This change affects only systems that have neither *at function support
nor the /proc/self/fd support required to emulate those *at functions.
* src/remove.c (write_protected_non_symlink): Call faccessat
unconditionally. Thus we no longer need euidaccess_stat, which was
the sole function used here to operate on a full relative file name.
Remove full_name parameter and update caller.
* lib/euidaccess-stat.h: Remove file.
* lib/euidaccess-stat.c: Likewise.
* m4/euidaccess-stat.m4: Likewise.
* po/POTFILES.in: Remove lib/euidaccess-stat.c.
* m4/prereq.m4 (gl_PREREQ): Don't require gl_EUIDACCESS_STAT.
Prompted by a report from Bruno Haible that the rm/deep-2
test was failing on HP-UX 11.31.
See http://thread.gmane.org/gmane.comp.gnu.coreutils.general/1748
* tests/df/total-verify: Use require_perl_, so that this test is
skipped when perl is not available.
* tests/rm/deep-2: Likewise, and fix wording in a comment.
Reported by Bruno Haible.
* bootstrap (download_po_files): Fallback to wget when downloading
the .po files via rsync failed. This is necessary to bootstrap behind
a strict firewall.
* tests/ls/slink-acl: New file.
* tests/Makefile.am (TESTS): Add it.
* tests/init.cfg (require_setfacl_): New function.
* gnulib: Update to latest, for file-has-acl changes.
* NEWS (Bug fixes): Mention it.
See http://thread.gmane.org/gmane.comp.lib.gnulib.bugs/28538. This
":>k; setfacl -m m::r k; ln -s k s; ls -Log s" should print e.g.,
-rw-r-----+ 1 0 Oct 5 19:22 s
With the ls from coreutils-8.13, it would print this (with "." or
nothing in place of the "+"):
-rw-r-----. 1 0 Oct 5 19:22 s
Interix provides faster replacements for getgr{gid,nam,ent} where
group member information is not fetched from domain controllers.
This makes 'id' usable on domain controlled interix boxes.
* m4/jm-macros.m4: Check for _nomembers functions.
* src/system.h: Redefine function to _nomembers when available.
Add a dummy, non-functional, always-successful replacement setgroups
function, to keep the original code untouched and simple.
* src/chroot.c (setgroups) [! HAVE_SETGROUPS]: Define.
This is related to commit b7f2b51c, 2010-01-01,
"ls: fix color of broken symlinks colored as target"
which didn't handle the --dereference case.
The simplest way to reproduce the resultant
erroneous "argetm" is as follows:
$ ln -s /no-such dangle
$ env LS_COLORS=ln=target ls --dereference --color
ls: cannot access dangle: No such file or directory
argetmdangle
This is also an issue with the `tree` utility,
reported here: http://bugs.debian.org/586765
* src/ls.c (print_color_indicator): Move the handling
of 'ln=target' in $LS_COLORS (color_symlink_as_referent == true)
to a higher scope, to handle all cases where type == C_LINK.
* tests/misc/ls-misc: Add a test case for the specific issue,
and 2 further test cases to verify other code paths in this area.
Reported by Jason Glassey.
At first this looked like a buffer overrun, since there was no test
to ensure that the buffer length was 6. However, since the LS_COLORS
string is NUL-terminated and since settings within it are separated by
":" there was neither the risk of reading beyond end of buffer nor risk
of a false-positive match.
* src/ls.c (print_color_indicator): Use color_symlink_as_referent
rather than manually comparing against "target" again.
* src/system.h (STRNCMP_LIT): Correct description in comment.
These commands would fail to terminate:
yes -- -nan | head -156903 | sort -g > /dev/null
echo nan > F; sort -m -g F F
That can happen with any strtold implementation that includes
uninitialized data in its return value. The problem arises in the
mergefps function when bubble-sorting the two or more lines, each
from one of the input streams being merged: compare(a,b) returns 64,
yet compare(b,a) also returns a positive value. With a broken
comparison function like that, the bubble sort never terminates.
Why do the long-double bit strings corresponding to two identical
"nan" strings not compare equal? Because some parts of the result
are uninitialized and thus depend on the state of the stack.
For more details, see http://bugs.gnu.org/9612.
* src/sort.c (nan_compare): New function.
(general_numcompare): Use it rather than bare memcmp.
Reported by Aaron Denney in http://bugs.debian.org/642557.
* NEWS (Bug fixes): Mention it.
* tests/misc/sort-NaN-infloop: New file.
* tests/Makefile.am (TESTS): Add it.
Pulling in the latest gnulib triggered a new false-positive
syntax-check failure.
* cfg.mk (exclude_file_name_regexp--sc_prohibit_always-defined_macros):
Exempt remove.c; its definitions of DT_UNKNOWN, DT_DIR and DT_LNK are
harmless.
* Makefile.am: add shortcuts to run (very) expensive tests.
Use "make check-expensive" to run tests with RUN_EXPENSIVE_TESTS=yes,
use "make check-very-expensive" to run tests with both
RUN_EXPENSIVE_TESTS=yes and RUN_VERY_EXPENSIVE_TESTS=yes.
Non-expensive tests are included in all cases.
On some systems like glibc on GNU/kFreeBSD, a thread is
implicitly created when timer_settime() is used.
This breaks our scheme to ignore signals we've
sent ourselves.
* src/timeout.c (send_sig): Change the scheme used to
ignore signals we've sent ourselves, to a more robust
but perhaps limited scheme of ignoring all signals of
a certain type after we've sent that signal to the job.
* NEWS: Mention the change in behavior.
This fixes a bug in pwd and all getcwd-using applications (for some
uses: df, readlink, stat) when run from a directory whose absolute name
contains more than PATH_MAX / 3 components. For more details, see
http://git.sv.gnu.org/cgit/gnulib.git/commit/?id=f6fe351fc534ae1
* gnulib: Update.
* NEWS (Improvements): Mention it.
* src/md5sum.c (split_3): Detect and handle BSD reversed
format checksums.
* tests/misc/md5sum-bsd: Add a new test.
* tests/Makefile.am: Reference new test.
* NEWS: Mention the improvement
Suggested by Rimas Kudelis.
* configure.ac (gl_GCC_VERSION_IFELSE): Define new macro.
(WERROR_CFLAGS): With --enable-gcc-warnings, use it to
add -Wsuggest-attribute=pure only with gcc 4.7 or newer.
* src/stat.c (USE_STATVFS): Adjust definition so that it is enabled
also on AIX 7.x systems that provide statvfs64 and no statvfs.
[USE_STATVFS && ! STAT_STATVFS && STAT_STATVFS64] (STATFS): Define
to statvfs64 in that precise case.
* m4/stat-prog.m4 (cu_PREREQ_STAT_PROG): Adjust the condition
here to match the new one in stat.c, to keep them in sync.
Reported by Bruno Haible. For details, see
http://article.gmane.org/gmane.comp.gnu.coreutils.general/1668
* tests/init.cfg (require_mkfs_PATH_): New function to test whether mkfs
is in PATH, otherwise adding /sbin to PATH. Needed for distributions
(OpenSuSE, Solaris) in which sudo does not include /sbin in PATH.
* tests/cp/cp-a-selinux: Use require_mkfs_PATH_.
* tests/cp/cp-mv-enotsup-xattr: Likewise.
* tests/cp/sparse-fiemap: Likewise.
* tests/mkdir/writable-under-readonly: Likewise.
* tests/rm/read-only: Likewise.
* NEWS (Improvements): Mention the cp-vs-NFSv4-ACL improvement we
have inherited via gnulib. For details, see
http://article.gmane.org/gmane.comp.lib.gnulib.bugs/28100
and the following messages in that thread.
* gnulib: Update to latest, for numerous ACL-related improvements.
* tests/init.sh: Relax check for diff -u support.
Rather than checking for GNU diff via --version, simply check
for support for -u itself. Useful at least on OpenBSD 4.9.
* tests/require-perl: Remove file.
* tests/Makefile.am (EXTRA_DIST): Remove it from this list.
* tests/init.cfg (require_perl_): New function.
* tests/misc/pwd-long: Use the new function, not the file.
* tests/ls/nameless-uid: Likewise.
* tests/misc/sum-sysv: Likewise.
* tests/misc/cut: Repeat each test using a multibyte locale,
if the configure-time test found such a locale.
Adjust the tests so that they also accept a slightly
different diagnostic that is specific to the MB-patched cut.
* HACKING (Add tests): Mention the variables and default values.
* README-release (Pre-release testing): Mention that setting
the SHELL variable may be required.
Suggested by Bruno Haible.
* tests/split/l-chunk: Don't use the `test "$var"`
idiom to test that var is set to something as
that's not supported by all shells.
The new style matches the usage is the rest of
the test in any case.
Reported by Bruno Haible on AIX 6.1 and 7.1
Without this change, we'd get use-of-uninit value warnings
and harder-to-diagnose failure down the road.
* tests/misc/pwd-long (normalize_to_cwd_relative): Diagnose stat
failure. This failed on AIX 6.1 and 7.1. Reported by Bruno Haible.
* tests/misc/printf-surprise: Also accept a strerror-style string
after the usual 'printf: write error:' diagnostic prefix.
Otherwise, this test would fail on HP-UX 11.
Reported by Bruno Haible.
* src/stdbuf.c (main): Pass the path of `stdbuf` rather than
the command it's running to the search function. This is
significant on platforms without /proc/self/exe
Reported by Bruno Haible
... which is not available on some platforms,
and the replacement currently requires linking
with threading libraries.
* src/split.c (closeout): Remove the call to strsignal()
which is largely redundant anyway as sig2str()
is already used to map number to name in the error.
Reported by Bruno Haible on AIX 6.1 and 7.1
Thanks to an improvement in gnulib's parse-datetime module,
commands like this now succeed (output manually indented):
$ ./date -u -d 2004-02-29T16:21:42.33+07:00 +%FT%T.%N%z
2004-02-29T09:21:42.330000000+0000
* tests/misc/date: Add a test to exercise the new-in-gnulib
parsing of ISO8601-with-"T" dates.
* NEWS (New features): Mention it.
* gnulib: Update, to pull in this parse-datetime improvement.
* gl/lib/heap.c: Move declaration to the top of scope.
Reported by Rob McMahon and Wolfgang Steinwender
in relation to "Sun WorkShop 6 update 2 C 5.3 Patch 111679-12
2003/05/18" and "GCC 2.95.3" respectively.
This effectively reverts the unreleased commit 5a647a05
* src/timeout.c (main): Don't propagate signals from the monitored
process, as on Linux /proc/sys/kernel/core_pattern could still
handle them and cause false reports against `timeout`
Lesson: do not include details like "4 million" in a file name.
* tests/rm/many-dir-entries-vs-OOM: Renamed from ...
* tests/rm/4-million-entry-dir: ...this.
* tests/Makefile.am (TESTS): Reflect renaming.
* tests/rm/4-million-entry-dir: Create only 200,000 files, rather
than 4 million. The latter was overkill, and was too likely to
fail due to inode exhaustion. Not everyone is using btrfs yet.
Now that this test doesn't take so long, label it as merely
"expensive", rather than "very expensive". Thanks to
Bernhard Voelker for pointing out the risk of inode exhaustion.
