The Git List

story 31 (HEAD -> project)

Thread: 8 messages from Scott Chacon, brian m. carlson and 3 others

Chacon proposes replacing Git's AI policy with Linux-style disclosure; carlson and Hamano push back

The RFC would let contributors submit AI-assisted work if they understand it, test it and disclose substantial help, but the DCO and copyright questions are unresolved.

Scott Chacon sent an RFC to SubmittingPatches that would allow AI-assisted contributions with human understanding and disclosure, instead of rejecting anything that looks AI-generated. brian m. carlson opposes it on DCO grounds, and Junio C Hamano is skeptical.

→ read the full story → open the thread

story 32 · patches

Two competing series would speed up SHA-1 collision detection by roughly 2.7x

Scott Chacon posted a four-patch series, mostly AI-generated, that ports a faster sha1dc approach to C, hashing about 2.7x faster on a Xeon. Johannes Schindelin has a competing series and review comments. Junio C Hamano asked how license provenance can be assured.

→ full story → thread (15)

story 33 · patches

http.sslVerifyStatus topic pulled from 'seen' while Hamano chases six-hour CI timeouts

Junio C Hamano ejected the http.sslVerifyStatus topic from 'seen' after bisecting CI jobs that spin for six hours. A run of the topic alone on Git 2.56 timed out many linux-* jobs, so the cause appears to lie in the topic itself.

→ full story → thread (40)

story 34 · bugs

Series would make branch rename and copy visible to the reference-transaction hook

Maciej Ciemborowicz's series moves ref copy and rename onto the transaction API so hooks see both endpoints. Junio C Hamano says it breaks t0600 and t5510 on current master, and has not had time to review patches that run past 1000 lines.

→ full story → thread (26)

story 35 · patches

Shallow-clone fetch series reaches v7, now built around remote.<name>.refmap

Harald Nordgren's series, now at v7, aims to stop a new remote in a shallow repo from fetching every branch. It adds remote.<name>.refmap, a refspec-less fetch that infers branches from upstreams, and a --limited-fetch option for git remote add.

→ full story → thread (52)

story 36 · bugs

repack --drop-filtered --dry-run still writes packs and, with -d, deletes old ones

Coy Geek reported that git repack --drop-filtered --dry-run prints candidates but also writes a new promisor pack, and with -d removes old packs. Siddharth Shrimali identified a fall-through in cmd_repack() and proposed a two-line fix.

→ full story → thread (3)

story 37 · discussion

Proposal to rewrite 'man git' navigation advice draws Hamano's questions on dropped references

Julia Evans's patch changes git(1) to explain -h versus --help and drops references to gittutorial, giteveryday and other pages. Junio C Hamano agrees with removing 'everyday' but wants the others justified in the next version.

→ full story → thread (17)

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

wednesday, october 7

Contributor's Summit notes cover Git 3.0, security reports, AI policy and funding The notes put a rough spring target on Git 3.0 and describe a backlog of outside security reports that maintainers say needs a better process.
Stale delta base cache entries can corrupt data after a pack is closed Patrick Steinhardt's fix addresses a bug reproducible with recursive submodule merges, where a new pack at a recycled address picks up another pack's cached data.
Patrick Steinhardt moves alternates into the files object-database backend The 13-patch series abandons the idea of mixing different backends within one object database, and reviewers are probing that design choice.
SHA-256 thread turns to the state of interoperability work and migration tooling Maintainers are asking what is still missing in brian m. carlson's interop branch and how ready Git is for users to migrate.
packed-refs rewrite copies unchanged refs verbatim, cutting deletion time about 20% Karthik Nayak's patch avoids reformatting unchanged refs when updating packed-refs; review has focused on comment wording and commit message detail.
Hash-transition document links to hijacked shattered.io domain A reader reported that the SHAttered link in the technical docs now leads to unrelated content, and a fix pointing to a Web Archive snapshot has been posted.
Julia Evans posts draft of a new beginner Git tutorial The work-in-progress two-part tutorial would replace gittutorial, with feedback from 22 beta testers behind it.

tuesday, october 6

Chacon's signed SHA-256 tree digest reopens the question of SHA-256 as the Git 3.0 default An RFC that would add an independent tree hash to signed commits and tags has turned into a debate over whether the 3.0 default should move to SHA-256 at all.
Phillip Wood's series makes `git checkout -m` restore the original conflict labels After `stash pop` or `checkout -m <branch>`, re-creating a conflict currently falls back to generic base/ours/theirs labels; the series records the real ones.
New merge-conflict manual page prompts questions about whether the code should change too Julia Evans's gitmergeconflicts series tries to fix confusing guidance on merge conflicts, and reviewers are asking how far to go beyond documentation.
Opt-in incremental connectivity check gets a design review from Patrick Steinhardt The series aims to make the post-fetch connectivity check scale with what changed rather than with everything already reachable.
Report of ZIP timestamp and strict fast-import date bugs prompts a clamp-or-reject question Out-of-range dates in `git archive --format=zip` produce misleading or wrapped timestamps, and strict raw `fast-import` accepts a date that `fsck --strict` rejects.
Proposal for porcelain to share stashes through remotes meets resistance Reviewers say branches already cover the use case, and the proposer has dropped a `stash publish` command.
Patch to validate `fetch.followRemoteHEAD` lazily is down to minor fixes The change stops `git fetch` from warning about a bad config value that the particular fetch never consults.
Git for Windows 2.56.0(2) fixes GIT_CONFIG_GLOBAL=NUL regression The follow-up release addresses problems introduced by the 2.56.0 move from MINGW64 to UCRT64.

git/list