2026-11-16         Arnold D. Robbins     <arnold@skeeve.com>

	* 5.4.2: Release tar ball made.

2026-11-15         Arnold D. Robbins     <arnold@skeeve.com>

	* README, NEWS: Updated before release.

2026-11-04         Arnold D. Robbins     <arnold@skeeve.com>

	* re.c (using_utf8): Check that nl_langinfo() doesn't return
	NULL. Backported from master.

	Unrelated.  Fix problems with indirect calls. Thanks to
	sunlv wang <wangsunlv@iscas.ac.cn> for the report.

	* interpret.h (r_interpret): Op_push, UPREF() and Node_var_new.
	* builtin.c (do_dcgettext, do_dcngettext): Check the number
	of arguments outside of the ifdefs for HAVE_NLS.

2026-11-01         Arnold D. Robbins     <arnold@skeeve.com>

	* main.c (copyleft): Use https in the license URL and enclose
	it in angle brackets. Thanks to Antonio Columbo.

2026-10-25         Paul Eggert  <eggert@cs.ucla.edu>

	* eval.c (calc_exp): Handle an exponent of LONG_MIN (-2^63).
	Thanks to Anton Khirnov <anton@khirnov.net> for the report.

2026-10-08         Arnold D. Robbins     <arnold@skeeve.com>

	* debug.c (CHECK_PROG_RUNNING): Don't use __FILE__ so that
	test output works on different systems. Mainly for z/OS.

2026-10-04         Arnold D. Robbins     <arnold@skeeve.com>

	* NEWS: Updated.

2026-10-01         Arnold D. Robbins     <arnold@skeeve.com>

	* builtin.c, node.c: Change spelling of `regex' to `regexp' in
	various messages. Thanks to Antonio Colombo for the report.

2026-09-29         Arnold D. Robbins     <arnold@skeeve.com>

	All of the following were reported by Daniel Koh Ji Yang
	<Daniel.Koh@mbzuai.ac.ae>. We thank him.

	* builtin.c (do_dcgettext, do_bindtextdomain): Rearrange the
	code slightly so that there are no references into freed
	memory.

	Unrelated:

	* builtin.c (call_sub): Make sure the regex isn't a plain
	regex constant, and that it compiles properly.
	(call_sub, call_match, call_split_func): Make sure the regex
	compiles properly.
	* awkgram.y (make_regnode): Check that exp->stptr isn't NULL.

	Unrelated:

	* interpret.h (Op_push): For a paramater that's a Node_elem_new,
	make sure valref is > 1 before derefing. See elemnew7.awk
	test case.

	Unrelated:

	* debug.c (do_commands): Check that arg is not NULL. See
	dbugbadcmd.awk test.
	(execute_commands): Use a local to walk the list in case the
	command deletes the breakpoint. See dbugbreak test case.
	(next_command): Check ret before checking if to delete a temporary
	breakpoint, as it might already have been deleted. Thanks to
	valgrind for the catch.

2026-09-29         Henner Zeller         <henner.zeller+gawk@gmail.com>

	* main.c (load_environ): Avoid writing into environment strings.
	Apparently some loaders add environment variables to read-only
	segments.

2026-09-24         Arnold D. Robbins     <arnold@skeeve.com>

	* AUTHORS: Add Mike Haertel.

2026-09-23         Arnold D. Robbins     <arnold@skeeve.com>

	* gawkapi.c (api_create_value): Allow AWK_BOOL also.

	Unrelated: Don't warn multiple times about the same
	unreferenced variable. Also, remove the warning about
	// looking like a C++ comment.  Thanks to Kpop 2GM
	<jason.cy.kwan@gmail.com> for the report.

	* awk.h [UNINIT_WARNED]: New flag.
	* awkgram.y (grammar): For REGEXP, remove warning about C++ comment.
	* eval.c (flags2str): Add an entry for UNINIT_WARNED.
	(r_get_lhs): Add check for flag being absent before issuing warning.
	Set the flag if it was clear.
	* interpret.h (r_interpret): Op_push* cases. Ditto.

	Unrelated:

	* re.c (using_utf8): Use nl_langinfo(CODESET) which is POSIX
	instead of the localeinfo struct, which isn't.

2026-09-20         Arnold D. Robbins     <arnold@skeeve.com>

	* bootstrap.sh: doc/notes.texi is now doc/gawknotes.texi. Ooops.
	Thanks to Daniel Richard G. for pointing out the problem.

2026-09-18         Arnold D. Robbins     <arnold@skeeve.com>

	* NEWS: Updated.

2026-09-14         Arnold D. Robbins     <arnold@skeeve.com>

	* awkgram.y: Adjust grammar to not print break/continue not allowed
	outside of a loop message twice.
	* builtin.c (do_exp): Set ERRNO if there's an error.

2026-09-07         Arnold D. Robbins     <arnold@skeeve.com>

	* configure.ac: Remove setting of ZOS_FAIL.  The test
	infrastructure now has *.ok-zos files to handle z/OS.

