summaryrefslogtreecommitdiffstats
Commit message (Collapse)AuthorAgeFilesLines
...
* streams: *stdnull* lazily opens /dev/nullKaz Kylheku2021-06-142-9/+75
| | | | | | | | | | | | | | | | | | | | | The *stdnull* stream has been purely a stream-level abstraction. To make it useful for redirecting real file descriptors around the execution of external programs, we endow it with the ability to open /dev/null when it is asked to provide its file descriptor. * stream.c (struct dev_null): New structure. (dev_null_close, dev_null_get_fd, dev_null_get_prop): New static functions. (null_ops): Wire in the above functions instead of null_close, null_get_fd and null_get_prop. We need new functions because null_close and others do not belong to just the null stream; they are base operations used by other streams as default implementations for some kinds of unimplemented functions. (make_null_stream): Alocate a struct dev_null instead of a struct strm_base. Set the fd to -1. * txr.1: Documented.
* pic: take advantage of recent format work.Kaz Kylheku2021-06-141-7/+2
| | | | | | | * share/txr/stdlib/pic.tl (expand-pic-num): format can now do everything internally that is required of a 0####.## type pattern; we don't have to generate the if logic with the gensym.
* pic: do mkstring cat macro time.Kaz Kylheku2021-06-141-1/+1
| | | | | | | | * share/txr/stdlib/pic.tl (expand-pic-num): The ##...## string that indicates an overflowing field can be created by the macro and inserted into the code as a literal object, rather than inserted as a mkstring call which calculates it run time each time the code is executed.
* format: revise numeric handling.Kaz Kylheku2021-06-143-85/+232
| | | | | | | | | | | | | | | | | | | | | | | There are a number of issues, such as left adjustment not working and such. This needs a better treatment from the requirements level, through to a set of test cases. * stream.c (max, min): Macros macros added, in their usual form. (vformat_num): Implement new rules which suppress the zero and space characters used in place of a sign if they overflow the field width. (formatv): Clamp integer precision field to width - 1 for integers, for consistency with floating-point handling. For floating-point values under ~a and ~s, do not force the second stage precision to width - 1; only clamp it if it is greater. * format.tl: Numerous new tests. * txr.1: Significant redocumenting of this area. The handling of numbers is described as a two stage process, clarifying the changing role of "precision" in the two stages.
* format: new precision - character for zero.Kaz Kylheku2021-06-143-0/+20
| | | | | | | | | | | * stream.c (formatv): The - precision option character produces a "sign" that is a zero. If this is used with leading zeros, it will avoid generating a space. The requirements can use improvement here, but one step at a time. * tests/018/format.tl: Some tests. * txr.1: Documented.
* doc: remove hyphenation.Kaz Kylheku2021-06-141-4/+4
| | | | * txr.1: non-negative changed to nonnegative.
* Version 262txr-262Kaz Kylheku2021-06-118-1262/+1349
| | | | | | | | | | | | | | | * RELNOTES: Updated. * configure, txr.1: Bumped version and date. * share/txr/stdlib/ver.tl: Bumped. * txr.vim, tl.vim: Regenerated. * protsym.c: Likewise. * y.tab.c.shipped: Regenerated. Good catch! forgot to do this for most recent change to parser.y.
* doc: add pic examples.Kaz Kylheku2021-06-111-0/+37
| | | | * txr.1: Examples added.
* pic: support quasiliteral as format string.Kaz Kylheku2021-06-113-22/+80
| | | | | | | | | | | | * share/txr/stdlib/pic.tl (pic): Refactor string compilation code into local function which has access to the overall argument list. Recognize the quasiliteral case and translate by compiling all the string parts, then forming a recombined quasiliteral where the compiled parts are substituted. * tests/018/format.tl: test case for this. * txr.1: Documented.
* doc: codify optional alternatives notation.Kaz Kylheku2021-06-111-1/+7
| | | | | | | * txr.1: The notation [ a | b | c | .. ] is used throughout the document, but the Conventions paragraph describes only square brackets around a single optional element. The notation is now officially codified.
* pic: allow trailing exclamation.Kaz Kylheku2021-06-113-16/+31
| | | | | | | | | | * share/txr/stdlib/pic.tl (expand-pic-num, expand-pic, pic): Allow ### to be followed by a single !. This is not counted toward the field width. * tests/018/format.tl: Cover with some tests. * txr.1: Doc updated.
