Auto-release follows CI instead of testing the release again

A version bump on main ran the whole suite twice: once in CI for the
push, once more in auto-release. Auto-release now starts when CI
finishes on main and releases only if that run was green, tagging the
exact commit CI tested. Started by hand, it still tests first, since
there is no CI result to go by.
This commit is contained in:
Frederic Jung 2026-11-07 10:00:15 +01:00
parent 0342d90657
commit 03523cb16b
No known key found for this signature in database
3 changed files with 34 additions and 14 deletions

View file

@ -5,8 +5,9 @@ description: Cut and ship a Series of Tubes release (patch or minor) from main v
# Releasing Series of Tubes
Bumping `version` in `package.json` on `main` is the release button. The
`auto-release.yml` workflow tests, tags `v<version>`, publishes a GitHub release with
Bumping `version` in `package.json` on `main` is the release button. CI tests the push
as usual; when it passes, `auto-release.yml` (which runs after every CI on `main`) sees
the new version, tags `v<version>` on the commit CI tested, publishes a GitHub release with
that version's CHANGELOG section as notes, and dispatches `release.yml`, which builds
`_site/` (`npm run build`: the page assembled from `src/` plus `site/`) and deploys it
to GitHub Pages (https://seriesoftub.es). GitHub Pages is the only host, ever: see the
@ -65,8 +66,8 @@ gh api repos/fjaeckel/seriesoftubes/releases/latest --jq '.tag_name+" "+.publish
## Steps
1. Don't run the test suite, the game or benchmarks before a release: CI and the
auto-release workflow test it, and the Performance workflow benchmarks it. Just
1. Don't run the test suite, the game or benchmarks before a release: CI tests the
version bump (a red CI releases nothing), and the Performance workflow benchmarks it. Just
check that the latest CI run on `main` is green (`gh run list --repo fjaeckel/seriesoftubes --limit 4`).
2. In `CHANGELOG.md`, move the bullets from `## [Unreleased]` into a new section
right below it: `## [x.y.z] — YYYY-MM-DD` (em dash, today's date). Keep an empty
@ -81,7 +82,7 @@ gh api repos/fjaeckel/seriesoftubes/releases/latest --jq '.tag_name+" "+.publish
--author="Frederic Jung <frederic@jc-ix.net>" -m "Short summary (x.y.z)"
git push origin main
```
6. Watch the workflows until both finish:
6. Watch the workflows until all three finish (CI, then Auto-release, then Release to GitHub Pages):
```bash
gh run list --repo fjaeckel/seriesoftubes --limit 4
gh run watch <id> --repo fjaeckel/seriesoftubes

