Wednesday, October 28, 2026 · macOS 15.7 (Darwin 24.6.0), APFS case-insensitive

Dogfood on scott-mbp: Individual developer

6 repos · 30 commands · 5 broken · 0 slow · 4 filed

As a solo developer on macOS (APFS, case-insensitive) I re-checked all six fixed issues (five are fixed; #2236 stat/numstat is still off by one against grit's own patch and was reopened), then ran the daily loop, macOS filesystem cases, interrupts, a 100k-file tree, hooks/odd content and a small grit-lib program. Speed and memory are mostly at or better than git; I filed four new bugs (#2244, #2247, #2253, #2262), mostly state/integrity problems around commit, hooks, binary patches and case handling.

os: macOS 15.7 (Darwin 24.6.0), APFS case-insensitive · arm64
machine: scott-mbp (agador.sparticus), Apple M3, 24 GB RAM
versions: grit 0.5.0 (PATH); tested build is main · main 7071b202 · 2.52.0

repositories6

repo	files	commits	.git	notes
rust-lang/cargo (git clone)	3,078	23,591	81 MB
rust-lang/cargo (grit clone)	3,078	23,591	90 MB	fsck clean; same 218 refs
scratch/wide-100k	100,000	3	16 MB	400 dirs x 250 small files
scratch/big-600mb	1	1	573 MB	one 600 MB random blob
scratch/bin	2	2	9 MB	binary patch repro
octocat/Hello-World	1	3	1 MB

commandsgrit vs. system git, wall time

command	repo	grit	git	grit/git	result	notes
grit status	rust-lang/cargo (git clone)	21 ms	40 ms	0.5×	ok
grit diff	rust-lang/cargo (git clone)	8.1 ms	7.5 ms	1.1×	ok
grit log -n 50 --oneline	rust-lang/cargo (git clone)	4.0 ms	5.2 ms	0.8×	ok
grit log --oneline -- src/util/workspace.rs	rust-lang/cargo (git clone)	76 ms	73 ms	1.0×	ok
grit blame src/util/workspace.rs	rust-lang/cargo (git clone)	45 ms	64 ms	0.7×	ok
grit log -S CargoResult -n 100000 --json	rust-lang/cargo (git clone)	1.06 s	1.93 s	0.5×	ok	1154 commits, same as git; -G, --grep -i, date range and --first-parent counts also identical
grit diff --stat A B (3000..2900 back)	rust-lang/cargo (git clone)	—	—	—	wrong	counts off by one on several files (#2236 not fixed)
grit reset --hard master (from master~3000)	rust-lang/cargo (git clone)	308 ms	—	—	ok	SIGINT/TERM: clean cancel
grit restore --source master~3000 -- .	rust-lang/cargo (git clone)	655 ms	—	—	ok	prints one 'Restored <path>' line per file (3000 lines)
grit stash / stash pop (2879 modified files)	rust-lang/cargo (git clone)	866 ms	—	—	ok	pop 271 ms; SIGINT ignored until done
grit gc	rust-lang/cargo (git clone)	230 ms	1.08 s	0.2×	ok	already-packed repo
grit clone https://github.com/rust-lang/cargo	rust-lang/cargo (grit clone)	32.51 s	107.97 s	0.3×	ok	network bound; git run was earlier and slower, not a controlled comparison
grit status (first, after clone)	rust-lang/cargo (grit clone)	22 ms	39 ms	0.6×	ok
grit commit -m big (100000 new files)	scratch/wide-100k	7.92 s	28.01 s	0.3×	ok	git add -A took 28 s (git commit extra 0.77 s)
grit status	scratch/wide-100k	114 ms	105 ms	1.1×	ok
grit diff --stat	scratch/wide-100k	68 ms	65 ms	1.0×	ok
grit add d010/f010.txt	scratch/wide-100k	71 ms	39 ms	1.8×	ok	1.8x; known #1682 family
grit commit -m x (1 modified file)	scratch/wide-100k	260 ms	110 ms	2.4×	ok	2.4x time, 2.0x memory; known #1884
grit clone <local 100k-file repo>	scratch/wide-100k	3.92 s	—	—	ok
grit diff / diff --stat / show / log -p after editing 100 bytes of a 600 MB blob	scratch/big-600mb	340 ms	11.98 s	0.0×	ok	#2237 fixed
grit diff --patch-text A B (modified binary)	scratch/bin	—	—	—	wrong	'literal 0' binary patch, git apply fails (#2244)
grit add blob.bin (600 MB)	scratch/big-600mb	9.81 s	11.27 s	0.9×	ok
grit commit x6 + grit gc (100 MB asset, 1 KB edits)	scratch/big-600mb	6.99 s	4.13 s	1.7×	ok	commits 7 s vs git 13 s; .git 96 MB vs 97 MB after gc
grit commit (auto-stage) after dir case rename	scratch/case	—	—	—	wrong	renames tracked paths (#2247)
grit commit with failing pre-commit hook	scratch/hk	—	—	—	wrong	index left auto-staged (#2253)
grit commit after resolving a pick conflict	scratch/pc	—	—	—	wrong	author lost, CHERRY_PICK_HEAD left (#2262)
grit push / pull / pull --ff-only / push --force-with-lease	scratch/solo	—	—	—	ok	local bare remote; exit codes 1 (rejected) and 128 (not ff); remote fsck clean
grit history reword/drop/undo/squash, rebase, pick	scratch/rb	—	—	—	ok	fsck clean, change-id preserved
grit blame --json vs git blame --porcelain (60 files)	rust-lang/cargo (git clone)	—	—	—	ok	17 files with default and -w: 0 differences; -C -C differs from git on 66 of 115 files (heuristic, not filed)
commit_worktree with FlagCancel (library)	scratch/wide-100k	352 ms	—	—	ok	cancel at 100 ms honoured at 352 ms; cancel at 20 ms at 77 ms; at 900 ms at 901 ms