2026-09-02         Arnold D. Robbins     <arnold@skeeve.com>

	* NEWS: Updated.

2026-08-27         Arnold D. Robbins     <arnold@skeeve.com>

	* re.c (reflags2str): Add table and logic for MinRX flags.

2026-08-26         Arnold D. Robbins     <arnold@skeeve.com>

	* field.c (choose_fs_function): FS = "" doesn't work if do_posix
	but does (now) if do_traditional, since BWK awk supports it.

2026-08-24         Arnold D. Robbins     <arnold@skeeve.com>

	* NEWS: Updated.

2026-08-10         Arnold D. Robbins     <arnold@skeeve.com>

	* interpret.h (r_interpret): Op_subscript. If we just went
	through SYMTAB and got a Node_var_new, AND the next thing we're
	about to do is typeof() or isarray(), don't add a var_value
	and use it. Otherwise, as before. Thanks to Jason C. Kwan
	for the report.

2026-08-08         Arnold D. Robbins     <arnold@skeeve.com>

	* NEWS: Updated.

2026-07-28         Arnold D. Robbins     <arnold@skeeve.com>

	* builtin.c (do_dump_node): Add sizeof(NODE) to the output.

2026-07-23         Arnold D. Robbins     <arnold@skeeve.com>

	* configure.ac: Don't check anymore for st_blksize in struct stat.
	Just assume it's there.

2026-07-20         Arnold D. Robbins     <arnold@skeeve.com>

	* array.c (value_info): Don't use n->stfmt if it isn't set.
	Thanks to Andrew Schorr for the report.

2026-07-14         Arnold D. Robbins     <arnold@skeeve.com>

	* builtin.c (do_dump_node): Add the size of each object to
	the printout.

	Unrelated: Make things works when built on systems without
	the GMP and MPFR libraries.  Thanks to Thomas Trepl <ttrepl@yahoo.de>
	and Bruce Dubbs <bdubbs@linuxfromscratch.org> for the reports.

	* awk.h (struct exp_node): Add alignment padding when we don't
	have MPFR. This is a hack, pending a total refactoring of
	the NODE structure.

2026-07-08         Arnold D. Robbins     <arnold@skeeve.com>

	* 5.4.1: Release tar ball made.

2026-07-08         Arnold D. Robbins     <arnold@skeeve.com>

	* Checklist: Revisions.
	* configure.ac: Set version to 5.4.1.

2026-07-06         Arnold D. Robbins     <arnold@skeeve.com>

	* awk.h (struct exp_node): For non-MPFR, keep alignment the same,
	especially on 32-bit systems, by adding an extra int member.
	Without this, gcc -m32 without the MFPR libraries causes several
	failures in the test suite. First seen on z/OS.

2026-07-03         Arnold D. Robbins     <arnold@skeeve.com>

	* array.c, cint_array.c, cmd.h, floatcomp.c, floatmagic.h,
	gawkapi.c, int_array.c, msg.c, str_array.c: Update copyright year.

2026-07-02         Arnold D. Robbins     <arnold@skeeve.com>

	* array.c (assoc_list): Move unsorted to the top of the list
	of options so that it will be found first, since it's
	the usual case.
	* interpret.h (r_interpret): Don't do a dupnode of Nnull_string
	when it's about to be upref'ed. Happened in two places.

	Unrelated:

	* NEWS: Updated. No LLM code allowed.

2026-06-30         Arnold D. Robbins     <arnold@skeeve.com>

	* str_array.c (str_lookup): Check for USER_INPUT too when
	freezing an index into a string. Thanks to Kpop 2GM
	<jason.cy.kwan@gmail.com> for the report.

2026-06-29         Arnold D. Robbins     <arnold@skeeve.com>

	* printf.c (format_out_of_range): Handle inf/nan correctly
	even if do_posix. Thanks to Kpop 2GM <jason.cy.kwan@gmail.com>
	for the report.

2026-06-24         Arnold D. Robbins     <arnold@skeeve.com>

	Make string formatting smarter by tracking whether it was OFMT
	or CONVFMT that was last used to create a string value.
	Thanks to Stephane CHAZELAS <stephane@chazelas.org>
	for the report.

	* awk.h: New flags: CONVFMT_FMT and OFMT_FMT.
	(format_val): Function now takes an additional argument.
	(r_format_val): Function now takes an additional argument.
	(force_string_fmt): Function now takes an additional argument, the
	flag. Adjust logic to check which one was used.
	(force_string, force_string_ofmt): Adjust calls.
	* eval.c (flags2str): Add new flags to the array of flags.
	* mpfr.c (mpg_format_val): Function now takes an additional argument.
	Adjust the logic to set the flags.
	* node.c (format_val): Function now takes an additional argument.
	(r_format_val): Function now takes an additional argument.
	Adjust the logic to set the flags.

	Unrelated:

	* NEWS: Updated.

2026-06-24         Arnold D. Robbins     <arnold@skeeve.com>

	* builtin.c (do_print_rec): Use force_string_ofmt() when
	printing $0. Thanks to Stephane CHAZELAS <stephane@chazelas.org>
	for the report.

