perf(findr): Replaced regex engine with glob.

Creating direct equivilant of fd for performance testing, before reducing
scope to needed features.

.gitignore

@@ -7,9 +7,12 @@ list.json
 man 
 
 # build artifacts
+*.spall
 builds
 envr
 envr-go
 findr/findr
+findr/findr-prof
+findr/bench-*.md
 result
 version.odin

findr/PERFORMANCE_IDEAS.md

@@ -0,0 +1,34 @@
+# Performance Ideas
+
+Current state after regex→glob migration. findr beats fd in 3/4 cases.
+
+## Benchmark results (2026-06-17)
+
+| Case | fd | findr | Ratio |
+|------|------|-------|-------|
+| 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** |
+
+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.
+
+## Completed
+
+1. **Per-thread result buffers** — each thread accumulates locally, merges once at exit. Eliminates per-result mutex contention.
+2. **Lean path join** — `join_path`/`join_path_dir` use stack buffer + `copy` + single alloc instead of `strings.Builder` + `fmt.sbprintf` + `clone`.
+3. **Regex→glob migration** — replaced regex NFA with backtracking glob matcher. Eliminated 27% of CPU spent on `add_thread`/`is_ignored`. Biggest win.
+
+## Remaining ideas
+
+1. **Larger getdents buffer** (8KB → 64KB+)
+   Fewer syscalls per directory with many entries. Low effort.
+
+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/PLAN.md

@@ -1,99 +1,115 @@
-# findr — Gitignored File Finder
+# findr — Native Odin File Finder (fd Replacement)
 
 ## Overview
 
-findr is a native Odin tool that finds **gitignored files** within git repositories. It replaces envr's current approach of running `fd` twice (all files vs. unignored files) and diffing the results.
+findr is a native Odin file finder that replaces `fd` in envr. It supports three ignore modes for A/B benchmarking against specific fd commands, plus a unique "emit ONLY gitignored files" mode that gives envr a single-pass advantage over fd's double-run-and-diff approach.
-
-**Simplified scope:** findr does one thing — walks directories, finds git repos, reads each repo's `.gitignore`, and prints every gitignored file. No flags, no filtering, no pattern matching. envr handles result filtering itself.
-
-## Current fd Usage in envr (being replaced)
-
-1. **`scan.odin:13-43`** (`scan_path`) — runs `fd` twice per search path:
-   - Run 1: `fd -a <matcher> [-E <exclude>]... -HI <path>` → all files including gitignored
-   - Run 2: `fd -a <matcher> [-E <exclude>]... -H <path>` → hidden but NOT gitignored
-   - Diff = gitignored files only
-2. Both go through `run_fd` (`scan.odin:68-118`), which spawns a subprocess and captures output via temp files.
-
-After findr integration, `scan_path` calls `findr.walk(path)` directly — no subprocess, no double-run, no diff.
 
 ## Directory Structure
 
 ```
 findr/
-  findr.odin           # main + CLI (positional dir args only)
+  findr.odin           # main + CLI (hand-rolled arg parsing)
-  walker.odin          # recursive directory walker using core:sys/linux getdents
+  walker.odin          # parallel directory walker (getdents + thread pool)
   gitignore.odin       # .gitignore parsing + glob→regex transpilation + matching
   test_env.odin        # test harness: temp dir, mock filesystem, assert helpers
-  findr_test.odin      # integration tests (10 tests)
+  findr_test.odin      # integration tests
   gitignore_test.odin  # transpilation + matching unit tests (22 tests)
 ```
 
-## Decisions
-
-- **Scope**: findr prints ALL gitignored files. No regex filtering, no exclude patterns, no type filters. envr post-processes the output.
-- **Gitignore matching**: Transpile gitignore glob patterns to regex, then use `core:text/regex`. No dedicated glob matcher.
-- **Stat avoidance**: Use `core:sys/linux` getdents directly — read `dirent.type` from the kernel, never call stat.
-- **Architecture**: Separate directory with its own `main`. Core logic (`walk` proc + `gitignore` package) designed to be importable into envr later.
-
 ## CLI Interface
 
 ```
-findr [dir1] [dir2] ...
+findr [-I] [--ignored] [--no-hidden] [-E <glob>]... [pattern] [path]...
 ```
 
-No flags. Defaults to `.` if no dirs given. Prints absolute or relative paths (as given) to stdout, one per line.
+Defaults: `include_hidden=true, ignore_mode=.Respected` (matches fd's `-H` behavior).
+
+| fd command | findr equivalent |
+|---|---|
+| `fd -a \.env -E ... -HI ~/` | `findr -I -E ... \.env ~/` |
+| `fd -a \.env -E ... -H ~/`  | `findr -E ... \.env ~/` |
+| `fd . -H ~/`                | `findr ~/` |
+| `fd . -HI ~/`               | `findr -I ~/` |
+| `fd . ~/` (no flags)        | `findr --no-hidden ~/` |
+| *(findr original)*          | `findr --ignored ~/` |
 
