Wednesday, October 28, 2026 · Windows 11 (10.0.26200)

Dogfood on win-x64: Team developer

6 repos · 33 commands · 5 broken · 1 slow · 5 filed

Played a team developer on win-x64: mixed git/grit clones of a shared bare remote and of grit serve over HTTP, rebase/merge/pick with conflicts, force-with-lease, prune, tags, worktree review, Windows-specific filesystem cases (autocrlf, case collisions, locked/read-only files, unicode), plus cargo and git/git scale checks. All 6 fixed issues re-verified as fixed. Core collaboration flows interoperate cleanly with git (fsck clean, identical trees to git rebase -r --update-refs); I filed 5 new problems (stuck rebase abort, D/F ref errors, history drop attribution, rebase marker labels, slow rebase apply).

os: Windows 11 (10.0.26200) · x86_64
machine: win-x64 self-hosted worker; Intel i5-12400, 16 GB RAM, NTFS
versions: grit 0.5.23 (built from source) · main eb2680ab · 2.55.0.windows.5 (system config: core.autocrlf=true, core.symlinks=false, lfs filter required)

repositories6

repo	files	commits	.git	notes
git/git	4,857	82,605	323 MB	cloned --no-checkout, then local clones via git --no-local and grit
rust-lang/cargo	3,078	23,588	81 MB
octocat/Hello-World	1	1	0.1 MB	depth 1, setup only
scratch/team (alice/bob/carol clones, remote.git)	6	12	0.1 MB	mixed git/grit clones, CRLF, rebase/force-with-lease
scratch/cc2 (6 concurrent pushers)	3	100	0.2 MB
scratch/misc (case collisions, unicode, md, ro, ab, hd, df*, cf)	8	6	0.1 MB

commandsgrit vs. system git, wall time

command	repo	grit	git	grit/git	result	notes
grit clone (local path, full)	rust-lang/cargo	1.84 s	1.08 s	1.7×	ok	git used hardlinked local clone
grit clone (local path vs git --no-local)	git/git	6.96 s	29.80 s	0.2×	ok	grit 4x faster
grit clone over HTTP from grit serve	rust-lang/cargo	4.50 s	5.96 s	0.8×	ok	git peak not measurable via wrapper; server peak WS 336 MB
grit push master over HTTP to grit serve	rust-lang/cargo	10.11 s	—	—	ok
grit clone --depth 5	rust-lang/cargo	1.55 s	2.33 s	0.7×	ok
grit clone --blobs none	rust-lang/cargo	11.27 s	19.80 s	0.6×	ok	same object counts as git --filter=blob:none (server ignores filter)
grit fetch (40 new commits, new branch)	rust-lang/cargo	300 ms	330 ms	0.9×	ok
grit fetch (no-op)	rust-lang/cargo	20 ms	305 ms	0.1×	ok
grit status (clean)	rust-lang/cargo	179 ms	155 ms	1.2×	ok
grit status (clean)	git/git	97 ms	36 ms	2.7×	ok	2.7x, under the 3x bar; see #1702
grit revs HEAD~20..HEAD	git/git	17 ms	33 ms	0.5×	ok
grit revs --count A...B	git/git	13 ms	44 ms	0.3×	ok
grit log --oneline -- builtin/commit.c	git/git	574 ms	596 ms	1.0×	ok
grit blame builtin/commit.c	git/git	667 ms	591 ms	1.1×	ok
grit blame Makefile	git/git	1.36 s	1.07 s	1.3×	ok	attribution of builtin/rebase.c identical to git (1938/1938 lines)
grit diff v2.40.0 v2.45.0 --stat	git/git	329 ms	379 ms	0.9×	ok
grit revs --count HEAD	git/git	734 ms	606 ms	1.2×	ok
grit switch --detach v2.30.0 / back to master	git/git	1.44 s	1.51 s	1.0×	ok	back: 1596 vs 1650
grit reset --hard across 1451 changed files	git/git	510 ms	790 ms	0.6×	ok
grit rebase master (60 commits, master +600)	git/git	9.04 s	3.17 s	2.8×	slow	#2211; 1 commit: 6547 vs 1106 ms
grit rebase master --dry-run / --check	git/git	885 ms	—	—	ok	planning is fast; apply phase is slow
grit rebase main-moving (9 commits, merge + dup, update-refs)	rust-lang/cargo	501 ms	—	—	ok	tree identical to git rebase -r --update-refs
grit merge --no-ff (121-file conflict) / -X theirs	rust-lang/cargo	457 ms	—	—	ok
grit pick -x with conflict + --continue	rust-lang/cargo	—	—	—	ok
grit fetch (D/F ref conflict, no prune)	scratch/df3	—	—	—	error	#2198 raw 'corrupt ref store: I/O error: Access is denied'
grit branch rel (rel/1.0 exists)	scratch/dfy	—	—	—	error	#2198
grit rebase --abort after exec_dirty + moved ref	scratch/ab2	—	—	—	error	#2197 repo stuck
grit history drop topic~1 (sibling branch conflict)	scratch/hd	—	—	—	wrong	#2204 unattributed conflict
grit rebase (conflict markers)	scratch/ab	—	—	—	wrong	#2207 label is file path
grit push --atomic over HTTP with update hook	scratch/srv	—	—	—	ok
grit push --force-with-lease (stale, then after fetch)	scratch/team	—	—	—	ok
grit worktree add/remove/list with spaces in path	scratch/cc2	—	—	—	ok
grit switch with login.txt locked (FileShare.None)	scratch/team	—	—	—	ok	clear error, nothing partially applied