View file

@ -5,10 +5,15 @@ name: Auto-release
# as notes, and starts the deploy. Releases created with the workflow token do
# not fire the "released" event, so the deploy is started explicitly, from the
# tag, which keeps the github-pages environment's v* tag rule happy.
#
# It runs when CI finishes on main, so a release never repeats the push's tests:
# CI green on the commit that bumped the version is the go. A red CI releases
# nothing. Started by hand, it has no CI result to go by, so it tests first.
on:
push:
workflow_run:
workflows: [CI]
types: [completed]
branches: [main]
paths: [package.json]
workflow_dispatch:
permissions:
@ -21,6 +26,10 @@ concurrency:
jobs:
check:
# After CI: only a green run of a push to main (not a PR's run, not a release's own call).
if: >-
github.event_name == 'workflow_dispatch' ||
(github.event.workflow_run.conclusion == 'success' && github.event.workflow_run.event == 'push')
runs-on: ubuntu-latest
env:
GH_TOKEN: ${{ github.token }}
@ -29,9 +38,12 @@ jobs:
ver: ${{ steps.v.outputs.ver }}
tag: ${{ steps.v.outputs.tag }}
pre: ${{ steps.v.outputs.pre }}
# The commit CI tested (main may have moved on since); by hand, main as it is now.
sha: ${{ github.event.workflow_run.head_sha || github.sha }}
steps:
- uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
with:
ref: ${{ github.event.workflow_run.head_sha || github.sha }}
fetch-depth: 0
- name: Work out the version and whether it is new
@ -58,23 +70,30 @@ jobs:
case "$ver" in *-*) echo "pre=true" >> "$GITHUB_OUTPUT";; *) echo "pre=false" >> "$GITHUB_OUTPUT";; esac
echo "skip=false" >> "$GITHUB_OUTPUT"
# The same tests as every push, split over several runners (ci.yml), before anything is tagged.
# Only when started by hand: after a push, CI has already tested this commit.
tests:
needs: check
if: needs.check.outputs.skip == 'false'
if: needs.check.outputs.skip == 'false' && github.event_name == 'workflow_dispatch'
uses: ./.github/workflows/ci.yml
with:
ref: ${{ needs.check.outputs.sha }}
release:
needs: [check, tests]
if: needs.check.outputs.skip == 'false'
if: >-
always() && needs.check.result == 'success' && needs.check.outputs.skip == 'false' &&
(needs.tests.result == 'success' || needs.tests.result == 'skipped')
runs-on: ubuntu-latest
env:
GH_TOKEN: ${{ github.token }}
TAG: ${{ needs.check.outputs.tag }}
VER: ${{ needs.check.outputs.ver }}
PRE: ${{ needs.check.outputs.pre }}
SHA: ${{ needs.check.outputs.sha }}
steps:
- uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
with:
ref: ${{ needs.check.outputs.sha }}
- name: Publish the release
run: |
@ -82,11 +101,11 @@ jobs:
awk -v v="$VER" 'index($0,"## ["v"]")==1{f=1;next} /^## \[/{f=0} f' CHANGELOG.md | sed -e '/./,$!d' > notes.md
pre=(); [ "$PRE" = "true" ] && pre=(--prerelease)
gh release create "$TAG" --repo "$GITHUB_REPOSITORY" \
--target "$GITHUB_SHA" --title "$VER" --notes-file notes.md "${pre[@]}"
--target "$SHA" --title "$VER" --notes-file notes.md "${pre[@]}"
- name: Deploy it
# Pre-releases are for testing and never go live. The tag was just tested above, so the deploy does not
# test it again (tested=true).
# Pre-releases are for testing and never go live. CI (or the tests above) already passed on this commit,
# so the deploy does not test it again (tested=true).
if: env.PRE == 'false'
run: gh workflow run release.yml --repo "$GITHUB_REPOSITORY" --ref "$TAG" -f tag="$TAG" -f tested=true

View file

@ -83,7 +83,7 @@ Development runs on `npm run dev` (assembles the development page on every reque
| `tests/harness.js` | Loads the assembled page in jsdom with a seeded `Math.random` and a fake canvas: the players' page, or the development page with `{agent: true}`. The page's scripts run straight in the window (jsdom's parser is slow on 2 MB of script) |
| `scripts/test.js`, `tests/durations.json` | Runs the tests, every file in its own process; splits them into parts of equal time for CI (`--plan`, `--shard`) by each file's seconds, which CI refreshes after every run on `main` |
| `.github/workflows/ci.yml` | Tests (split over 6 runners), build and translations on every push and PR; the releases call it too, so the tests run one way everywhere |
| `.github/workflows/auto-release.yml` | On a `package.json` version bump on `main`: test (ci.yml), tag, GitHub release from the CHANGELOG, start the deploy (already tested, so it does not test again) |
| `.github/workflows/auto-release.yml` | After every green CI on `main`: if `package.json` has a new version, tag the commit CI tested, publish the GitHub release from the CHANGELOG, start the deploy (`tested=true`, so it does not test again) |
| `.github/workflows/release.yml` | Builds and deploys to GitHub Pages, the only workflow that can |
| `.github/workflows/weekly-puzzle.yml` | Sunday 03:37 UTC: makes the week's puzzle, commits it to `main` (data only, no release) and starts `release.yml` for the live release tag with `weekly=true` |
| `scripts/weekly-puzzle.js`, `site/puzzles/weekly/` | The weekly puzzle generator and its output, one `<key>.json` per week plus `index.json` (see below) |