[CI][Docs] Pull requests closed by the PR limit workflow cannot be reopened by their authors #54696

Activity

1. github-actions

   added

   Component: Documentation

   Component: Continuous Integration

   on Oct 20, 2026
2. pearu commented on Oct 20, 2026

   

   pearu

   on Oct 20, 2026

   Last edited by pearu

   ContributorAuthor

   More actions

   Some data on the bot workflow so far, covering all 232 pull requests created since it was introduced on 2026-09-14:

   • 35 pull requests were closed by the workflow.
   • 2 were reopened, both on the first day: one by its author while the limit still applied to collaborators, and one by a maintainer. None was reopened by an author without repository access.
   • For 5 of them, the author opened a new pull request for the same issue instead (4 resubmissions).
   • The remaining 28 have not been reopened or resubmitted.

   Using a previously merged pull request in apache/arrow as a sign of an established contributor:

   • 12 of the 35 closures (34%) hit authors with at least one merged pull request before the closure;
   • 23 of the 35 closures (66%) hit authors without a previously merged pull request. This group also includes genuine first-time contributors.

   In summary, the bot catches likely AI slop that is about 10 % of all PRs, while about 70 % of the genuine contributors it hits have not returned so far.
3. pearu commented on Oct 21, 2026

   

   pearu

   on Oct 21, 2026

   ContributorAuthor

   More actions

   A possible improvement, based on the data above: exempt contributors with a few merged pull requests in this repository from the limit. In the period above, all 12 closures that hit an author with a previously merged pull request would have been avoided with a threshold of 1, and 6 of them with a threshold of 3, while all 23 other closures would have been kept. A threshold higher than 1 makes it harder to get exempted with a single trivial pull request.

   The merged count is available with the same search API that the workflow already uses for the open count, and is only needed when a contributor is over the limit. A sketch with a threshold of 3:

   Diff

   diff --git a/.github/workflows/pr_limit/check.js b/.github/workflows/pr_limit/check.js
   --- a/.github/workflows/pr_limit/check.js
   +++ b/.github/workflows/pr_limit/check.js
   @@ -62,6 +62,18 @@ async function countOpenPullRequests(github, context, username) {
      return data.total_count;
    }
    
   +/**
   + * Returns the number of merged pull requests authored by the user in this
   + * repository.
   + */
   +async function countMergedPullRequests(github, context, username) {
   +  const {data} = await github.rest.search.issuesAndPullRequests({
   +    q: `repo:${context.repo.owner}/${context.repo.repo} is:pr is:merged author:${username}`,
   +    per_page: 1
   +  });
   +  return data.total_count;
   +}
   +
    /**
     * Comments on the pull request explaining the limit, then closes it.
     *
   @@ -125,6 +137,13 @@ module.exports = async ({github, context, core}) => {
        return;
      }
    
   +  // Contributors with at least 3 merged pull requests are not limited.
   +  const merged = await countMergedPullRequests(github, context, user.login);
   +  if (merged >= 3) {
   +    core.info(`Skipping: ${user.login} has ${merged} merged pull request(s).`);
   +    return;
   +  }
   +
      core.info(`Closing #${pullRequestNumber}: over the limit.`);
      await commentAndClose(github, context, pullRequestNumber, user.login, limit, count);
    };