worked10

›Mixed git/grit collaboration on a bare remote: refs, history and fsck always clean; system autocrlf=true honored and index stays LF
›Stacked rebase with merge commit, cherry-picked duplicate and update-refs gives the same tree as git rebase -r --update-refs on cargo
›merge --check, --no-ff, -X theirs, delete/modify, rename/rename and binary conflicts match git merge-tree's index; git merge --abort works on a grit merge
›Case-colliding paths: warning like git, grit commit does not record phantom deletions
›Exec bit (100755) and gitattributes eol preserved when committing from Windows
›Locked and read-only files during switch/reset/stash handled with clear errors and no partial state
›Atomic push over grit serve HTTP with update hook, concurrent pushers: no partial publication, remote fsck clean
›pull.rebase / branch.<name>.rebase honored; commit-msg/pre-commit/pre-push hooks run via sh on Windows; --json hook failure is structured
›Shallow clone, blobless clone parity with git; fetch no-op 20 ms vs git 305 ms; clone git/git 4x faster than git --no-local
›Unicode file and branch names, spaces in paths, worktree review of a teammate's branch

problems11

›#2197 rebase stopped by exec_dirty can't be aborted/continued/skipped once the branch ref moves (filed)
›#2198 ref dir/file conflicts surface as 'I/O error: Access is denied' / 'corrupt ref store'; fetch aborts without hint (filed)
›#2204 history drop/reword conflicts from a sibling branch are unattributed; default --update-refs branches silently rewrites other local branches (filed)
›#2207 rebase conflict markers use the file path as the label (filed)
›#2211 rebase apply phase 3-6x slower than git on git/git (filed)
›Not filed (design/UX): rebase --exec failure exit code rolls back the whole rebase instead of stopping for a fix; exec_dirty only offers stash+continue so a formatter's edits cannot be amended in after the stop ('HEAD changed')
›Not filed (too minor): fetch rejection text for a moved tag says 'tag already exists locally and did not move' (the remote tag moved; git says would clobber existing tag)
›Not filed (duplicate of #2151): grit remote add prints the URL with the password in cleartext
›Not filed (feature gap): no range-diff style pairing/interdiff of rewritten commits; revs --cherry-mark only pairs identical patches
›Not filed: grit clone writes a stale .git/config.lock if the pipe is closed early (killed mid-clone) - expected for a killed process
›Not filed (known #1702 class): grit status 97 ms vs git 36 ms on git/git (2.7x)

bugs filed5

#2197 #2198 #2204 #2207 #2211already reported: #2151 #1702 #1014 #1909

next time

›Profile and fix #2211 apply phase, then retest rebase --update-refs stacks with 3+ branches on git/git and cargo
›Real SSH server (Windows OpenSSH sshd or WSL) with 50k+ refs, concurrent pushers and annotated tag moves
›grit pull --rebase conflict and recovery across tools (#1909) plus abort recovery matrix: move HEAD/branch/index during every stopped state (merge, pick, revert, rebase exec/non-exec)
›Windows long paths (>260), reserved names and symlink-heavy repos in a team flow (#2173, #1812, #1813) with the persona's mixed-OS teammates
›Library persona: program that reproduces what grit rebase does (revwalk range + patch-id + replay) to see which parts need grit-cli
›Large binary/LFS-required repos (system git config has filter.lfs.required=true) through grit commit/switch/rebase