The Git List

story 229 (HEAD -> patches)

Thread: 9 messages from Kristofer Karlsson via GitGitGadget, Junio C Hamano and 3 others

Connectivity check would stop following replace refs under Karlsson patch

Fsck, prune, pack-objects and upload-pack already ignore replacements; check_connected() was the odd one out, and reviewers are now weighing the edge cases.

Kristofer Karlsson's patch makes the rev-list connectivity check ignore replace refs, as other repository-completeness operations do. Patrick Steinhardt says he is aligned with the change after some questions. Karthik Nayak asks for a documentation update, and a test change and a commit split are still being discussed.

→ read the full story → open the thread

story 230 · bugs

Recent graph indentation change garbles log --graph when a root has columns to its right

A report that git log --graph --all draws an orphan branch's root with parents has been traced to the visual-root indentation added in a34c00d96. A patch was posted, but Pablo Sabater says it was malformed and asked for git send-email. He also says the graph should render the same with or without indentation here.

→ full story → thread (9)

story 231 · patches

Rewrite of git-diff EXAMPLES draws objections over --staged and table layout

Julia Evans proposed a rewritten EXAMPLES section for git-diff based on what users say they use. Junio Hamano objected to changing one example to --staged without matching changes elsewhere, and to wrapped sentences in tables. Evans offered to revert that and suggested horizontal definition lists.

→ full story → thread (7)

story 232 · discussion

Pre-commit hook edits are lost with git commit <pathspec>, prompting a docs question

Kevin Lyles found that a pre-commit hook that modifies staged content does not take effect with git commit <pathspec>. Chris Torek explained the index handling involved. Junio Hamano said hooks are meant for validation and floated documenting that and possibly detecting mutation.

→ full story → thread (5)

story 233 · bugs

Bare --force-with-lease is ignored when pushing to a remote group; fix posted

Patrick Peroša reported that bare --force-with-lease has no effect when pushing to a remote group. Tyler Cipriani posted a patch for push_multiple() that fixes this and a second case in which --force-with-lease=<ref>: could overwrite a remote branch.

→ full story → thread (2)

story 234 · patches

Outreachy notes-utils series removes the_repository from the file, with caveats from Junio

Okhuomon Ajayi's series makes notes-utils.c use the repository it is passed and, in v2, drops USE_THE_REPOSITORY_VARIABLE from the file. Patrick Steinhardt and Karthik Nayak found it good, while Junio Hamano cautioned that the notes code has a long way to go before it is free of the_repository.

→ full story → thread (10)

story 235 · discussion

Proposal for `git branch -s` to create and switch meets resistance

Johannes Welsch proposed git branch -s <name> to create a branch and switch to it. Junio Hamano asked why git checkout -b is not enough, and Pushkar Singh said it would overlap with git switch -c and blur the split that motivated git switch.

→ full story → thread (3)

story 236 · project

Cirrus CI config likely to be removed as the service winds down

Tuomas Ahola asked whether anyone still uses .cirrus.yml now that Cirrus CI appears to have shut down. Carlo Arenas sees no reason to keep it. Patrick Steinhardt agrees it should go but wants to give Arenas a few days before a removal patch.

→ full story → thread (8)

$ git log --since=1.week.ago stories/

tuesday, november 3

Shallow-clone hint patch turns into a debate on review, AI and test coverage Junio Hamano says he now leans heavily on human reviews after trusting AI-generated ones too much, and others discuss what review and tooling the project can realistically rely on.
Junio weighs how to handle topics that sit unreviewed in 'seen' After a period of queuing many topics, the maintainer says 'seen' is crowded with unreviewed work and is now warning contributors that stale topics will be dropped.
Reference-transaction hook reports zero object IDs on branch and tag deletion The behaviour dates to Git 2.31 and remains in 2.55, but after a month with little interest Junio says the topic may be dropped unless someone steps up.
range-diff --matched-only draws suggestions to pick better inputs instead Reviewers say the noise from one-sided commits mostly comes from how range-diff is invoked, while Harald Nordgren wants an easy way to hide commits added by a rebase.
Jasper Mayone's worktree work: --merged filter RFC and a series on in-progress operations Related patches would make `git worktree remove` refuse worktrees mid-rebase or mid-bisect and make %(worktreepath) report the worktree rebasing or bisecting a branch.
Patch proposes documenting how the Git project is governed Antonin Delpeuch's new technical document describes current practice without setting it as policy; Junio Hamano's first comments concern wording and omissions.
Junio proposes freezing the repo_config_values structure The maintainer wants to stop moving more settings into the per-repository structure until it can work on repositories other than the_repository.
Patrick Steinhardt moves loose-object logic into its own backend file The refactoring is meant to show which parts of Git still depend on loose-backend internals, ahead of similar series for the packed and files backends.

