perf(findr): Replaced regex engine with glob.

PERFORMANCE_IDEAS.md

@@ -1,27 +1,34 @@
-findr is ~2.3x slower than fd (case 1: 547ms vs 241ms). Opportunities:
+# Performance Ideas
 
-1. Per-thread result buffers (DONE)
+Current state after regex→glob migration. findr beats fd in 3/4 cases.
-Each thread accumulates results locally, then merges once at exit. Eliminates per-result mutex contention.
 
-2. Batched channel (fd's approach)
+## Benchmark results (2026-06-17)
-Replace global results array + merge with a buffered channel of batches. Each worker fills a local batch (~256 items), sends it to a `chan.Chan([]string)` (capacity = 2 × threads). A receiver thread drains batches and collects/prints. Provides backpressure, streaming output, and per-batch (not global) synchronization. Enables sorting like fd does (buffer first 1000 results or 100ms, then stream).
 
-3. Path allocation waste (join_path/join_path_dir)
+| Case | fd | findr | Ratio |
-Every path construction spins up a strings.Builder, does fmt.sbprintf, to_string, clone, then builder_destroy — 2 heap allocs + 2 frees per path. Could be a simple memcpy into a stack buffer with a single alloc.
+|------|------|-------|-------|
+| 1 `-E .jj` | 172ms | 135ms | **1.27x faster** |
+| 2 `-H` | 1.184s | 1.097s | **1.08x faster** |
+| 3 `-HI` | 1.251s | 1.670s | **1.34x slower** |
+| 4 `-E .git` | 274ms | 202ms | **1.36x faster** |
 
-4. Larger getdents buffer
+Case 3 (`-HI`) skips gitignore entirely, so it's pure I/O + allocation. System time is 2x fd's (12.1s vs 5.5s), pointing to syscall/allocation overhead.
-Currently 8KB. Increasing to 64KB+ means fewer syscalls per directory with many entries.
 
-5. Eliminate entry name cloning
+## Completed
-strings.clone(name) in read_dir_entries heap-allocates per dirent. Names are valid in the getdents buffer during process_dir, so the clone may be unnecessary.
 
-6. Arena allocator per thread
+1. **Per-thread result buffers** — each thread accumulates locally, merges once at exit. Eliminates per-result mutex contention.
-Replace the default allocator for transient strings with a bump allocator — allocate in bulk, free all at once.
+2. **Lean path join** — `join_path`/`join_path_dir` use stack buffer + `copy` + single alloc instead of `strings.Builder` + `fmt.sbprintf` + `clone`.
-2. Path allocation waste (join_path/join_path_dir)
+3. **Regex→glob migration** — replaced regex NFA with backtracking glob matcher. Eliminated 27% of CPU spent on `add_thread`/`is_ignored`. Biggest win.
-Every path construction spins up a strings.Builder, does fmt.sbprintf, to_string, clone, then builder_destroy — 2 heap allocs + 2 frees per path. Could be a simple memcpy into a stack buffer with a single alloc.
+
-3. Larger getdents buffer
+## Remaining ideas
-Currently 8KB. Increasing to 64KB+ means fewer syscalls per directory with many entries.
+
-4. Eliminate entry name cloning
+1. **Larger getdents buffer** (8KB → 64KB+)
-strings.clone(name) in read_dir_entries heap-allocates per dirent. Names are valid in the getdents buffer during process_dir, so the clone may be unnecessary.
+   Fewer syscalls per directory with many entries. Low effort.
-5. Arena allocator per thread
+
-Replace the default allocator for transient strings with a bump allocator — allocate in bulk, free all at once.
+2. **Eliminate entry name cloning**
+   `strings.clone(name)` in `read_dir_entries` heap-allocates per dirent. Names are valid in the getdents buffer during `process_dir`, so the clone may be unnecessary. Low effort.
+
+3. **Arena allocator per thread**
+   Bump allocator for all transient strings, free once at exit. Bigger change, helps everywhere.
+
+4. **Batched channel** (fd's approach)
+   Replace global results array with buffered channel of batches. Enables streaming output and sorting like fd does.

