Paul Eggert [Tue, 30 Aug 2016 15:23:44 +0000 (08:23 -0700)]
strverscmp: sync with glibc
Although this doesn't exactly synchronize with glibc
byte-for-byte, it makes the code behave the same as glibc.
* lib/strverscmp.c (S_I, S_F, S_Z): Now masks, not powers of 2.
(ISDIGIT): Remove, as glibc is sticking with isdigit, and the
difference shouldn't matter in practical use. All uses changed
back to isdigit.
(__strverscmp, strverscmp): Use new glibc method for weak aliases.
(next_state): Now unsigned char array; redo elements.
(result_type): Now signed char array; redo elements.
(__strverscmp): Fix glibc bug 9913 by using new states.
* tests/test-strverscmp.c (main): Test glibc bug 9913.
Paul Eggert [Mon, 29 Aug 2016 19:03:51 +0000 (12:03 -0700)]
intprops.h: port recent changes to GCC 6.2.0
* lib/intprops.h (__has_builtin): Move earlier.
(_GL_HAS_BUILTIN_OVERFLOW): Rename from
_GL_HAS_BUILTIN_OVERFLOW_WITH_NULL and don't worry about whether
the last argument can be null. All uses changed.
(_GL_HAS_BUILTIN_OVERFLOW_P): Also test __has_builtin.
(_GL_ADD_OVERFLOW, _GL_SUBTRACT_OVERFLOW, _GL_MULTIPLY_OVERFLOW):
Don't try to use 3rd arg null, as this doesn't work on GCC 6.2.0
and it's not clear which GCC versions it works for.
(_GL_INT_OP_WRAPV): Use _GL_HAS_BUILTIN_OVERFLOW instead of
its definiens.
Paul Eggert [Mon, 29 Aug 2016 17:08:32 +0000 (10:08 -0700)]
intprops.h: use __typeof__ with GCC 7
* lib/intprops.h (_GL_ADD_OVERFLOW, _GL_SUBTRACT_OVERFLOW)
(_GL_MULTIPLY_OVERFLOW): Use __typeof__ as in the GCC manual.
This avoids computing the expression's value (which might overflow!).
Jim Meyering [Mon, 29 Aug 2016 16:45:18 +0000 (09:45 -0700)]
intprops.h, xalloc-oversized.h: work with gcc 7
In gcc 6, __builtin_add_overflow, __builtin_sub_overflow and
__builtin_mul_overflow each accept a NULL pointer as the third
argument. However in gcc 7, that is no longer accepted.
Instead, one must use the "_p"-suffixed names, with which, the
third parameter is no longer a pointer.
* lib/intprops.h (_GL_HAS_BUILTIN_OVERFLOW_WITH_NULL): Correct
the definition: not true for gcc 7 and subsequent.
(_GL_HAS_BUILTIN_OVERFLOW_P): Define.
(_GL_ADD_OVERFLOW, _GL_SUBTRACT_OVERFLOW, _GL_MULTIPLY_OVERFLOW):
Provide new definitions for gcc 7 and subsequent.
* lib/xalloc-oversized.h (xalloc_oversized): Provide a definition
that works with gcc-7.
Paul Eggert [Thu, 25 Aug 2016 03:48:32 +0000 (20:48 -0700)]
intprops: port to OpenVMS
Problem reported by John E. Malmberg in: https://bugs.gnu.org/24300
* doc/posix-headers/limits.texi: Document the problem.
* lib/intprops.h (LLONG_MAX, LLONG_MIN) [__INT64_MAX]:
Define if not already defined.
Assaf Gordon [Mon, 15 Aug 2016 01:29:50 +0000 (21:29 -0400)]
parse-datetime: improve debug implementation
Follow-up to commit 12ad79069 ("add optional debug printing").
Improve parse-datetime's debug implementation: remove macros,
replace global debug flag variable with a function parameter,
use nstrftime for formatting.
See: https://lists.gnu.org/archive/html/bug-gnulib/2016-08/msg00021.html
* lib/parse-datetime.h: (parse_datetime_debug): Remove global extern
declaration.
(parse_datetime2): New function, accepts 'flags' parameter, supporting
debug flag. Existing interface 'parse_datetime' left unmodified.
* lib/parse-datetime.c: (parse_datetime_debug): Remove global variable.
(struct parser_control): add 'parse_datetime_debug' member variable.
(parse_datetime): Call new function 'parse_datetime2' without debug.
