The series comes from qeesung and Qin ShiCheng, who say they hit the problem in production: a ref pointing at a commit that no longer exists, on Git 2.43. They say it also reproduces on master.

The cover letter describes the sequence. Repack scans for .keep files and decides which packs are redundant, then starts pack-objects, which scans again when it opens the packs. In between, a push of content identical to an earlier push can finish migrating its quarantine. Its duplicate pack is dropped, but its .keep file is linked into place next to the old pack.

Pack-objects then sees the .keep and treats the old pack as kept, so its objects are not copied into the new pack. They are also not loosened for -A, or kept for -k or --cruft. Repack still deletes the pack from its earlier list, because repack_remove_redundant_pack() passes force_delete, which skips the .keep check.

The fix in the latest cover letter is that repack now tells pack-objects which packs it may delete, and pack-objects ignores .keep files on those packs. A separate patch covers a related problem with cruft packs written with an expiration. There a cache of kept-pack state, built on first use, goes stale after a pack is unmarked, so the later walk gets wrong answers.

Justin Tobler and Junio C Hamano took part in the thread, but the excerpts do not show their positions. Justin's own series in a nearby thread addresses a different .keep problem. The review outcome for this series is not visible in the excerpts.