Ever since the LAST_PAGE functionality was added
in commit ed0923a1, 1996-12-05,
"Apply big patch (patch-20) from Roland Huebner"
it was ignored when -t or -T were specified or
when -l <= 10.
* src/pr.c (print_page): Increment the current page here.
(print_header): Don't increment the current page here.
* tests/pr/pr-tests: Add a test case.
* NEWS: Mention the fix
Reported at http://debbugs.gnu.org/cgi/bugreport.cgi?bug=9347
For details, see the gnulib commit,
http://git.sv.gnu.org/cgit/gnulib.git/commit/?id=47cb657e
* tests/rm/4-million-entry-dir: New test.
* tests/Makefile.am (TESTS): Add it.
* NEWS (Bug fixes): Mention it.
* gnulib: Update to latest to get the required fts fixes.
* src/head.c (elide_tail_bytes_pipe): Remove the empty statement, ";"
after a jump label; it is needed only when a declaration follows.
* src/kill.c (main): Likewise.
* src/od.c (main): Likewise.
* src/paste.c (collapse_escapes): Likewise.
* src/printf.c (print_formatted): Likewise.
This problem was discovered when trying to build git coreutils on
Solaris 8 sparcv9 with Sun C 5.8: the 3rd argument to mbrlen was
of type int * (a pointer to the gnulib replacement mbstate_t) but
the system mbrlen wants the system mbstate_t *.
* bootstrap.conf (gnulib_modules): Add mbrlen.
* src/pathchk.c (mbrlen, mbstate_t) [!(HAVE_MBRLEN && HAVE_MBSTATE_T)]:
Remove macros, which are wrong now that gnulib supplies
replacements for mbstate_t and mbrlen.
* bootstrap: Sync from gnulib. This removes an obsolescent
gettext.m4 patch, along with some other changes that do not
seem to affect coreutils.
* bootstrap.conf (gnulib_modules): Use gettext, not gettext-h.
Current gnulib gettext seems to work without needing special hacking.
* configure.ac (AM_GNU_GETTEXT_VERSION): Now 0.18.1, not 0.17.
* gnulib: Update to latest.
* bootstrap.conf (gnulib_modules): Add largefile.
This is useful to Mac OS X 10.5 users if/when configure
is generated using autoconf prior to v2.68-80-gdb2f2e0.
* gnulib: Update to latest.
* tests/du/inaccessible-cwd: Skip this test on systems like HP-UX 11.31
that lack both the *at functions and the /proc/self/fd-based support
we might have used to emulate them. Reported by Bruno Haible in
http://debbugs.gnu.org/8846
* src/join (g_names): New global (was main's "names").
(main): Update all uses of "names".
(line_no[2]): New globals.
(get_line): Increment after reading each line.
(check_order): Print the standard "file name:line_no: " prefix
as well as the offending line when reporting disorder.
Here is a sample old/new comparison:
-join: file 1 is not in sorted order
+join: in:4: is not sorted: contents-of-line-4
* tests/misc/join: Change the two affected tests to expect
the new diagnostic.
Add new tests for more coverage: mismatch in file 2,
two diagnostics, zero-length out-of-order line.
* NEWS (Improvements): Mention it.
Suggested by David Gast in http://debbugs.gnu.org/9236
This is especially important for an error-handling shell function
like this that is actually called only rarely.
* cfg.mk (sc_prohibit_framework_failure): New rule, to prevent
use of the old name.
These changes were mostly mechanical, made by running the
following command:
git grep -lw framework_failure | grep -v ChangeLog \
| xargs perl -pi -e 's/\b(framework_failure)\b/${1}_/'
and then editing init.cfg and `tests/cp/cp-a-selinux' by hand.
* tests/init.cfg (framework_failure): Remove, `framework_failure_'
from init.sh should be used instead in the tests.
Remove now-obsolete "FIXME" comment.
(is_local_dir_, require_strace_, require_membership_in_two_groups_,
require_sparse_support_, skip_if_mcstransd_is_running_,
mkfifo_or_skip_) Use `framework_failure_', not `framework_failure'.
* Many test scripts: Likewise.
* tests/init.cfg (framework_failure, getlimits_): Use 'fatal_'
instead of 'error_'.
(error_): Delete, it's not used anymore (and one is anyway
advised to use 'fatal_' instead).
Update heading comments.
* tests/shell-or-perl (error_): Renamed ...
(fatal_): ... to this, for consistency. Also, add a useful
comment.
* bootstrap.conf (gnulib_modules): Explicitly use the fclose module.
This is required, now that gnulib's close module no longer depends
on fclose. Without this module, we would be exposed to fclose bugs
affecting at least glibc 2.13 and FreeBSD.
Reported by Bernhard Voelker. Suggested by Eric Blake.
* src/copy.c (create_hard_link): A new function refactored
from existing code.
(copy_internal): Call the new function from all 3 locations
that create hard links.
* tests/cp/same-file: Amend to match the adjusted diagnostic.
* tests/cp/preserve-link: Add test cases for when a missing
link in the destination tree is encountered first and second.
Also add cases for old and new separate files in the destination
tree, both to make the clobbering behavior explicit, and to
test any changes in this area in future.
* src/copy.c (copy_internal): Adjust formatting style to conform with
guidelines in HACKING: put braces around two one-line "else" blocks.
* tests/cp/existing-perm-dir: Use $(...), not `...`, and
stat rather than ls+cut to get the mode string.
mode=$(stat --p=%A dst/dir)
* src/copy.c (copy_internal): If we don't create the directory,
then we cannot have omitted permissions. Problem and trivial
fix reported by Eric Lammerts.
* tests/Makefile.am (TESTS): Add cp/existing-perm-dir.
* tests/cp/existing-perm-dir: New file.
* src/copy.c (copy_internal): With --update (-u), this function would
return early once it found that the destination is not older than the
source, *without* recording the source-dev/ino--to--dest_name mapping.
That mapping is required in order to preserve src hard links in the
destination tree, so when using cp with --update and --preserve=links
(perhaps via -p or -a), cp could fail to preserve one hard link
per inode when at least one of the hard-linked names already exists
in the destination tree.
Reported by Odd Harry Mannsverk in http://debbugs.gnu.org/8419.
* tests/cp/preserve-link: New file. Exercise the flaw/fix.
* tests/Makefile.am (TESTS): Add it.
* NEWS (Bug fixes): Mention it.
* src/timeout.c (settimeout): A new function to convert
from a floating point duration and call alarm() or
timer_settime() if that's available.
(parse_duration): Return a double rather than unsigned int.
(usage): Mention floating point is supported.
(main): Pass the double to settimeout() rather than
calling alarm() directly with the parsed int.
(cleanup): Likewise.
* doc/coreutils.texi (timeout invocation): Say floating point timeouts
now supported, and mention the caveat with resolution.
* bootstrap.conf: Include the timer-time gnulib module.
* tests/misc/timeout-parameters: Add a test with nanoseconds.
* NEWS: Mention the improvement.
This change derives from improvements to gnulib's fsusage module.
* NEWS (Improvements): df now supports disk partitions larger than
4 TiB on MacOS X 10.5 or newer and on AIX 5.2 or newer.
Alphabetize entries.
* gnulib: Update to latest.
* m4/jm-macros.m4 (coreutils_MACROS): Use fdatasync only if declared.
MacOS X 10.7 has an fdatasync that is not declared, and is rumored to
be ineffective. (Bug#9141)
The following dropped the space from the first field
printf "1234567 \t1\n" | unexpand -a
Note POSIX says that spaces should not precede tabs.
Also a single trailing space should not be converted
if the next field starts with non blank characters.
So we enforce those rules too, with this change.
* src/unexpand.c (unexpand): Implement as per POSIX rules.
* tests/misc/unexpand: Add tests, and adjust existing
tests as per POSIX rules.
* NEWS: Mention the fix.
Reported by Hallvard B Furuseth
Treat fractions as a request to round up to the next representable
value, and treat out-of-range values as maximal ones. This is
consistent with how "sleep" works. And this way, "timeout
999999999999999999d FOO" and "timeout 4.5 foo" are more likely to
do what the user wants.
* src/timeout.c: Include c-strtod.h and xstrtod.h, not xstrtol.h.
(apply_time_suffix): Change it to the way sleep.c's time_suffix
does things. Maybe this function (identical in both programs,
other than its name) should be moved to a library?
(parse_duration): Return a maximal value on overflow. Return
unsigned int, not unsigned long. Allow fractions, which round
up to the next integer value.
* tests/misc/timeout-parameters: Adjust tests to match new behavior.
Add a very large number.
* src/csplit.c (interrupt_handler): Reset signal to SIG_DFL
after deleting the files, so that a second interrupt won't
prematurely terminate cleanup.
(main): Don't use SA_NODEFER | SA_RESETHAND, as that might
allow premature termination of cleanup. Also, this ports better
to platforms like NonStop, which don't ahve SA_RESETHAND.
* src/stat.c (print_stat): Use ST_BLKSIZE() rather than
accessing st_blksize directly, which is not present on
NonStop at least. Reported by Joachim Schmitz.
* src/fiemap.h (struct fiemap): Adjust the previous change
to the fiemap_extents array, which would also require changes
to the sizeof calculations in extent_scan_read().
Instead, only declare the fiemap_extents zero length array
on linux, which is the only platform that references this member.
This avoids a compilation failure on systems that don't support
this non standard construct. We don't use the equivalent C99
flexible array construct so as to have maximum portability.
* src/extent-scan.c: Cleanup. Remove a redundant #ifndef.
* src/mktemp.c: maint: avoid warning by using the comma operator
rather than an always-true conditional (as suggested by Eric Blake).
Reported by Joachim Schmitz in http://debbugs.gnu.org/9064.
This information has already been added to the Texinfo manual, but was
missing from the --help output.
* src/wc.c (usage): As above, for --help.
Reported by Vincent Lefevre in http://bugs.debian.org/395430.
* cfg.mk (sc_prohibit_man_see_also_period): Prohibit a period at
the end of the first line after a "SEE ALSO" marker in man/*.x.
With this, we shouldn't have to make any more changes like those
in today's commit, f2dabd68.
man-pages(7) explicitly says about SEE ALSO sections: "Do not terminate
this with a period." This is also in line with all the other man pages
in coreutils.
* man/cat.x: Remove period at the end of the SEE ALSO section.
* man/tac.x: Likewise.
* m4/jm-macros.m4: Define HAVE_SETRLIMIT.
* src/timeout.c: If the child exited with a signal,
raise that signal to the timeout process itself,
so that callers may also see the signal status.
Use setrlimit to disable core dumps for the timeout
process, which would be generated by some signals.
* src/timeout.c (cleanup): Send signals directly to the child
in case it has started its own process group (like a cascaded
timeout command would for example).
* test/misc/timeout-group: Add a test case.
* NEWS: Mention the fix.
Or more accurately, commands not started from the shell prompt,
that are interactive, or need to receive Ctrl-C etc. from the terminal.
* doc/coreutils.texi (timeout invocation): Document --foreground.
* src/timeout.c (main): Set the foreground flag and don't create
a separate group.
(cleanup): Only send a signal directly to the monitored command
when the foreground flag is set.
(usage): Describe --foreground.
* tests/misc/timeout-group: Add a new test.
* tests/Makefile.am: Reference new test.
NEWS: Mention the new option.
