`tail -n2 in >> in` already operated in a bounded manner,
so this makes tailing from the start with `tail -n+2 in >> in`
bounded also, operating only on the original file contents
and not increasing the size of 'in' forever.
Note this bounded operation does not apply with -f.
* src/tail.c (remaining_input): A new helper to return
the limit of the original file when output is the same file.
(tail_bytes): Use the new limit from remaining_input().
(tail_lines): Likewise.
* tests/tail/append-to-input.sh: Adjust accordingly.
Note this test was failing on Alpine due to its
different buffering behavior. We're now consistent
over all file sizes and buffering behaviour in this case.
* NEWS: Mention the improvement.
Test failure reported by Bruno Haible.
* src/tail.c (tail_forever_inotify): Size the buffer based on NAME_MAX
instead of the operands. Reduce the scope of some variables.
* tests/tail/inotify-long-name.sh: New file.
* tests/local.mk (all_tests): Add the new test.
* NEWS: Mention the bug fix. Adjust the reference in a previously fixed
bug to mention it was introduced with inotify support.
Discovered by Anthropic's OSS Scanner, as bug ANT-2026-S9L24ERC.
* src/tail.c (tail_forever_inotify): Fallback to polling upon overlow.
* tests/tail/inotify-overflow.sh: Add a new test.
* tests/local.mk: Reference the new test.
* NEWS: Mention the bug fix.
Discovered by Anthropic's OSS Scanner, as bug ANT-2026-UUNT73BK.
Replace the O(N) scan of watched names on each directory inotify event
with a Hash_table keyed by (parent_wd, basename), addressing the
long-standing FIXME in tail_forever_inotify.
Keep ownership consistent with wd_to_name: parent_by_name is created
only for Follow_name and returned to the caller, so fallback paths can
return without local hash cleanup.
Hash basenames with hash_string() and mix in parent_wd, matching the
existing gnulib helper rather than a hand-rolled string hash.
Matching cost for N directory events that each compare against N names
(illustrative host timings of the scan vs hash lookup alone):
N=1000: ~0.018s -> ~0.000s
N=5000: ~0.39s -> ~0.000s
N=10000: ~1.5s -> ~0.000s
End-to-end recreate storms remain largely file-system bound on this
host; the hash removes the quadratic name-matching CPU from that path.
* src/tail.c (parent_name_hasher, parent_name_comparator): New.
(tail_forever_inotify): Build and use parent_by_name; return it to the
caller.
(main): Free parent_ht when non-NULL.
* tests/tail/inotify-parent-hash.sh: Require notification mode; cover
multiple names in one directory and the same basename in two
directories.
* tests/local.mk: Reference the new test.
* NEWS: Mention the improvement.
This was seen with a NetBSD failure in the pipe-f2.sh test
due to it clearing the EOF indication on a fifo, and thus
blocking in the subsequent read. That was discussed at:
https://lists.gnu.org/r/coreutils/2026-04/msg00129.htmlhttps://bugs.gnu.org/81923
Note the NetBSD issue may be addressed in NetBSD 12 as
discussed in the bug above.
This is more defensive/consistent in the case
where a subsequent read on a file blocks.
We might be operating on non fifos after all.
* src/tail.c (tail_forever): Flush output before we might block.
* tests/tail/pipe-f2.sh: Test non inotify path on all platforms.
* NEWS: Mention the improvement.
`strace -e tee true` will run the tee command on cygwin,
rather than filtering for the tee syscall.
Reported by Bruno Haible on Cygwin 2.9.0
* init.cfg: Use -e trace=foo where foo might be a command,
so that strace probes will fail rather than invoke foo on Cygwin.
* tests/cat/cat-distinct-err.sh: Likewise.
* tests/cat/splice.sh: Likewise.
* tests/ls/no-cap.sh: Likewise.
* tests/misc/yes.sh: Likewise.
* tests/mv/atomic.sh: Likewise.
* tests/mv/atomic2.sh: Likewise.
* tests/nproc/nproc-quota.sh: Likewise.
* tests/pwd/pwd-long.sh: Likewise.
* tests/stty/stty.sh: Likewise.
* tests/tail/debug.sh: Likewise.
* tests/tail/inotify-only-regular.sh: Likewise.
It seems that NetBSD requires a fifo to be open, not merely opened,
to ensure a subsequent open proceeds. Immediately closing the opened
fifo caused intermittent failures on NetBSD 10.0. While that seems
like a bug in NetBSD, we can avoid the issue.
* tests/tail/pid-pipe.sh: Hold the fifo open until our writer
has opened it. This also allows specifying a single timeout
(of 12.7s), rather than 2 separate timeouts.