 ## Build
 
 ```bash
 odin build findr -o:speed -out:findr/findr
+odin test findr
 ```
 
-## How It Works
+## Architecture
 
-```
+### Two Orthogonal Axes (matching fd's semantics)
-walk(dir):
+
-  entries = getdents(dir)         # via core:sys/linux, zero stat calls
+1. **Hidden files** (`.` prefix): `include_hidden=true` includes them, `false` excludes them
-  if entries contains ".git/":
+2. **Gitignore**: three modes (see `IgnoreMode` below)
-    gi = parse(.gitignore)        # if present
+
-    for entry in entries:
+### Types
-      if entry is gitignored file:
+
-        emit entry path
+```odin
-      if entry is dir (not ignored):
+IgnoreMode :: enum {
-        walk(entry)               # recurse to find nested repos
+    Respected,  // skip gitignored, prune ignored dirs (fd -H default)
-  else:
+    All,        // ignore .gitignore entirely, descend everywhere (fd -HI)
-    for entry in entries:
+    Ignored,    // emit ONLY gitignored files, prune ignored dirs (findr original)
-      if entry is dir:
+}
-        walk(entry)               # descend looking for repos
+
+WalkOptions :: struct {
+    pattern:        string,       // regex on basename; "" = match all
+    excludes:       []string,     // glob patterns to skip entirely (fd -E)
+    include_hidden: bool,         // true = include dotfiles (fd -H)
+    ignore_mode:    IgnoreMode,
+}
 ```
 
-Key behaviors:
+### process_dir Filtering Order Per Entry
-- **Nested repos**: When a repo is found, subdirectories are still traversed to find nested repos. Gitignored directories are pruned (not descended into).
-- **Flat gitignore**: Only the root `.gitignore` is read. `.gitignore` files in subdirectories of a repo are ignored.
-- **Non-repo dirs**: Traversed recursively to find repos. No gitignore rules apply.
 
-## Performance Architecture
+Each directory traversal carries a `WorkItem` with the absolute path, a relative path from repo root, and a `^GIContext` linked list of gitignore contexts (one per ancestor directory with a `.gitignore`).
 
-### Implemented
+1. Skip `.git` directory
+2. **Load nested `.gitignore`**: If this directory has a `.gitignore`, push a new `GIContext` onto the chain (tracked in `pool.all_contexts` for cleanup)
+3. **Per entry**:
+   - Skip non-regular files (symlinks, sockets, etc. — parity with `fd -t f`)
+   - **Excludes**: if entry matches any exclude glob → skip entirely
+   - **Hidden**: if `!include_hidden && name[0] == '.'` → skip entirely
+   - **Gitignore status**: check `GIContext` chain deepest-to-root via `check_chain`, passing the **relative path** (not basename). First match wins (correct gitignore precedence). Nested negation overrides parent rules.
+   - **Mode-based decision**:
 
-- **Stat avoidance via `dirent.type`** — Uses `core:sys/linux` getdents directly, bypassing `core:os` which calls `openat` + `fstat` per entry. File type comes free from the directory entry.
+| Mode | gitignored file | gitignored dir | normal file | normal dir |
-- **Prune ignored directories** — When a directory matches a gitignore pattern, it is not descended into. Skips potentially thousands of readdir calls.
+|---|---|---|---|---|
-- **Parallel traversal** — 8-worker thread pool with shared LIFO queue and futex-based semaphore signaling. 5.4x speedup over serial on home directory.
+| `.All` | emit if pattern matches | descend | emit if pattern matches | descend |
+| `.Respected` | skip | prune | emit if pattern matches | descend |
+| `.Ignored` | emit if pattern matches | prune | skip | descend |
 
-### Future (if needed)
+**Nested repos**: When a directory contains `.git/`, the gitignore context chain is reset (new repo root). The relative path resets to `""`. Nested repos are always traversed to find deeper repos.
 
-- BufWriter on stdout for large result sets
+### Performance Architecture
-- Arena allocators for path strings
+
+- **Stat avoidance via `dirent.type`** — Uses `core:sys/linux` getdents directly, bypassing `core:os` which calls `openat` + `fstat` per entry.
+- **Prune ignored directories** — When a directory matches a gitignore/exclude pattern, it is not descended into.
+- **Parallel traversal** — Worker thread pool with shared LIFO queue and futex-based semaphore signaling. 5.4x speedup over serial on home directory.
+
+## Decisions
+
+- **Gitignore matching**: Transpile gitignore glob patterns to regex, then use `core:text/regex`. No dedicated glob matcher.
+- **Pattern matching**: Pattern is a regex (same as fd), matched against basename via `regex.match` (unanchored search).
+- **Excludes**: Glob patterns compiled via the same gitignore transpiler (`parse()`). Reuses tested transpilation logic.
+- **Nested gitignore**: Every `.gitignore` file within a repo is read, not just the root. Each directory's rules are scoped relative to that directory's path. Negation in a child overrides parent rules (correct gitignore precedence).
+- **Stat avoidance**: Use `core:sys/linux` getdents directly — read `dirent.type` from the kernel, never call stat. `DT_UNKNOWN` treated as regular file (correct for ext4/tmpfs; may miss dirs on XFS/BTRFS/FUSE — Phase 7 concern).
 
 ## Testing Strategy
 
 - **In-process integration tests** — Tests call `walk()` directly (not via subprocess), build mock filesystems in temp dirs, and compare sorted output.
 - **Unit tests** — Pure-function tests for glob→regex transpilation and gitignore matching.
 - **Output sorting for determinism** — Always sort output lines before comparison.
-- **Memory tracking** — Odin's test runner reports leaks automatically. All 32 tests pass with zero leaks.
+- **Memory tracking** — Odin's test runner reports leaks automatically.
 
 ### Test Coverage (findr_test.odin)
 
+**`.Ignored` mode (original findr behavior):**
+
 | Test | What it covers |
 |---|---|
 | `test_basic_gitignored` | Repo with `.gitignore`, gitignored files emitted, normal files skipped |
@@ -102,14 +118,35 @@ Key behaviors:
 | `test_dir_only_pattern` | `node_modules/` pattern doesn't emit file results |
 | `test_multiple_repos` | Multiple repos in one tree, each with its own `.gitignore` |
 | `test_nested_repos` | Repo inside a repo, both scanned independently |
-| `test_gitignore_in_subdir_ignored` | Subdirectory `.gitignore` files are not read |
 | `test_no_gitignore_file` | Repo with `.git/` but no `.gitignore` produces nothing |
 | `test_empty_gitignore` | Comments and blank lines only → no results |
 | `test_multiple_search_dirs` | Multiple top-level search dirs in one call |
+| `test_nested_gitignore_read` | Nested `.gitignore` rules applied (subdir patterns work) |
+| `test_nested_gitignore_negation` | Nested negation overrides parent pattern |
+| `test_multisegment_pattern` | `build/output.txt` matches relative path, not just basename |
 
-### Gitignore Unit Tests (gitignore_test.odin)
+**`.All` mode (fd -HI parity):**
 
-22 tests covering: simple/anchored patterns, `*`, `?`, `[abc]`, `[!abc]`, dot escaping, globstar variants, backslash escapes, empty patterns, basic matching, negation, dir-only, comments, blank lines, last-match-wins, env patterns.
+| Test | What it covers |
+|---|---|
+| `test_all_mode_emits_all_files` | All files emitted regardless of gitignore |
+| `test_all_mode_descends_everywhere` | Gitignored dirs still descended |
+
+**`.Respected` mode (fd -H parity):**
+
+| Test | What it covers |
+|---|---|
+| `test_respected_mode_skips_gitignored` | Gitignored files skipped |
+| `test_respected_mode_prunes_ignored_dirs` | Gitignored dirs pruned |
+| `test_nested_gitignore_respected_mode` | Nested negation respected in `.Respected` mode |
+
+**Filters:**
+
+| Test | What it covers |
+|---|---|
+| `test_excludes_prune_dirs` | Excluded dirs not descended |
+| `test_pattern_filters_results` | Only pattern-matching files emitted |
+| `test_no_hidden_skips_dotfiles` | Hidden files skipped when include_hidden=false |
 
 ## Glob→Regex Transpilation Rules
 
@@ -130,52 +167,176 @@ Key behaviors:
 
 ### Phase 1: Gitignore Transpiler + Tests ✅
 
-**Goal:** Isolated, fully-tested glob→regex transpiler.
+22 tests, all passing, zero leaks.
-
-**Result:** 22 tests, all passing, zero leaks.
-
----
 
 ### Phase 2: findr Walker + Tests ✅
 
-**Goal:** Working tool that finds gitignored files in git repos.
