Repository navigation
ci: check the UI mock in the nightly release routine (#777) - #912
Conversation
|
Written-by: HugoXK @HugoXK-bot open PR queue, oldest first:
Please take them in that order and name the blocking sentence when you request changes. |
HugoXK-bot
left a comment
There was a problem hiding this comment.
Verdict: PASS at 73ee997
Written-by: Fledge Platform Reviewer
Posted-by: HugoXK-bot
I read the tip from the API this turn (73ee9971a494791d0bd84cdc5a50a2043b50839c, base 8213cec5, merge base 482488ad) and re-ran the gates myself rather than restating the body.
The gate itself is right: the new step sits after the opener precondition and is gated on steps.opener.outputs.ready == 'true' && inputs.dry_run != true (.github/workflows/daily-release.yml:359-370), so a dry run, a slot-shipped night and a publish night all skip it, and a stale snapshot exits 1 before the branch push — nothing is pushed and the slot is not consumed, which is the failure shape #333 wanted. set -euo pipefail plus the explicit ::error:: names the repair command. The step has no token in its environment: RELEASE_OPEN_TOKEN is referenced only in the opener step's env: even though the job is attached to the release environment, so npm ci never sees it. docs/deploy.md:342-348 matches the step.
Two notes that do not block this tip:
- After the flight-deck UI lands (
#911),mock/is stale by construction and the next night reds until the docs pull request regenerates it; that is the documented policy, but it means the first night after that merge should be expected red rather than treated as an incident. - If
npm ciitself fails, the step exits 1 without the::error::repair hint; the log still names the command, so it is readable.
Ran:
npm run check— pass, exit 0,# tests 2434/# pass 2434/# fail 0npm run mock:check—mock-snapshot: ok (12 files)node scripts/workflow-injection.mjs --workflows .github/workflows --allowlist .github/agent/workflow-injection-allowlist.json—workflow-injection: ok (9 workflow file(s), 2 allowlisted expression(s))node scripts/risk-gate.mjs --base 482488ad --head HEAD—risk-tier: T3; basis: .github/workflows/daily-release.ymlGITHUB_REPOSITORY=HugoXK/fledge GITHUB_EVENT_NAME=merge_group … node scripts/pr-lint.mjs --pr 912—pr-lint: ok (#912, T3, 12 files; tracks #777, closes nothing)node scripts/changelog.mjs check—changelog: 27 fragment(s) oknode scripts/version.mjs check—version: 0.4.0 oknode scripts/secret-scan.mjs—secret:scan: no suspected value in the tracked files
Not run: a real nightly run (the step first executes on the next planned release), actionlint 1.7.12 (not installed here; the job's only new shell is npm ci plus the mock:check branch, with no ${{ }} inside it).
HugoXK-bot
left a comment
There was a problem hiding this comment.
Verdict: PASS at 73ee997
Written-by: HugoXK-bot
Posted-by: HugoXK-bot
Replaces: 5917443193 (same tip; its second line named a writer the policy does not list)
I read the tip from the API this turn (73ee9971a494791d0bd84cdc5a50a2043b50839c, base 8213cec5, merge base 482488ad) and re-ran the gates myself rather than restating the body.
The gate itself is right: the new step sits after the opener precondition and is gated on steps.opener.outputs.ready == 'true' && inputs.dry_run != true (.github/workflows/daily-release.yml:359-370), so a dry run, a slot-shipped night and a publish night all skip it, and a stale snapshot exits 1 before the branch push — nothing is pushed and the slot is not consumed, which is the failure shape #333 wanted. set -euo pipefail plus the explicit ::error:: names the repair command. The step has no token in its environment: RELEASE_OPEN_TOKEN is referenced only in the opener step's env: even though the job is attached to the release environment, so npm ci never sees it. docs/deploy.md:342-348 matches the step.
Two notes that do not block this tip:
- After the flight-deck UI lands (
#911),mock/is stale by construction and the next night reds until the docs pull request regenerates it; that is the documented policy, but it means the first night after that merge should be expected red rather than treated as an incident. - If
npm ciitself fails, the step exits 1 without the::error::repair hint; the log still names the command, so it is readable.
Ran:
npm run check— pass, exit 0,# tests 2434/# pass 2434/# fail 0npm run mock:check—mock-snapshot: ok (12 files)node scripts/workflow-injection.mjs --workflows .github/workflows --allowlist .github/agent/workflow-injection-allowlist.json—workflow-injection: ok (9 workflow file(s), 2 allowlisted expression(s))node scripts/risk-gate.mjs --base 482488ad --head HEAD—risk-tier: T3; basis: .github/workflows/daily-release.ymlGITHUB_REPOSITORY=HugoXK/fledge GITHUB_EVENT_NAME=merge_group … node scripts/pr-lint.mjs --pr 912—pr-lint: ok (#912, T3, 12 files; tracks #777, closes nothing)node scripts/changelog.mjs check—changelog: 27 fragment(s) oknode scripts/version.mjs check—version: 0.4.0 oknode scripts/secret-scan.mjs—secret:scan: no suspected value in the tracked files
Not run: a real nightly run (the step first executes on the next planned release), actionlint 1.7.12 (not installed here; the job's only new shell is npm ci plus the mock:check branch, with no ${{ }} inside it).
#921) `daily-release.yml`'s mock-snapshot step was the only workflow install that #916's sweep did not cover, because #912 landed while #916's checks were computed against an earlier base. The step runs `npm run mock:check` over the installed tree, so a dependency lifecycle script must not run first. Docs-not-needed: the workflow's behaviour is unchanged; only the install flag follows the rule the sweep already enforces. Fixes #920
Written-by: HugoXK
Part of #777. Makes the generated mock a real release artifact instead of a silent local command.
What changed:
npm ciandnpm run mock:checkafter the opener precondition and before it pushesrelease/v<version>. A stale mock is a red night that namesnpm run mock:snapshotand the docs pull request.docs/deploy.mdsays the nightly check is the gate; the release routine regenerates the snapshot in a docs pull request, and normal feature PRs do not touchmock/.mock/is regenerated frommainat482488aso the check is green at this tip. After the flight-deck UI (ui: give the console a flight-deck visual system (#547) #911) merges, a docs PR will regenerate the snapshot again; until then the nightly correctly reds instead of shipping a stale snapshot.Why:
mock:checkwas a local-only command; no workflow called it, so the snapshot could drift silently.docs/deploy.md; this keeps it once per release.Ran:
npm run check— 2432/2434 pass; the 2 failures are pre-existing macOS environment failures (test/gateBaseTree.test.ts,test/nginxCheck.test.ts) that also fail on a cleanmain.npm run mock:check— ok (12 files)node scripts/risk-gate.mjs --base server-gh/server-deploy—risk-tier: T3(base482488a=main)node scripts/version.mjs check— oknpm run secret:scan— cleannode scripts/changelog.mjs check— 27 fragments okNot run:
Risk: T3 (
.github/workflows/daily-release.yml). I did not post a PASS, addreview:approved/agent-ready, or merge this branch.