findr.odin

@@ -1,7 +1,6 @@
 package findr
 
 import "core:bufio"
-import "core:fmt"
 import "core:os"
 import "core:strings"
 
@@ -89,3 +88,4 @@ main :: proc() {
 	}
 	bufio.writer_flush(&w)
 }
+

gitignore.odin

@@ -1,112 +1,36 @@
 package findr
 
-import "core:fmt"
 import "core:strings"
-import "core:text/regex"
 
-// FIXME: Use a const bit_set[0..<128; u128] here when we start doing optimizations
+Gitignore :: struct {
-is_regex_meta :: proc(c: u8) -> bool {
+	rules: [dynamic]Rule,
-	switch c {
-	case '.', '+', '(', ')', '{', '}', '^', '$', '|', '#':
-		return true
-	}
-	return false
-}
-
-glob_to_regex :: proc(pattern: string, anchored: bool) -> string {
-	// TODO: Attempt to pre-allocate the string builder when we start doing optimizations
-	sb: strings.Builder
-	strings.builder_init(&sb)
-	defer strings.builder_destroy(&sb)
-
-	if anchored {
-		fmt.sbprintf(&sb, "^")
-	} else {
-		fmt.sbprintf(&sb, "(^|/)")
-	}
-
-	i := 0
-	for i < len(pattern) {
-		c := pattern[i]
-
-		if c == '*' {
-			if i + 1 < len(pattern) && pattern[i + 1] == '*' {
-				prev_slash := i == 0 || pattern[i - 1] == '/'
-				at_end := i + 2 >= len(pattern)
-				next_slash := !at_end && pattern[i + 2] == '/'
-
-				if prev_slash && (next_slash || at_end) {
-					if next_slash {
-						i += 3
-						fmt.sbprintf(&sb, "(.*/)?")
-					} else {
-						i += 2
-						fmt.sbprintf(&sb, ".*")
-					}
-				} else {
-					fmt.sbprintf(&sb, "[^/]*")
-					i += 2
-				}
-			} else {
-				fmt.sbprintf(&sb, "[^/]*")
-				i += 1
-			}
-		} else if c == '?' {
-			fmt.sbprintf(&sb, "[^/]")
-			i += 1
-		} else if c == '[' {
-			append(&sb.buf, '[')
-			i += 1
-			if i < len(pattern) && pattern[i] == '!' {
-				append(&sb.buf, '^')
-				i += 1
-			}
-			if i < len(pattern) && pattern[i] == ']' {
-				append(&sb.buf, ']')
-				i += 1
-			}
-			for i < len(pattern) && pattern[i] != ']' {
-				append(&sb.buf, pattern[i])
-				i += 1
-			}
-			if i < len(pattern) {
-				append(&sb.buf, ']')
-				i += 1
-			}
-		} else if c == '\\' {
-			i += 1
-			if i < len(pattern) {
-				if is_regex_meta(pattern[i]) {
-					append(&sb.buf, '\\')
-				}
-				append(&sb.buf, pattern[i])
-				i += 1
-			}
-		} else if is_regex_meta(c) {
-			append(&sb.buf, '\\')
-			append(&sb.buf, c)
-			i += 1
-		} else {
-			append(&sb.buf, c)
-			i += 1
-		}
-	}
-
-	fmt.sbprintf(&sb, "$")
-
-	s := strings.to_string(sb)
-	result, _ := strings.clone(s)
-	return result
 }
 
 Rule :: struct {
-	regex:    regex.Regular_Expression,
+	pattern:  GlobPattern,
 	negated:  bool,
 	dir_only: bool,
 }
 
