chore(dev): release 0.3.1

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

.gitignore

@@ -7,8 +7,12 @@ list.json
 man 
 
 # build artifacts
+*.spall
 builds
 envr
 envr-go
+findr/findr
+findr/findr-prof
+findr/bench-*.md
 result
 version.odin

CHANGELOG.md

@@ -1,5 +1,22 @@
 # Changelog
 
+## [0.3.1](https://github.com/sbrow/envr/compare/v0.3.0...v0.3.1) (2026-06-17)
+
+
+### Bug Fixes
+
+* **find:** Ignored volatile dirs in benchmarks. ([6fe6125](https://github.com/sbrow/envr/commit/6fe61251ee2fe86bca3b50a5893699b5a71b1af8))
+
+
+### Performance Improvements
+
+* **findr:** Added spall support. ([ce57009](https://github.com/sbrow/envr/commit/ce57009b9217926c9e171cb86043463706882bc3))
+* **findr:** Each thread gets its own buffer. ([cbab562](https://github.com/sbrow/envr/commit/cbab562d62d74595ce5578ebaba8d839affa4e7f))
+* **findr:** Improved performance of `join_path*` procedures. ([f51c0d6](https://github.com/sbrow/envr/commit/f51c0d6755a3ba6a54d9292c1a95e13d460df7b9))
+* **findr:** Replaced regex engine with glob. ([19d03ff](https://github.com/sbrow/envr/commit/19d03ff71a3ae0e2f8f1d4fe8e7e0cd11165405f))
+* **findr:** Use buffered writer. ([116ed6d](https://github.com/sbrow/envr/commit/116ed6de4c3e08909b25f35c458913022cafefdd))
+* Replaced `fd` with custom internals. ([440f944](https://github.com/sbrow/envr/commit/440f944b33b20cc5f86a7716af0ba6e793c6c094))
+
 ## [0.3.0](https://github.com/sbrow/envr/compare/v0.2.1...v0.3.0) (2026-06-16)
 
 Version 0.3.0 represents a significant departure (and improvement) for envr.

TODOS.md

@@ -32,8 +32,6 @@
 
 18. 2 scan tests silently skip when fd isn't installed, tests pass without actually testing anything. These should use #assert to be sure that fd is in path.
 
-19. Try to do all encryption / decryption in memory - only read / write encrypted data to disk.
-
 20. add --format -f flag to commands that draw tables.
 
 21. Replace `testing.expect` calls with `testing.expect_value` calls where appropriate.

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

@@ -0,0 +1,342 @@
+# findr — Native Odin File Finder (fd Replacement)
+
+## Overview
+
+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.
+
+## Directory Structure
+
+```
+findr/
+  findr.odin           # main + CLI (hand-rolled arg parsing)
+  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
+  gitignore_test.odin  # transpilation + matching unit tests (22 tests)
+```
+
+## CLI Interface
+
+```
+findr [-I] [--ignored] [--no-hidden] [-E <glob>]... [pattern] [path]...
+```
+
+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
+```
+
+## Architecture
+
+### Two Orthogonal Axes (matching fd's semantics)
+
+1. **Hidden files** (`.` prefix): `include_hidden=true` includes them, `false` excludes them
+2. **Gitignore**: three modes (see `IgnoreMode` below)
+
+### Types
+
+```odin
+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,
+}
+```
+
+### process_dir Filtering Order Per Entry
+
+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`).
+
+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**:
+
+| Mode | gitignored file | gitignored dir | normal file | normal dir |
+|---|---|---|---|---|
+| `.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 |
+
+**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.
+
+### Performance Architecture
+
+- **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.
+
+### 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 |
+| `test_non_repo_not_scanned` | Dirs without `.git/` produce no output |
+| `test_negation_pattern` | `!prod.env` un-ignores a file |
+| `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_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 |
+
+**`.All` mode (fd -HI parity):**
+
+| 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
+
+| Gitignore pattern | Regex | Notes |
+|---|---|---|
+| `foo` | `(^|/)foo(/.*)?$` | matches at any depth |
+| `/foo` | `^foo(/.*)?$` | anchored to gitignore dir |
+| `foo/` | `(^|/)foo/.*$` | directory only |
+| `*.log` | `(^|/)[^/]*\.log$` | `*` = any chars except `/` |
+| `**/foo` | `(^|/)(.*/)?foo(/.*)?$` | `**` = any chars including `/` |
+| `foo/**/bar` | `(^|/)foo/(.*/)?bar(/.*)?$` | `**` between segments |
+| `!pattern` | (handled by layer) | negation flag, not regex |
+| `#comment` | (skipped) | |
+| `[abc]` | `[abc]` | same regex syntax |
+| `?` | `[^/]` | single char, no `/` |
+
+## Implementation Phases
+
+### Phase 1: Gitignore Transpiler + Tests ✅
+
+22 tests, all passing, zero leaks.
+
+### Phase 2: findr Walker + Tests ✅
+
+Parallel DFS using getdents with worker thread pool. 32 total tests pass, zero leaks.
+
+### Phase 3: Parallel Traversal ✅
+
+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.
+
+### Phase 4: Benchmark ✅
+
+findr found 227 gitignored files on `~` in 852ms. fd's double-run walked ~1.1M entries.
+
+### Phase 5: fd-Parity API ✅
+
+**Goal:** Make findr replicate specific fd commands for A/B benchmarking, plus keep the unique gitignored-only mode.
+
+**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
+
+**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 |
+|---|---|
+| Gitignore edge cases (`**/foo`, `foo/**/bar`) | Comprehensive gitignore_test.odin with spec examples |
+| `DT_UNKNOWN` on XFS/BTRFS/FUSE | Phase 6 Step 3: stat fallback for unknown types |
+| 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). |
+| 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

