tail: fix -f/-F not following files on Windows (#4827)

[crutkas]

Problem

Fixes #4827 — tail -f / tail -F does not follow a growing file on Windows. tail prints the initial tail, then never emits appended content.

Root cause

The follow event loop in src/uu/tail/src/follow/watch.rs (Observer::handle_event) matched:

• Modify(Data(DataChange::Any))
• Modify(Metadata(Any | WriteTime))
• Modify(Name(RenameMode::To))
• Create(..)

…but not the bare Modify(ModifyKind::Any).

notify's Windows ReadDirectoryChangesW backend maps every file write (FILE_ACTION_MODIFIED) to exactly EventKind::Modify(ModifyKind::Any) (notify/src/windows.rs). So on Windows every append/truncate event fell through to _ => {} and was silently dropped — the file was watched, events arrived, paths matched, but nothing was ever read/printed.

Linux inotify emits the more specific Modify(Data(..)), which is why the bug is Windows-only. The macOS/BSD kqueue backend also emits Modify(ModifyKind::Any) for NOTE_WRITE, so this fix hardens follow there too. The polling fallback only emits Data/Metadata, so this arm is a no-op for it and for inotify.

Fix

Add ModifyKind::Any to the match arm (one line + explanatory comment). It routes through the existing tailable/truncation logic, so appends and in-place truncation are both handled.

Tests

Re-enabled the seven Windows follow tests that were disabled with // FIXME: test times out / for currently not working platforms — they timed out because of this bug and now pass:

• Append following: test_follow_single, test_follow_multiple, test_follow_name_multiple, test_n0_with_follow
• Truncation: test_follow_name_truncate1, test_follow_name_truncate2, test_follow_name_truncate3

These are genuine regression guards (verified on Windows):

• Without the fix: test_follow_single and test_follow_name_truncate1 FAIL.
• With the fix: full Windows follow suite is green.

Known limitation (out of scope for this PR)

test_follow_name_truncate4 and the move/rename follow tests remain Windows-disabled for a different reason: MetadataExtTail::file_id_eq() is stubbed to false on Windows in paths.rs (it needs GetFileInformationByHandle / the windows_by_handle API to compare file identity). Truncate-to-same-content and rename detection depend on that, so tail spuriously reports "file has been replaced". That is a separate follow-up.

[codspeed-hq]

Merging this PR will degrade performance by 4.62%

❌ 1 regressed benchmark
✅ 330 untouched benchmarks
⏩ 46 skipped benchmarks¹

Warning

Please fix the performance issues or acknowledge them on CodSpeed.

Performance Changes

	Mode	Benchmark	BASE	HEAD	Efficiency
❌	Simulation	sort_ascii_utf8_locale	15.4 ms	16.1 ms	-4.62%

Tip

Investigate this regression by commenting @codspeedbot fix this regression on this PR, or directly use the CodSpeed MCP with your agent.

---

_{Comparing crutkas:fix/tail-follow-windows-4827 (120a4e8) with main (d967066)}

Footnotes

1. 46 benchmarks were skipped, so the baseline results were used instead. If they were deleted from the codebase, click here and archive them to remove them from the performance reports. ↩

[github-actions]

GNU testsuite comparison:

Skip an intermittent issue tests/date/date-locale-hour (fails in this run but passes in the 'main' branch)
Congrats! The gnu test tests/cp/sparse-to-pipe is no longer failing!
Note: The gnu test tests/tail/pipe-f is now being skipped but was previously passing.