summaryrefslogtreecommitdiffstats
Commit message (Collapse)AuthorAgeFilesLines
* compiler: cosmetic: merge set assignments.Kaz Kylheku2021-09-301-7/+7
| | | | | | * stdlib/optimize.tl (basic-blocks join-block): Merge set forms into one. (basic-blocks elim-dead-code): Likewise.
* compiler: improvement in next-block linking.Kaz Kylheku2021-09-291-3/+3
| | | | | | | * stdlib/optimize.tl (basic-blocks link-graph): Do not search the entire list for a block's successor. Iterate over the cdr of the list in parallel, so that the next block is directly available at each iteration.
* compiler: remove impossible cases in jump threading.Kaz Kylheku2021-09-291-8/+4
| | | | | | | | | | | | | | * stdlib/optimize.tl (basic-blocks thread-jumps-block): There can't be any instructions in a basic block after an if or ifq, so in these cases, jrest is always nil. Let's ignore that nil efficiently with @nil, and get rid of the cut-block branches of the code. There is a similar case in peephole-block, but the target of the jump is an (end ...) which doesn't necessarily end a basic block. I temporarily put in an (assert (null jrest)), and, surprisingly, it never went off during a rebuild of the library or running of the test case. Still, only a jend ends a basic block; it would not be correct to simplify it like these two cases in thread-jumps-block.
* compiler: peephole: merge basic blocks when jmp removed.Kaz Kylheku2021-09-291-5/+16
| | | | | | | | | | | | | | | | | | | | | | | | | | | When a jmp instruction is removed from (necessarily) the end of a basic block, that basic block can be merged with the next one, and marked for re-scanning. A test case where this eliminates wasteful register-register move instruction is (match #(@a) #(3) a). * stdlib/optimize.tl (basic-blocks): New slot, tryjoin. (basic-blocks join-block): Null out the instruction list of the joined block. This helps if we do this during peephole processing, because it happens in the middle of an iteration over a list of blocks which can still visit the next block that has been merged into its predecesor; we don't want to be processing instructions that are no longer relevant. (basic-blocks peephole-block): In the one case where a conditional instruction is deleted from the end of the basic block, we add the block to the rescan list, and also to the tryjoin list. If the block can be merged with the next one, that can create more opportunities for peephole optimization. (basic-blocks peephole): Use zap in a few places to condense the logic of sampling a state variable that needs to be nulled out. Add the processing of the tryjoin list: pop basic blocks from the list, and try to merge them with their successor, if possible. We handle cases here where the next block could itself be in tryjoin. Also, if we join any blocks, we set the recalc flag to recalculate the liveness info.
* compiler: code clean-up in peephole optimizer.Kaz Kylheku2021-09-281-5/+5
| | | | | | | | | | | * stdlib/optimize.tl (basic-blocks peephole-block): When we match a branching instruction, including jend, we know that's the end of the basic block. So there is no need to splice the (rest insns) into the output; let's get rid of that. On the other hand, there is also no need to have a specific pattern match for the end of the list such as ((jmp @label)). This costs extra cycles to validate. Let's consistently match these basic-block terminating instructions using prefix patterns like ((jmp @label) . @nil)).
* quantile: fix test needing nonportable amount of precision.Kaz Kylheku2021-09-272-11/+14
| | | | | | | | | | | | | | * tests/common.tl (sstest): New macro. Like stest, but the right hand side is an object which the macro turns to a string, rather than expecting a string. * tests/016/arith.tl: Use the sstest macro for the main quantile test to compare the result and expected value as character strings rather than objects. Specify the expected values using no more than 14 decimal digits of precision, and over the scope of the test case, restrict floating-point printing to 14 digits. Thus, we effectively have quick and dirty epsilon comparison to 14 digits that recurses over the list, without having to write that as a function.
* quips: one in French.Kaz Kylheku2021-09-261-0/+1
| | | | * quips.tl: New quip.
* New variable: *child-env*.Kaz Kylheku2021-09-266-5/+95
| | | | | | | | | | | | | | | | | | | | | | | | | | This specifies the environment to be used for executing programs. * stream.c (open_subprocess, run): Check *child-env* variable and if other than t, then install the environment before execvp. In the spawn-based version of run, we save and restore the environment around the spawn call, if *child-env* is in effect. * sysif.c (child_env_s): New symbol variable. (exec_wrap): If *child-env* is other than t, then save the environment in a list, and install the specified environment before calling execvp. If that function returns, restore the environbment. * sysif.h (child_env_s): Declared. (child_env): New macro. * tests/018/process.tl: New tests. * txr.1: Documented. * stdlib/doc-syms.tl: Updated.
