.
Firstly,
.code load-time
is useful for staging effects, like definitions, to load time, even when the
resulting value is not used.
Secondly, unlike \*(TL, ANSI CL features multiple values: a form
can yield zero or more values. The ANSI CL
.code load-time-value
operator, however, is restricted to yielding a single value, and its
name may have been chosen to emphasize this aspect/restriction.
That doesn't apply in the context of \*(TL in which all expressions
which terminate normally yield exactly one value, making
.str -value
a suffix that adds no value. Lastly,
.code load-time
is shorter, and harmonizes with
.codn macro-time ,
which preceded it by four years.
.coNP Function @ disassemble
.synb
.mets (disassemble << function-name )
.mets (disassemble << function )
.mets (disassemble << compiled-expression )
.syne
.desc
The
.code disassemble
function presents a disassembly listing of the virtual-machine
code of a compiled function or form. It also presents the literal data
contained in that compiled object in a tabular form which is readily
cross-referenced with the disassembly listing.
If the argument is a
.meta function-name
then the function object is retrieved from the binding indicated
by the name, in the global namespace. That object is then treated
as if it were the
.meta function
argument.
A
.meta function
argument is one that is a function object. Only compiled virtual-machine
functions can be disassembled; other kinds of functions are rejected by
.codn disassemble .
The
.code disassemble
function will also process the
.meta complied-expression
object that is returned by the
.code compile-toplevel
function.
In the case of
.metn function ,
the entire compiled form containing
.meta function
is disassembled. That form usually contains code which is external
to the function, even possibly other functions.
The disassembly listing indicates the entry point in the code
block where the execution of
.meta function
begins.
The
.code disassemble
function returns its argument.
.coNP Function @ dump-compiled-objects
.synb
.mets (dump-compiled-objects < stream << object *)
.syne
.desc
The
.code dump-compiled-objects
function writes compiled objects into
.meta stream
in the same format as the
.code compile-file
function.
Unlike under
.codn compile-file ,
the output is written into an arbitrary stream rather than a named file.
The objects aren't specified by the to-be-compiled syntax processed from a
source file, but rather as zero or more arguments which specify objects that
are already compiled.
Each
.meta object
must be be one of three kinds of values:
.RS
.IP 1.
a virtual-machine-description object returned by
.code compile-toplevel
function; or
.IP 2.
a compiled function object, satisfying the function
.codn vm-fun-p ;
or else
.IP 3.
the name of a compiled function object, which may take any of the
forms suitable as arguments to the
.code symbol-function
function.
.RE
.IP
First,
.code dump-compiled-objects
writes some preamble information into
.metn stream .
Then, for each
.meta object
that is not already a virtual-machine description, its corresponding
virtual-machine description is retrieved. The virtual-machine description
is converted into the externalized format required for the object format
and that externalized format is written into
.metn stream .
The
.code object
argument are thus processed in left-to-right order.
If exactly one call to
.code dump-compiled-objects
is used to populate an initially empty file, and no other data are
written into the file, then that file is a valid compiled file.
If that file is processed by
.code load-file
then each of the externalized forms is converted to a virtual-machine
description and executed.
Note that virtual-machine descriptions are not functions. A function's
virtual-machine description is the compiled version of the top-level form
whose evaluation produced that function.
For example, if the following top-level form is compiled and executed,
two functions are defined:
.verb
(let ()
(defun a ())
(defun b ()))
.brev
Then, the following two expressions all have the same effect on
stream
.codn s :
.verb
(dump-compiled-objects s 'a)
(dump-compiled-objects s 'b)
.brev
Whether the
.code a
or
.code b
symbol is used to specify the object to be dumped, the same virtual-machine
description is externalized and deposited into the stream. That machine
description, when loaded and executed, defines two functions.
.SH* INTERACTIVE LISTENER
.SS* Overview
On some target platforms, \*(TX provides an interactive listener, which is
invoked using the
.code -i
command-line option, or by executing
.code txr
with no arguments. The interactive listener provides features like visual
editing of the command line, tab completion on \*(TL symbols, and history
recall.
.SS* Basic Operation
The interactive listener prints a numbered prompt. The number in the prompt increments with every
command. The first command line is numbered 1, the second one 2 and so forth.
The listener accepts input characters from the terminal. Characters are either
interpreted as editing commands or other special characters, or else are
inserted into the editing buffer. However, control characters which don't
correspond to commands are silently rejected.
The carriage return character generated by the
.key Enter
key indicates that a
complete line has been entered, and it is to be interpreted. The listener
parses the line as a \*(TL expression, evaluates it, and prints the resulting
value. If the evaluation of the line throws an exception, the listener
intercepts the exception and prints information about it preceded by
two asterisks and a space. These asterisks distinguish an exception from a
result value.
If an empty line is entered, or a line containing only spaces, tabs
or embedded carriage returns or linefeeds, the prompt is repeated without
incrementing the number. Such a line is not entered into the history.
A line which only contains a \*(TL comment (optional spaces, tabs or embedded
carriage returns or linefeeds, followed by a semicolon), also causes
the prompt to be repeated without incrementing the number. However,
such a line
.B is
entered into the history.
The listener does not allow lines containing certain bad syntax to be submitted
with
.keyn Enter .
If the buffer contains an expression with unbalanced parentheses
or brackets, or unterminated literals, then
.key Enter
generates a newline character
which is inserted into the buffer. In that situation, if that newline character
is being added at the very end of the buffer, the listener flashes the
exclamation mark character (!) two times to warn the user that line has not
been submitted: no computation is taking place, and the listener is waiting for
more input. It is possible to force the submission of an unbalanced line using
the sequence
.key Ctrl-X
.keyn Ctrl-F .
.SS* Limitations
The interactive listener can only accept up to 4095 abstract characters of
input in a single command line.
Though the edit buffer is referred as the "command line", it may contain
multiline input. The carriage return characters which separate multiple lines
count as one abstract character each, and are understood to occupy two display
positions.
The command line must contain exactly one complete \*(TL expression, or a
comment. Multiple expressions will not be evaluated.
In multiline mode, if the number of lines exceeds the number of lines
of the terminal display, the editing experience is adversely affected
in unspecified ways.
The screen updating logic in the listener is based on the assumption that
the display terminal uses ANSI emulation. No other terminal emulation
is supported. The
.code TERM
environment variable is ignored.
.SS* Ways to Quit
Pressing
.key Ctrl-D
in a completely empty command line terminates the listener.
Another way to quit is to enter the
.code :quit
keyword symbol. When the form input into the listener consists of this symbol,
the listener will terminate:
.verb
1> (+ 2 2)
4
2> :quit
os-shell $
.brev
Another way to terminate is to evaluate a call to the
.code exit
function. This method allows a termination status to be specified:
.verb
1> (exit 1)
os-shell $
.brev
However, if a \*(TX interactive session is terminated this way, it will not
save the listener history.
Raising a fatal signal with the
.code raise
function is another way to quit:
.verb
1> (raise sig-abrt)
Aborted (core dumped)
os-shell $
.brev
The previous remark about not saving the listener history applies here also.
.SS* Interrupting Evaluation
.key Ctrl-C
typed while editing a command line is interpreted as an editing command
which causes that command line to be canceled. The listener prints the string
.str ** intr
and repeats the same prompt.
If a command line is submitted for evaluation, the evaluation might take
a long time or block for input. In these situations, typing
.key Ctrl-C
will issue
an interrupt signal. The listener has installed a handler for this signal which
generates an exception of type
.code error
which is caught by the listener. The exception's message is the string
.str intr
so that the listener ends up printing
.str intr **
like in the case of the
.key Ctrl-C
editing command. In this situation, though, a new
command-line prompt is issued with an incremented number, and the exception
is recorded as a value.
.SS* Listener Variables
.coNP Variables @, *0 @, *1 @, *2 ..., @ *99
.desc
The listener provides useful variables which allow commands to reference
the results of previous commands. As noted previously, the commands
are enumerated with an incrementing number. Each command's number, modulo 100,
corresponds to one of the variables
.codn *0 ,
.codn *1 ,
.codn *2 ,
\&...,
.codn *99 .
Thus, up to the previous hundred results can be referenced:
.verb
...
99> (+ 2 2) ;; stored in *99
4
100> (* 3 2) ;; stored in *0
6
101> (+ *99 *0) ;; i.e. (+ 4 6)
10
.brev
.coNP Symbol macros @, *-1 @, *-2 ..., @ *-20
The listener provides small number of symbol macros for referencing the
results of previous commands in a relative. The macro
.code *-1
refers to the value of the immediately previous command. The macro
.code *-2
refers to the value of the command before that one and so on.
Note: each of these macros expands to a reference to the
.code *r
vector, according to the following pattern:
.mono
*-1 --> [*r (mod (- *v 1) 100)]
*-2 --> [*r (mod (- *v 2) 100)]
...
*-20 --> [*r (mod (- *v 20) 100)]
.onom
.coNP Variable @ *n
.desc
The listener variable
.code *n
evaluates to the current command-line number: the number of the command in
which the variable occurs:
.verb
5> *n
5
6> (* 2 *n)
12
.brev
.coNP Variable @ *v
.desc
The listener variable
.code *v
evaluates to the current variable number: the command number modulo 100:
.verb
103> *v
3
104> *v
4
.brev
.coNP Variable @ *r
.desc
The listener variable
.code *r
evaluates to a hash table which associates variable numbers with command
results:
.verb
213> 42
42
214> [*r 13]
42
.brev
The result hash allows relative addressing. For instance the expression
.code "[*r (mod (pred *v) 100)]"
refers to the result of the previous command.
.SS* Exceptions
The interactive listener catches all exceptions. Each caught exception is
associated with the command's variable number, and stored as a value
in the appropriate listener variable as well as the
.code *r
result hash. Exceptions are turned into values by creating a cons cell
whose
.code car
is the exception symbol and whose
.code cdr
holds the exception's arguments.
For each caught exception, a message
is printed beginning with the sequence
.strn "** " .
Exactly how the message appears depends on the type and content of
the exception.
.SS* Editing
The following sections describe the interactive editing commands
available in the listener.
Terminals can often be configured with different choices of cursor
shape: such as a block-shaped cursor, an underline cursor or a
vertical line or "I-beam" cursor. In the following sections, the
phrase "character under the cursor" refers to the character that is
currently covered by a block cursor, underlined by an underline cursor,
or that is immediately to the right of an I-beam cursor.
.NP* Move Left and Right
Moving within the line is achieved using the left and right arrow keys
.key \[<-]
and
.keyn \[->] .
In addition,
.key Ctrl-B
("back") and
.key Ctrl-F
("forward") perform this movement.
.NP* Jump to Beginning and End of Line
The
.key Ctrl-A
command moves to the beginning of the line. ("A" is the beginning
of the alphabet). The
.key Ctrl-E
("end") command jumps to the end of the line,
such that the last character of the line is to the left of the cursor
position. On terminals which have the Home and End keys, these may also
be used instead of
.key Ctrl-A
and
.keyn Ctrl-E .
In line mode, these commands move the cursor to the beginning or end of the
edit buffer.
In multiline mode, if the cursor is not already at the beginning of a physical
line, then
.key Ctrl-A
moves it to the first character of the physical line.
Otherwise,
.key Ctrl-A
moves the cursor to the beginning of the edit buffer.
Similarly, in multiline mode, if the cursor not already at the end of a
physical line,
.key Ctrl-E
moves it there. Otherwise, the cursor moves to the
end of the edit buffer.
.NP* Jump to Matching Parenthesis
If the cursor is on an opening or closing parenthesis, brace or bracket,
the
.key Ctrl-]
command tries to jump to the matching character. The logic for
finding the matching character is identical to that of the Parenthesis Matching
feature. If no matching character is found, then no movement takes place.
If the cursor is not on an opening or closing parenthesis, brace or bracket,
then the closest such character is found. The cursor is moved to that character
and then an attempt is made to jump to the matching one from that new
position.
If the cursor is equidistant to two such characters, then one of them
is chosen as follows. If the two characters are oriented in the same way (both
are opening and closing), then that one is chosen whose convex side faces the
cursor position. Thus, effectively, an inner enclosure is favored over an
outer one. Otherwise, if the two characters have opposite orientation (one is
opening and the other closing), then the one which is to the right of the
cursor position is chosen.
