On IRIX 6.5, link(file, "dangling") creates the target of dangling
as a link to file, rather than failing with EEXIST.
* m4/link.m4 (gl_FUNC_LINK): Expose the bug.
* lib/link.c (rpl_link): Work around it.
* tests/test-link.h (test_link): Enhance test.
* doc/posix-functions/link.texi (link): Document the bug.
Signed-off-by: Eric Blake <eblake@redhat.com>
On OpenBSD, strerror_r(1000,buf,19) gives "Unknown error: " rather
than "Unknown error: 100" while failing with ERANGE. Admittedly,
this behavior is nice, since a truncated integer is misleading,
but all other platforms use maximal strings on ERANGE and we are
already replacing strerror_r for other reasons, so it is easier
to work around this behavior than to adjust the testsuite (how
do you quickly decide if the only reason that the ERANGE string
was shorter than maximal was because the implementation avoided
truncating an integer?).
This patch intentionally avoids dragging in the strnlen module.
* lib/strerror_r.c (strerror_r): Always use maximal string.
* doc/posix-functions/strerror_r.texi (strerror_r): Document it.
Signed-off-by: Eric Blake <eblake@redhat.com>
OpenBSD treats strerror_r(0,,) as a success, but with a message
"Undefined error: 0"; while this is distinct from strerror_r(-1,,)
returning "Unknown error: -1", it does not imply success. Meanwhile,
if buf is short enough for ERANGE, then we can't use strstr to look
for "Unknown" or "Undefined" in the resulting message, like we had
been doing for strerror(). Fix this by shifting the burden - now
the strerror-override code guarantees that 0 will have an
override when needed.
* lib/strerror-override.c (strerror_override): Also override 0
when needed.
* lib/strerror-override.h (strerror_override): Likewise.
* lib/strerror.c (strerror): Simplify, now that 0 override is done
earlier.
* lib/strerror_r.c (strerror_r): Likewise.
* m4/strerror.m4 (gl_FUNC_STRERROR): Split detection of 0
behavior...
(gl_FUNC_STRERROR_0): ...into new macro.
* m4/strerror_r.m4 (gl_FUNC_STRERROR_R): Replace strerror_r if 0
is overridden.
(gl_FUNC_STRERROR_R_WORKS): Avoid extra tests if 0 is broken.
* modules/strerror-override (Files): Add strerror.m4.
(configure.ac): Also provide override for 0 when needed.
* doc/posix-functions/strerror.texi (strerror): Document this.
* doc/posix-functions/perror.texi (perror): Likewise.
Signed-off-by: Eric Blake <eblake@redhat.com>
* modules/round-ieee (Depends-on): Add floor-ieee, ceil-ieee.
* doc/posix-functions/round.texi: Mention problem with negative
arguments.
* doc/posix-functions/ceil.texi: Mention problem on OSF/1 5.1.
* m4/roundf.m4 (gl_FUNC_ROUNDF): Test also the sign of roundf (-0.3f).
* modules/roundf-ieee (Depends-on): Add floorf-ieee, ceilf-ieee.
* doc/posix-functions/roundf.texi: Mention problem with negative
arguments.
* doc/posix-functions/ceilf.texi: Mention problem on OSF/1 5.1.
MacOS X 10.5 strerror(0) is "Unknown error: 0", which is not distinguished
from "Unknown error: -1" for out-of-range. Worse, strerror_r(0,,)
is "Undefined error: 0", although strerror_r for all other out-of-range
values matches strerror.
* m4/strerror.m4 (gl_FUNC_STRERROR): Flush out MacOS bug.
* m4/strerror_r.m4 (gl_FUNC_STRERROR_R_WORKS): Likewise, and fix
logic bug.
* lib/strerror_r.c (strerror_r): Fix the bug.
* lib/strerror.c (strerror): Likewise.
* doc/posix-functions/strerror_r.texi (strerror_r): Document the
problem.
* doc/posix-functions/strerror.texi (strerror): Likewise.
* doc/posix-functions/perror.texi (perror): Likewise.
* tests/test-strerror.c (main): Enhance test.
* tests/test-strerror_r.c (main): Likewise.
Signed-off-by: Eric Blake <eblake@redhat.com>
Cygwin __xpg_strerror_r clobbers strerror, until cygwin 1.7.10; in
this case, we must replace strerror, but there was nothing in
strerror.m4 finding fault with cygwin's native strerror. Solve
this by splitting strerror_r.m4 into two parts, one for probing
for the strerror_r bug, and one for replacing strerror_r based
on the presence of bugs unrelated to signature issues; that way,
the 'strerror' module in isolation will not replace strerror, but
if strerror_r is in use for any other reason, then the two use
separate buffers.