* New function: replace-env.Kaz Kylheku2021-09-256-0/+98
| | | | | | | | | | | | | | | | | | | Using this new function together with env, it's now possible to save the set of environment variables, clobber it to a specified set (possibly empty) and then restore it. Useful for improved security in running child processes. * lib.[ch] (chk_substrdup_utf8): New function. * sysif.c (replace_env): New function. (sysif_init): Register replace-env intrinsic. * sysif.h (replace_env): Declared. * txr.1: Documented. * stdlib/doc-syms.tl: Updated.
* path access tests: use real credentials.Kaz Kylheku2021-09-252-28/+33
| | | | | | | | | | | | | | | | | The various accessibility functions like path-writable-to-me should use the real credentials, the same way that the POSIX access function does. This makes them much more useful and secure in setuid programs, since they answer the question "does the underlying user, without these elevated privileges, have this access". * stdlib/path-test.tl (path-mine-p): Use getuid, not geteuid. (path-my-group-p): Use getgid, not getegid. (sys:path-access, path-private-to-me, path-strictly-private-to-me): Use getuid, getgid and rename euid variable to uid. * txr.1: Updated.
* quantile: fix failing test.Kaz Kylheku2021-09-251-1/+1
| | | | | * arith.tl: Somehow I committed some wrong expected numbers in a quantile test, yet didn't catch the test failure.
* path-search: rewrite in C with saner semantics.Kaz Kylheku2021-09-256-69/+93
| | | | | | | | | | | | | | | | | | | | | | | | | | The new function: - just returns the name if it contains path name components. - returns nil if the name is "." or "..". - tests for existence only, not permission to execute. * lisplib.c (path_test_set_entries): Do not auto-load path-test module on the path-search symbol, since it is no longer implemented there. * stdlib/path-test.tl (path-search): Function removed. * stream.c (path_var_sep_char): New global variable. (path_search): New function. (detect_path_separators): Also set path_var_sep_char to semicolon on Cygnal. (stream-init): Register path-search intrinsic here now. * stream.h (path_var_sep_char, path_search): Declared. * tests/018/path-test.tl: New tests. * txr.1: Documentation revised for path-search.
* New variants of each operator for sum and product.Kaz Kylheku2021-09-236-1/+290
| | | | | | | | | | | | | | | | | | | | | | | | | * lisplib.c (arith_each_instantiate, arith_each_set_entries): New functions. (each_prod_set_entries): Add sum-each-prod, sum-each-prod*, mul-each-prod and mul-each-prod* as autoload triggers for each-prod.tl, where those macros are now defined. (lisplib_init): Register autoloading of arith-each.tl via the two new functions. * stdlib/arith-each.tl: New file. * stdlib/each-prod.tl (sys:expand-each-prod*): Handle sum-each-prod* and mul-each-prod* in the same way, by mapping to their parallel binding counterparts. (sys:expand-arith-each-prod): New function. (sym-each-prod, mul-each-prod, sum-each-prod*, mul-each-prod*): New macros. * tests/016/arith.tl: New tests. * txr.1: Documented. * stdlib/doc-syms.tl: Updated.
* math: quantile estimator using P-Squared algorithm.Kaz Kylheku2021-09-228-1/+452
| | | | | | | | | | | | | | | | | | | * Makefile (psquare.o): New object file. * arith.c (psq_ops): New static structure. (quant_fun): New static function. (quantile): New function. (arith_init): Register quantile intrinsic. * arith.h (quantile): Declared. * psquare.c, psquare.h: New files. * tests/016/arith.tl: New tests. * txr.1: Documented. * stdlib/doc-syms.tl: Updated.
* doc: sock-opt: found proper way to escape ...Kaz Kylheku2021-09-221-2/+2
| | | | | | | * txr.1: To begin a troff line with a period that is to be treated as literal, we use \& before that dot, not simply \. This was giving warnings and not rendering properly. The \& produces good output in man, HTML and PDF.
* doc: fix problem in errno retrieval example.Kaz Kylheku2021-09-221-2/+0
| | | | * txr.1: Remove stray comment from string-get-code example.
