This bug showed up via valgrind as a "Conditional jump or move
depends on uninitialized value(s)" error.
* src/tail.c (ignore_fifo_and_pipe): New function.
(main): Use it only when tailing forever.
The code to compute n_viable and mark some F[i] as ignored would call
isapipe on an uninitialized file descriptor. But n_viable and those
.ignored marks are useful/used only when tailing forever. This bug
was introduced via commit f0ff8c73 (7.6), "tail: make the new
piped-stdin test as portable as the old one".
* NEWS (Bug fixes): Mention it.
* NEWS (Bug fixes): Mention implications of the "make distcheck" change.
This was introduced on 2008-07-22 by commit 9bb0d576, "tests: ensure
"make check" w/tainted build dir no longer impacts $HOME".
* src/sort.c (main): Reset the SIGCHLD handler to the default
as otherwise wait() could return an error.
* tests/misc/sort-compress: Set the CHLD handler in a subshell
to SIG_IGN to ensure the sort command resets it to SIG_DFL.
* NEWS: Mention the fix.
* src/timeout.c (main): Reset the SIGCHLD handler to the default
as otherwise wait() could return -1 and set errno to ECHILD.
This condition was ignored until commit 0b1dcf33, on 31-08-2009,
"timeout: defensive handling of all wait() errors"
but subsequently timeout would run the command correctly
but then fail with an error message.
* tests/misc/timeout: In a subshell set the CHLD handler to
SIG_IGN to ensure the timeout command resets it to SIG_DFL.
* NEWS: Mention the fix.
* tests/misc/stty-row-col: Linux virtual consoles at least,
issue an error if you try to increase their size, so skip the
test if we can't increase the dimensions of the tty by 1 cell.
Reported by Matthew Burgess.
* dist-check.mk (tmpdir): Rename from TMPDIR. Use ./tests/torture
unconditionally, rather than $TMPDIR-with-default-to-/tmp.
Otherwise, running "make distcheck" could leave an empty /tmp/coreutils
directory behind.
(tp): Simplify, now that it's always in the build-dir.
(taint-distcheck): Set HOME earlier, in case $(MAKE) misbehaves.
(my-instcheck, coreutils-path-check): Add diagnostics, so it's easier
to diagnose when each runs.
(coreutils-path-check): Run configure with --quiet, to reduce output.
Inspired by Ralf Wildenhues' report of /tmp/coreutils being left behind.
* tests/ls/readdir-mountpoint-inode: With some systems, stat can
succeed on a mount point and report that the inode number is 0.
Since ls displays "?" for those, that would otherwise show up as a
difference. Skip such mount points. Reported by Sergei Steshenko
in http://thread.gmane.org/gmane.comp.gnu.coreutils.bugs/19142
* tests/tail-2/inotify-race: Note the caveats of the test.
I.E. the intermittent skips and the gdb hang reported
by Alan Curry. Add extra info to the log on why the test
is skipped as it may be due to multiple reasons. Mark
the test as very expensive so that it's not normally run.
If getgroups failed with ENOSYS, mgetgroups would unnecessarily
fail, and that provoked id into freeing an uninitialized pointer.
Meanwhile, we were not using xalloc_die properly. Both issues
are better solved in gnulib, by introducing xgetgroups; this
patch uses the new interface.
Regression introduced by commit 6a31fd8d7.
* gnulib: Update, for mgetgroups improvments.
* src/id.c (print_full_info): Adjust caller to die on allocation
failure, and no longer worry about ENOSYS.
* src/group-list.c (print_group_list): Likewise.
* src/setuidgid.c (main): Likewise.
* NEWS: Mention the fix.
* THANKS: Update.
Reported by Scott Harrison.
While "rm ''" would properly fail, "rm F1 '' F2" would fail
to remove F1 and F2, due to the empty string argument.
This bug was introduced on 2009-07-12, via commit 4f73ecaf,
"rm: rewrite to use fts".
* gnulib: Update to latest, for fixed fts.c.
