Avoid dismissing approvals on restacks alone

[lunelson]

Summary

• Avoid dismissing approvals solely because restacking/rebasing rewrote commit IDs
• Defer rewritten-history cases to the existing range-diff stale-check logic
• Add self-test coverage for safe restacks and changed patch series

Tests

• npm ci
• npx renovate-config-validator
• .github/actions/dismiss-stale-approvals/self-test.sh

[cursor]

PR Summary

Medium Risk
Changes the stale-approval decision logic in the GitHub Action, which can affect whether PR approvals are dismissed; mistakes could leave stale approvals in place or dismiss valid ones. Scope is small and covered by expanded self-tests.

Overview
Stops treating “approval commit is not an ancestor of HEAD” as automatically stale in check-manual-merge-resolutions.sh, so simple rebases/restacks no longer dismiss approvals and the decision is deferred to the existing git range-diff check.

Updates self-test.sh to reflect the new behavior and adds coverage to ensure rewritten history only becomes stale when range-diff detects a changed patch series (and that the final decision output points reviewers to the range-diff output).

^{Reviewed by Cursor Bugbot for commit befef1c. Bugbot is set up for automated code reviews on this repo. Configure here.}

[augmentcode]

🤖 Augment PR Summary

Summary: This PR refines the “dismiss stale approvals” composite action to avoid invalidating reviews due to restacks/rebases that rewrite commit SHAs.

Changes:

• Updates check-manual-merge-resolutions.sh so “approval commit is not an ancestor of HEAD” no longer marks approvals stale by itself.
• Defers rewritten-history situations to the existing git range-diff-based staleness check.
• Expands self-test.sh coverage to validate that safe rewrites stay non-stale, while changed patch series are still detected as stale via range-diff.

Technical Notes: The action now treats “non-ancestor” as informational (common after rebases) and relies on range-diff output to decide whether the reviewed patch series materially changed.

_{🤖 Was this summary useful? React with 👍 or 👎}

[augmentcode]

Review completed. No suggestions at this time.

Comment augment review to trigger a new review at any time.

[lunelson]

augment review

[augmentcode]

Review completed. 1 suggestion posted.

Comment augment review to trigger a new review at any time.

[augmentcode]

git merge-base --is-ancestor can exit with an error status (e.g., missing history in a shallow fetch), and this block currently treats any nonzero exit as a harmless “non-ancestor” case. Consider distinguishing a real command failure from a true non-ancestor result so the action doesn’t silently skip the manual-merge-resolution check when it can’t be evaluated.

Severity: medium

_{🤖 Was this useful? React with 👍 or 👎, or 🚀 if it prevented an incident/outage.}