* vm, structs: use FLEX_ARRAY macro.Kaz Kylheku2021-09-192-2/+2
| | | | | | | | * vm.c (struct vm_closure): Use the FLEX_ARRAY macro to define the trailing array at the end of the structure instead of hard-coding [1]. * struct.c (struct struct_inst): Likewise.
* vm: fix self name of vm-desc-nlevels.Paul A. Patience2021-09-171-1/+1
| | | | * vm.c (vm_desc_nlevels): vm_desc_nlevels -> vm-desc-nlevels.
* lib: fix self name of cmp-str.Paul A. Patience2021-09-171-3/+3
| | | | * lib.c (cmp_str): Fix self name and use it in uw_throwf call.
* eval: fix single-list-argument case in maprodo.Paul A. Patience2021-09-171-1/+1
| | | | | | | Calling maprodo with one list argument would fall back on mappend rather than mapdo. * eval.c (maprodo): mappendv -> mapdov.
* doc: interchange buf-str and str-buf descriptions.Paul A. Patience2021-09-171-15/+15
| | | | | * txr.1: Interchange the first arguments of buf-str and str-buf, and their descriptions also.
* doc: fix various trivial typos.Paul A. Patience2021-09-171-7/+7
| | | | | | * txr.1: Fix typos in .meIP and .mets lines. Fix a .code that should be .codn. Add missing closing parenthesis in description of greater function. Unparenthesize make-random-state in a .code.
* compiler: reduce single-arg logior and logand.Paul A. Patience2021-09-141-1/+1
| | | | | * stdlib/compiler.tl (compiler comp-fun-form): Reduce single-argument logior and logand calls to just the argument.
* defset: add set-mask and clear-mask.Paul A. Patience2021-09-147-24/+76
| | | | | | | | | | | | | | | | | | | | * stdlib/defset.tl (set-mask, clear-mask): New update macros. * stdlib/optimize.tl (calc-liveness): Use the new macros. * stdlib/socket.tl (sys:str-inaddr-net-impl, str-in6addr-net): Same. * stdlib/termios.tl (set-iflags, set-oflags, set-cflags, set-lflags, clear-iflags, clear-oflags, clear-cflags, clear-lflags): Same. * lisplib.c (defset_set_entries): Add set-mask and clear-mask to autoload symbols for defset. * txr.1: Documented. * stdlib/doc-syms.tl: Updated.
* compiler: fix random perturbance in reg allocation.Kaz Kylheku2021-09-141-2/+2
| | | | | | | | | | | | | | | At optimization level 2 or higher, an issue occurs whereby code generation exhibits instabilities. The same code is compiled slightly differently (but not incorrectly) depending on irrelevant circumstances, due to some different registers being used. * stdlib/compiler.tl (compiler eliminate-frame): Do not free the newly allocated t-registers inside a dohash loop. We have a separate list of them in order; just hand that off to free-tregs. The dohash loop is not ordered, because it traverses a hash, which is keyed by object identities; i.e. machine addresses assigned by memory allocation.
* seq_iter: gc crash marking vector iterator.Kaz Kylheku2021-09-131-4/+4
| | | | | | | | | * lib.c (si_vec_ops): This must be initialized with seq_iter_ops_init_nomark, since it uses a cnum index, and not a val iter; the seq_iter_mark_op will pass the cnum bit pattern to gc_mark an cause a crash. (si_null_ops): While we are at it, this should also use seq_iter_ops_init_nomark, because it->ui.iter is always nil.
* hash: gc problem in copy-hash.Kaz Kylheku2021-09-131-1/+1
| | | | | | | | | * hash.c (copy_hash): The order of allocating the hash object and vector is incorrect. The hash must be allocated last, like it is in do_make_hash and make_similar_hash. If the vector is allocated after the hash, it can trigger gc, and then the garbage collector will traverse the uninitialized parts of the hash object.
* ffi, sockets: add sock-opt and sock-set-opt.Paul A. Patience2021-09-126-5/+329
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | The new sock-opt and sock-set-opt functions are wrappers around getsockopt and setsockopt, respectively. All POSIX socket options are registered. Platform-specific options may be added in the future. * ffi.c (sock_opt, sock_set_opt): New functions. (ffi_init): Register sock-opt, sock-set-opt, sol-socket, ipproto-ip, ipproto-ipv6, ipproto-tcp, ipproto-udp, 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, so-type, ipv6-join-group, ipv6-leave-group, ipv6-multicast-hops, ipv6-multicast-if, ipv6-multicast-loop, ipv6-unicast-hops, ipv6-v6only, tcp-nodelay. * lisplib.c (sock_set_entries): Add sock-opt and sock-set-opt. * stdlib/socket.tl (sock-opt): Define as syntactic place. * tests/014/socket-misc.tl: New cases, for sock-opt. (set-and-get): New macro. * txr.1: Documented. Also, mention that sock-bind enables so-reuseaddr. * stdlib/doc-syms.tl: Updated.
