* configure.ac: When --enable-gcc-warnings is used, omit
-Wno-type-limits. The need for -Wno-type-limits has passed, now
that intprops.h uses builtin primitives for GCC 5 and later, given
that recent GCCs issue type-limits warnings only for non-constant
expressions. --enable-gcc-warnings is not intended for use with
old compilers, so we can drop -Wno-type-limits now.
‘shuf -r -n 0 file’ would mistakenly read from standard input.
Problem reported by my student Jingnong Qu while reimplementing a
shuf subset in Python as an exercise in UCLA Computer Science 35L:
https://web.cs.ucla.edu/classes/fall19/cs35L/assign/assign3.html
* NEWS: Mention the fix. Also, ASCIIfy a previous item.
* src/shuf.c (main): Fix bug.
* tests/misc/shuf.sh: Add a test case for the bug.
statx allows ls to indicate interest in only certain inode metadata.
This is potentially a win on networked/clustered/distributed
file systems. In cases where we'd have to do a full, heavyweight stat()
call we can now do a much lighter statx() call.
As a real-world example, consider a file system like CephFS where one
client is actively writing to a file and another client does an
ls --color in the same directory. --color means that we need to fetch
the mode of the file.
Doing that with a stat() call means that we have to fetch the size and
mtime in addition to the mode. The MDS in that situation will have to
revoke caps in order to ensure that it has up-to-date values to report,
which disrupts the writer.
This has a measurable affect on performance. I ran a fio sequential
write test on one cephfs client and had a second client do "ls --color"
in a tight loop on the directory that held the file:
Baseline -- no activity on the second client:
WRITE: bw=76.7MiB/s (80.4MB/s), 76.7MiB/s-76.7MiB/s (80.4MB/s-80.4MB/s),
io=4600MiB (4824MB), run=60016-60016msec
Without this patch series, we see a noticable performance hit:
WRITE: bw=70.4MiB/s (73.9MB/s), 70.4MiB/s-70.4MiB/s (73.9MB/s-73.9MB/s),
io=4228MiB (4433MB), run=60012-60012msec
With this patch series, we gain most of that ground back:
WRITE: bw=75.9MiB/s (79.6MB/s), 75.9MiB/s-75.9MiB/s (79.6MB/s-79.6MB/s),
io=4555MiB (4776MB), run=60019-60019msec
* src/stat.c: move statx to stat struct conversion to new header...
* src/statx.h: ...here.
* src/ls.c: Add wrapper functions for stat/lstat/fstat calls,
and add variants for when we are only interested in specific info.
Add statx-enabled functions and set the request mask based on the
output format and what values are needed.
* NEWS: Mention the Improvement.
* src/truncate.c (do_ftruncate): Simplify overflow checking,
and don’t rely on theoretically-nonportable assumptions
like assuming that OFF_MAX < UINTMAX_MAX.
* src/seq.c: (seq_fast): Accept STEP as a parameter and use that
to skip the output of generated numbers.
(main): Relax to using seq_fast for integer steps between 1 and 200.
For larger steps the throughput was faster using the standard
incrementing procedure.
(cmp): Use the equivalent but faster memcmp for equal len strings.
* tests/misc/seq.pl: Update fast path cases.
Addresses https://bugs.gnu.org/37241
The previous commit introduced a bug into the following syntax-check,
and thus effectively turned it off:
$ make sc_prohibit_test_calls_print_ver_with_irrelevant_argument; \
echo $?
prohibit_test_calls_print_ver_with_irrelevant_argument
fatal: cannot change to 'grep': No such file or directory
0
* cfg.mk (sc_prohibit_test_calls_print_ver_with_irrelevant_argument):
Remove changing directory, and pass $(srcdir) as argument to 'git -C'.
* cfg.mk (sc_prohibit_colon_redirection): Don't expect `|` to denote
the pipe character in git grep.
(sc_tests_executable)
(sc_case_insensitive_file_names)
(sc_some_programs_must_avoid_exit_failure)
(sc_prohibit_test_background_without_cleanup_)
(sc_prohibit_test_calls_print_ver_with_irrelevant_argument)
(sc_prohibit_test_ulimit_without_require_)
(sc_prohibit_test_background_without_cleanup_)
(sc_THANKS_in_duplicates)
*sc_prohibit_test_calls_print_ver_with_irrelevant_argument):
Don't expect builddir to be a descendant of srcdir.
(sc_strftime_check): Don't check file size against 0 when "N\nq\n" was
already put in the file.
* THANKS.in: Remove me.
Under certain circumstances seq prints an extra line when the output
format has custom format with characters following the printed numbers:
$ seq -f "%g " 1000000 1000000
1e+06
1e+06
This is due to the "print_extra_number" logic using strings to determine
whether a 'extra number' is needed, but only one string was trimmed
when using a custom printf format.
Prompted by https://lists.gnu.org/r/coreutils/2019-08/msg00001.html
* NEWS: Mention fix.
* src/seq.c (print_numbers): Trim the 'x0_str' string before comparing
it to the previous 'x_str' string.
* tests/misc/seq-extra-number.sh: Add this scenario.
* tests/local.mk (all_tests): Add new test.