2026-06-22         Arnold D. Robbins     <arnold@skeeve.com>

	* field.c (get_field): Add special case handling when $0
	has been assigned a numeric value. See test/assignnumfield3.awk
	Thanks to Stephane CHAZELAS <stephane@chazelas.org> for the
	report and discussion. The full thread can be found starting at
	https://lists.gnu.org/archive/html/bug-gawk/2026-06/msg00001.html.

2026-06-19         Arnold D. Robbins     <arnold@skeeve.com>

	* eval.c (elem_new_to_scalar): Set the type to Node_var *before*
	unrefing the value.  See test case mdim9.awk. Thanks to Andrew
	Schorr for the report.

2026-06-15         Arnold D. Robbins     <arnold@skeeve.com>

	Thoroughly fix lint of plus on two strings with Op_elem_new.

	* awk.h (OPCODE): Remove Op_lint_plus.
	* awkgram.y (isnoeffect): Ditto.
	(mk_binary): Just do Op_plus; lint check moved to interpret.h.
	* eval.c (optypes): Remove Op_lint_plus.
	* interpret.h (r_interpret): Remove Op_lint_plus case. Add lint
	check to Op_plus case, rearranging code a bit.
	* mpfr.c (mpg_interpret):. Add lint check to Op_plus case.
	* profile.c (pprint): Remove Op_lint_plus.

2026-06-14         Arnold D. Robbins     <arnold@skeeve.com>

	* eval.c (fmt_ok): Don't complain if no width or precision. Thanks
	to Stephane Chazelas <stephane@chazelas.org> for the report.
	* interpret.h (r_interpret): For Op_lint_plus, call
	elem_new_to_scalar() on the values. Thanks to Hermann Peifer
	for the report.

2026-05-26         Arnold D. Robbins     <arnold@skeeve.com>

	* awkgram.y (mk_binary): Keep Doug McIlroy happy. Make division
	by zero a lint warning.  Applies to division and modulus.
	* NEWS: Updated.

2026-05-25         Arnold D. Robbins     <arnold@skeeve.com>

	* awkgram.y (mk_binary): Clean up code for when not optimizing.
	Motivated by a bug report from Douglas McIlroy
	<douglas.mcilroy@dartmouth.edu>.
	(GRAMMAR): Add an additional lint check for plain print in a
	BEGIN or END rule for when not optimizing. Thanks to the test
	suite and GAWK_TEST_ARGS=-s.

2026-05-11         Arnold D. Robbins     <arnold@skeeve.com>

	* builtin.c (do_exp): Issue warning for ERANGE only if the
	result is not zero. Thanks to joerg van den hoff
	<veedeehjay@gmail.com> for the report.

2026-05-07         Arnold D. Robbins     <arnold@skeeve.com>

	* eval.c (fmt_index): If OFMT or CONVFMT value is bad, always
	warn.  Using something like %sf for CONVFMT causes a core dump
	due to infinite recursion, but mawk and nawk also fail this way.
	It's a user error, we should not correct the value for them but
	instead have them fix their code.  Thanks to Michael D. Setzer II
	<mikes@guam.net> for the report.

2026-04-28         Arnold D. Robbins     <arnold@skeeve.com>

	* str_array.c (STR_CHAIN_MAX): Document the change in size
	back to 10 from 2025-02-16.

2026-04-25         Arnold D. Robbins     <arnold@skeeve.com>

	* NEWS: Updated.

2026-04-16         Arnold D. Robbins     <arnold@skeeve.com>

	* NEWS: Updated.
	* main.c (usage): Make the text more generic so that I don't
	have to keep changing test/badargs.ok upon every beta or
	real release.

2026-04-11         Arnold D. Robbins     <arnold@skeeve.com>

	* interpret.h (r_interpret): For Op_unary_{plus,minus}, force
	the top of the stack to be a number. This fixes issues with
	Op_elem_new.  See test/uplus2.awk. Thanks to
	Lewis Hyatt <lhyatt@gmail.com> for the report.
	* mpfr.c (mpg_interpret): Ditto.

2026-04-10         Arnold D. Robbins     <arnold@skeeve.com>

	* builtin.c (do_split): Refactor the logic a bit to make sure
	that single-character regexs as the third argument are handled
	correctly. Thanks to Ben Hoyt <benhoyt@gmail.com> for the
	report and test case. New test case is splitany.

2026-04-06         Arnold D. Robbins     <arnold@skeeve.com>

	* builtin.c (do_sub): Check for overflow in calculation of size
	of result buffer by doing the math in 64 bits. This provides
	a fatal message on 32 bit systems if overflow happens instead
	of letting dynamic memory get corrupted and likely causing
	a core dump. Thanks to Michał Majchrowicz <mmajchrowicz@afine.com>
	for the report and fix.

2026-04-05         Arnold D. Robbins     <arnold@skeeve.com>

	* NEWS: Updated.