* sockets: add test for recent sock-peer place issue.Kaz Kylheku2021-09-111-1/+5
| | | | | * tests/014/socket-misc.tl: Test that we can perform a place update operation on a (sock-peer s) place.
* sockets: bug in clearing SOCK_* type flags.Kaz Kylheku2021-09-112-6/+9
| | | | | | | | | | | | | | | | | | | | Paul A. Patience reports a regression in 3d5d525eb525cfad8f643917c31e3d9fedce2874, whereby the code marked with #if SOCK_NONBLOCK || SOCK_CLOEXEC is being excluded by the preprocoessor, and so those flags are not being cleared from the socket type that we retain. This is because these preprocessor symbols are not necessarily integer constants. They may expand to C enum identifiers, in which case, in preprocessor expressions they appear to take on the value zero. * socket.c (open_socket, socketpair_wrap): Use the if statement to test for SOCK_NONBLOCK or SOCK_CLOEXEC being nonzero, rather than a preprocessor #if. This should still be optimized away as unreachable code if they are zero. * tests/014/socket-misc.tl: New file.
* sockets: remove excess apostrophe in sock-peer.Paul A. Patience2021-09-111-1/+1
| | | | | * stdlib/socket.tl (sock-peer): Remove excess apostrophe in getter.
* doc: fix various important typos.Paul A. Patience2021-09-111-10/+10
| | | | | | | | | * txr.1: use-symbol -> use-sym; socket-open -> open-socket; af-inet -> af-unix in description of open-socket; r+ -> r+b in description of sock-accept; get-obj -> fill-obj in description of fill-obj; offs -> offset in description of carray-buf; caddr-ref -> carray-ref; caddr-refset -> carray-refset; fix int-carray's argument list.
* sockets: throw exception if socket call fails.Paul A. Patience2021-09-111-0/+5
| | | | | | | | | | The open_socket function was not checking the result of the socket call for failure, which would cause datagram socket streams to be initialized with an invalid file descriptor (-1) and, when opening stream sockets, throw an error not representative of the actual problem. * socket.c (open_socket): Throw exception if socket call fails.
* mmap: fix typo in error message.Paul A. Patience2021-09-111-1/+1
| | | | * ffi.c (mmap_op): mmaped -> mmapped.
* matcher: use function for match and match-ecase error.Kaz Kylheku2021-09-101-2/+8
| | | | | | | * stdlib/match.tl (match-pat-error, match-error): New functions. (match, match-ecase): Generate more compact code which just calls match-pat-error rather than throwf, and doesn't contain any string literals.
* oop: remove repeated code in slot lookup.Kaz Kylheku2021-09-091-24/+8
| | | | | | | | * struct.c (lookup_slot): Do not repeat the slot lookup logic for the case when the symbol's slot_cache must be boostrapped, which only happens exactly once in the life of a symbol. Just allocate the cache if necessary and fall through to the have-cache case.
* gcc11: warnings related to struct args allocation..Kaz Kylheku2021-09-084-7/+19
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | As reported by Paul A. Patience, GCC 11 warns about situations in a few places where we do args_decl(args, 0) to allocate an absolutely empty argument list object. (This by the way, is only done in places where we are absolutely sure that the function we call will not be accessing the arguments. The usual rule is that there have to be at least ARGS_MIN arguments allocated, currently 4, and code relies on that being the case! So the places which call args_decl(args, 0) are coded carefully, checking that args is not passed anywhere where ARGS_MIN space is required.) Anyway, in the 0 case, the val arg[1] member of struct args is not allocated at all, because we call alloca(offsetof (struct args, arg)). Now GCC 11 notices this and complains that accesses to the other members like args->fill or args->list are using a struct that has not been entirely allocated. This, even though those members lie entirely within the allocated area. The fix for this is two faceted. Firstly, on C99, this diagnostic goes away if we make one simple change: declare the arg array as a flexible array member: val arg[]. However, we still support C90 in maintainer mode. So in maintainer mode, we stick with the 1. But we ensure that the places which call args_decl(args, 0) will pass 1 instead of 0, so the whole structure is allocated. * lib.h (FLEX_ARRAY): New macro: empty definition in C99 or later, otherwise 1. * args.h (struct args): Declare the size of the arg member using the new FLEX_ARRAY macro from lib.h. (ARGS_ABS_MIN): New macro, the absolute args minimum: zero in C99 mode, 1 in maintainer C90 mode. * ffi.c (ffi_struct_in, ffi_struct_get, make_zstruct): Use ARGS_ABS_MIN instead of 0 when preparing dummy args for make_struct call. * struct.c (struct_from_plist, struct_from_args, make_struct_lit): Use ARGS_ABS_MIN instead of zero when preparing dummy args for make_struct_impl or make_struct call.