(parse_datetime2): Adapted from previous 'parse_datetime', initialize
pc.parse_datetime_debug variable as needed.
(to_year): Accept new flags parameter, instead of using global variable.
(debug_print_current_time,debug_print_relative_time,debug_mktime_not_ok):
use struct 'debug' variable instead of global variable.
(DEBUG,DEBUG_PRINT_CURRENT_TIME,DEBUG_PRINT_RELATIVE_TIME,
DEBUG_MKTIME_NOT_OK,PROGRESS,PROGRESS0): Remove macros. Call
correspnding functions directly instead of using macros.
* modules/parse-datetime: Add gnulib's strftime module.
Paul Eggert [Thu, 18 Aug 2016 18:56:07 +0000 (11:56 -0700)]
canonicalize-lgpl: fix errno after malloca fails
This fixes a typo I recently introduced. Suggested by Bruno Haible in:
http://lists.gnu.org/archive/html/bug-gnulib/2016-08/msg00039.html
* lib/canonicalize-lgpl.c (__realpath):
Don't assume malloca sets errno on failure.
Daniel Richard G [Wed, 17 Aug 2016 23:57:35 +0000 (16:57 -0700)]
maint: port tests to z/OS errno behavior
* tests/test-nonblocking-reader.h:
* tests/test-nonblocking-writer.h:
Accommodate z/OS errno code preferences. (I believe this should
still be within spec; IBM is good at following the letter if not
the spirit of such things.)
Daniel Richard G [Wed, 17 Aug 2016 23:09:38 +0000 (16:09 -0700)]
fclose, strstr-simple, wchar: port to z/OS
* m4/fclose.m4, m4/strstr.m4, m4/wchar_h.m4:
Changes to the Autoconf M4 code to support z/OS. Note that
fclose() is broken in a different way on z/OS than it is on other
systems, thus the special-case in fclose.m4.
Daniel Richard G [Wed, 17 Aug 2016 23:05:02 +0000 (16:05 -0700)]
iconv_open-utf-tests, iconv-tests: port to EBCDIC
* tests/test-iconv-utf.c, tests/test-iconv.c:
Added appropriately conditional #pragmas so that the test strings
in test-iconv-utf.c are correctly interpreted in ASCII instead of
EBCDIC (i.e. 'J' == 0x4A and not 0xD1). This issue could be
addressed in a more portable way by simply rewriting all the ASCII
literal characters as octal escapes, but then you would lose the
partial readability that the strings have now. Also, iconv_open()
on z/OS does not recognize "ISO-8859-1", but "ISO8859-1" works.
Paul Eggert [Wed, 17 Aug 2016 22:33:06 +0000 (15:33 -0700)]
stdbool: don't require _Bool for C++
Problem reported by David Seifert in:
http://lists.gnu.org/archive/html/bug-gnulib/2016-06/msg00005.html
* NEWS, doc/posix-headers/stdbool.texi (stdbool.h): Document this.
* m4/stdbool.m4 (AC_CHECK_HEADER_STDBOOL): Make the check
more-forgiving for C++, in that it requires only 'bool'. Be a bit
stricter about checking that bool and _Bool are compatible in C.
Paul Eggert [Wed, 17 Aug 2016 22:01:50 +0000 (15:01 -0700)]
getdelim: remove dependency on realloc-posix
* lib/canonicalize-lgpl.c (alloc_failed)
[!FUNC_REALPATH_WORKS || defined _LIBC]: New function,
(__realpath) [!FUNC_REALPATH_WORKS || defined _LIBC]: Use it.
Use __set_errno where needed, for consistency.
* lib/getdelim.c (alloc_failed): New function.
(getdelim): Use it.
* modules/getdelim (Depends-on): Remove realloc-posix.
Assaf Gordon [Sun, 7 Aug 2016 00:08:32 +0000 (20:08 -0400)]
parse-datetime: add optional debug printing
Print parsing information, warnings, and errors to stderr.
* lib/parse-datetime.h (parse_datetimte_debug): New global variable.
* lib/parse-datetime.y:
(DEBUG_*): Macros calling debug functions if debugging is enabled.
(PROGRESS*): Same as DEBUG, for progress reporting.
(dbg_printf): Print message to stderr, with 'date' prefix.
(struct parser_control): Add 'debug_*_seen' variables.
(str_days): Converts day ordinal/number to string (e.g. 'last wed').
(debug_print_current_time, debug_print_relateive_time): Prints the
current/relative date/time value of parser_control.