2026-04-04         Arnold D. Robbins     <arnold@skeeve.com>

	* builtin.c (do_sub): Make `sofar' be size_t to avoid
	integer overflows.  Thanks to Michał Majchrowicz
	<mmajchrowicz@afine.com> for the report.
	* node.c (parse_escape): Change `i' to int64_t to avoid
	overflows. Thanks to ASan with gcc -m32.

2026-04-03         Arnold D. Robbins     <arnold@skeeve.com>

	* io.c (do_getline_redir): Don't DEREF redir_exp too early.
	Thanks to Michał Majchrowicz <mmajchrowicz@afine.com>
	for the report.

2026-03-30         Arnold D. Robbins     <arnold@skeeve.com>

	* builtin.c (do_sub): Add error checking on multibyte char matching.
	* interpret.h (r_interpret): Op_field_spec, use dupnode() instead
	of UPREF on the pointer.  Thanks to Marc Schönefeld
	<marc.schoenefeld@gmx.org> for the report and test case.

2026-03-13         Arnold D. Robbins     <arnold@skeeve.com>

	* NEWS: Updated.

2026-03-12         Arnold D. Robbins     <arnold@skeeve.com>

	All files updated, replacing the FSF address with a
	link to the licenses: https://www.gnu.org/licenses/.
	Thanks to Xose Vazquez Perez <xose.vazquez@gmail.com>
	for pointing out the need.

2026-03-11         Arnold D. Robbins     <arnold@skeeve.com>

	* NEWS: Updated.

2026-03-02         Arnold D. Robbins     <arnold@skeeve.com>

	* Checklist: Revisions.

2026-03-01         Weixie Cui		 <cuiweixie@gmail.com>

	* array.c (make_aname): Use sizeof(char) instead of
	sizeof(char *) when allocating memory for the aname
	string buffer. The buffer holds characters, not
	pointers, so the previous sizeof(char *) caused
	over-allocation.

2026-02-27         Arnold D. Robbins     <arnold@skeeve.com>

	* buitin.c (do_sub): Pass searchflags to research in
	loop over the whole line so that gensub()'s global works
	with \N construct. Thanks to Koichi Murase <myoga.murase@gmail.com>
	and Finn Steffens <finn.steffens@protonmail.com> for
	reporting the bug.

2026-02-25         Arnold D. Robbins     <arnold@skeeve.com>

	* builtin.c (do_dump_node): Add print of gawk_mb_cur_max.

2026-02-22         Arnold D. Robbins     <arnold@skeeve.com>

	* 5.4.0: Release tar ball made.

2026-02-22         Arnold D. Robbins     <arnold@skeeve.com>

	* README: Updated.

2026-02-15         Arnold D. Robbins     <arnold@skeeve.com>

	* NEWS: Updated.

2026-02-13         Arnold D. Robbins     <arnold@skeeve.com>

	* Checklist: Add more stuff.

2026-02-12         Arnold D. Robbins     <arnold@skeeve.com>

	* Checklist: Add more stuff.
	* Makefile.am, Makefile.in, awk.h, awkgram.y, command.y,
	debug.c, eval.c, ext.c, field.c, gawkapi.h, gawkmisc.c,
	interpret.h, io.c, mpfr.c, node.c, nonposix.h, printf.c,
	profile.c, replace.c, symbol.c: Update copyright year.

2026-02-10         Arnold D. Robbins     <arnold@skeeve.com>

	* NEWS: Updated.
	* main.c (UPDATE_YEAR): Update to 2026.

2026-02-09         Arnold D. Robbins     <arnold@skeeve.com>

	Bug fix for untyped elements used in redirections.
	Thanks to Ed Morton for the report.

	* builtin.c (do_print): Call elem_new_to_scalar() on the redirection.
	(do_getline_redir): Ditto.

2026-02-09         Arnold D. Robbins     <arnold@skeeve.com>

	* main.c (__dso_handle): Remove. PCC no longer needs it.
	* printf.c (isinf): Remove macros, PCC now handles it okay.

2026-02-05         Arnold D. Robbins     <arnold@skeeve.com>

	* main.c (__dso_handle): Define if compiling with PCC.
	* printf.c (isinf): For PCC undef it and define it as __builtin_isinf
	which PCC supports.

2026-02-04         Arnold D. Robbins     <arnold@skeeve.com>

	* configure.ac: Look for <sys/procctl.h> and <sys/sysctl.h>, and
	for procctl() and sysctl(), on FreeBSD.
	* NEWS: Updated.

2026-02-03         Arnold D. Robbins     <arnold@skeeve.com>

	* Makefile.am: For NetBSD 10, turn off ASLR via paxctl.
	* NEWS: Updated.

2026-02-02         Arnold D. Robbins     <arnold@skeeve.com>

	* NEWS: Updated.
	* re.c (regexflags2str): Remove declaration of function that
	doesn't exist.
	(resetup): Clean up the code a little.
	(minrxcompflags2str,  minrxexecflags2str): New functions.