+Parallel DFS using getdents with worker thread pool. 32 total tests pass, zero leaks.
-
-**Built:**
-- `walker.odin` — Parallel DFS using `core:sys/linux` getdents with 8-worker thread pool. Finds repos, reads `.gitignore`, emits gitignored files, recurses into subdirs for nested repos.
-- `findr.odin` — Minimal CLI: `findr [dirs...]`, no flags.
-- `test_env.odin` — Test harness with temp dirs and mock filesystems.
-- `findr_test.odin` — 10 integration tests.
-
-**Result:** All 32 tests pass (22 gitignore + 10 walker), zero leaks.
-
----
 
 ### Phase 3: Parallel Traversal ✅
 
-**Goal:** Parallelize directory descent for large trees.
+8-worker thread pool, shared LIFO queue, futex-based semaphore. 852ms vs 4.57s serial (5.4x speedup). Serial code removed — parallel is the only implementation.
-
-**Result:** Worker pool with shared LIFO queue, 8 threads, futex-based semaphore signaling. 852ms vs 4.57s serial (5.4x speedup) on `~`. Serial code has been removed — parallel is the only implementation.
-
----
 
 ### Phase 4: Benchmark ✅
 
-**Goal:** Quantify performance vs fd on large directory trees.
+findr found 227 gitignored files on `~` in 852ms. fd's double-run walked ~1.1M entries.
 
-**Result:** findr found 227 gitignored files on `~` in 852ms. fd's double-run (all vs unignored) walked ~1.1M entries. findr's pruning of ignored directories (node_modules, dist, etc.) gives a massive advantage.
+### Phase 5: fd-Parity API ✅
 
----
+**Goal:** Make findr replicate specific fd commands for A/B benchmarking, plus keep the unique gitignored-only mode.
 
-### Phase 5: Integrate into envr (future)
+**Built:**
+- `IgnoreMode` enum (`.Respected`, `.All`, `.Ignored`) and `WalkOptions` struct
+- New `walk` signature: `walk(root, results, opts: WalkOptions, thread_count)`
+- Rewritten `process_dir` with centralized mode-based filtering
+- Pattern matching via `core:text/regex` on basenames
+- Exclude patterns compiled via existing `gitignore.parse()`
+- CLI arg parsing: `-I`, `--ignored`, `--no-hidden`, `-E <glob>`
+- 7 new integration tests (17 total) covering all three modes, excludes, pattern, and hidden filtering
 
-**Goal:** Replace `run_fd` in `scan.odin`. `scan_path` calls `findr.walk()` directly instead of two subprocess runs + diff.
+**Result:** All tests pass (22 gitignore + 20 walker = 42), zero leaks.
+
+### Phase 6: Parity (partially done)
+
+**Goal:** Achieve file-count parity with fd. An invalid benchmark (different result sets) is useless.
+
+#### Steps 1-2: Nested gitignore + relative path matching ✅
+
+**What was done:**
+
+1. **`Match` enum + `check_match`** in `gitignore.odin` — Tri-state return (`None`/`Ignored`/`Unignored`) so nested negation overrides work correctly. `is_ignored` wraps it as before.
+
+2. **`GIContext` linked list** in `walker.odin` — Each context holds a `^Gitignore`, `base_rel` (relative path from repo root to this dir), and `parent: ^GIContext`. `process_dir` loads `.gitignore` in every directory within a repo (not just roots). `check_chain` walks deepest-to-root, first match wins (correct gitignore precedence).
+
+3. **`WorkItem` struct** replaced plain `string` in the work queue:
+   ```odin
+   WorkItem :: struct {
+       path:    string,       // absolute directory path
+       rel:     string,       // relative path from repo root ("" = root)
+       gi_ctx:  ^GIContext,   // gitignore chain (nil = outside any repo)
+   }
+   ```
+
+4. **Relative path matching** — `check_chain` strips each context's `base_rel` prefix to get the locally-scoped relative path. Multi-segment patterns like `build/output.txt` now match correctly.
+
+5. **Symlink filtering** — Only `DT_REG` and `DT_UNKNOWN` entries are emitted (matching `fd -t f`). Symlinks (`DT_LNK`) are skipped.
+
+6. **`DT_UNKNOWN` handling** — Treated as regular files (no stat fallback). Correct for ext4/tmpfs; may miss directories on XFS/BTRFS/FUSE.
+
+**Memory management:** All `GIContext` objects tracked in `pool.all_contexts` (mutex-protected append). Gitignore objects and context structs freed in bulk when `walk` completes.
+
+**Parity achieved** (`~`, 5M+ files):
+
+| Mode | findr | fd equivalent | diff |
+|---|---|---|---|
+| `.All` (-I) | 5,426,451 | `fd -HI -t f --exclude .git` | **0 (exact)** |
+| `.Respected` | 4,442,505 | `fd -H -t f --exclude .git` | +1,417 (0.03%) |
+| `--no-hidden` | 393,605 | `fd -t f --exclude .git` | +17 (0.004%) |
+
+On the envr repo itself, all three modes are **exact match (0 diffs)**. The tiny residual diffs on `~` are likely from global gitignore (`~/.config/git/ignore`) and `.git/info/exclude` which fd reads but findr doesn't.
+
+#### Step 3: DT_UNKNOWN stat fallback (TODO)
+
+On XFS/BTRFS/FUSE filesystems, `dirent.type` returns `DT_UNKNOWN`. Currently findr treats these as regular files, which means directories may be missed (not descended into). Add a stat fallback in `read_dir_entries` when `d.type == .UNKNOWN` to determine the real type before proceeding. This is not needed for ext4/tmpfs (what tests and most Linux systems use).
+
+### Phase 7: Performance Optimization (next)
+
+**Goal:** Make findr competitive with or faster than fd across all modes. Current benchmark (`~`, hyperfine 5 runs):
+
+| Command | Mean | vs fd equivalent |
+|---|---|---|
+| `findr --ignored` | 984ms | *(no fd equivalent)* |
+| `findr --no-hidden` | 542ms | 3.2x slower than `fd -t f` (170ms) |
+| `findr` (respected) | 4.134s | 2.4x slower than `fd -H -t f` (1.745s) |
+| `findr -I` (all) | 3.821s | 1.9x slower than `fd -HI -t f` (1.972s) |
+
+**Bottleneck analysis:**
+
+1. **Mutex contention on result collection** — Every file append goes through `sync.mutex_lock(&pool.results_mutex)` → `append` → `sync.mutex_unlock`. With 5M+ files across 16 threads, workers serialize on the mutex.
+
+2. **`--ignored` regression** — Was 402ms before nested gitignore support, now 984ms. The overhead comes from loading `.gitignore` in every directory and checking the context chain per entry. Since `--ignored` mode prunes gitignored dirs, many of these `.gitignore` loads are wasted (the dir won't be descended into anyway). Optimization: skip loading `.gitignore` for directories that will be pruned.
+
+3. **Per-string heap allocation** — Every path string is individually `strings.clone`'d and `delete`'d. Millions of alloc/free calls.
+
+**Optimization plan:**
+
+1. **Per-thread result buffers** — Each worker accumulates results in a thread-local `[dynamic]string`. Merge into shared array once at the end (single-threaded concat).
+
+2. **Lazy gitignore loading for `.Ignored` mode** — Only load `.gitignore` when we need to decide whether to emit or descend. In `.Ignored` mode, we can check the parent context first and skip loading if the directory itself is already ignored.
+
+3. **Arena allocator for paths** — Replace per-string `strings.clone` with a bump allocator. Free everything in one `arena_destroy` at the end.
+
+4. **Larger getdents buffer** — Increase from 8KB to 64KB to reduce syscall count.
+
+5. **BufWriter on stdout** — Batch `write` syscalls instead of per-line `fmt.println`.
+
+**Success criteria:**
+- `.All` mode faster than `fd -HI -t f --exclude .git`
+- `.Respected` mode faster than `fd -H -t f --exclude .git`
+- `--ignored` mode faster than `fd -HI -t f --exclude .git` (restore pre-regression advantage)
+- Re-benchmark after each step using `findr/bench.sh`
+
+### Phase 8: Integrate into envr
+
+**Goal:** Replace ALL `fd` subprocess usage in envr with in-process findr calls. Remove `Feature.Fd` entirely.
+
+#### Part A: Rewrite `scan_path` (`scan.odin`)
+
+Replace the double-run-and-diff approach with a single `findr.walk` call using `.Ignored` mode:
+
+```odin
+// Before: fd -HI + fd -H, then diff
+// After:
+findr.walk(search_path, &paths, WalkOptions{
+    pattern = cfg.ScanConfig.Matcher,
+    excludes = cfg.ScanConfig.Exclude[:],
+    include_hidden = true,
+    ignore_mode = .Ignored,
+}, thread_count)
+```
+
+**Delete:** `build_fd_args`, `run_fd`, `next_fd_tmp_path`, `fd_counter`, `fd_seq`, `cant_scan`.
+
+#### Part B: Add `find_repos` and rewrite `find_git_roots` (`config.odin`)
+
+Add a `find_repos` proc to findr that walks a tree and collects directories containing `.git/`:
+
+```odin