* poll: don't free array from alloca.Kaz Kylheku2021-09-081-2/+0
| | | | | | | * sysif.c (poll_wrap): April 23, 2020 commit 7fbf6b853893f65193ea9c81cf467be08c651244 left behind some stray free calls, which happen only in some error cases. Reported by Paul A. Patience, by way of GCC 11 testing.
* exceptions: hack to store errno in string object.Kaz Kylheku2021-09-0714-187/+365
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | Basic idea: when we throw an exception that pertains to a system error which has an errno code, we can stick the errno into the memory area of the character string, into the wchar_t that immediately follows the null terminator. We can do this because strings track their actual allocation size. A pair of setter/getter functions to set and retrieve this value are provided, and all functions in the code which can set such a code are updated to do so, simply by calling the newly added uw_ethrowf that drop-in replaces for uw_throwf. * lib.[ch] (string_set_code, string_get_code): New functions. * unwind.[ch] (uw_ethrowf): New function. * eval.c (eval_init): Register string-set-code and string-get-code intrinsics. * ftw.c (ftw_wrap): Switch to uw_ethrowf. * parser.c (open_txr_file): Likewise. * socket.c (dgram_overflow): Store the ENOBUFS error in errno, and use uw_ethrowf instead uw_throwf. (dgram_get_byte_callback, dgram_flush, sock_bind, to_connect, open_sockfd, sock_connect, sock_listen, sock_accept, sock_shutdown, sock_timeout, socketpair_wrap): Switch to uw_ethrowf. * stream.c (dev_null_get_fd, stdio_maybe_read_error, stdio_maybe_error, stdio_close, pipe_close, open_directory, open_file, open_fileno, open_tail, fds_subst, open_subprocess, open_command, remove_path, rename_path, tmpfile_wrap, mkdtemp_wrap, mkstemp_wrap): Switch to uw_ethrowf. * sysif.c (mkdir_wrap, ensure_dir, chdir_wrap, getcwd_wrap, rmdir_wrap, mknod_wrap, mkfifo_wrap, chmod_wrap, do_chown, symlink_wrap, link_wrap, readlink_wrap, close_wrap, val exec_wrap, stat_impl, do_utimes, pipe_wrap, poll_wrap, getgroups_wrap, setuid_wrap, seteuid_wrap, setgid_wrap, setegid_wrap, setgroups_wrap, getresuid_wrap, setresuid_wrap, setresgid_wrap, crypt_wrap, uname_wrap, opendir_wrap, getrlimit_wrap, setrlimit_wrap): Likewise. * termios.c (tcgetattr_wrap, tcsetattr_wrap, tcsendbreak_wrap, tcdrain_wrap, tcflush_wrap, tcflow_wrap): Likewise. * tests/018/errno.tl: New file. * txr.1: Documented. * stdlib/doc-syms.tl: Updated.
* unwind: remove unused uw_errorf.Kaz Kylheku2021-09-072-14/+0
| | | | * unwind.[ch] (uw_errorf): Function removed.
* string-extend: use num, not num_fast.Kaz Kylheku2021-09-071-3/+3
| | | | | | | | * lib.c (string_extend, string_finish): When we update the alloc value in the string, we should be using num, because the cnum value is not necessarily in the fixnum range. That's the whole reason we are using the set macro: because the assigned value could be a heap-allocated bignum. Which it will never be, if we use num_fast.
* string-finish: new function.Kaz Kylheku2021-09-075-0/+42
| | | | | | | | | | | | * eval.c (eval_init): Register string-finish intrinsic. * lib.c (string_finish): New function. * lib.h (string_finish): Declared. * txr.1: Documented. * stdlib/doc-syms.tl: Updated.
