Follow-up to #258 with a fresh reproduction, filed as a bug.
Repro: 5-PR stack (bottom based on main, each PR based on the branch below). Rebased the stack with gh-stack. Afterwards every PR's REST base.sha matched its base branch tip, but refs/pull/N/merge did not:
- Bottom+1 PR: base-side parent is the pre-rebase tip of its base branch (no longer reachable from any ref).
- Each PR above that: base-side parent is the
refs/pull/N/merge commit of the PR below it, not its base branch tip.
- Bottom PR (based on
main): unaffected.
CI that builds from the merge ref (Jenkins GitHub Branch Source, "merge with target") fetches only branch refs, so every affected PR fails checkout with merge: <sha> - not something we can merge. Per #258, rescans and close/reopen don't recover it; retargeting the base does.
Expected: after a restack, each PR's test-merge ref is recomputed against the current tip of its base branch.
Happy to share SHAs and logs privately.
Follow-up to #258 with a fresh reproduction, filed as a bug.
Repro: 5-PR stack (bottom based on
main, each PR based on the branch below). Rebased the stack with gh-stack. Afterwards every PR's RESTbase.shamatched its base branch tip, butrefs/pull/N/mergedid not:refs/pull/N/mergecommit of the PR below it, not its base branch tip.main): unaffected.CI that builds from the merge ref (Jenkins GitHub Branch Source, "merge with target") fetches only branch refs, so every affected PR fails checkout with
merge: <sha> - not something we can merge. Per #258, rescans and close/reopen don't recover it; retargeting the base does.Expected: after a restack, each PR's test-merge ref is recomputed against the current tip of its base branch.
Happy to share SHAs and logs privately.