guardians_of_the_galaxy/actions

Shared Forgejo Actions composite actions for the ecosystem.

dispatch-and-wait

Dispatches a workflow_dispatch in another repo and blocks until it reports back — via a commit status on the caller's own commit, not by scanning the target repo's job list for "the one we just started". That search is what this replaces: Forgejo's dispatch API doesn't hand back the run it just created (same limitation as GitHub's), so finding it again meant guessing "the newest job named X, dispatched after timestamp Y" — racy, and wrong if the target runner is busy (a real run took over 2 minutes just to get a task record). A commit status has no such ambiguity: it's keyed by repo+sha+context, and this action owns that key from the moment it posts the initial pending state.

- name: Trigger Filao/e2e and wait for it to finish
  uses: https://git.stationmyr.net/guardians_of_the_galaxy/actions/dispatch-and-wait@v1
  with:
    token: ${{ secrets.PACKAGES_TOKEN }}
    target-repo: Filao/e2e
    target-ref: master
    target-inputs: '{"backend_ref": "${{ github.sha }}"}'
    status-context: filao-e2e

The target workflow must declare three extra workflow_dispatch inputs (status_repo, status_sha, status_context) — this action merges them into whatever target-inputs you pass — and report back using report-status (below) once it's done, if: always()-style (two steps, if: success() / if: failure()).

report-status

The callee's half of the contract above. Call it from the dispatched workflow, using the status_repo/status_sha/status_context inputs the dispatcher added:

on:
  workflow_dispatch:
    inputs:
      backend_ref: {required: false, default: '', type: string}
      status_repo: {required: false, default: '', type: string}
      status_sha: {required: false, default: '', type: string}
      status_context: {required: false, default: '', type: string}

jobs:
  e2e:
    steps:
      - run: ./run-the-actual-tests.sh
      - name: Report success
        if: success() && inputs.status_repo != ''
        uses: https://git.stationmyr.net/guardians_of_the_galaxy/actions/report-status@v1
        with:
          token: ${{ secrets.PACKAGES_TOKEN }}
          repo: ${{ inputs.status_repo }}
          sha: ${{ inputs.status_sha }}
          context: ${{ inputs.status_context }}
          state: success
      - name: Report failure
        if: failure() && inputs.status_repo != ''
        uses: https://git.stationmyr.net/guardians_of_the_galaxy/actions/report-status@v1
        with:
          token: ${{ secrets.PACKAGES_TOKEN }}
          repo: ${{ inputs.status_repo }}
          sha: ${{ inputs.status_sha }}
          context: ${{ inputs.status_context }}
          state: failure

The inputs.status_repo != '' guard keeps the workflow usable on its own (plain push/pull_request runs, where those inputs are empty) without failing on a missing repo/sha.

The token needs write:repository access to both repos — secrets. PACKAGES_TOKEN already has it for every repo sharing an org, which is the only case this has been used for so far.

Testing

sh tests/run.sh   # needs python3, curl, jq -- spins up a fake Forgejo API

Releasing

Manual workflow_dispatch with a version (X.Y.Z) tags vX.Y.Z, creates its release, and moves the floating vX major tag callers pin against — see .forgejo/workflows/ci.yml, same mechanism as argus/upload-action.