coreutils-9.6

[Xi Ruoyao]

New minor version.

[Xi Ruoyao]

As usual the i18n patch needs to be updated.

[Xi Ruoyao]

I managed to rebase the i18n patch but now autoreconf -fiv blows up. It invokes autopoint, and autopoint extracts threadlib.m4 from /usr/share/gettext/archive.dir.tar.xz, overwriting the (newer) one shipped in the package. The bootstrap script in the package says:

  # NOTE: we have to be careful to run both autopoint and libtoolize
  # before gnulib-tool, since gnulib-tool is likely to provide newer
  # versions of files "installed" by these two programs.
  # Then, *after* gnulib-tool (see below), we have to be careful to
  # run autoreconf in such a way that it does not run either of these
  # two just-pre-run programs.
}}

and maybe we are hitting this.  But why this wasn't an issue for coreutils-9.5??

[Xi Ruoyao]

Also the issue above isn't related to the patch. Just run autoreconf -fiv in the original release tarball and it'll blow up.

[Xi Ruoyao]

Remving -i from the autoreconf command line does allow to build the package, but then the test suite blows up because Makefile.in is generated with /usr/share/automake-1.17/am/check.am which passes --collect-skipped-logs to test-driver, but without -i the shipped test-driver isn't updated by the automake-1.17 copy and it doesn't support --collect-skipped-log.

[Xi Ruoyao]

autoreconf -fv && automake -af works.

[Xi Ruoyao]

​https://www.linuxfromscratch.org/patches/downloads/coreutils/coreutils-9.6-i18n-1.patch

[Xi Ruoyao]

Replying to Xi Ruoyao:

​https://www.linuxfromscratch.org/patches/downloads/coreutils/coreutils-9.6-i18n-1.patch

Fedora has updated coreutils to 9.6 and I've over-wrote the patch with the Fedora one.

Also Fedora does:

# FIXME: Force a newer gettext version to workaround `autoreconf -i` errors
# with coreutils 9.6 and bundled gettext 0.19.2 from gettext-common-devel.
sed -i 's/0.19.2/0.22.5/' bootstrap.conf configure.ac
 
autoreconf -fiv

for the autoreconf issue.

[Xi Ruoyao]

It seems tests/cp/preserve-mode.sh and tests/mv/acl.sh now can pass normally in the chroot environment.

[Bruce Dubbs]

Noteworthy changes in release 9.6 (2025-01-17) [stable]

Bug fixes

• cp fixes support for --update=none-fail, which would have been rejected as an invalid option. [bug introduced in coreutils-9.5]

• cp,mv --update no longer overrides --interactive or --force. [bug introduced in coreutils-9.3]

• csplit no longer creates empty files given empty input. [This bug was present in "the beginning".]

• ls and printf fix shell quoted output in the edge case of escaped first and last characters, and single quotes in the string. [bug introduced in coreutils-8.26]

• ls -l no longer outputs "Permission denied" errors on NFS which may happen with files without read permission, and which resulted in inaccurate indication of ACLs (missing '+' flag after mode). [bug introduced in coreutils-9.4]

• ls -l no longer outputs "Not supported" errors on virtiofs. [bug introduced in coreutils-9.4]

• mv works again with macFUSE file systems. Previously it would have exited with a "Function not implemented" error. [bug introduced in coreutils-8.28]

• nproc gives more consistent results on systems with more than 1024 CPUs. Previously it would have ignored the affinity mask on such systems. [bug introduced with nproc in coreutils-8.1]

• numfmt --from=iec-i now works with numbers without a suffix. Previously such numbers were rejected with an error. [bug introduced with numfmt in coreutils-8.21]

• printf now diagnoses attempts to treat empty strings as numbers, as per POSIX. For example, "printf '%d' " now issues a diagnostic and fails instead of silently succeeding. [This bug was present in "the beginning".]

• pwd no longer outputs an erroneous double slash on systems where the system getcwd() was completely replaced. [bug introduced in coreutils-9.2]

• 'shuf' generates more-random output when the output is small. [bug introduced in coreutils-8.6]

• tail --follow=name no longer waits indefinitely for watched file names that are moved elsewhere within the same file system. [bug introduced in coreutils-8.24]

• tail --follow without --retry, will consistently exit with failure status where inotify is not used, when all followed files become inaccessible. [This bug was present in "the beginning".]

• tail --follow --pid=PID will now exit when the PID dies, even in the presence of blocking inputs like unopened fifos. [This bug was present in "the beginning".]

• 'tail -c 4096 /dev/zero' no longer loops forever. [This bug was present in "the beginning".]

Changes in behavior

• 'factor' now buffers output more efficiently in some cases.

• install -C now dereferences symlink sources when comparing, rather than always treating as different and performing the copy.

• kill -l and -t now list signal 0, as it's a valid signal to send.

• ls's -f option now simply acts like -aU, instead of also ignoring some earlier options. For example 'ls -fl' and 'ls -lf' are now equivalent because -f no longer ignores an earlier -l. The new behavior is more orthogonal and is compatible with FreeBSD.

• stat -f -c%T now reports the "fuseblk" file system type as "fuse", given that there is no longer a distinct "ctl" fuse variant file system.

New Features

• cksum -a now supports the "crc32b" option, which calculates the CRC of the input as defined by ITU V.42, as used by gzip for example. For performance pclmul instructions are used where supported.

• ls now supports the --sort=name option, to explicitly select the default operation of sorting by file name.

• printf now supports indexed arguments, using the POSIX:2024 specified %<i>$ format, where '<i>' is an integer referencing a particular argument, thus allowing repetition or reordering of printf arguments.

• test supports the POSIX:2024 specified '<' and '>' operators with strings, to compare the string locale collating order.

• timeout now supports the POSIX:2024 specified -f, and -p short options, corresponding to --foreground, and --preserve-status respectively.

Improvements

• cksum -a crc, makes use of AVX2, AVX512, and ARMv8 SIMD extensions for time reductions of up to 40%, 60%, and 80% respectively.

• 'head -c NUM', 'head -n NUM', 'nl -l NUM', 'nproc --ignore NUM', 'tail -c NUM', 'tail -n NUM', and 'tail --max-unchanged-stats NUM’ no longer fail merely because NUM stands for 2⁶⁴ or more.

• sort operates more efficiently when used on pseudo files with an apparent size of 0, like those in /proc.

• stat and tail now know about the "bcachefs", and "pidfs" file system types. stat -f -c%T now reports the file system type, and tail -f uses inotify for these file systems.

• wc now reads a minimum of 256KiB at a time. This was previously 16KiB and increasing to 256KiB was seen to increase wc -l performance by about 10% when reading cached files on modern systems.

[Bruce Dubbs]

Fixed at commit 4f401fb372.