Thursday, October 29, 2026 · Ubuntu 24.04.5 LTS (kernel 6.12.94+)

Dogfood on linux-cloud: Individual developer

7 repos · 34 commands · 4 broken · 0 slow · 3 filed

As a solo developer I re-checked all six fixed issues on current main (all fixed), then went deeper on interrupts (SIGINT/TERM/HUP/KILL across switch, reset, restore, commit, stash, merge, pick, gc, clone) on a 100k-file repo, a bare-remote push/pull/rebase loop, history rewrite/undo, interop with git-tuned repos, big-asset gc, and parity sweeps on git/git and cargo. Everyday speed and parity are very good (status/diff/log/blame/clone at or faster than git, log/diff/blame filters identical to git on git/git); I filed three bugs: stale index.lock after interrupting commit/stash/merge/pick, no log --follow, and commit --dry-run ignoring mode-only changes.

os: Ubuntu 24.04.5 LTS (kernel 6.12.94+) · x86_64
machine: Cursor cloud VM, 4 vCPU Intel Xeon, 15 GB RAM
versions: grit 0.5.24 · main 1d054b51 · 2.43.0

repositories7

repo	files	commits	.git	notes
octocat/Hello-World	1	3	1 MB
rust-lang/cargo	3,078	23,596	81 MB	also used for fresh clones, shallow, --blobs none, git-tuned interop copy
git/git	4,857	82,616	324 MB
scratch/wide-100k-files	100,000	1	27 MB	500 dirs x 200 files; branches A (20k files edited) and B (20k other files edited)
scratch/deep-30k-commits	50	30,000	9 MB
scratch/churn-100mb-asset	1	13	102 MB
scratch/solo	9	12	2 MB	daily-loop project with bare remote and two peer clones

commandsgrit vs. system git, wall time

command	repo	grit	git	grit/git	result	notes
grit status (clean)	rust-lang/cargo	13 ms	11 ms	1.2×	ok	hyperfine
grit diff / log -n 20	rust-lang/cargo	4.8 ms	2.7 ms	1.8×	ok	log -n 20: 2.3 vs 1.4 ms
grit clone https://github.com/rust-lang/cargo	rust-lang/cargo	3.46 s	5.02 s	0.7×	ok	fsck --strict clean
grit clone --depth 1 / fetch --depth 50 / --unshallow / --blobs none	rust-lang/cargo	—	—	—	ok	depth 50 -> 722 commits, same as git; unshallow -> 23596
grit status (after touching all files)	scratch/wide-100k-files	55 ms	2.07 s	0.0×	ok
grit status (dirty, 2 changes)	scratch/wide-100k-files	39 ms	65 ms	0.6×	ok
grit diff (dirty)	scratch/wide-100k-files	27 ms	23 ms	1.2×	ok
grit add (2 changes)	scratch/wide-100k-files	107 ms	101 ms	1.1×	ok
grit commit -m x (2 changes, auto-stage)	scratch/wide-100k-files	171 ms	66 ms	2.6×	ok	2.6x, known #1884
grit commit -m one -- <1 path>	scratch/wide-100k-files	155 ms	66 ms	2.3×	ok	2.3x, known #1884
grit switch A / switch main (20k-file checkout)	scratch/wide-100k-files	976 ms	1.19 s	0.8×	ok	RSS 1.9x git
grit merge B into A (clean, 20k files)	scratch/wide-100k-files	4.0 ms	8.0 ms	0.5×	ok	merge was already done in this state; measured fast path only
grit stash (100 dirty files)	scratch/wide-100k-files	130 ms	—	—	ok
grit add node_modules (90k untracked files)	scratch/solo	996 ms	597 ms	1.7×	ok	1.7x slower, 2.3x RSS
grit commit -m nm (90k new files)	scratch/solo	1.16 s	—	—	ok
grit blame builtin/commit.c	git/git	621 ms	550 ms	1.1×	ok	line attribution identical to git blame --porcelain on 5 files
grit log -n 50 -- builtin/commit.c	git/git	38 ms	45 ms	0.8×	ok
grit log -- Documentation/git-commit.txt (full)	git/git	480 ms	464 ms	1.0×	ok
grit log filters (13 combos)	git/git	—	—	—	ok	oid lists identical to git
grit diff --patch-text v2.10.0 v2.43.0	git/git	1.29 s	950 ms	1.4×	ok	hunk-header function labels differ in a few hunks
grit diff --patch-text v2.40.0 v2.41.0	git/git	109 ms	—	—	wrong	byte-identical except 7 hunk headers ('out:' vs signature); not filed (cosmetic)
grit log --follow -- builtin/commit.c	git/git	—	—	—	error	option does not exist; filed #2316
grit log (30k linear commits)	scratch/deep-30k-commits	140 ms	108 ms	1.3×	ok
grit log -- f7.txt (30k linear commits)	scratch/deep-30k-commits	257 ms	216 ms	1.2×	ok
grit merge-base HEAD HEAD~20000	scratch/deep-30k-commits	103 ms	58 ms	1.8×	ok
grit diff/show/status on 600 MB binary	scratch/big-asset	3.0 ms	1.0 ms	3.0×	ok
grit gc (100 MB asset, 12 edits)	scratch/churn-100mb-asset	23.07 s	18.09 s	1.3×	ok	pack 102 MB vs 101 MB, fsck clean
grit gc while committing 30 times	rust-lang/cargo	2.09 s	—	—	ok
grit stash / commit / merge / pick killed by SIGINT, SIGTERM, SIGHUP	scratch/wide-100k-files	—	—	—	wrong	index.lock + index~pid.lock left behind; filed #2315
grit switch / reset --hard / restore killed by SIGINT, TERM, HUP	scratch/wide-100k-files	—	—	—	ok	rolled back, 'operation cancelled', no locks
grit commit --dry-run (mode-only change)	scratch/solo	—	—	—	wrong	says nothing to commit; filed #2318
grit push/pull/pull --rebase/rebase --continue/push --delete/--force-with-lease	scratch/solo	—	—	—	ok	bare remote fsck --strict clean; rejected push exit 1 with JSON
grit history reword/drop/fixup/undo	scratch/hist	—	—	—	ok	change-ids preserved
grit bundle create/verify/list, notes add, tag	scratch/solo	—	—	—	ok	git bundle verify and git clone of grit bundle work