* pic: clarification and tests.Kaz Kylheku2021-06-112-3/+13
| | | | | | | | * tests/018/format.tl: Add tests which combine overflow flagging with +/-. The space generated by - contributes to overflow. * txr.1: Clarify overflow issue in documentation.
* tests: remove *stderr* to *stdnull* redirection.Kaz Kylheku2021-06-113-17/+7
| | | | | | | | | | | | | | The recent commit 225ff2fa2fdb9e5169db5e2c06dc3b0053b775bb titled "errors: avoid premature release of deferred warnings." obviates the need for dealing with noise when detecting errors from test cases. * patmatch.tl: Remove macro-time-let around several test cases. * tests/012/ifa.tl: Likewise. * tests/common.tl (macro-time-let): Macro removed.
* pic: test cases and fixes.Kaz Kylheku2021-06-112-8/+98
| | | | | | | | | | | | | | | | | | * share/txr/stdlib/pic.tl (expand-pic-num): Bug: when a field overflows, the (rest ...) call truncates the leftmost digit. A failing test case is (pic "#.#" 12) which produces "2.0" instead of "12.0". Firstly, we only need that logic at all in the zero padding case. When the number is positive, we stick in the + request, so we are sure to get a + character. The rest call then predictably chops off the + rather than a digit. (pic-join-opt): Fix two bugs here in the string-string combine case: using s2 instead of s1, and not splicing in rest. (expand-pic, pic): Implement tightening for escape sequences. If ~ is not followed by anything, or not followed by the documented characters for escaping, it is erroneous. * format.tl: Battery of new tests.
* tests: support testing for warnings.Kaz Kylheku2021-06-111-5/+15
| | | | | | | | * tests/common.tl (error-to-sym): Catch warnings, and convert to :warning symbol. (vtest): Support new kinds of expected value: :warning. The test is satisfied if it throws a warning at expansion time or during evaluation.
* errors: avoid premature release of deferred warnings.Kaz Kylheku2021-06-112-4/+5
| | | | | | | | | | | | | | | | | | | | | | | | | We don't want to be unconditionally releasing deferred warnings when error exceptions occur in evaluation or compilation. The reason is that the error might be handled, for instance by a speculative expansion. Then unwanted noise occurs, because deferred warnings have been released prematurely. * eval.c (eval_exception): Do not call uw_release_deferred_warnings here. (error_trace): But do call uw_release_deferred_warnings here, before printing anything else. We want to preserve the behavior that when error information is actually being printed to a stream, any deferred warnings are dumped first because they might pertain to the error. (I may revisit this requirement; perhaps deferred warnings rarely, if ever, pertain to an error). * share/txr/stdlib/error.tl (compile-error): Do not dump deferred warnings unconditionally. Only dump them if we are also printing the error message to stderr. Secondly, do not output the error message at all, unless there is no handler for the error.
* doc: document pic macro.Kaz Kylheku2021-06-102-0/+194
| | | | | | * txr.1: pic macro documented. * share/txr/stdlib/doc-syms.tl: Updated.
* format: fix precision field leading zero problems.Kaz Kylheku2021-06-102-8/+27
| | | | | | | | | | | | | | | | | * stream.c (formatv): Do not recognize multiple leading zeros as a single one; once the zero flag is set, if another zero is seen, it must be treated as one of the digits specifying the precision value. New requirement: before processing a format specifier, check for the situation that the leading zero has been specified, but no precision. Convert this situation to that of a precision of zero being given, with no leading zero. * txr.1: Document the ambiguity around the leading zero and how it is being handled when only the leading zero flag is given, and no actual precision. Add a note about what happens when zero precision is specified in ~a in conjunction with a floating-point value. Misspelled "pas" word fixed.
* New macro: pic.Kaz Kylheku2021-06-102-0/+125
| | | | | | | | | * lisplib.c (pic_instantiate, pic_set_entries): New static functions. (lisplib_init): Register autoloading of pic.tl module via new functions. * share/txr/stdlib/pic.tl: New file.
