[ricorna]

Summary

• Preserve native find visibility when native flags are used by disabling hidden, .ignore, and git-ignore pruning.
• Keep the existing ignore-aware behavior for compact RTK syntax.
• Support positive, ranged, and negated bracket classes such as [123], [0-9], and [!0-9].
• Support ] as the first class member, including patterns such as []a] and [!]].
• Preserve unmatched [ as a literal instead of introducing a malformed-pattern regression.

This addresses the hidden/gitignored and bracket-class false negatives documented in #2995 and its follow-up report. It overlaps the visibility portion of draft #2996, while also covering the bracket-class cases that PR does not address.

Root cause

The ignore::WalkBuilder always pruned hidden and ignored entries, even when RTK was transparently rewriting native find ... -name/-type/-maxdepth syntax. The custom glob matcher only implemented * and ?, so valid native bracket expressions silently returned no matches with exit 0.

Verification

• Added end-to-end tests for hidden and gitignored matches under native syntax.
• Added end-to-end tests for ranged, negated, and literal-] bracket classes.
• Added a regression test for an unclosed literal bracket.
• cargo fmt --all -- --check
• cargo clippy --locked --all-targets -- -D warnings
• cargo test --locked

The complete locked suite passed locally: 2,512 unit tests passed, 8 ignored, plus all integration targets.

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}

[KuSh]

Thanks for the bracket-class work and the fixture-backed tests. Bracket patterns have been handled since #3603 (merged 2026-08-26): the subset rtk walks itself now refuses any -name value containing "[" and hands the command to real find, so on current develop "a[0-9].txt", "a[!0-9].txt" and "[]a]*.txt" all return the right files without rtk implementing fnmatch itself. The same merge removed has_native_find_flags and rewrites the compact rtk syntax into find syntax before dispatch, so the native-versus-compact distinction this PR keys its visibility carve-out on no longer exists at that point, and #3854 (merged 2026-09-03) then chose to disclose pruned entries as "... (N filtered: hidden or gitignored)" with a tee file rather than return them. That last decision is still contested: hidden and gitignored matches are indeed missing from the results, and open issue #2995 tracks it, so a narrow PR against the current walker would be the way back in. Please comment or reopen if a case here was missed.