mirror of
https://github.com/ankitects/anki.git
synced 2026-09-10 06:20:47 -04:00
4aa9c08369
## Linked issue Fixes #5418 ## Summary / motivation Every push to `main` creates a new `coverage-baseline-linux-<sha>` cache entry. Without a pruning step, these entries accumulate until GitHub's LRU eviction removes old ones. When the baseline a PR needs has been evicted, `check-coverage-regression.py` silently skips the check instead of failing, which could mask genuine coverage regressions. Two changes together close the gap: 1. **`prune-coverage-cache.yml`**: a new workflow triggered by `workflow_run` on every successful CI run on `main`. It keeps exactly one `coverage-baseline-linux-*` cache (the one just created) and deletes all older entries via the GitHub Actions API. With a single, always-fresh entry, eviction becomes practically impossible. 2. **`check-coverage-regression.py`**: turns the silent "no baseline — skipping" path into an explicit failure (`exit 2`). If somehow a baseline is still absent, the check now surfaces the problem visibly instead of passing silently. ## Steps to reproduce 1. Allow several pushes to `main` to accumulate `coverage-baseline-linux-*` cache entries until GitHub evicts the most recent one. 2. Open a PR: `check-coverage-regression.py` prints `no baseline — skipping` and exits 0, even if coverage dropped. ## How to test ### Details - After merging, verify via **Actions → Prune coverage baseline caches** that the job runs after a `CI` completion on `main` and leaves exactly one `coverage-baseline-linux-*` entry (`gh cache list -R ankitects/anki --key coverage-baseline-linux-`). - Confirm that a PR with a missing baseline now receives exit code 2 (visible failure) instead of exit code 0. ## Before / after behavior **Before:** evicted baseline → `check-coverage-regression.py` skips silently, coverage regressions go undetected. **After:** only one baseline exists at a time (no eviction risk); if it is somehow absent, the script exits with code 2 and the CI step fails visibly.