* m4/stdlib_h.m4 (gl_STDLIB_H): Check for qsort_r.
(gl_STDLIB_H_DEFAULTS): Initialize HAVE_QSORT_R.
* modules/stdlib (Makefile.am): Substitute HAVE_QSORT_R.
* lib/stdlib.in.h (qsort_r): Provide declaration if the function does
not exist.
* m4/qsort_r.m4 (gl_FUNC_QSORT_R): Use AC_CHECK_FUNCS to test whether
the function exists.
* modules/qsort_r: Add comments.
This works even on FreeBSD, which has an incompatible qsort_r API.
* MODULES.html.sh: Add it.
* doc/glibc-functions/qsort_r.texi: It's now supported.
* lib/qsort.c: New file, taken from glibc with minor changes
inside "#ifndef _LIBC" and with an unnecessary "#include <alloca.h>"
removed.
* lib/qsort_r.c: New file, compiled only on FreeBSD.
* lib/stdlib.in.h (qsort_r): Declare in the usual way.
* m4/stdlib_h.m4 (gl_STDLIB_H_DEFAULTS):
* modules/qsort_r, modules/qsort_r-tests: New files.
* modules/stdlib (Makefile): Set up its defaults.
* tests/test-qsort_r.c: New file.
* MODULES.html.sh (Extra functions based on ANSI C 89):
Add secure_getenv.
* doc/glibc-functions/secure_getenv.texi: New file.
* doc/gnulib.texi: Include it.
* lib/secure_getenv.c, m4/secure_getenv.m4, modules/secure_getenv:
New files.
* lib/stdlib.in.h (secure_getenv): New decl.
* m4/stdlib_h.m4 (gl_STDLIB_H, gl_STDLIB_H_DEFAULTS):
* modules/stdlib (stdlib.h):
Add secure_getenv checks.
POSIX left errno undefined on ptsname() failure, although there
has at least been an effort to specify reasonable values to use:
http://www.austingroupbugs.net/view.php?id=503
However, our tests for ptsname and ptsname_r already require errno
to be set to useful values (as in glibc), so it is worth replacing
ptsname on FreeBSD 8.2 in order to get better QoI and pass the test.
* m4/ptsname.m4 (gl_FUNC_PTSNAME): Probe for FreeBSD bug.
* m4/stdlib_h.m4 (gl_STDLIB_H_DEFAULTS): Add new witness.
* modules/stdlib (Makefile.am): Replace witness.
* lib/stdlib.in.h (ptsname): Allow for replacement.
* modules/ptsname (configure.ac): Trigger replacement.
* doc/posix-functions/ptsname.texi (ptsname): Document this.
* lib/stdlib.in.h (ptsname_r): Override if REPLACE_PTSNAME_R is 1.
* m4/stdlib_h.m4 (gl_STDLIB_H_DEFAULTS): Initialize REPLACE_PTSNAME_R.
* m4/ptsname_r.m4 (gl_FUNC_PTSNAME_R): Set REPLACE_PTSNAME_R if the
function is not declared or incompatibly declared.
* modules/stdlib (Makefile.am): Substitute REPLACE_PTSNAME_R.
* modules/ptsname_r (Depends-on, configure.ac): Update.
* doc/glibc-functions/ptsname_r.texi: Mention the OSF/1 problems.
For now, this replacement focuses solely on compilation
compatibility, and assumes that isatty() and ttyname_r() work
on a master side pty; if this assumption fails, or if
thread-safety is also required, then a later patch can follow
the lead of strerror_r.c in wrapping the system ptsname()
with a lock.
* modules/ptsname_r: New module.
* m4/ptsname_r.m4 (gl_FUNC_PTSNAME_R): New file.
* lib/ptsname.c (__ptsname_r): Split...
* lib/ptsname_r.c: ...into new file.
* m4/stdlib_h.m4 (gl_STDLIB_H): Check for decl.
(gl_STDLIB_H_DEFAULTS): Set witness defaults.
* modules/stdlib (Makefile.am): Substitute witnesses.
* lib/stdlib.in.h (ptsname_r): Declare it.
* doc/glibc-functions/ptsname_r.texi (ptsname_r): Document it.
* MODULES.html.sh (Misc): Likewise.
* modules/ptsname (Depends-on): Alter dependency.
* doc/posix-functions/ptsname.texi (ptsname): Mention new module.
Signed-off-by: Eric Blake <eblake@redhat.com>
* lib/stdlib.in.h (_GL_ATTRIBUTE_NORETURN): Remove.
(_Exit): Use _Noreturn rather than _GL_ATTRIBUTE_NORETURN.