worked9

›Cancellation of switch/reset --hard/restore/clone via SIGINT/TERM/HUP is now clean with a clear message
›Stale-lock error names the dead pid; hook failures leave the staged/unstaged split intact
›Change-id header survives history reword/drop/fixup/undo and git fsck --strict accepts it
›Log filters, blame and numstat match git exactly on git/git and cargo; --patch-text patches apply with git apply, including binaries
›Status after touching 100k files is 55 ms (git 2.1 s on first run); big-asset (600 MB) diff/status now 3 ms / 15 MB
›Interop: grit commits keep git's index v4, untracked cache, commit-graph and midx valid; linked worktrees and submodule bumps fine
›Three concurrent pushers to a bare repo end consistent with clean fsck; pull --rebase conflicts now leave resumable state
›Hostile trees (.git/, .GIT/ entries) are rejected at clone
›--json outputs parse for every command tried; JSON errors carry kind and hint

problems12

›Filed #2315: commit/stash/merge/pick interrupted by SIGINT/TERM/HUP leave index.lock (also release 0.5.24)
›Filed #2316: grit log has no --follow; renamed-file history stops at the rename
›Filed #2318: commit --dry-run omits mode-only changes and reports 'nothing to commit'
›Not filed (cosmetic): rename display is 'a => b' with full paths vs git's brace form in --stat/--numstat
›Not filed (cosmetic): --patch-text hunk header function labels can differ ('out:' label vs function signature) because diff drivers like cpp are not applied
›Not filed (minor): diff --patch-text for a modified binary always emits full 'literal N' data (258 KB for 100 KB random file vs git --binary 264 B delta)
›Not filed (minor): human 'grit add' prints non-UTF-8 paths as 'grit+pathhex:...' and newline names raw
›Not filed (minor UX): grit pull 'local changes would be overwritten' error lists no files; rebase --continue with unstaged resolved file says 'unmerged paths remain in the index' without 'grit add' hint; 'grit push nowhere' reports 'src refspec nowhere does not match any ref' (no 'no such remote' hint, JSON error has no kind); 'grit blame nofile' prints a raw os error 2; 'grit cat HEAD:a' on a corrupt blob says unknown revision instead of corrupt object
›Not filed (minor): after kill -9 mid-switch the leftover half-checkout is reported as 10k user edits and the hint says 'commit them' which would commit the partial checkout (same exposure as git)
›Not filed: fsmonitor could not be exercised (git 2.43 fsmonitor--daemon unsupported on this VM)
›Not filed: git 2.43 aborting with 'BUG: wt-status.c:579' after git update-index --split-index is a git-side problem
›Single-file commit on a 100k-file tree is 2.3-2.6x git (already #1884); add of 90k new files 1.7x time and 2.3x RSS vs git (below the 3x bar)

bugs filed3

#2315 #2316 #2318already reported: #2142 #1884 #1859 #1883 #2303 #1014 #2155 #1909 #2154

next time

›Real ssh push/fetch with two concurrent pushers (needs sshd) and thin packs, fsck after each
›Case-insensitive filesystem checks for merge/switch/pick when a branch adds a file differing only by case (needs macOS/Windows)
›Interrupt matrix for rebase --continue, fetch/pull/push mid-pack and gc on a 200k-file repo with core.lockfilePid, after #2315 is fixed (re-run commit/stash/merge/pick under SIGINT/TERM/HUP)
›core.fsmonitor hook usage and timings (needs a git with fsmonitor--daemon or a watchman-style hook)
›log --follow equivalent after #2316; also test log -L across renames and show --follow-style rename display
›10k-commit churn with 100-200 MB binaries and gc.autoPackLimit consolidation; gc --auto never runs after grit commit (#1859)
›Library variant: tiny solo tool over typed APIs (reflog after rebase, custom ObjectStore, commit/stash cancellation per #2142)