@@ -0,0 +1,91 @@
+package findr
+
+import "core:bufio"
+import "core:os"
+import "core:strings"
+
+main :: proc() {
+	prof_init()
+	defer prof_destroy()
+
+	args := os.args
+
+	opts: WalkOptions
+	opts.include_hidden = false
+	opts.ignore_mode = .Respected
+
+	excludes := make([dynamic]string)
+	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(paths) == 0 && pattern != "" && os.exists(pattern) {
+		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)}
+		delete(results)
+	}
+
+	thread_count := os.get_processor_core_count()
+	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 {
+		bufio.writer_write_string(&w, r)
+		bufio.writer_write_byte(&w, '\n')
+	}
+	bufio.writer_flush(&w)
+}
+

findr/findr_test.odin

@@ -0,0 +1,474 @@
+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()
+	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 = .Ignored}, {
+		"repo/.env", "repo/secrets.env",
+	})
+}
+
+@(test)
+test_non_repo_not_scanned :: proc(t: ^testing.T) {
+	env := create_test_env()
+	defer destroy_test_env(&env)
+
+	create_dir(env, "norepo")
+	create_file(env, "norepo/.gitignore", "*.env\n")
+	create_file(env, "norepo/.env")
+
+	assert_output_empty(t, env, nil, {include_hidden = true, ignore_mode = .Ignored})
+}
+
+@(test)
+test_negation_pattern :: 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!prod.env\n")
+	create_file(env, "repo/.env")
+	create_file(env, "repo/secrets.env")
+	create_file(env, "repo/prod.env")
+
+	assert_output(t, env, nil, {include_hidden = true, ignore_mode = .Ignored}, {
+		"repo/.env", "repo/secrets.env",
+	})
+}
+
+@(test)
+test_dir_only_pattern :: proc(t: ^testing.T) {
+	env := create_test_env()
+	defer destroy_test_env(&env)
+
+	create_git_repo(env, "repo")
+	create_file(env, "repo/.gitignore", "node_modules/\n")
+	create_file(env, "repo/node_modules", "should not match (it's a file)")
+
+	create_dir(env, "repo/ignored_dir")
+	create_file(env, "repo/.gitignore", "ignored_dir/\n")
+
+	assert_output(t, env, nil, {include_hidden = true, ignore_mode = .Ignored}, {
+		"repo/ignored_dir/",
+	})
+}
+
+@(test)
+test_multiple_repos :: proc(t: ^testing.T) {
+	env := create_test_env()
+	defer destroy_test_env(&env)
+
+	create_git_repo(env, "repo1")
+	create_file(env, "repo1/.gitignore", "*.env\n")
+	create_file(env, "repo1/a.env")
+
+	create_git_repo(env, "repo2")
+	create_file(env, "repo2/.gitignore", "*.key\n")
+	create_file(env, "repo2/secret.key")
+
+	assert_output(t, env, nil, {include_hidden = true, ignore_mode = .Ignored}, {
+		"repo1/a.env", "repo2/secret.key",
+	})
+}
+
+@(test)
+test_nested_repos :: proc(t: ^testing.T) {
+	env := create_test_env()
+	defer destroy_test_env(&env)
+
+	create_git_repo(env, "parent")
+	create_file(env, "parent/.gitignore", "*.env\n")
+	create_file(env, "parent/top.env")
+
+	create_git_repo(env, "parent/child")
+	create_file(env, "parent/child/.gitignore", "*.key\n")
+	create_file(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_nested_gitignore_read :: 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_dir(env, "repo/sub")
+	create_file(env, "repo/sub/.gitignore", "*.txt\n")
+	create_file(env, "repo/sub/secret.txt")
+	create_file(env, "repo/sub/.env")
+
+	// Both root and nested .gitignore are read.
+	// secret.txt: ignored by sub/.gitignore (*.txt)
+	// .env: ignored by root .gitignore (*.env)
+	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)
+test_no_gitignore_file :: proc(t: ^testing.T) {
+	env := create_test_env()
+	defer destroy_test_env(&env)
+
+	create_git_repo(env, "repo")
+	create_file(env, "repo/.env")
+
+	assert_output_empty(t, env, nil, {include_hidden = true, ignore_mode = .Ignored})
+}
+
+@(test)
+test_empty_gitignore :: proc(t: ^testing.T) {
+	env := create_test_env()
+	defer destroy_test_env(&env)
+
+	create_git_repo(env, "repo")
+	create_file(env, "repo/.gitignore", "\n\n# comment\n\n")
+	create_file(env, "repo/.env")
+
+	assert_output_empty(t, env, nil, {include_hidden = true, ignore_mode = .Ignored})
+}
+
+@(test)
+test_multiple_search_dirs :: proc(t: ^testing.T) {
+	env := create_test_env()
+	defer destroy_test_env(&env)
+
+	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")
+	create_file(env, "dir2/repo/b.env")
+
+	dir1 := join_path(env.temp_dir, "dir1")
+	defer delete(dir1)
+	dir2 := join_path(env.temp_dir, "dir2")
+	defer delete(dir2)
+
+	results := make([dynamic]string)
+	defer {
+		for r in results {delete(r)}
+		delete(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

