Skip to content

README.tests.md

Latest commit

 

History

History
113 lines (86 loc) · 4.5 KB

README.tests.md

File metadata and controls

113 lines (86 loc) · 4.5 KB
Raw

Testing

Frontend and backend each have their own suite. They share one stack and one seed corpus (seed-data/README.md).

Frontend

cd frontend
npm run check      # lint + tsc + unit/integration — run before committing
npm run test       # unit + integration, no Docker
npm run test:e2e   # brings the stack up, runs the specs, destroys it
frontend/tests/
├─ unit/          *.test.ts    pure logic, mirrors src/
├─ integration/   *.test.tsx   components + real store + MSW; flows/ uses the real routes
├─ e2e/           *.spec.ts    Playwright
└─ support/       NO TESTS LIVE HERE

Which tier? No DOM → unit. MSW can answer everything → integration. Only true against a real backend → e2e, and prefer a step in an existing spec.

  • MSW runs with onUnhandledRequest: 'error'; handlers are built from the app's own URL builders, so a renamed path breaks them instead of silently passing.
  • jsdom gaps are patched only in vitest.config.ts and tests/support/setup/.
  • E2E specs share one database and run alphabetically on one worker, so anything destructive uses its own account.

Backend

cd backend
uv sync --group dev       # once
./bin/test.sh             # starts the test database, runs everything
./bin/test.sh tests/api   # or -k <subject>, --create-db, --cov
uv run pytest tests/unit  # no database, no Docker
backend/
├─ conftest.py                  seeding, API clients, the throttle switch
└─ tests/
   ├─ unit/          test_*.py  the django_db marker is rejected here
   ├─ integration/   test_*.py  models, signals, tasks, serializers, consumers
   ├─ api/           test_*.py  DRF client against mounted endpoints
   └─ support/       NO TESTS LIVE HERE

Which tier? Decided by the boundary a test crosses, not by its subject. Runs without the database → unit. Needs the ORM, channel layer, filesystem or mail but still calls the code directly → integration. Routing, permissions, status codes, wire format → api.

The first rule is enforced: collection fails if anything under tests/unit/ uses django_db. The ORM is never mocked — a rule that needs a query is an integration test; extract the pure part instead.

Database

pytest creates test_backend_db beside the stack's backend_db, so a Playwright run and a pytest run can share one container. The corpus is seeded once per session and each test rolls back onto it. --reuse-db is the default; --create-db rebuilds.

django_db(transaction=True) commits for real and flushes afterwards, which empties the corpus. conftest.py runs those last and re-seeds if needed, so such a test must build its own data with factories. Only the cases that reach the channel layer through async_to_sync need it.

Fixtures

alice / bob / carol are signed-in clients for seeded accounts; client_for(seed.DAVE) for any other, client anonymous, client_as(user) for a factory-built user. All carry the auth cookies and echo X-CSRFToken, so popping csrftoken is how you assert a 403.

tests/support/helpers/frames.py builds board headers: signature_headers, token_headers, hmac_headers. serial_for(seed_b64) derives the serial rather than hard-coding it. tests/support/seed.py reads seed-data/*.json, so the corpus and the tests cannot drift.

Test settings

config/settings_test.py swaps Redis, mail and Celery for in-memory equivalents and moves MEDIA_ROOT aside. Two consequences:

  • DEBUG=False on purpose — the media views are only mounted when it is false; the trade is that api/schema/ is not routed.
  • Throttles are off. The throttled fixture restores the real rates.

The test stack

docker-compose.test.yml + .env.test, always under -p shareframe-test: backend, postgres (tmpfs), redis, mailhog. Ports are offset from dev, so both stacks can run at once. .env.test is committed; every value in it is fake.

Playwright brings up everything; pytest needs only backend_db.

Don't start an e2e run while pytest is mid-session — its teardown runs down -v. E2E_KEEP_STACK=1 covers the other direction. Always pass -p shareframe-test, or a down -v hits the dev database.

npm run e2e:up / npm run e2e:down      # from frontend/
npm run db:reset                       # flushes and re-seeds the DEV database

Conventions

Comments say why a test looks unusual, never what the code does. A test that pins a known bug rather than the intended behaviour says so — grep -rn "Known bug".