[michel2323]

Re-enables the ALCF bridge that 55bb6d1 turned off, and brings back alcf-report.yml (removed in bccaac1) together with the GitLab webhook it was waiting for.

Why the pipelines failed "without output"

Over the last few days the GitLab jobs failed in two different ways that had nothing to do with the code under test:

• stuck_pending_no_matching_runners (38 jobs, e.g. build:support on a0adaa1 and tb/defer-free): no Aurora runner was online on 2026-10-19 from about 13:30 to 19:25 UTC. GitLab marks those jobs as failed without a log.
• no_updates_running: PBS never started the batch job ("Not enough free nodes available and terminated"), and the runner gave up after 100 min.

The real test failures (script_failure) were separate.

What changes

• alcf.yml starts the pipeline as before (mirror sync, fork push behind the alcf-ci label, a fresh pipeline on re-run), then sets the commit status ALCF / LTS stack (Aurora) to pending and exits. No runner waits for hours any more. If the commit's pipeline had already finished, it dispatches the report straight away.
• alcf-report.yml is dispatched by a GitLab pipeline webhook and sets the final status:
  • If the only failures are infrastructure ones (stuck_pending_no_matching_runners, no_updates_running, runner_system_failure, …), it starts one fresh pipeline and leaves the status pending.
  • If that fails the same way, the status becomes an error that names the infrastructure failure, not a test failure.
  • Skipped ([skip ci]) pipelines are left alone.
• Concurrency moves to the job, as in ALCF CI: keep labeling a PR from cancelling its LTS run #722, which this supersedes.

Verification so far

• Both scripts were run against real ALCF pipelines, with gh and the trigger call stubbed. Cases covered: a pass (70952), a test failure (70931), an infra failure on a commit that is no longer the branch tip (70764 → error), an infra failure on a branch tip (70792, 70790 → retry), a cancel, a skip, no pipeline yet, and an already finished pipeline (re-dispatched).
• GitLab can reach api.github.com: a test delivery from a temporary webhook got GitHub's 401, and that webhook has been deleted.

workflow_dispatch only works for workflows on the default branch, so the end-to-end path can only be tested after merging:

1. Create a fine-grained token (JuliaGPU/oneAPI.jl only, Actions: read and write).
2. Add the pipeline webhook to GitLab project 238. Its URL, template and header are described at the top of alcf-report.yml.
3. Report a known pipeline by hand: gh workflow run alcf-report.yml -f pipeline_id=70952.

Until the webhook exists, the ALCF status on this PR stays pending.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 84.11%. Comparing base (70d831a) to head (071b88d).
⚠️ Report is 5 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #726      +/-   ##
==========================================
- Coverage   84.66%   84.11%   -0.55%     
==========================================
  Files          52       51       -1     
  Lines        4499     4375     -124     
==========================================
- Hits         3809     3680     -129     
- Misses        690      695       +5     

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.