@@ -0,0 +1,88 @@
+package findr
+
+import "core:strings"
+
+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)
+
+	remaining := content
+	for {
+		line, ok := strings.split_lines_iterator(&remaining)
+		if !ok do break
+
+		s := strings.trim_space(line)
+		if len(s) == 0 do continue
+		if s[0] == '#' do continue
+
+		negated := false
+		if s[0] == '!' {
+			negated = true
+			s = s[1:]
+		}
+
+		if len(s) > 0 && s[0] == '\\' {
+			if len(s) > 1 && (s[1] == '#' || s[1] == '!') {
+				s = s[1:]
+			}
+		}
+
+		dir_only := false
+		if len(s) > 0 && s[len(s) - 1] == '/' {
+			dir_only = true
+			s = s[:len(s) - 1]
+		}
+
+		anchored := false
+		if len(s) > 0 && s[0] == '/' {
+			anchored = true
+			s = s[1:]
+		}
+
+		if len(s) == 0 do continue
+
+		gp := glob_compile(s, anchored)
+		append(&gi.rules, Rule{pattern = gp, negated = negated, dir_only = dir_only})
+	}
+
+	return gi
+}
+
+destroy :: proc(gi: ^Gitignore) {
+	for &rule in gi.rules {
+		glob_destroy(&rule.pattern)
+	}
+	delete(gi.rules)
+}
+