Note: the
.key Ctrl-]
character can be produced on some terminals using
.key Ctrl-5
(using the keyboard home row 5, not the numeric keypad 5). This the same
key which produces the % character when Shift is used. The % character is
used in the Vi editor for parenthesis matching.
.NP* Character Swap
The
.key Ctrl-T
(twiddle) command exchanges the character under the cursor with the
previous character.
.NP* Delete Character Left
The Backspace key erases the character to the left of the cursor, and moves the
cursor to the position which that character occupied.
It doesn't matter whether this key generates ASCII
characters 8 (BS) or 127 (DEL): either one is acceptable. The
.key Ctrl-H
command
also performs the same action, since it corresponds to ASCII BS.
.NP* Delete Character Right
The
.key Ctrl-D
command deletes the character under the cursor, if the cursor
is block-shaped, or to the right of the cursor if the cursor is an I-beam.
the cursor maintains its current character position relative to the
start of the line. In multiline mode, if
.key Ctrl-D
is at the end of a line that
is not the last line, it deletes the newline character, causing the
following line to be joined to the end of the current line.
If the cursor is at the end of the buffer, then
.key Ctrl-D
does nothing,
except if the buffer is completely empty, in which case it is a quit
indication. The Delete key, if available on the terminal, is a near synonym of
.keyn Ctrl-D .
It performs all the same functions, except that it does not
act as a quit indication; Delete has no effect when the buffer is empty.
When a visual selection is in effect, then
.key Ctrl-D
and
.key Del
delete
that selection, and copy it to the clipboard.
.NP* Delete Word Left
The
.key Ctrl-W
("word") command deletes the word to the left of the cursor
position. More precisely, this command first deletes any consecutive whitespace
characters (spaces or tabs) to the left of the cursor. Then, it deletes
consecutive non-whitespace characters. Material under the cursor or to the
right remains. The deleted material is copied into the clipboard.
.NP* Delete to Beginning of Line
The
.key Ctrl-U
("undo typing") command is a "super backspace" operation: it deletes
all characters to the left of the cursor position. The cursor is moved to
the leftmost position. In multiline mode,
.key Ctrl-U
deletes only to the beginning of the current
physical line, not all the way to the first position of the buffer.
.key Ctrl-U
copies the deleted material into the clipboard.
.NP* Delete to End of Line
The
.key Ctrl-K
("kill") command deletes the character under the cursor position
and all subsequent characters. The cursor position doesn't change.
In multiline mode,
.key Ctrl-K
deletes only until the end of the current
physical line, not the entire buffer.
The material deleted by
.key Ctrl-K
is copied into the clipboard.
.NP* Verbatim Character Insert
The
.key Ctrl-V
("verbatim") command places the listener's input editor into
a mode in which the next character is interpreted literally and inserted
into the line, even if that character is a special character such as
.keyn Enter ,
or a command character.
.NP* Verbatim Insert Mode
The two-character sequence
.key Ctrl-X
.key Ctrl-V
("extended verbatim", "super paste")
enters into an verbatim insert mode useful for entry of free-form text. It is
particularly useful in multiline mode. In this mode, almost every character
is inserted verbatim, including
.keyn Enter .
The only commands recognized are:
.keyn Ctrl-X ,
which terminates this mode,
.key Backspace
(whether that key generates ASCII BS or DEL) and arrow key navigation.
.key Enter
inserts a line break, which
appears as such in multiline mode, or as
.code ^M
in line mode.
.NP* Delete Current Line
The
.key Ctrl-X
.key Ctrl-K
command sequence may be used in multiline mode
to delete the entire physical line under the cursor. Any lines below that
line move up to close the gap. In line mode, the command has no effect,
other than canceling select mode. The deleted line, including the
terminating newline character, if it has one, is copied into the
clipboard.
.NP* History Recall
By default, the most recent 500 lines submitted to the interactive listener are
remembered in a history. This history is available for recall, making it
convenient to repair mistakes, or compose new lines which are based on previous
lines. Note that the history suppresses consecutive, duplicate lines.
The number of lines retained may be customized using the
.code *listener-hist-len*
variable.
If the
.key \[ua]
key is used while editing a line, the contents of the line are
placed into a temporary save area. The line display is then updated to
show the most recent line of history. Using
.key \[ua] additional times
will recall successively less recent lines.
The
.key \[da]
key navigates in the opposite direction: from older lines to
newer lines. When
.key \[da]
is invoked on the most recent history line,
then the current line is restored from the temporary save area.
Instead of
.key \[ua]
and
.keyn \[da] ,
the commands
.key Ctrl-P
("previous")
and
.key Ctrl-N
("next") may be used.
If the
.key Enter
key is pressed while a recalled history line is showing, then that
line will be submitted as if it were a newly composed line. The originally
edited line which had been placed in the save area is discarded.
When a recalled line is showing, it may be edited. There are two important
behaviors to note here. If a recalled history line is edited, and then
.key \[ua]
or
.key \[da]
or a navigation command is used to show a different
history line, or to restore the original current line, then the edit is made
permanent: the edited line replaces its original version in the same
position in the history. This feature allows corrections to be made to the
history.
The edit is recorded in the line's undo history as a single change; if the
edited line is visited again, then a single
.key Ctrl-O
command will revert all the
edits that were made.
However, if a recalled line is edited and submitted without navigating to
another line, then it is submitted as a newly composed line, without replacing
the original in the history.
Each submitted line is entered into the history, if it is different
from the most recent line already in history. This is true whether it
is a freshly composed line, a recalled history line, or an edited
history line.
.NP* History Search
It is possible to search backwards through the history interactively
for a line containing a substring. The
.key Ctrl-R
command is used to initiate
search. The command prompt is replaced with the prefix
.code search:
next to which a pair of empty square brackets appears, indicating
that the listener is in search mode. The square brackets are the
search box, enclosing the search text, which is initially empty.
In search mode, characters may be typed. They accumulate inside the search
box, and constitute the string to search for. The listener instantly
navigates to the most recent line which contains a substring match for the
search string, and places the cursor on the first character of the
match. Control characters entered directly are ignored. The
.key Ctrl-V
command be
used to add a character verbatim, as in edit mode.
To remove characters from the search box, Backspace can be used. The
search is not repeated with the shortened search text: the same line
continues to show until a character is added, at which point
a new search is issued.
Search mode has a "home position": a starting point for searches.
The initial home position is whatever line of history is selected
when search mode is initiated. Searches work backward in history from
that line. If search text is edited by deleting characters and then
adding new ones, the new search proceeds from the home position.
The
.key Ctrl-R
command can be used in search mode. It registers the currently
showing line as the new home position, and then repeats the search using the
existing search text backwards from the new position. If the search text
is empty,
.key Ctrl-R
has no effect.
The
.key Ctrl-C
command leaves search mode at any time and causes the
listener to resume editing the original input at the original character
position. The
.key Enter
key accepts the result of a search and submits it
as if it were a newly composed line.
Navigation and editing keys may be used in search mode. A navigation or editing
key immediately cancels search mode, and is processed in edit mode, using
whatever line was located by the search, at the matching character position.
The
.key Ctrl-L
(Clear Screen and Refresh), as well as
.key Ctrl-Z
(Suspend to Background) commands are available in search mode. Their effects
takes place without leaving search mode.
Navigating to a history line manually using
.key \[ua]
or
.key \[da]
(or
.key Ctrl-P
and
.keyn Ctrl-N )
has the same net effect same as locating that line using
.key Ctrl-R
search.
.NP* Submit and Stay in History
Normally when the
.key Enter
key is used on a recalled history line,
the next time the listener is reentered, it jumps back to the
newest history position where a new line is about to be composed.
The alternative command sequence
.key Ctrl-X
.key Enter
provides a useful alternative
behavior. After the submitted line is processed, the listener doesn't jump to
the newest history position. Instead, it stays in the history, advancing
forward by one position to the successor of the submitted line.
.key Ctrl-X
.key Enter
can be used to conveniently submit a range of lines
from the history, one by one, in their original order.
.NP* Insert Previous Word
The equivalent command sequences
.key Ctrl-X
.key w
and
.key Ctrl-X
.key Ctrl-W
insert
a word from the previous line at the cursor position. A word is defined
as a sequence of non-whitespace characters, separated from other words
by whitespace. By default, the last word of the previous line is inserted.
Between the
.key Ctrl-X
and the following
.key Ctrl-W
or
.keyn w ,
a decimal number can be entered.
The number 1 specifies that the last word is to be inserted, 2 specifies
the second last word, 3 the third word from the right and so on.
Only the most recent three decimal digits are retained, so the number can range
from 0 to 999. A value of 0, or a value which exceeds the number of words
causes the
.key Ctrl-W
or
.key w
to do nothing. Note that "previous line" means
relative to the current location in the history. If the 42nd most recent
history line is currently recalled, this command takes material from the 43rd
history line.
.NP* Insert Previous Atom
The equivalent command sequences
.key Ctrl-X
.key a
and
.key Ctrl-X
.key Ctrl-A
insert
an atom from the previous line at the cursor position. A line only
makes atoms available if it expresses a valid \*(TX form, free of syntax
errors. A line containing only whitespace or a comment makes no atoms
available. For the purposes of this editing feature, an atom is defined
as the printed representation of a Lisp atom taken from the Lisp form
specified in the previous line. The line is flattened into atoms
as if by the
.code flatcar
function. By default, the last atom is extracted. A numeric argument
typed between the
.key Ctrl-X
and
.key Ctrl-A
or a can be used to select a
atoms by position from the end. The number 1 specifies the last atom,
2 the second last and so on.
Only the most recent three decimal digits are retained, so the number can range
from 0 to 999. A value of 0, or a value which exceeds the number of words
causes the
.key Ctrl-A
or a to do nothing. Note that "previous line"
has the same meaning as for the
.key Ctrl-X
.key Ctrl-W
(insert previous word) command.
.NP* Insert Previous Line
The command sequences
.key Ctrl-X
.key Ctrl-R
("repeat") and
.key Ctrl-X
.keyn r ,
which are
equivalent, insert an entire line of history into the current buffer. By
default, the previous line is inserted. A less recent line can be selected by
typing a numeric argument between the
.key Ctrl-X
and the
.key Ctrl-R
or
.keyn r .
The immediately
previous history line is numbered 1, the one before it 2 and so on.
If this command is used during history navigation, it references previous
lines relative to the currently recalled history line.
.NP* Symbolic Completion
If the Tab key is pressed while editing a line, it is interpreted as a
request for completion. There is a second completion command: the
sequence
.key Ctrl-X
.keyn Tab .
When completion is invoked with
.key Tab
or
.key Ctrl-X
.keyn Tab ,
the listener looks at a few
of the trailing characters to the left of the cursor position to determine the
applicable list of completions. Completions are determined from among the \*(TL symbols which have
global variable, function, macro and symbolic macro bindings, as well
as the static and instance slots of structures. Symbols which
have operator bindings are also taken into consideration. If a
package-qualified symbol is completed, then completion is restricted to that
package. Keyword symbol completion is restricted to the contents of the keyword
package. The namespaces which are searched for symbols are restricted according
to preceding character syntax. For instance if the characters
.code ".("
or
.code ".["
immediately precede the prefix, then only those symbols are considered
which are methods: that is, each is the static slot of at least one structure,
in which that static slots holds a function.
The difference between
.key Tab
and
.key Ctrl-X
.key Tab
is that Tab completion looks only for
prefix matches among the eligible identifiers. Thus it is a pure completion in
the sense that it suggests additional material that may follow what has been
typed. If the buffer contains
.code (list
it will only suggest completions which can be endings for
.code list
such as
.codn list* ,
.codn listp ,
and
.codn list-str .
It will not suggest identifiers which rewrite the
.code list
prefix. By contrast, the
.key Ctrl-X
.key Tab
completion suggests not only pure
completions but also alternatives to the partial identifier, by looking for
substring matches. For instance
.code copy-list
is a possible completion for
.codn list ,
as is
.codn proper-list-p .
If no completions are found, then the BEL character is sent to the terminal
to generate a beep or a visual alert indication. The listener returns to
editing mode.
If completions are found, listener enters into completion selection mode.
The first available completion is placed into the line as if it had been typed
in. The other completions may be viewed one by one using the Tab key.
(Note that the
.key Ctrl-X
is not used, only Tab, even if completion mode had been
entered via
.key Ctrl-X
.keyn Tab ).
When the completions are exhausted, the original uncompleted line is shown
again, and Tab can continue to be used to cycle through the completions again.
