Writing unit tests (webui)

Stack

• vitest (yarn test runs vitest run), jsdom environment, React Testing Library (@testing-library/react, @testing-library/user-event, @testing-library/jest-dom).
• Config: vite.config.mts → test.include: ['test/unit/**/*.spec.{ts,tsx}'], setupFiles: ['./test/setup.ts'].
• test/setup.ts registers jest-dom matchers and an afterEach(cleanup). Vitest globals are off: import describe/it/expect/vi from 'vitest', and don't add your own RTL cleanup — setup.ts owns it.

Layout

• Mirror the source path under test/unit/ (e.g. src/context/search/… → test/unit/context/…).
• Reusable harnesses/mocks go in test/unit/support/.

Test what you own, not your dependencies

Assert the logic and wiring that belong to this codebase — not behavior a dependency already guarantees. A test whose only assertion is that MUI Tabs switch tabs, react-router navigates, or TanStack caches is testing someone else's library; it earns nothing and breaks on their upgrades.

• Exercising that behavior incidentally, as a step in a flow that asserts your own logic, is good — click the tab, then assert the panel your code renders. What you don't do is write a dedicated test for the third-party behavior alone.
• When you catch yourself asserting a library did its job, move the assertion to the part you own: the props you passed it, the state/callback you wired, the value you derived from its output.

Patterns, simplest first

• Pure functions → call and assert. See test/unit/utils.spec.ts.
• Components → render + screen + jest-dom matchers (toBeInTheDocument, …). See test/unit/components/kbd-key.spec.tsx. Prefer @testing-library/user-event for real interaction over synthetic events.
• Anything that needs the app's providers (a TanStack hook reading service from MainContext, a component using search, keyboard shortcuts, the theme, or routing) → render it through the shared harness test/unit/support/test-providers.tsx: renderWithProviders(ui, { route, mainContext: { service } }) for components, renderHookWithProviders(hook, …) for hooks (the idiomatic way to read a hook's return via result.current), or withProviders(Component, opts) for the HOC form. The harness reuses the app's real AppProviders (so it tracks the app automatically — query client, MainContext, keyboard shortcuts, search) and adds the entry-shell bits AppProviders doesn't own (a fresh no-retry QueryClient, the MUI theme, a MemoryRouter). Reach for this first — don't hand-roll a provider stack per spec.
• Routing / history → don't mock it. test-providers mounts a real MemoryRouter; drive back/forward with the useNavigate() you pull from the hook under test (navigate(-1) / navigate(1) are POPs), and pass the starting URL via route.
• An external dependency whose timing or return you must control (a slow client, a flaky global) → vi.mock('pkg', …) it with a small controllable stand-in, kept generic and reusable in test/unit/support/. Reach for this only when the real thing genuinely can't be driven — prefer the real router/providers first.

Grow the harness, don't duplicate it

The suite gets easier to extend only if shared test code is treated like production code.

• Before writing a mock, stub, or render helper, check test/unit/support/ for one that already fits and reuse it.
• If an existing helper almost fits, extend or refactor it — kept generic and dependency-agnostic — rather than forking a near-duplicate. A second caller is the signal to generalize, not to copy.
• When an inline mock or helper in a spec would help another test, promote it to test/unit/support/ before it gets copied.
• Refactoring a shared harness is part of adding the test that needs it: update every call site and leave the whole suite green. Never let two helpers do the same job.
• Don't write tests for the test helpers themselves. A harness or mock in test/unit/support/ is exercised by the specs that use it; it never gets its own spec.

The fail-first check (required for regression tests)

A test that passes tells you nothing until you've seen it fail. After writing a test that guards a fix, temporarily break the code it protects, run the test, confirm it goes red, then restore. This is the only thing that proves the test is wired to the behavior — it matters most for timing/race fixes, where a naive test passes with or without the fix.

Run

• One file while iterating: yarn vitest run test/unit/<path>.spec.tsx.
• Whole suite: yarn test. Iterate on test or implementation until green, then yarn lint.