Note that cygwin's strerror is thread-safe while the gnulib
replacement is not; but this is no worse than any other
platform where strerror is not thread-safe; in a single-threaded
program, the difference is not observable, and in a multi-threaded
program, you really shouldn't be using strerror in the first place.
Also note that this ends up replacing glibc 2.13 strerror as it
deems __xpg_strerror_r broken on that platform, which isn't
technically necessary for strerror, but doesn't hurt too much.
Meanwhile, glibc 2.14 fixed __xpg_strerror_r, and strerror is not
replaced in that scenario.
* m4/strerror_r.m4 (gl_FUNC_STRERROR_R): Split...
(gl_FUNC_STRERROR_R_WORKS): ...into new macro, to detect Cygwin
bug without replacing strerror_r.
* m4/strerror.m4 (gl_FUNC_STRERROR): Replace strerror if
strerror_r is buggy, but without requiring strerror_r compilation.
* doc/posix-functions/strerror_r.texi (strerror_r): Fix docs.
Signed-off-by: Eric Blake <eblake@redhat.com>
glibc was not the only platform where fprintf(fopen(,"r"))
fails to detect errors; cygwin 1.7.9 is another culprit
(although it will be fixed for 1.7.10), and I suspect that
several other platforms were failing perror2 for the same
reason.
At this point, there are so many functions affected, and
the way to avoid the bug is easy enough (don't pass bogus
streams to output-producing functions), that I'm not worried
about fixing things other than to document them.
* tests/test-perror2.c (main): Relax test on requiring detection
of stream errors, and use unbuffered stream.
* doc/posix-functions/dprintf.texi (dprintf): Document bug.
* doc/posix-functions/fprintf.texi (fprintf): Likewise.
* doc/posix-functions/fputc.texi (fputc): Likewise.
* doc/posix-functions/fputs.texi (fputs): Likewise.
* doc/posix-functions/fputws.texi (fputws): Likewise.
* doc/posix-functions/fwprintf.texi (fwprintf): Likewise.
* doc/posix-functions/fwrite.texi (fwrite): Likewise.
* doc/posix-functions/getopt.texi (getopt): Likewise.
* doc/posix-functions/perror.texi (perror): Likewise.
* doc/posix-functions/printf.texi (printf): Likewise.
* doc/posix-functions/psiginfo.texi (psiginfo): Likewise.
* doc/posix-functions/psignal.texi (psignal): Likewise.
* doc/posix-functions/putc.texi (putc): Likewise.
* doc/posix-functions/putc_unlocked.texi (putc_unlocked):
Likewise.
* doc/posix-functions/putchar.texi (putchar): Likewise.
* doc/posix-functions/putchar_unlocked.texi (putchar_unlocked):
Likewise.
* doc/posix-functions/puts.texi (puts): Likewise.
* doc/posix-functions/putwc.texi (putwc): Likewise.
* doc/posix-functions/putwchar.texi (putwchar): Likewise.
* doc/posix-functions/vdprintf.texi (vdprintf): Likewise.
* doc/posix-functions/vfprintf.texi (vfprintf): Likewise.
* doc/posix-functions/vfwprintf.texi (vfwprintf): Likewise.
* doc/posix-functions/vprintf.texi (vprintf): Likewise.
* doc/posix-functions/vwprintf.texi (vwprintf): Likewise.
* doc/posix-functions/wordexp.texi (wordexp): Likewise.
* doc/posix-functions/wprintf.texi (wprintf): Likewise.
Signed-off-by: Eric Blake <eblake@redhat.com>
AIX 7.1 realpath() gets horribly confused by consecutive /
in the name to be resolved.
* m4/canonicalize.m4 (gl_FUNC_REALPATH_WORKS): Expose AIX bug.
* doc/posix-functions/realpath.texi (realpath): Document it.
Reported by Bruno Haible.
Signed-off-by: Eric Blake <eblake@redhat.com>
* m4/setlocale.m4 (gl_FUNC_SETLOCALE): Set REPLACE_SETLOCALE also on
Cygwin 1.5.x.
* doc/posix-functions/setlocale.texi: Mention that the problem with the
LC_CTYPE category also exists on Cygwin 1.5.x.
No need to make a wrapper that burns static storage when we can
just use stack storage.
* modules/perror (Files): Drop strerror-impl.h.
* lib/perror.c (perror): Use our own stack buffer, rather than
calling a wrapper that uses static storage.