+find_repos :: proc(root: string, results: ^[dynamic]string, thread_count: int)
+```
+
+- Reuses worker pool architecture
+- `process_dir` emits `dir_path` when `has_git == true`
+- Always descends into subdirs (except `.git`) to find nested repos
+- No gitignore/exclude/pattern processing
+
+Replace `find_git_roots`'s `run_fd` call with `findr.find_repos`.
+
+#### Part C: Remove `Feature.Fd` everywhere
+
+| File | Change |
+|---|---|
+| `features.odin` | Remove `Fd` from enum, remove fd binary check |
+| `cmd_scan.odin` | Remove feats/cant_scan guard + "install fd" error |
+| `cmd_check.odin` | Same removal |
+| `cmd_deps.odin` | Remove fd table row |
+| `db.odin` | Change check to `.Git not_in feats` only; update error message |
+| `scan_test.odin` | Remove `cant_scan` tests and assertions |
+
+#### Part D: Verification
+
+```bash
+odin build findr -o:speed -out:findr/findr
+odin test findr
+odin build . -o:speed -out:envr
+odin test .
+```
 
 ## Risks
 
 | Risk | Mitigation |
 |---|---|
-| Single-threaded may be slow on huge trees | Resolved — parallel traversal implemented (Phase 3) |
 | Gitignore edge cases (`**/foo`, `foo/**/bar`) | Comprehensive gitignore_test.odin with spec examples |
-| dirent.type may be UNKNOWN on some filesystems | Fall back to stat only when type is UNKNOWN |
+| `DT_UNKNOWN` on XFS/BTRFS/FUSE | Phase 6 Step 3: stat fallback for unknown types |
-| Missing nested `.env` files in monorepos | Accepted limitation — flat gitignore model |
+| Global gitignore (`~/.config/git/ignore`) and `.git/info/exclude` not read | Causes ~0.03% delta vs fd. Acceptable for envr's use case (finds `.env` files in repos). |
-| Memory allocation churn from path strings | Use thread-local arena allocators in Phase 3 |
+| Thread safety of `regex.match` on shared `Regular_Expression` | Odin regex is read-only after compilation; `match` returns per-call `Captures` |

findr/bench.sh

@@ -0,0 +1,71 @@
+#!/usr/bin/env bash
+set -euo pipefail
+
+BENCH_DIR="$(cd "$(dirname "$0")" && pwd)"
+TARGET="${1:-$HOME}"
+RESULTS_FILE="$BENCH_DIR/bench-results.md"
+FINDR="$BENCH_DIR/findr"
+
+echo "=== findr benchmark suite ==="
+echo "Target: $TARGET"
+echo
+
+# --- pre-flight checks ---
+if ! command -v fd &>/dev/null; then
+    echo "ERROR: fd is not on PATH" >&2
+    exit 1
+fi
+if ! command -v hyperfine &>/dev/null; then
+    echo "ERROR: hyperfine is not on PATH" >&2
+    exit 1
+fi
+
+# --- build findr if missing or stale ---
+NEEDS_BUILD=false
+if [[ ! -f "$BENCH_DIR/findr" ]]; then
+    NEEDS_BUILD=true
+else
+    # rebuild if any .odin source is newer than the binary
+    if find "$BENCH_DIR" -name '*.odin' -newer "$BENCH_DIR/findr" | grep -q .; then
+        NEEDS_BUILD=true
+    fi
+fi
+if $NEEDS_BUILD; then
+    echo "Building findr..."
+    odin build "$BENCH_DIR" -o:speed -out:"$BENCH_DIR/findr"
+fi
+echo
+
+# --- file counts ---
+echo "=== File counts ==="
+printf "  fd -a -E .jj .                : %8d\n" "$(fd -a -E .jj . "$TARGET" 2>/dev/null | wc -l)"
+printf "  findr -E .jj                  : %8d\n" "$("$FINDR" -E .jj "$TARGET" 2>/dev/null | wc -l)"
+echo
+printf "  fd -a -E .git -E .jj -H .     : %8d\n" "$(fd -a -E .git -E .jj -H . "$TARGET" 2>/dev/null | wc -l)"
+printf "  findr -E .git -E .jj -H       : %8d\n" "$("$FINDR" -E .git -E .jj -H "$TARGET" 2>/dev/null | wc -l)"
+echo
+printf "  fd -a -E .git -E .jj -HI .    : %8d\n" "$(fd -a -E .git -E .jj -HI . "$TARGET" 2>/dev/null | wc -l)"
+printf "  findr -E .git -E .jj -HI      : %8d\n" "$("$FINDR" -E .git -E .jj -HI "$TARGET" 2>/dev/null | wc -l)"
+echo
+printf "  fd -a -E .git -E .jj .        : %8d\n" "$(fd -a -E .git -E .jj . "$TARGET" 2>/dev/null | wc -l)"
+printf "  findr -E .git -E .jj          : %8d\n" "$("$FINDR" -E .git -E .jj "$TARGET" 2>/dev/null | wc -l)"
+echo
+
+# --- benchmarks ---
+echo "=== Benchmarks (hyperfine, 5 runs, 2 warmups) ==="
+echo
+hyperfine \
+    --warmup 2 \
+    --runs 5 \
+    --export-markdown "$RESULTS_FILE" \
+    "fd -a -E .jj . \"$TARGET\" > /dev/null" \
+    "$FINDR -E .jj \"$TARGET\" > /dev/null" \
+    "fd -a -E .git -E .jj -H . \"$TARGET\" > /dev/null" \
+    "$FINDR -E .git -E .jj -H \"$TARGET\" > /dev/null" \
+    "fd -a -E .git -E .jj -HI . \"$TARGET\" > /dev/null" \
+    "$FINDR -E .git -E .jj -HI \"$TARGET\" > /dev/null" \
+    "fd -a -E .git -E .jj . \"$TARGET\" > /dev/null" \
+    "$FINDR -E .git -E .jj \"$TARGET\" > /dev/null"
+echo
+
+echo "=== Results written to $RESULTS_FILE ==="

findr/f.nu

@@ -0,0 +1,27 @@
+#!/usr/bin/env nu
+
+def main [] {
+  let all = (fd -HI -a .env . ~/ | lines | sort)
+  let unignored = (fd -H -a .env ~/ | lines | sort)  
+
+  $all | filter { |it| not ($it in $unignored) } | str join "\n"
+  # sorted_list_intersect $all $unignored | str join "\n"
+}
+
+def sorted_list_intersect [xs1: list, xs2: list] {
+  let len1 = ($xs1 | length)
+  let len2 = ($xs2 | length)
+  mut i = 0
+  mut j = 0
+  while ($i < $len1 and $j < $len2) {
+    if ($xs1 | get $i) < ($xs2 | get $j) {
+      $i = $i + 1
+    } else if ($xs2 | get $j) < ($xs1 | get $i) {
+      $j = $j + 1
+    } else {
+      echo ($xs2 | get $j)
+      $i = $i + 1
+      $j = $j + 1
+    }
+  }
+}

findr/findr.odin

@@ -1,33 +1,91 @@
 package findr
 
-import "core:fmt"
+import "core:bufio"
 import "core:os"
+import "core:strings"
 
 main :: proc() {
+	prof_init()
+	defer prof_destroy()
+
 	args := os.args
 
-	search_dirs := make([dynamic]string)
+	opts: WalkOptions
-	defer delete(search_dirs)
+	opts.include_hidden = false
+	opts.ignore_mode = .Respected
 
-	for i in 1..<len(args) {
+	excludes := make([dynamic]string)
-		append(&search_dirs, args[i])
+	defer delete(excludes)
+
+	pattern := ""
+	paths := make([dynamic]string)
+	defer delete(paths)
+
+	i := 1
+	for i < len(args) {
+		arg := args[i]
+		switch {
+		case arg == "--ignored":
+			opts.ignore_mode = .Ignored
+		case arg == "-E":
+			i += 1
+			if i < len(args) {
+				append(&excludes, args[i])
+			}
+		case strings.has_prefix(arg, "-E"):
+			append(&excludes, arg[2:])
+		case len(arg) > 1 && arg[0] == '-':
+			for c, j in arg[1:] {
+				switch c {
+				case 'H':
+					opts.include_hidden = true
+				case 'I':
+					opts.ignore_mode = .All
+				case 'a':
+				// no-op: accepted for fd compatibility
+				}
+			}
+		case:
+			if pattern == "" {
+				pattern = arg
+			} else {
+				append(&paths, arg)
+			}
+		}
+		i += 1
 	}
 
-	if len(search_dirs) == 0 {
+	if len(paths) == 0 && pattern != "" && os.exists(pattern) {
-		append(&search_dirs, ".")
+		append(&paths, pattern)
+		pattern = ""
+	}
+
+	opts.pattern = pattern
+	if len(excludes) > 0 {
+		opts.excludes = excludes[:]
+	}
+
+	if len(paths) == 0 {
+		append(&paths, ".")
 	}
 
 	results := make([dynamic]string)
 	defer {
-		for r in results { delete(r) }
+		for r in results {delete(r)}
 		delete(results)
 	}
 
-	for dir in search_dirs {
+	thread_count := os.get_processor_core_count()
-		walk(dir, &results)
+	walk(paths[:], &results, opts, thread_count)
-	}
+
+	w: bufio.Writer
+	bufio.writer_init(&w, os.to_stream(os.stdout), 1 << 13)
+	defer bufio.writer_destroy(&w)
 
 	for r in results {
-		fmt.println(r)
+		bufio.writer_write_string(&w, r)
+		bufio.writer_write_byte(&w, '\n')
 	}
+	bufio.writer_flush(&w)
 }
+

findr/findr_test.odin

@@ -1,7 +1,15 @@
 package findr
 
+import "core:os"
+import "core:sort"
+import "core:strings"
+import "core:sys/linux"
 import "core:testing"
 
+// ============================================================================
+// .Ignored mode tests (original findr behavior — emit ONLY gitignored files)
+// ============================================================================
+
 @(test)
 test_basic_gitignored :: proc(t: ^testing.T) {
 	env := create_test_env()
@@ -13,7 +21,9 @@ test_basic_gitignored :: proc(t: ^testing.T) {
 	create_file(env, "repo/secrets.env")
 	create_file(env, "repo/normal.txt")
 
-	assert_output(t, env, nil, {"repo/.env", "repo/secrets.env"})
+	assert_output(t, env, nil, {include_hidden = true, ignore_mode = .Ignored}, {
+		"repo/.env", "repo/secrets.env",
+	})
 }
 
 @(test)
@@ -25,7 +35,7 @@ test_non_repo_not_scanned :: proc(t: ^testing.T) {
 	create_file(env, "norepo/.gitignore", "*.env\n")
 	create_file(env, "norepo/.env")
 
-	assert_output_empty(t, env, nil)