* listener: complete macros and operators after quote.Kaz Kylheku2021-06-101-6/+11
| | | | | | | | | | | This fixes the problem that (doc 'wh[Tab] will not complete the macro name while. * parser.c (find_matching_syms): Recognize the kind context symbol 'Q', under which a which a macro or operator are eligible for completion. (provide_completions): Calculate kind as 'Q' if the previous character is a ' quote or ^ backquote.
* ffi: lazily calculate libffi type descriptors.Kaz Kylheku2021-06-101-122/+110
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | I think this is a nicer treatment of the libffi situation. We only need the type descriptors for arrays, structs and unions if they are ever used as return values or arguments in a libffi call descriptor. We don't have to waste time and memory allocating them otherwise, and in the case of arrays, we can allocate the full descriptor that includes every member, when that is needed. Structures that are involved in calls, but are passed and returned by pointer, not directly, will not have this cruft instantiated at all. * ffi.c (struct txr_ffi_type): New function pointer member calcft. This specifies the lazy initialization funtion for the libffi stuff. Only arrays, structs and unions have a non-null pointer here. If the pointer is null, it indicates that the type doesn't require lazy initialization: either because it was initialized eagerly, or because lazy initialization was already done. (ffi_get_type): If the type has a calcft, then call it and null it out, to ensure tffi->ft. (ffi_struct_clone): Do not deal with ft and elements here at all. (ffi_union_clone): Short-lived static function removed, because it is identical to ffi_struct_clone again. (ffi_struct_calcft, ffi_union_calcft, ffi_array_calcft): New static functions, implementing different strategies for calculating the libffi elements array. For unions, I invented this strategy: we pretend that a union is atually a structure made of repetitions of the type which has the strictest alignment. This is uncharted territory because libffi doesn't support unions at all. (make_ffi_type_struct, make_ffi_type_union, make_ffi_type_array): Do not deal with tf or elems, except for deleting them if we are re-initializing the object. Leave these pointers null. Install the appropriate calcft function. (ffi_array_clone): Function removed; the array type can use ffi_simple_clone, since the clone function doesn't deal with the libffi stuff. Why was that needed until several commits ago? Because, subject to HAVE_FFI, it performed the assignment ft->elements = copy->elements.
* arith: switch sum and prod to seq_iter.Kaz Kylheku2021-06-093-34/+72
| | | | | | | | | | | | | | * arith.c (nary_op_keyfun): Static function removed. (nary_op_seq, nary_op_seq_keyfun): New static functions. (sumv, prodv): Static functions removed. (sum, prod): Reimplement using nary_op_seq and nary_op_seq_keyfun. Conversion of sequence to list smuggled via args is gone. * tests/016/arith.tl: new sum and prod tests. * txr.1: Note about sum and prod taking an iterable sequence added.
* reduce-left: rewrite using seq_iter.Kaz Kylheku2021-06-093-6/+32
| | | | | | | * lib.c (reduce_left): Use sequence iteration instead of list operations. * txr.1: Add a note to the documentation.
* parser: bug #; and #S don't play.Kaz Kylheku2021-06-091-6/+15
| | | | | | | | | When #; is used to comment out syntax containing #S, an attempt is made to instantiate a structure using nil as the type, and an exception occurs. * parser.y (hash, struct, tree): Check for parser->ignore and produce nil instead of making any of these objects.
* ffi: bug: always using ffi_prep_cif_var.Kaz Kylheku2021-06-093-7/+10
| | | | | | | | | | | | | | | * share/txr/stdlib/ffi.c (deffi): Fix misnamed variable. The second value coming from sys:analyze-argtypes is the number of fixed arguments, not the number of variadic arguments. Furthermore, if this number is equal to nargs, we were supposed to have been passing nil instead to ffi-make-call-desc, which indicates the ordinary non-variadic function. We were always always passing a non-nil value, so always requesting variadic. That is fixed now thanks to the change to ffi_make_call_desc. * ffi.c (ffi_make_call_desc): Register the function as variadic if either nfixed is specified as nil, or if it is equal to ntotal. * txr.1: Document the convention change for ffi-make-call-desc.