findr/gitignore_test.odin

@@ -0,0 +1,219 @@
+package findr
+
+import "core:testing"
+
+@(test)
+test_glob_simple :: proc(t: ^testing.T) {
+	testing.expect(t, glob_match("foo", "foo", false))
+	testing.expect(t, glob_match("foo", "bar/foo", false))
+	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) {
+	testing.expect(t, glob_match("foo", "foo", true))
+	testing.expect(t, !glob_match("foo", "bar/foo", true))
+	testing.expect(t, !glob_match("foo", "foobar", true))
+}
+
+@(test)
+test_glob_star :: proc(t: ^testing.T) {
+	testing.expect(t, glob_match("*.log", "test.log", false))
+	testing.expect(t, glob_match("*.log", ".log", false))
+	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) {
+	testing.expect(t, glob_match("?.log", "a.log", false))
+	testing.expect(t, !glob_match("?.log", "ab.log", false))
+	testing.expect(t, !glob_match("?.log", ".log", false))
+}
+
+@(test)
+test_glob_char_class :: proc(t: ^testing.T) {
+	testing.expect(t, glob_match("[abc].log", "a.log", false))
+	testing.expect(t, glob_match("[abc].log", "b.log", false))
+	testing.expect(t, !glob_match("[abc].log", "d.log", false))
+}
+
+@(test)
+test_glob_negated_class :: proc(t: ^testing.T) {
+	testing.expect(t, glob_match("[!abc].log", "d.log", false))
+	testing.expect(t, !glob_match("[!abc].log", "a.log", false))
+}
+
+@(test)
+test_glob_dot_literal :: proc(t: ^testing.T) {
+	testing.expect(t, glob_match(".env", ".env", false))
+	testing.expect(t, glob_match(".env", "dir/.env", false))
+	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) {
+	testing.expect(t, glob_match("**/foo", "foo", false))
+	testing.expect(t, glob_match("**/foo", "a/b/foo", false))
+	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) {
+	testing.expect(t, glob_match("abc/**", "abc/x", false))
+	testing.expect(t, glob_match("abc/**", "abc/x/y", false))
+	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) {
+	testing.expect(t, glob_match("foo/**/bar", "foo/bar", false))
+	testing.expect(t, glob_match("foo/**/bar", "foo/x/bar", false))
+	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) {
+	testing.expect(t, glob_match("\\!foo", "!foo", false))
+	testing.expect(t, !glob_match("\\!foo", "foo", false))
+}
+
+@(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) {
+	testing.expect(t, glob_match("", "", false))
+	testing.expect(t, !glob_match("", "foo", false))
+}
+
+@(test)
+test_is_ignored_basic :: proc(t: ^testing.T) {
+	gi := parse("*.env\n")
+	defer destroy(&gi)
+
+	testing.expect_value(t, is_ignored(&gi, ".env", false), true)
+	testing.expect_value(t, is_ignored(&gi, "foo.env", false), true)
+	testing.expect_value(t, is_ignored(&gi, ".env.local", false), false)
+	testing.expect_value(t, is_ignored(&gi, "config.yaml", false), false)
+}
+
+@(test)
+test_is_ignored_negation :: proc(t: ^testing.T) {
+	gi := parse("*.env\n!.env.production\n")
+	defer destroy(&gi)
+
+	testing.expect_value(t, is_ignored(&gi, ".env", false), true)
+	testing.expect_value(t, is_ignored(&gi, ".env.production", false), false)
+}
+
+@(test)
+test_is_ignored_dir_only :: proc(t: ^testing.T) {
+	gi := parse("node_modules/\n")
+	defer destroy(&gi)
+
+	testing.expect_value(t, is_ignored(&gi, "node_modules", true), true)
+	testing.expect_value(t, is_ignored(&gi, "node_modules", false), false)
+}
+
+@(test)
+test_is_ignored_anchored :: proc(t: ^testing.T) {
+	gi := parse("/secret.key\n")
+	defer destroy(&gi)
+
+	testing.expect_value(t, is_ignored(&gi, "secret.key", false), true)
+}
+
+@(test)
+test_is_ignored_comments_skipped :: proc(t: ^testing.T) {
+	gi := parse("# this is a comment\n#another\n*.tmp\n")
+	defer destroy(&gi)
+
+	testing.expect_value(t, len(gi.rules), 1)
+	testing.expect_value(t, is_ignored(&gi, "file.tmp", false), true)
+}
+
+@(test)
+test_is_ignored_blank_lines_skipped :: proc(t: ^testing.T) {
+	gi := parse("\n\n  \n*.log\n\n")
+	defer destroy(&gi)
+
+	testing.expect_value(t, len(gi.rules), 1)
+}
+
+@(test)
+test_is_ignored_last_match_wins :: proc(t: ^testing.T) {
+	gi := parse("*.env\n!*.env\n")
+	defer destroy(&gi)
+
+	testing.expect_value(t, is_ignored(&gi, ".env", false), false)
+}
+
+@(test)
+test_is_ignored_no_rules :: proc(t: ^testing.T) {
+	gi := parse("")
+	defer destroy(&gi)
+
+	testing.expect_value(t, is_ignored(&gi, "anything", false), false)
+}
+
+@(test)
+test_is_ignored_env_pattern :: proc(t: ^testing.T) {
+	gi := parse(".env*\n")
+	defer destroy(&gi)
+
+	testing.expect_value(t, is_ignored(&gi, ".env", false), true)
+	testing.expect_value(t, is_ignored(&gi, ".env.local", false), true)
+	testing.expect_value(t, is_ignored(&gi, ".envrc", false), true)
+}
+
+@(test)
+test_is_ignored_globstar :: proc(t: ^testing.T) {
+	gi := parse("**/cache\n")
+	defer destroy(&gi)
+
+	testing.expect_value(t, is_ignored(&gi, "cache", false), true)
+	testing.expect_value(t, is_ignored(&gi, "foo/cache", false), true)
+	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