monday, november 2

Three patches aim to fix CI jobs broken by Ubuntu 26.04 runner image Four CI jobs have failed since ubuntu-latest moved to Ubuntu 26.04, and a new series proposes workarounds rather than waiting for a fixed runner image.
Two simultaneous credential-cache daemons can clobber each other's socket When two commands store a password at the same moment, one daemon takes over the other's socket and the first credential is lost.
Elia Pinto's size_t series finds 32-bit sizes on every platform Three patches widen types in hashwrite(), count_lines() and hash_chars(), found with -Wshorten-64-to-32 on a 64-bit Linux build.
imap-send fix for reordered headers reaches its fifth revision git imap-send could not take patches fetched with b4 because it expected Date before Subject, while git send-email writes them the other way round.
Series would align 'pseudoref' terminology with its 2024 redefinition Docs, comments, messages and tests still use 'pseudoref' in its old sense; the series adds a 'root ref' glossary entry and replaces outdated uses.
Junio declines to move Rust crate into rust/ subdirectory Mike Hommey's proposal to relocate Cargo.toml and the gitcore crate was rejected because the maintainer is trying to limit topics in flight.
Junio plans to drop 'git repo info' path keys series after a month without review The seven-patch GSoC series adding path.* keys has drawn no review, and 'seen' is crowded with topics that are collecting dust.

sunday, november 1

Shallow-boundary check for revert moves to command setup in v2 After review, the patch that stops "git revert" from emptying a shallow clone now covers cherry-pick and rebase and refuses before any commit is made.
Torvalds patches git describe distance count for skewed commit dates The Linux merge window exposed a 'git describe' distance that disagrees with the kernel's own version calculation.
HTTP retry series adds backoff, configurable status codes and POST retries The four patches extend the HTTP 429 retry support added in March, which currently covers only GET requests and 429 responses.
More size_t conversions for parsing and checking objects, with a note on off_t A new series from Elia Pinto continues the effort to avoid truncating object sizes where "long" is 32 bits, and Simon Richter added a caveat about on-disk sizes.
Valueless merge.renames and diff.renames settings to be accepted by merge-ort "git -c merge.renames merge" currently fails even though the documentation says a setting with no value means true.
Review of Couder's early option scanning API turns up a documentation mismatch The series adds an early-scan sub-API to parse-options so commands stop mis-handling option values and "--" in hand-rolled scans.
Includes in .git/config are used in practice, brian carlson says A discussion of sharp edges in Git configuration files raises whether hand-edited include directives in a repository's config should be restricted.

saturday, october 31

Patrick Steinhardt fixes git grep failing on nested submodule trees A 2.56 regression made `git grep --recurse-submodules` against a tree die on directories inside submodules; a three-patch fix is out and reviewed.
Elia Pinto posts six patches on loose-object size checks and size truncations The series tightens size handling that could read or write out of bounds, including a loose-object reader that returns uninitialized heap memory for truncated objects.
Git test suite patched for coreutils 9.12's quoted `env` output GNU `env` now shell-quotes values, which breaks grep patterns in t5702 that check whether the v2 protocol is requested.
Hamano series makes four unnecessarily public symbols static again Leftovers from the removal of the recursive merge strategy and other refactors are being hidden; reviewers like the cleanup.
Patch proposes a document defining Git's security boundary With a rising volume of security reports, the patch would give reporters guidance on what Git treats as a vulnerability.
Size_t widening series reaches v2 after Steinhardt review, with AI-style prose questioned The patches stop sizes above 4 GiB being truncated on 64-bit Windows, but a reviewer asked for human-written explanations.
Gitignore docs to say explicitly that `*` can match nothing A user's question about `foo*.log` led Junio C Hamano to draft a rewrite of the wildcard description that no longer depends on fnmatch(3).

friday, october 30

Fix for transaction commit failing when a compatibility hash is configured With batch fsync and a compatibility hash both enabled, migrating the temporary object directory cannot handle an existing loose-object-idx file.
Fetch fix would commit refs before tag backfill to stop oversized shallow transfers Since batched ref updates arrived in 2025, backfilling tags in a shallow repository can make the server resend objects the client already has.
Quarantine's own refs get listed as alternate refs during receive-pack connectivity check A patch proposes a fix, while reviewers ask whether the real fix is for child processes to recognize which alternate is the repository's own object directory.
git grep --recurse-submodules on a tree dies on directories inside a submodule A reporter says the regression appeared in Git 2.56.0 and did not occur in 2.55.0.
MIDX write fix skips empty incremental layers that broke later bitmap writes A zero-object layer published without a reverse index made a subsequent incremental write with --bitmap fail.
Proposal to flag concerning repository structure in 'git repo structure' A three-patch series adds a degree-of-concern indicator and config-based thresholds; Junio Hamano spotted a copy-and-paste error.
Patch would let @{push} resolve under push.default=simple with a separate push remote The change makes `git status` hints match where `git push` actually sends a branch in triangular setups.
What's cooking, Oct 2026 #10: several topics marked for 'next' Junio Hamano's latest topic report drew review comments from Patrick Steinhardt, Harald Nordgren and Karthik Nayak.

git/list

$ git log --since=1.week.ago stories/