In completion mode, the
.key Ctrl-C
character acts as a command to cancel completion mode
and return to editing the original uncompleted line. Any other input character causes
the listener to keep the currently shown completion, and return to edit mode,
where that character is immediately processed as if it had been typed in
edit mode.
.NP* Edit with External Editor
The two character command
.key Ctrl-X
.key Ctrl-E
launches an external editor to
edit the current command line. The command line is stored in a temporary
file first, and the editor is invoked on this file. When the editor
terminates, the file is read into the editing buffer.
The editor is determined from the
.code EDITOR
environment variable. If this variable is unset or empty,
the command does nothing.
The temporary file is created in the home directory, if that can
be determined. Otherwise it is created in the current working directory. If the
creation of the file fails, then the command silently returns to edit mode.
The home directory is determined from the
.code HOME
environment variable in POSIX environments. On MS Windows, the
.code USERPROFILE
variable is probed for the user's directory.
If the command line contains embedded carriage returns (which denote
line breaks in multiline mode) these are replaced with newline characters
when written out to the file. Conversely, when the edited file is read
back, its newlines are converted to carriage returns, so that multiline
content is handled properly. (See the following section, Multiline Mode.)
.NP* Undo Editing
The listener provides an undo feature. The
.key Ctrl-O
command ("old", "oops")
restores the edit buffer contents and cursor position to a previous state.
There is a single undo history which records up the 200 most recent edit
states. However, the states are associated with history lines, so that it
appears that each line has its own, independent undo history.
Undoing the edits in one line has no effect on the undo history of another
line.
Undo also records edits for lines that have been canceled with
.key Ctrl-C
and are
not entered into the history, making it possible to recall canceled lines.
The undo history is lost when \*(TX terminates.
Undo doesn't save and restore previous contents of the clipboard buffer.
There is no redo. When undo removes an edit to restore to a prior edit state,
the removed edit is permanently discarded.
Note that if undo is invoked on a historic line, each undo step updates that
history entry instantly to the restored state, not only the visible edit
buffer. This is in contrast to the way new edits work. New edits are not
committed to history until navigation takes place to a different history line.
Also note that when new edits are performed on a historic line and it is
submitted with
.key Enter
without navigating to another line, the undo information
for those edits is retained, and belongs to the newly submitted line. The
historic line hasn't actually been modified, and so it has no new undo
information. However, if a historic line is edited, and then navigation takes
place to a different historic line, then the undo information is committed to
that line, because the modifications to the line have been placed back
in the history entry.
.SS* Visual Selection Mode
The interactive listener supports visual copy and paste operation.
Text may be visually selected for copying into a clipboard or
or for deletion. In visual selection mode, the actions of some editing
commands are modified so that they act upon the selection instead
of their usual target, or upon both the target and the selection.
.NP* Making a Selection
The
.key Ctrl-S
command enters into visual selection mode and marks the
starting point of the selection, which is considered the position
immediately to the left of the current character.
While in visual selection mode, it is possible to move around using
the usual movement commands. The ending point of the selection
tracks the movement.
The selected text is displayed in reverse video.
Typing
.key Ctrl-S
again while in visual selection mode cancels
the mode.
Tab completion, history navigation, history search and editing in an external
editor all cancel visual selection mode.
By default, the selection excludes the character which lies to the right of
the rightmost endpoint. Thus, the selection simply consists of the text
between these two positions, whether or not they are reversed. This style of
selection pairs excellently with an I-beam style cursor, and has clear
semantics. The endpoints are referenced to the positions between the
characters, and everything between them is selected.
The selection behavior may be altered using the Boolean configuration variable
.codn *listener-sel-inclusive-p* .
This variable is
.code nil
by default. If it is changed to true, then the selection includes the
character to the right of the rightmost endpoint, if there is such a
character within the current line. This style of selection
pair well with a block-shaped cursor. It creates the apparent semantics that
the endpoints of the selection are characters, rather than points
between characters, and that these characters are included in the selection.
.NP* Selection Endpoint Toggle
In visual selection, the starting point of the selection remains fixed, while
the ending point tracks the movement of the cursor. The
.key Ctrl-^
command will
exchange the two points. The effect is that the cursor jumps to the opposite
end of the selection. That end is now the ending point which tracks the cursor
movement.
.NP* Visual Copy
The
.key Ctrl-Y
command ("yank") copies the selected text into a clipboard buffer.
The previous contents of the clipboard buffer, if any, are discarded.
Unlike the history, the clipboard buffer is not persisted.
If \*(TX terminates, it is lost.
.NP* Visual Cut
If the
.key Ctrl-D
command is invoked while a selection is in effect, then
instead of deleting the character under the cursor, it deletes the
selection, and copies it to the clipboard. The Delete key has the same
effect.
.key Ctrl-D
and
.key Del
have no effect on the clipboard when visual selection is not in
effect, and they operate on just one character.
.NP* Clipboard Paste
The
.key Ctrl-Q
command ("quote the clipboard") inserts text from the clipboard
at the current cursor position. The cursor position is updated to
be immediately after the inserted text. The clipboard text remains available
for further pasting.
If nothing has been yet been copied to the clipboard in the current
session, then this command has no effect.
.NP* Clipboard Swap Paste
The
.key Ctrl-X
.key Ctrl-Q
command sequence ("exchange quote") exchanges the
selected text with the contents of the clipboard. The selection is
copied into the clipboard as if by
.key Ctrl-Y
and replaced by the
previous contents of the clipboard.
If the clipboard has not yet been used in the current session,
If nothing has been yet been copied to the clipboard in the current
session, then this command behaves like
.keyn Ctrl-Y :
text is yanked into the clipboard, but not deleted.
.NP* Visual Replace
In visual selection mode, an editing commands may be used which insert new
text, or a character may be typed in order to insert it. When this happens, the
selection is first deleted and visual mode is canceled. Then the insertion
takes place and visual mode is canceled. The effect is that the newly inserted
text replaces the selected text.
This applies to the Clipboard Paste
.key Ctrl-Q
command also. If a
selection is effect when
.key Ctrl-Q
is invoked, the selected text
is replaced with the clipboard buffer contents.
When a selection is replaced in this manner, the contents
of the clipboard are unaffected.
.NP* Delete in Selection Mode
In visual mode, it is possible to issue commands which delete text.
One such command is
.keyn Ctrl-D .
Its special behavior in selection mode,
Visual Cut, is described above.
The
.key Backspace
key and
.key Ctrl-H
also have a special behavior in select mode. If
the cursor is at the rightmost endpoint of the selection, then these commands
delete the selection and nothing else. If the cursor is at the leftmost
endpoint of the selection, then these commands delete the selection, and take
their usual effect of deleting a character also. In both cases, selection mode
is canceled. The clipboard is not affected.
The
.key Ctrl-W
command for deleting the previous word, when used in visual
selection mode, deletes the selection and cancels selection mode,
and then deletes the word before the selection. Only the deleted
selection is copied into the clipboard, not the deleted word.
All other deletion commands such as
.key Ctrl-K
simply cancel visual
selection mode and take their usual effect.
.SS* Multiline Mode
The listener operates in one of two modes: line mode and multiline mode.
This is determined by the special variable
.code *listener-multi-line-p*
whose default value is
.code t
(multiline mode). It is possible to toggle between
line mode and multiline mode using the
.key Ctrl-J
command.
In line mode, all input given to a single prompt appears to be on a single
line. When the line becomes longer than the screen width, it scrolls
horizontally. In line mode, carriage return characters embedded in a line
are displayed as
.codn ^M .
In multiline mode, when the input exceeds the screen width, it simply wraps to
take up additional lines rather than scrolling horizontally. Furthermore,
multiline mode not only wraps long lines of input onto multiple lines of
the display, but also supports true multiline input. In multiline
mode, carriage return characters embedded in input are treated as line
breaks rather than being rendered as
.codn ^M .
Because carriage returns are not line terminators in text files,
lines which contain embedded carriage returns are correctly saved
into and retrieved from the persistent history file.
When
.key Enter
is typed in multiline mode, the listener tries to determine whether
the current input, taken as a whole, is an incomplete expression which requires
closing punctuation for elements like compound expressions and string literals.
If the input appears incomplete, then the
.key Enter
is inserted verbatim at
the current cursor position, rather than signaling that the line is
being submitted for evaluation. The
.key Ctrl-X
.key Enter
command sequence also has this behavior.
.SS* Reading Forms Directly from the Terminal
In addition to multiline mode, the listener provides support
for directly parsing input from the terminal, suitable for processing
large amounts of pasted material.
If the
.code :read
keyword is entered into the listener, it will temporarily suspend
interactive editing and allow the \*(TL parser to read
directly from standard input. The reading stops when an error occurs,
or EOF is indicated by entering
.keyn Ctrl-D .
In direct parsing mode, each expression which is read is evaluated, but its
value is not printed. However, the value of the last form evaluated is returned
to the interactive listener, which prints the value and accepts it as if
it as the result value of the
.code :read
command.
Note that none of the material read from the terminal is entered into the
interactive history. Only the
.code :read
command which triggers this parsing mode appears in the history.
.SS* Clear Screen and Refresh
The
.key Ctrl-L
command clears the screen and redraws the line being edited.
This is useful when the display is disturbed by the output of some
background process, or serial line noise.
.SS* Suspend to Background
The
.key Ctrl-Z
("Zzzz... (sleep)") command causes \*(TX to be placed into the
background in a suspended, and control returned to the system shell.
Bringing the suspended \*(TX back into the foreground is achieved with a shell
job-control command such as the
.code fg
command in GNU Bash.
When \*(TX is resumed, the interactive listener will redisplay the edited
line and restore the previous cursor position.
Making full use of this feature requires a POSIX job control shell,
in the sense that without job control support in the shell, there may not be a
way to restore \*(TX into the terminal session's foreground, causing the
user to lose interactive control over that \*(TX instance.
.SS* Editing Help
The
.key Ctrl-X
.key ?
command shows a summary of commands, in a four-line
display which temporarily replaces the editing area. The help text
is divided into several pages.
.key Ctrl-C
dismisses the display, and
returns to editing. The
.keyn Ctrl-P ,
.key \[<-]
and
.key \[ua]
keys return
to the previous screen. The
.key Ctrl-Z
and
.key Ctrl-L
commands are available,
having their usual meaning of suspending and refreshing the display.
Any other key advances to the next screen.
Advancing from the last screen, dismisses the display, and returns
to editing. Navigating to the previous screen when the first screen
is being shown also dismisses the display and returns to editing.
.SS* Print the Prompt
The
.code :prompt
command prints the current prompt, followed by a newline, without
incrementing the prompt number. The
.code :p
command prints just the current prompt number, followed by a newline,
without incrementing the number.
In plain mode, the
.code :prompt-on
command enables the printing of prompts. The full prompt is printed before
reading each new expression. An abbreviated prompt is printed before reading
the continuation lines of an incomplete expression. The printing of prompts
is automatically enabled if the input device is an interactive terminal.
None of these prompt-related commands are entered into the history.
.SS* Plain Mode
When the input device isn't an interactive terminal, or if the
.code -n
or
.code --noninteractive
command-line operations are used when invoking \*(TX,
the listener operates in
.IR "plain mode" .
It reads input without providing any editing features: no completion,
history recall, selection, or copy and paste. Only the line editing
features provided by the operating system are available.
Prompts appear if standard input is an interactive terminal, or
if explicitly enabled. There is still an incrementing counter,
and the numbered variables
.codn *1 ,
.codn *2 ,
.code ...
for accessing evaluation results are established.
Lines are still entered into the history, and the interactive profile
is still processed, as usual.
Plain mode reads whole lines of input, yet recognizes multi-line expressions.
Whenever a line of input is read which represents incomplete syntax, another
line of input is read and appended to that line. This repeats until the
accumulated input represents complete syntax, and is then processed as a unit.
Each unit of input is expected to represent a single expression, otherwise
an error is diagnosed.
.SS* Interactive Profile File
Unless the
.code --noprofile
option has been used, when the listener starts up, it looks for file called
.code .txr_profile
in the user's home directory, as determined by the
.code HOME
environment variable in POSIX environments or the
.code USERPROFILE
environment variable on MS Windows. If that variable doesn't exist, no further
attempt is made to locate this file.
If the file exists, it is subject to a security check.
The function
.code path-private-to-me-p
is applied to the file. If it returns
.code nil
then an error message is displayed and the file is not loaded.
