Repository navigation
docs(ops,quality): the smoke test these pages still describe was deleted before they were written - #122
Merged
Conversation
…e it was written "What the pipeline does" describes three platform builds, "each installing or extracting its bundle, launching it, asking its health endpoint, and comparing the reported versions before upload", and a publish step gated on all three. None of that is the pipeline. There are four platform builds — Android joined on 2026-08-17 — and publish declares five dependencies, the fifth being a self-host smoke job added on 2026-09-05. Neither change reached this page. The launch-and-probe is older than a drift. Per-platform smoke probes existed for one day: they went in on 2026-05-28 and came out the same day, because a hosted runner cannot start these bundles — the Linux runner has no display server, the Windows runner loses the silent-install to GUI-launch race, and macOS trips ad-hoc-signing friction on `open`. This page was written two months later, on 2026-07-27, and described the morning shape rather than the one that survived the afternoon. What the platform builds actually do is interrogate the artifact they produced: signing identity, forbidden contents, and the manifest written beside it. What is launched and probed is the self-host recipe, which is the one shape CI can run. The page already asks, on its own checklist, whether the ruleset's required checks still name the jobs that actually run (OPS-R18), on the grounds that a renamed job silently stops being required. Its own list of jobs had gone stale the same way. Verify-published is added to the list for the same reason: it is the step that re-reads the manifests and the checksum file off the release page and re-checks every signature, and the page's "verify the signed manifest by hand" item is easier to read next to it. This commit changes the description only. Whether OPS-R5 — "Every platform build MUST pass its smoke test before the publish step runs" — still says what it should, now that no platform build launches anything, is a question for the rule and not for this page. Signed-off-by: Wessel Verheij <info@nightworks.io>
github-actions
Bot
added
documentation
`40-quality/ci-cd.md` carries an ASCII drawing of the release pipeline holding the same three errors as `70-operations/releasing.md`: three platform builds rather than four, a publish gated on "all three" rather than on five, and a per-platform smoke test — "install or extract, launch, ask the health endpoint, compare the reported versions" — that was removed on the day it was added, because a hosted runner cannot launch these bundles. A diagram is read more readily than the prose beside it and checked less often, so the same wrong shape survives longer here. Corrected to match: what each platform build checks is the artifact it produced, what gets launched and probed is the self-host recipe, and the verify-published step is drawn rather than left off the end. `40-quality/tooling.md` also says three platform runners, and is right to: it is describing desktop packaging, and the fourth build is mobile. Left alone. Signed-off-by: Wessel Verheij <info@nightworks.io>
lessevv
changed the title
|
github-actions
Bot
added
awaiting-maintainer
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Two pages describe the release pipeline. Both describe three platform builds gated on "all three", and both describe a per-platform smoke test that "installs or extracts its bundle, launches it, asks its health endpoint, and compares the reported versions".
None of that is the pipeline.
The counts
publishdeclares five dependencies:needs: [build-macos, build-windows, build-linux, build-android, smoke-server]. The fifth is a self-host smoke job added 2026-09-05 (beatrax#374).Neither change reached either page. "Nothing is published — the publish step requires all three" is the sentence an operator reads when four builds are green and publish never ran.
The launch-and-probe is older than a drift
Per-platform smoke probes existed for one day. In on 2026-05-28 (
1c184c5, "per-platform smoke probes"), out the same day (8dd6811, "drop smoke-test steps from release.yml") — quoting that commit: the Linux runner has no display server, the Windows runner loses the silent-install-to-GUI-launch race, and macOS trips ad-hoc-signing friction onopen -a.70-operations/releasing.mdwas written two months later, on 2026-07-27, in the founding commit. It documented the shape that existed that morning rather than the one that survived the afternoon — so this half has never been true of the pipeline it describes.What the platform builds actually do is interrogate the artifact they produced: signing identity, forbidden contents, and the update manifest written beside it. What is launched and probed is the self-host recipe, the one shape CI can run end to end.
Two files, because a diagram outlives the prose
40-quality/ci-cd.mdholds the same three errors as an ASCII drawing of the pipeline. A diagram is read more readily than the prose beside it and checked less often, so the wrong shape survives longer there. Both are corrected, andverify what was publishedis drawn rather than left off the end.40-quality/tooling.mdalso says three platform runners and is right to — it describes desktop packaging, and the fourth build is mobile. Left alone.Why these pages, of all pages
releasing.mdalready asks, on its own pre-tag checklist, whether the ruleset's required checks "still name the jobs that actually run" (OPS-R18) — because a renamed job silently stops being required. Its own list of jobs had gone stale the same way.One thing deliberately not changed
OPS-R5 — "Every platform build MUST pass its smoke test before the publish step runs."
The rule is defined once and elaborated nowhere else in the spec; the only gloss on it was the step this PR corrects. With no platform build launching anything, OPS-R5 either wants restating against what a build now verifies, or it stands as a requirement the pipeline does not meet and cannot meet on hosted runners. That is a decision about the rule, not about these pages, so OPS-R5 is untouched and flagged for a separate call.
Descriptions only — no rule added, removed, renumbered or reworded.
integrity.pyclean, markdownlint clean.