(YACC parser syntax): Print parsed parts with DEBUG_* macros.
(to_year): Warn about 2-digit year parsing.
(yylex): Warn about unrecognized words.
(get_effective_timezone): Returns current timezone in minutes.
(debug_strf{time,date,datetime}): Convert 'struct tm' to string as
clearly and unambigiously as possible.
(debug_mktime_not_ok): Print detailed information about failed
date/time values.
(parse_datetime): Add DEBUG messages for failures, warnings. Add
PROGRESS messages for status messages.
* modules/parse-datetime: Add 'timegm', 'gettext-h' dependencies.
Jim Meyering [Mon, 1 Aug 2016 16:37:15 +0000 (09:37 -0700)]
tests/init.sh: exclude dash with bad "local" semantics
* tests/init.sh (gl_shell_test_script_): Add a function to
eliminate a shell like "dash" (unlike bash, zsh) that has
surprising/risky "local var='...'" semantics. Inspired by
the problem and discussion in https://bugs.gnu.org/24116#11.
Ján Tomko [Tue, 2 Aug 2016 13:41:13 +0000 (15:41 +0200)]
maint.mk: expand the prohibit_doubled_word regex
This check has a static list of words that are checked for
repetitions. Expand it before running the perl script to
avoid using expensive captures. This decreases the cost
for libvirt from 1.66s to 0.66s.
* top/maint.mk (prohibit_doubled_word_expanded_): Define.
(sc_prohibit_doubled_word): Use it.
Ján Tomko [Mon, 1 Aug 2016 17:54:47 +0000 (10:54 -0700)]
useless-if-before-free: skip non-matching lines early
* build-aux/useless-if-before-free: First match each line with the
simple/quick /\bif\b/ and reject if there is no match. This often
saves the cost of the much more involved regular expression.
For libvirt, this decreases the cost from 1.44s to 1.02s.
Ján Tomko [Sat, 30 Jul 2016 05:39:31 +0000 (22:39 -0700)]
maint.mk: speed up sc_po_check
sc_po_check would skip files based on their names, or on the
existence of files with derived names. Rewrite it to use perl
instead of shell to make the check faster.
* top/maint.mk (perl_translatable_files_list_): Define.
(sc_po_check): Use it.
Ján Tomko [Tue, 26 Jul 2016 14:28:16 +0000 (16:28 +0200)]
maint.mk: speed up require_config_h_first
Instead of spawning three processes per file,
rewrite the check in perl and run it once for all the files.
* top/maint.mk (perl_config_h_first_): Define.
(sc_require_config_h_first): Use it in place of shell code.
Ján Tomko [Sat, 30 Jul 2016 05:39:31 +0000 (22:39 -0700)]
maint.mk: speed up sc_po_check
sc_po_check would skip files based on their names, or on the
existence of files with derived names. Rewrite it to use perl
instead of shell to make the check faster.
* top/maint.mk (perl_translatable_files_list_): Define.
(sc_po_check): Use it.
Paul Eggert [Fri, 15 Jul 2016 08:35:22 +0000 (10:35 +0200)]
obstack: pacify GCC 6 with -Wnull-dereference
Problem reported by Assaf Gordon in:
http://lists.gnu.org/archive/html/bug-gnulib/2016-07/msg00028.html
* lib/obstack.c, lib/obstack.h (obstack_alloc_failed_handler):
Declare with __attribute_noreturn__.
* lib/obstack.h (__attribute_noreturn__): New macro.
Commit 54615b95ff238e235e806855efc46a9abad09f2e changed the regular
expression for detecting C symbol prefixes but forgot to qoute square
brackets in the command line arguments for grep. That way when
building with mingw the condition was false although it ought to be
true instead. In particular scenarios this led to the following
compile error:
Cannot export rpl_printf: symbol not found
Cannot export rpl_scanf: symbol not found
collect2: error: ld returned 1 exit status
Pádraig Brady [Wed, 22 Jun 2016 12:49:53 +0000 (13:49 +0100)]
fts: handle readdir() errors
* lib/fts.c (fts_build): readdir(3) returns NULL when finished,
but also upon error when it will also set errno. Therefore
flag the error case from readdir(). We treat the case where
no items are read the same as if the dir can't be accessed,
i.e. by setting fts_errno to FTS_DNR.
The bug was initially reported by Peter Benie
http://bugzilla.opensuse.org/show_bug.cgi?id=984910
where it was mentioned that readdir() may fail
when an NFS server has a poor readdir cookie implementation.