If the file passes the security check, it is expected to be readable and
to contain
\*(TL forms, which are read and evaluated.
Syntax errors encountered while reading the profile file are displayed
on standard output, and any exceptions thrown that are derived from
.code error
are caught and displayed. The interactive listener starts in spite of these
situations. Exceptions not derived from
.code error
will terminate the process.
The profile file is not read by noninteractive invocations of \*(TX:
that is, when the
.code -i
option isn't present.
.SS* History Persistence
The history is maintained in a text file called
.code .txr_history
in the user's home directory. Whenever the interactive listener terminates,
this file is updated with the history contents stored in the listener's
memory. The next time the listener starts, it first reloads the history from
this file, making the most recent
.code *listener-hist-len*
expressions of a previous session available for recall.
The history file is maintained in a way that is somewhat
robust against the loss of history arising from the situation that a user
manages multiple simultaneous \*(TX sessions. When a session terminates, it
doesn't blindly overwrite the history file, which may have already been updated
with new history produced by another session. Rather, it appends new entries
to the history file. New entries are those that had not been previously read
from the history file, but have been newly entered into the listener.
An effort is made to keep the history file trimmed to no more than
twice the number of entries specified in
.codn *listener-hist-len* .
The terminating session first makes a temporary copy of the existing
history, which is trimmed to the most recent
.code *listener-hist-len*
entries. New entries are then appended to this temporary file.
Finally, the actual history file is replaced with this temporary file by a
.code rename-path
a rename operation. This algorithm doesn't use locking, and is therefore not
robust against the situation when a two or more multiple interactive \*(TX
sessions belonging to the same user terminate at around the same time.
The home directory is determined from the
contents of the
.code HOME
environment variable in POSIX environments or
.code USERPROFILE
on MS Windows. If this variable doesn't exist, or the user doesn't
have permissions to write to this directory or to an existing history file
in that directory, then the history isn't saved.
It is possible to save the history without terminating the interactive
session, using the
.code :save
command. This saves the history in the manner described above.
Each invocation of
.code :save
only adds to the history file new input since the most recent
.code :save
command.
.SS* Parenthesis Matching
A feature of the listener is visual parenthesis matching in the form of a
brief forward or backward jump of the cursor. This provides a hint to the programmer,
helping to prevent avoid parenthesis balancing errors.
When any of the three closing characters
.codn ) ,
.code ]
or
.code }
is inserted, the listener scans backward for the matching opening
character. Likewise, if any of the three opening characters
.codn ( ,
.code [
or
.code {
is inserted in the middle of text, the listener scans forward for the matching
closing character.
If the matching character is found, the cursor jumps to that
character and then returns to the original position a brief moment later. If a
new character is typed during the brief time delay, the delay is immediately
canceled, so as not to hinder rapid typing.
This back-and-forth jump behavior also occurs when a character is erased using
Backspace, and the cursor ends up immediately to the right of a
parenthesis.
Note that the matching is unsophisticated; it doesn't observe the
lexical conventions and syntax of the \*(TL programming language. For
instance, a closing parenthesis outside a string literal may match
match an opening one inside a string literal.
.SS* Listener Configuration Variables
The listener's behavior can be influenced through values of certain
global variables. The settings can be made persistent by means
of setting these variables in the interactive profile file.
.coNP Special variable @ *listener-hist-len*
.desc
This special variable determines how many lines of history are
retained by the listener. Changing this variable from within the listener
has an instant effect. If the number is reduced from its current value,
history lines are immediately discarded. The default value is 500.
.coNP Special variable @ *listener-multi-line-p*
.desc
This is a Boolean variable which indicates whether the listener is
in multiline mode. The default value is
.codn nil .
Changing this variable from within the listener takes effect
immediately for the next line of input.
If multiline mode is toggled interactively from within the listener,
the variable is updated to reflect the latest state. This happens
when the command is submitted for evaluation.
.coNP Special variable @ *listener-sel-inclusive-p*
.desc
This Boolean variable controls the behavior of visual selection.
It is
.code nil
by default.
A visual selection is determined by endpoints, which are abstract positions
understood as being between characters. When a visual selection begins,
it marks an endpoint immediately to the left of a block-shaped cursor,
or precisely at the in-between position of an I-beam cursor.
The end of the visual selection is similarly determined from the ending
cursor position. The selection consists of those characters which lie
between these positions. This style of selection pairs well with an I-beam
style cursor shape.
If the
.code *listener-sel-inclusive-p*
variable is set true, then the selection also includes one more
character to the right of the rightmost endpoint, if there is
such a character within the current line, giving rise to the appearance
that the selection is determined by the starting and ending character,
and includes them. This type of selection pairs well with a block-shaped
cursor.
.coNP Special variable @ *listener-pprint-p*
.desc
This Boolean variable controls how the listener prints the results
of evaluations.
It is
.code nil
by default.
When the variable is
.codn nil ,
the evaluation result of each line entered into the listener is printed
using the
.code prinl
function. Thus values are rendered in a machine-readable syntax, ensuing
read/print consistency.
If the variable is set true, the evaluation result of each line is printed
using the
.code pprinl
function.
.coNP Special variable @ *listener-greedy-eval-p*
.desc
The special variable
.code *listener-greedy-eval-p*
controls whether or not a "greedy evaluation" feature is enabled
in the listener. The default value is
.codn nil ,
disabling the feature.
Greedy evaluation means that after the listener evaluates an expression
successfully and prints its value, it then checks whether that value is
an expression that may be further subject to nontrivial evaluation.
If so, it evaluates that expression, and prints the resulting value.
The process is then repeated with the resulting value. It keeps repeating until
evaluation throws an error, or produces a self-evaluating object.
These additional evaluations are performed in such a way that all warnings are
suppressed and all other exceptions are intercepted.
Greedy evaluation doesn't affect the state of the listener.
Only the original expression is entered into the
history. Only the value of the original expression is saved in the result hash
or a numbered variable. The command-line number
.code *n
is incremented by one. The additional evaluations are only performed for
the purpose of producing useful output. The evaluations may
have side effects.
.TP* Example:
.verb
1> (set *listener-greedy-eval-p* t)
t
2> 'a
a
3> (defvar b 2)
b
2
4> (defvar c '(+ 2 2))
c
(+ 2 2)
4
5> (defvar d '(list '+ 2 2))
d
(list '+ 2 2)
(+ 2 2)
4
.brev
The
.code "(defvar d ...)"
form produces
.code d
symbol as its result value. That symbol has a variable binding as a result
of that
.code defvar
and so evaluates; that evaluation produces
.codn "(list '+ 2 2)" ,
the contents of
.codn d .
That object is a Lisp expression and is evaluated, producing
.code "(+ 2 2)"
and that is also an expression, which reduces to
.codn 4 .
The object
.code 4
is self-evaluating, and so the greedy evaluation process stops.
.coNP Special variable @ *doc-url*
.desc
The special variable
.code *doc-url*
holds a character string representing a web URL intended to point to the HTML
version of this document. The initial value points to the publicly hosted
document on the Internet. The user may change this to point to another
location, such as a locally hosted copy of the document.
This variable is used by the
.code doc
function.
.SS* Listener-Related Functions
.coNP Function @ doc
.synb
.mets (doc <> [ symbol ])
.syne
.desc
The
.code doc
function provides help for the library symbol
.metn symbol .
If information about
.meta symbol
is available in the HTML version of this document, and is indexed, then this
function causes that document to be opened using a web browser,
such that the browser navigates to the appropriate section of
the manual.
If the
.meta symbol
argument is omitted, then the document is opened without navigating to a
particular section.
The base URL for the document is configured by the
.code *doc-url*
variable.
If
.meta symbol
is successfully found, or else not specified, and
.code doc
successfully invokes the URL-opening mechanism, it returns
.codn t .
Otherwise, it throws an error exception.
The web browser is invoked using a system-dependent strategy.
On MS Windows, the
.code ShellExecuteW
function is relied upon to open the URL.
On other platforms, if the
.code BROWSER
environment variable exists and is nonempty,
its value is assumed to indicate the name or path
of the web-browsing program which can accept the URL as an argument.
If this variable doesn't exist or is empty, then
.code doc
searches for a system-dependent URL-opening utility, such as
.codn xdg-open .
If this utility is not found, then
.code doc
falls back to searching for a browser using one of several names.
If no URL-opening mechanism is identified using the above strategies, an error
exception is thrown. However, if the mechanism is identified, but does not
successfully dispatch the URL to a browser, there is no requirement to throw
an error exception. It may appear that the
.code doc
function returns
.code t
but has no effect.
.coNP Function @ quip
.synb
.mets (quip)
.syne
.desc
The
.code quip
function returns a randomly selected string containing a humorous quip,
quote or witticism. The following code may be added to
.code .txr_profile
to produce a random quip on startup:
.verb
(put-line (quip))
.brev
The
.code quip
function was introduced in \*(TX 244. If the
.code .txr_profile
is used with installations of older \*(TX versions, it is recommended to use
the following, to avoid calling the undefined function, as well as to
prevent a warning:
.verb
(if (fboundp 'quip)
(put-line (quip))
(defun quip ()))
.brev
.SH* SETUID/SETGID OPERATION
On platforms with the Unix filesystem and process security model, \*(TX has
support for executing setuid/setgid scripts, even on platforms whose operating system
kernel does not honor the setuid/setgid bit on hash-bang scripts. On these
systems, taking advantage of the feature requires \*(TX to be installed as a
setuid/setgid executable. For this reason, \*(TX is aware when it is executed
setuid and takes care to manage privileges. The following description about
the handling of setuid applies to the parallel handling of setgid also.
When \*(TX starts, early in its execution it determines whether or not is
is executing setuid. If so, it temporarily drops privileges, as a precaution.
This is done before processing the command-line arguments.
When \*(TX determines that it is executing a setuid script (a file marked
executable to its owner and attributed with the set-user-ID bit), it then
attempts to impersonate the owner of the script file by changing to
effective user ID to that owner just before executing the file. It retains
the real and saved user ID. If the attempt to assume that user ID is
unsuccessful, then \*(TX permanently drops setuid privileges before executing
the script. Likewise, before executing any code other than a setuid
script, \*(TX also drops privileges.
\*(TX tries to honor and implement the setuid permissions on a script
whether or not it is running setuid. When not running setuid, it nevertheless
tries to change its effective user ID to that of the owner of the setuid
script. This will succeed if it has sufficient permissions to do so.
To rephrase: in order for \*(TX to execute a file which is setuid root,
it has to be running with a root effective user ID somehow. In order
to execute a file which is setuid to a non-root user, \*(TX has to be
running effectively as root or else as that user. It doesn't matter whether
these privileges are achieved effectively using the setuid mechanism, or
whether \*(TX is running with the required user ID as its real ID.
However, if \*(TX is running setuid, it takes special care to temporarily
drop the privileges as early as possible, and eventually to drop the
privileges permanently before executing any code, other that the setuid
script. If the setuid script cannot be executed with the privileges it
calls for, \*(TX also drops privileges and executes it anyway, strictly as the
real user who invoked the \*(TX executable.
What it means to drop privileges is to change the effective user ID
and the saved user ID to be equal to the real user ID. On platforms
where the
.code setresuid
function is available, \*(TX uses that function to drop privileges.
On platforms where
.code setresuid
is not available, \*(TX tries to drop privileges using the
C language function call
.codn "setuid(r)" ,
where
.code r
is the previously noted real user ID obtained from
.codn getuid() .
On some platforms, this only works for dropping root privileges: it
overwrites the real and saved ID only if the caller is effectively root.
On those platforms, this approach does not drop non-root privileges.
\*(TX tries to detect whether this approach worked by evaluating
the C language expression
.codn "seteuid(e)" ,
where
.code e
is the previously noted effective user ID. In other words, it
attempts to regain the dropped privilege by recovering the previous
effective ID. If this attempt succeeds, \*(TX immediately aborts.
Dropping setgid privileges is similar. Where
.code setresgid
is available it is used, otherwise an attempt is made with
.code "setegid(r)"
where
.code r
is the previously noted real group ID. Then a test using
.code "setegid(e)"
is performed using the original effective group ID as
.codn e .
This is done after dropping any setuid root user ID privilege
which would allow such a test to succeed.
If \*(TX is running both setuid and setgid, and execute a script
which is setuid only, it will still drop group privileges, and vice
versa: if it executed a setgid script, it will drop user privileges.