Reported by Shay Shimony
Analysis by Alan Curry
Fix suggested by Paul Eggert
This fact was already noted in the Texinfo manual, but not in the
output of --help.
* src/cp.c (usage): As above, for --help.
Reported by Jari Aalto in http://bugs.debian.org/294327.
Use this new option with --check when the input is expected to
consist solely of checksum lines. With only --check, an invalid
line evokes a warning, but the program can still exit successfully.
With --strict, any invalid line makes the program exit non-zero.
* src/md5sum.c (strict, STRICT_OPTION): Declare/define.
(long_options): Add "strict".
(usage): Describe --strict.
(digest_check): Count improperly_formatted lines, too, and use
that number and the global "strict" to determine the return value.
(main): Handle STRICT_OPTION.
Reject --strict without --check.
* doc/coreutils.texi: Describe it.
* NEWS (New features): Mention it.
* src/date.c (usage): As above, for --help.
* doc/coreutils.texi (Time conversion specifiers): Likewise.
Reported by Britton Leo Kerin in http://bugs.debian.org/115833.
* bootstrap.conf (gnulib_modules): Remove pathmax.
* src/system.h: Don't include "pathmax.h".
(PATH_MAX) [!PATH_MAX]: Define to 8192. Defining it to a constant
is preferable to using a definition from pathmax.h that might expand
to pathconf ("/", _PC_PATH_MAX). Prompted by discussion leading to:
http://thread.gmane.org/gmane.comp.lib.gnulib.bugs/27183/focus=27269
* tests/ls/stat-free-color: The system may perform additional stat
calls upon loading (seen on OpenSuSE-11.4). Count only the number
of stat calls compared to --help.
This also reduces back to "1" the number of expected calls,
effectively reverting part of 2011-06-01 commit, ccf2d9a4.
* tests/init.sh (warn_): Use "sed 1q" in place of "head -1".
The latter is officially obsolete but more portable than "head -n1".
Reported by Bernhard Voelker.
* tests/shell-or-perl: Prefer the `read' builtin over `grep' to
look at the shebang line of test scripts. Since `read' is a
special builtin, it might abort the whole program upon failures,
so add extra sanity checks, verifying that the test script exists
and is readable, before trying to read from it.
This change implements a more correct and idiomatic use of the
features of the Automake-provided 'parallel-tests' harness.
Moreover, this change is required in order for the testsuite to
continue to work with the new testsuite harness that is planned
to be introduced in Automake 1.12 (which, as of the writing date,
is still under development and in alpha state).
* tests/shell-or-perl: New auxiliary script.
* tests/Makefile.am (EXTRA_DIST): Distribute it.
* tests/check.mk (TESTS_ENVIRONMENT): Remove definition of the
`shell_or_perl_' shell function, whose code has been moved in
the new script above (with a few improvements and extensions).
Do not use it to run the test scripts.
(LOG_COMPILER): New, properly invoking `shell-or-perl'.
* src/Makefile.am (pkglibexec_PROGRAMS): Rename from pkglib_PROGRAMS.
The latter is invalid. Without this change, automake
v1.11-373-g9ca6326 and newer (on master) would fail with this:
`pkglibdir' is not a legitimate directory for `PROGRAMS'
This changes the default installation directory of libstdbuf.so from
$prefix/lib/coreutils/ to
$prefix/libexec/coreutils/
* src/stdbuf.c (set_LD_PRELOAD): Search in PKGLIBEXECDIR, not PKGLIBDIR,
since that's where we install libstdbuf.so.
Do not search in "", the system default search path.
Ensure that English diagnostics are emitted even when using
French sorting rules.
* tests/misc/sort-debug-keys: Unset LC_ALL and set LC_COLLATE,
LC_CTYPE and LC_NUMERIC to the fr_FR.UTF-8 locale, while setting
LC_MESSAGES=C. Reported by Stefano Lattarini.
Using a .diff is much more maintainable. Otherwise, changes in
the gnulib module description file may not be noticed and merged
promptly and may even result in subtle errors. Luckily, this time,
the failure to propagate gnulib's changes to modules/tempname resulted
only in an obvious link failure.
* gl/modules/tempname: Remove file.
* gl/modules/tempname.diff: Use a .diff file instead.
* gnulib: Update submodule to latest.
* tests/init.cfg (skip_test_): Remove function.
Use skip_ in place of skip_test_ everywhere else.
* cfg.mk (sc_prohibit_skip_): Remove rule.
* tests/**: Use skip_, not skip_test_, everywhere.
* tests/init.sh (warn_): Use printf, not echo. The latter would
misbehave when given strings containing a backslash or starting
with e.g., -n. James Youngman suggested setting IFS.
Running "make check" normally prints a diagnostic to the outermost
stderr (usually a tty) to explain why a test is skipped. It did this
by redirecting FD 9 to stderr (via "exec 9>&2") before invoking the
shell script. Shell scripts write skip-explanation to FD 9 via
init.sh's skip_ function. However, with ksh and HP-UX's /bin/sh,
the effects of "exec 9>&2" are canceled upon fork-and-exec, so we
would get a "Bad file number" diagnostic and no skip explanation on
those systems.
* tests/check.mk (TESTS_ENVIRONMENT): Redirect more portably, via
"$(SHELL) 9>&2", rather than the prior "exec 9>&2; $(SHELL) ..."
Actually, we use "shell_or_perl_ 9>&2", to make this effective
also for the perl-based tests.
* tests/init.sh (stderr_fileno_): Update the advice in comments.
See http://thread.gmane.org/gmane.comp.gnu.coreutils.bugs/22488
for lots of discussion. Stefano Lattarini suggested the solution
of putting "9>&2" after the command. Reported by Bruno Haible.
* gnulib: Update to latest.
* src/system.h: Definitions of ST_* macros have moved into the
gnulib module stat-size (specifically, the header file
stat-size.h), so remove them from here.
* src/truncate.c: Include stat-size.h.
* src/stat.c: Likewise.
* src/shred.c: Likewise.
* src/ls.c: Likewise.
* src/du.c: Likewise.
* src/ioblksize.h: New file. Move definition of io_blksize out of
system.h so that system.h does not have to include stat-size.h.
* src/cat.c: Include ioblksize.h.
* src/split.c: Likewise.
* src/copy.c: Include both stat-size.h and ioblksize.h.
* src/Makefile.am (noinst_HEADERS): Add ioblksize.h.
* tests/dd/nocache: Relax the test, as the system
may return various errors from posix_fadvise().
HPUX 11.31 returns ENOTTY for example.
Reported by Bruno Haible
* src/date.c (usage): Add examples for TZ handling,
and "seconds since epoch" parsing, neither of which
was mentioned in the man page until now.
* THANKS.in: Add Rick.
Suggested by Rick Stanley.
* configure.ac: Require autoconf-2.64, which is nearly two years old.
* src/system.h (emit_ancillary_info): Use PACKAGE_URL, now that we
require autoconf-2.64.
* tests/ls/stat-free-color: This test recently began to fail on
rawhide because dynamic library start-up code now stats "/selinux",
making the total number of calls 2 rather than the prior 1.
Create two more dangling symlinks, so that any erroneous stat-
or lstat-calling code will get at least those three.
I think it would be better to exit through the error() and not
to call the exit() after the error(). This way we can get rid of
one function call (and curly brackets).
* src/yes.c (main): Exit through the error(), remove exit() call
after error().
* src/tail.c (start_bytes): Increase *READ_POS (not READ_POS)
by the number of bytes read. This is a real bug that happens
to have no consequence in practice. First, this code is exercised
only when tailing-forever a non-regular file by bytes, and with a
start-relative offset, e.g., "mkfifo f; tail -f -c +3 f", but even
then, the invalid READ_POS value does not influence how tail works.
It is stored in the File_spec.size member, but that member is not
used at all in tail_forever_inotify, and in tail_forever, it is
used only when the File_spec refers to a regular file.
* src/chown-core.c (describe_change): Output the
original owner if possible.
(user_group_str): Handle the case when neither
owner or group are passed.
* NEWS: Mention the change in behavior.
* src/chown_core.c (describe_change): Accept the ownership of
the original file and output that when not changing.
This is significant when --from is specified as then
the original and specified ownership may be different.
(user_group_str): A new helper function refactored from
describe_change().
(change_file_owner): Pass the original user and group
strings to describe_change().
* test/chown/basic: Add a test case.
* NEWS: Mention the fix.
* configure.ac (WARN_CFLAGS): Don't turn off -Wstrict-overflow.
(GNULIB_WARN_CFLAGS): Remove -Wstrict-overflow from the list of
warning options used in lib/.
Normally I find that -Wstrict-overflow produces too many false
positives, but considering that it warns of the bug reported in
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=33498, I now think
it is worthwhile. The lesser of two evils.
Thanks to Daniel Veillard for showing me the gcc bug report.
* src/factor.c (factor_using_pollard_rho): Change type of "i"
to unsigned to avoid warning from gcc's -Wstrict-overflow.
* src/expr.c: Use an unsigned intermediate.
* src/dircolors.c (main): Reorder operations to avoid the risk of
pointer overflow.
* src/tr.c (squeeze_filter): Change NOT_A_CHAR from an anonymous
"enum" to an "int", to avoid this warning:
tr.c:1624:10: error: assuming signed overflow does not occur when
simplifying conditional to constant [-Werror=strict-overflow]
* src/pr.c (main): Make index "i" unsigned.
* cfg.mk (sc_preprocessor_indentation): New test, from libvirt.
Exempt 3 files from new cppi test.
* gl/lib/randread.c: Adjust cpp indentation to comply.
* src/extent-scan.c (extent_need_sync): Likewise.
* src/shred.c (incname): Add an assertion to tell static analyzers
that we know this particular use of strchr never returns NULL.
Finish incomplete sentence in function-describing comment.
* src/split.c (lines_chunk_split): Don't use ignore_error() which
is redundant and confusing when not running with --filter.
(lines_rr): Likewise.
(ofile_open): Likewise. Add a comment to clarify that
filters aren't restarted under file descriptor pressure.
* src/split.c (main): Exit with a diagnostic if --filter
is specified along with a specific chunk number.
* test/split/filter: Ensure this combination fails.
* src/split.c (bytes_split): Stop reading when we
can no longer write to a child process.
(lines_rr): Likewise.
(lines_bytes_split): No change is made here since
input is bounded by the original file size.
* test/split/filter: Add test cases.
src/split.c (main): Don't unblock SIGPIPE before cleanup,
as then any pending signals will be sent and cause
the main split process to exit with a non zero status (141).
* test/split/filter: Add a test for this case.
* src/split.c (lines_bytes_chunk): Handle the edge case
where the file is truncated as we read.
* tests/misc/split-lchunk: Cleanup; no functional change.
* src/chmod.c (describe_change): Pass in the original mode,
and output this in the messages.
* tests/chmod/c-option: Adjust as per the new message.
* THANKS.in: Remove the now auto-generated name.
* NEWS: Mention the change in behavior.
Signed-off-by: Pádraig Brady <P@draigBrady.com>
* cfg.mk: Include via "-include", to accommodate new tight-scope rule.
(sc_check-AUTHORS): Change the name of the rule in src/Makefile.am
to _sc_check-AUTHORS, so it doesn't conflict with this one when
this file is included into the sub-make's context.