-Gitignore :: struct {
+Match :: enum {
-	rules: [dynamic]Rule,
+	None,
+	Ignored,
+	Unignored,
+}
+
+is_ignored :: proc(gi: ^Gitignore, path: string, is_dir: bool) -> bool {
+	return check_match(gi, path, is_dir) == .Ignored
+}
+
+check_match :: proc(gi: ^Gitignore, path: string, is_dir: bool) -> Match {
+	result := Match.None
+	for &rule in gi.rules {
+		if rule.dir_only && !is_dir do continue
+		if glob_match_compiled(&rule.pattern, path) {
+			result = rule.negated ? .Unignored : .Ignored
+		}
+	}
+	return result
 }
 
 parse :: proc(content: string) -> Gitignore {
@@ -148,43 +72,16 @@ parse :: proc(content: string) -> Gitignore {
 
 		if len(s) == 0 do continue
 
-		regex_str := glob_to_regex(s, anchored)
+		gp := glob_compile(s, anchored)
-		re, err := regex.create(regex_str, {regex.Flag.No_Capture})
+		append(&gi.rules, Rule{pattern = gp, negated = negated, dir_only = dir_only})
-		delete(regex_str)
-		if err != nil do continue
-
-		append(&gi.rules, Rule{regex = re, negated = negated, dir_only = dir_only})
 	}
 
 	return gi
 }
 
-Match :: enum {
-	None,
-	Ignored,
-	Unignored,
-}
-
-check_match :: proc(gi: ^Gitignore, path: string, is_dir: bool) -> Match {
-	result := Match.None
-	for rule in gi.rules {
-		if rule.dir_only && !is_dir do continue
-		cap, ok := regex.match(rule.regex, path)
-		regex.destroy(cap)
-		if ok {
-			result = rule.negated ? .Unignored : .Ignored
-		}
-	}
-	return result
-}
-
-is_ignored :: proc(gi: ^Gitignore, path: string, is_dir: bool) -> bool {
-	return check_match(gi, path, is_dir) == .Ignored
-}
-
 destroy :: proc(gi: ^Gitignore) {
-	for rule in gi.rules {
+	for &rule in gi.rules {
-		regex.destroy(rule.regex)
+		glob_destroy(&rule.pattern)
 	}
 	delete(gi.rules)
 }

gitignore_test.odin

@@ -4,100 +4,103 @@ import "core:testing"
 
 @(test)
 test_glob_simple :: proc(t: ^testing.T) {
-	result := glob_to_regex("foo", false)
+	testing.expect(t, glob_match("foo", "foo", false))
-	defer delete(result)
+	testing.expect(t, glob_match("foo", "bar/foo", false))
-	testing.expect_value(t, result, "(^|/)foo$")
+	testing.expect(t, !glob_match("foo", "foobar", false))
+	testing.expect(t, !glob_match("foo", "foo/bar", false))
 }
 
 @(test)
 test_glob_anchored :: proc(t: ^testing.T) {
-	result := glob_to_regex("foo", true)
+	testing.expect(t, glob_match("foo", "foo", true))
-	defer delete(result)
+	testing.expect(t, !glob_match("foo", "bar/foo", true))
-	testing.expect_value(t, result, "^foo$")
+	testing.expect(t, !glob_match("foo", "foobar", true))
 }
 
 @(test)
 test_glob_star :: proc(t: ^testing.T) {
-	result := glob_to_regex("*.log", false)
+	testing.expect(t, glob_match("*.log", "test.log", false))
-	defer delete(result)
+	testing.expect(t, glob_match("*.log", ".log", false))
-	testing.expect_value(t, result, "(^|/)[^/]*\\.log$")
+	testing.expect(t, !glob_match("*.log", "test.txt", false))
+	testing.expect(t, !glob_match("*.log", "dir/test", false))
 }
 
 @(test)
 test_glob_question :: proc(t: ^testing.T) {
-	result := glob_to_regex("?.log", false)
+	testing.expect(t, glob_match("?.log", "a.log", false))
-	defer delete(result)
+	testing.expect(t, !glob_match("?.log", "ab.log", false))
-	testing.expect_value(t, result, "(^|/)[^/]\\.log$")
+	testing.expect(t, !glob_match("?.log", ".log", false))
 }
 
 @(test)
 test_glob_char_class :: proc(t: ^testing.T) {
-	result := glob_to_regex("[abc].log", false)
+	testing.expect(t, glob_match("[abc].log", "a.log", false))
-	defer delete(result)
+	testing.expect(t, glob_match("[abc].log", "b.log", false))
-	testing.expect_value(t, result, "(^|/)[abc]\\.log$")
+	testing.expect(t, !glob_match("[abc].log", "d.log", false))
 }
 