For instance, if a root-owned \*(TX runs a setgid script which is owned by
user
.code 10
and group-owned by group
.codn 20 ,
that script will run with an effective group ID of 20. The effective user ID
will be that of the user who invoked the script: \*(TX will drop the root
privilege to the original real ID of the user, and while for the setgid
operation, it will change to the group ID of the script.
The setuid/setgid privilege machinery in \*(TX does not manipulate
the list of supplementary ("ancillary", in the language of POSIX) group IDs.
It is unnecessary for security because the list does not change while
running with setuid privilege. No group IDs are added to the list which
need to be retracted when privileges are dropped. The supplementary
groups also persist across the execution of a setuid/setgid script.
.SH* STANDALONE APPLICATION SUPPORT
The \*(TX executable image supports a general mechanism by means of which
a custom program can be packaged as an apparent standalone executable.
.SS* The Internal Argument String
The \*(TX executable contains a 128 byte data area preceded by the
seven-byte ASCII character sequence
.strn @(txr): .
The 128 byte data area which follows this identifying prefix
represents a null-terminated UTF-8 string. In the stock executable,
this area is filled with null bytes.
If the \*(TX executable is edited such that this area is replaced
with a nonempty, null-terminated UTF-8 string, the program will,
for the purposes of command-line-argument processing, treat this string as if
it were the one and only command-line argument.
(The original command-line arguments are still retained in the
.code *args*
and
.code *args-full*
variables).
The function
.code save-exe
creates a copy of the \*(TX executable with a custom internal argument.
.TP* Example:
Suppose that \*(TX is copied to an executable in the same directory called
.code myapp
(or
.code myapp.exe
on an operating system which requires the
.code .exe
suffix). Also suppose that in the same directory, there exists a file
called
.codn myscript.tl .
This
.code myapp
executable can then be edited so that the data area which follows the
.code @(txr):
bytes contains the following string:
.verb
--args|-e|(load (path-cat (dir-name txr-exe-path) "main.tl"))
.brev
When the
.code myapp
executable is invoked, it will process the above string as a single
command-line argument, causing the
.code main.tl
\*(TL source file to be loaded.
Any arguments passed to
.code myapp
are ignored and available to
.code main.tl
via the
.code *args*
variable.
.SS* Deployment Directory Structure
The \*(TX executable may require library files, depending on the
functionality invoked by the program code. Library files are located
relative to the installation directory, called the
.IR sysroot .
The executable tries to dynamically determine the sysroot from
its own location, according to the following directory structure.
The executable may be renamed, it need not be called
.codn txr :
.verb
/path/to/sysroot/bin/txr
.../share/txr/stdlib/cadr.tl
.../stdlib/cadr.tlo
.../stdlib/except.tl
...
.brev
The above structure is assumed if the executable finds itself
in a directory named
.strn bin .
Otherwise, if the executable finds itself in a directory not
named
.strn bin ,
the following structure is expected:
.verb
/path/to/installation/txr
.../stdlib/cadr.tl
.../stdlib/cadr.tlo
.../stdlib/except.tl
...
.brev
Note that this has changed starting in \*(TX 264. Older versions
of \*(TX, when the executable is not in a directory named
.strn bin ,
expect the following structure:
.verb
/path/to/installation/txr
.../share/txr/stdlib/cadr.tl
.../share/txr/stdlib/cadr.tlo
.../share/txr/stdlib/except.tl
...
.brev
When a custom application is deployed using a possibly renamed
.code txr
executable, one of the above structures should be observed:
either the sysroot with a
.code bin
subdirectory where the executable is located, on the
same level with the
.code share
directory, or else the second structure in which the
.code stdlib
directory is a direct subdirectory of the executable directory.
If one of these structures is not observed, the application
may fail due to the failure of a library file to load.
.coSS Function @ save-exe
.synb
.mets (save-exe < path << arg-string )
.syne
.desc
The
.code save-exe
function produces an edited copy of the \*(TX executable at the specified
.metn path ,
inserting
.meta arg-string
as the internal argument string.
In order for the copied executable to be useful, the required installation
directory structure must be provided around it, as described in the
previous section, Deployment Directory Structure.
The return value of
.code save-exe
is unspecified.
The
.code arg-string
should encode to 127 bytes of UTF-8 or less, or else it will be abruptly
truncated, possibly in the middle of a UTF-8 sequence.
.TP* Example:
Create a copy of \*(TX called
.code myapp
which will load a file called
.code main.tl
that is located in the same directory.
.verb
(save-exe
"myapp"
"--args|-e|(load (path-cat (dir-name txr-exe-path) \e
\e \e"main.tl\e"))")
.brev
.SH* DEBUGGER
\*(TX had a simple, crude, built-in debugger, which was removed.
.SH* COMPATIBILITY
.SS* Overview
New \*(TX versions are usually intended to be backward-compatible with prior
releases in the sense that documented features will continue to work in
the same way. Due to new features, new versions of \*(TX will supply new
behaviors where old versions of \*(TX would have produced an error, such as a
syntax error. Though, strictly speaking, this means that something is working
differently in a new version, replacing an error situation with functionality
is usually not considered a deviation from backward-compatibility.
There is one notable deviation from this general requirement for backwards
compatibility: the handling of compiled files. For pragmatic reasons,
from time to time \*(TX may break backward compatibility, such that a newer
version of \*(TX will not load compiled files produced by an older version.
The files will have to be recompiled with the new \*(TX. More details
are given in the section
.B "Compiled File Compatibility"
under the major section
.BR "LISP COMPILATION" .
The rationale for not requiring backward compatibility support for older
compiled files is that older files require the older implementation of the
virtual machine which they target. In some cases the differences between
the older virtual machine and new is so great that \*(TX would have to carry a
whole separate virtual-machine implementation for the sake of the older files,
which is a significant burden.
.coSS The @ -C compatibility option
When a change is introduced which is not backward compatible, \*(TX's
.code -C
option can be used to request emulation of old behavior.
The option was introduced in \*(TX 98, and so the oldest \*(TX version which
can be emulated is \*(TX 97.
Side effects occur in the processing of the option. If the option is specified
multiple times, the behavior is unspecified.
.coSS Environment variable @ TXR_COMPAT
If the
.code TXR_COMPAT
environment variable exists, and its value is not en empty string,
it must contain a decimal integer. Its value is taken by \*(TX as a request
to emulate old behaviors, just like the value of the
.code -C
option.
If the variable has incorrect contents or an out-of-range value,
\*(TX will print an error diagnostic and exit.
If both
.code -C
and the
.code TXR_COMPAT
environment variable are supplied, the behavior is unspecified.
.SS* Compatibility Version Values
The following version values which have a special meaning as arguments to the
.code -C
option, along with a description of what behaviors are affected. For each
of these version values, the described behaviors are provided if
.code -C
is given an argument which is equal or lower. For instance
.code "-C 103"
selects the behaviors described below for version 105, but not those for 102.
.IP 273
In \*(TX 273 and older versions,
.code lazy-str-get-trailing-list
has a flaw, which causes it to produce an extra empty string. Because the
.code @(freeform)
directive in the pattern language is based on lazy strings, and depends
on this function, it is affected by this issue.
The extra empty string is produced because the materialized prefix of the lazy
string is split on the terminator without regard for the fact that it ends in
the terminator, producing an extra empty piece. For instance, if the terminator is
.strn \en
the materialized prefix of the lazy string is
.strn foo\en
and the remaining list of not-yet-materialized lazy string material is
.codn "(\(dqbar\(dq \(dqbaz\(dq)" ,
then the returned list is
.codn "(\(dqfoo\(dq \(dq\(dq \(dqbar\(dq \(dqbaz\(dq)" ,
rather than
.codn "(\(dqfoo\(dq \(dqbar\(dq \(dqbaz\(dq)" .
Whenever the lazy string's
.meta terminator
is non-empty, this issue reproduces in almost all instances, because
the materialized prefix, unless it is empty, is always terminated by the
.meta terminator
and so the split always produces the extra empty string. This is not a rare edge case.
Compatibility values of 273 and lower restore this behavior.
.IP 272
The compatibility version value 272 restores old behaviors in the pattern
language with regard to the regex and function cases of positive match variables.
\*(TX 273, several semantic improvements took place in this area, which
can break existing code. Pattern variables of the form
.mono
.meti >> @{ bident >> ( fun >> [ args ...])}
.onom
can now invoke a vertical function against the full input, and the variable
consequently to be bound to multiple lines. Previously this syntax invoked
only horizontal functions or else vertical functions in a single-line
horizontal mode. That behavior is restored by 272 or lower compatibility.
Secondly, the function is now always invoked, whether or not the variable
has a binding. The variable is then matched against the text spanned
by the function to either give it a new binding or match the existing binding.
The old behavior, restored by 272 or lower compatibility, is that the
function is not invoked when the variable has a binding; the
variable's value is instead used to match text. Lastly, a similar change
took place in positive match regular expression variables of the
.mono
.meti >> @{ bident <> / regex /}
.onom
form.
Prior to 273, when a variable of this form has an existing binding, the regex
is ignored, and the situation is treated as a match for the variable content.
This old behavior is also restored.
.IP 265
Until \*(TX 265, the
.code with-resources
macro exhibited an undocumented behavior: the three-element binding expression
.mono
.meti >> ( var < init << cleanup )
.onom
immediately caused the
.code with-resources
form to terminate with a return value of
.code nil
if the
.meta init
form returned
.codn nil .
Neither the
.meta cleanup
in the same expression, nor any subsequent binding expressions or the body
of the construct, would be evaluated.
Prior cleanup forms would be evaluated in reverse order, as documented.
A compatibility value of 265 or less restores this behavior.
.IP 262
Selection 262 compatibility restores a wrong behavior which existed between
versions 191 and 262 due to a regression. The wrong behavior is that the
.code defsymacro
operator macro-expanded the replacement form, instead of associating the
macro symbol with the unexpanded form. This makes a crucial difference
to symbol macros which rely on expansion-time effects, such as producing a
different expansion each time they are used.
.IP 258
Selecting 258 or lower compatibility causes
.code abs-path-p
to behave like
.codn portable-abs-path-p .
.IP 257
Until \*(TX 257, the function
.code lexical-var-p
returned
.code t
for not only lexical variables, but also for locally bound special variables,
which are not lexical. The behavior is restored if 257 or older compatibility
is selected.
.IP 251
Until \*(TX 251, the syntax
.code "obj.[fun arg]"
was equivalent to
.codn "[obj.fun arg]" ,
providing little utility.
A compatibility value of 251 or lower restores that behavior. The new
behavior is that
.code "obj.[fun arg]"
is equivalent to
.codn "obj.[fun obj arg]" ,
with
.code obj
evaluated only once, performing method dispatch.
.IP 248
Until \*(TX 248, the
.code hash-revget
function defaulted to using
.code eql
equality for searching the hash table for matching values rather than the
current
.codn equal .
Also, until 248, the
.code @
token for denoting meta-expressions was treated with a low precedence
relative to the range dot
.code ..
token. This led to strange results, such as
.code @(a)..@(b)
parsing in a way equivalent to
.code "@(rcons a @(b))"
rather than
.codn "(rcons @(a) @(b))" .
Not is that undesirable due to the lack of symmetry, it's also
inconsistent with
.code "@a..@b"
denoting
.codn "(rcons @a @b)" .
The latter is because in that case the
.code @
is handled as part of the symbol token as a token, and not as a separate operator.
A compatibility value of 248 or lower restores the above old behaviors of
.code @
and
.codn hash-revget .
.IP 244
Until \*(TX 244, the
.code env-hash
function returned a new hash table each time it was called. The behavior is
restored if 244 or older compatibility is selected.
.IP 243
Two mistakes in the pseudorandom number generator (PRNG) were discovered,
affecting \*(TX 243 and older. Using this compatibility value, or lower, will
restore the buggy behavior, allowing pseudorandom number sequences produced
by those older versions can be reproduced. The PRNG is intended to be an
implementation of the WELL512a PRNG described by Panneton and L'Ecuyer.
The coding mistakes, however, resulted in the PRNG being an implementation of
something other than WELL512a.
.IP 242
In \*(TX 242 and older, the instantiation of an object whose type inherits
from the same supertype more than once resulted in duplicate execution
of the supertype's initialization. This was a documented behavior.
