A user reports that since Git 2.54.0, a cherry-pick in a blob:none partial clone can try to lazy-fetch a blob it has just written to the local object store. The server never had the blob, so the fetch fails with "not our ref" and then "could not fetch ... from promisor remote".

The commit itself is written. The failure happens while the commit summary is printed. The object in question, 201d4490, is the result of a content merge of dev-tools/go.mod and is stored as a loose object in the clone. The subject line says the problem was bisected to 5946a564cd.

Patrick Steinhardt replied that this is a known issue and that a fix is in flight (message ID 20261104-pks-odb-loose-stale-cache-v1-0-a00d5cdb5e7d@pks.im). He asked the reporter to test that change, and later thanked the reporter for confirming.

The fix is Steinhardt's two-patch series "odb/source-loose: fix failing existence checks caused by stale cache". It describes the cause: when git-merge writes merged blobs it may be unable to read them back because the loose object cache is not updated. Git then thinks the blobs do not exist locally, and in a partial clone it performs a backfill fetch the remote cannot satisfy.

In a side discussion on that series, Steinhardt said the cache cannot simply be dropped as is. He wondered whether the existing approach is correct in all cases with concurrent maintenance and promisor remotes, and said he does not plan to look into it now.