Harald Nordgren's patch adds a hint when <ref>~N or <ref>^N fails in a shallow clone because the walk ran out of parents at a recorded shallow boundary. The hint explains that history was truncated there and suggests a command, including an exact --deepen for <ref>~N. Where the ref looks like <remote>/<branch> and the remote is configured, it names that remote and branch.
Junio Hamano pinged the thread again and told Nordgren that if the change were acceptable it would already be in 'next' and 'master'. He said one criterion for acceptability is whether others find it a reasonable change, and he knows of no way to show that other than positive reviews.
The discussion then turned to AI-generated reviews. Junio said that a few release cycles ago he trusted them too much and let topics without human review into 'next' and 'master'. That produced quite a few immediate reverts. He now bases his decision to mark a topic for 'next' heavily on human reviews. He also questioned whether a contributor who submits a lot of low-quality AI output should add AI-generated reviews to other people's threads.
Nordgren said review should not depend on a single individual and pointed to tools such as Codecov that hook into GitHub pull requests. He also proposed a soft rule against nitpicking until the idea itself is settled, since nitpicks waste reviewer time and produce new versions that get nitpicked again.
Phillip Wood replied that CI shows tests pass but not whether test coverage is sensible. He said Derrick Stolee used to run a coverage report but stopped because it was time-consuming. Even coverage numbers would not reveal many pointless tests that slow the suite, which still takes a human reader. Patrick Steinhardt agreed that automated checks would help but said they would need to be self-executable scripts, since the project lacks computing resources to run them. He noted that clang-format discussions never went anywhere, and said a contact has shown interest in adapting a tool to Git. He also said not all contributions are code and that documentation work is welcome.
Nothing was decided. The original patch still awaits positive review.