Testing¶

Root task ci runs formatting, Clippy linting, database-free unit tests, PostgreSQL 18 and RustFS integration tests, and Rust Playwright browser journeys against the Loco application.

Loco test stages¶

Local integration runs provision disposable PostgreSQL 18 and RustFS containers through compose.yaml; browser runs provision PostgreSQL only. Both ignore an ambient DATABASE_URL. The browser lane is a single job: Playwright is driven by the playwright-rs crate, which assembles its driver in the playwright-rust cache directory and Chromium in the usual ms-playwright cache. The driver supplies its own Node runtime; frontend:prepare still uses Node and npm to build the application CSS and scanner assets before the journeys run. Journeys are marked #[ignore] so the shared integration lanes skip them; task test:browser selects them with nextest --run-ignored only. Both integration and browser tasks fail when a filter selects no tests.

The Rust browser lane is deliberately smaller than the retired JavaScript suite; it does not claim test-for-test parity. Each journey chooses its viewport rather than inheriting an automatic desktop/mobile matrix. Retained journeys now include native passkeys and account forms, OAuth consent, care/scanner flows, administration and location-deletion protection. CI uploads captures under docs/screenshots/browser-journeys/ as the browser-journey-screenshots artifact.

At main 957e57390 on 23 October 2026, CI executed 34 browser journeys successfully, with one unignored harness unit test excluded by --run-ignored only. The same run passed 153 Rust library tests, 19 JavaScript unit tests and 1,169 tests in the integration lane (including the 153 library tests; 38 additional tests skipped), plus the separately invoked template-concurrency guard. Counts describe that run, not universal page, device or live-provider coverage.

The remaining concrete gaps have existing owners: typed browser fixtures, passkey cancellation/reuse browser paths, live provider/passkey acceptance and accessibility. Do not recreate the deleted suite or count API/render assertions as browser proof.

Running cargo test --tests directly requires an owned DATABASE_URL; the empty medtracker_reference base database is created lazily by tests/support/template.rs when absent. task test:integration provisions the owned instance. The ignored template_concurrency guardrail runs automatically on an unfiltered task test:integration.

Integration fixtures clone a migrated and seeded template for each fixture family into private databases. Short-lived OAuth fixture credentials are created for each clone so they cannot expire while waiting in a shared template. Each browser journey clones the seeded browser_template into a private database with a restricted runtime role, spawns the compiled med-tracker server on a dynamic port, and drops the database, role and server process on completion — including failure paths. Storage runtimes get private buckets.

Browser journeys¶

Browser journeys live in tests/browser/journeys.rs as ordinary #[tokio::test]s marked #[ignore] so the shared --tests lanes skip them. Each journey supplies an explicit Viewport to harness::journey(...), which clones the seeded browser_template into a private database, spawns the compiled med-tracker server, and drops everything on exit.

• Choose task test:integration for HTTP-level behaviour and task test:browser for a real browser driving the UI.
• To add a retained journey, write a test in journeys.rs, sign in with harness::sign_in(&journey), drive pages with semantic locators (get_by_role/get_by_label) and expect assertions, and mark it #[ignore] like the others.
• Fixture data comes from SeaORM entity seeds in tests/support/fixtures.rs; add a seeding helper there rather than raw SQL. Template-level rows go through harness's prepare callback; per-journey rows (for example seed_barcode_metadata) run inside the journey itself.
• playwright-rs assembles its driver at build time and install_browsers fetches Chromium on first run; both land in the user cache and need no separate Node install for Playwright. Application assets still use Node and npm through frontend:prepare.
• On failure the harness prints the tail of the server log; Playwright traces can be enabled per journey when more detail is needed.

Loco database setup¶

Start with an empty PostgreSQL 18 database. The setup connection must be able to create the required non-login roles and standard citext, pg_trgm and pgcrypto extensions, assign ownership and assume med_tracker_owner. Keep these credentials separate from the restricted application login.

Loco owns schema installation through versioned SeaORM migrations. Ordinary tables, indexes and foreign keys use builders; PostgreSQL-specific definitions are bundled with their migration. The seed hook invokes the migrator once before loading data.

Repeated migration is a no-op. Synthetic seeding refuses production environments and populated application databases. Reset and down commands remain disabled; create another empty database for a fresh start.

TDD workflow¶

MedTracker follows Red-Green-Refactor:

1. Write a failing test first.
2. Make the smallest change that passes.
3. Refactor while keeping tests green.

For a manual screen-reader and keyboard pass over browser journeys, use the manual accessibility smoke-test checklist.

Test fixture users use the password password.