* src/Makefile.am (_sc_check-AUTHORS): Rename from sc_check-AUTHORS.
* gnulib: Update to latest.
* src/touch.c (main): Avoid even the hint of possibility that
we'd dereference NULL upon localtime failure. Coverity reported
the potential, but it appears not to be possible, since posixtime
rejects any time for which the subsequent localtime would return NULL.
See http://thread.gmane.org/gmane.comp.gnu.coreutils.general/1253
* src/split.c (lines_chunk_split): Ensure that data is only
written to stdout when k specified. Also ensure that
extra files are not created when there is more data available
than reported in the file size.
* tests/misc/split-lchunk: Verify that split -n l/k/n doesn't
generate any files, and that -n l/n always generates n files.
* NEWS: Mention the fix.
* src/timeout.c (usage): Add a space to be consistent
with other uses of "(the default)" in the documentation.
Signed-off-by: Pádraig Brady <P@draigBrady.com>
* tests/misc/tac-continue: Fix typo in usually-skipped test:
s/mkfifo_or_skip/mkfifo_or_skip_/ (i.e., append "_").
This test is usually skipped, because I'm probably the only
one to set the FULL_PARTITION_TMPDIR envvar, and recently the
one I'd been using ceased to exist, so this test was skipped
even for me. Good argument for making this a root-only test
and creating a full partition just for this test case.
* NEWS: "an misleading"
* src/expr.c: "a integer
* src/ptx.c (find_occurs_in_text): "a end"
* src/shred.c (do_wipefd): "a infinite"
* src/sort.c (SUBTHREAD_LINES_HEURISTIC): "an dual-core"
(compare_random): "an checksum"
* cfg.mk (old_NEWS_hash): Update, since the typo was in old news.
* src/printf.c (STRTOX): Don't access memory after a
string containing a single quote character.
* tests/misc/printf: Add tests for various combinations
of single quote characters combined with a numeric format.
* THANKS.in: Add bug reporter.
* NEWS: Mention the fix.
Reported-by: Paul Marinescu <paul.marinescu@imperial.ac.uk>
* gl/lib/randperm.c (randperm_new): When the number of items
to return H, is much smaller than the total number of items N,
use a hash to represent the sparse permutations of the set N.
This is currently enabled for N > 128K and N/H > 32.
* tests/misc/shuf: Ensure shuf can quickly return 2 numbers
from a large range.
* gl/modules/randperm: Depend on hash.
* NEWS: Mention the change.
skip_test_ emits its diagnostic both to FD 9 (tty), and to
FD 2 (usually the log file), whereas init.sh's skip_ emits
only to FD 9. Without that, the log is slightly less useful.
* tests/cp/fiemap-2: Use skip_test_, not skip_.
* tests/cp/fiemap-perf: Likewise.
* tests/du/bigtime: Likewise.
* tests/du/files0-from-dir: Likewise.
* tests/du/move-dir-while-traversing: Likewise.
* tests/init.sh: Likewise.
* tests/misc/sort-stale-thread-mem: Likewise.
* tests/misc/stat-nanoseconds: Likewise.
* tests/mv/i-3: Likewise.
* tests/mv/sticky-to-xpart: Likewise.
* tests/split/filter: Likewise.
Prompted by a report from Pádraig Brady.
Even on a system with d_type support, the default use of --color
makes ls stat every file in order to be able to honor settings like
EXEC, STICKY, ORPHAN, SETUID, etc., because those settings require
information that is not provided by dirent.d_type. However, if
for a potentially large performance gain, you are willing to disable
those settings, you can now make ls --color give type-related coloring
and perform no stat calls at all (other than the unavoidable call-per-
command-line argument). Before this change, even with all of those
attributes disabled, ls --color would still stat every directory.
Now, we're down to the minimum of one stat call per command-line arg.
* src/ls.c (gobble_file): With --color, don't stat a
non-command-line-specified directory when no directory-coloring
attribute is enabled.
* tests/init.cfg (require_dirent_d_type_): New function.
* tests/d_type-check: New script, mostly from Pádraig Brady.
* tests/Makefile.am (EXTRA_DIST): Add it.
* tests/ls/stat-free-color: New test.
* tests/Makefile.am (TESTS): Add it.
* doc/coreutils.texi (General output formatting): Describe how
to use dircolors to make ls --color refrain from calling stat
on a d_type-enabled file system.
Prompted by a query from Josef Bacik.
gnulib's group-member module now ensures that the group_member
function is declared in <unistd.h>, just like it is glibc.
* lib/euidaccess-stat.c: Remove inclusion of "group-member.h".
* src/chgrp.c: Likewise.
* configure.ac: Rather than disabling -Wmissing-field-initializers,
use the fact that gnulib now disables it automatically when required
(on versions of GCC older than 4.7).
* src/system.h: Remove the no longer needed DECLARE_ZEROED_AGGREGATE.
* src/ls.c: Likewise.
* src/pathchk.c: Likewise.
* src/shred.c: Likewise.
* src/stty.c: Likewise.
* src/wc.c: Likewise.
* src/split.c: Include <signal.h>, <sys/wait.h> and "sig2str.h".
(FILTER_OPTION): New anonymous enum member.
(filter_command, filter_pid): New globals.
(open_pipes, open_pipes_alloc, n_open_pipes): Likewise.
(oldblocked, newblocked): Likewise.
(longopts): Add "filter".
(usage): Document --filter.
(create): Extend to create a pipe and fork "sh -c CMD".
(closeout): Adapt to close a pipe and wait for child process.
(cwrite): Call closeout, not just close.
(lines_chunk_split): FIXME
(bytes_chunk_extract): FIXME
(opid, ofile_open, lines_rr, main): FIXME
(ignorable): New function, to encapsulate EPIPE test.
* src/sort.c (key_warn): `sort -k2,1n --debug` would output
warnings about being both "zero width" and "spanning multiple fields".
Suppress the latter one.
* tests/misc/sort-debug-warn: Add a couple of test cases.
* src/copy.c (copy_reg): Fix my typo (mis-applied patch).
The patch by Jeff Liu was fine, but I mis-applied it
and introduced a compilation error in commit efa479c1.
2011-05-03 Jim Meyering <meyering@redhat.com>
Running "GZIP=-9 ./bootstrap" would fail right away, because the
tool-version-checking code would treat the upper-cased program name
as an environment variable name and if that has a value use the
result as the application name. That works fine for automake,
autoconf, etc. but not for gzip.
* bootstrap (check_versions): Do not treat $GZIP as a program name.
If defined at all, it is supposed to list gzip options.
Reported by Alan Curry in http://debbugs.gnu.org/8609
* tests/Makefile.am (TESTS): Remove misc/pwd-unreadable-parent.
This test was misleading and useless (was always skipped).
Inspired by a report from Bruno Haible: http://debbugs.gnu.org/8570
* tests/misc/pwd-unreadable-parent: Remove file.
* tests/CuSkip.pm (skip): New file/module/function, to help
the perl test scripts "skip" a test consistently, emitting
a diagnostic both into the log file and into the outermost
stderr stream that is more likely to be seen by a human.
* tests/check.mk (TESTS_ENVIRONMENT): Add -MCuSkip.
* tests/misc/date-next-dow: Use CuSkip::skip in place of warn+exit-77.
* tests/misc/tty-eof: Likewise.
* tests/misc/uniq: Likewise.
* tests/rm/fail-eperm: Likewise.
* tests/misc/md5sum-newline: Likewise. Also, s/program_name/ME/.
* tests/misc/ls-misc (setuid_setup, main): Likewise.
* tests/misc/pwd-long: Likewise, and add -I"$abs_srcdir" -MCuSkip
to the $PERL invocation command.
Inspired by a request from Bruno Haible regarding misc/tty-eof:
http://debbugs.gnu.org/8570
* src/dd.c (O_NOCACHE): Undefine. This symbol is defined
via AIX's <fcntl.h>, yet used as an enum name in dd.c.
Reported by Gary V. Vaughan in http://debbugs.gnu.org/8555
* NEWS (Portability): Mention this.
* src/tail.c (tail_forever_inotify): Just as without inotify,
tail --follow=name now terminates when the last tailed-by-name file
is unlinked or moved aside. This bug was introduced on 2009-06-15
via commit ae494d4b, "tail: use inotify if it is available".
Reported by Tim Underwood in
http://thread.gmane.org/gmane.comp.gnu.coreutils.bugs/22286
* NEWS (Bug fixes): Mention this.
* tests/tail-2/follow-name: Test for this.
* tests/Makefile.am (TESTS): Add it.
* src/copy.c (copy_reg): Upon btrfs clone failure, print not just
the destination file name, but also the source file name.
That may be useful upon failure of a cross-device clone attempt.
Otherwise, this would fail (albeit rarely) on a "make -j24 check" run.
* tests/tail-2/pipe-f2: Increase timeout from 1 second to 10,
to avoid false positive failure.
* bootstrap.conf (gnulib_modules): Include full-read and full-write
explicitly. Before, we'd get them via safe-read, but with newer
gnulib, that is no longer enough: link failure due to undefined
references to full_write.
* tests/cp/sparse-fiemap: Parenthesize ternary expression used
as an argument to awk's printf. Otherwise, gawk 3.0.1 and the
one from debian stable's original-awk would get a syntax error.
Reported by Dennis Clarke.
Copyright note: tiny change
* tests/cp/sparse-fiemap: When this test was run as root on an ext3
file system, (ext3 had known problems), it would trickily create and
mount a loopback ext4 file system and use that instead. However, due
to a bug in 2.6.39-rc1..rc3, this loopback test (when run in another
loopback FS) exposed a bug with 1k-blocksize ext4 whereby non-NUL
data would be read from a hole. For details, see this:
http://thread.gmane.org/gmane.comp.file-systems.ext4/24495
to avoid the expense of extent_copy's unconditional use of
FIEMAP_FLAG_SYNC.
* src/copy.c (copy_reg): Do not attempt extent_copy on a file
that appears to have no holes.
* NEWS (Changes in behavior): Document this. At first I labeled this
as a bug fix, but that would be inaccurate, considering there is no
documentation of FIEMAP semantics, nor even consensus among kernel
FS developers. Here's hoping SEEK_HOLE/SEEK_DATA support will soon
make it into the linux kernel.
* src/copy.c (HAVE_STRUCT_STAT_ST_BLOCKS): Define to 0 if undefined,
so we can use it in the return expression, here:
(is_probably_sparse): New function, factored out of...
(copy_reg): ...here. Use the new function.
* src/copy.c (extent_copy): Do not treat "unwritten extents" specially.
Otherwise, with a release-candidate 2.6.39-rc3 kernel, XFS or ext4,
when using gold as your linker, and if you forget to run "make check",
you could end up installing files full of zeros instead of the expected
binaries. For a lot of discussion, see
http://thread.gmane.org/gmane.comp.file-systems.xfs.general/37895
* tests/cp/fiemap-empty: Disable this test.
* src/extent-scan.c (extent_need_sync): Always return true,
to make the sole caller always use FIEMAP_FLAG_SYNC.
This will doubtless have an undesirable performance impact,
but we'll mitigate that shortly, by using extent_copy only on
files with holes.
* tests/filefrag-extent-compare: Don't check the length of the
last extent, as this was seen to vary on XFS, where it leaves
trailing blocks allocated for performance reasons.