@@ -0,0 +1,152 @@
+package findr
+
+import "core:fmt"
+import "core:log"
+import "core:os"
+import "core:sort"
+import "core:strings"
+import "core:testing"
+
+TestEnv :: struct {
+	temp_dir: string,
+}
+
+create_test_env :: proc() -> (env: TestEnv) {
+	tmp, err := os.mkdir_temp("", "findr-test-*", context.allocator)
+	if err != nil {
+		log.error("Failed to create temp dir:", err)
+		panic("Failed to create temp dir")
+	}
+
+	env.temp_dir = tmp
+	return
+}
+
+destroy_test_env :: proc(env: ^TestEnv) {
+	os.remove_all(env.temp_dir)
+	delete(env.temp_dir)
+}
+
+create_dir :: proc(env: TestEnv, path: string) {
+	full := join_path(env.temp_dir, path)
+	defer delete(full)
+	os.mkdir_all(full, os.Permissions_Default_Directory)
+}
+
+create_file :: proc(env: TestEnv, path: string, content: string = "") {
+	full := join_path(env.temp_dir, path)
+	defer delete(full)
+
+	dir_end := strings.last_index(full, "/")
+	if dir_end >= 0 {
+		dir_path := full[:dir_end]
+		os.mkdir_all(dir_path, os.Permissions_Default_Directory)
+	}
+
+	f, err := os.create(full)
+	if err != nil {
+		log.error("Failed to create file:", full, err)
+		return
+	}
+	if len(content) > 0 {
+		os.write_string(f, content)
+	}
+	os.close(f)
+}
+
+create_git_repo :: proc(env: TestEnv, path: string) {
+	sub := join_path(path, ".git")
+	defer delete(sub)
+	create_dir(env, sub)
+}
+
+assert_output :: proc(
+	t: ^testing.T,
+	env: TestEnv,
+	args: []string,
+	opts: WalkOptions,
+	expected: []string,
+) {
+	results := collect_results(env, args, opts)
+	defer {
+		for r in results {delete(r)}
+		delete(results)
+	}
+
+	sorted_expected := make([dynamic]string, 0, len(expected))
+	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)}
+	defer delete(sorted_actual)
+
+	sort.quick_sort(sorted_expected[:])
+	sort.quick_sort(sorted_actual[:])
+
+	if len(sorted_expected) != len(sorted_actual) {
+		testing.fail(t)
+		log.error(
+			fmt.tprintf("Expected %d results, got %d", len(sorted_expected), len(sorted_actual)),
+		)
+		log.error("Expected:", sorted_expected[:])
+		log.error("Actual:  ", sorted_actual[:])
+		return
+	}
+
+	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))
+			log.error("Expected:", sorted_expected[:])
+			log.error("Actual:  ", sorted_actual[:])
+			return
+		}
+	}
+}
+
+assert_output_empty :: proc(
+	t: ^testing.T,
+	env: TestEnv,
+	args: []string,
+	opts: WalkOptions,
+) {
+	results := collect_results(env, args, opts)
+	defer {
+		for r in results {delete(r)}
+		delete(results)
+	}
+	if len(results) > 0 {
+		testing.fail(t)
+		log.error(fmt.tprintf("Expected no results, got %d:", len(results)))
+		for r in results {
+			log.error("  ", r)
+		}
+	}
+}
+
+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)}
+
+	thread_count := os.get_processor_core_count()
+	walk(full_args[:], &results, opts, thread_count)
+
+	for i in 0 ..< len(results) {
+		r := results[i]
+		if strings.has_prefix(r, env.temp_dir) {
+			stripped := r[len(env.temp_dir):]
+			if len(stripped) > 0 && stripped[0] == '/' {
+				stripped = stripped[1:]
+			}
+			new_r, _ := strings.clone(stripped)
+			delete(r)
+			results[i] = new_r
+		}
+	}
+
+	return results
+}

