globset: add fast path for case-insensitive literal/basename patterns

[dhmztr]

Summary

Addresses #1086: case-insensitive glob patterns that are pure literals (no wildcards) — e.g. a bare filename with .case_insensitive(true) — were always falling through to full regex matching, even though there's no wildcard to justify it. On a large pattern set this is orders of magnitude slower than the literal fast path already used for case-sensitive literals (see the benchmark in the issue).

Scope note: this covers the Literal and BasenameLiteral strategies, which is what the issue's own ignore::overrides benchmark exercises. ext()/prefix()/suffix() are intentionally left as-is (still falling back to regex for case-insensitive patterns) — ExtensionStrategy and the prefix/suffix MultiStrategyBuilder have no case-insensitive comparison mode, and wiring case-insensitive literals into them without that support would silently produce wrong matches (e.g. a case-insensitive *.TXT pattern failing to match a real photo.TXT file). Extending those is a reasonable follow-up but is a larger, separate change — didn't want to bundle a correctness risk into a perf fix.

Changes

• literal() / basename_literal() no longer bail out early on case_insensitive; they ASCII-lowercase the extracted literal and let it flow into the existing Literal/BasenameLiteral match strategies instead of forcing MatchStrategy::Regex.
• LiteralStrategy / BasenameLiteralStrategy gained a case-insensitive mode: literals are lowercased on insert, lookups lowercase the candidate path/basename before hashing.
• GlobSet::new() now builds separate case-sensitive and case-insensitive instances of each strategy and routes each pattern into the right one.
• Lowercasing is ASCII-only (to_ascii_lowercase, not to_lowercase) to match this crate's actual case-insensitive semantics — regex-syntax is built here without the unicode-case feature, so even the pre-existing (?i) regex fallback can't fold non-ASCII case pairs (confirmed directly: (?i) on a pattern with a non-ASCII literal fails to compile at all in this build with UnicodeCaseUnavailable). Using Unicode-aware lowercasing in the new fast path would have made it behave differently from the regex path it replaces for the same logical pattern.

Test plan

• cargo test -p globset — 293 passed, 0 failed (added case-insensitive literal/basename correctness tests, plus fixed two pre-existing tests that encoded the old bail-out-to-None behavior as expected)
• cargo test -p ignore (direct consumer) — 182 unit + 5 integration + 7 doctests, all pass
• cargo fmt -p globset -- --check — clean
• cargo clippy -p globset --all-targets -- -D warnings — no new warnings introduced by this diff (24 pre-existing warnings elsewhere in the crate, unrelated to this change, left untouched)
• Full workspace build not runnable in this environment (local rustc 1.95, workspace ripgrep/grep-* crates pin 1.96 — unrelated to this change; globset itself only requires 1.88 and builds/tests fine)

[Copilot]

Pull request overview

Improves globset performance for case-insensitive glob patterns that are pure literals (full-path literals and basename-only literals) by avoiding the regex-based fallback and using dedicated literal match strategies instead.

Changes:

• Route case-insensitive literal and basename-literal patterns into dedicated LiteralStrategy / BasenameLiteralStrategy instances.
• Add case-insensitive support to LiteralStrategy and BasenameLiteralStrategy by ASCII-lowercasing stored keys and lowercasing candidates on lookup.
• Update Glob literal/basename-literal extraction to support case-insensitive literals (ASCII-only) and adjust tests accordingly.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 3 comments.

File	Description
crates/globset/src/lib.rs	Splits literal/basename-literal strategies into case-sensitive and case-insensitive variants and wires routing in GlobSet::new().
crates/globset/src/glob.rs	Enables extracting literals/basename-literals for case-insensitive globs (ASCII lowercased) and updates matching/tests.

Comments suppressed due to low confidence (2)

crates/globset/src/lib.rs:771

• matches_into also allocates a lowercased Vec for every case-insensitive candidate lookup. Similar to is_match, it can avoid allocating for already-lowercase candidates by attempting get with the original bytes first and only lowercasing on a miss.

        let hits = if self.1 {
            self.0.get(&candidate.path.to_ascii_lowercase())
        } else {
            self.0.get(candidate.path.as_bytes())
        };

crates/globset/src/lib.rs:823

• BasenameLiteralStrategy::matches_into allocates a lowercased Vec on every case-insensitive lookup. As with is_match, you can avoid allocations for already-lowercase basenames by attempting get with the original bytes first and only lowercasing on miss.

        let hits = if self.1 {
            self.0.get(&candidate.basename.to_ascii_lowercase())
        } else {
            self.0.get(candidate.basename.as_bytes())
        };

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.