perf: per-goroutine DFA cache + AC prefilter + integrated loop

CHANGELOG.md

@@ -12,6 +12,82 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 - ARM NEON SIMD support (Go 1.26 `simd/archsimd` intrinsics — [#120](https://github.com/coregx/coregex/issues/120))
 - SIMD prefilter for CompositeSequenceDFA (#83)
 
+## [0.12.15] - 2026-03-21
+
+### Performance
+- **Per-goroutine DFA cache** (Rust approach) — split lazy DFA into immutable `DFA`
+  (shared across goroutines) + mutable `DFACache` (pooled via `sync.Pool`).
+  Eliminates all data races on concurrent `IsMatch`/`Find`/`FindSubmatch`.
+  Concurrent case-insensitive IsMatch: data race → **450 ns/op** (zero-contention).
+
+- **Pre-computed word boundary match flags** — `matchAtWordBoundary` /
+  `matchAtNonWordBoundary` flags computed during `determinize()`, replacing
+  per-byte `checkWordBoundaryMatch` (was 30% CPU). Word boundary check:
+  **30% → 0.3% CPU**.
+
+- **Aho-Corasick DFA prefilter for >64 literals** — for patterns exceeding
+  FatTeddy's 64-pattern limit, AC DFA provides zero false positives with
+  O(n) flat transition table scan. FatTeddy retained for 33-64 patterns.
+
+- **Integrated prefilter+DFA loop** (Rust approach) — replaced two-pass
+  `prefilter.Find → DFA.searchAnchored → repeat` with single DFA loop where
+  dead-state transitions trigger prefilter skip-ahead. Eliminates function call
+  overhead between passes.
+
+- **Eliminated double prefilter scan** — `isMatchDFA` called prefilter externally,
+  then `DFA.IsMatch` called it again internally. Removed redundant external call.
+
+- **Strategy DFA caches in SearchState** — reverse searchers (ReverseSuffix,
+  ReverseInner, etc.) had per-call `sync.Pool` access for DFA caches. Now
+  strategy-specific caches live in `SearchState`, eliminating N×pool.Get/Put
+  per FindAll iteration.
+
+### Fixed
+- **`.*` newline boundary** — ReverseSuffix/ReverseSuffixSet `matchStartZero`
+  path crossed `\n` boundaries. Pattern `.*@example.com` on multi-line input
+  matched from line 1 across newline into line 2. Fix: `lineStartBefore()`
+  constrains `.*` to single line (matching stdlib behavior).
+
+- **Teddy ignoring anchors** — `(?m)^(GET|POST|...)` selected UseTeddy which
+  doesn't verify `^` anchor, returning matches mid-line. Fix:
+  `hasAnchorAssertions()` prevents UseTeddy for anchored patterns.
+
+- **`(?m)^` multiline anchor in DFA** — DFA can't verify multiline line-start
+  assertions. Fix: `hasMultilineLineAnchor()` routes to UseNFA with prefilter
+  skip-ahead. `http_methods` on 6MB: 73ms (DFA) → 1.56ms (NFA+prefilter).
+
+- **`.*` FindAll with DFA cache clear** — `UseBoth` strategy caused incorrect
+  match positions when DFA cache cleared mid-FindAll for empty-matchable
+  patterns. Fix: `canMatchEmpty()` routes to UseNFA.
+
+- **Partial prefilter on `(?i)` alternation overflow** — literal extractor
+  truncated alternation branches on cross-product overflow (>250), producing
+  a partial prefilter that missed uncovered branches. Fix: return empty Seq
+  on overflow (Rust approach) — NFA handles all branches without prefilter.
+
+- **`expandCaseFoldLiteral` incomplete foldSets** — early exit on
+  CrossProductLimit left trailing nil entries in foldSets array.
+  `generateCaseFoldVariants` produced 0 results. Fix: track `filledCount`.
+
+- **Prefilter `IsComplete` with anchors** — prefilter marked `complete=true`
+  skipped anchor verification. Fix: `WrapIncomplete()` forces engine
+  verification when pattern has anchors.
+
+### Added
+- **Stdlib compatibility test** — 38 patterns (regex-bench + LangArena + edge
+  cases) compared against `regexp` for IsMatch, Find, FindAllIndex, Count.
+  Runs in CI on every PR. 38/38 PASS.
+
+### Refactored
+- `dfa/lazy/cache.go` — `Cache` renamed to `DFACache`, `sync.RWMutex` removed
+  (single-owner per goroutine, no locking needed).
+- `dfa/lazy/start.go` — `StartTable` moved into `DFACache` (mutable start states).
+- `meta/search_state.go` — `SearchState` carries `dfaCache`, `revDFACache`,
+  and strategy-specific caches (`stratFwdCache`, `stratRevCache`).
+- All DFA call sites in `meta/` updated to pass pooled cache (14+ call sites).
+- `prefilter/ahocorasick.go` — AC DFA prefilter for >64 patterns.
+- `prefilter/wrap.go` — `WrapIncomplete()` for anchor-aware prefilter gating.
+
 ## [0.12.14] - 2026-03-19
 
 ### Fixed

ROADMAP.md

@@ -2,7 +2,7 @@
 
 > **Strategic Focus**: Production-grade regex engine with RE2/rust-regex level optimizations
 