After 242, duplicate initialization is suppressed. For more information, see
the section
.BR "Duplicate Supertypes" . A compatibility value of 242 or lower restores
the duplicate initialization behavior.
.IP 237
Compatibility values of 237 or lower restore the destructive behavior of the
.code sort
and
.code shuffle
functions.
.IP 234
In \*(TX 234 and older versions, the exception throwing functions
.code throw
and
.code throwf
did not return, regardless of the exception type. All unhandled exceptions
triggered internal handling leading to unwinding and termination.
The current behavior is that only
.code error
exceptions lead to termination. When a non-error exception isn't intercepted
by a catch or handler, the
.code throw
or
.code throwf
returns normally, yielding the value
.codn nil .
If a compatibility value equal to or lower than 234 is requested,
the old behavior occurs: all unhandled exceptions terminate.
.IP 231
Versions of \*(TX until 231 contained an undocumented feature: some
library functions which are documented as having parameters that must be of
string type were allowing the arguments to be symbols. For such symbolic
arguments, the name of the symbol obtained from
.code symbol-name
was implicitly taken as the required string value. This behavior was removed:
passing symbolic arguments to library function parameters documented as
strings will cause an exception to the thrown. If a compatibility value
of 231 or lower is specified, however, the tolerant behavior is restored.
.IP 227
In \*(TX 227 and older versions, the functions
.codn carray-uint ,
.codn carray-int ,
.code uint-carray
and
.code int-carray
had different names, namely
.codn carray-unum ,
.codn carray-num ,
.code unum-carray
and
.codn num-carray ,
respectively.
If 227 or lower compatibility is selected, these functions become
available under their old names in addition to their new names.
.IP 225
After \*(TX 225, the behavior of the
.code do
operator was adjusted. Previously, a form like
.code "(do set x)"
which contains no variable references like
.codn @1 ,
.code @2
or
.codn @rest ,
generated a function similar to
.codn "(lambda (. rest) (set x))" .
This was contrary to documentation, which states that
.code "(do set x)"
should produce a variadic function which has one required argument,
and which assigns that argument to the variable
.code x
when invoked. The current implementation is that
.code "(do set x)"
is equivalent to
.code "(do set x @1)"
which produces the documented behavior. If 225 or lower compatibility is
selected, then the old behavior of
.code do
takes effect.
.IP 224
After \*(TX 224, the treatment of certain special structure functions
has changed. Selecting 224 compatibility or lower restores that behavior.
The specification given in the
.B "Special Structure Functions"
paragraph has always stated that special functions must be static slots,
and that the behavior is unspecified if they are instance slots.
The behavior of \*(TX 224 and earlier was that these functions worked anyway
if they were instance slots; after \*(TX 224, they some special functions
will no longer be recognized if bound to instance slots.
.IP 222
After \*(TX 222, the behavior of
.code :vars
in
.code @(collect)
was subject to an adjustment. Previously, if the collect body
didn't bind any variables, and both required and optional variables
were specified in
.codn :vars ,
it would still bind all of the optional ones to their default values.
This was a poor behavior which violated the idea that
.code :vars
enforces an all-or-nothing binding discipline to keep the collected
lists consistent. Selecting 222 compatibility or lower restores this
behavior.
.IP 215
After \*(TX 215, the behavior of the
.code load
function changed with respect to its treatment of the
.code *load-path*
variable. In cases where
.code load
resolved the path by adding a suffix,
.code *load-path*
was bound to the unsuffixed name, which was a documented behavior.
After \*(TX 215, also, the behavior of the
.code sub-str
function changed. When the arguments implicate the entire string,
.code sub-str
started just returning the original string, and not making a copy.
The old behavior was to always make a copy.
The above old behaviors of
.code load
and
.code sub-str
are restored if 215 or lower compatibility
is requested. Note, however, that the restoration of the
.code sub-str
behavior in response to the compatibility option was only
introduced in \*(TX 251. In \*(TX 249 and older, the
compatibility value has no effect on the behavior of
.codn sub-str .
.IP 202
Up to \*(TX 202, the
.code logxor
function was incorrectly implemented, producing wrong results when both
arguments are the same fixnum integer, or the same bignum object.
The incorrect behavior is restored if 202 or earlier compatibility is
requested. After 202, the behavior of the
.code print
function changed with regard to symbols in the keyword package.
Regardless of the
.meta pretty-p
flag, keywords are printed with the leading colon. Compatibility with
202 or earlier restores the behavior that when the
.meta pretty-p
flag is true, symbols are printed without package prefixes.
.IP 199
After \*(TX 199, certain global variables that had been deprecated
for a long time, and no longer documented, were removed. Requesting 199 or
earlier compatibility restores those variables.
.IP 190
Until \*(TX 190, the
.code reset-struct
function neglected to perform
.code :postinit
initializations, and didn't invoke finalization on the structure object
if an exception was thrown during reinitialization. Thus, contrary
to documented requirements, reinitialization of a structure didn't behave
like fresh construction. Also, until \*(TX 190, macro parameter
lists implemented the requirement that a
.code :
(colon keyword symbol) argument to an optional
was treated as a missing argument, triggering argument-defaulting behavior.
That requirement was removed; the colon symbol behaves as an ordinary value
under destructuring with macro parameter lists.
Moreover, until \*(TX 190, the
.code pub
symbol package didn't exist; the
.code *package*
variable was initialized to the user package and so symbols introduced
by application code were interned in the same package as the \*(TL
library.
Until \*(TX 190,
.code defmacro
and
.code defsymacro
forms were evaluated immediately during macro expansion; in \*(TX 191
or later, this eager evaluation was abandoned.
Unfortunately, this change introduced
a regression, causing the replacement form of a
.code defsymacro
to be macro-expanded at the time that form is traversed by the
expander, so that the macro is associated with the expanded version
of that form. This is something which had been fixed in 137.
It went unnoticed until much later, after the 262 release.
All the above old behaviors are restored in compatibility
with version 190 or earlier.
Finally, one more change after \*(TX 190 that is controlled by the
compatibility mechanism was a critical redesign of the requirements
for the behavior of the
.code ldiff
function. Version 190 compatibility causes the
.code ldiff
symbol to refer to the old implementation of
.codn ldiff .
.IP 188
Until \*(TX 188,
.codn equal -based
hash tables printed using the notation
.code "#H((:equal-based ...) ...)"
whereas
.codn eql -based
hash tables simply omitted the
.code :equal-based
keyword. Changes were introduced in \*(TX 187 which gave rise to a read/print
inconsistency with printing behavior. In \*(TX 189, further changes were
introduced to fix this inconsistency:
.codn equal -based
hash tables print without any keyword indicating equality, and
.codn eql -based
hash tables print as
.codn "#H((:eql-based) ...)" .
If 188 or compatibility is selected, hash tables are printed
in the old way.
.IP 187
Until \*(TX 187, hash tables constructed by the
.code hash
function were based on
.code eql
equality by default; the
.code :equal-based
keyword argument had to be specified to override this default, and the
.code :eql-based
keyword didn't exist. Selecting 187 or lower compatibility restores the
behavior of
.code eql
equality being default, and the
.code :eql-based
keyword being unrecognized. This affects all functions which implicitly
rely on
.codn hash ,
those being :
.codn uniq ,
.codn unique ,
and
.codn group-by .
In spite of these changes, the printed representation of hash tables continues
to use the
.code :equal-based
keyword to indicate hash tables based on
.code equal
and its absence to indicate
.code eql
equality. The new
.code :eql-based
keyword may be used in hash literals (unless 187 compatibility is
in effect, in which case it is ignored).
.IP 184
A value of 184 or lower switches to the old implementation of the
.code op
and
.code do
macros which was replaced starting in \*(TX 185. Also, this has the
effect of disabling the special recognition of meta-expressions
and meta-symbols in the dot position of function calls, and
the macro expansion of meta-symbols in quasiliterals. This is
because the old
.code op
implementation implements these behaviors itself. The implication
is that user code which binds custom macros to
.code sys:var
or
.code sys:expr
may be affected by 184 or lower compatibility.
.IP 185
A value of 185 or lower restores the old precedence of the
double dot notation for expressing ranges, relative to the
referencing dot. Until \*(TX 185, the expression
.code a.b..c.d
parsed as
.codn "(qref a (rcons b c) d)" .
What is worse, it parsed this way even if written as
.codn "a.b .. c.d" .
Starting in \*(TX 186,
.code ..
has a lower precedence, producing the more useful and intuitive parse
.codn "(rcons (qref a b) (qref c d))" :
in other words, the range with endpoints given by
.code a.b
and
.codn c.d .
.IP 183
A value of 183 or lower restores an inconsistent behavior in the
.code "@(bind)"
directive and other places in the \*(TX pattern language where binding
takes place. Prior to version 184, a string-tree match was only tried in
both directions when the left-hand side of a binding (the "pattern") was a
variable. For non-variable pattern terms, such as Lisp expressions or atoms,
the string-tree match was tried in one direction only: a string tree arising
out of the pattern could match a string atom value on the right side.
A string tree is a nested list structure whose leaves are strings: a list
of strings, a list of lists of strings, and so on, in any mixture.
Concretely, before \*(TX 184,
.mono
@(bind "a" ("a" "b" "c"))
.onom
didn't match, but
.mono
@(bind ("a" "b" "c") "a")
.onom
did. However, if the variable
.code a
contained
.strn a
then
.mono
@(bind a ("a" "b" "c"))
.onom
did match: an inconsistency.
.IP 177
A value of 177 or lower causes the emulation of a bug which was present in the
.code rng
awk macro. A range whose start and end condition matched on the same record failed
to activate for that record, even though
.code rng
is inclusive. The behavior is incompatible with POSIX Awk.
.IP 174
A value of 174 or lower restores a previous behavior of variable substitution
in the
.code output
directive and in quasiliterals in both the \*(TX pattern language and \*(TL.
The behavior in question is the evaluation of the element indexing or
range selection modifier, exemplified by
.codn "@{a [2]}" .
The previous behavior was that if the variable is of any type other
than list, it is converted to a string (unless it already is one).
The indexing then applies to the string. If it is a list then the
indexing or range selection applies to the original list value,
prior to conversion to text. The current behavior is that indexing
and range selection is applied to the original value if that value
is any sequence type which satisfies the
.code seqp
function, otherwise to the string representation.
.IP 172
A value of 172 or lower restores a behavior of the \*(TX pattern
matching language when matching a variable followed by a directive, such as
.codn "@a@(fun b)" .
The old behavior is that the scan for a match for the directive
takes place in an environment in which a binding for
.code a
has not yet been established. The new behavior is that the variable
is always bound prior to the processing of the directive. During
the search, it is bound to the range of text spanning between the
starting position and the position being tried.
.IP 170
A value of 170 or lower disables the behavior that \*(TX scans standard input
when no input sources are specified on the command line. Standard input must
be requested explicitly using the
.code -
argument. This is how it was in all versions of \*(TX up to 170.
Some programs may behave differently because of this. Specifically, programs
which do not take any arguments, and do not select an input source using the
.code @(next)
directive, or suppress the use of an input source using
.codn "@(next nil)" ,
may now accidentally read from standard input.
Until version 170, the functions
.codn split ,
.codn split* ,
.code partition
and
.code partition*
ignored negative indices in their
.meta index-list
argument. The new behavior is that the length of the input sequence
is added to any negative index values. The resulting values are then
ignored if they are still negative.
.IP 165
A value of 165 restores the following behaviors, which changed starting in 166.
There was a change in Lisp evaluation support of the \*(TX pattern language.
Specifically, Lisp argument forms were not subject to expansion prior to
evaluation in these directives:
.codn output ,
.codn mod ,
.codn modlast ,
.codn skip ,
.codn fuzz ,
.codn load ,
.codn close ,
.codn call ,
.code cat
and
.codn next .
.IP 161
Version 161 was the last version in which a bug existed in the
.code handle
macro. In spite of the documentation claiming that
.code handle
has the same syntax as
.codn catch ,
the clauses of
.code handle
were being passed the exception symbol as the leftmost argument, followed
by the exception arguments. This convention is different from
.code catch
clauses which do not receive the exception symbol, only the arguments.
The discrepancy was corrected by making
.code handle
behave like
.codn catch ,
as documented. Requesting compatibility with 161 or earlier restores
the previous behavior of the
.code handle
macro.