* doc/sort-version.texi: Fix some typos, avoid overly long lines in
the generated PDF, enclose some sample strings in @samp{...} for better
readability, etc. This also avoids an sc-avoid-builtin error:
s/builtin/built-in/
* doc/sort-version.texi: New file.
* doc/local.mk (doc_coreutils_TEXINFOS): Add new file.
* doc/coreutils.texi: @include new file, replace previous "Details about
version sort" section.
When calling 'stat -c %N' to print the filename, don't explicitly
request the size of the file via statx(), as it may add overhead on
some filesystems. The size is only needed to optimize an allocation
for the relatively rare case of reading a symlink name, and the worst
effect is a somewhat-too-large temporary buffer may be allocated for
areadlink_with_size(), or internal retries if buffer is too small.
The file size will be returned by statx() on most filesystems, even
if not requested, unless the filesystem considers this to be too
expensive for that file, in which case the tradeoff is worthwhile.
* src/stat.c: Don't explicitly request STATX_SIZE for filenames.
Problem reported by Szőts Ákos (Bug#36291).
* NEWS: Mention this.
* src/od.c (skip): Try fseek even on files that do not have usable
sizes, falling back on fread if fseek fails.
When debugging an invalid date due to DST switching, the intermediate
'normalized time' should not be checked - its value can differ between
systems (e.g. glibc vs musl).
Reported by Niklas Hambüchen in
https://lists.gnu.org/r/coreutils/2019-05/msg00031.html
Analyzed by Rich Felker in
https://lists.gnu.org/r/coreutils/2019-05/msg00039.html
* tests/misc/date-debug.sh: Replace the exact normalized time
with 'XX:XX:XX' so different values would not trigger test failure.
* src/stat.c: Drop statbuf argument from out_epoch_sec().
Use statx() rather than [lf]stat() where available,
so a separate call is not required to get birth time.
Set STATX_* mask bits only for things we want to print,
which can be more efficient on some file systems.
Add a new --cache= command-line option that sets the appropriate hint
flags in the statx call. These are primarily used with network
file systems to indicate what level of cache coherency is desired.
The new option is available unconditionally for better portability,
and ignored where not implemented.
* doc/coreutils.texi: Add documention for --cached.
* man/stat.x (SEE ALSO): Mention statx().
* NEWS: Mention the new feature.
* src/split.c (set_suffix_length): Use a more standard
zero based logN calculation for the number of units.
* tests/split/suffix-auto-length.sh: Add a test case.
* THANKS.in: Mention the reporter.
* NEWS: Mention the fix.
Fixes https://bugs.gnu.org/35291
Problem reported by Hans Henrik Bergan (Bug#36007).
* NEWS: Mention this.
* src/dd.c (iclose, ifdatasync, ifstat, ifsync):
New functions, which are more careful about SIGINT.
(cleanup): Use iclose instead of close.
(finish_up): Process signals first.
(skip, dd_copy, main): Use ifstat instead of fstat.
(dd_copy): Use ifdatasync and ifsync instead of fdatasync and fsync.
Problem reported by Jakub Kulik (Bug#35713).
* NEWS: Mention this.
* configure.ac (DEV_FD_MIGHT_BE_CHR): New macro.
* src/copy.c (DEV_FD_MIGHT_BE_CHR): Default to false.
(follow_fstatat): New function.
(copy_internal): Use it.
* src/copy.h (XSTAT): Remove; no longer used.
The wording of the dd --help text suggests that output will be skipped
for sparse *input* blocks (i.e. that NUL-checking is done on input
blocks) while the code actually checks/skips all-NUL *output* blocks.[1]
* src/dd.c (usage): Update the --help text to clarify the above.
* tests/dd/sparse.sh: Ensure sparseness is controlled with obs.
[1]: https://superuser.com/a/1136358
Its support for the -include option is flaky. Problem reported by
Michael Osipov (Bug#35650). Plus, we could run into other
compilers that don’t support any option like -include. Change the
code so that -include is not needed. Although this causes us to
depart from the upstream version, we’re already doing that for
other reasons.
* configure.ac (USE_XLC_INCLUDE): Remove, as there’s no
guarantee a compiler will support something like -include.
* src/blake2/b2sum.c [HAVE_CONFIG_H]: Include <config.h>.
* src/local.mk (src_b2sum_CPPFLAGS): Add -DHAVE_CONFIG_H.
Do not use -include or a substitute.
Problem reported by Michael Osipov (Bug#35650).
* configure.ac: Use AC_LANG_WERROR to pay attention to compiler
and linker warnings when testing whether stdbuf will work.
* src/blake2/blake2.h (BLAKE2_PACKED):
Don’t assume __attribute__ ((packed)) works on non-Microsoft
compilers. Instead, assume it works only if we have good
reason to assume so, and fall back on Microsoft (or not packing)
otherwise. In practice, not packing is good enough and the
BLAKE2_PACKED macro is mostly just for documentation.
* src/basenc.c: Various minor style cleanups.
(struct base_decode_context): Do not use anonymous unions, as
they’re not in C99. Use a named union instead. All uses changed.
* src/system.h (X2NREALLOC, X2REALLOC, DECIMAL_DIGIT_ACCUMULATE):
Use verify_expr instead of verify_true, which has been removed.
(DECIMAL_DIGIT_ACCUMULATE): Remove unnecessary size check.
