ci-gate has no way to take an expected head, so kanon awaitCi cannot stop a stale-head judgement #690

New issue

Closed

opened 2026-10-03 01:17:31 +00:00 by claude · 1 comment

claude commented

2026-10-03 01:17:31 +00:00

Member

Found during IW-607 (follow-up).

core/commands/CiGateCommand.scala:217 sets expectedHead = None. kanon awaitCi (~/ops/kanon/iw-plugin/lib/batch/CiGate.scala:70-113) runs ./iw ci-gate --input <card> right after the ci-fix session pushes, so it can judge the head the forge held before that push and use up a ci-fix attempt on a stale red.

IW-607 fixed the phase-pr half only. The kanon fix needs a ci-gate option that passes the expected head into CiGateTarget.expectedHead. IW-607 did not add it (YAGNI; the kanon change owns the option's shape). See project-management/issues/IW-607/analysis.md, "Out of scope / Interface observation".

Found during IW-607 (follow-up). `core/commands/CiGateCommand.scala:217` sets `expectedHead = None`. kanon `awaitCi` (`~/ops/kanon/iw-plugin/lib/batch/CiGate.scala:70-113`) runs `./iw ci-gate --input <card>` right after the ci-fix session pushes, so it can judge the head the forge held before that push and use up a ci-fix attempt on a stale red. IW-607 fixed the `phase-pr` half only. The kanon fix needs a `ci-gate` option that passes the expected head into `CiGateTarget.expectedHead`. IW-607 did not add it (YAGNI; the kanon change owns the option's shape). See project-management/issues/IW-607/analysis.md, "Out of scope / Interface observation".