 @(test)
 test_glob_negated_class :: proc(t: ^testing.T) {
-	result := glob_to_regex("[!abc].log", false)
+	testing.expect(t, glob_match("[!abc].log", "d.log", false))
-	defer delete(result)
+	testing.expect(t, !glob_match("[!abc].log", "a.log", false))
-	testing.expect_value(t, result, "(^|/)[^abc]\\.log$")
 }
 
 @(test)
-test_glob_dot_escaped :: proc(t: ^testing.T) {
+test_glob_dot_literal :: proc(t: ^testing.T) {
-	result := glob_to_regex(".env", false)
+	testing.expect(t, glob_match(".env", ".env", false))
-	defer delete(result)
+	testing.expect(t, glob_match(".env", "dir/.env", false))
-	testing.expect_value(t, result, "(^|/)\\.env$")
+	testing.expect(t, !glob_match(".env", "env", false))
+	testing.expect(t, !glob_match(".env", "x.env", false))
 }
 
 @(test)
 test_glob_globstar_prefix :: proc(t: ^testing.T) {
-	result := glob_to_regex("**/foo", false)
+	testing.expect(t, glob_match("**/foo", "foo", false))
-	defer delete(result)
+	testing.expect(t, glob_match("**/foo", "a/b/foo", false))
-	testing.expect_value(t, result, "(^|/)(.*/)?foo$")
+	testing.expect(t, !glob_match("**/foo", "foobar", false))
+	testing.expect(t, !glob_match("**/foo", "a/foobar", false))
 }
 
 @(test)
 test_glob_globstar_suffix :: proc(t: ^testing.T) {
-	result := glob_to_regex("abc/**", false)
+	testing.expect(t, glob_match("abc/**", "abc/x", false))
-	defer delete(result)
+	testing.expect(t, glob_match("abc/**", "abc/x/y", false))
-	testing.expect_value(t, result, "(^|/)abc/.*$")
+	testing.expect(t, !glob_match("abc/**", "abc", false))
+	testing.expect(t, !glob_match("abc/**", "abcd/x", false))
 }
 
 @(test)
 test_glob_globstar_middle :: proc(t: ^testing.T) {
-	result := glob_to_regex("foo/**/bar", false)
+	testing.expect(t, glob_match("foo/**/bar", "foo/bar", false))
-	defer delete(result)
+	testing.expect(t, glob_match("foo/**/bar", "foo/x/bar", false))
-	testing.expect_value(t, result, "(^|/)foo/(.*/)?bar$")
+	testing.expect(t, !glob_match("foo/**/bar", "foo/barx", false))
+	testing.expect(t, !glob_match("foo/**/bar", "foo/x/y/baz", false))
 }
 
 @(test)
 test_glob_backslash_escape :: proc(t: ^testing.T) {
-	result := glob_to_regex("\\!foo", false)
+	testing.expect(t, glob_match("\\!foo", "!foo", false))
-	defer delete(result)
+	testing.expect(t, !glob_match("\\!foo", "foo", false))
-	testing.expect_value(t, result, "(^|/)!foo$")
 }
 
 @(test)
-test_glob_hash_escaped :: proc(t: ^testing.T) {
+test_glob_hash_literal :: proc(t: ^testing.T) {
-	result := glob_to_regex("#foo", false)
+	testing.expect(t, glob_match("#foo", "#foo", false))
-	defer delete(result)
+	testing.expect(t, !glob_match("#foo", "foo", false))
-	testing.expect_value(t, result, "(^|/)\\#foo$")
 }
 
 @(test)
-test_glob_hash_in_pattern :: proc(t: ^testing.T) {
+test_glob_hash_pattern :: proc(t: ^testing.T) {
-	result := glob_to_regex("#*#", false)
+	testing.expect(t, glob_match("#*#", "#test#", false))
-	defer delete(result)
+	testing.expect(t, glob_match("#*#", "##", false))
-	testing.expect_value(t, result, "(^|/)\\#[^/]*\\#$")
+	testing.expect(t, !glob_match("#*#", "test", false))
+	testing.expect(t, !glob_match("#*#", "#test", false))
 }
 