* modules/stdlib (Depends-on): Add _Noreturn.
(stdlib.h): Depend on $(_NORETURN_H), and copy it in.
* lib/stdlib.in.h (mbtowc): New declaration.
* lib/mbtowc.c: New file.
* lib/mbtowc-impl.h: New file, from libutf8 with modifications
* m4/mbtowc.m4: New file.
* m4/stdlib_h.m4 (gl_STDLIB_H_DEFAULTS): Initialize GNULIB_MBTOWC,
REPLACE_MBTOWC.
* modules/stdlib (Makefile.am): Substitute GNULIB_MBTOWC,
REPLACE_MBTOWC.
* modules/mbtowc: New file.
* tests/test-stdlib-c++.cc: Test signature of mbtowc.
* doc/posix-functions/mbtowc.texi: Mention the new module.
* modules/btowc (Depends-on): Add mbtowc.
Fix a serious and tricky problem encountered when attempting to
add the getloadavg module to Emacs. Emacs worked fine on RHEL
5.5, but it crashed due to memory corruption on Solaris 10 with
Sun C 5.11. Emacs normally ORs 3-bit tags into their low-order
bits that are otherwise zero. This tagging is optional inside
Emacs but is preferred and is used when __attribute__ ((__aligned
(8))) works, as it does with both recent-enough GCC and with Sun C
5.11. However, Sun C 5.11 is not GCC and does not #define
__GNUC__ and __GNUC_MINOR__.
When I added the getloadavg module to Emacs, it brought in
stdlib.in.h, which contained this fragment:
#ifndef __attribute__
# if __GNUC__ < 2 || (__GNUC__ == 2 && __GNUC_MINOR__ < 8)
# define __attribute__(Spec) /* empty */
# endif
#endif
When files that include <stdlib.h> were compiled with Sun C 5.11,
the above code disabled __attribute__ ((__aligned (8))), which
caused variables to not be properly aligned, which eventually led
to the pointer corruption mentioned above. (This was a bit hard
to diagnose, unfortunately.)
Several "#define __attribute__(X) /* empty */" code snippets need
to be eradicated from Gnulib to work with non-GCC compilers that
support __attribute__. The Autoconf way to do this is to test for
each kind of attribute that we want support for, and selectively
enable that in source code.
Fix this problem just for stdlib.h, by adding a test for the
__noreturn__ attribute, and change stdlib.in.h to use that test
when needed. This technique can be easily generalized to the
other *.in.h files and attributes, and a similar technique can be
used for *.h and *.c files. This patch is enough to solve the
problem for Emacs + getloadavg, and I thought I'd publish it for
feedback before undertaking further, similar fixes in other
modules.
This patch does not arrange to #define HAVE_ATTRIBUTE_NORETURN
because it's not needed for stdlib.h. It merely substitutes the
value directly into stdlib.h. We may well need to #define it, or
similar symbols, for other modules, but it's nice to also have an
option to not #define it for applications like Emacs that do not
need it.
* lib/stdlib.in.h (__attribute__): Do not #define.
(_GL_ATTRIBUTE_NORETURN): New macro, which in stdlib.h needs to
be defined only if the _Exit module is also used.
* m4/_Exit.m4 (gl_FUNC__EXIT): Require gl_ATTRIBUTE_NORETURN.
* m4/stdlib_h.m4 (gl_STDLIB_H_DEFAULTS): Subst
HAVE_ATTRIBUTE_NORETURN and default it to 1, its value on GNU
platforms.
* modules/_Exit (Files): Add m4/attribute.m4.
* modules/stdlib (Makefile.am): Substitute HAVE_ATTRIBUTE_NORETURN.
* m4/attribute.m4: New file.
* lib/stdlib.in.h: Don't include <stdint.h> merely because
GNULIB_POSIXCHECK is defined. GNULIB_POSIXCHECK seems to
be independent of whether stdint.h is needed.
* m4/random_r.m4 (gl_FUNC_RANDOM_R): Check for struct random_data
here, instead of ...
* m4/stdlib_h.m4 (gl_STDLIB_H): ... here. Applications that need
struct random_data should be using the random_r module, not just
the stdlib module (which wouldn't make sense: what package needs
just struct random_data without also needing random_r?).
* modules/stdlib (Depends-on): Remove stdint.
Back out yesterday's big merge from the libposix branch, except for the
openat-die.c change that was reviewed and approved by Paul Eggert.
Put the ChangeLog entry at the top, to reflect the order of changes in the
master branch.
Modify gnulib-tool to strip them from generated Makefile.am-s where
they are not needed. Also, replace M4 macro prefixes when requested.