* doc/posix-functions/perror.texi (perror): Document a limitation
of our replacement.
Signed-off-by: Eric Blake <eblake@redhat.com>
mingw getcwd(buf, 0) fails with ERANGE, instead of the required
EINVAL. Since we're already replacing getcwd on mingw, the
workaround is trivial.
* lib/getcwd-lgpl.c (rpl_getcwd): Guarantee correct error.
* doc/posix-functions/getcwd.texi (getcwd): Document it.
Reported by Matthias Bolte.
Signed-off-by: Eric Blake <eblake@redhat.com>
Document it as a known bug, but one where we don't provide a
workaround since programmers are unlikely to hit it in practice.
* doc/posix-functions/fprintf.texi (fprintf): Document it.
* doc/posix-functions/printf.texi (printf): Likewise.
* doc/posix-functions/vfprintf.texi (vfprintf): Likewise.
* doc/posix-functions/vprintf.texi (vprintf): Likewise.
Signed-off-by: Eric Blake <eblake@redhat.com>
Solaris 10 populates buf on EINVAL, but not on ERANGE.
* lib/strerror_r.c (strerror_r): Partially populate buf on ERANGE
failures.
* doc/posix-functions/strerror_r.texi (strerror_r): Document this.
Signed-off-by: Eric Blake <eblake@redhat.com>
POSIX recommends (but does not require) that strerror_r populate
buf even on error. But since we guarantee this behavior for
strerror, we might as well also guarantee it for strerror_r.
* lib/strerror_r.c (safe_copy): New helper method.
(strerror_r): Guarantee a non-empty string.
* tests/test-strerror_r.c (main): Enhance tests to incorporate
recent POSIX rulings and to match our strerror guarantees.
* doc/posix-functions/strerror_r.texi (strerror_r): Document this.
Signed-off-by: Eric Blake <eblake@redhat.com>
Avoid cygwin strerror_r, since it clobbers strerror buffer through
cygwin 1.7.9. __xpg_strerror_r is okay, but if a program is
compiled on cygwin 1.7.8 or earlier, it is not available.
* lib/strerror_r.c (strerror_r): Don't use cygwin's strerror_r;
fall back instead to sys_errlist.
* modules/strerror (configure.ac): Add witness.
* tests/test-strerror_r.c (main): Enhance test.
* doc/posix-functions/strerror_r.texi (strerror_r): Document it.
* tests/test-perror2.c (main): Free memory before exit.
Signed-off-by: Eric Blake <eblake@redhat.com>
On Solaris, a non-zero return was also reflected into errno.
Leaving errno unchanged is a useful feature worth guaranteeing.
* m4/strerror_r.m4 (gl_FUNC_STRERROR_R): Check for Solaris behavior.
* doc/posix-functions/strerror_r.texi (strerror_r): Document it.
Signed-off-by: Eric Blake <eblake@redhat.com>
POSIX requires that 'errno = 0; perror ("")' print the same message
as strerror(0), but this failed if we were replacing strerror to work
around the FreeBSD bug of treating 0 as a failure.
The goal here is to _not_ replace perror on glibc, even though
strerror_r has to be replaced, because the strerror_r replacement is
only for the sake of correcting the signature rather than working
around bugs in the handling of any particular errnum value. Recall
that $gl_cv_func_strerror_r_works is only set if the POSIX signature
was detected in the first place.
* m4/perror.m4 (gl_FUNC_PERROR): Also replace perror if strerror_r
is broken. Move AC_LIBOBJ...
* modules/perror (configure.ac): Here.
* doc/posix-functions/perror.texi (perror): Document this.
* tests/test-perror2.c (main): Enhance test.
Signed-off-by: Eric Blake <eblake@redhat.com>
This uncovered a glibc bug, although not many people check for
perror failures, so for now I'm not working around it.
http://sourceware.org/bugzilla/show_bug.cgi?id=12792
* tests/macros.h (STREQ) Add macro.
* modules/perror-tests (Files): Add second test.
* tests/test-perror2.c (main): New file.
* doc/posix-functions/perror.texi (perror): Document glibc bug.
Signed-off-by: Eric Blake <eblake@redhat.com>
getcwd(NULL, 1) mallocs a larger buffer on BSD, rather than failing
with ERANGE as on glibc. This behavior difference is not worth
coding around, as it is an uncommon use of getcwd in the first place.
* doc/posix-functions/getcwd.texi (getcwd): Document portability
issue.
* tests/test-getcwd-lgpl.c (main): Relax test.
Reported by Matthias Bolte.