* tests/cp/fiemap-empty: Though not seen as an issue in practise,
try to avoid possible issues with the allocator in file systems,
by requesting to allocate a power of 2.
* NEWS: Slightly obfuscate a line to avoid a false-positive
doubled-word ("is-is") match.
Fix a grammar error in news for 8.2.
* cfg.mk (old_NEWS_hash): Resync.
and to avoid a false-positive "TO to" in new doubled word check.
* src/install.c (change_timestamps): Rename parameters for
readability. Make the comment match the code.
* extent-scan.h (extent_scan_free): Init the pointer to NULL,
and reset the count to 0, so that we can realloc the buffer.
* src/extent-scan.c (extent_scan_init): Likewise.
(extent_scan_read): Loop over multiple fiemap scans, so we handle
mergeable extents that span across fiemap scan boundaries. Once
we have enough unique extents, return so as to minimize memory use.
Running the new fiemap-empty test uses 600MB of disk space via
fallocate, and in so doing caused failure in unrelated tests that
were running in parallel on a small file system. Rather than
simply running fallocate (which allocates the space, inducing
disk full when it fails), skip the test if there is less than
800MB of free space, as computed via stat and awk.
* tests/init.cfg (require_file_system_bytes_free_): New function.
* tests/cp/fiemap-empty: Use it.
* tests/misc/help-version: Sleep only ~30s, not 10m.
The latter was a problem when somehow that sleep process would
hang around and thereby prevent (for up to 10m) a normal unmount
of the temporary partition in which I'd run the tests.
* tests/cp/preserve-gid: Simply chmod a+rx instead.
That is safer, in case the nameless UID actually has an account,
and might take advantage of root running a program in a directory
under its control
* src/Makefile.am (sc_tight_scope): Adjust rule to use an eval-based
trap-setting for-loop rather than 4x hard-coded 128+N constants.
Also catch SIGQUIT (3). Tweak comments.
* src/extent-scan.c (extent_need_sync): Require sync also for 2.6.38.
Without this, part of the cp/fiemap-empty test would fail both on
F15-to-be (2.6.38.1-6.fc15.x86_64) and rawhide. For details, see
http://thread.gmane.org/gmane.comp.gnu.coreutils.bugs/22190
When running the erroneous command, cp -rl A D D, and depending on the
structure of directories A and D and the file system type (because that
changes order of dir. entry traversal), cp would sometimes fail to
detect that D was being copied into D, and would create D/D/D/D/D/...
until it hit PATH_MAX or exhausted some resource.
I noticed this via the occasional failure of the cp/into-self test
when run using a ZFS file system. It is occasional because the bug
is dependent on the order in which directory entries are traversed,
and that is apparently indeterminate with ZFS.
Technically, with the current recursive implementation, there is no
risk of an infinite loop, due to stack limitations, but with an
eventual fts-based implementation, it might have iterated until
disk space or inodes are exhausted.
* src/copy.c (copy_dir): Avoid copy-into-self interminable loop on
systems with large PATH_MAX. On other systems, diagnose the copy-into-
self error consistently. Handle the parameter,
first_dir_created_per_command_line_arg, correctly when there are two
or more sub-directories.
* src/copy.c (extent_copy): Treat an allocated but empty extent
much like a hole. I.E. don't read data we know is going to be NUL.
Also we convert the empty extent to a hole only when SPARSE_ALWAYS
so that the source and dest have the same allocation. This will
be improved soon, when we use fallocate() to do the allocation.
* tests/cp/fiemap-empty: A new test for efficiency and correctness
of copying empty extents.
* tests/Makefile.am: Reference the new test.
* NEWS: Mention the change in behavior.
* src/extent-scan.c (extent_scan_read): Add a more stringent check
for OFF_T overflow, to ensure subsequent code is immune.
Detect overlapping extents and adjust, so as files always copied.
Detection using a single scan with fallback to a standard copy
was thought too expensive in memory or time.
* NEWS: Mention the fix
This bug was introduced in commit ca9e212c, 2009-09-24,
"cp, mv: use linkat to guarantee semantics", which
inadvertently disabled the creation of hardlinks to symlinks.
However rather than implementing the intention of that commit
and relying on gnulib linkat emulation, we'll revert to the
previous emulation as that maintains ownership and timestamps.
* src/copy.c (copy_internal): Use our existing hardlink to
symlink emulation when link() might dereference the symlink.
Also ensure that we copy the timestamps of the original symlink
when we use the emulation.
* tests/cp/link-symlink: Add a test to ensure timestamps copied.
* tests/Makefile.am: Reference the new test.
* NEWS: Mention the fix.
Reported by Ruediger Meier
* tests/tail-2/inotify-rotate: Wait 50% longer for grep to succeed.
Without this change, this test would fail consistently when using
"make -j25 check" with F15 in a virtio- and spinning-rust-backed
virtual machine.
* src/extent-scan.h (struct extent_scan): Add the fm_flags member to
pass to the fiemap scan.
* src/extent-scan.c (extent_need_sync): A new function used to
detect Linux kernels before 2.6.38.
(extent_scan_init): Add FIEMAP_FLAG_SYNC when needed.
* tests/cp/sparse-fiemap: Adjust comment.
* NEWS: Mention the change in behavior.
Indirectly suggested by Mike Frysinger
* tests/init.cfg (require_selinux_): Skip the test also when
/proc/filesystems does not list selinuxfs.
Add comments.
* cfg.mk (exclude_file_name_regexp--sc_file_system): Exempt
tests/init.cfg, with its use of /proc/filesystems.
Based on the patch by Mathieu Bridon in http://debbugs.gnu.org/8359.
More discussion in http://bugzilla.redhat.com/573111
* src/find-mount-point.h: Move "*" to where it belongs.
Move "const", too.
* src/find-mount-point.c: Move "const" to conform.
* src/Makefile.am (sc_tight_scope): Allow `*'s before the function name.
Use perl's -l option and drop the \n after (and quotes around) $1.
* src/df.c (alloc_table_row): A new function to allocate storage
for a row of strings.
(print_table): A new function to interate over all stored strings in
the table, and apply alignment honoring the max width of each column.
(get_header): Renamed from print_header, and adjusted accordingly.
(get_dev): Renamed from show_dev. Also we no longer wrap longer
device names over two lines, which can be an unexpected issue for
scripts parsing the output from df.
(get_disk): s/show_/get_/
(get_point): Likewise.
(get_entry): Likewise.
(get_all_entries): Likewise.
* NEWS: Mention the change.
Make GNU coreutils' test recognize "==" as a synonym for "=".
This is already the case in GNU coreutils' expr, bash, ksh,
busybox ash, FreeBSD-current /bin/sh and /bin/test, and
OpenBSD's /bin/sh.
Before, env test a '==' a would fail with this diagnostic:
"test: ==: binary operator expected". Now, it succeeds.
* src/test.c: Accept "==" as a synonym for "=".
* doc/coreutils.texi (String tests): Document it.
Reported as http://debbugs.gnu.org/8263
Also see http://austingroupbugs.net/view.php?id=375
* tests/cp/preserve-gid: Ensure that every process under test uses
the cp binary we've just built. Before this fix, with a restrictive
umask or build-dir permissions, the UID-changing tests would end up
using whatever cp happened to be available through $PATH
Analysis by arbogast.cedric@gmail.com in http://debbugs.gnu.org/8292.
The names in THANKS are generated from two sources: the hard-coded
list, THANKS.in, and the names of committers from the git log.
When a contributor on the hard-coded list commits a change,
we remove their now-redundant name from THANKS.in.
* THANKS.in: Remove a now-duplicate name.
* tests/filefrag-extent-compare: Merge adjacent extents in
each list before processing, so we correctly account for
split extents in either list.
* tests/cp/sparse-fiemap: Remove the explicit syncing,
which was only changing the way extents were arranged,
and thus working around the extent comparison issue
that was seen on ext4 loop back.
* gl/lib/mbsalign.c (rpl_wcswidth): Remove this in favor
of the equivalent wcswidth replacement in gnulib.
* bootstrap.conf: Depend on the wcswidth module.
Suggested by Bruno Haible.
* gnulib: Update to latest, to address http://debbugs.gnu.org/8230.
When built on Solaris 9 and run on Solaris 10, touch would segfault.
Reported by Ben Walton.
* bootstrap: Update from gnulib.
* tests/init.sh: Likewise.
* NEWS (Bug fixes): Mention this.
* src/sort.c (SUBTHREAD_LINES_HEURISTIC): Do not spawn a new thread
for every 4 lines. Increase this from 4 to 128K. 128K lines seems
appropriate for a 5-year-old dual-core laptop, but it is too low for
some common combinations of short lines and/or newer systems.
* NEWS (Bug fixes): Mention it.
* src/extent-scan.c (extent_scan_read): Merge adjacent extents
that vary only in size, so that we may process them more efficiently.
This will be especially useful when we introduce fallocate()
so that we don't reproduce fragmentation in the destination.
* src/dd.c (warn_partial_read): New static var.
(iread): Diagnose partial reads if needed.
(iwrite): Don't diagnose them here; not needed any more.
(scanargs): Determine whether partial reads should be diagnosted.
* src/dd.c (FFS_MASK): A new macro (Find First Set) refactored
from the following enum as it's now used twice.
(usage): Mention the new 'nocache' flag.
(cache_round): A new function to help ignore requests
to drop cache, that are less than page_size.
(invalidate_cache): A new function to call posix_fadvise()
with the appropriate offset and length. Note we don't
use fdadvise() so we can detect errors when count=0.
(dd_copy): Call invalidate_cache() for the portions read.
(iwrite): Likewise for the portions written.
(main): Call invalidate_cache for page_size slop or
for full file when count=0.
* cfg.mk (sc_dd_O_FLAGS): Adjust to pass.
* doc/coreutils.texi (dd invocation): Describe the 'nocache' flag,
and give some examples of how it can be used.
* tests/dd/nocache: A new test.
* tests/Makefile.am: Reference the new test.
* NEWS: Mention the new feature.
* doc/coreutils.texi (dd invocation): Clarify that bs= can
cause parital reads to be immediately written to output.
* src/dd.c (usage): Hint that bs= can cause partial writes.
See http://debbugs.gnu.org/cgi/bugreport.cgi?bug=8171
* src/dd.c (usage): Move 'sync' up with other data transformation
options. Having it alongside 'fsync' and 'fdatasync' is
particularly confusing. Also the double line description of
the 'sync' option, serves as a visual break from the "flag"
type options that follow.
* doc/coreutils.texi (dd invocation): Apply the same grouping
as above, by splitting the "conv=" table in two.
An alternative to this is to auto enable iflag=fullblock
when oflag=direct and bs= is specified.
It was thought better though, to warn about the specific issue,
and give full control of dd's options to the user.
* src/dd.c (iwrite): Warn, when we write after having
disabled O_DIRECT.
See https://bugzilla.redhat.com/show_bug.cgi?id=614605
* src/du.c (main): Fail on AI_ERR_READ error, rather than merely
diagnosing and continuing. Based on a patch by Stefan Vargyas.
Also move the handling of AI_ERR_EOF into the case stmt.
Do not report ferror/fclose(stdin) failure when we've
already diagnosed e.g., failure to read the DIR, above.