+	assert_output_empty(t, env, nil, {include_hidden = true, ignore_mode = .Ignored})
 }
 
 @(test)
@@ -39,7 +49,9 @@ test_negation_pattern :: proc(t: ^testing.T) {
 	create_file(env, "repo/secrets.env")
 	create_file(env, "repo/prod.env")
 
-	assert_output(t, env, nil, {"repo/.env", "repo/secrets.env"})
+	assert_output(t, env, nil, {include_hidden = true, ignore_mode = .Ignored}, {
+		"repo/.env", "repo/secrets.env",
+	})
 }
 
 @(test)
@@ -54,8 +66,9 @@ test_dir_only_pattern :: proc(t: ^testing.T) {
 	create_dir(env, "repo/ignored_dir")
 	create_file(env, "repo/.gitignore", "ignored_dir/\n")
 
-	// dir-only patterns don't produce file results
+	assert_output(t, env, nil, {include_hidden = true, ignore_mode = .Ignored}, {
-	assert_output(t, env, nil, {})
+		"repo/ignored_dir/",
+	})
 }
 
 @(test)
@@ -71,7 +84,9 @@ test_multiple_repos :: proc(t: ^testing.T) {
 	create_file(env, "repo2/.gitignore", "*.key\n")
 	create_file(env, "repo2/secret.key")
 
-	assert_output(t, env, nil, {"repo1/a.env", "repo2/secret.key"})
+	assert_output(t, env, nil, {include_hidden = true, ignore_mode = .Ignored}, {
+		"repo1/a.env", "repo2/secret.key",
+	})
 }
 
 @(test)
@@ -87,11 +102,13 @@ test_nested_repos :: proc(t: ^testing.T) {
 	create_file(env, "parent/child/.gitignore", "*.key\n")
 	create_file(env, "parent/child/api.key")
 
-	assert_output(t, env, nil, {"parent/top.env", "parent/child/api.key"})
+	assert_output(t, env, nil, {include_hidden = true, ignore_mode = .Ignored}, {
+		"parent/top.env", "parent/child/api.key",
+	})
 }
 
 @(test)
-test_gitignore_in_subdir_ignored :: proc(t: ^testing.T) {
+test_nested_gitignore_read :: proc(t: ^testing.T) {
 	env := create_test_env()
 	defer destroy_test_env(&env)
 
@@ -102,10 +119,73 @@ test_gitignore_in_subdir_ignored :: proc(t: ^testing.T) {
 	create_file(env, "repo/sub/secret.txt")
 	create_file(env, "repo/sub/.env")
 
-	// .gitignore in subdir is not read (flat model).
+	// Both root and nested .gitignore are read.
-	// secret.txt should NOT appear (subdir .gitignore ignored).
+	// secret.txt: ignored by sub/.gitignore (*.txt)
-	// .env should NOT appear (it's nested, not top-level of repo).
+	// .env: ignored by root .gitignore (*.env)
-	assert_output(t, env, nil, {})
+	assert_output(t, env, nil, {include_hidden = true, ignore_mode = .Ignored}, {
+		"repo/sub/secret.txt", "repo/sub/.env",
+	})
+}
+
+@(test)
+test_nested_gitignore_negation :: proc(t: ^testing.T) {
+	env := create_test_env()
+	defer destroy_test_env(&env)
+
+	create_git_repo(env, "repo")
+	create_file(env, "repo/.gitignore", "*.log\n")
+	create_dir(env, "repo/sub")
+	create_file(env, "repo/sub/.gitignore", "!important.log\n")
+	create_file(env, "repo/sub/important.log")
+	create_file(env, "repo/sub/debug.log")
+
+	// Nested negation overrides root pattern.
+	// important.log: un-ignored by sub/.gitignore → NOT emitted in .Ignored mode
+	// debug.log: still ignored by root → emitted
+	assert_output(t, env, nil, {include_hidden = true, ignore_mode = .Ignored}, {
+		"repo/sub/debug.log",
+	})
+}
+
+@(test)
+test_nested_gitignore_respected_mode :: proc(t: ^testing.T) {
+	env := create_test_env()
+	defer destroy_test_env(&env)
+
+	create_git_repo(env, "repo")
+	create_file(env, "repo/.gitignore", "*.log\n")
+	create_dir(env, "repo/sub")
+	create_file(env, "repo/sub/.gitignore", "!important.log\n")
+	create_file(env, "repo/sub/important.log")
+	create_file(env, "repo/sub/debug.log")
+
+	// In .Respected mode:
+	// important.log: un-ignored by nested negation → emitted
+	// debug.log: ignored by root → skipped
+	assert_output(t, env, nil, {include_hidden = true, ignore_mode = .Respected}, {
+		"repo/", "repo/.gitignore", "repo/sub/", "repo/sub/.gitignore", "repo/sub/important.log",
+	})
+}
+
+@(test)
+test_multisegment_pattern :: proc(t: ^testing.T) {
+	env := create_test_env()
+	defer destroy_test_env(&env)
+
+	create_git_repo(env, "repo")
+	create_file(env, "repo/.gitignore", "build/output.txt\n")
+	create_dir(env, "repo/build")
+	create_file(env, "repo/build/output.txt")
+	create_file(env, "repo/build/other.txt")
+	create_file(env, "repo/output.txt")
+
+	// Multi-segment pattern matches relative path, not just basename.
+	// build/output.txt: matches → ignored
+	// build/other.txt: doesn't match → not ignored
+	// output.txt: doesn't match (needs build/ prefix) → not ignored
+	assert_output(t, env, nil, {include_hidden = true, ignore_mode = .Ignored}, {
+		"repo/build/output.txt",
+	})
 }
 
 @(test)
@@ -116,7 +196,7 @@ test_no_gitignore_file :: proc(t: ^testing.T) {
 	create_git_repo(env, "repo")
 	create_file(env, "repo/.env")
 
-	assert_output_empty(t, env, nil)
+	assert_output_empty(t, env, nil, {include_hidden = true, ignore_mode = .Ignored})
 }
 
 @(test)
@@ -128,7 +208,7 @@ test_empty_gitignore :: proc(t: ^testing.T) {
 	create_file(env, "repo/.gitignore", "\n\n# comment\n\n")
 	create_file(env, "repo/.env")
 
-	assert_output_empty(t, env, nil)
+	assert_output_empty(t, env, nil, {include_hidden = true, ignore_mode = .Ignored})
 }
 
 @(test)
@@ -139,6 +219,7 @@ test_multiple_search_dirs :: proc(t: ^testing.T) {
 	create_git_repo(env, "dir1/repo")
 	create_file(env, "dir1/repo/.gitignore", "*.env\n")
 	create_file(env, "dir1/repo/a.env")
+	create_file(env, "dir1/repo/normal.txt")
 
 	create_git_repo(env, "dir2/repo")
 	create_file(env, "dir2/repo/.gitignore", "*.env\n")
@@ -151,10 +232,243 @@ test_multiple_search_dirs :: proc(t: ^testing.T) {
 
 	results := make([dynamic]string)
 	defer {
-		for r in results { delete(r) }
+		for r in results {delete(r)}
 		delete(results)
 	}
-	walk(dir1, &results)
+
-	walk(dir2, &results)
+	opts := WalkOptions{include_hidden = true, ignore_mode = .Ignored}
+	thread_count := os.get_processor_core_count()
+	walk({dir1, dir2}, &results, opts, thread_count)
+
 	testing.expect_value(t, len(results), 2)
+
+	actual := make([dynamic]string, 0, len(results))
+	for r in results {
+		stripped := r
+		if strings.has_prefix(stripped, env.temp_dir) {
+			stripped = stripped[len(env.temp_dir):]
+			if len(stripped) > 0 && stripped[0] == '/' {
+				stripped = stripped[1:]
+			}
+		}
+		append(&actual, stripped)
+	}
+	defer delete(actual)
+
+	expected := []string{"dir1/repo/a.env", "dir2/repo/b.env"}
+
+	sort.quick_sort(actual[:])
+	sort.quick_sort(expected[:])
+
+	for i in 0 ..< len(expected) {
+		testing.expect_value(t, actual[i], expected[i])
+	}
+}
+
+// ============================================================================
+// .All mode tests (fd -HI parity — ignore gitignore entirely)
+// ============================================================================
+
+@(test)
+test_all_mode_emits_all_files :: proc(t: ^testing.T) {
+	env := create_test_env()
+	defer destroy_test_env(&env)
+
+	create_git_repo(env, "repo")
+	create_file(env, "repo/.gitignore", "*.env\n")
+	create_file(env, "repo/.env")
+	create_file(env, "repo/secrets.env")
+	create_file(env, "repo/normal.txt")
+
+	assert_output(t, env, nil, {include_hidden = true, ignore_mode = .All}, {
+		"repo/", "repo/.env", "repo/.gitignore", "repo/secrets.env", "repo/normal.txt",
+	})
+}
+
+@(test)
+test_all_mode_descends_everywhere :: proc(t: ^testing.T) {
+	env := create_test_env()
+	defer destroy_test_env(&env)
+
+	create_git_repo(env, "repo")
+	create_file(env, "repo/.gitignore", "build/\n")
+	create_dir(env, "repo/build")
+	create_file(env, "repo/build/output.txt")
+
+	assert_output(t, env, nil, {include_hidden = true, ignore_mode = .All}, {
+		"repo/", "repo/.gitignore", "repo/build/", "repo/build/output.txt",
+	})
+}
+
+// ============================================================================
+// .Respected mode tests (fd -H parity — skip gitignored, prune ignored dirs)
+// ============================================================================
+
+@(test)
+test_respected_mode_skips_gitignored :: proc(t: ^testing.T) {
+	env := create_test_env()
+	defer destroy_test_env(&env)
+
+	create_git_repo(env, "repo")
+	create_file(env, "repo/.gitignore", "*.env\n")
+	create_file(env, "repo/.env")
+	create_file(env, "repo/secrets.env")
+	create_file(env, "repo/normal.txt")
+
+	assert_output(t, env, nil, {include_hidden = true, ignore_mode = .Respected}, {
+		"repo/", "repo/.gitignore", "repo/normal.txt",
+	})
+}
+
+@(test)