-**Last Updated**: 2026-03-19 | **Current Version**: v0.12.14 | **Target**: v1.0.0 stable
+**Last Updated**: 2026-03-20 | **Current Version**: v0.12.15 | **Target**: v1.0.0 stable
 
 ---
 
@@ -12,7 +12,7 @@ Build a **production-ready, high-performance regex engine** for Go that matches
 
 ### Current State vs Target
 
-| Metric | Current (v0.12.14) | Target (v1.0.0) |
+| Metric | Current (v0.12.15) | Target (v1.0.0) |
 |--------|-------------------|-----------------|
 | Inner literal speedup | **280-3154x** | ✅ Achieved |
 | Case-insensitive speedup | **263x** | ✅ Achieved |
@@ -23,7 +23,8 @@ Build a **production-ready, high-performance regex engine** for Go that matches
 | OnePass DFA | **Yes** | ✅ Achieved |
 | Slim Teddy (2-32 patterns) | **Yes (SSSE3, 9GB/s)** | ✅ Achieved |
 | Fat Teddy (33-64 patterns) | **Yes (AVX2, 9GB/s)** | ✅ Achieved |
-| Aho-Corasick (>64 patterns) | **Yes** | ✅ Achieved |
+| Aho-Corasick (>64 patterns) | **Yes (AC DFA prefilter)** | ✅ Achieved |
+| Per-goroutine DFA cache | **Yes (Rust approach)** | ✅ Achieved |
 | BoundedBacktracker | **Yes** | ✅ Achieved |
 | CharClassSearcher | **Yes (35% faster than Rust!)** | ✅ Achieved |
 | **Patterns faster than Rust** | **5 patterns** | ✅ Achieved |
@@ -84,7 +85,9 @@ v0.12.12 ✅ → Prefix trimming for case-fold literals (Teddy instead of AC)
          ↓
 v0.12.13 ✅ → FatTeddy fix, prefilter acceleration, AC v0.2.1
          ↓
-v0.12.14 (Current) ✅ → Concurrent safety fix for isMatchDFA prefilter (#137)
+v0.12.14 ✅ → Concurrent safety fix for isMatchDFA prefilter (#137)
+         ↓
+v0.12.15 (Current) ✅ → Per-goroutine DFA cache, word boundary 30%→0.3% CPU, AC prefilter
          ↓
 v1.0.0-rc → Feature freeze, API locked
          ↓
@@ -121,7 +124,13 @@ v1.0.0 STABLE → Production release with API stability guarantee
 - ✅ **v0.12.6**: BoundedBacktracker span-based CanHandle, ReplaceAllStringFunc O(n) (#127)
 - ✅ **v0.12.7**: PikeVM sparse-dispatch for `.` patterns, 2.8-4.8x speedup (#132)
 - ✅ **v0.12.8**: Streaming ReplaceAll + DFA-first FindSubmatchAt, Rust-style two-phase search (#135)
-- ✅ **v0.12.8**: Streaming ReplaceAll + DFA-first FindSubmatchAt, Rust-style two-phase search (#135)
+- ✅ **v0.12.9**: Bidirectional DFA, Teddy/reverse NFA fixes
+- ✅ **v0.12.10**: Case-insensitive literals, DigitPrefilter O(n²) fix (#137)
+- ✅ **v0.12.11**: ReverseSuffix multi-wildcard, COREGEX_DEBUG logging
+- ✅ **v0.12.12**: Prefix trimming for case-fold literals
+- ✅ **v0.12.13**: FatTeddy fix (ANDL→ORL, VPTEST), prefilter acceleration, AC v0.2.1
+- ✅ **v0.12.14**: Concurrent safety fix for isMatchDFA prefilter (#137)
+- ✅ **v0.12.15**: Per-goroutine DFA cache (Rust approach), word boundary 30%→0.3% CPU, AC DFA prefilter for >32 literals (7-13x faster)
 
 ---
 

dfa/lazy/accel_test.go

@@ -132,6 +132,7 @@ func TestDFAAccelerate(t *testing.T) {
 	if err != nil {
 		t.Fatalf("Failed to compile DFA: %v", err)
 	}
+	_ = dfa.NewCache() // accelerate is an immutable DFA method, cache not needed here
 
 	haystack := []byte("aaaaaaaaaafoobar")
 
@@ -175,6 +176,7 @@ func TestAccelerableStateInSearch(t *testing.T) {
 	if err != nil {
 		t.Fatalf("Failed to compile DFA: %v", err)
 	}
+	cache := dfa.NewCache()
 
 	// Test that search still works correctly
 	tests := []struct {
@@ -188,7 +190,7 @@ func TestAccelerableStateInSearch(t *testing.T) {
 	}
 
 	for _, tc := range tests {
-		got := dfa.Find([]byte(tc.haystack))
+		got := dfa.Find(cache, []byte(tc.haystack))
 		if got != tc.want {
 			t.Errorf("Find(%q) = %d, want %d", tc.haystack, got, tc.want)
 		}
@@ -200,6 +202,7 @@ func BenchmarkAccelerate(b *testing.B) {
 	if err != nil {
 		b.Fatalf("Failed to compile DFA: %v", err)
 	}
+	_ = dfa.NewCache() // accelerate is an immutable DFA method
 
 	haystack := make([]byte, 4096)
 	for i := range haystack {