Paul Eggert [Fri, 24 Jun 2016 11:35:12 +0000 (13:35 +0200)]
intprops: port better to GCC 7
GCC 7 __builtin_add_overflow supports a new usage form, where the
last argument is a null pointer, and which merely returns 1 if an
overflow would occur. This is a constant expression if all
arguments are constants, and should generate faster code when code
needs to be generated.
* lib/intprops.h (_GL_HAS_BUILTIN_OVERFLOW_WITH_NULL): New macro.
(_GL_ADD_OVERFLOW, _GL_SUBTRACT_OVERFLOW, _GL_MULTIPLY_OVERFLOW):
Use builtin operations if available.
(INT_NEGATE_OVERFLOW): Prefer INT_SUBTRACT_OVERFLOW if builtin
operations are available, as it's almost surely faster.
Paul Eggert [Mon, 13 Jun 2016 18:17:03 +0000 (11:17 -0700)]
xalloc-oversized: port to GCC 7; fewer warnings
GCC 7 will have a better way to deal with integer overflow.
Plus, fix a warnings problem reported by Tim Ruehsen in:
http://lists.gnu.org/archive/html/bug-gnulib/2016-06/msg00022.html
* lib/xalloc-oversized.h (__xalloc_oversized): New macro.
(xalloc_oversized): Use plain __builtin_mul_overflow if GCC 7 or later.
For GCC 5, use __xalloc_oversized if both args are constants,
or if pedantic.
Paul Eggert [Wed, 8 Jun 2016 08:46:35 +0000 (01:46 -0700)]
regex: port to Sun C
Reported by Daiki Ueno.
* lib/regcomp.c (regcomp, regerror): Use _Restrict_, not
__restrict, in prototype. This fixes a problem I introduced in
the 2016-02-19 merge from glibc.
Bruno Haible [Sun, 29 May 2016 10:54:32 +0000 (12:54 +0200)]
secure_getenv: Port to many more platforms.
* m4/secure_getenv.m4 (gl_PREREQ_SECURE_GETENV): Also check for get*id
functions.
* lib/secure_getenv.c (secure_getenv): Add alternate implementations
for non-BSD Unix platforms and for native Windows.
* doc/glibc-functions/secure_getenv.texi: Remove known issue.
Prompted by a request from Nikos Mavrogiannopoulos.
Eric Blake [Thu, 26 May 2016 14:47:27 +0000 (08:47 -0600)]
unsetenv: relax to LGPLv2+
Especially since setenv is also LGPLv2+, and it makes sense for the
two modules to stay in sync, rather than one to be usable only in
LPGLv3+ projects.
* modules/unsetenv (License): Match setenv license.
Paul Eggert [Tue, 17 May 2016 16:08:02 +0000 (09:08 -0700)]
manywarnings: update for GCC 6.1
* m4/manywarnings.m4 (gl_MANYWARN_ALL_GCC):
Add GCC 6.1 options that apply to C.
* build-aux/gcc-warning.spec: Add GCC 6.1 options that
do not apply to C, are obsolescent, etc.
Paul Eggert [Fri, 13 May 2016 06:08:22 +0000 (23:08 -0700)]
glob: size_t overflow checks
* lib/glob.c (__has_builtin): New macro.
(size_add_wrapv, glob_use_alloca): New static functions.
(glob, glob_in_dir): Check for size_t overflow in several places,
and fix some size_t checks that were not quite right.
Bruno Haible [Mon, 9 May 2016 07:29:35 +0000 (09:29 +0200)]
Fix undefined behaviour in gettext.h.
* lib/gettext.h (dcpgettext_expr, dcnpgettext_expr): Avoid accessing a
pointer's value after the storage it points to has been freed.
Reported by Michael Pyne in https://savannah.gnu.org/bugs/?47847.
Spotted by Coverity.
Paul Eggert [Sun, 8 May 2016 18:38:22 +0000 (11:38 -0700)]
git-version-gen: avoid undefined shift
Problem reported by Mosè Giordano in:
http://lists.gnu.org/archive/html/bug-gnulib/2016-05/msg00012.html
* build-aux/git-version-gen: Avoid undefined behavior if invoked
with --prefix or --fallback but without a later argument. While
we're at it, omit unnecessary quotes.
