phase-merge can read the forge before a recovery push moves the pull request head, and spends a retry on the old red #594

New issue

Closed

opened 2026-09-21 14:50:30 +00:00 by claude · 1 comment

claude commented

2026-09-21 14:50:30 +00:00

Member

Found during IW-491.

After a recovery push, phase-merge may read the forge before the forge has moved the pull request head to the pushed commit. The read then returns the old head and its red verdict, and phase-merge counts that as a failure of the new head and spends a retry.

PROPOSED FIX. Before judging a reading, compare the pull request head with the local pushed sha. When they differ, the reading is not yet about the pushed commit: wait and read again, and do not spend a retry.

Found during IW-491. After a recovery push, `phase-merge` may read the forge before the forge has moved the pull request head to the pushed commit. The read then returns the old head and its red verdict, and `phase-merge` counts that as a failure of the new head and spends a retry. PROPOSED FIX. Before judging a reading, compare the pull request head with the local pushed sha. When they differ, the reading is not yet about the pushed commit: wait and read again, and do not spend a retry.

claude referenced this issue

2026-09-21 14:52:17 +00:00

IW-491: phase-pr pushes a commit after reading CI, so its green can describe a stale head #596

claude referenced this issue

2026-09-21 19:32:19 +00:00

The phase gate reads and writes cards that do not belong to the head under test #601

claude commented

2026-09-21 19:32:25 +00:00

Author

Member

Closed as a duplicate of #601, which takes this defect and three others as one investigation on one branch. This one is phase 3 there.

The four share a cause: the card and the CI reading do not agree on which head they speak about. Fixing them apart would read the same code four times.

Closed as a duplicate of #601, which takes this defect and three others as one investigation on one branch. This one is phase 3 there. The four share a cause: the card and the CI reading do not agree on which head they speak about. Fixing them apart would read the same code four times.

claude closed this issue

2026-09-21 19:32:25 +00:00