* src/fts-missing.h: Add helper routines to determine
if an FTS entry is skippable. I.e., that is traversed
and not specified on the command line.
* src/chcon.c (process_file): Setup to ignore skippable files.
(change_file_context): Ignore missing files if appropriate.
* src/chmod.c (process_file): Setup to ignore skippable files.
Ignore missing files if appropriate.
(mode_changed): Ignore missing files if appropriate.
* src/chown-core.c (change_file_owner): Setup to ignore skippable files.
Ignore missing files if appropriate.
* src/du.c (process_file): Likewise.
* src/ls.c (print_dir): Ignore missing files if appropriate.
(get_link_name): Likewise.
(gobble_file): Likewise. Also handle the dangling symlink case.
* src/system.h (ignorable_traversal_errno): Match ENOENT for missing
files.
* tests/chcon/traversal-missing.sh: A new test.
* tests/misc/traversal-missing.sh: Likewise.
* tests/local.mk: Reference the new tests.
* NEWS: Mention the bug fix.
Fixes https://bugs.gnu.org/81444
* tests/rm/empty-immutable-skip.sh: Use cleanup_() to
remove the immutable attribute, so that false failures
or stale files do not result from e.g. a killed test run.
"install -C" (--compare) has always called chown/chmod on the
destination even after determining that copying was unnecessary,
ever since --compare was added in commit v7.0-182-gdac5f12c6.
need_copy() already verifies that the destination's mode, owner, and
group match (and, with --preserve-context, that the SELinux context
matches too) before skipping the copy, so re-applying them is
redundant, and can needlessly fail on a destination that cannot be
chown/chmod'd even though nothing would actually change, e.g. one
marked immutable with chattr +i.
* src/install.c (install_file_in_file): Skip metadata updates
if we already skipped the data, as previous checks ensure
that the metadata is already as required.
* tests/install/install-C-immutable.sh: Add a test.
(based on tests/rm/empty-immutable-skip.sh).
* tests/local.mk: Reference the new test.
* NEWS: Mention the improvement.
* tests/dd/nocache.sh: Whether fadvise() is called (uutils)
or avoided (GNU), ensure there is no failure when nocache
is only the ancillary operation being requested.
Link: https://github.com/coreutils/coreutils/pull/276
* tests/cat/splice.sh: Move read/error distinction test to ...
* tests/cat/cat-distinct-err.sh: ... here, and augment.
* tests/local.mk: Reference the new file.
* tests/cut/cut.pl: Add cases verifying that in the C locale -c and
-b -n select a single byte instead of decoding a multi-byte sequence.
Link: https://github.com/coreutils/coreutils/pull/316
* src/pr.c (char_to_clump): Use ckd_add() and report integer overflow.
* tests/pr/options.sh: Add a test case.
* NEWS: Mention the bug fix.
Fixes https://bugs.gnu.org/81393
With an ASAN build we can trigger the issue like:
$ src/factor 22486472939460651857
src/factor.c:888:36: runtime error:
index 675 out of bounds for type 'int_least16_t [675]'
* src/factor.c (factor_using_division): Guard the final unrolled block
before reading a subsequent prime.
* tests/factor/factor.pl: Add a test case.
* NEWS: Mention the bug fix.
Problem caught by Coverity analysis.
* NEWS: Mention the bug fix.
* src/du.c (main): Prefer ckd_add to check for overflow. Emit an error
if the value is negative.
* tests/du/max-depth.sh: Add a test case for the bug.
* tests/nproc/nproc-quota.sh: Use command -v nproc instead of
$abs_top_builddir for external usage of tests.
Also quote "$nproc" usage, to support spaces in path.
Link: https://github.com/coreutils/coreutils/pull/310
* tests/mv/mv-special-1.sh: Adjust the python probe to more
closely match how copy.c creates sockets, as otherwise we
get a false failure on systems where mknod() fails with S_IFSOCK,
which was seen on Solaris and FreeBSD at least.
* tests/mv/mv-special-2.sh: Adjust similarly as this
was a latent issue on this root only test.
Reported by Bruno Haible.
* tests/mv/mv-special-1.sh: Moving a socket onto an existing file
on another file system must replace that file, and not for example
unlink the destination first and then fail, leaving the user
with neither file.
Link: https://github.com/coreutils/coreutils/pull/303
A failed shuf invocation should leave an existing output file untouched
rather than truncating it before the failure is detected.
* tests/shuf/shuf.sh: Verify that a missing input file or an unreadable
random source leaves an existing -o file unchanged.
Link: https://github.com/coreutils/coreutils/pull/304
* tests/ls/quoting-utf8.sh: A non-ASCII C1 control character (U+0085
NEL) is not printable, so locale/clocale quoting must octal-escape its
bytes (\302\205) rather than emit them literally. Add a file named
with U+0085 and verify the escaped form for both styles.
gap identified here: https://github.com/uutils/coreutils/pull/13150
Link: https://github.com/coreutils/coreutils/pull/306
A trailing slash only asserts that the operand is a directory. That
is a stat-level check, needing search permission on the parent, not
read permission on the directory itself.
* tests/misc/realpath.sh: Check that a searchable but unreadable
directory still resolves with a trailing slash, with and without -e.
Link: https://github.com/coreutils/coreutils/pull/307
* tests/cp/same-file.sh: A different spelling of the same path
('foo' vs './foo') must still be detected as the same file, so
--remove-destination refuses instead of unlinking the source.
Link: https://github.com/coreutils/coreutils/pull/302
This increases throughput on systems with unlocked stdio functions.
E.g., 2.5x in the following case:
$ timeout 1m taskset 1 shuf -r -i 0-100 \
| taskset 2 uniq -c \
| taskset 3 pv -r > /dev/null
[ 100MiB/s]
$ timeout 1m taskset 1 shuf -r -i 0-100 \
| taskset 2 uniq -c \
| taskset 3 pv -r > /dev/null
[ 246MiB/s]
* src/uniq.c: Include assure.h.
(writeline): Convert the value to a string by hand and use fwrite, which
may be unlocked, instead of printf.
* tests/uniq/uniq-c-width.sh: New test to verify the padding logic.
* tests/local.mk (all_tests): Add the test case.
* NEWS: Mention the improvement.
In legacy CJK locales a single input byte can decode to a wide character
above UCHAR_MAX, e.g. SHIFT-JIS bytes 0xA1..0xDF become U+FF61..U+FF9F.
wc then indexed its 256-entry wc_isprint/wc_isspace tables with that
value and read past their end.
* src/wc.c (wc): Take the byte-indexed fast path only when the decoded
character fits in a byte; wider values fall through to c32width and
c32isspace.
* tests/wc/wc-sjis.sh: New test.
* tests/local.mk (all_tests): Add it.
* NEWS: Mention the fix.
Link: https://github.com/coreutils/coreutils/pull/298
* tests/nproc/nproc-quota-systemd.sh: Explicitly set OMP_NUM_THREADS
to an ignorable value, to ensure scheduler processing is independent.
* tests/nproc/nproc-quota.sh: Likewise for the SCHED_FIFO check.
https://github.com/coreutils/coreutils/pull/290
This imports a xstrtol-error change from Gnulib that fixes some error
messages that were not properly escaped when given an invalid integer
argument.
* .gitignore: Add /lib/thread-creators.h which is generated by Gnulib.
* gnulib: Update to the latest commit.
* tests/od/od.pl (@Tests): Add some test cases.
* NEWS: Mention the improvement.