+test_respected_mode_prunes_ignored_dirs :: proc(t: ^testing.T) {
+	env := create_test_env()
+	defer destroy_test_env(&env)
+
+	create_git_repo(env, "repo")
+	create_file(env, "repo/.gitignore", "build/\n")
+	create_dir(env, "repo/build")
+	create_file(env, "repo/build/output.txt")
+	create_file(env, "repo/main.txt")
+
+	assert_output(t, env, nil, {include_hidden = true, ignore_mode = .Respected}, {
+		"repo/", "repo/.gitignore", "repo/main.txt",
+	})
+}
+
+// ============================================================================
+// Filter tests (excludes, pattern, hidden)
+// ============================================================================
+
+@(test)
+test_excludes_prune_dirs :: proc(t: ^testing.T) {
+	env := create_test_env()
+	defer destroy_test_env(&env)
+
+	create_git_repo(env, "repo")
+	create_file(env, "repo/.gitignore", "*.env\n")
+	create_file(env, "repo/.env")
+	create_dir(env, "repo/vendor")
+	create_file(env, "repo/vendor/lib.env")
+
+	assert_output(t, env, nil,
+		{include_hidden = true, ignore_mode = .Ignored, excludes = {"vendor"}},
+		{"repo/.env"},
+	)
+}
+
+@(test)
+test_pattern_filters_results :: proc(t: ^testing.T) {
+	env := create_test_env()
+	defer destroy_test_env(&env)
+
+	create_git_repo(env, "repo")
+	create_file(env, "repo/.gitignore", "*.env\n*.key\n")
+	create_file(env, "repo/.env")
+	create_file(env, "repo/secrets.env")
+	create_file(env, "repo/master.key")
+
+	assert_output(t, env, nil,
+		{pattern = "\\.env$", include_hidden = true, ignore_mode = .Ignored},
+		{"repo/.env", "repo/secrets.env"},
+	)
+}
+
+@(test)
+test_no_hidden_skips_dotfiles :: proc(t: ^testing.T) {
+	env := create_test_env()
+	defer destroy_test_env(&env)
+
+	create_git_repo(env, "repo")
+	create_file(env, "repo/.gitignore", "*.env\n")
+	create_file(env, "repo/.env")
+	create_file(env, "repo/secrets.env")
+	create_file(env, "repo/.hidden.env")
+
+	assert_output(t, env, nil,
+		{include_hidden = false, ignore_mode = .Ignored},
+		{"repo/secrets.env"},
+	)
+}
+
+// ============================================================================
+// Special file type tests (SOCK, FIFO, CHR, BLK parity with fd)
+// ============================================================================
+
+@(test)
+test_fifo_emitted :: proc(t: ^testing.T) {
+	env := create_test_env()
+	defer destroy_test_env(&env)
+
+	create_git_repo(env, "repo")
+	create_file(env, "repo/.gitignore", "*.env\n")
+
+	fifo_path := join_path(env.temp_dir, "repo/test.fifo")
+	defer delete(fifo_path)
+	cpath := strings.clone_to_cstring(fifo_path)
+	defer delete(cpath)
+	linux.mknod(cpath, linux.S_IFIFO | linux.Mode{.IRUSR, .IWUSR}, 0)
+
+	assert_output(t, env, nil,
+		{include_hidden = true, ignore_mode = .All},
+		{"repo/", "repo/.gitignore", "repo/test.fifo"},
+	)
+}
+
+// ============================================================================
+// in_repo propagation tests
+// ============================================================================
+
+@(test)
+test_repo_without_root_gitignore :: proc(t: ^testing.T) {
+	env := create_test_env()
+	defer destroy_test_env(&env)
+
+	create_git_repo(env, "repo")
+	create_dir(env, "repo/sub")
+	create_file(env, "repo/sub/.gitignore", "*.tmp\n")
+	create_file(env, "repo/sub/file.tmp")
+	create_file(env, "repo/sub/file.txt")
+
+	assert_output(t, env, nil,
+		{include_hidden = true, ignore_mode = .Respected},
+		{"repo/", "repo/sub/", "repo/sub/.gitignore", "repo/sub/file.txt"},
+	)
+}
+
+// ============================================================================
+// .ignore file support tests (fd respects .ignore in addition to .gitignore)
+// ============================================================================
+
+@(test)
+test_ignore_file_respected :: proc(t: ^testing.T) {
+	env := create_test_env()
+	defer destroy_test_env(&env)
+
+	create_git_repo(env, "repo")
+	create_file(env, "repo/.ignore", "*.tmp\n")
+	create_file(env, "repo/file.tmp")
+	create_file(env, "repo/file.txt")
+
+	assert_output(t, env, nil,
+		{include_hidden = true, ignore_mode = .Respected},
+		{"repo/", "repo/.ignore", "repo/file.txt"},
+	)
+}
+
+@(test)
+test_ignore_overrides_gitignore :: proc(t: ^testing.T) {
+	env := create_test_env()
+	defer destroy_test_env(&env)
+
+	create_git_repo(env, "repo")
+	create_file(env, "repo/.gitignore", "*.log\n")
+	create_file(env, "repo/.ignore", "important.log\n")
+	create_file(env, "repo/debug.log")
+	create_file(env, "repo/important.log")
+
+	assert_output(t, env, nil,
+		{include_hidden = true, ignore_mode = .Respected},
+		{"repo/", "repo/.gitignore", "repo/.ignore"},
+	)
 }

findr/gitignore.odin

@@ -1,112 +1,38 @@
 package findr
 
-import "core:fmt"
 import "core:strings"
-import "core:text/regex"
-
-is_regex_meta :: proc(c: u8) -> bool {
-	switch c {
-	case '.', '+', '(', ')', '{', '}', '^', '$', '|':
-		return true
-	}
-	return false
-}
-
-glob_to_regex :: proc(pattern: string, anchored: bool) -> string {
-	sb: strings.Builder
-	strings.builder_init(&sb)
-	defer strings.builder_destroy(&sb)
-
-	if anchored {
-		fmt.sbprintf(&sb, "^")
-	} else {
-		fmt.sbprintf(&sb, "(^|/)")
-	}
-
-	i := 0
-	for i < len(pattern) {
-		c := pattern[i]
-
-		if c == '*' {
-			if i + 1 < len(pattern) && pattern[i + 1] == '*' {
-				prev_slash := i == 0 || pattern[i - 1] == '/'
-				at_end := i + 2 >= len(pattern)
-				next_slash := !at_end && pattern[i + 2] == '/'
-
-				if prev_slash && (next_slash || at_end) {
-					if next_slash {
-						i += 3
-						fmt.sbprintf(&sb, "(.*/)?")
-					} else {
-						i += 2
-						fmt.sbprintf(&sb, ".*")
-					}
-				} else {
-					fmt.sbprintf(&sb, "[^/]*")
-					i += 2
-				}
-			} else {
-				fmt.sbprintf(&sb, "[^/]*")
-				i += 1
-			}
-		} else if c == '?' {
-			fmt.sbprintf(&sb, "[^/]")
-			i += 1
-		} else if c == '[' {
-			append(&sb.buf, '[')
-			i += 1
-			if i < len(pattern) && pattern[i] == '!' {
-				append(&sb.buf, '^')
-				i += 1
-			}
-			if i < len(pattern) && pattern[i] == ']' {
-				append(&sb.buf, ']')
-				i += 1
-			}
-			for i < len(pattern) && pattern[i] != ']' {
-				append(&sb.buf, pattern[i])
-				i += 1
-			}
-			if i < len(pattern) {
-				append(&sb.buf, ']')
-				i += 1
-			}
-		} else if c == '\\' {
-			i += 1
-			if i < len(pattern) {
-				if is_regex_meta(pattern[i]) {
-					append(&sb.buf, '\\')
-				}
-				append(&sb.buf, pattern[i])
-				i += 1
-			}
-		} else if is_regex_meta(c) {
-			append(&sb.buf, '\\')
-			append(&sb.buf, c)
-			i += 1
-		} else {
-			append(&sb.buf, c)
-			i += 1
-		}
-	}
-
-	fmt.sbprintf(&sb, "(/.*)?$")
-
-	s := strings.to_string(sb)
-	result, _ := strings.clone(s)
-	return result
-}
-
-Rule :: struct {
-	regex: regex.Regular_Expression,
-	negated: bool,
-	dir_only: bool,
-}
 
 Gitignore :: struct {
 	rules: [dynamic]Rule,
 }
 
+Rule :: struct {
+	pattern:  GlobPattern,
+	negated:  bool,
+	dir_only: bool,
+}
+
+Match :: enum {
+	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 {
 	gi: Gitignore
 	gi.rules = make([dynamic]Rule)
@@ -146,37 +72,17 @@ 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
 }
 
-is_ignored :: proc(gi: ^Gitignore, path: string, is_dir: bool) -> bool {
-	matched := false
-	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 {
-			matched = !rule.negated
-		}
-	}
-	return matched
-}
-
 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)
 }
+

findr/gitignore_test.odin

@@ -4,86 +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_literal :: proc(t: ^testing.T) {
+	testing.expect(t, glob_match("#foo", "#foo", false))
+	testing.expect(t, !glob_match("#foo", "foo", false))
+}
+
+@(test)
+test_glob_hash_pattern :: proc(t: ^testing.T) {
+	testing.expect(t, glob_match("#*#", "#test#", false))
+	testing.expect(t, glob_match("#*#", "##", false))
+	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)
@@ -176,3 +193,27 @@ test_is_ignored_globstar :: proc(t: ^testing.T) {
 	testing.expect_value(t, is_ignored(&gi, "foo/bar/cache", false), true)
 }
 
+@(test)
+test_star_negation_subpath :: proc(t: ^testing.T) {
+	gi := parse("*\n!public/\n")
+	defer destroy(&gi)
+
+	// public dir itself is un-ignored
+	testing.expect_value(t, is_ignored(&gi, "public", true), false)
+	// children of public/ should still be ignored by *