* NEWS (Bug fixes): Describe it.
* tests/rm/empty-name: Adjust for changed diagnostic.
(mk_file): Define, copied from misc/ls-misc.
(empty-name-2): New test, for today's fix.
* lib/xfts.c (xfts_open): Reflect the change in fts_open, now that
it no longer fails immediately when one argument is the empty string.
Assert that the bit flags were not the cause of failure.
* po/POTFILES.in: Remove xfts.c.
* THANKS: Update.
Reported by Ladislav Hagara.
* bootstrap (get_version): Don't use perl's $] special
variable, as that requires updating all bootstrap.conf files to
use perl's x.yyyzzz version format. Instead make the regular
expression more general to support version formats from older
perl-5.005_002 (5.5.2) and perl-5.11 which has other numbers
in the version line.
Counterpart to commit 8fe40b84bd, since test-link.c uses rename,
and we override gnulib with a rename() replacement that can xalloc_die.
* gl/modules/link-tests.diff: New file.
Cygwin 1.5 has a broken sleep, and the gnulib tests dragged in
rpl_sleep which then caused a link failure because it wasn't in
libcoreutils.a. We could solve it by using the gnulib sleep module.
However, sleep and usleep may interact poorly with SIGALRM, and they
have less granularity; so it is better to adopt a policy that if we
must sleep, prefer xnanosleep.
* src/sort.c (pipe_fork): Use xnanosleep, to avoid the need for
rpl_sleep on cygwin, and to reduce granularity.
(MAX_FORK_TRIES_COMPRESS, MAX_FORK_TRIES_DECOMPRESS): Increase,
to account for reduction in granularity.
* src/tail.c (tail_file): Use xnanosleep in debug code.
* cfg.mk (sc_prohibit_sleep): New rule.
* cfg.mk (gnu_ftp_host-alpha, gnu_ftp_host-beta gnu_ftp_host-stable):
(gnu_rel_host, url_dir_list): Remove definitions. The defaults,
now provided by maint.mk, are the same.
* gnulib: Update for latest, including those maint.mk additions.
These tests perform no PATH search, and used to simply delete PATH from
the environment. However, that is not portable, as seen on Cygwin,
where cygwin.dll must be resolvable via PATH when starting a sub-shell.
With commit 0cc04241, we took the alternate approach of untainting the
incoming $ENV{PATH}, but that fails when it contains an other-writable
directory. Instead, now we hard code it to '/bin:/usr/bin'.
* tests/misc/pwd-long: Hard code $ENV{PATH} to a safe value.
* tests/rm/fail-eperm: Likewise.
Reported by Gilles Espinasse, Andreas Schwab, and Bauke Jan Douma.
* src/true.c (main): There is no reason to examine argv[0],
call atexit, etc., in the usual case in which we're about to exit.
This has the side effect of making it so that these programs
no longer segfault when subjected to execve abuse.
Before this change, these commands would make "true" segfault:
printf '%s\n' '#include <unistd.h>' 'int main(int c, char**v)' \
'{ execve (v[1], 0, 0); }' > k.c && gcc k.c && ./a.out $PWD/true
Now it succeeds. Reported by Tetsuo Handa and Bart Van Assche
via Ondřej Vašík in http://bugzilla.redhat.com/537684.
Tailing forever and by-name (--follow=name, -F), tail would
sometimes fail to follow a file that had been removed via rename.
If you can't apply this patch and have tail 7.6 or newer, you can
work around the bug via the undocumented --disable-inotify option.
* src/tail.c (tail_forever_inotify): When tailing by name (-F),
do not un-watch a file upon receipt of the IN_MOVE_SELF event.
Reported by Arjan Opmeer in http://bugs.debian.org/548439.
* NEWS (Bug fixes): Mention it.
Also see http://marc.info/?l=coreutils-bug&m=125829031916515
* tests/Makefile.am (TESTS): Add tail-2/inotify-rotate.
* tests/tail-2/inotify-rotate: New test.
Jim Meyering