Paul Eggert [Wed, 4 May 2016 16:57:48 +0000 (09:57 -0700)]
glob: merge glibc changes into lib/glob.c
* lib/glob.c (glob_in_dir): Sync with glibc/posix/glob.c,
dated 2016-05-04 12:09:35 2016 +0200. Here are the changes:
2016-05-04 CVE-2016-1234: glob: Do not copy d_name field of
struct dirent [BZ #19779]
2016-04-29 glob: Simplify the interface for the GLOB_ALTDIRFUNC
callback gl_readdir
2015-10-20 Convert miscellaneous function definitions to prototype style
2015-10-20 Convert 113 more function definitions to prototype style
(files with assertions)
2015-06-12 Fix getlogin_r namespace (bug 18527).
2014-02-10 Use glibc_likely instead __builtin_expect.
2013-10-20 When glob pattern contains a trailing slash match only
directories. Fixes bug 10278.
2013-09-04 glob: silence -Wattribute warnings
2013-06-07 Avoid use of "register" as optimization hint.
2012-09-25 Use size_t instead of int for internal variables in glob
(bug 14621)
2011-07-20 Check for overflows in expressions
2011-05-28 Remove unused variable
2011-05-22 Add a few more alloca size checks
2010-03-27 Whitespace fixes
2010-03-27 Fix one more issue with the glob patch
2010-03-24 Fix glob with empty pattern
2008-05-27 Remove useless more "if" tests before "free"
* modules/glob (Depends-on): Add stdint.
Paul Eggert [Mon, 2 May 2016 00:25:16 +0000 (17:25 -0700)]
mktime: port to stricter signed overflow checking
* lib/mktime.c: Omit 'pragma GCC optimize ("wrapv")'.
(long_int): Require width for INT_MAX * 3 * (seconds per year),
instead of merely for INT_MAX * 2. In practice platforms that
do the latter also do the former.
(TIME_T_MIN, TIME_T_MAX, TIME_T_MIDPOINT, SHR): Remove.
(shr): New static function, replacing SHR. All uses changed.
(mktime_min, mktime_max): New constants, replacing TIME_T_MIN
and TIME_T_MAX. All uses changed.
(ydhms_diff, guess_time_tm, ranged_convert, __mktime_internal):
Use long_int, not time_t.
(long_int_avg): New static function, replacing time_t_avg.
All uses changed. Round toward positive infinity, as that
generates slightly better code.
(time_t_add_ok, time_t_int_add_ok): Remove. All uses replaced
by INT_ADD_WRAPV.
(guess_time_tm): Accept time, not a pointer to it. All uses changed.
(convert_time): New static function.
(ranged_convert): Use it
(ranged_convert): Check for *T out of [mktime_min, mktime_max] range.
Use simpler test for loop exit.
(__mktime_internal): Store negative of guessed offset, to simplify
overflow checking. Remove no-longer-needed test for small time_t
overflows.
Paul Eggert [Sun, 1 May 2016 19:49:21 +0000 (12:49 -0700)]
mktime: speed up DEBUG_MKTIME benchmarks
Call tzset just once, at the start, rather than for every test
case. This lets us measure the CPU cost of mktime as opposed to
that of tzset. This is relevant when TZ is not set and glibc is
being used. This speeds up tests by a factor of 40 on my Fedora
23 x86-64 platform.
* lib/mktime.c (main) [DEBUG_MKTIME]: Call localtime at the start,
to call tzset and as a sanity check. Later on, use localtime_r
instead of localtime.
Paul Eggert [Sun, 1 May 2016 18:44:51 +0000 (11:44 -0700)]
Port mktime_internal offset to unsigned time_t
This avoids some assumptions about wraparound arithmetic on
signed integer overflow.
* lib/mktime-internal.h (mktime_offset_t): New type.
(mktime_internal): Use it in decl.
* lib/mktime.c, lib/timegm.c (mktime_offset_t) [_LIBC]: New type.
* lib/mktime.c (__mktime_internal, localtime_offset):
* lib/timegm.c (timegm): Use it.
* m4/mktime.m4 (gl_TIME_T_IS_SIGNED): New macro.
(gl_FUNC_MKTIME): Require it.
Paul Eggert [Wed, 27 Apr 2016 19:10:54 +0000 (12:10 -0700)]
xstrtol: prohibit monstrosities like "1bB"
Problem reported by Young Mo Kang in: http://bugs.gnu.org/23388
* lib/xstrtol.c (__xstrtol): Allow trailing second suffixes like
"B" only if the first suffix needs a base.