2026-02-03         John E. Malmberg      <wb8tyw@qsl.net>

	* builtin.c (do_dump_node): Honor the HAVE_LONG_LONG_INT macro.

2026-01-28         John E. Malmberg      <wb8tyw@qsl.net>

	* ChangeLog: Spelling fixes
	* custom.h: OpenVMS Posix is long gone.

2026-01-27         Arnold D. Robbins     <arnold@skeeve.com>

	* NEWS: Updated.

2026-01-24         Arnold D. Robbins     <arnold@skeeve.com>

	* TODO: Updated.

2026-01-20         Arnold D. Robbins     <arnold@skeeve.com>

	* TODO: Updated.

2026-01-20         J.			<thirdedition@gmx.net>

	* configure.ac: Remove \ continuation in lists inside m4
	constructs. Needed for MSYS2.

2026-01-17         Arnold D. Robbins     <arnold@skeeve.com>

	* eval.c (posix_compare): Simplify further to always use
	strcoll() everywhere but MinGW.
	* NEWS: Updated. Only OpenBSD 7 works with PMA.

2026-01-16         Arnold D. Robbins     <arnold@skeeve.com>

	* printf.c (format_args): Revert fix of 2026-01-15.
	* eval.c (posix_compare): For Cygwin, always uses strcoll().

2026-01-15         Corinna Vinschen      <vinschen@redhat.com>

	UTF-8 and GB18030 allow 4 byte multibyte sequences which
	have to be converted to surrogate pairs on systems with
	wchar_t == UTF-16.  That only affects Cygwin for now, given
	AIX is ignored and native Windows is handled separately.

	This affects the printf() functions because the character and byte
	counting in mbc_char_count() and mbc_byte_count() can't handle
	characters yet which have to be converted to surrogate pairs.

	First of all, both functions do not expect the mbstate_t to be
	in a shift state after it read the first character.  However, this
	is essential to keep track of the full character during surrogate
	handling.  When the code restarts reading the string from the
	beginning for counting, it neglects to reset the shift state,
	so the next read encounters an invalid character sequence.  So make
	sure to reset the shift state when reading from the beginning.

	For Cygwin:

	Add code to mbc_char_count() to count a surrogate pair as a single
	character, not as two characters.

	Add code to mbc_byte_count() to handle only a full surrogate as a
	full character.

2026-01-15         Arnold D. Robbins     <arnold@skeeve.com>

	* printf.c (format_args): For %c, check for -2 return value
	from c32rtomb().

2026-01-14         Arnold D. Robbins     <arnold@skeeve.com>

	* builtin.c (do_dump_node): Change %zd to %zu.
	For MB_CUR_MAX, cast to int; helps on Cygwin.

2026-01-12         Arnold D. Robbins     <arnold@skeeve.com>

	* builtin.c (do_substr): Add casts to silence clang warnings.
	* node.c (r_format_val): Ditto.
	* main.c (load_procinfo): Ditto.
	(initfds): Change use of newfd to silence set but not used warning.

2026-01-09         Arnold D. Robbins     <arnold@skeeve.com>

	* mpfr.c (force_mpnum): Use bool instead of int for parameters.
	(mpg_force_number): Sync some with r_force_number. Accept
	non-decimal data if --posix, as in r_force_number. Thanks to
	Jason C. Kwan for the report.

2026-01-06         Arnold D. Robbins     <arnold@skeeve.com>

	* NEWS: Edited.

2026-01-04         Arnold D. Robbins     <arnold@skeeve.com>

	* bootstrap.sh: No need to touch doc/gawk.texi. Add touch
	of pc/Makefile.tst.

2026-01-04         Eli Zaretskii  <eliz@gnu.org>

	* NEWS: Updated with info about UTF-8 on Windows.

2025-12-30         Arnold D. Robbins     <arnold@skeeve.com>

	* NEWS: Updated.

2025-12-30         Arnold D. Robbins     <arnold@skeeve.com>

	* awk.h: Style edit: POSIX in all caps.
	* main.c: Style edits: No space before parens in function calls.
	* node.c (parse_escape): Move memset() of mbs to outside ifdefs.
	Factors out duplicate code.

2025-12-28         Eli Zaretskii  <eliz@gnu.org>

	* awk.h: Remove MinGW stuff that is no longer needed.
	* custom.h [__MINGW32__]: Add redirections for char32_t functions
	and their prototypes.  Override MB_CUR_MAX with a MinGW-specific
	function.  Add prototypes and redirections for stdio functions to
	work around a bug in MSVCRT.
	* gawkmisc.c [__MINGW32__]: Include pc/mbc32.c.
	* main.c (main) [__MINGW32__]: Adjust the value of gawk_mb_cur_max
	when the locale changes.
	* node.c: Remove MinGW-specific code, which is not longer needed.
	* nonposix.h: Add prototype of 'mingw_using_utf8'.
	* printf.c: Fix a typo: wcrtomb should not be called directly, but,
	if needed, via c32rtomb.
	* profile.c: Avoid compiler warnings about functions that are
	unused when SIGHUP and SIGUSR1 are not defined.
	* replace.c [__MINGW32__ && !_UCRT]: Include stdio-consolesafe.c
	to work around MSVCRT bugs when writing multibyte characters to
	the console.