* tests/tail/inotify-rotate-resources.sh: Use require_inotify_supported_
instead of require_local_dir_ and the recently added check for
overlayfs.
Suggested by Pádraig Brady.
* src/tail.c (write_header): Use quotef() to quote
file names when needed.
* src/head.c (write_header): Likewise.
* tests/head/quote-headers.sh: Add a new test.
* tests/tail/quote-headers.sh: Add a new test.
* tests/local.mk: Reference the new test.
Reported by Michał Majchrowicz.
I noticed a single failure on cfarm29 (Linux 6.12 ppc64le),
where the output was missing:
-tail: directory containing watched file was removed
-tail: inotify cannot be used, reverting to polling
-tail: 'dir/file' has appeared; following new file
* tests/tail/inotify-dir-recreate.sh: Leverage the new tail --debug
output to sync to a point where we know the inotify watches are in
place. Also normalize the file argument quoting a bit.
This avoids the test harness being terminated like:
make[1]: *** [Makefile:24419: check-recursive] Hangup
make[3]: *** [Makefile:24668: check-TESTS] Hangup
make: *** [Makefile:24922: check] Hangup
make[2]: *** [Makefile:24920: check-am] Hangup
make[4]: *** [Makefile:24685: tests/misc/usage_vs_refs.log] Error 129
...
This happened sometimes when the tests were being run non interactively.
For example when run like:
setsid make TESTS="tests/timeout/timeout.sh \
tests/tail/overlay-headers.sh" SUBDIRS=. -j2 check
Note the race window can be made bigger by adding a sleep
after tail is stopped in overlay-headers.sh
The race can trigger the kernel to induce its job control
mechanism to prevent stuck processes.
I.e. where it sends SIGHUP + SIGCONT to a process group
when it determines that group may become orphaned,
and there are stopped processes in that group.
* tests/tail/overlay-headers.sh: Use setsid(1) to keep the stopped
tail process in a separate process group, thus avoiding any kernel
job control protection mechanism.
* tests/timeout/timeout.sh: Use setsid(1) to avoid the kernel
checking the main process group when sleep(1) is reparented.
Fixes https://bugs.gnu.org/80477
* tests/tail/overlay-headers.sh (cleanup_): Ensure we send SIGCONT
to the tail process, otherwise we would hang if the test is terminated
while the tail process is in stopped state.
(wait4stopped_): A new function to ensure tail is in the stopped state
before we start writing to the monitored files.
Also remove "---disable-inotify" from $fastpoll so we actually
test the inotify code (where supported).
Also remove the timeout(1) wrapper, so we actually suspend tail(1).
Reported by Bruno Haible on macOS 26
* init.cfg (require_inotify_supported_): A new function that
indicates inotify support for a file in the current directory,
using tail --debug, rather than awkward strace checks etc.
* tests/tail/inotify-dir-recreate.sh: Call require_inotify_supported_.
tests/tail/inotify-only-regular.sh: Likewise.
tests/tail/inotify-race.sh: Likewise.
tests/tail/inotify-race2.sh: Likewise.
* doc/coreutils.texi (tail invocation): Describe --debug.
* src/tail.c (tail_forever, tail_forever_inotify): Output
which --follow implementation is being used.
* tests/tail/debug.sh: Add a new test.
* tests/local.mk: Reference the new test.
* NEWS: Mention the new feature.
* src/tail.c (file_lines): Seek to the previous block instead of the
beginning (or a little before) of the block that was just scanned.
Otherwise, the same block is read and scanned (at least partially)
again. This bug was introduced by commit v9.7-219-g976f8abc1.
* tests/tail/basic-seek.sh: Add a new test.
* tests/local.mk: Reference the new test.
* NEWS: mention the bug fix.
* tests/tail/inotify-dir-recreate.sh: Add an extra check
that inotify is in use, as it's required for this test.
Inotify is avoided with overlayfs for which the
df --local check is not sufficient exclusion for.
* tests/tail/overlay-headers.sh: Protect tail invocation with timeout,
and extend the possible running period to 60 seconds like other tests.
Reported by Brudno Haible on T2SDE Linux/alpha
* tests/tail/wait.sh: This test was seen to hang occasionally
on an Alpine Linux 3.20 system, so protect the tail(1) call
with `timeout 60` as done in similar tests.
Reported by Bruno Haible.
* src/tail.c (tail_bytes): New function.
(tail_bytes, tail_lines, tail): Accept struct stat pointer from
caller instead of calling fstat ourselves. All callers changed.
(tail_file): Skip a call to fstat if fstat already failed.
* tests/tail/follow-stdin.sh: Adjust to match new behavior
on failure, which omits a redundant diagnostic.