Bug introduced by 2008-11-24 commit 031e2fb5, "du: read and
process --files0-from= input a name at a time,".
* src/wc.c: Handle read failure as with du: do not exit
immediately, but rather go on to print any total and to clean-up.
As above, move the handling of AI_ERR_EOF into the case stmt.
* tests/du/files0-from-dir: New file, to test both du and wc.
* tests/Makefile.am (TESTS): Add it.
* NEWS (Bug fixes): Mention it.
* src/wc.c (main): Diagnose failed argv_iter_init_* failure,
rather than falling through and dereferencing NULL.
Bug introduced by 2008-11-25 commit c2e56e0d,
"wc: read and process --files0-from= input a name at a time,".
* NEWS (Bug fixes): Mention it.
* tests/cp/sparse-fiemap: Move the PERL check to the top,
since we don't test anything without it. In the loop,
don't use skip_test_ as it exits the test completely.
There was an execution path by which "libstdbuf" could be used after
being freed, but that would happen only if there were no libstdbuf.so
alongside the stdbuf program and there had been an installation error
leading to absence of the file, PKGLIBDIR/libstdbuf.so.
* src/stdbuf.c (set_LD_PRELOAD): Rearrange loop to make it perfectly
clear that there is no possibility of use-after-free.
Steve Grubb reported this possible use-after-free of "libstdbuf".
* src/uptime.c (print_uptime): Omit unnecessary "#if defined
HAVE_GETLOADAVG || defined C_GETLOADAVG". This #if is always
true, and removing it will help us simplify the gnulib getloadavg
module.
* tests/cp/sparse-fiemap: Check for fiemap support against a file
rather than a directory to enable tests on BTRFS for example.
Explicity disable the test on ext3 or file systems where we
can't determine the type.
* tests/cp/fiemap-perf: Likewise. Also disable the test on older
BTRFS (like in Fedora 14), where extents are returned for holes.
* tests/init.cfg: Comment that BTRFS only supports fiemap
for regular files.
Don't depend on heuristics to detect sparse files
if fiemap is available. Also don't scan for new
holes unless --sparse=always has been specified.
* src/copy.c (extent_copy): Pass the user specified
sparse mode, and handle as described above.
Do not run a sub-make to set up the environment for each
and every test script. Instead, run it just once and store
the result in a file.
* tests/check.mk (built_programs): Remove definition.
(.built-programs): New rule to create the temporary file.
(CLEANFILES): Arrange to remove it.
(TESTS_ENVIRONMENT): Simply cat .built-programs, rather than
running the sub-make.
* .gitignore: Ignore it.
* tests/cp/fiemap-2: Enable the fiemap check for files, which
will enable the test for files on ext3.
* tests/cp/fiemap-perf: Comment why we're not enabling for ext3.
* tests/cp/sparse-fiemap: Ditto. Also sync the files before
doing a fiemap which was needed for ext4 loop back at least.
Add a comment that FIEMAP_FLAG_SYNC is ineffective, thus
requiring the explicit syncs.
* tests/fiemap-capable: A new python script to determine
if a specified path supports fiemap.
* tests/init.cfg (fiemap_capable_): Use the new python script.
* tests/Makefile.am (EXTRA_DIST): Include the new python script.
This is required for patch, and hence is about to move to gnulib.
* gl/lib/di-set.c (di_set_lookup): New function.
* gl/lib/di-set.h: Declare it.
* gl/tests/test-di-set.c (main): Exercise it.
The bug was introduced on 2004-12-04 via commit 7380cf79.
* src/cut.c (set_fields): When computing the maximum range endpoint,
take into consideration the start of any unbounded range, like "999-".
* NEWS (Bug fixes): Mention it.
* tests/misc/cut (big-unbounded-b,c,f): Add tests.
Reported by Paul Marinescu in http://debbugs.gnu.org/7993
The bug was introduced on 2004-12-04 via commit 7380cf79.
Upon failed lseek, sparse_copy_finalize would mistakenly return true.
Admittedly, that is very unlikely, since that particular lseek
is attempted only if the preceding call to sparse_copy induced
a hole at EOF (via lseek on the destination FD). However, now
that sparse_copy has an output parameter, N_READ, there is no
longer any reason to call lseek (fd, 0, SEEK_CUR), so...
* src/copy.c (sparse_copy_finalize): Remove the function.
(copy_reg): Call ftruncate with n_read, rather than
sparse_copy_finalize with its now-unnecessary lseek.
Lasse Collin spotted the bug in sparse_copy_finalize.
This case was overlooked in commit bdde34f9, 2010-08-05,
"sort: tune and refactor --debug code, and fix minor underlining bug"
* src/sort.c (debug_key): Don't adjust the key end when
it's before the key start.
* tests/misc/sort-debug-keys: Add a test case.
* tests/cp/fiemap-perf: Correct erroneous added test.
Since nonexistent names were used, the final test ended up
being "test =", which would always "succeed".
* src/copy.c (write_zeros): This bug caused 4 or 8 bytes to
be written at a time which is very inefficient. One could
trigger the issue with `cp --sparse=never sparse non-sparse`
on a file system that supports fiemap.
The recent FIEMAP-related changes made it so the unusual case of
copying a sparse file to a non-regular destination (e.g., a pipe)
would erroneously write one byte too many to that destination.
That happened because extent_copy assumed that it could use lseek
to obtain the number of bytes written to the output file descriptor.
That was valid only for regular files.
* src/copy.c (sparse_copy): Add a parameter, to be used by extent_copy,
but not by reg_copy. Adjust callers.
(extent_copy): Maintain new local, dest_pos, using new arg, n_read.
Don't call lseek on dest_fd; use new var, dest_pos, instead.
(copy_reg): Add unused arg.
* tests/cp/fiemap-perf: Copy block-comparing code from sparse-fiemap.
* tests/cp/sparse-fiemap: The same test was here, alongside a much
more involved test. Remove it, now that it is in its own file.
Don't use "#ifdef EINTR". dd.c has been doing that since 2004.
* src/copy.c (sparse_copy): Remove #ifdef...#endif around EINTR use.
* src/tee.c (tee_files): Remove #ifdef...#endif around EINTR use.
If we need it, add something like this in system.h:
/* When EINTR is not defined, define it to an improbable value
so that each use does not have to be #ifdef'd. */
#ifndef EINTR
# define EINTR 999988
#endif
* src/copy.c (extent_copy): Before this change, extent_copy would fail
to create holes, thus breaking --sparse=auto and --sparse=always.
I.e., copying a large enough file of all zeros, cp --sparse=always
should introduce a hole, but with extent_copy, it would not.
we're going to have to use it from within extent_copy, too.
* src/copy.c (sparse_copy): New function, factored out of...
(copy_reg): ...here.
Remove now-unused locals.
* src/copy.c (extent_copy): Don't let what should have been a
temporary reduction of buf_size (to handle a short ext_len) become
permanent and thus impact the performance of all further iterations.
* src/copy.c (copy_reg): Move use of extent_scan to just *after*
we allocate the main copying buffer, so we can...
(extent_scan): Take a new parameter, BUF, and use that rather
than allocating a private buffer. Update caller.
* src/copy.c (copy_reg): Rename a variable to make more sense from
caller's perspective: s/require_normal_copy/normal_copy_required/.
This is an output-only variable, and the original name could make
it look like an input (or i&o) variable.
so that we benefit from using extents also when reading a sparse
input file with --sparse=never.
* src/copy.c (copy_reg): Remove erroneous test of "make_holes"
so that we call extent_copy also when make_holes is false.
Otherwise, what's the point of that parameter?
* src/extent-scan.h [struct extent_scan]: Rename member:
s/hit_last_extent/hit_final_extent/. "final" is clearer,
since "last" can be interpreted as "preceding".
Rename extent-scan functions to start with extent_scan_.
* src/Makefile.am (copy_sources): Also distribute extent-scan.h.
* src/extent-scan.c: Don't include error.h or quote.h. Neither is used.
* src/copy.c: shorten a comment to fit in 80 columns
* src/extent-scan.c, src/extent-scan.h: Correct formatting.
* src/copy.c (write_zeros): Add comments.
(extent_copy): Move decls of "ok" and "i" down to scope where used.
Adjust comments.
Rename local: s/holes_len/hole_size/
Print a diagnostic upon failure to write zeros.
Changes:
========
1. fix write_zeros() per Jim's comments.
2. remove char const *fname from struct extent_scan.
3. change the signature of open_extent_scan() from
"void open_extent_scan(struct extent_scan **scan)" to
"void open_extent_scan(struct extent_scan *scan)" to avoid having
to malloc the extent_scan variable; instead save it on the stack.
4. move close_extent_scan() from a function defined in extent-scan.c
to extent-scan.h as a macro definition, but it does nothing for now,
since initial extent scan defined at stack.
5. add a macro "free_extents_info()" defined at extent-scan.h to
release the memory allocated to extent info which should be called
combine with get_extents_info(), it just one line, so IMHO, define
it as macro should be ok.
* src/extent-scan.c: New file; functions to read "extents".
* src/extent-scan.h: Header file of extent-scan.c.
* src/Makefile.am: Reference it and link it to copy_source.
* src/copy.c: Use the new functions and avoid double-free.
* src/copy.c (fiemap_copy): Ensure that our fiemap buffer
is large enough and well-aligned.
Replace "0LL" with equivalent "0" as 3rd argument to lseek.
* src/copy.c (fiemap_copy): Rename from fiemap_copy_ok.
Add/improve comments.
Remove local, "fail".
(fiemap_copy): Do not require caller to set
"normal_copy_required" before calling fiemap_copy.
Report ioctl failure if it's the 2nd or subsequent call.
* tests/cp/sparse-fiemap: Ensure that a file with many extents
(more than fit in copy.c's internal buffer) is copied properly.
Don't require root access if current partition is btrfs or xfs.
Use init.sh, not test-lib.sh.
* tests/filefrag-extent-compare: New file.
* tests/cp/sparse-fiemap: Add a new test for FIEMAP-copy against a
loopbacked ext4 partition.
* tests/Makefile.am (sparse-fiemap): Reference the new test.
* src/fiemap.h: Add fiemap.h for fiemap ioctl(2) support. Copied
from linux's include/linux/fiemap.h, with minor formatting changes.
* src/copy.c (copy_reg): Now, when `cp' is invoked with --sparse=[WHEN],
we will try to do FIEMAP-copy if the underlying file system
supports it, and fall back to a normal copy if it fails.
* doc/coreutils.texi (cut invocation): Remove the tr -s '[:blank:]'
example, as it doesn't handle leading and trailing blanks. Add `awk`
examples for common field processing operations often asked about.
Also document a `join` hack, to achieve the same thing. Note the
join options are ordered so as to be compatible with other systems.
This allows one to use join as a field extractor like:
join -a1 -o 1.3,1.1 - /dev/null
* src/join.c (join): Don't flag unpairable lines when
one of the files is empty.
* tests/misc/join: Add a new test for empty input, and adjust
a previous test that was only checking against empty input.
* doc/coreutils.texi (join invocation): Document the change.
* NEWS: Likewise.
Lines with a different number of fields than the first line,
will be truncated or padded.
* src/join.c (prfields): A new function refactored from prjoin(),
to output all but the join field.