* string-extend: third optional argument.Kaz Kylheku2021-09-078-30/+49
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | A Boolean optional argument to string-extend indicates whether this is likely the last call to string-extend, so memory can be trimmed accordingly. * eval.c (eval_init): Update string-extend registration. * filter.c (trie_filter_string): Pass nil for new argument of string_extend. * lib.c (str_seq, replace_str, lazy_str_force, lazy_str_force_upto): Pass nil for new argument of string_extend. (rem_impl, remove_if, separate): Pass t for new argument of string_extend on last iteration, nil otherwise. (string_extend): Implement new third argument, defaulted to nil. Switch from chk_grow_vec to the more specific chk_wrealloc, which simplifies the code. * lib.h (string_extend): Declaration updated. * parser.y (litchars): Pass t as last argument of string_extend since we know syntactically that these reductions finalize the string. (restlitchar): Pass nil as the last argument of string_extend, since we know syntactically that it isn't the last. * regex.c (scan_until_common): Pass nil for new argument of string_extend. * txr.1: Documented.
* tests: no gc torture for tests/019.Kaz Kylheku2021-09-071-0/+1
| | | | | | * Makefile (tst/tests/019/%): Clear TXR_DBG_OPTS for this recently introduced directory, so txr isn't run with --gc-debug.
* ffi: ffi macro uses load-time.Kaz Kylheku2021-09-062-2/+2
| | | | | | | | | | | | | | | A number of functions take an argument which is a ffi type. Typically, this argument is produced using by a ffi-type-compile call which is produced by the ffi macro. But this ffi-type-compile call is invoked at run time, each time such a function is called. A solution for this is to have the ffi macro hoist the compilation to load time. * stdlib/ffi.tl (ffi): Add load-time wrapping to generated expression. * txr.1: Updated correspondence between (ffi ...) form and equivalent (ffi-type-compile form).
* sockets: improve socked type patch.Kaz Kylheku2021-09-061-10/+12
| | | | | | | | * socket.c (SOCK_NONBLOCK, SOCK_CLOEXEC): #define these as 0 if they are missing. (open_socket, socketpair_wrap): Clear the SOCK_NONBLOCK and SOCK_CLOEXEC flags in a single operation instead of two. Use C native arithmetic instead of Lisp logand and lognot.
* sockets: clear non-type bits after fd creation.Paul A. Patience2021-09-061-0/+13
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | The SOCK_NONBLOCK and SOCK_CLOEXEC non-type bits are a convenience allowed by the socket and socketpair functions, but they prevent simple comparisons with the actual socket types SOCK_STREAM, SOCK_DGRAM, etc. Before this commit, opening datagram sockets with any of these bits ORed into the type would result in TXR not realizing that such sockets were datagram sockets because the type would be compared directly with SOCK_DGRAM. Various problems would ensue, such as calling listen on a datagram socket. This commit clears the SOCK_NONBLOCK and SOCK_CLOEXEC bits after calling socket or socketpair so that the socket streams get the actual socket type. A potentially better solution would be to manually define SOCK_TYPE_MASK as 0xF (because although Linux defines it as such in include/linux/net.h, it doesn't export it to userspace) and AND it with the socket type, because clearing SOCK_NONBLOCK and SOCK_CLOEXEC will not be enough if new flags are added in the future, and it's probably more likely for that to happen than for SOCK_TYPE_MASK to be changed. * socket.c (open_socket, socketpair_wrap): Clear the SOCK_NONBLOCK and SOCK_CLOEXEC bits from the socket type after calling socket or socketpair, and before initializing the socket streams.
* sockets: make error messages more consistent.Paul A. Patience2021-09-061-17/+17
| | | | | | | | | | * socket.c (dgram_set_sock_peer): set-sock-peer -> sock-set-peer. (sock_bind, to_connect, sock_accept, socketpair_wrap): Use ~a and self in uv_throwf calls. (to_connect): Add colon after self in a uv_throwf call not reporting "failed". (sock_listen): Remove colon after self from uv_throwf call reporting "failed".
* rand: remove redundant inclusion of buf.h.Paul A. Patience2021-09-061-1/+0
| | | | * rand.c: Remove redundant inclusion of buf.h.
* doc: fix description of carray-uint type argument.Paul A. Patience2021-09-061-1/+1
| | | | | | | | The optional type argument of carray-uint and carray-int was documented as defaulting to uint, but it actually defaults to uchar. * txr.1: uint -> uchar.