.IP 156
After version 156, two behaviors changed in the in the macro expander for
.codn caseq ,
.code caseql
and
.codn casequal :
one outright bug was fixed, and one hitherto undocumented behavior
was changed and specified in the documentation at the same time.
Selecting a compatibility value of 156 or less restores the previous
behaviors. The bug was that single-atom case keys were undergoing
evaluation. For instance
.code "(caseql x (a 0))"
would arrange for the evaluation of
.code a
as a variable, rather than treating it as the symbol
.code a
itself. Though the compatibility mechanism restores the behavior,
applications depending on the evaluating behavior should be changed to
instead use
.codn caseq* ,
.code caseql*
or
.codn casequal .
A workaround for this bug for \*(TX versions 156 or older is to replace
simple keys with a key list of length one, exemplified by a rewrite of the
foregoing expression to
.codn "(caseql x ((a) 0))" .
Here
.code a
is not evaluated.
The undocumented behavior was that a matching clause which has no forms
to be evaluated was producing a result value of
.codn t .
For example
.code "(case 1 (1))"
previously yielded
.codn t ,
but now yields
.codn nil ,
and this behavior is documented.
.IP 155
After version 155, the
.code tok-str
and
.code tok-where
functions changed semantics. Previously, these functions exhibited the
flaw that under some conditions they extracted an empty token immediately
following a nonempty token. This behavior was working as designed and
documented, but the design was flawed, creating a major difficulty in simple
tokenizing tasks when tokens may be empty strings. Requesting compatibility
with version 155 or earlier restores the behavior.
.IP 154
After version 154, changes were introduced in the semantics of struct
literals. Previously, the syntax
.code "#S(abc x a y b)"
denoted the construction of an instance of
.code abc
with
.code "x a y b"
as the constructor parameters, similarly to
.codn "(new abc x 'a y 'b)" .
The new behavior is that
.code abc
is constructed using no parameters, as if by
.code "(new abc)"
and then the slot values are assigned. This means that the values
specified in the literal override any manipulations of those slots by
the type's user-defined
.code :postinit
handlers. Also, after 154,
.code print
methods are expected to take three arguments and are invoked for both
pretty printing and regular machine-readable printing. Until 154, a struct's
.code print
methods was called only when that struct was being pretty-printed, and
only with two arguments; ordinary printing side-stepped the method and rendered
the standard
.code #S
syntax featuring all instance slots.
.IP 151
After version 151, changes were implemented to the way static slots work
in \*(TL structs. Selecting compatibility with 151 restores most of the behaviors.
Until 151, each structure type had its own instance of static slots whether
they were newly defined or inherited. Under the new scheme, a derived struct
shares one instance of each inherited static slot with its base type.
Under the old scheme, a struct inherits the static
initialization functions of its bases (the
.meta static-initfun
argument passed in
.codn make-struct-type ).
These are invoked because they are relied upon by the
.code defstruct
macro to perform the initializations of all the inherited static slots.
Under the new scheme, the static initialization functions are not inherited.
Only the type's own
.meta static-initfun
is invoked to initialize its newly defined static slots that it doesn't
share with the parent. The inherited static slots simply preserve their
current values they have in the base type; their values are untouched by
the introduction of a derived type. The
.code static-slot-ensure
also changed semantics after version 151. The old behavior was problematic
because it affected all static slots throughout the inheritance hierarchy
matching the name passed in by argument. Since this function is the basis
for redefining methods, its behavior broke the semantics of overriding.
Selecting 151 compatibility only restores the behavior of this
function and macros based on it like
.codn defmeth :
in the situation when it introduces a new static slot into one or more
struct types, in compatibility mode it introduces the slot separately into each
type without sharing, and it recurses over the entire type hierarchy,
storing
.meta new-val
into all static slots which match
.metn name .
.IP 150
Until version 150, the
.code match-regex
function behaved in a different way from what was documented. Rather
than returning the length of the match, it returned the index one
past the last matching character. In the case when the starting position
is zero, these values coincide; they are different if the match begins
at some position inside the string. Compatibility with 150 restores
the behavior. The
.code match-regst
function was also affected by this issue; however, since it returned nonsense
result not corresponding to the matching text, it was repaired without
backward compatibility.
Also affected by version 150 compatibility are the
.code match-regex-right
and
.code match-regst-right
functions. These functions worked as documented; however, their
specification changes after version 150 to a semantics which is
more useful and less surprising to the programmer.
.IP 148
Up until version 148, the
.code :postinit
handlers specified in a
.code defstruct
were executed in derived-to-base order, opposite to the
order of execution of
.code :init
handlers. Though described in terms of
.code defstruct
syntax and concepts, this is actually a change in how
.code make-struct-type
treats its
.meta postinitfun
argument.
Specifying 148 or earlier compatibility provides this
old behavior. Also, until version 148, the
.code trim-str
function stripped leading and trailing whitespace from a string
consisting of not only spaces, tabs and newlines, but also carriage
returns, vertical tabs and form feeds.
.IP 145
In versions 144 and 145, \*(TX opened files in text mode on Cygwin,
enabling conversion between CR-LF line endings and abstract newline
characters. This behavior change was retracted, so that files on Cygwin are
opened without specifying text mode, causing the streams to be effectively
binary. The intended "Windows native" behavior of streams being text mode is
instead provided in the Windows version of \*(TX by the Cygnal library.
.IP 143
Until version 143, the
.code stdlib
variable didn't include the trailing slash. The
.code makunbound
function semantics changed after version 143 to be more
compatible with ANSI Common Lisp. Until 143, that function removed
only the global binding, leaving the dynamic rebinding of a variable
intact. The
.code defsymacro
operator neglected to remove the symbol's special variable
mark, if the symbol was previously defined as a special variable.
Also, until version 143 many more places in the \*(TX pattern language used
bind expressions rather than Lisp expressions. The compatibility
option restores these behaviors.
.IP 142
Until version 142, the \*(TX pattern language supported a prefix
convention on data sources. Data sources beginning with the character
.code !
were treated as system command pipes, and data sources beginning with
.code $
indicated that a directory is to be scanned. This convention was recognized
both for command-line arguments, the arguments of the
.code @(next)
directive, and of the
.code @(output)
directive, whether or not the argument was a literal or a computed
value. This feature was dropped from the language after version 142.
Also, until version 142, the
.code @(next)
directive recognized the name
.str -
as denoting standard input, and
.code @(output)
recognized it as standard output. These behaviors were also removed;
versions after 142 recognize this convention only when it appears
as a command-line argument. Lastly, until version 142, the
.code @(output)
directive evaluated the
.meta destination
argument as an expression of the \*(TX pattern language, requiring
.code @
to be used to denote a Lisp expression. This is no longer required.
All these old behaviors are provided
if compatibility with 142 or earlier is requested.
.IP 139
After \*(TX 139, changes were implemented in the area of pseudorandom
number generation. Compatibility with 139 brings back the previous
seeding algorithm used by
.codn make-random-state ,
allowing the old pseudorandom sequences to be reproduced. This is only
the case if the default value of 8 is used for the
.meta warmup-period
argument of that function (which didn't exist in 139 or earlier versions).
.IP 138
After \*(TX 138, the variable name lookup rules in the \*(TX pattern language
changed for greater utility and consistency. Compatibility with 138 or later
restores the previous rules under which most accesses to a \*(TL variable from
\*(TL require the
.code @
prefix denoting Lisp evaluation, but some do not.
.IP 137
Compatibility with \*(TX 137 restores the behavior of not expanding
symbol macros in the dot position of a function call form. For instance
if
.code x
is a symbol macro, in this compatibility mode it is not recognized
in a form like
.codn "(list 1 2 . x)" .
This preserves the behavior of code which depends on
.code x
in such a form to refer to a variable that is being otherwise shadowed by the
symbol macro. \*(TX 137 compatibility also restores a particular behavior
of the global and local macro defining operators
.code defsymacro
and
.codn symacrolet :
in compatibility mode, these operators macro-expand the replacement forms
of symbol macros at expansion time, and then bind the resulting expanded
forms to their respective macro symbols. The forms are then potentially
expanded again when the symbol macros are substituted. This wrong behavior was
never implied by the documentation. The
.code with-slots
macro is also affected by this, because it is implemented in terms of
.codn symacrolet .
Lastly, \*(TX 137 compatibility mode also restores another behavior
of the dot position in function call forms: if the dot position of a
function call form produces a sequence that is not a list, that sequence
is converted to a list so that
.mono
(list . "abc")
.onom
produces
.codn "(#\ea #\eb #\ec)" .
After 137, no such treatment is applied to the value and the same form now
yields
.strn abc .
.IP 136
A request for compatibility with \*(TX 136 or earlier restores the old behavior
of the
.code if
directive, which in used to be a syntactic sugar for a
.code cases
directive with
.code require
at the top of each block. Though semantically well-defined and working as
documented, the behavior was confusing, since failed matching caused potential
evaluation of multiple clauses, whereas programmers expect an if/elif/else
ladder to select exactly one clause.
.IP 128
Compatibility with \*(TX 128 or earlier brings back the behavior that
expressions in quasiliterals are evaluated according to \*(TX evaluation
rules for quasiliterals which occur in the \*(TX pattern language.
Similarly, expressions in
.code @(output)
blocks are treated \*(TX pattern language expressions.
.IP 127
In versions of \*(TX until 127, the functions
.codn symbol-function ,
.code fboundp
and
.code fmakunbound
behaved similarly to their Common Lisp counterparts. See the Dialect Notes
under these functions.
.IP 124
In \*(TX 124 and earlier versions, the
.code @(next)
directive didn't evaluate the
.meta source
argument as a Lisp expression, but as a \*(TX pattern language
expression. Lisp expressions thus had to be delimited by
.codn @ .
The current behavior is that the argument is treated as Lisp.
If the compatibility option is set to 124 or lower, the old behavior
is restored. However, even without the presence of the compatibility option,
if the
.meta source
argument is a meta-expression or meta-symbol (denotes by the
.code @
prefix in front of a compound expression or symbol, respectively)
it is also treated in the old way. This latter behavior is obsolescent
and will eventually disappear, and the compatibility option will be
the only way to get the old behavior.
.IP 123
In \*(TX 123 and earlier, the variable initialization forms of a
.code for
or
.code for*
loop were evaluated outside of the scope of the implicit
.code nil
block. They are now inside the block. The compatibility option will
restore the old behavior.
.IP 121
In \*(TX 121 and earlier versions, \*(TL expressions evaluated in the
pattern language were placed in a lexical environment in which the
pattern variables were visible as lexical variables. The meant that
these variables could be directly captured in lexical closures. On the other
hand, it meant that a Lisp function defined in a
.code @(do)
block could not access a variable established by a later
.codn @(bind) .
It doesn't make sense for dynamically captured variables to be lexical,
so the rule was changed. The backward compatibility switch will enable
the old scoping behavior. Capturing the values of pattern variables in
closures is possible indirectly under the new rule: simply bind new lexical
variables with their values.
.IP 118
The
.code slot-p
function's name changed to
.code slotp
after 118. The compatibility option causes
.code slot-p
to be defined also.
.IP 117
The arguments of the
.code make-struct-type
acquired changed after version 117. 117 compatibility brings back the old
interface.
.IP 114
\*(TX until version 114 reported parse errors in this format:
.verb
./txr: (file.txr:123): syntax error
.brev
The new format omits the program name prefix and parentheses.
Also, the
.code kill
function returned an integer, obtained from the return
value of the underlying C function, rather than converting
that value to a Boolean. The old behavior was not documented,
and 114 compatibility restores it.
Lastly, prior to 115, random state objects were of type
.code *random-state*
(the same symbol as the special variable name)
rather than of type
.codn random-state .
This is a bug whose behavior is simulated by 114 compatibility.
.IP 113
Version 113 is the last version in which the
.codn stat ,
.codn lstat ,
and
.code fstat
functions returned a property list rather than a structure.
Requesting 113 compatibility restores the behavior of returning
a property list. However, the filesystem testing functions like
.code path-exists-p
will not work, because they rely on these functions returning
a structure.
.IP 109
The optional trailing semicolon on hex and octal codes in the \*(TX
pattern language was introduced in 110. The feature is disabled
with 109 or lower compatibility, so that
.code @\ex21;a
encodes
.code !;a
rather than the current behavior of encoding
.codn !a .
Also, in 109 and earlier, newlines were allowed in word list literals and
word list quasiliterals. They were treated as a word-separating space.