findr/walker.odin

@@ -0,0 +1,434 @@
+package findr
+
+import "core:fmt"
+import "core:os"
+import "core:strings"
+import "core:sync"
+import "core:sys/linux"
+import "core:text/regex"
+import "core:thread"
+
+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]WorkItem,
+	queue_mutex:   sync.Mutex,
+	queue_sema:    sync.Atomic_Sema,
+	results:       ^[dynamic]string,
+	results_mutex: sync.Mutex,
+	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(roots: []string, results: ^[dynamic]string, opts: WalkOptions, thread_count: int) {
+	if len(roots) == 0 do return
+
+	pool := new(WalkerPool)
+	pool.queue = make([dynamic]WorkItem)
+	pool.results = results
+	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
+
+	if len(opts.pattern) > 0 {
+		re, err := regex.create(opts.pattern, {regex.Flag.No_Capture})
+		if err == nil {
+			pool.pattern_re = re
+			pool.has_pattern = true
+		}
+	}
+
+	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
+		thread.start(t)
+		append(&pool.threads, t)
+	}
+
+	sync.one_shot_event_wait(&pool.done)
+
+	for _ in 0 ..< thread_count {
+		sync.atomic_sema_post(&pool.queue_sema)
+	}
+
+	for t in pool.threads {
+		thread.destroy(t)
+	}
+	delete(pool.threads)
+	for item in pool.queue {
+		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)
+
+		sync.mutex_lock(&pool.queue_mutex)
+		if len(pool.queue) == 0 {
+			sync.mutex_unlock(&pool.queue_mutex)
+			if sync.atomic_load_explicit(&pool.active, .Acquire) == 0 {
+				sync.one_shot_event_signal(&pool.done)
+			}
+			break
+		}
+		last := len(pool.queue) - 1
+		item := pool.queue[last]
+		ordered_remove(&pool.queue, last)
+		sync.mutex_unlock(&pool.queue_mutex)
+
+		process_dir(pool, item, &local_results)
+		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, 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_ctx = nil
+		rel = ""
+	}
+
+	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
+		}
+
+		if !pool.opts.include_hidden && len(entry.name) > 0 && entry.name[0] == '.' {
+			continue
+		}
+
+		entry_rel := build_rel(rel_buf[:], rel, entry.name)
+
+		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,
+					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)
+			}
+		}
+	}
+}
+
+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, item)
+	sync.mutex_unlock(&pool.queue_mutex)
+	sync.atomic_sema_post(&pool.queue_sema)
+}
+
+read_dir_entries :: proc(dir_path: string, has_git: ^bool) -> [dynamic]RawEntry {
+	entries := make([dynamic]RawEntry)
+
+	cpath := strings.clone_to_cstring(dir_path)
+	if cpath == nil do return entries
+
+	fd, err := linux.open(cpath, {.DIRECTORY, .CLOEXEC})
+	delete(cpath)
+	if err != .NONE do return entries
+
+	buf: [8192]u8
+	has_git^ = false
+
+	for {
+		n, errno := linux.getdents(fd, buf[:])
+		if n <= 0 || errno != .NONE do break
+
+		offs := 0
+		for d in linux.dirent_iterate_buf(buf[:n], &offs) {
+			name := linux.dirent_name(d)
+			if name == "." || name == ".." do continue
+
+			if name == ".git" && d.type == .DIR {
+				has_git^ = true
+			}
+
+			cloned := strings.clone(name)
+			append(&entries, RawEntry{name = cloned, type = d.type})
+		}
+	}
+
+	linux.close(fd)
+	return entries
+}
+
+free_entries :: proc(entries: ^[dynamic]RawEntry) {
+	for &entry in entries {
+		delete(entry.name)
+	}
+	delete(entries^)
+}
+
+load_ignore_patterns :: proc(dir_path: string, in_repo: bool) -> ^Gitignore {
+	has_patterns := false
+	sb: strings.Builder
+	strings.builder_init(&sb)
+	defer strings.builder_destroy(&sb)
+
+	if in_repo {
+		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(content)
+	return gi
+}
+
+join_path :: proc(parent, child: string) -> string {
+	need_sep := len(parent) == 0 || parent[len(parent) - 1] != '/'
+	total := len(parent) + len(child)
+	if need_sep do total += 1
+	buf := make([]u8, total, context.allocator)
+	pos := copy(buf, parent)
+	if need_sep {
+		buf[pos] = '/'
+		pos += 1
+	}
+	copy(buf[pos:], child)
+	return string(buf)
+}
+
+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

@@ -106,6 +106,10 @@
               # Build tools
               zip
 
+              # Helper tools
+              delta
+              hyperfine
+
               # IDE
               unstable.helix
               typescript-language-server

version.txt

@@ -1 +1 @@
-0.3.0
+0.3.1