Signed-off-by: Eric Blake <eblake@redhat.com>
* modules/fclose (Depends-on): Add fflush.
* doc/posix-functions/fclose.texi (fclose): Document this.
* tests/test-fclose.c (main): Make test for this unconditional.
Signed-off-by: Eric Blake <eblake@redhat.com>
This fixes the fclose failures detected in the previous patch,
but only when the GPL fflush module is also in use. That is
because the need for behavior of resetting seekable input streams
is much less common, and the fix more complex. The LGPLv2+ test
for fclose() in isolation is relaxed to pass if fflush is not
being replaced to cater to input streams.
* modules/fflush (Depends-on): Add fclose.
* m4/fflush.m4 (gl_FUNC_FFLUSH): Also replace fclose.
* lib/fclose.c (rpl_fclose): Don't cause spurious failures on
memstreams with no backing fd.
* doc/posix-functions/fclose.texi (fclose): Document the use of
fflush module to fix the bug.
* tests/test-fclose.c (main): Relax test when fclose is used in
isolation.
Signed-off-by: Eric Blake <eblake@redhat.com>
Several platforms declare gai_strerror to return char* rather than
const char*. Worse, on mingw, if UNICODE is defined, it is defined
to return WCHAR*, which means the result is in unicode but an
application expecting bytes for characters will only see a one-byte
answer.
* m4/getaddrinfo.m4 (gl_GETADDRINFO): Detect broken signatures,
and work around mingw with UNICODE defined.
(gl_PREREQ_GETADDRINFO): Drop redundant decl check.
* m4/netdb_h.m4 (gl_NETDB_H_DEFAULTS): Add witness.
* modules/netdb (Makefile.am): Substitute it.
* lib/netdb.in.h (gai_strerror): Declare replacement.
* lib/gai_strerror.c (rpl_gai_strerror): Fix signature.
* doc/posix-functions/gai_strerror.texi (gai_strerror): Document
the fix.
Signed-off-by: Eric Blake <eblake@redhat.com>
On mingw, getcwd(NULL,1) succeeds, even though glibc documents that
with a non-zero size, the allocation will not exceed that many bytes.
On mingw, getcwd has the wrong signature. However, we don't have
to check for this if anything else triggers the replacement.
Also, fix a type bug that crept into the original getcwd-lgpl commit.
* m4/getcwd.m4 (gl_FUNC_GETCWD_NULL): Detect one mingw bug.
* doc/posix-functions/getcwd.texi (getcwd): Document the problems.
* lib/getcwd-lgpl.c (rpl_getcwd): Fix return type.
Signed-off-by: Eric Blake <eblake@redhat.com>
For programs that aren't worried about being invoked from an
current working directory longer than PATH_MAX (perhaps because
the program always does chdir to a sane location first), the
getcwd module is overkill, given that all modern portability
targets have a getcwd that works on short names.
* modules/getcwd-lgpl: New module.
* lib/getcwd-lgpl.c: New file.
* doc/posix-functions/getcwd.texi (getcwd): Document it.
* MODULES.html.sh (lacking POSIX:2008): Likewise.
* modules/getcwd (configure.ac): Set C witness.
* m4/getcwd.m4 (gl_FUNC_GETCWD_LGPL): New macro.
Signed-off-by: Eric Blake <eblake@redhat.com>
* m4/mkstemp.m4 (gl_FUNC_MKSTEMP): Add test for non-owner
read/write mode bits set in file created by mkstemp.
* doc/posix-functions/mkstemp.texi (mkstemp): Document the fix.
Signed-off-by: Eric Blake <eblake@redhat.com>
* modules/strtod-obsolete: New file.
* m4/strtod-obsolete.m4: New file.
* m4/strtod.m4 (gl_FUNC_STRTOD): Don't check whether strtod is declared
if gl_FUNC_STRTOD_OBSOLETE is not also defined.
* modules/strtod (Depends-on): Add strtod-obsolete.
* doc/posix-functions/strtod.texi: Mention module strtod-obsolete.
* modules/nonblocking (Depends-on): Remove 'open'.
* m4/nonblocking.m4 (gl_NONBLOCKING_IO_BODY): Set
gl_cv_have_open_O_NONBLOCK.
* m4/open.m4 (gl_FUNC_OPEN): Replace open() also when required for
O_NONBLOCK support.
* doc/posix-functions/open.texi: Document support for O_NONBLOCK.
* lib/write.c (rpl_write): Split a write request that failed merely
because the byte count was larger than the pipe buffer's size.
* doc/posix-functions/write.texi: Mention the problem with large byte
counts.