This better matches the treatment of POSIX form,
e.g., ‘tail +Nc’ is now like ‘tail -c +N’ even when N is large.
* src/tail.c: Don’t include xstrtol.h.
(parse_obsolete_option): Treat numbers greater than UINTMAX_MAX as
if they are UINTMAX_MAX. Parse the number by hand with saturating
arithmetic; nowadays that’s simpler than using xstrtoumax. There
is no need for a diagnostic now, as the error cannot happen any more.
* tests/tail/tail.pl (obs-plus-c3): New test.
* tests/tail/tail-c.sh: On Solaris 11, tail -c 4096 /dev/urandom,
will induce an lseek(,-4096,SEEK_END) which returns -4096 without
setting errno, and a subsequent read() then gives EINVAL.
Since tailing the end of a psuedo device is an edge case,
we just verify that we don't spin reading the device forever.
* src/tail.c (tail_file): Open files with O_NONBLOCK
if we might need async processing.
(pipe_bytes): Ignore EAGAIN read() errors.
(pipe_lines): Likewise.
* tests/tail/pid-pipe.sh: Add a new test.
* tests/local.mk: Reference the new test.
* NEWS: Mention the bug fix.
Reported by Berhard Voelker.
Update to latest gnulib with new copyright year.
Run "make update-copyright" and then...
* gnulib: Update included in this commit as copyright years
are the only change from the previous gnulib commit.
* tests/init.sh: Sync with gnulib to pick up copyright year.
* bootstrap: Likewise.
* tests/sample-test: Adjust to use the single most recent year.
* src/tail.c (tail_forever): Without --retry, exit with failure
status like we do for the inotify case (since v8.11-15-g61de57cd2).
This is also consistent with the failure exit if no file was
accessible at tail startup.
* tests/tail/follow-stdin.sh: Tweak due to earlier exit.
* tests/tail/follow-name.sh: Test with and without inotify.
* NEWS: Mention the bug fix.
Require --retry to continue to track files upon rename.
We already unfollowed a file if it was renamed
to another file system (unlinked), so this makes the behavior
consistent if renaming to a file in the same file system.
I.e. --follow=name without --retry, means unfollow if the
name is unlinked or moved, so this change ensures that
behavior for all rename cases.
Related commits: v8.0-121-g3b997a9bc, v8.23-161-gd313a0b24
* src/tail.c (tail_forever_notify): Remove watch for a renamed file
if --retry is not specified.
* tests/tail/F-vs-rename.sh: Related test cleanup.
* tests/tail/follow-name.sh: Add a test case.
* NEWS: Mention the bug fix.
Fixes https://bugs.gnu.org/74653
* src/tail.c (tail_lines): If skipping all input, use lseek if
possible.
(parse_options): Allow counts to exceed 2**64.
(main): Don’t subtract 1 from UINTMAX_MAX, since it stands
for infinity in this context.
(main): Also don’t read anything when given infinite elisions.
* tests/tail/tail.pl: Adjust to match new behavior. Rename err-5
test to big-c and expect the invocation to succeed, since ‘tail
-c99999999999999999999’ now succeeds instead of (unnecessarily)
failing.
Problem reported by Ionut Nicula in:
https://bugs.gnu.org/70477
* src/tail.c (tail_bytes): Do not loop forever on commands
like 'tail -c 4096 /dev/zero'.
* tests/tail/tail-c.sh: Test this fix.
Update to latest gnulib with new copyright year.
Run "make update-copyright" and then...
* gnulib: Update included in this commit as copyright years
are the only change from the previous gnulib commit.
* tests/init.sh: Sync with gnulib to pick up copyright year.
* bootstrap: Manually update copyright year,
until we fully sync with gnulib at a later stage.
* tests/sample-test: Adjust to use the single most recent year.
* src/tail.c (file_lines): Ensure we use a buffer size >= PAGE_SIZE when
searching backwards to avoid seeking within a file,
which on sysfs files is accepted but also returns no data.
* tests/tail/tail-sysfs.sh: Add a new test.
* tests/local.mk: Reference the new test.
* NEWS: Mention the bug fix.
Fixes https://bugs.gnu.org/67490
tail can watch multiple files, but currently only a single writer. It
can be useful to watch files from multiple writers, or even processes
not directly related to the files (e.g. watch log files written by a
server process, for the duration of a test driven by a separate
client).
* src/tail.c (writers_are_dead): New function.
(tail_forever): Use it to wait for writers.
(tail_forever_inotify): As above.
(parse_options): Manage --pid options in an array.
* doc/coreutils.texi: Update documentation.
* tests/tail/pid.sh: Add a variant with two PIDs.
* News: Mention the new feature.