(prjoin): Don't swap line1 and line2 when line1 is blank
so that the padding is applied to the right place.
(main): Handle the -o 'auto' option.
* tests/misc/join: Add 6 new cases to test the auto format.
* NEWS: Mention the change in behavior.
Suggestion from Assaf Gordon
Those were useful when tests might have been run in the same
directory and in parallel. Now, each test is run in a newly-
created empty directory.
* tests/cp/backup-1: Remove obsolete uses of "$$".
* tests/cp/same-file: Likewise.
* tests/dd/misc: Likewise.
* tests/mv/part-symlink: Likewise.
* tests/mv/to-symlink: Likewise.
* tests/touch/fail-diag: Likewise.
There was a non-negligible delay after running a single test.
Now, you'll know why when you see this test's name.
* tests/check.mk (vc_exe_in_TESTS): Don't @-hide commands.
Use $(AM_V_GEN) instead.
* tests/du/move-dir-while-traversing: Ignoring SIGTSTP is enough;
don't also attempt to ignore SIGSTOP, it cannot be handled or ignored.
Spotted by Andreas Schwab.
* tests/du/move-dir-while-traversing: Create an even larger tree
to avoid a false-positive failure due to du terminating before
the rename is triggered.
* src/tail.c (fremote): Do not print a diagnostic when
fstatfs (pipe_FD, &buf) fails, as it now does on linux-2.3.38.
This avoids the spurious failure of tests/misc/tail's f-pipe-1
test, when running in input-from-pipe mode.
The previous gnulib submodule reference was *still* to a
non-public commit. My submodule had a stray commit, so
the reference was always to a local merge commit.
Reported by Rob Vermaas.
* src/uniq.c (find_field): Remove the byte-skipping loop altogether.
Instead, perform the simple calculation. This results in a 10%
performance improvement for large byte offsets.
* NEWS (Bug fixes): Mention it.
* src/uniq.c (find_field): Stop processing loop when end of line
is reached. Before this fix, 'uniq -f 10000000000 /etc/passwd'
would run for a very long time.
* doc/coreutils.texi (tr's Character sets): Document how a 9-bit
octal value is interpreted. tr does not ignore the ninth bit.
(echo invocation, printf invocation): Document that any ninth
bit in \OOO is ignored. (http://debbugs.gnu.org/7574)
* bootstrap.conf: Add the read-file module
* src/ptx.c: Replace the original code which would
needlessly read SIZE_MAX bytes of files larger than this.
* src/shuf.c: Replace the original code.
* src/system.h: Note where it should be included, and
make ordering check portable to GLIBC > 2
* src/copy.c: Move <sys/ioctl.h> along with other system headers
as is done elsewhere.
* src/install.c: Move <sys/wait.h> along with other system headers
as is done elsewhere.
* src/ptx.c: Include <regex.h> rather than "regex.h" as
is done elsewhere. Note <regex.h> is kept after "system.h"
as per commit dba300a0.
* HACKING: Remove mention of "indent-tabs-mode: nil", since
we've remove all of those directives. No longer needed.
Remove dated (pre-emacs-23) reference regarding WhiteSpace mode.
* NEWS (Bug fixes): Mention it.
* src/du.c (prev_level): Move declaration "up" to file-scope global.
(du_files): Reset prev_level to 0 upon abnormal fts_read termination.
Reported by Johathan Nieder in http://bugs.debian.org/609049
Also, improve a diagnostic.
* tests/du/move-dir-while-traversing: Test for the above.
* tests/Makefile.am (TESTS): Add it.
Before this change, we had a tendency to manually list each
contributor's name in THANKS. Now, each commit "Author" is
included in the generated THANKS file automatically, and most
of the old THANKS file is now a template, THANKS.in.
We'll still have to manually list the names of people who report
problems without a usable patch.
* THANKS.in: New file, derived from THANKS, but removing names of
those who are listed as git log 'Author:'s.
* THANKS: Remove file.
* thanks-gen: New file.
* Makefile.am (THANKS): New rule.
(EXTRA_DIST): Add .mailmap, THANKS.in and thanks-gen.
* .gitignore: Add THANKS and THANKS-to-translators.
* .mailmap: Unify on single address and name-spelling per contributor.
Commit 041c9c47 traded the 'gettext' module for the lighter 'gettext-h'
module, so as to not require the latest gettext release (we only need
the latest release if we ship gettext as a dependent library, but
coreutils has long preferred to use it as an external library).
But that commit overlooked two places necessary to allow the use of
gettext 0.17.
This does not force you to downgrade (using gettext 0.18.1.1 is still
just fine), nor does it affect tarballs (once a tarball is built
with a given gettext version, it can be built on other machines
regardless of what gettext version is present).
* bootstrap.conf (buildreq): Relax prerequisite.
* configure.ac (AM_GNU_GETTEXT_VERSION): Likewise.
* src/split.c (set_suffix_length): Only auto-calculate
the suffix length when the number of files is specified.
* tests/misc/split-a: Add a case to trigger the bug,
and exercise the suffix length auto-calculation.
* NEWS: Mention the fix.
Reported by Dmitry V. Levin and Sergey Vlasov at
https://bugzilla.altlinux.org/show_bug.cgi?id=24841
* src/csplit.c (create_output_file): Detect overflow when the
file counter wraps around, and exit with a diagnostic. Formerly
the code silently wrapped around and wrote to the wrong file,
losing output data.
* src/getlimits.c (decimal_ascii_add): Remove, replacing with ...
(decimal_absval_add_one): New function, with different signature,
which does not assume ASCII. All callers changed.
(print_int): Remove assumptions that integers fit in 206 bits, and
that characters are ASCII. These assumptions are portable in
practice but are easy to remove here.
* tests/mv/i-3: Adjust comment to match just-changed code.
Spotted by Pádraig Brady.
* tests/init.cfg (retry_delay_): Correct spelling of function name
in usage example.
* gnulib-tests/Makefile.am (test_xvasprintf_CFLAGS):
(test_lock_CFLAGS, test_tls_CFLAGS): Avoid a syntax error when
$(WERROR_CFLAGS) expands to more than one token.
* gnulib-tests/Makefile.am (test_xvasprintf_CFLAGS):
(test_lock_CFLAGS, test_tls_CFLAGS): Do not append GCC-specific
flags like -Wno-format-security unless the GCC-specific flag
-Werror is also specified. This avoids a "make check" failure on
Solaris when using Sun C 5.8.
This prevents a compilation failure on Solaris 8, GCC 4.4.2, with
"configure --enable-gcc-warnings".
* src/who.c (MAXHOSTNAMELEN): Remove; no longer needed.
* src/pinky.c: Likewise.
* tests/init.sh (setup_): Initialize fail=0 before invoking mktempd_.
Ensure that IFS is defined initially.
(mktempd_): Remove fail=0 initialization; no longer needed.
* src/cp.c (do_copy): When -T is specified, initialize
the NEW_DST and SB variables, which are checked when
running: cp -T --force --backup file file
* tests/cp/backup-1: Add the -T case
* src/sort.c (main): If --parallel isn't specified,
restrict the number of threads to 8 by default.
If the --parallel option is specified, then
allow any number of threads to be set, independent
of the number of processors on the system.
* doc/coreutils.texi (sort invocation): Document the changes
to determining the number of threads to use.
Mention the memory overhead when using multiple threads.
* tests/misc/sort-spinlock-abuse: Allow single core
systems that support pthreads.
* tests/misc/sort-stale-thread-mem: Likewise.
* tests/misc/sort-unique-segv: Likewise.
* NEWS: Mention the change in behaviour.
Without this change, tests/misc/sort-compress-hang would consume
more than 10,000 process slots on my RHEL 5.5 x86-64 server,
making it likely for other applications to fail due to lack of
process slots. With this change, the same benchmark causes 'sort'
to consume at most 19 process slots. The change also improved
wall-clock time by 2% and user+system time by 14% on that benchmark.
* NEWS: Document this.
* src/sort.c (MAX_PROCS_BEFORE_REAP): Remove.
(reap_exited): Renamed from reap_some; this is a more accurate name,
since "some" incorrectly implies that it reaps at least one process.
All uses changed.
(reap_some): New function: it *does* reap at least one process.
(pipe_fork): Do not allow more than NMERGE + 2 subprocesses.
(mergefps, sort): Omit check for exited processes: no longer needed,
and anyway the code consumed too much CPU per line when 2 < nprocs.
* NEWS: Document this.
* src/sort.c (UNCOMPRESSED, UNREAPED, REAPED): New constants.
(struct tempnode): New member 'state', to hold these constants.
The pid member is now undefined if state == UNCOMPRESSED.
(struct sortfile): Replace member 'pid' with member 'temp'.
(uintptr): Remove.
(proctab_hasher, proctab_comparator, register_proc, delete_proc):
Proctab entries are now struct tempnode *, not pid_t, to handle
the case where multiple tempnode objects correspond to the same
pid. This avoids a race condition that can cause a hang.
(register_proc): Arg is now struct tempnode *, not pid_t. All
callers changed.
(delete_proc): Set tempnode state to REAPED.
(create_temp_file): No need to set pid member here; it's now
done when the pid is known.
(maybe_create_temp, create_temp): Remove PPID arg. Return struct
tempnode *, not char *. All callers changed.
(maybe_create_temp): Set node state to UNCOMPRESSED or UNREAPED.
No need to set node->pid to 0.
(open_temp): Replace NAME and PID args with a single TEMP arg.
All callers changed. Wait only for unreaped children.
(zaptemp): Wait for decompressor to finish before removing its
temporary-file input. This avoids .nfsXXXX hassles with NFS
and fixes a race (leading to a hang) regardless of NFS.
(open_input_files): Adjust to new way of dealing with temp files
and their subprocesses.
* tests/Makefile.am (TESTS): Add misc/sort-compress-hang.
* tests/misc/sort-compress-hang: New file.
* NEWS: Document this.
* src/sort.c (avoid_trashing_input): The previous fix to this
function didn't fix all the problems with this code. Replace it
with something simpler: just copy the input file. This doesn't
change the number of files, so return void instead of the updated
file count. Caller changed.
* tests/misc/sort-merge-fdlimit: Test for the bug.
Those options are useful only on systems that lack inotify support
and in the unusual event that a system with inotify support must
resort to polling.
* src/tail.c (usage): Note that the --max-unchanged-stats=N and
--sleep-interval=N options are rarely useful on systems with
inotify support.
* doc/coreutils.texi (tail invocation): Likewise.
* src/sort.c (avoid_trashing_input): Fix a typo that could cause a
buffer overrun in theory. In practice this is extremely unlikely,
as it requires running out of file descriptors in a small merge,
presumably because some other process is hogging all the OS's file
descriptors.
* tests/check.mk (TESTS_ENVIRONMENT): Default TMPDIR to /tmp,
rather than to the working directory; this is more common in
practice, which makes the tests more real-worldish; and it is
often faster. Also, it avoids some problems with NFS cleanups.
* tests/misc/sort-compress: Remove unnecessary code setting TMPDIR.
* tests/misc/sort-compress-proc: Likewise. Do the final sleep
only if TMPDIR is relative, which should be rarely given the
change to TESTS_ENVIRONMENT.
* src/sort.c (uintptr): New type.
(enum procstate, struct procnode, update_proc): Remove.
