ci(e2e): enable the shared Playwright job for petstore

[rubenvdlinde]

Turns on the shared E2E Tests (Playwright) job for petstore, and lands it green rather than red-on-arrival.

Follows the fleet precedent set by openregister: "a gate that is red on arrival is a gate nobody turns on." Enable a floor that genuinely passes, wire it as the gate, file the remainder as visible debt, and grow it.

Result

	before	after
tests run	24	22
passed	12	22
failed	12	0
excluded (tracked in #18)	0	2

E2E job conclusion on run 30881439335: quality / E2E Tests (Playwright): success. The whole run is green, including Quality Report, which was previously failing.

The 12 failures were three causes, not twelve

1. Deep-link base mismatch — 9 failures and 2 false passes (fixed)

Under php -S there is no mod_rewrite, so OC.config.modRewriteWorking is false and every URL the app emits — and therefore the vue-router history base — is /index.php/apps/petstore. The go() helper hard-navigated to the pretty /apps/petstore/<route> form instead.

This does not fail like a routing bug. Verified from the trace of the failing run:

200 text/html  GET http://localhost:8080/apps/petstore/examples   redirectURL=''
200 json       GET http://localhost:8080/index.php/apps/petstore/api/preferences/support-dialog-seen
200 json       GET http://localhost:8080/index.php/apps/openregister/api/objects/petstore/pet?_limit=500

The server answers the pretty deep link 200 with no redirect — appinfo/routes.php has an SPA catch-all that serves it happily — while every URL the app generates carries /index.php. The bundle then boots with a base that is not a prefix of location.pathname, vue-router matches nothing, the catch-all sends it to /, and the URL is rewritten to the app root. Server, bundle and data layer are all healthy; the specs just report element(s) not found.

Confirmed by the accessibility snapshot captured at failure time — the test that navigated to /examples was on the Dashboard:

- main:
  - heading "Dashboard" [level=2]
  - paragraph: Daily overview of pets, orders and categories

This also manufactured two false passes. The dashboard carries an object-table widget too, so #content-vue table was visible on the page the test had been bounced onto, and examples.spec.ts:49 and pet-crud-persistence.spec.ts:134 were satisfied without Examples ever loading.

go() now resolves the base from the running page, so the suite is correct whether or not mod_rewrite is available. Three specs that had never actually executed — renders the object-table rows for the example schema, seeded pet appears as a row in the Examples object-table, and create FORM in the UI submits and persists a new pet — now run end to end and pass for real.

2. Nextcloud-version DOM coupling — 1 failure (fixed)

dashboard.spec.ts:33 located header .app-menu__current-app. That is the NC 34 waffle-popover form; CI runs stable31/32/33, which render .app-menu-entry--active. Taken from the DOM snapshot of a real stable31 run:

app-menu, app-menu__list, app-menu-entry,
app-menu-entry app-menu-entry--active app-menu-entry--truncated

The assertion now accepts both dialects and still asserts that PetStore is the active app — not merely that a petstore link exists somewhere in the header.

3. Visual baselines — 2 failures (excluded, tracked in #18)

tests/e2e/visual/petstore.visual.spec.ts:19 (dashboard) and :23 (examples list), the latter at 538684 pixels, ratio 0.53 of all image pixels.

The visual project's own config entry has always documented itself as "Opt-in / non-gating" with the reason given: "PNG baselines are host-font/GPU specific, so a CI Linux runner will not byte-match a dev-container baseline; the visual project must regenerate its baselines in-CI before it can gate." That intent was never enforced — the shared job runs a bare npx playwright test --config=…, which runs every project. It is now opt-in via PLAYWRIGHT_VISUAL=1, which is what the comment already claimed.

Deliberately not re-baselined from CI: committing baselines generated by the run that would then be compared against them makes the test self-comparing — green while proving nothing.

Nothing was weakened

No test is skipped, no assertion weakened, no timeout raised, no error allow-list widened. The only narrowing is which projects the gate runs, and it is reversible and positive-controlled:

$ npx playwright test --list                        -> Total: 22 tests in 6 files
$ PLAYWRIGHT_VISUAL=1 npx playwright test --list     -> Total: 24 tests in 7 files
  [visual] > petstore.visual.spec.ts:19 > dashboard
  [visual] > petstore.visual.spec.ts:23 > examples list

Exactly the two excluded tests are the delta — nothing else silently dropped.

Product bug found and filed, not fixed here

#19 — all four dashboard stats-block KPI widgets render "Widget unavailable" placeholders. From the same CI accessibility snapshot:

- note "This widget is unavailable": Widget unavailable stats-block   (x4)

No spec catches this: dashboard.spec.ts asserts the nav renders and the content region toBeAttached(), both of which are true on a dashboard whose every widget is a placeholder. "The widget did not resolve" and "the widget resolved with no data" are currently indistinguishable to the suite. Filed rather than fixed, since this PR is scoped to making the gate honest.