* tests/test-xstrtol.sh: Test this.
Paul Eggert [Wed, 13 Apr 2016 17:53:56 +0000 (10:53 -0700)]
mktime: improve integer overflow checking
* lib/mktime.c: Include stdbool.h, intprops.h, verify.h.
(WRAPV): Remove; no longer needed.
(verify): Remove. Replace all uses with call to verify.h 'verify'.
(TYPE_IS_INTEGER, TYPE_SIGNED, TYPE_MINIMUM, TYPE_MAXIMUM):
Remove. Use intprops.h defns instead.
(leapyear, isdst_differ, time_t_add_ok, time_t_int_ok):
Use bool for Boolean, for clarity.
(time_t_add_ok, time_t_int_add_ok): Use INT_ADD_WRAPV to
detect integer overflow.
* modules/mktime (Depends-on): Add intprops, stdbool, verify.
Paul Eggert [Wed, 13 Apr 2016 17:49:08 +0000 (10:49 -0700)]
intprops: check two's complement assumption
Suggested by Eric Blake in:
http://lists.gnu.org/archive/html/bug-gnulib/2016-04/msg00016.html
* lib/intprops.h: Include <verify.h>. Verify that signed char,
short, int, long, and (if available) long long are two's complement.
* modules/intprops (Depends-on): Add 'verify'.
Paul Eggert [Wed, 13 Apr 2016 07:06:36 +0000 (00:06 -0700)]
intprops, mktime, strtol: assume two's complement
These macros were not portable to every conforming C11 ones'
complement platform. It's not worth the hassle of porting to some
platforms that use ones' complement or signed magnitude, as such
platforms are almost purely theoretical nowadays and porting even
to some of them makes the code harder to review for little
practical benefit. Problem reported by Florian Weimer in:
https://sourceware.org/ml/libc-alpha/2016-04/msg00295.html
* lib/intprops.h (TYPE_TWOS_COMPLEMENT, TYPE_ONES_COMPLEMENT)
(TYPE_SIGNED_MAGNITUDE, _GL_INT_TWOS_COMPLEMENT):
* lib/mktime.c (TYPE_TWOS_COMPLEMENT):
* lib/strtol.c (TYPE_TWOS_COMPLEMENT, TYPE_ONES_COMPLEMENT)
(TYPE_SIGNED_MAGNITUDE):
Remove. All uses rewritten to assume two's complement, which is
all we can reasonably test nowadays anyway.
* top/maint.mk (_intprops_names): Remove the removed macros.
Paul Eggert [Mon, 11 Apr 2016 15:42:02 +0000 (08:42 -0700)]
stdint: port to strict C11 left shift
* lib/stdint.in.h (_STDINT_MIN, _STDINT_MAX):
Pacify clang -Wshift-negative-value, which should be an issue only
on clang setups where stdint.h does not conform to C11 or to C++11.
Problem reported by Philipp Stephani in: http://bugs.gnu.org/23261
Paul Eggert [Sat, 9 Apr 2016 08:28:36 +0000 (01:28 -0700)]
mbrtowc: work around glibc bug#19932
Fix mbrtowc so that it never returns -1 in the C locale,
as this conflicts with a future version of POSIX
http://austingroupbugs.net/view.php?id=663#c2738
and causes problems with GNU grep: http://bugs.gnu.org/23234
See glibc bug 19932:
https://sourceware.org/bugzilla/show_bug.cgi?id=19932
* doc/posix-functions/mbrlen.texi (mbrlen):
* doc/posix-functions/mbrtowc.texi (mbrtowc):
Document the glibc bug.
* lib/mbrtowc.c [C_LOCALE_MAYBE_EILSEQ]:
Include hard-locale.h, locale.h.
(rpl_mbrtowc): Work around the C_LOCALE_MAYBE_EILSEQ bug,
if the bug is possible.
* m4/mbrtowc.m4 (gl_MBRTOWC_C_LOCALE): New macro.
(gl_FUNC_MBRTOWC): Use it, and define C_LOCALE_MAYBE_EILSEQ as needed.
* modules/hard-locale (License): Now LGPLv2+, for mbrtowc.
* modules/mbrtowc (Depends-on): Add hard-locale.
* modules/mbrtowc-tests (Files, TESTS): Add tests/test-mbrtowc5.sh.
* tests/test-mbrtowc.c (main): Test for bug fix if arg is '5'.
* tests/test-mbrtowc5.sh: New file.