A backslash-escaped newline, and all whitespace around it, was deleted
just like in ordinary literals, and did not separate words. The old
behavior is emulated.
.IP 107
Up through \*(TX 107, by accident, there was a function called
.code flip
as well as an operator by the same name. The function was renamed to
.codn flipargs .
Version 107 compatibility or earlier provides the
function under the original name also. Also, up until this version,
\*(TX allowed functions and macros to be defined with the same names
as built-in operators, and macros. Newer versions reject this as an error.
Requesting compatibility to 107 or earlier suppresses the rejection,
though without introducing any requirement that redefinition will work as
expected.
.IP 105
Provides the behavior that the
.code open-file
function automatically marks a stream open on a TTY devices as a real-time stream
(subject to the availability of the POSIX
.code isatty
function).
Also allows unrecognized backslash escape sequences in regular
expression syntax to simply denote the escaped character literally,
as was historically the case prior to \*(TX 106, so that
.code \ez
for instance denotes
.codn z .
As of \*(TX 106, these are diagnosed as errors.
.IP 102
Up to \*(TX 102, the
.code get-string
function did not close the stream. This old behavior is emulated.
.IP 101
Up to \*(TX 101, the
.code make-like
function incorrectly returned
.code nil
when converting the empty list
.code nil
to string type. This affects numerous generic sequence functions,
causing their result to be
.code nil
instead of an empty string.
.IP 100
Up to \*(TX 100, the
.code split-str
function had an undocumented behavior. When the
.code sep
argument was an empty string, it split the string into
individual characters as if by calling
.codn list-str .
This behavior changed to the currently
documented behavior starting in \*(TX 101.
Also, the arguments of the
.code where
function, which introduced in \*(TX 91, were reversed starting
in \*(TX 101.
.IP 99
Up to \*(TX 99, the substitution of TXR Lisp expressions in
.code @(output)
directives and in the quasistrings of the pattern language
exhibited the buggy behavior that if the TXR Lisp expression
produced a list, the list was rendered as a parenthesized
representation, or the text
.code nil
in the empty list case. Moreover, in the
.code @(output)
case, the value of TXR Lisp expressions was not subject to filtering.
Starting with \*(TX 100, these issues
are fixed, making the behavior consistent with
that of TXR Lisp quasiliterals.
.IP 97
Up to \*(TX 97, the error exception symbols such as
.code file-error
were named with underscores, as in
.codn file_error .
These error symbols existed:
.codn type_error ,
.codn internal_error ,
.codn numeric_error ,
.codn range_error ,
.codn query_error ,
.code file_error
and
.codn process_error .
.coSS Variables @ txr-version and @ lib-version
.desc
The
.code txr-version
variable gives the version of the \*(TX executable. Programs can express
conditional variable based on detecting the version.
The
.code lib-version
variable gives the version of the installed library of \*(TX code accompanying
the executable.
It is expected that these two variables have an identical value. Any
discrepancy in their value indicates an installation whose library or \*(TX
executable were upgraded independently. Should such a situation arise in any
system and cause a problem, \*(TX programs can be defensively coded against it
with the help of these variables.
Some features of the \*(TX library are built into the executable, whereas
others are in the library directory. This aspect of library symbols isn't
specified in this manual; knowing which of these two variables is relevant
to a library feature requires familiarity with the implementation.
.SH* APPENDIX
.SS* A. NOTES ON EXOTIC REGULAR EXPRESSIONS
Users familiar with regular expressions may not be familiar with the complement
and intersection operators, which are often absent from text processing tools
that support regular expressions. The following remarks are offered in the
hope that they may be of some use.
.TP* "Equivalence to Sets"
Regexp intersection is not essential; it may be obtained from complement and
union as follows, since De Morgan's law applies to regular-expression algebra:
.code (R1)&(R2)
=
.codn ~(~(R1)|~(R2)) .
(The complement of the union of the complements of
R1 and R2 constitutes the intersection.) This law works because the regular
expression operators denote set operations in a straightforward way. A regular
expression denotes a set of strings (a potentially infinite one) in a condensed
way. The union of two regular expressions
.code R1|R2
denotes the union of the set
of texts denoted by
.code R1
and that denoted by
.codn R2 .
Similarly
.code R1&R2
denotes a set intersection, and
.code ~R
denotes a set complement. Thus algebraic laws
that apply to set operations apply to regular expressions. It's useful to keep
in mind this relationship between regular expressions and sets in understanding
intersection and complement.
Given a finite set of strings, like the set
.mono
{ "abc", "def" }
.onom
which corresponds to the regular expression
.codn (abc|def) ,
the complement is the set which contains
an infinite number of strings: it consists of all possible strings except
.str abc
and
.strn def .
It includes the empty string, all strings of length 1, all strings
of length 2, all strings of length 3 other than
.str abc
and
.strn def ,
all strings of
length 4, etc. This means that a "harmless looking" expression like
.code ~(abc|def)
can actually match arbitrarily long inputs.
.TP* "Set Difference"
How about matching only three-character-long strings other than
.str abc
or
.strn def ?
To express this, regex intersection can be used: these strings are the
intersection of the set of all three-character strings, and the set of all
strings which are not
.str abc
or
.strn def .
The straightforward set-based reasoning
leads us to this:
.codn ...&~(abc|def) .
This
.code A&~B
idiom is also called set
difference, sometimes notated with a minus sign:
.code A-B
(which is not supported in \*(TX regular-expression syntax). Elements which
are in the set
.codn A ,
but not
.codn B ,
are those elements which are in the intersection of
.code A
with the complement of
.codn B .
This is similar to the arithmetic rule
.codn "A - B = A + -B" :
subtraction is
equivalent to addition of the additive inverse. Set difference is a useful
tool: it enables us to write a positive match which captures a more general set
than what is intended (but one whose regular expression is far simpler
than a positive match for the exact set we want), then we can
intersect this over-generalized set with the complemented set of
another regular expression which matches the particulars that we wish excluded.
.TP* "Expressiveness versus Power"
It turns out that regular expressions which do not make use of the
complement or intersection operators are just as powerful as expressions
that do. That is to say, with or without these operators, regular expressions
can match the same sets of strings (all regular languages). This means that
for a given regular expression which uses intersection and complement, it is
possible to find a regular expression which doesn't use these operators, yet
matches the same set of strings. But, though they exist, such equivalent
regular expressions are often much more complicated, which makes them difficult
to design. Such expressions do not necessarily
. B express
what it is they match; they merely capture the equivalent set. They
perform a job, without making it obvious what it is they do. The use of
complement and intersection leads to natural ways of expressing many kinds of
matching sets, which not only demonstrate the power to carry out an operation,
but also easily express the concept.
.TP* "Example: Matching C Language Comments"
For instance, using complement, we can write a straightforward regular
expression which matches C language comments. A C language
comment is the digraph
.codn /* ,
followed by any string which does not contain the
closing sequence
.codn */ ,
followed by that closing sequence.
Examples of valid comments are
.codn /**/ ,
.code "/* abc */"
or
.codn /***/ .
But C
comments do not nest (cannot contain comments), so that
.code "/* /* nested */ */"
actually consists of the comment
.codn "/* /* nested */" ,
which is followed by the trailing junk
.codn */ .
Our simple characterization of the interior part of a C comment as a string
which does not contain the terminating digraph makes use of the
complement, and can be expressed using the complemented regular expression like
this:
.codn (~.*[*][/].*) .
That is to say, strings which contain
.code */
are matched by
the expression
.codn .*[*][/].* :
zero or more arbitrary characters, followed by
.codn */ ,
followed by zero or more arbitrary characters. Therefore, the complement of
this expression matches all other strings: those which do not contain
.codn */ .
These strings make up the inside of a C comment between the
.code /*
and
.codn */ .
The equivalent simple regex is quite a bit more complicated.
Without complement, we must somehow write a positive match for all strings such
that we avoid matching
.codn */ .
Obviously, sequences of characters other than
.code *
are included:
.codn [^*]* .
Occurrences of
.code *
are also allowed, but only if followed
by something other than a slash, so let's include this via union:
.verb
([^*]|[*][^/])*.
.brev
Alas, we already have a bug in this expression. The
subexpression
.code [*][^/]
can match
.codn ** ,
since a
.code *
is not a
.codn / .
If the next character in the input is
.codn / ,
we missed a comment close. To fix the problem we
revise to this:
.verb
([^*]|[*][^*/])*
.brev
(The interior of a C language comment is any
mixture of zero or more non-asterisks, or digraphs consisting of an asterisk
followed by something other than a slash or another asterisk). Oops, now we
have a problem again. What if two asterisks occur in a comment? They are not
matched by
.codn [^*] ,
and they are not matched by
.codn [*][^*/] .
Actually, our regex must not simply match asterisk-non-asterisk digraphs, but
rather sequences of one or more asterisks followed by a non-asterisk:
.verb
([^*]|[*]*[^*/])*
.brev
This is still not right, because, for instance, it fails to match the interior
of a comment which is terminated by asterisks, including the simple test cases
where the comment interior is nothing but asterisks. We have no provision in
our expression for this case; the expression requires all runs of asterisks to
be followed by something which is not a slash or asterisk. The way to fix this
is to add on a subexpression which optionally matches a run of zero or more
interior asterisks before the comment close:
.verb
([^*]|[*]*[^*/])*[*]*
.brev
Thus the semi-final regular expression is
.verb
[/][*]([^*]|[*]*[^*/])*[*]*[*][/]
.brev
(Interpretation: a C comment is an interior string enclosed in
.codn "/* */" ,
where this interior part
consists of a mixture of non-asterisk characters, as well as runs of asterisk
characters which are terminated by a character other than a slash, except for
possibly one rightmost run of asterisks which extends to the end of the
interior, touching the comment close. Phew!) One final simplification is
possible: the tail part
.code [*]*[*][/]
can be reduced to
.code [*]+[/]
such that the
final run of asterisks is regarded as part of an extended comment terminator
which consists of one or more asterisks followed by a slash. The regular
expression works, but it's cryptic; to someone who has not developed it, it
isn't obvious what it is intended to match. Working out complemented matching
without complement support from the language is not impossible, but it may be
difficult and error-prone, possibly requiring multiple iterations of
trial-and-error development involving numerous test cases, resulting in an
expression that doesn't have a straightforward relationship to the original
idea.
.TP* "The Non-Greedy Operator"
The non-greedy operator
.code %
is actually defined in terms of a set difference,
which is in turn based on intersection and complement. The uninteresting case
.code (R%)
where the right operand is empty reduces to
.codn (R*) :
if there is no trailing
context, the non-greedy operator matches
.code R
as far as possible, possibly to the
end of the input, exactly like the greedy operator. The interesting case
.code (R%T)
is defined as a "syntactic sugar" which expands to the expression
.code ((R*)&(~.*(T&.+).*))T
which means: match the longest string which is matched
by
.codn R* ,
but which does not contain a non-empty match for
.codn T ;
then, match
.codn T .
This is a useful and expressive notation. With it, we can write the regular
expression for matching C language comments simply like this:
.code [/][*].%[*][/]
(match the opening sequence
.codn /* ,
then match a sequence of zero or more
characters non-greedily, and then the closing sequence
.codn */ .
With the non-greedy
operator, we don't have to think about the interior of the comment as set of
strings which excludes
.codn */ .
Though the non-greedy operator appears expressive,
its apparent simplicity may be deceptive. It looks as if it works "magically"
by itself; "somehow" this
.code .%
part "knows" only to consume enough characters so that
it doesn't swallow an occurrence of the trailing context. Care must be taken
that the trailing context passed to the operator really is the correct text
that should be excluded by the non-greedy match. For instance, take the
expression
.codn .%abc .
If you intend the trailing context to be merely
.codn a ,
you must be careful to write
.codn (.%a)bc .
Otherwise, the trailing context is
.codn abc ,
and this means that the
.code .%
match will consume the longest string that does not contain
.codn abc ,
when in fact what was intended was to consume the longest string that
does not contain
.codn a .
The change in behavior of the
.code %
operator upon modifying the
trailing context is not as intuitive as that of the * operator, because the
trailing context is deeply involved in its logic.
On a related note, for single-character trailing contexts, it may be a good
idea to use a complemented character class instead. That is to say, rather than
.codn (.%a)bc ,
consider
.codn [^a]*abc .
The set of strings which don't contain the
character a is adequately expressed by
.codn [^a]* .