GNU coreutils 9.12: remaining test failures - Nov 2026 #14705

This catalogues the remaining GNU coreutils 9.12 test-suite failures so
they can be tracked and worked on individually. It replaces #13841.

Refreshed 2026-11-03 from the GnuTests CI run on main (c6367b8,
native + root + SELinux + SMACK runs aggregated):
720 tests, 665 PASS, 37 FAIL, 3 ERROR, 15 SKIP (September: 571 PASS, 63 FAIL, 5 ERROR, 78 SKIP).
The failure list is identical on the two previous main runs.

Entries fixed since September are folded at the bottom. A PR number next to an
open entry means a fix is under review.

FAILED Tests

• cp/cp-mv-enotsup-xattr.sh (root): mv to a filesystem without xattr support must warn setting attributes for ...: Operation not supported - PR mv: warn when the destination doesn't support xattrs #15101
• cut/bounded-memory.sh: cut must use bounded memory - PR cut: bound memory usage for streamable selections #15094
• date/date-debug.sh: date --debug output (none of the parsed ... part / warning: lines are printed)
• dd/fail-ftruncate-fstat.sh (strace): dd must stop copying when ftruncate/fstat fail
• dd/misc.sh: dd and --, plus the strace-injected error paths
• df/over-mount-device.sh (root): df /dev/loopN must report cannot access ...: over-mounted by another device - PR df: report an over-mounted device instead of falling back to statfs #15977
• env/env-signal-handler.sh: env --default-signal=PIPE, --list-signal-handling - PR env: list the state of the signals, not the actions taken #15186 (part of it)
• join/join.pl: -o field lists given as separate, unquoted operands (-o 1.1 2.2) - PR join: accept separate -o field operands #15408
• misc/close-stdin.sh: programs run with stdin closed must diagnose EBADF - still dd, factor, numfmt, stty, tsort
• misc/io-errors.sh: programs diagnose write errors (intermittent)
• misc/printenv.sh: printenv must honour QUOTING_STYLE - PR printenv: Honor QUOTING_STYLE environment variable #14857
• misc/read-errors.sh: programs diagnose read errors
• misc/responsive.sh: programs stay responsive to input - PR make the input responsive  #16088
• misc/stdbuf.sh: stdbuf line/block buffering (intermittent)
• misc/traversal-missing.sh: traversal races must be ignored (chmod reports Permission denied where GNU reports the race)
• misc/warning-errors.sh: failure to write warnings is handled gracefully
• misc/write-errors.sh: ~40 utilities still fail to diagnose EBADF/ENOSPC, and cut -b1- /dev/zero aborts on a 16MiB allocation
• mv/mv-special-1.sh, mv/mv-special-2.sh (root): moving a large file, a device node and a socket across filesystems fails with Permission denied - see PR mv: recreate special files in directories moved across devices #15123, PR mv: report specific errors instead of generic permission denied on cross-device fallback #16057
• nproc/nproc-quota.sh (strace): cgroup-quota injection for nproc
• od/od.pl: bad -j/-N arguments quoted as $'a\t' instead of 'a\t', and -j/-N/-S overflow is accepted (exit 0) instead of argument ... too large - claimed by @ChihweiLHBird
• pr/bounded-memory.sh: pr on files larger than memory (intermittent)
• pr/options.sh: pr numeric option diagnostics (invalid page range '0', '-l PAGE_LENGTH' invalid number of lines, ...) - PR pr: report invalid numeric options and page ranges as GNU does #15176
• pr/pr-tests.pl: comprehensive pr page-formatting tests
• ptx/ptx.pl: invalid -S/-W regex must print a one-line Invalid regular expression error and exit 1 - PR ptx: single-line error for invalid -S regexp #14904
• rm/rm-readdir-fail.sh (LD_PRELOAD): simulates readdir failures - PR rm: overriding the libc call when implementation does not match gnu coreutils exactly #9604
• shred/fifo.sh: shred on a FIFO - invalid file type / failed to open for writing diagnostics - PR shred: report a FIFO, a directory and a missing file as GNU does #15182
• shred/shred-remove.sh: shred --remove in an unwritable directory (failed to remove: Permission denied)
• tail/debug.sh: tail --debug output
• tail/follow-name.sh: --follow=name must not imply --retry (intermittent)
• tail/inotify-dir-recreate.sh: switch to polling when the directory is recreated (intermittent)
• tail/inotify-rotate-resources.sh: tail -F must not leak inotify resources
• tail/pid-pipe.sh: tail --pid=PID fifo exits responsively (intermittent) - see PR tail: keep named inputs nonblocking with --pid #15479
• wc/wc.pl: wc -c -m on multibyte input reports the wrong count - PR wc: honor single-byte character locales (fixes the wc test) #16087

SMACK (qemu run)

• id/smack.sh, mkdir/smack-no-root.sh, mkdir/smack-root.sh: the qemu image lacks the test helpers (print_ver_: not found, require_smack_: not found) - a harness problem, not a utility bug

ERROR Tests

• env/env.sh, env/env0-from.sh: env --debug must quote the executed command (executing: './truetrue') - NOTE: upstream is working on changing this
• env/printenv.sh (LD_PRELOAD): printenv with a corrupted environ

SKIPPED Tests

• id/gnu-zero-uids.sh: id must fail for processes with zero UIDs
• install/stdin.sh: install reading from /dev/stdin
• misc/tty-quoting.sh: quoting when stdout is a terminal
• sort/sort-locale.sh: locale-aware sort ordering
• wc/wc-sjis.sh: Shift_JIS counting (locale missing on the runner)

Skipped on purpose or for lack of runner support: misc/getopt_vs_usage.sh
and misc/usage_vs_refs.sh (patched out in util/gnu-patches/, PR #14717),
chcon/traversal-missing.sh, df/skip-rootfs.sh, ls/no-cap.sh.

INCOMPATIBLE Tests

Tests using LD_PRELOAD or gdb breakpoints cannot be fixed without matching
internal implementation details (skipped in CI):

• cp/nfs-removal-race.sh (LD_PRELOAD): NFS client removal race during cp S D
• df/skip-duplicates.sh (LD_PRELOAD): duplicate mount-list entry injection
• rm/r-root.sh (LD_PRELOAD + gdb): tries to remove / recursively
• tail/inotify-race.sh, tail/inotify-race2.sh (gdb): races during the initial read

Several failures above (dd/fail-ftruncate-fstat, dd/misc,
misc/read-errors, misc/traversal-missing, nproc/nproc-quota,
shred/fifo, tail/debug) use strace fault injection; they are not
structurally impossible - they fail because our syscall sequence or our error
messages differ.