Repository navigation
fix(release): refuse a release without notes - #343
Conversation
0.42.0 shipped with an empty GitHub Release. Its CHANGELOG section was empty because nothing had landed since the v0.41.0 tag: #340 and #341 merged while the 0.41.0 release PR waited for review, so they were inside that tag but outside its notes, and git-cliff's next run (starting at the tag) never saw them either. scripts/check-release-notes.py checks a version's section. `prepare` runs it after cargo-release and fails on a section with no bullets. `finish` runs it with --since <previous tag> before tagging and fails when a pull request merged since then is missing from the section, with a message to cut the release again from current main.
![claude[bot]](https://avatars.githubusercontent.com/in/1236702?s=60&v=4){kind=small}
| missing = [ | ||
| s for s in subjects | ||
| if not RELEASE_COMMIT.match(s) | ||
| and (m := PR_NUMBER.search(s)) and m.group(1) not in noted | ||
| ] |
There was a problem hiding this comment.
Exempt every subject that git-cliff drops from the --since check, not only release commits. The current check demands subjects that git-cliff never writes, so finish can fail permanently.
cliff.toml drops these commits:
- non-conventional subjects (
filter_unconventional = true, line 42) docs(changelog)(line 68)chore: bump version(line 75)Merge pull request(line 76)
Failure scenario: this PR squash-merges as Release scripts refuse a release without notes (#343). That subject is not conventional, so git-cliff leaves #343 out of the 0.43.0 section. finish then reports #343 as missing and refuses to tag. The error message says to cut the release again from main. A new cut runs the same git-cliff filters, so it omits #343 again, and the release never tags.
Fix options:
- Generate the expected set with git-cliff itself, for example
git cliff <since>..HEAD --context, and compare the PR numbers it emits. - Or mirror the filters: require only subjects that match
^\w+(\([^)]*\))?!?:and do not match theskip = trueparsers.
Also retitle this PR to a conventional subject, for example ci: release scripts refuse a release without notes. Without that change, #343 is dropped from the next changelog even after the fix.
| print(__doc__, file=sys.stderr) | ||
| sys.exit(2) | ||
| version = args[0] | ||
| since = args[args.index("--since") + 1] if "--since" in args else None |
There was a problem hiding this comment.
nit: Use argparse here, or check that a value follows --since (not blocking). A trailing --since with no value raises IndexError and prints a traceback instead of the usage text. release.sh always passes a value, so only manual runs hit this case.
There was a problem hiding this comment.
Review
Blocking Issues
scripts/check-release-notes.py:68-72: The--sincecheck requires every(#N)subject since the previous tag.cliff.tomldrops non-conventional subjects (filter_unconventional = true),docs(changelog),chore: bump version, andMerge pull request. Those PRs never reach the changelog, sorelease.sh finishrefuses to tag. Cutting the release again produces the same changelog, so the release never tags. This PR's own subject,Release scripts refuse a release without notes (#343), is non-conventional and triggers the failure on 0.43.0.
Action Required
- Exempt from the
--sincecheck every subject that git-cliff drops. Either derive the expected PR numbers from git-cliff output or mirror thecliff.tomlfilters. - Retitle this PR to a conventional subject, for example
ci: release scripts refuse a release without notes.
eddietejeda
changed the title
Mirror cliff.toml: a non-conventional subject, docs(changelog), the release and bump-version commits, and merge commits never reach the changelog, so the --since check must not demand them.
Codecov Report✅ All modified and coverable lines are covered by tests. 📢 Thoughts on this report? Let us know! |
1 participant
0.42.0 shipped with an empty GitHub Release. Its CHANGELOG section was empty because nothing had landed since the v0.41.0 tag: #340 and #341 merged while the 0.41.0 release PR waited for review, so they were inside that tag but outside its notes, and git-cliff's next run (which starts at the tag) never saw them either.
scripts/check-release-notes.py <version> [--since <ref>]: the section must exist and have bullets; with--since, every PR merged since that ref must appear in it by(#N).release.sh prepareruns it after cargo-release (bullets required).release.sh finishruns it with--since <previous v* tag>before tagging, and tells you to cut again from current main when PRs are missing.Checked against history: 0.40.0 passes, 0.41.0 fails (misses #340, #341), 0.42.0 fails (no bullets).