Finally, omit calls to error() for libposix.
* lib/stdlib.in.h (setenv): Test HAVE_DECL_UNSETENV, not HAVE_UNSETENV.
* m4/setenv.m4 (gl_FUNC_UNSETENV): Test whether unsetenv is declared.
Don't set HAVE_UNSETENV. In the test program, set _BSD.
* m4/stdlib_h.m4 (gl_STDLIB_H_DEFAULTS): Initialize HAVE_DECL_UNSETENV,
not HAVE_UNSETENV.
* modules/stdlib (Makefile.am): Substitute HAVE_DECL_UNSETENV, not
HAVE_UNSETENV.
* doc/posix-functions/unsetenv.texi: Mention the OSF/1 5.1 problem.
* lib/stdlib.in.h (setenv): Test HAVE_DECL_SETENV, not HAVE_SETENV.
* m4/setenv.m4 (gl_FUNC_SETENV_SEPARATE): Test whether setenv is
declared. Don't set HAVE_SETENV.
* m4/stdlib_h.m4 (gl_STDLIB_H_DEFAULTS): Initialize HAVE_DECL_SETENV,
not HAVE_SETENV.
* modules/stdlib (Makefile.am): Substitute HAVE_DECL_SETENV, not
HAVE_SETENV.
* doc/posix-functions/setenv.texi: Mention the OSF/1 5.1 problem.
* modules/system-posix: New file.
* lib/stdlib.in.h: Include <sys/wait.h> only when the 'system-posix'
module is present.
* m4/stdlib_h.m4 (gl_STDLIB_H_DEFAULTS): Initialize
GNULIB_SYSTEM_POSIX.
* modules/stdlib (Depends-on): Remove sys_wait.
(Makefile.am): Substitute GNULIB_SYSTEM_POSIX.
* doc/posix-functions/system.texi: Mention the new module.
* doc/posix-headers/stdlib.texi: Likewise.
* tests/test-stdlib.c: If GNULIB_TEST_SYSTEM_POSIX is not defined,
define test_sys_wait_macros to a no-op.
Reported by Sam Steingold <sds@gnu.org>.
Each *.in.h file serves two purposes - provide enough self-contained
content to serve as a replacement for the system header regardless of
the user's compiler, and offer the developer the ability to detect any
gnulib modules that might have been inadvertantly excluded. The
former requires independence from config.h, and routes everything
through a Makefile.am snippet that uses @@ substitution specific to
the modules that were in use, with details learned at configure time.
The latter works by poisoning anything that gnulib did not replace,
but which the developer had natively available, explaining why their
tests passed during development. Poisoning relies on gcc features,
and requires manual triggering by adding -DGNULIB_POSIXCHECK to
CFLAGS; it assumes that <config.h> is properly included. In fact, we
do not want to use @@ substitution for HAVE_DECL_* during poisoning,
because the warning is only relevant for the gnulib modules which were
not included, and thus where the m4 macros to set proper @@ values
have not been run. Furthermore, we only need to poison interfaces
that already have a declaration; if something is not declared, then
the developer wouldn't have been able to link, so their code won't be
using the problematic interface in the first place.
* modules/stdio (Depends-on, Makefile.am): Drop link-warning.
* modules/unistd (Depends-on, Makefile.am): Likewise.
* modules/arpa_inet (Depends-on): Replace link-warning with
warn-on-use.
(Makefile.am): Update rules accordingly.
* modules/ctype (Depends-on, Makefile.am): Likewise.
* modules/dirent (Depends-on, Makefile.am): Likewise.
* modules/fcntl-h (Depends-on, Makefile.am): Likewise.
* modules/inttypes (Depends-on, Makefile.am): Likewise.
* modules/langinfo (Depends-on, Makefile.am): Likewise.
* modules/locale (Depends-on, Makefile.am): Likewise.
* modules/math (Depends-on, Makefile.am): Likewise.
* modules/search (Depends-on, Makefile.am): Likewise.
* modules/signal (Depends-on, Makefile.am): Likewise.
* modules/spawn (Depends-on, Makefile.am): Likewise.
* modules/stdlib (Depends-on, Makefile.am): Likewise.
* modules/string (Depends-on, Makefile.am): Likewise.
* modules/strings (Depends-on, Makefile.am): Likewise.
* modules/sys_file (Depends-on, Makefile.am): Likewise.
* modules/sys_ioctl (Depends-on, Makefile.am): Likewise.
* modules/sys_select (Depends-on, Makefile.am): Likewise.
* modules/sys_socket (Depends-on, Makefile.am): Likewise.