 @(test)
 test_glob_empty :: proc(t: ^testing.T) {
-	result := glob_to_regex("", false)
+	testing.expect(t, glob_match("", "", false))
-	defer delete(result)
+	testing.expect(t, !glob_match("", "foo", false))
-	testing.expect_value(t, result, "(^|/)$")
 }
 
 @(test)

glob.odin

@@ -0,0 +1,210 @@
+package findr
+
+Range :: struct {
+	lo: u8,
+	hi: u8,
+}
+
+Class_Data :: struct {
+	negated: bool,
+	ranges:  [dynamic]Range,
+}
+
+Token_Kind :: enum u8 { Char, Star, Globstar, Question, Class }
+
+Token :: struct {
+	kind:      Token_Kind,
+	byte:      u8,
+	class_idx: u16,
+}
+
+GlobPattern :: struct {
+	tokens:   [dynamic]Token,
+	classes:  [dynamic]Class_Data,
+	anchored: bool,
+}
+
+glob_compile :: proc(pattern: string, anchored: bool) -> GlobPattern {
+	gp: GlobPattern
+	gp.tokens = make([dynamic]Token)
+	gp.classes = make([dynamic]Class_Data)
+	gp.anchored = anchored
+
+	i := 0
+	for i < len(pattern) {
+		c := pattern[i]
+
+		if c == '*' {
+			if i + 1 < len(pattern) && pattern[i + 1] == '*' {
+				prev_slash := i == 0 || pattern[i - 1] == '/'
+				at_end := i + 2 >= len(pattern)
+				next_slash := !at_end && pattern[i + 2] == '/'
+
+				if prev_slash && (next_slash || at_end) {
+					append(&gp.tokens, Token{kind = .Globstar})
+					if next_slash {
+						i += 3
+					} else {
+						i += 2
+					}
+				} else {
+					append(&gp.tokens, Token{kind = .Star})
+					i += 2
+				}
+			} else {
+				append(&gp.tokens, Token{kind = .Star})
+				i += 1
+			}
+		} else if c == '?' {
+			append(&gp.tokens, Token{kind = .Question})
+			i += 1
+		} else if c == '[' {
+			i += 1
+			negated := false
+			if i < len(pattern) && pattern[i] == '!' {
+				negated = true
+				i += 1
+			}
+
+			ranges := make([dynamic]Range)
+
+			if i < len(pattern) && pattern[i] == ']' {
+				append(&ranges, Range{lo = ']', hi = ']'})
+				i += 1
+			}
+
+			for i < len(pattern) && pattern[i] != ']' {
+				if i + 2 < len(pattern) && pattern[i + 1] == '-' && pattern[i + 2] != ']' {
+					append(&ranges, Range{lo = pattern[i], hi = pattern[i + 2]})
+					i += 3
+				} else {
+					append(&ranges, Range{lo = pattern[i], hi = pattern[i]})
+					i += 1
+				}
+			}
+
+			if i < len(pattern) {
+				i += 1
+			}
+
+			class_idx := u16(len(gp.classes))
+			append(&gp.classes, Class_Data{negated = negated, ranges = ranges})
+			append(&gp.tokens, Token{kind = .Class, class_idx = class_idx})
+		} else if c == '\\' {
+			i += 1
+			if i < len(pattern) {
+				append(&gp.tokens, Token{kind = .Char, byte = pattern[i]})
+				i += 1
+			}
+		} else {
+			append(&gp.tokens, Token{kind = .Char, byte = c})
+			i += 1
+		}
+	}
+
+	return gp
+}
+
+match_tokens :: proc(tokens: []Token, classes: []Class_Data, ti: int, path: string, pi: int) -> bool {
+	if ti >= len(tokens) {
+		return pi == len(path)
+	}
+
+	tok := tokens[ti]
+	switch tok.kind {
+	case .Char:
+		if pi < len(path) && path[pi] == tok.byte {
+			return match_tokens(tokens, classes, ti + 1, path, pi + 1)
+		}
+		return false
+
+	case .Question:
+		if pi < len(path) && path[pi] != '/' {
+			return match_tokens(tokens, classes, ti + 1, path, pi + 1)
+		}
+		return false
+
+	case .Star:
+		max_end := pi
+		for max_end < len(path) && path[max_end] != '/' {
+			max_end += 1
+		}
+		for end := max_end; end >= pi; end -= 1 {
+			if match_tokens(tokens, classes, ti + 1, path, end) {
+				return true
+			}
+		}
+		return false
+
+	case .Globstar:
+		if ti + 1 >= len(tokens) {
+			return true
+		}
+		if match_tokens(tokens, classes, ti + 1, path, pi) {
+			return true
+		}
+		for end := pi + 1; end <= len(path); end += 1 {
+			if path[end - 1] == '/' {
+				if match_tokens(tokens, classes, ti + 1, path, end) {
+					return true
+				}
+			}
+		}
+		return false
+
+	case .Class:
+		if pi >= len(path) {
+			return false
+		}
+		cd := classes[tok.class_idx]
+		ch := path[pi]
+		in_range := false
+		for r in cd.ranges {
+			if ch >= r.lo && ch <= r.hi {
+				in_range = true
+				break
+			}
+		}
+		if in_range != cd.negated {
+			return match_tokens(tokens, classes, ti + 1, path, pi + 1)
+		}
+		return false
+	}
+	return false
+}
+
+glob_match_compiled :: proc(gp: ^GlobPattern, path: string) -> bool {
+	tokens := gp.tokens[:]
+	classes := gp.classes[:]
+
+	if gp.anchored {
+		return match_tokens(tokens, classes, 0, path, 0)
+	}
+
+	if match_tokens(tokens, classes, 0, path, 0) {
+		return true
+	}
+	for i := 1; i < len(path); i += 1 {
+		if path[i - 1] == '/' {
+			if match_tokens(tokens, classes, 0, path, i) {
+				return true
+			}
+		}
+	}
+	return false
+}
+
+glob_destroy :: proc(gp: ^GlobPattern) {
+	for &cd in gp.classes {
+		delete(cd.ranges)
+	}
+	delete(gp.classes)
+	delete(gp.tokens)
+}
+
+glob_match :: proc(pattern: string, path: string, anchored: bool) -> bool {
+	gp := glob_compile(pattern, anchored)
+	result := glob_match_compiled(&gp, path)
+	glob_destroy(&gp)
+	return result
+}