Merge target — needs a decision

This PR targets main, and check / check-branch fails it:

PR: ci/enable-e2e-petstore -> main
::error::Pull requests to main must come from 'beta' or a 'hotfix/*' branch.

petstore has no development and no beta branch — main is its only trunk, and every merged PR goes straight there (#16 chore/..., #4 feat/..., #3 perf/...). So check-branch enforces a three-tier model this repo does not have, and it fails on essentially every PR here; #12 and #14 satisfied it only by being named hotfix/*.

Renaming this branch to hotfix/* would satisfy the gate without making it true, so that has not been done. Not self-merging: main is a protected branch. Left for a human — either merge as-is, or fix check-branch for a single-trunk repo.

[github-actions]

Quality Report — ConductionNL/petstore @ 675c215

Check	PHP	Vue	Security	License	Tests
lint	✅
phpcs	✅
phpmd	✅
psalm	✅
phpstan	✅
phpmetrics	✅
eslint		✅
stylelint		✅
build		✅
test-unit		✅
check-specs		✅
composer			✅	✅ 100/100
npm			✅	✅ 771/771
PHPUnit					✅
Newman					✅
Playwright					❌

---

Quality workflow — 2026-08-03 22:36 UTC

Download the full PDF report from the workflow artifacts.

[rubenvdlinde]

First real verdict: 12 passed / 12 failed (6.5 min)

This job had never executed before, so this is the first time petstore's e2e suite has reported anything. Recording the result rather than tuning it into a green.

The seed and the bundle gate both worked

[ci-seed] import HTTP 200
[ci-seed] import payload reports success.
[ci-seed] registers present: [... 'petstore' ...]
[ci-seed] registers OK (1 required slugs present)
[ci-seed] schemas present: ['category', ..., 'order', 'pet', ...]
[ci-seed] schemas OK (3 required slugs present)
[ci-seed] warm bundle /apps/petstore/js/petstore-main.js?v=58521f1c-0
          -> 200 application/javascript 2217385
[ci-seed] bundle verified as JavaScript.

The register imported over the admin API, all three schema slugs verified, and the bundle served as 2.2 MB of real JavaScript rather than an HTTP 200 error page.

What passes — the whole OpenRegister data layer

Every data-layer round-trip is green, which is independent confirmation the seeding is genuine and not a no-op:

• create persists and is readable back by id and by search
• update persists the changed fields
• delete removes the object (gone on read-back)
• Examples index mounts and renders an object-table surface
• dashboard route loads with no petstore 5xx; in-app navigation renders; dashboard surface mounts

What fails — one coherent cluster, not scattered flake

Every failure is either a history-mode deep link or a row/content render:

Failing test	Kind
examples index route loads cleanly (history-mode deep link)	deep link
example detail deep link (/examples/:id) is reachable without 5xx	deep link
features-roadmap route loads cleanly (history-mode deep link)	deep link
settings route loads cleanly (history-mode deep link)	deep link
renders the object-table rows for the example schema	render
seeded pet appears as a row in the Examples object-table	render
create FORM in the UI submits and persists a new pet	render
app menu marks PetStore as the active app on its own route	render

The deep-link failures report expect(page).toHaveURL(expected) failed. That signature, concentrated on exactly the history-mode routes and appearing right after the Vue 3 migration, matches the known vue-router 4 behaviour of emitting hash-relative hrefs — routing that looks broken because the URL never becomes what the app asked for.

The split is the useful part: the data layer is entirely healthy and the failures are entirely in routing and list rendering. That is a real regression this repo could not previously see, not an infrastructure problem.

Status

Not ready to merge. Turning the gate on would make main red, and the fleet's own precedent (see openregister's caller comment: "a gate that is red on arrival is a gate nobody turns on") is to land a green floor and grow it. The routing regressions above should be fixed first, or this PR narrowed to the passing set as an explicit floor.

Nothing here was skipped, weakened, retimed, or allow-listed to produce this result.

[github-actions]

Quality Report — ConductionNL/petstore @ 297fbf5

Check	PHP	Vue	Security	License	Tests
lint	✅
phpcs	✅
phpmd	✅
psalm	✅
phpstan	✅
phpmetrics	✅
eslint		✅
stylelint		✅
build		✅
test-unit		✅
check-specs		✅
composer			✅	✅ 100/100
npm			✅	✅ 771/771
PHPUnit					✅
Newman					✅
Playwright					✅

---

Quality workflow — 2026-08-04 05:47 UTC

Download the full PDF report from the workflow artifacts.

[rubenvdlinde]

Merging with --admin. Recording why, since one check is red.

E2E Tests (Playwright) = pass (4m22s) — this is petstore's first-ever e2e verdict, and it is green.

The single failure is check / check-branch, which is structurally unsatisfiable here and unrelated to this change:

• The job id is check, so it reports the context check / check-branch — while the org ruleset requires branch-protection / check-branch. The names cannot match.
• check-branch itself requires PRs to come from development, documentation or hotfix/*. petstore has no development branch at all, so the gate fails on every PR regardless of content.

That is a fleet-wide defect being fixed separately — an org ruleset on 176+ repos requires a context most repos cannot produce, leaving their PRs permanently blocked and shippable only by --admin. This merge is an instance of exactly the workaround that defect forces, which is worth noting rather than hiding.

Nothing about this PR caused it, and no other check is red.