* modules/sys_stat (Depends-on, Makefile.am): Likewise.
* modules/sys_times (Depends-on, Makefile.am): Likewise.
* modules/sys_utsname (Depends-on, Makefile.am): Likewise.
* modules/wchar (Depends-on, Makefile.am): Likewise.
* m4/arpa_inet_h.m4 (gl_HEADER_ARPA_INET): Check which functions
should be poisoned.
* m4/ctype.m4 (gl_CTYPE_H): Likewise.
* m4/dirent_h.m4 (gl_DIRENT_H): Likewise.
* m4/fcntl_h.m4 (gl_FCNTL_H): Likewise.
* m4/inttypes.m4 (gl_INTTYPES_H): Likewise.
* m4/langinfo_h.m4 (gl_LANGINFO_H): Likewise.
* m4/locale_h.m4 (gl_LOCALE_H): Likewise.
* m4/math_h.m4 (gl_MATH_H): Likewise.
* m4/search_h.m4 (gl_SEARCH_H): Likewise.
* m4/signal_h.m4 (gl_SIGNAL_H): Likewise.
* m4/spawn_h.m4 (gl_SPAWN_H): Likewise.
* m4/stdio_h.m4 (gl_STDIO_H): Likewise.
* m4/stdlib_h.m4 (gl_STDLIB_H): Likewise.
* m4/string_h.m4 (gl_HEADER_STRING_H_BODY): Likewise.
* m4/strings_h.m4 (gl_HEADER_STRINGS_H_BODY): Likewise.
* m4/sys_file_h.m4 (gl_HEADER_SYS_FILE_H_DEFAULTS): Likewise.
* m4/sys_ioctl_h.m4 (gl_SYS_IOCTL_H): Likewise.
* m4/sys_select_h.m4 (gl_HEADER_SYS_SELECT): Likewise.
* m4/sys_socket_h.m4 (gl_HEADER_SYS_SOCKET): Likewise.
* m4/sys_stat_h.m4 (gl_HEADER_SYS_STAT_H): Likewise.
* m4/sys_times_h.m4 (gl_SYS_TIMES_H): Likewise.
* m4/sys_utsname_h.m4 (gl_SYS_UTSNAME_H): Likewise.
* m4/unistd_h.m4 (gl_UNISTD_H): Likewise.
* m4/wchar.m4 (gl_WCHAR_H): Likewise.
* lib/arpa_inet.in.h: Use _GL_WARN_ON_USE instead of
GL_LINK_WARNING.
* lib/ctype.in.h: Likewise.
* lib/dirent.in.h: Likewise.
* lib/fcntl.in.h: Likewise.
* lib/inttypes.in.h: Likewise.
* lib/langinfo.in.h: Likewise.
* lib/locale.in.h: Likewise.
* lib/math.in.h: Likewise.
* lib/search.in.h: Likewise.
* lib/signal.in.h: Likewise.
* lib/spawn.in.h: Likewise.
* lib/stdio.in.h: Likewise.
* lib/stdlib.in.h: Likewise.
* lib/string.in.h: Likewise.
* lib/strings.in.h: Likewise.
* lib/sys_file.in.h: Likewise.
* lib/sys_ioctl.in.h: Likewise.
* lib/sys_select.in.h: Likewise.
* lib/sys_socket.in.h: Likewise.
* lib/sys_stat.in.h: Likewise.
* lib/sys_times.in.h: Likewise.
* lib/sys_utsname.in.h: Likewise.
* lib/unistd.in.h: Likewise.
* lib/wchar.in.h: Likewise.
Signed-off-by: Eric Blake <ebb9@byu.net>
POSIX requires setenv(NULL,"",0), setenv("a=b","",0),
unsetenv(NULL), and unsetenv("a=b") to fail with EINVAL, but
many BSD implementations lack validation. The gnulib
replacement for void unsetenv did not do validation, and the
replacement for setenv was out of sync with glibc. Also, some
BSD implementations of setenv("a","==",1) eat the leading '='.
See also some recent Austin Group interpretations on environ:
http://austingroupbugs.net/view.php?id=167http://austingroupbugs.net/view.php?id=185
* lib/setenv.c (setenv) [!HAVE_SETENV]: Resync from glibc.
(setenv) [HAVE_SETENV]: Work around bugs.
* lib/unsetenv.c (unsetenv) [HAVE_UNSETENV]: Work around bugs.
* m4/setenv.m4 (gl_FUNC_SETENV_SEPARATE, gl_FUNC_UNSETENV): Check
for bugs.
(gl_FUNC_SETENV): Write in terms of gl_FUNC_SETENV_SEPARATE.