+	testing.expect_value(t, is_ignored(&gi, "public/uuid-dir", true), true)
+	testing.expect_value(t, is_ignored(&gi, "public/uuid-dir/file.txt", false), true)
+}
+
+@(test)
+test_is_ignored_hash_pattern :: proc(t: ^testing.T) {
+	gi := parse("\\#*\\#\n")
+	defer destroy(&gi)
+
+	testing.expect_value(t, is_ignored(&gi, "#foo#", false), true)
+	testing.expect_value(t, is_ignored(&gi, "#test#", false), true)
+	testing.expect_value(t, is_ignored(&gi, "AUTHORS", false), false)
+	testing.expect_value(t, is_ignored(&gi, "build.zig", false), false)
+	testing.expect_value(t, is_ignored(&gi, "ChangeLog", false), false)
+}
+

findr/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
+}

findr/prof.odin

@@ -0,0 +1,64 @@
+package findr
+
+import "base:runtime"
+import "core:prof/spall"
+import "core:sync"
+
+SPALL_ENABLED :: #config(SPALL_ENABLED, ODIN_DEBUG)
+
+spall_ctx: spall.Context
+
+@(thread_local) spall_buffer: spall.Buffer
+@(thread_local) spall_backing: []u8
+
+@(instrumentation_enter)
+spall_enter :: proc "contextless" (
+	proc_address, call_site_return_address: rawptr,
+	loc: runtime.Source_Code_Location,
+) {
+	when SPALL_ENABLED {
+		spall._buffer_begin(&spall_ctx, &spall_buffer, "", "", loc)
+	}
+}
+
+@(instrumentation_exit)
+spall_exit :: proc "contextless" (
+	proc_address, call_site_return_address: rawptr,
+	loc: runtime.Source_Code_Location,
+) {
+	when SPALL_ENABLED {
+		spall._buffer_end(&spall_ctx, &spall_buffer)
+	}
+}
+
+prof_init :: proc() {
+	when SPALL_ENABLED {
+		spall_ctx = spall.context_create_with_scale("findr.spall", false, 1.0)
+		spall_backing = make([]u8, spall.BUFFER_DEFAULT_SIZE)
+		spall_buffer = spall.buffer_create(spall_backing, u32(sync.current_thread_id()))
+		spall._buffer_name_thread(&spall_ctx, &spall_buffer, "main")
+	}
+}
+
+prof_destroy :: proc() {
+	when SPALL_ENABLED {
+		spall.buffer_destroy(&spall_ctx, &spall_buffer)
+		delete(spall_backing)
+		spall.context_destroy(&spall_ctx)
+	}
+}
+
+prof_thread_init :: proc(name: string) {
+	when SPALL_ENABLED {
+		spall_backing = make([]u8, spall.BUFFER_DEFAULT_SIZE)
+		spall_buffer = spall.buffer_create(spall_backing, u32(sync.current_thread_id()))
+		spall._buffer_name_thread(&spall_ctx, &spall_buffer, name)
+	}
+}
+
+prof_thread_destroy :: proc() {
+	when SPALL_ENABLED {
+		spall.buffer_destroy(&spall_ctx, &spall_buffer)
+		delete(spall_backing)
+	}
+}

findr/profile.sh

@@ -0,0 +1,13 @@
+#!/usr/bin/env bash
+set -euo pipefail
+
+DIR="$(cd "$(dirname "$0")" && pwd)"
+
+echo "Building findr-prof..."
+odin build "$DIR" -debug -out:"$DIR/findr-prof"
+
+echo "Running profiler..."
+"$DIR/findr-prof" -E .git -E .jj -HI ~/git.verticalaxion.com
+
+echo
+echo "Spall trace: $DIR/findr.spall"

findr/test_env.odin

@@ -61,23 +61,24 @@ create_git_repo :: proc(env: TestEnv, path: string) {
 }
 
 assert_output :: proc(
-	t:        ^testing.T,
+	t: ^testing.T,
-	env:      TestEnv,
+	env: TestEnv,
-	args:     []string,
+	args: []string,
+	opts: WalkOptions,
 	expected: []string,
 ) {
-	results := collect_results(env, args)
+	results := collect_results(env, args, opts)
 	defer {
-		for r in results { delete(r) }
+		for r in results {delete(r)}
 		delete(results)
 	}
 
 	sorted_expected := make([dynamic]string, 0, len(expected))
-	for e in expected { append(&sorted_expected, e) }
+	for e in expected {append(&sorted_expected, e)}
 	defer delete(sorted_expected)
 
 	sorted_actual := make([dynamic]string, 0, len(results))
-	for a in results { append(&sorted_actual, a) }
+	for a in results {append(&sorted_actual, a)}
 	defer delete(sorted_actual)
 
 	sort.quick_sort(sorted_expected[:])
@@ -93,7 +94,7 @@ assert_output :: proc(
 		return
 	}
 
-	for i in 0..<len(sorted_expected) {
+	for i in 0 ..< len(sorted_expected) {
 		if sorted_expected[i] != sorted_actual[i] {
 			testing.fail(t)
 			log.error(fmt.tprintf("Mismatch at index %d", i))
@@ -104,10 +105,15 @@ assert_output :: proc(
 	}
 }
 
-assert_output_empty :: proc(t: ^testing.T, env: TestEnv, args: []string) {
+assert_output_empty :: proc(
-	results := collect_results(env, args)
+	t: ^testing.T,
+	env: TestEnv,
+	args: []string,
+	opts: WalkOptions,
+) {
+	results := collect_results(env, args, opts)
 	defer {
-		for r in results { delete(r) }
+		for r in results {delete(r)}
 		delete(results)
 	}
 	if len(results) > 0 {
@@ -119,18 +125,17 @@ assert_output_empty :: proc(t: ^testing.T, env: TestEnv, args: []string) {
 	}
 }
 
-collect_results :: proc(env: TestEnv, args: []string) -> [dynamic]string {
+collect_results :: proc(env: TestEnv, args: []string, opts: WalkOptions) -> [dynamic]string {
 	results := make([dynamic]string)
 
 	full_args := make([dynamic]string, 0, len(args) + 1, context.temp_allocator)
 	append(&full_args, env.temp_dir)
-	for a in args { append(&full_args, a) }
+	for a in args {append(&full_args, a)}
 
-	for dir in full_args {
+	thread_count := os.get_processor_core_count()
-		walk(dir, &results)
+	walk(full_args[:], &results, opts, thread_count)
-	}
 
-	for i in 0..<len(results) {
+	for i in 0 ..< len(results) {
 		r := results[i]
 		if strings.has_prefix(r, env.temp_dir) {
 			stripped := r[len(env.temp_dir):]

findr/walker.odin

@@ -5,17 +5,42 @@ import "core:os"
 import "core:strings"
 import "core:sync"
 import "core:sys/linux"
+import "core:text/regex"
 import "core:thread"
 
-THREAD_COUNT :: 8
+IgnoreMode :: enum {
+	Respected, // skip gitignored, prune ignored dirs (fd -H default)
+	All, // ignore .gitignore entirely, descend everywhere (fd -HI)
+	Ignored, // emit ONLY gitignored files, prune ignored dirs (findr original)
+}
+
+WalkOptions :: struct {
+	pattern:        string, // regex on basename; "" = match all
+	excludes:       []string, // glob patterns to skip entirely (fd -E)
+	include_hidden: bool, // true = include dotfiles (fd -H)
+	ignore_mode:    IgnoreMode,
+}
 
 RawEntry :: struct {
 	name: string,
 	type: linux.Dirent_Type,
 }
 
+GIContext :: struct {
+	gi:       ^Gitignore, // nil if this dir had no .gitignore
+	base_rel: string, // relative path from repo root to this dir
+	parent:   ^GIContext, // parent context (nil if repo root)
+}
+
+WorkItem :: struct {
+	path:    string, // absolute directory path
+	rel:     string, // relative path from repo root ("" = root)
+	gi_ctx:  ^GIContext, // gitignore chain (nil = outside any repo)
+	in_repo: bool, // true if inside a git repo
+}
+
 WalkerPool :: struct {
-	queue:         [dynamic]string,
+	queue:         [dynamic]WorkItem,
 	queue_mutex:   sync.Mutex,
 	queue_sema:    sync.Atomic_Sema,
 	results:       ^[dynamic]string,
@@ -23,20 +48,54 @@ WalkerPool :: struct {
 	active:        i64,
 	done:          sync.One_Shot_Event,
 	threads:       [dynamic]^thread.Thread,
+	opts:          WalkOptions,
+	pattern_re:    regex.Regular_Expression,
+	has_pattern:   bool,
+	exclude_gi:    ^Gitignore,
+	all_contexts:  [dynamic]^GIContext,
+	contexts_lock: sync.Mutex,
 }
 
-walk :: proc(root: string, results: ^[dynamic]string) {
+walk :: proc(roots: []string, results: ^[dynamic]string, opts: WalkOptions, thread_count: int) {
+	if len(roots) == 0 do return
+
 	pool := new(WalkerPool)
-	pool.queue = make([dynamic]string)
+	pool.queue = make([dynamic]WorkItem)
 	pool.results = results
-	pool.active = 1
+	pool.active = i64(len(roots))
 	pool.threads = make([dynamic]^thread.Thread)
+	pool.all_contexts = make([dynamic]^GIContext)
+	pool.opts = opts
+	pool.exclude_gi = nil
+	pool.has_pattern = false
 
-	root_clone, _ := strings.clone(root)
+	if len(opts.pattern) > 0 {
-	append(&pool.queue, root_clone)
+		re, err := regex.create(opts.pattern, {regex.Flag.No_Capture})
-	sync.atomic_sema_post(&pool.queue_sema)
+		if err == nil {
+			pool.pattern_re = re
+			pool.has_pattern = true
+		}
+	}
 
-	for i in 0 ..< THREAD_COUNT {
+	if len(opts.excludes) > 0 {
+		sb: strings.Builder
+		strings.builder_init(&sb)
+		for ex in opts.excludes {
+			fmt.sbprintf(&sb, "%s\n", ex)
+		}
+		content := strings.to_string(sb)
+		pool.exclude_gi = new(Gitignore)
+		pool.exclude_gi^ = parse(content)
+		strings.builder_destroy(&sb)
+	}
+
+	for root in roots {
+		root_clone, _ := strings.clone(root)
+		append(&pool.queue, WorkItem{path = root_clone})
+		sync.atomic_sema_post(&pool.queue_sema)
+	}
+
+	for i in 0 ..< thread_count {
 		t := thread.create(walk_worker)
 		t.data = rawptr(pool)