2025-12-28         Arnold D. Robbins     <arnold@skeeve.com>

	* TODO: Updated.

2025-12-27         Arnold D. Robbins     <arnold@skeeve.com>

	* awkgram.y (yylex): Add support for hexadecimal floating point.
	* builtin.c (nondec2awknum): Improve hex floating point support.

2025-12-24         Arnold D. Robbins     <arnold@skeeve.com>

	* configure.ac: New option, --enable-O3 to use -O3 at compile
	time. This restores the default of -O2.
	* NEWS: Updated.

	Unrelated:

	* Checklist: Updated with more items.

2025-12-23         Eli Zaretskii  <eliz@gnu.org>

	* awk.h [_UCRT]: Use MinGW64-provided <uchar.h> when possible.
	(char32_t): Use the same typedef as in uchar.h.  Include
	<stdint.h> that's required for uint_least32_t.

2025-12-23         Arnold D. Robbins     <arnold@skeeve.com>

	* bootstrap.sh: Be smarter about touching the doc/*.info files.

2025-12-22         Arnold D. Robbins     <arnold@skeeve.com>

	* awk.h: Fixes for systems without uchar.h.

2025-12-22         Arnold D. Robbins     <arnold@skeeve.com>

	Switch to using the char32_t and associated routines.

	* configure.ac: Add checks for uchar.h, mbrtoc32(), c32rtomb().
	* awk.h: Add declarations and macros, based on configuration.
	(NODE, wstrstr, wcasestrstr): Switch to char32_t from wchar_t.
	* builtin.c: Ditto, use mbrtoc32(), across the board.
	* eval.c: Ditto, use c32slen(), c32scoll() in posix_compare().
	* interpret.h: Use char32_t.
	* node.c: Use char32_t and associated functions.
	* printf.c: Ditto.

2025-12-18         Arnold D. Robbins     <arnold@skeeve.com>

	* awk.h: Add a pragma for z/OS so that enum sizes are the same
	size as int. Prevents a core dump in the randtest test. Much
	thanks to Daniel Richard G. <skunk@iskunk.org> for the fix.

2025-12-17         Arnold D. Robbins     <arnold@skeeve.com>

	* builtin.c (do_match): If $0 was the regex, after unrefing
	tre->re_exp, set it to NULL so that it won't get unref'ed
	a second time. Thanks to valgrind.

	Unrelated:

	* node.c (parse_escape): Set type of `i' to long. Avoids
	undefined behavior warning from UBSan.

2025-12-16         Arnold D. Robbins     <arnold@skeeve.com>

	* awk.h, configure.ac, main.c, re.c, version.c: Minor edits
	before merging into master.

2025-12-15         Arnold D. Robbins     <arnold@skeeve.com>

	* builtin.c (do_dump_node): Add additional information, fix
	print param_cnt member.

	Unrelated:

	* configure.ac: Add strftime back to the list of functions
	we search for. It's needed on MinGW.
	* replace.c: Include missing_d/strftime.c if needed, along
	with MinGW stuff. Thanks to Eli Zaretskii <eliz@gnu.org>.

	Unrelated:

	* TODO: Updated.

2025-12-12         Eli Zaretskii  <eliz@gnu.org>

	* awk.h: Add declaration for timegm if HAVE_TIMEGM not defined.
	Needed for MinGW.
	* io.c (wait_any_block_signals): Add ifdefs as needed for MinGW.
	(file_can_timeout): Ditto.

2025-12-12         Arnold D. Robbins     <arnold@skeeve.com>

	* builtin.c (do_dump_node): Add an additional ifdef for MPFR.

2025-12-11         Arnold D. Robbins     <arnold@skeeve.com>

	* debug.c (CHECK_PROG_RUNNING): Add file, line number and function
	names to error message.
	(serialize_list): Move history related variables and code into
	#ifdefs so that things compile without readline.

2025-12-10         Arnold D. Robbins     <arnold@skeeve.com>

	Add a function to dump the NODE structure. Should have done
	this ages ago.

	* awk.h (do_dump_node): Declare the function.
	* builtin.c (do_dump_node): New function.
	* awkgram.y (tokentab): Add entry for _dump_node(). Note the "_".

2025-12-09         Arnold D. Robbins     <arnold@skeeve.com>

	* NEWS: Updated with info about MinRX.

2025-12-09         Arnold D. Robbins     <arnold@skeeve.com>

	* mpfr.c (mpg_mod): Keep only comments relevant to compatibility.
	And refer to the test.
	* builtin.c (do_intdiv): Fix typos in a comment.
	* mpfr.c (do_mpfr_intdiv): Ditto.
	Thanks to Nethox <nethox+awk@gmail.com> for the patch.

