Ignore previously failed checks on pull_request edited

[perryao]

We have a workflow that checks the format of a PR title and it runs on

on:
  pull_request:
    types:
      - opened
      - synchronize
      - reopened
      - edited

If I open a PR with a title that fails this check, our summary workflow below fails as expected:

name: Summary
on:
  pull_request:
    types: [opened, synchronize, reopened, edited]
jobs:
  require-all-checks:
    name: Require All Checks
    runs-on: ubuntu-latest
    permissions:
      checks: read
    steps:
      - name: Wait for Status Checks
        uses: poseidon/[email protected]
        with:
          token: ${{ secrets.GITHUB_TOKEN }}
          ignore: Require All Checks
          ignore_pattern: '\[Optional\].*'

However, if I edit the PR title and get the PR title workflow to run again and pass, the require-all-checks workflow still fails.

Is there a way to ignore previous failures on pull request edited triggers?

[tylermichael]

It looks like this API call returns the last runs and all their attempts by default, not just the last runs and their last attempt.

There would need to be code added to filter out just the previous runs and only look at the most recent attempts. IMO, that makes sense for this check to do.

[bjhargrave]

For type edited, the event payload.pull_request.head.sha is unchanged from the previous workflow run on the PR. Since this is the commit SHA used to locate the check runs, the check runs from the previous workflow run are visible. So the failed check-pr-title check run from the prior execution is visible before the new execution completes and updates the check run result to success. You probably need some other synchronization to ensure the check-pr-title job_id runs before the check status action checks it.

[tylermichael]

It fails even after you manually re-run the wait for status check. Only pushing a new commit allows it to pass.

[cwrau]

@devodev you closed your PR, did you find a workaround?

[devodev]

@devodev you closed your PR, did you find a workaround?

No workaround, I simply don't use it anymore and therefore have no need for this. Also the PR has been open for months with no activity and now has a merge conflict, which I don't feel like maintaining.

[pksunkara]

I think we would encounter this issue even if you manually retry a flaky failed job.