 		t.init_context = context
@@ -46,7 +105,7 @@ walk :: proc(root: string, results: ^[dynamic]string) {
 
 	sync.one_shot_event_wait(&pool.done)
 
-	for _ in 0 ..< THREAD_COUNT {
+	for _ in 0 ..< thread_count {
 		sync.atomic_sema_post(&pool.queue_sema)
 	}
 
@@ -54,16 +113,44 @@ walk :: proc(root: string, results: ^[dynamic]string) {
 		thread.destroy(t)
 	}
 	delete(pool.threads)
-	for path in pool.queue {
+	for item in pool.queue {
-		delete(path)
+		delete(item.path)
+		if len(item.rel) > 0 {delete(item.rel)}
 	}
 	delete(pool.queue)
+
+	for ctx in pool.all_contexts {
+		if ctx.gi != nil {
+			destroy(ctx.gi)
+			free(ctx.gi)
+		}
+		if len(ctx.base_rel) > 0 {
+			delete(ctx.base_rel)
+		}
+		free(ctx)
+	}
+	delete(pool.all_contexts)
+
+	if pool.has_pattern {
+		regex.destroy(pool.pattern_re)
+	}
+	if pool.exclude_gi != nil {
+		destroy(pool.exclude_gi)
+		free(pool.exclude_gi)
+	}
+
 	free(pool)
 }
 
 walk_worker :: proc(t: ^thread.Thread) {
 	pool := cast(^WalkerPool)t.data
 
+	prof_thread_init("walker")
+	defer prof_thread_destroy()
+
+	local_results := make([dynamic]string, 0, 256)
+	defer delete(local_results)
+
 	for {
 		sync.atomic_sema_wait(&pool.queue_sema)
 
@@ -76,63 +163,166 @@ walk_worker :: proc(t: ^thread.Thread) {
 			break
 		}
 		last := len(pool.queue) - 1
-		dir_path := pool.queue[last]
+		item := pool.queue[last]
 		ordered_remove(&pool.queue, last)
 		sync.mutex_unlock(&pool.queue_mutex)
 
-		process_dir(pool, dir_path)
+		process_dir(pool, item, &local_results)
-		delete(dir_path)
+		delete(item.path)
+		if len(item.rel) > 0 {delete(item.rel)}
 
 		old := sync.atomic_sub_explicit(&pool.active, 1, .Release)
 		if old == 1 {
 			sync.one_shot_event_signal(&pool.done)
 		}
 	}
+
+	if len(local_results) > 0 {
+		sync.mutex_lock(&pool.results_mutex)
+		for res in local_results {
+			append(pool.results, res)
+		}
+		sync.mutex_unlock(&pool.results_mutex)
+	}
 }
 
-process_dir :: proc(pool: ^WalkerPool, dir_path: string) {
+process_dir :: proc(pool: ^WalkerPool, item: WorkItem, local_results: ^[dynamic]string) {
+	dir_path := item.path
 	has_git := false
 	entries := read_dir_entries(dir_path, &has_git)
 	defer free_entries(&entries)
 
+	gi_ctx := item.gi_ctx
+	rel := item.rel
+
 	if has_git {
-		gi := load_gitignore(dir_path)
+		gi_ctx = nil
-		defer if gi != nil {
+		rel = ""
-			destroy(gi)
+	}
-			free(gi)
+
+	child_in_repo := has_git || item.in_repo
+
+	gi := load_ignore_patterns(dir_path, child_in_repo)
+	if gi != nil {
+		new_ctx := new(GIContext)
+		new_ctx.gi = gi
+		if len(rel) > 0 {
+			new_ctx.base_rel, _ = strings.clone(rel)
+		}
+		new_ctx.parent = gi_ctx
+
+		sync.mutex_lock(&pool.contexts_lock)
+		append(&pool.all_contexts, new_ctx)
+		sync.mutex_unlock(&pool.contexts_lock)
+
+		gi_ctx = new_ctx
+	}
+
+	rel_buf: [4096]u8
+
+	for entry in entries {
+		if entry.name == ".git" do continue
+
+		is_dir := entry.type == .DIR
+		is_nondir := entry.type != .DIR
+
+		if pool.exclude_gi != nil && is_ignored(pool.exclude_gi, entry.name, is_dir) {
+			continue
 		}
 
-		for entry in entries {
+		if !pool.opts.include_hidden && len(entry.name) > 0 && entry.name[0] == '.' {
-			if entry.name == ".git" do continue
+			continue
-			is_dir := entry.type == .DIR
-			if gi != nil && is_ignored(gi, entry.name, is_dir) {
-				if !is_dir {
-					full_path := join_path(dir_path, entry.name)
-					sync.mutex_lock(&pool.results_mutex)
-					append(pool.results, full_path)
-					sync.mutex_unlock(&pool.results_mutex)
-				}
-				continue
-			}
-			if is_dir {
-				child_path := join_path(dir_path, entry.name)
-				push_work(pool, child_path)
-			}
 		}
-	} else {
+
-		for entry in entries {
+		entry_rel := build_rel(rel_buf[:], rel, entry.name)
-			if entry.type == .DIR {
+
+		ignored := false
+		if gi_ctx != nil && pool.opts.ignore_mode != .All {
+			ignored = check_chain(gi_ctx, entry_rel, is_dir)
+		}
+
+		should_emit: bool
+		if ignored {
+			should_emit = pool.opts.ignore_mode == .Ignored
+		} else {
+			should_emit = pool.opts.ignore_mode != .Ignored
+		}
+
+		if is_dir {
+			if should_emit && matches_pattern(pool, entry.name) {
+				dir_path_out := join_path_dir(dir_path, entry.name)
+				append(local_results, dir_path_out)
+			}
+			if !ignored {
+				child_rel, _ := strings.clone(entry_rel)
 				child_path := join_path(dir_path, entry.name)
-				push_work(pool, child_path)
+				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) {
+				full_path := join_path(dir_path, entry.name)
+				append(local_results, full_path)
 			}
 		}
 	}
 }
 
-push_work :: proc(pool: ^WalkerPool, path: string) {
+check_chain :: proc(ctx: ^GIContext, entry_rel: string, is_dir: bool) -> bool {
+	c := ctx
+	for c != nil {
+		if c.gi != nil {
+			rel := relative_to(entry_rel, c.base_rel)
+			match := check_match(c.gi, rel, is_dir)
+			if match != .None {
+				return match == .Ignored
+			}
+		}
+		c = c.parent
+	}
+	return false
+}
+
+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] == '/' &&
+	   strings.has_prefix(entry_rel, base_rel) {
+		return entry_rel[prefix_len + 1:]
+	}
+	return entry_rel
+}
+
+build_rel :: proc(buf: []u8, rel, name: string) -> string {
+	if len(rel) == 0 do return name
+	pos := copy(buf, rel)
+	if pos < len(buf) {
+		buf[pos] = '/'
+		pos += 1
+		pos += copy(buf[pos:], name)
+	}
+	return string(buf[:pos])
+}
+
+matches_pattern :: proc(pool: ^WalkerPool, name: string) -> bool {
+	if !pool.has_pattern do return true
+	cap, ok := regex.match(pool.pattern_re, name)
+	regex.destroy(cap)
+	return ok
+}
+
+push_work :: proc(pool: ^WalkerPool, item: WorkItem) {
 	sync.atomic_add_explicit(&pool.active, 1, .Relaxed)
 	sync.mutex_lock(&pool.queue_mutex)
-	append(&pool.queue, path)
+	append(&pool.queue, item)
 	sync.mutex_unlock(&pool.queue_mutex)
 	sync.atomic_sema_post(&pool.queue_sema)
 }
@@ -179,32 +369,66 @@ free_entries :: proc(entries: ^[dynamic]RawEntry) {
 	delete(entries^)
 }
 
-load_gitignore :: proc(dir_path: string) -> ^Gitignore {
+load_ignore_patterns :: proc(dir_path: string, in_repo: bool) -> ^Gitignore {
-	gi_path := join_path(dir_path, ".gitignore")
+	has_patterns := false
-	defer delete(gi_path)
+	sb: strings.Builder
+	strings.builder_init(&sb)
+	defer strings.builder_destroy(&sb)
 
-	data, err := os.read_entire_file_from_path(gi_path, context.allocator)
+	if in_repo {
-	if err != .NONE do return nil
+		gi_path := join_path(dir_path, ".gitignore")
+		data, err := os.read_entire_file_from_path(gi_path, context.allocator)
+		delete(gi_path)
+		if err == .NONE {
+			fmt.sbprintf(&sb, "%s", string(data))
+			delete(data)
+			has_patterns = true
+		}
+	}
 
+	ig_path := join_path(dir_path, ".ignore")
+	idata, ierr := os.read_entire_file_from_path(ig_path, context.allocator)
+	delete(ig_path)
+	if ierr == .NONE {
+		fmt.sbprintf(&sb, "%s", string(idata))
+		delete(idata)
+		has_patterns = true
+	}
+
+	if !has_patterns do return nil
+
+	content := strings.to_string(sb)
 	gi := new(Gitignore)
-	gi^ = parse(string(data))
+	gi^ = parse(content)
-	delete(data)
 	return gi
 }
 
 join_path :: proc(parent, child: string) -> string {
-	b: strings.Builder
+	need_sep := len(parent) == 0 || parent[len(parent) - 1] != '/'
-	strings.builder_init(&b)
+	total := len(parent) + len(child)
-	defer strings.builder_destroy(&b)
+	if need_sep do total += 1
-
+	buf := make([]u8, total, context.allocator)
-	fmt.sbprintf(&b, "%s", parent)
+	pos := copy(buf, parent)
-	if len(parent) == 0 || parent[len(parent) - 1] != '/' {
+	if need_sep {
-		fmt.sbprintf(&b, "/")
+		buf[pos] = '/'
+		pos += 1
 	}
-	fmt.sbprintf(&b, "%s", child)
+	copy(buf[pos:], child)
-
+	return string(buf)
-	s := strings.to_string(b)
+}
-	result, _ := strings.clone(s)
+
-	return result
+join_path_dir :: proc(parent, child: string) -> string {
+	need_sep := len(parent) == 0 || parent[len(parent) - 1] != '/'
+	total := len(parent) + len(child) + 1 // +1 for trailing '/'
+	if need_sep do total += 1
+	buf := make([]u8, total, context.allocator)
+	pos := copy(buf, parent)
+	if need_sep {
+		buf[pos] = '/'
+		pos += 1
+	}
+	pos += copy(buf[pos:], child)
+	buf[pos] = '/'
+	return string(buf)
 }
 

flake.nix

@@ -107,6 +107,7 @@
               zip
 
               # Helper tools
+              delta
               hyperfine
 
               # IDE