walker.odin

@@ -256,7 +256,15 @@ process_dir :: proc(pool: ^WalkerPool, item: WorkItem, local_results: ^[dynamic]
 			if !ignored {
 				child_rel, _ := strings.clone(entry_rel)
 				child_path := join_path(dir_path, entry.name)
-				push_work(pool, WorkItem{path = child_path, rel = child_rel, gi_ctx = gi_ctx, in_repo = child_in_repo})
+				push_work(
+					pool,
+					WorkItem {
+						path = child_path,
+						rel = child_rel,
+						gi_ctx = gi_ctx,
+						in_repo = child_in_repo,
+					},
+				)
 			}
 		} else if is_nondir {
 			if should_emit && matches_pattern(pool, entry.name) {
@@ -285,7 +293,8 @@ check_chain :: proc(ctx: ^GIContext, entry_rel: string, is_dir: bool) -> bool {
 relative_to :: proc(entry_rel, base_rel: string) -> string {
 	if len(base_rel) == 0 do return entry_rel
 	prefix_len := len(base_rel)
-	if len(entry_rel) > prefix_len && entry_rel[prefix_len] == '/' &&
+	if len(entry_rel) > prefix_len &&
+	   entry_rel[prefix_len] == '/' &&
 	   strings.has_prefix(entry_rel, base_rel) {
 		return entry_rel[prefix_len + 1:]
 	}
@@ -422,3 +431,4 @@ join_path_dir :: proc(parent, child: string) -> string {
 	buf[pos] = '/'
 	return string(buf)
 }
+