A failing stage prints FAILED with its name and exits non-zero; e2e:shell is reported NOT RUN (it needs the isolated stack); both suites keep their own commands.
6.8 KiB
Browser tests (Playwright and axe)
Status: V1-1.3, 2026-10-25 (decision D14, V1-1_OWNERS_ANIMALS.md §12.5).
Chromium only until V1-5. The suites live in apps/app/e2e.
1. Suites
| Command | Backend | Dev server | Content |
|---|---|---|---|
pnpm e2e:ui |
none | Vite on 127.0.0.1:5191 |
e2e/ui/dialog.spec.ts: the D13 contract of the shared Dialog on a test page; e2e/ui/local-tool.spec.ts: the disclaimer, edit and replace dialogs on the real local-tool routes |
pnpm e2e:shell |
the isolated stack of pnpm db:sandbox |
Vite on 127.0.0.1:5192 |
e2e/shell/shell.spec.ts: sign-in, dashboard, the account photo confirmation, the phone drawer |
Each suite runs on two Chromium projects: chromium-desktop (1280 × 800) and
chromium-375 (375 × 667, mobile viewport, touch). A test tagged @phone
runs on the phone project only (the drawer has no desktop counterpart). Both
languages run: the local tool and the shell through the browser locale and
the account's locale, the test page through ?lang=en.
The test page e2e/harness/dialog.html is served by the dev server only
(/e2e/harness/dialog.html); the build has one entry, index.html, so it
never ships. It holds the D13 cases (basic, long content, refused close,
vanishing opener) and no product journey.
The shell suite never reaches DEV. Its dev server is built with the Supabase
URL http://sandbox.faunosteo.invalid (a name that resolves nowhere,
RFC 6761); e2e/support/sandbox.ts replays /auth/v1, /rest/v1 and
/storage/v1 on the sandbox's loopback ports, the way Kong maps them on the
CLI stack. Each test creates its own confirmed account through the sandbox's
GoTrue admin API (random address under sandbox.local, random password, a
fictitious display name, the profile locale of the test), and seeds the clean
avatar the worker would have produced (the assets row as postgres, the
object with the service key). pnpm db:sandbox down removes all of it.
2. Prerequisites
- The repository's Node and pnpm,
pnpm install. - The Chromium build of the pinned Playwright (1.64.1, Chromium 1248):
pnpm --filter @faunosteo/app exec playwright install chromium. It downloads into~/.cache/ms-playwrightand needs nosudo. Never pass--with-deps: it installs system packages withsudo, an owner action. On the Ubuntu development machine the headless shell runs with the libraries already present. pnpm wasm:build: the local tool runs the core, and the UI suite stops on a missing or stale module (packages/core-wasm/tests/global-setup.ts).- For
pnpm e2e:shell: Docker (rootless works) and the sandbox up,pnpm db:sandbox up fresh(orup upgrade <version>). - Ports 5191 and 5192 free: each server uses
--strictPortand is never reused, so the owner'spnpm devon 5173 is never picked up by mistake.
3. Guarantees
- Request guard (
e2e/support/browser.ts, automatic in every test): an allowlist of origins per suite (the dev server, plus the fake Supabase origin for the shell). Any other request, DEV's API on the LAN address included, is aborted and fails the test with its URL. - No fixture value in a URL: the local tool's names and the shell account's email, password and display name are checked against every request URL; a match fails the test and the report masks the value.
- axe (
@axe-core/playwright4.14.0, rules tagged WCAG 2.0 to 2.2 A and AA plus best practices): zero serious or critical violation on each scanned state, colours measured once animations have settled. Three pre-existing findings, outside the V1-1.3 code, are listed with their reason inKNOWN_VIOLATIONSand in TODO RR-7: the language switcher's names (« FR » shown, « Passer en français » announced), the top bar account link (display name shown, « Mon compte » announced), the dashboard'sdlholding a status row withoutdt/dd. Each entry names the axe rule, the selector, the opening markup of the element and the route where it lives; a node is let through only when all four match, so a new screen or a new element never inherits an exception (owner brief 2026-10-26). - Reports:
apps/app/e2e/test-results/andapps/app/e2e/playwright-report/, git-ignored. The UI suite keeps a trace and a screenshot of a failed test (no backend, no secret). The shell suite keeps a screenshot only: a trace would hold the session tokens and the typed password.
4. Reading a failure
The terminal shows each step and the failing assertion. The HTML report:
pnpm --filter @faunosteo/app exec playwright show-report e2e/playwright-report/ui
(or .../shell); a UI trace: pnpm --filter @faunosteo/app exec playwright show-trace <path printed in the failure>. Setup failures are named: a missing
Chromium (Playwright prints the install command), a stale core module
(run pnpm wasm:build), the shell suite outside its wrapper
(SANDBOX_ANON_KEY missing: run pnpm e2e:shell), the sandbox down
(db-sandbox: not running).
5. In pnpm check:all and a future CI
pnpm check:all runs pnpm e2e:ui when the pinned Chromium is installed;
otherwise it prints SKIPPED with the reason and the install command, and
goes on. pnpm e2e:shell needs the sandbox and stays a separate run, like
pnpm e2e:avatar. The last lines of a green run state each browser suite:
PASSED, SKIPPED (environment missing, with the reason) or NOT RUN
(e2e:shell). A suite that runs and fails stops check:all with
== FAILED: <stage> and a non-zero exit, never a warning. Both suites keep
their own commands and run alone. A CI runner installs the same Chromium
(playwright install chromium in its image) and runs pnpm e2e:ui; the shell
suite needs Docker in the runner for the sandbox.
6. Limits
- Chromium only. Firefox, WebKit and the 412 × 915 viewport join at V1-5
(contract §12.5). WebKit's system libraries may need
sudo(owner action). Desktop WebKit is not iOS Safari: the owner's manual check on a real iPhone and on a computer over the LAN stays a validation criterion of every mobile step. - Tab past a modal dialog's last control reaches the browser's own interface for one step, then the first control (the platform's behaviour, D13); the page behind is never reached. The tests assert exactly that walk.
- Safari does not focus a button on a pointer click, so a dialog opened by a click has no opener to return to there; a keyboard opening always has one.
- The vitest suites (happy-dom) cover the React wiring of the dialogs only: happy-dom has no top layer, no inertness and no dialog focusing steps.
7. Adding a test
Import test and expect from e2e/support/browser.ts (the request guard
comes with them), read every visible string from @/i18n/{fr,en}, use
fictitious values only, and pass the values a test creates to
requests.forbid(...). A new backend origin goes into the suite's
allowedOrigins, never into a wildcard.