2025-12-06         Arnold D. Robbins     <arnold@skeeve.com>

	* awk.h: If using gettimeofday(), include <sys/time.h>.
	This is helpful on old macOS, at least.

2025-11-30         Arnold D. Robbins     <arnold@skeeve.com>

	Improve the timeout check to be made when the IOP is created
	instead of on every record read.

	* awk.h (IOBUF): Add a can_timeout member.
	* io.c (get_a_record): Use it instead of calling file_can_timeout().
	(file_can_timeout): Adjust to take fd and mode arguments.
	(iop_alloc): Set the can_timeout member by calling file_can_timeout.

2025-11-24         Arnold D. Robbins     <arnold@skeeve.com>

	* README: Updated some more.

2025-11-22         Arnold D. Robbins     <arnold@skeeve.com>

	Improve raw I/O speed by not checking for a timeout
	when there's no need.  Issue noticed during profiling
	of something unrelated.

	* io.c (file_can_timeout): New routine. Returns true if an
	input file is one that can timeout. Regular files cannot.
	(get_a_record): Use it.

2025-11-16         Arnold D. Robbins     <arnold@skeeve.com>

	* io.c (inetfile): Issue a warning if udp is used. It's
	now obsolete.
	* NEWS: Updated.

2025-11-04         Arnold D. Robbins     <arnold@skeeve.com>

	Continue the cleanup:

	* configure.ac: Don't look for system().
	* replace.c: Don't include missing_d/system.c.  It was for DJGPP
	which is not supported anymore.
	* custom.h: Remove the bit for QNX. Improve the declaration of
	strsignal().

2025-11-03         Arnold D. Robbins     <arnold@skeeve.com>

	Considerably simplify use of replacement functions. Many of the
	replacement functions are for routines that now widely standard,
	so we assume they're available.

	* Makefile.am (base_sources): Remove protos.h.
	* protos.h: Remove this file.
	* awk.h: Don't include protos.h. Don't do gawk_strtod() stuff.
	* configure.ac: Remove checks for a whole bunch of standard functions.
	Remove AC_FUNC_MKTIME check.
	* custom.h: Remove old checks for HP/UX and macOS.
	* replace.c: Remove checks and includes of files that are no
	longer needed.

2025-10-31         Arnold D. Robbins     <arnold@skeeve.com>

	After discussion with Nelson Beebe, allow hexadecimal
	floating point for strtonum() and --non-decimal-data.
	NOTE: NOT in program source code! (At least, not yet. Sigh.)

	* NEWS: Updated.
	* builtin.c (nondec2awknum): Revise logic.
	* mpfr.c (force_mpnum): Ditto.

2025-10-29         Arnold D. Robbins     <arnold@skeeve.com>

	* re.c (make_regexp): Remove assertion that string is zero
	terminated. It isn't always. Instead, save and restore the
	character past the value of `len'. This required making the
	first parameter not be const anymore. Thanks to Ben Hoyt
	<benhoyt@gmail.com> for the report.
	* awk.k (make_regexp): Adjust declaration.

2025-10-28         Arnold D. Robbins     <arnold@skeeve.com>

	* awkgram.y: Remove never-documented support for tawk extension
	`delete(a)' and lint warnings for regexp modifiers. Tawk is dead,
	it's time to bury the corpse.

2025-10-27         Arnold D. Robbins     <arnold@skeeve.com>

	* interpret.h (r_interpret): Op_push_arg_untyped: Don't issue
	lint warning for untyped variables, since isarray() and typeof()
	don't change the type of the parameter. Thanks to
	Greg Minshall <minshall@umich.edu> for the suggestion.

2025-10-25         Arnold D. Robbins     <arnold@skeeve.com>

	* field.c (do_split, do_patsplit): Use do_lint_extensions before
	warning about extensions.
	(set_FPAT): Clean up a lint warning.
	* io.c (set_RS): Ditto.

2025-10-23         Arnold D. Robbins     <arnold@skeeve.com>

	* field.c (do_split): If someone passed a Node_elem_new
	or subarray as the second argument to split(), issue a
	lint warning. Thanks to Hermann Peifer for the suggestion.
	(do_patsplit): Add the same warning.

2025-10-22         Arnold D. Robbins     <arnold@skeeve.com>

	* interpret.h (r_interpret): For Op_sub_array, don't warn
	about using an uninitialized element, as there is no
	way to create the parent of a subarray first. Thanks to
	Greg Minshall <minshall@umich.edu> for the report.

	Unrelated:

	* Checklist: Add an item to run the test suite with MPFR.

2025-10-19         Arnold D. Robbins     <arnold@skeeve.com>

	* configure.ac: Don't disable assertions anymore. They provide
	value, and on modern systems don't affect performance.
	* NEWS: Updated.

2025-10-19         Arnold D. Robbins     <arnold@skeeve.com>

	* re.c (re_update): Revert previous fix, it didn't do the trick.
	* builtin.c (do_match): If someone passed $0 as the second
	argument, deref it.  It's a special case, but it keeps the
	assertions happy.