* ffi: earnest implementation of FFI struct elements.Kaz Kylheku2021-06-091-23/+72
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | This commit tries to address a flaw in our use of libffi. When we construct a structure type, we do not fill in the "elements" array of the libffi type to describe every element. Instead we calculate the size and alignment ourselves. This breaks when passing small structures by value on some platforms. There are some areas where we are still winging it. We allow passing of arrays by value, which are simulated using FFI structure types. To allocate the descriptor earnestly for such an object means allocating an array of ffi_type pointers as large as the array, which is wasteful. So we cap it to 20. This 20 comes form the idea that anything bigger than a 16 byte structure is passed on the stack rather than in registers. For unions, we choose the largest member and inform libffi of its type. * ffi.c (struct txr_ffi_type): elements member changes to from array of 1 to pointer. (ffi_type_struct_destroy_op): Consolidate HAVE_LIBFFI code. Don't bother setting tft->ft->elements to null. Free tft->elements. (ffi_struct_clone): Clone the elements array. (ffi_union_clonse): New static function. Unions need their own clone function now because their elements array has a size not related to the number of members. It contains two pointers, including the null terminating one. (make_ffi_type_struct): Allocate the elements array, and fill it. We wing it around bitfields, which libffi doesn't support. This will be revisited, but it's likely a bad idea to be passing bitfield-endowed structures by value. (make_ffi_type_union): Allocate the elements array to a fixed size of two pointers. Fill it in with the largest member's type. (ffi_array_clone): Clone the elements array, which is up two 20 pointers plus a null terminator. (make_ffi_type_array): Allocate the elements array up to 20 elements plus null pointer, and fill it with repetitions of the element libffi type.
* ffi: fix leak on struct/union redefinition.Kaz Kylheku2021-06-091-0/+6
| | | | | | | * ffi.c (make_ffi_type_struct, make_ffi_type_union): When we are replacing the existing type, if HAVE_FFI is true, we must free the tft->ft object before clobbering the tft with zeros.
* ffi: leak fix on !HAVE_FFI builds.Kaz Kylheku2021-06-091-2/+18
| | | | | | | | | | The ffi_type_struct_destroy_op only frees tft->ft if HAVE_FFI is true. But the object is allocated in several places without regard for HAVE_FFI. * ffi.c (ffi_struct_clone, make_ffi_type_struct, make_ffi_type_union, ffi_array_clone, make_ffi_type_array): Do not allocate a ffi_type object if HAVE_FFI is false.
* tags: recognize deffi-struct.Kaz Kylheku2021-06-081-1/+1
| | | | | * tags.tl (process-form): Recognize deffi-struct together with defstruct.
* tags: handle literals for undefined structures.Kaz Kylheku2021-06-081-0/+1
| | | | | * tags.tl: bind *read-unknown-structs* to t around the file walk.
* parser: new *read-unknown-structs* variable.Kaz Kylheku2021-06-086-283/+310
| | | | | | | | | | | | | | | | | | | | | * parser.c (read_unknown_structs_s): New symbol variable. (parser_common_init): Initialize read_unknown_structs flag member of the parser structure from the new special variable. (parse_init): Initialize read_unknown_struct_s variable. Register the *read-unknown-structs* dynamic variable. * parser.h (struct parser): New member, read_unknown_structs. (read_unknown_structs_s): Declared. * parser.y (struct): Generate the struct literal syntax not only for quasiquoted structures, but for structures with an unknown type name, if the read_unkonwn_structs flag is set. * txr.1: Documented. * share/txr/stdlib/doc-syms.tl: Regenerated. * y.tab.c.shipped: Regenerated.
* lib: new function, fill-vec.Kaz Kylheku2021-06-086-0/+157
| | | | | | | | | | | | | | * eval.c (eval_init): Register fill-vec intrinsic. * lib.c (fill_vec): New function. * lib.h (fill_vec): Declared. * tests/010/vec.tl: New file. * txr.1: Documented. * share/txr/stdlib/doc-syms.tl: Updated.
* ffi: new macros deffi-struct and deffi-union.Kaz Kylheku2021-06-084-1/+40
| | | | | | | | | | | | * lisplib.c (ffi_set_entries): Autoload for deffi-struct and deffi-union symbols. * share/txr/stdlib/ffi.tl (deffi-struct, deffi-union): New macros. * txr.1: Documented. * share/txr/stdlib/doc-syms.tl: Updated.
