review.knotless.dev asked for changes on at://did:plc:bcgltzqazw5tb6k2g3ttenbj/dev.knotless.pull/3n2w2polrosmx (HUD, app/shell into main) twice, each time because it could not read the diff: git diff: fatal: unable to read tree 3e2c0deae1b4a5ed09bbdf6e56ab83b12de3b7e6, and reading AGENTS.md at main failed with Needed a single revision.
The tree is main's (3a99df9). knotless.dev's own mirror (/mnt/data/knotless/cache/mirrors/did_plc_bcgltzqazw5tb6k2g3ttenbj/3n2kdckpvmnlo.git) has it, and so does a public git clone https://knotless.dev/chadtmiller.com/hud. The reviewer's own copy in its container (/tmp/knotless-review/cache/mirrors/did_plc_bcgltzqazw5tb6k2g3ttenbj/3n2kdckpvmnlo.git) did not: git cat-file -t 3e2c0de… there failed with could not get object info, its refs all present.
The repository had been made public that morning (repo publicize), and repo compact ran on it in between. Clearing the reviewer's copy and pushing an empty commit then got: "Carried forward: the change this pull makes at cf38fa9 is the one reviewed at d5c022f, so that review stands." So a review that said it could not read the diff was carried forward as if it had read it.
How to see it
Unsure of the exact trigger. Likely: a private repository mirrored by the reviewer, made public (its refs and packs move out of the space), compacted, then a pull into a branch whose tree only the moved packs hold.
What you expected
A mirror that misses an object refetches it (or rebuilds) rather than failing the review.
A review that could not read the change is not carried forward to a head with the same change.
Where
knotless.dev; review.knotless.dev's container knotless-review-1 on knotless's box, image knotless:current, 2026-11-22.
What happened
review.knotless.dev asked for changes on at://did:plc:bcgltzqazw5tb6k2g3ttenbj/dev.knotless.pull/3n2w2polrosmx (HUD, app/shell into main) twice, each time because it could not read the diff:
git diff: fatal: unable to read tree 3e2c0deae1b4a5ed09bbdf6e56ab83b12de3b7e6, and reading AGENTS.md at main failed withNeeded a single revision.The tree is main's (3a99df9). knotless.dev's own mirror (
/mnt/data/knotless/cache/mirrors/did_plc_bcgltzqazw5tb6k2g3ttenbj/3n2kdckpvmnlo.git) has it, and so does a publicgit clone https://knotless.dev/chadtmiller.com/hud. The reviewer's own copy in its container (/tmp/knotless-review/cache/mirrors/did_plc_bcgltzqazw5tb6k2g3ttenbj/3n2kdckpvmnlo.git) did not:git cat-file -t 3e2c0de…there failed withcould not get object info, its refs all present.The repository had been made public that morning (
repo publicize), andrepo compactran on it in between. Clearing the reviewer's copy and pushing an empty commit then got: "Carried forward: the change this pull makes at cf38fa9 is the one reviewed at d5c022f, so that review stands." So a review that said it could not read the diff was carried forward as if it had read it.How to see it
Unsure of the exact trigger. Likely: a private repository mirrored by the reviewer, made public (its refs and packs move out of the space), compacted, then a pull into a branch whose tree only the moved packs hold.
What you expected
Where
knotless.dev; review.knotless.dev's container
knotless-review-1on knotless's box, image knotless:current, 2026-11-22.