* m4/environ.m4 (gl_ENVIRON): Avoid expand-before-require.
* m4/stdlib_h.m4 (gl_STDLIB_H_DEFAULTS): Update defaults.
* modules/stdlib (Makefile.am): Update substitutions.
* lib/stdlib.in.h (setenv, unsetenv): Update prototypes.
* doc/posix-functions/setenv.texi (setenv): Document the bugs.
* doc/posix-functions/unsetenv.texi (unsetenv): Likewise.
* modules/setenv-tests: New test.
* modules/unsetenv-tests: Likewise.
* tests/test-setenv.c: New file.
* tests/test-unsetenv.c: Likewise.
Signed-off-by: Eric Blake <ebb9@byu.net>
Consolidate the m4 macros into a single file, since both modules
now have to worry about replacing canonicalize_file_name on
buggy glibc.
* m4/canonicalize-lgpl.m4: Move contents...
* m4/canonicalize.m4: ...here.
(gl_CANONICALIZE_LGPL): Factor realpath check...
(gl_FUNC_REALPATH_WORKS): ...into new macro. Enhance to catch
glibc 2.3.5 bug, fixed 2005-04-27.
(gl_FUNC_CANONICALIZE_FILENAME_MODE): Use it.
(gl_PREREQ_CANONICALIZE_LGPL): Inline...
(gl_CANONICALIZE_LGPL_SEPARATE): ...into this macro.
* modules/canonicalize-lgpl (Files): Manage file rename.
* m4/stdlib_h.m4 (gl_STDLIB_H_DEFAULTS): Provide default.
* modules/stdlib (Makefile.am): Substitute witness.
* lib/stdlib.in.h (canonicalize_file_name): Declare if replacement
is needed.
* lib/canonicalize-lgpl.c: Also compile if canonicalize_file_name
replacement is required.
* lib/canonicalize.c (canonicalize_file_name): Likewise.
* doc/glibc-functions/canonicalize_file_name.texi
(canonicalize_file_name): Document this.
* doc/posix-functions/realpath.texi (realpath): Likewise.
Signed-off-by: Eric Blake <ebb9@byu.net>
Forward-looking to when more platforms comply with POSIX 2008,
but don't provide glibc extensions. For example, this could
fix // handling in cygwin 1.7 (well, if cygwin didn't have bugs
in .. handling). canonicalize can't use native realpath, for
the same reason that it does not use resolvepath.
* lib/canonicalize-lgpl.c (realpath): Guard with
FUNC_REALPATH_WORKS.
* lib/stdlib.in.h (realpath): Make declaration optional based on
HAVE_REALPATH.
* m4/canonicalize-lgpl.m4 (gl_CANONICALIZE_LGPL): Check whether
native realpath works.
* m4/stdlib_h.m4 (gl_STDLIB_H_DEFAULTS): Provide default.
* modules/stdlib (Makefile.am): Substitute witness.
Signed-off-by: Eric Blake <ebb9@byu.net>
Match current gnulib conventions of glibc functions being declared in
the same header as glibc. This makes it easier to replace broken
canonicalize_file_name. canonicalize still requires "canonicalize.h"
because canonicalize_filename_mode is not in glibc. With this patch,
canonicalize-lgpl always provides realpath and canonicalize_file_name,
while canonicalize can provide canonicalize_file_name but not
realpath; if both modules are in use, canonicalize_file_name comes
from the LGPLv2+ sources, but the testing comes from canonicalize.
* modules/canonicalize-lgpl (Files): Drop canonicalize.h.
(Include): Mention <stdlib.h>.
(configure.ac): Mention functions we provide.
* modules/canonicalize (configure.ac): Likewise.
* m4/canonicalize-lgpl.m4 (gl_CANONICALIZE_LGPL): Always replace
realpath if canonicalize_file_name is missing.
* m4/stdlib_h.m4 (gl_STDLIB_H_DEFAULTS): Provide defaults.
* modules/stdlib (Makefile.am): Substitute witnesses.
* lib/stdlib.in.h (canonicalize_file_name, realpath): Declare.
* lib/canonicalize-lgpl.c (includes): Adjust accordingly.
* lib/canonicalize.h (canonicalize_file_name): Drop declaration.
* NEWS: Document this.
* doc/glibc-functions/canonicalize_file_name.texi
(canonicalize_file_name): Likewise.
* doc/posix-functions/realpath.texi (realpath): Likewise.
* tests/test-canonicalize-lgpl.c (includes): Use <stdlib.h>.
Signed-off-by: Eric Blake <ebb9@byu.net>