(proctab_hasher, proctab_comparator, register_proc, wait_proc):
(reap_some): The proctab is now simply a hash of process-IDs
rather than of pointers to objects with reference counts and
states; this is smaller and faster and easier to understand.
(nprocs): Now pid_t, not size_t, since one cannot have more than
PID_MAX children.
(reap): If the argument is -1, wait; if 0 (a new value), do not.
Delete pid from proctab as needed. Ignore children that are not
in proctab, as they are from the program that exec'ed us and are
irrelevant to our success or failure.
(delete_proc, reap_all): New functions.
(open_temp): Register the child.
(sort): Clean up all children afterwards; without this patch,
'sort' sometimes missed failures in children due to race conditions.
* tests/Makefile.am (TESTS): Add misc/sort-compress-proc.
* tests/misc/sort-compress-proc: New file, to test for the
bugs fixed above.
* tests/misc/sort-stale-thread-mem: Don't initialize fail=0 here;
that is done in init.sh. This avoids a syntax-check failure.
Invoke "Exit $fail" at end, too.
Mark as a very expensive test.
This change does not fix the actual bug. That was done by commit
c9db0ac6, "sort: preallocate merge tree nodes to heap". The fix
was to store each "node" structure on the heap, not on the stack.
Otherwise, a node from one thread's stack could be used in another
thread after the first thread had expired (via pthread_join).
This bug was very hard to trigger when using spinlocks, but
easier once we began using mutexes.
* NEWS (Bug fixes): Mention it.
For details, see http://debbugs.gnu.org/7597.
* src/sort.c (specify_nthreads, merge_tree_init, init_node):
(queue_init, sortlines, struct thread_args, sort, main):
Use size_t, not unsigned long int, for thread counts, since thread
counts are now used to compute sizes.
(specify_nthreads): Check for size_t overflow.
(merge_tree_init, sort): Shorten name of local variable, for
readability.
(merge_tree_init): Move constants next to each other in product,
so that the constant folding is easier to see.
(init_node): Now static. Add 'restrict' only where it might
be helpful for compiler optimization.
(queue_init): 2nd arg is now nthreads, not "reserve", which is
a bit harder to follow. All uses changed.
(struct thread_args): Rename lo_child to is_lo_child, so that
it's obvious to the reader when we're talking about this boolean
as opposed to the new lo_child member of the other structure.
All uses changed.
(sort): Remove unused local variable end_node.
(main): Don't allow large thread counts to cause undefined behavior
later, due to integer overflow.
* src/sort.c: (merge_tree_init) New function. Allocates memory for
merge tree nodes.
(merge_tree_destory) New function.
(init_node) New function.
(sortlines) Refactor node creation code to init_node. Remove now
superfluous arguments. All callers changed.
(sort) Initialize/destory merge tree. Refactor root node creation
to merge_tree_init.
Running a command like this on a multi-core system
sort < big-file | less
would peg all processors at near 100% utilization.
* src/sort.c: (struct merge_node) Change member lock to mutex.
All uses changed.
* tests/Makefile.am (XFAIL_TESTS): Remove definition, now that
this test passes once again. I.e., the sort-spinlock-abuse test
no longer fails.
* NEWS (Bug reports): Mention this.
Reported by DJ Lucas in http://debbugs.gnu.org/7489.
When -n l/N is used and long lines are present that both
span partitions and multiple buffers, one would get
inconsistent chunk sizes.
* src/split.c (main): Add a new undocumented ---io-blksize option
to support full testing with varied buffer sizes.
(cwrite): Refactor most handling of --elide-empty to here.
(bytes_split): Remove handling of --elide-empty.
(lines_chunk_split): Likewise. The specific issue here
was the first handling of elide_empty_files interfered
with the replenishing of the input buffer.
* test/misc/split-lchunk: Add -e and the new ---io-blksize
combinations to the test.
* tests/misc/sort-spinlock-abuse: On a busy system, with only 12
pauses of length 0.1 seconds, the buggy (busy-spinlock blocked)
sort would fail to accumulate 1 second of CPU time, and hence
would mistakenly pass. Increase from 12 to 50.
* src/sort.c (struct thread_args, sortlines_thread, sortlines, sort):
Rename "merge_queue" to "queue", for consistency with other functions
that just use the name "queue" for these things.
* src/sort.c (mergelines_node): Return void, not size_t. All
callers changed. Change *node->dest here, not in caller.
Do not change node->dest: it's not needed and could cause problems
on (mostly theoretical) hosts that do not allow adding integers to
null pointers.
(queue_check_insert_parent): Omit MERGED parameter; no longer needed.
All callers changed.
* src/sort.c (queue_check_insert, queue_check_insert_parent): Make
the queue arg first, for consistency with other functions such as
queue_insert that put the queue arg first. Rename from
check_insert and update_parent, respectively. All callers
changed.
* src/sort.c (struct merge_node): 'lock' is now the actual lock,
not a pointer to the lock; there's no need for indirection here.
Make 'level' unsigned int instead of size_t, since it is a
bit-shift count; also, move it next to a bool so that it's more
likely to take less space. All uses changed.
(sortlines, sort): Spell out initialization instead of using an
initializer. This makes the initializer a bit easier to understand,
and avoids unnecessary stores into the spin lock.
This problem was observed on RHEL 5.5 x86-64 when running as a
client of a NetApp FAS2050.
* tests/cp/cp-mv-backup: Don't leave a file descriptor open to
a file in a directory that will be cleaned up with "rm -rf".
Under NFS, when the rm unlinks that file, it is instead renamed
to .nfsXXXX and then rm cannot remove the parent directory,
and the test fails.
* tests/cp/same-file: Likewise.
* src/sort.c (MAX_MERGE): Avoid integer overflow when on a machine
with (say) 32-bit int and 64-bit size_t and when level == 15.
Without this fix, on such a machine with 32768 or more processors,
the level computation could overflow on large input, and this
would result in division by zero.
* src/sort.c (write_unique): Save the entire "struct line", not
just a pointer to one. Otherwise, with a multi-thread run,
sometimes, with some inputs, fillbuf would would win a race
and clobber a "saved->text" pointer in one thread just before
it was dereferenced in a comparison in another thread.
* NEWS (Bug fixes): Mention it.
* README-valgrind: Include the "noinst" programs in
those wrapped by valgrind. Update $PATH in check.mk
rather than Makefile.am. Make wrapper scripts
work when suppressions not setup. Keep lines < 80 chars.
This patch was written by Jim Meyering and myself.
* src/copy.c (copy_reg): Turn EISDIR to ENOTDIR to improve the
quality of diagnostics for commands like "cp a nosuch/". Reported
by Марк Коренберг and Alan Curry in the thread starting at:
http://lists.gnu.org/archive/html/bug-coreutils/2010-11/msg00178.html
* THANKS: Update.
* tests/mv/trailing-slash: Add a test.
* src/split.c (usage, long_options, main): New options --number,
--unbuffered, --elide-empty-files.
(set_suffix_length): New function to auto increase suffix length
to handle a specified number of files.
(create): New function. Refactored from cwrite() and ofile_open().
(bytes_split): Add max_files argument to support byte chunking.
(lines_chunk_split): New function. Split file into chunks of lines.
(bytes_chunk_extract): New function. Extract a chunk of file.
(of_info): New struct. Used by functions lines_rr and ofile_open
to keep track of file descriptors associated with output files.
(ofile_open): New function. Shuffle file descriptors when there
are more output files than available file descriptors.
(lines_rr): New function to distribute lines round-robin to files.
(chunk_parse): New function. Parses K/N syntax.
* tests/misc/split-bchunk: New test for byte chunking.
* tests/misc/split-lchunk: New test for line delimited chunking.
* tests/misc/split-rchunk: New test for round-robin chunking.
* tests/Makefile.am: Reference new tests.
* tests/misc/split-fail: Add failure scenarios for new options.
* tests/misc/split-l: Fix a typo. s/ln/split/.
* doc/coreutils.texi (split invocation): Document --number.
* NEWS: Mention the new feature.
* .mailmap: Map new email address for shortlog.
Signed-off-by: Pádraig Brady <P@draigBrady.com>
* NEWS: Describe patch.
* bootstrap.conf (gnulib_modules): Add ftoastr.
* src/od.c: Include ftoastr.h, not float.h.
(FLT_DIG, DBL_DIG): Remove. No need to verify LDBL_DIG.
(FMT_BYTES_ALLOCATED): No need to worry about floating point now,
since this format is no longer used for floating point.
(PRINT_FIELDS): New macro, with most of the guts of the old PRINT_TYPE.
(PRINT_TYPE): Rewrite to use PRINT_FIELDS.
(PRINT_FLOATTYPE): New macro. This uses the new functions from
ftoastr.
(print_float, print_double, print_long_double): Reimplement
using PRINT_FLOATTYPE.
(decode_one_format): Calculate field widths based on ftoastr-supplied
macros.
* tests/Makefile.am (TESTS): Add misc/od-float.
* tests/misc/od-float: New file.
Initially, I did this,
git grep -l srcdir/test-lib.sh|xargs perl -p0i -e '~180-byte script'
but that line would have been much longer than the maximum permitted
by coreutils' commit hook, and wasn't readable besides, so here's a
more readable version:
lhs=$(printf '%s\\n' \
'if test "$VERBOSE" = yes; then' \
' set -x' \
' touch --version' \
'fi' \
'' \
'. $srcdir/test-lib.sh' \
| sed 's/\$/\\\$/g;s/touch/(\\w+)/')
rhs=$(printf '%s\\n' \
'. "${srcdir=.}/init.sh"; path_prepend_ ../src' \
'test "$VERBOSE" = yes && FIXME --version' \
| sed 's/\$/\\\$/g;s/FIXME/\$1/')
git grep -l srcdir/test-lib.sh|xargs perl -p0i -e "s,$lhs,$rhs,"
src/truncate.c (main): Use a bool to store if an error occurred,
rather than an int, to protect against overflow.
(do_ftruncate): Likewise. Also change 0/false to mean failure
rather than success.
... when configured with the --enable-gcc-warnings option.
This follows on from commit 34ef0a01, 2010-10-14,
"sort: fix unportable cast of unsigned char * -> char *"
* configure.ac: -Wall implicitly enables this warning
so remove the explicit disabling.
* gl/modules/fadvise-tests: Add the module previously missed
in commit 63b5e816, 2010-07-14, "fadvise: new module ...".
* gl/tests/test-fadvise.c: Add a comment as to why we don't
check return values.
* doc/coreutils.texi (stat invocation): Add sentence-ending period.
* NEWS: Correct stat change description: s/floating point //.
* cfg.mk (old_NEWS_hash): Update, to match this NEWS change.
Copyright (C) 2005-2006, 2009-2010 Free Software Foundation, Inc.
Copyright (C) 2005-2006, 2009-2011 Free Software Foundation, Inc.
This program is free software: you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
@@ -23,7 +23,7 @@
/* Externally-visible name for numcompare. */
int
int_GL_ATTRIBUTE_PURE
strnumcmp(charconst*a,charconst*b,
intdecimal_point,intthousands_sep)
{
Some files were not shown because too many files have changed in this diff
Show More
Reference in New Issue
Block a user
Blocking a user prevents them from interacting with repositories, such as opening or commenting on pull requests or issues. Learn more about blocking a user.
