Repository navigation
[PR] ci(release): ship-time audit — flag backport-labeled PRs missing from the train (#4474) - #4475
Conversation
… the train (#4474) A conflicted backport leaves only a PR comment; CI green on the train head says nothing about fixes that are ABSENT. Before opening the ship PR, reconcile every backport-labeled merge since the train's base tag (merge-commit ranges + squash-shape subjects) against the cherry-pick -x trailers on the train; any no-show renders as a prominent warning section at the top of the ship PR body. Never fails the run — the owner may ship without the straggler, but must see it first. Verified end-to-end against the live repo: stale train (missing the #4470 pick) -> flags {#4473, #4470}; real train (pick present) -> flags {#4473} only, both directions.
📦 Built Package ArtifactBranch: Latest build — install this exact versionnpm install -g billion-context@0.0.5376-pr.4475.4d05ac1This exact version is immutable and reproducible. The rolling Recent builds (commit → version)
Download artifact
tar xzf billion-context-pr4475.tgz
npm install -g packageInstall instructions are refreshed on each push. |
📦 Merged Build PublishedThis merge to master was published as npm install -g billion-context@0.0.785-master.2d51e5d # this exact build
npm install -g billion-context@master # always the newest merged build
|
|
Backport to |
Fixes #4474.
The gap (owner-spotted in-session): a backport cherry-pick conflict leaves only a PR comment. Ship-time CI validates the train TREE is green — never that every backport-labeled fix actually MADE IT onto the train. So a patch release could silently ship without a fix the owner labeled for it.
The fix: a reconciliation step in the one-click train flow (
release-manual.yml), running just before the ship PR opens:Merge pull request #N) via theirM^1..M^2ranges AND squash-shape subjects ending(#N);-xtrailers on the train (cherry picked from commit <sha>);Never fails the run — the owner may legitimately ship without the straggler (e.g. #4473, a CI-only change that cannot ride a tag-based train), but it becomes impossible to MISS at the single human decision point.
Verification (live repo, both directions):
{#4473, #4470}— absence detected;466fc281present) → flags{#4473}only — presence detected via trailer, no false positive.YAML parses (20 steps), audit
runblock passesbash -n, and the shell logic executed verbatim against real refs + real PR labels.
')reference/release.mddocuments the audit in the dual-lane ship step.$(printf '