* expander: expand must only ignores unbound warnings.Kaz Kylheku2021-06-073-6/+38
| | | | | | | | | | | | | | | | | | | The expand function must not muffle all deferred warnings. That causes the problem that a form like (inc var a.bar) fails to produce a warning due to bar not being the slot of any structure. The expand function must only muffle warnings about undefined functions and variables. * eval.c (muffle_unbound_warning): New static function. (no_warn_expand): Use muffle_unbound_warning as handler, rather than uw_muffle_warning. * tests/012/struct.tl: Fix two test cases here which test the expand function using a form that references a nonexistent slot. These now generate a warning, so we use the slot name b rather than d, which is defined. * txr.1: Documented change to expand.
* args: correction in assertion.Kaz Kylheku2021-06-071-1/+1
| | | | | | | * args.c (args_normalize_least): The bug_unless here should be checking that minfill is not beyond argc->argc. That's the situation that would cause the loop to overflow the fixed argument argument storage in args indicated by args->argc.
* ffi: allow conversion of carray objects under cptr.Kaz Kylheku2021-06-062-4/+41
| | | | | | | | | | | | | | * ffi.c (ffi_cptr_put): If the object is not a cptr, try it as a carray. This requires the tft to have an eltype, requiring a change in ffi_type_compile. (ffi_type_compile): When compiling a parametrized cptr object, we now look up the type symbol as a FFI type, and store the result as the tft->eltype. If the symbol is not an FFI type, then this lookup returns nil. If the eltype is non-nil, then there is the possibility it can match the element type of a carray. * txr.1: Documented.
* ffi: allow nil object to convert under closure type.Kaz Kylheku2021-06-062-4/+31
| | | | | | | | * ffi.c (ffi_closure_put): Only diagnose a bad object if it's not nil, otherwise let the null value of p through. This is useful because there are sometimes C interfaces which take an optional function pointer, whereby a null value indicates that the pointer is not specified.
* ffi: add space in ffi-call-desc's printed rep.Kaz Kylheku2021-06-061-1/+1
| | | | | * ffi.c (ffi_call_desc_print_op): Print a space between the class symbol and name so they are not run together.
* dlsym: improve diagnostic.Kaz Kylheku2021-06-051-1/+2
| | | | | | | * sysif.c (dlsym_error): Show the library object and symbol even when the system has a diagnotic. On Cygwin/Cygnal there is a generic "No such process" error that doesn't mention the library or symbol.
* matcher: doc fix, quasiquote examples, new test.Kaz Kylheku2021-06-042-1/+42
| | | | | | | * tests/011/patmatch.tl: New test case. * txr.1: Heading fix: Quasiquote matching notation, not quasiliteral. Examples of quasiquote notation added.
* bugfix: do not expand defun body with name in scope.Kaz Kylheku2021-06-041-4/+1
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | This is a revert of November 2016 commit 606132c336dbeb0dd8bb851a64c97f2c11b76a85. The commit claims that it fixes a bug, but in fact it introduces one. There is no discussion in that commit about what motivated it. The commit which follows that one introduces a naivey-implemented diagnostic for catching unbound functions at macro-expansion time, so the likely motivation for this wrong fix was to suppress false positives from that naive diagnostic for recursive functions. That has long since been replaced by a better approach. Because of the bug, we cannot do this very useful thing: we cannot write an inline version of a funtion as a macro first, and then a real function which just calls that macro: (defmacro foo (arg) ...) (defun foo (arg) (foo arg)) The bug causes the (foo arg) call in this function not to be expanded, due to the shadowing. * eval.c (do_expand): When expanding defun, do not introduce the function's name as a lexical function binding, because it isn't one.