	Thanks to Eli Zaretskii <eliz@gnu.org> for the report.

2025-10-17         Arnold D. Robbins     <arnold@skeeve.com>

	Fix a bug when match() is called with $0 as the second argument.
	The bug only showed up with assertions enabled.  At the same time,
	add a warning if a regexp constant is used as the first argument.
	Thanks to Daniel David Goren <danieldavidgoren@gmail.com> for
	the report.

	* awkgram.y (snode): For do_match, add a warning if the first
	argument is a regexp constant.
	* re.c (re_update): Unref the original text of the regexp
	if the copy is different from it.

	Unrelated:

	* mpfr.c (mpg_mod): Use mpz_tdiv_r() instead of mpz_tdiv_qr()
	to simplify the code. Suggested by Jason C. Kwan on the
	help-gawk list.

2025-10-04         Arnold D. Robbins     <arnold@skeeve.com>

	* printf.c (format_args): Fix call to fatal() to not have "fatal:"
	in it.

2025-09-27         Arnold D. Robbins     <arnold@skeeve.com>

	Fix configure --disable-extensions. Thanks to
	Frank Busse <f.busse@imperial.ac.uk> for the report.

	* configure.ac: When modifying the Makefile for different CFLAGS,
	use test -f to make sure that one is there.
	* ext.c (init_extensions_list): Add dummy version for when extensions
	are disabled.
	(load_ext): Update the function argument list for the dummy
	version to match the real one and the declaration in awk.h.
	* Checklist: Add an item to check `configure --disable-extensions'
	so that this won't happen again.

2025-09-01         Arnold D. Robbins     <arnold@skeeve.com>

	* NEWS: Fix spacing / indentation.
	* builtin.c (do_typeof): Memory leak fix for Node_elem_new. This
	showed up with valgrind in debug build on the test/elemnew4.awk
	file.
	* builtin.c: Move commented out main() test program for add_thousands()
	to ...
	* printf.c: ... here from builtin.c.

2025-07-30         Andrew J. Schorr      <aschorr@telemetry-investments.com>

	Add #ifdef'ed out patch to io.c to use a hashtable to store pipe
	pids to fix a problem with a docker container that defaulted to
	a ulimit of a billion open fds.
	* io.c (write_pipe): If hashing, add a next field.
	(save_pipe): New function to add a pipe to the hash table.
	(gawk_popen_write): If hashing, allocate index of PIPE_HASH_SIZE
	instead of getdtablesize(), and use save_pipe to store the pipe info
	instead of sticking it directly into the open_pipes array.
	(gawk_popen_write_close): If hashing, search the bucket list
	for a matching process. If the match is not at the head of the
	linked list, remove it from the list and free it.

2025-07-29         Arnold D. Robbins     <arnold@skeeve.com>

	Make gawk's system() function smarter. It has to ignore SIGINT
	and SIGQUIT and block SIGCHLD (per POSIX for the C system()
	function.) Thanks to Finn Magnusson <finnmag@yahoo.com> for
	the report.  The test case was:

		gawk 'BEGIN { system("cat"); print "\n1" }'

	Then immediately interrupt the "cat" with a Control-C.
	Older versions of gawk that used the C library system() would
	still print the "1".  Gawk 5.3.1 would not. These changes
	restore the old behavior.

	Note that different versions of awk do different things
	in this test case. BWK awk does print the "1", mawk does not.

	* io.c (wait_any): Now has just the code to do the wait and
	update the status.
	* io.c (wait_any_block_signals): This was originally wait_any().
	Simplified it by having just the signal handling here and making
	a call to wait_any().
	* builtin.c (gawk_system): Added signal handling code to
	ignore/block signals in the parent before the fork, restore
	things in the child before the exec, and restore them in
	the parent after calling wait_any().

2025-07-29         Arnold D. Robbins     <arnold@skeeve.com>

	* awk.h (emalloc_real, ezalloc_real, erealloc_real): Use %zu
	in the error message instead of %ld and casting to long.

2025-07-28         Arnold D. Robbins     <arnold@skeeve.com>

	Rationalize do_traditional throughout the code base:

	1. Bring things into sync with current BWK awk.
	2. Change `do_posix || do_traditional' tests to just `do_traditional'
	since --posix automatically also turns on do_traditional.

	* builtin.c (do_system): Remove old code that divided the
	exit status by 256 if do_traditional. BWK awk no longer
	does that.
	* ext.c (load_ext): Simplify check for do_traditional.
	* field.c (do_split): Ditto.
	* main.c (is_std_var): Ditto.
	* re.c (make_regexp): Ditto.
	* symbol.c (lookup): Ditto.
	* io.c (set_RS): Change check from do_traditional to do_posix
	in case where RS->stlen > 1.

2025-07-24         Andrew J. Schorr      <aschorr@telemetry-investments.com>

	* io.c (redirect): Don't issue a lint complaint if the
	redirect string is a STRNUM, because a filename could be
	numeric, and that shouldn't be flagged by lint if it came
[--snip--]