worked9

›status, diff, log, blame, path-log and -S/-G/--grep/date filters on cargo match git and are as fast or faster
›grit add/commit of 100k files is 3.5x faster than git; clone of 100k files in 3.9 s
›Cancellation of reset --hard, restore, clone and library commit leaves no lock and an unchanged repository
›Error messages name the problem and the next command (dirty switch, untracked overwrite, rejected push, ff-only, tag -a without message)
›Safe defaults: clean is a dry run, reset warns when commits become unreachable, switch refuses to overwrite
›change-id headers kept across rebase, history reword/drop/squash/undo; history undo restores refs exactly
›NFC normalisation, symlinks, exec bit, odd file names and worktrees on APFS interoperate with git (fsck clean)
›--json validated on 26 commands; --markdown correctly widens fences around embedded code fences
›grit-lib quick start compiles from the docs; typed Error variants (NothingToCommit, Hook with output, Cancelled) are usable

problems11

›#2244 diff/show --patch-text of a modified binary emits 'literal 0' (and an empty body for worktree/staged diffs); git apply rejects it (filed)
›#2236 stat/numstat still off by one vs patch/git on commit-pair diffs (verify_fix: not fixed, reopened)
›#2247 core.ignorecase auto-stage renames tracked dirs/files to the on-disk case, rewriting unrelated paths (filed)
›#2253 failing pre-commit/commit-msg hook leaves auto-staged index and staged untracked files (filed; same at library level)
›#2262 grit commit after a pick/revert conflict loses the picked author and leaves CHERRY_PICK_HEAD; --continue/--abort fail (filed)
›grit blame -C -C differs from git on 66 of 115 cargo files, mostly blank/trivial lines attributed to other files (git ignores lines below its interesting-line threshold); not filed, low value and heuristic
›grit restore prints one 'Restored <path>' line per file (3000 lines on cargo); not filed, cosmetic
›grit log on an unborn branch prints the 'make one with grit commit' hint twice; grit init has no -q; numstat renames print 'old => new' instead of git's {a => b} form; not filed, cosmetic
›stash ignores SIGINT until finished (~0.9 s on 2.9k files) and library commit cancel has a ~250 ms un-polled window; both covered by #2142
›gc of 6x100 MB churn is 1.7x slower than git (7.0 s vs 4.1 s); within threshold, not filed
›1-file commit in 100k-file repo is 2.4x time / 2x memory vs git; known #1884

bugs filed4

#2244 #2247 #2253 #2262already reported: #1884 #1682 #2142 #2183 #1014 #2155 #2201

next time

›Case-insensitive checkout paths beyond identical leaves: merge/switch/pick when a branch adds a file differing only by case, and git-created repos with differing dir case
›Test grit commit/pick/revert/merge finishing paths for authorship and state cleanup (pick -x, revert --mainline, merge --continue vs commit) and the stale-state recovery commands
›Real ssh push/fetch with thin packs and two concurrent pushers (needs sshd) and fsck after each
›Interrupt matrix for merge, pick, rebase --continue, gc and push on a repo with 100k+ files, including kill -9 and index.lock held by an editor
›fsmonitor: core.fsmonitor hook usage and timings on a 200k-file tree
›Library: Remote::fetch into a custom ObjectStore and a TUI-style loop over typed APIs; measure the un-polled cancel window in commit_worktree
›Full blame -C parity sweep on git/git with a minimal repro for the blank-line attribution difference