* FFI: big improvement in bad call diagnosis.Kaz Kylheku2021-06-044-14/+34
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | FFI has the problem that when things go wrong in calls, or in specifications of functions, the diagnostics refer to an internal function like ffi-call or ffi-make-call-desc, which is not helpful in identifying the error. We want the diagnostics to refer to the foreign function, or foreignb callback wrapper, to which the problem pertains. The approach taken is to stick the name symbol into the ffi-call-desc object. Functions which work with a ffi-call-desc can pull out the name and use it for reporting. * ffi.c (struct txr_ffi_call_desc): Add name member. (ffi_call_desc_print_op): Include name in printed representation. (ffi_desc_mark_op): Mark the name. (ffi_make_call_desc): Take new argument to specify the name, storing it into the structure. If it is specified,then use that name for reporting errors, otherwise stick with ffi-make-call-desc. (ffi_call_wrap, ffi_closure_dispatch, ffi_closure_dispatch_safe, ffi_make_closure): Use the name from the call descriptor, or else the function's own name if that is nil. (ffi_init): Update registration of ffi-make-call-desc intrinsic to five arguments with four required. * ffi.h (ffi_make_call_desc): Declaration updated. * share/txr/stdlib/ffi.tl (deffi, deffi-cb-expander): Pass the name symbol down to ffi-make-call-desc. * txr.1: Documented.
* lib: oversight, neglected struct literal printing.Kaz Kylheku2021-06-031-0/+3
| | | | | * lib.c (obj_print_impl): print (sys:struct-lit ...) syntax as #S(...).
* json: fix quasiquote print-read consistency issueKaz Kylheku2021-06-031-10/+24
| | | | | | | | | | | | * lib.c (out_json_rec): When printing keys that might be potentially quasiquoted symbols that look like ~abc, we must avoid the condensed {~abc:~def} style. This is because abd:~def looks like a single symbol, where abc is a package qualifier. To be safe and readable at the same time, we add spaces whenever either the key or the value are conses, indicating non-JSON syntax. Spaces are added on both sides of the colon, and also after the preceding comma, if there is a previous item.
* json: pattern matching test cases and bugfix.Kaz Kylheku2021-06-033-257/+276
| | | | | | | | | | | | | * parser.y (json_val): We must nreverse the json_pairs which were pushed in right to left order. This didn't matter for constructing hashes so it was left out, but under quasiquoting the order matters: it determines the order of evaluation and of pattern matching. * tests/011/patmatch.tl: New quasiquoting pattern matching cases, including JSON. * y.tab.c.shipped: Regenerated.
* quips: new one.Kaz Kylheku2021-06-031-0/+1
| | | | * share/txr/stdlib/quips.tl (sys:%quips%): New entry.
* json: improve escaping for script tags.Kaz Kylheku2021-06-033-14/+58
| | | | | | | | | | | * lib.c (out_json_str): Strengthen the test for escaping the forward slash. It has to occur in the sequence </script rather than just </. Recognize <!-- and --> in the string, and encode them. * tests/010/json.tl: Cover this area with some tests. * txr.1: Documented.
* matcher: better error handling for backquotes.Kaz Kylheku2021-06-022-6/+20
| | | | | | | * share/txr/stdlib/match.tl (transform-qquote): Handle hash error case with separate pattern. Use compile-error and *match form instead of error. Diagnose splicing unquote and nested quasiquote.
* matcher: quasiquote matching.Kaz Kylheku2021-06-022-10/+148
| | | | | | | | | | | | | | | | | | | | | | | This allows (when-match ^(,a ,b) '(1 2) (list a b)) -> (1 2) which is a nice alternative that is supported by some Lisp pattern matchers. We don't need it since we have (@a @b). The motivation is JSON matching. (when-match ^#J{"foo" : {"x" : ~val}} #J{"foo" : {"x" : "y"}} val) -> "y" * share/txr/stdlib/match.tl (compile-match): Recognize qquote case and handle via transform-qquote function. (non-triv-pat-p): Let's declare quasiquotes to be nontrivial. (transform-qquote): New function: transform quasi-quoted syntax into regular pattern matching syntax. * txr.1: Documented.
* json: fix two test cases for Windows.Kaz Kylheku2021-06-021-2/+5
| | | | | | | | | | | | * tests/010/json.tl: on Windows characters are limited to the BMP range 0 to #\xFFFF. The character escape \x10437 is out of range, and so throws an error, simply from that syntax being read. The two test cases which use this character are clumped into their own test form, which is executed conditionally on wide characters being more than two bytes. Because the expression is still parsed on Windows, we read the troublesome character from a string at run-time, and interpolate it.
generated by cgit v1.2.3 (git 2.25.1) at 2025-03-19 10:07:56 +0000