test: incremental unit test coverage (67.7% -> 83.7%)

[dkapanidis]

What

This is the first automated coverage run on this repo (no prior .coverage-ledger.json).

Config fix first: ui/vite.config.ts's test.coverage block had no include, so the v8 provider only reported files a test happened to import — 15 of 41 source files — and showed a misleading 99.5%. Added coverage.include: ['src/**/*.{ts,tsx}'] (with tests/setup/main.tsx excluded) so the report reflects the whole src/ tree. That alone dropped the honest starting baseline to ui/: 220/370 stmts (59.5%). Combined with the backend's unchanged 250/324 (77.2%), the true starting total was 470/694 = 67.7%.

New tests, following the repo's existing testify/vitest conventions:

• ui/src/actions/utils.test.ts — the fetch wrapper's get/post/put/delete, including the non-JSON error-body fallback path
• ui/src/hooks/use-mousetrap.test.ts — bind/unbind/rebind lifecycle, mocking the mousetrap package
• ui/src/shortcuts/KeyboardShortcuts.test.tsx — g h / g m navigate wiring
• ui/src/pages/movies/list.test.tsx — loading/error/empty/populated states and the create-dialog toolbar button, rendered against the real DataTable
• ui/src/routes.test.tsx — route param parsing for /movies/:id and /movies/:id/edit, exercised through the real router and SideMenu layout

Also added a ResizeObserver stub to ui/src/test/setup.ts — jsdom doesn't implement it, and SideMenu's react-resizable-panels needs it just to mount, so nothing rendering that layout could be tested before this.

Rendering list.tsx and routes.tsx against the real component tree (rather than mocking child components) transitively covered several previously-0% shadcn primitives they actually use: data-table.tsx, table.tsx, resizable.tsx, toast.tsx, toaster.tsx.

No production code was changed — only the coverage config and new/modified test files.

Coverage delta

	Statements	%
Backend (unchanged)	250/324	77.2%
UI before	220/370	59.5%
UI after	331/370	89.5%
Combined before	470/694	67.7%
Combined after	581/694	83.7%

+16.0pp against the +15pp target — met.

Measured with:

cd backend && go test -coverprofile=cov.out ./... && go tool cover -func=cov.out | tail -1
cd ui && yarn test:coverage

What was skipped, and why

• ui/src/components/ui/alert-dialog.tsx (27 stmts, 0%) is dead code — not imported anywhere in src/. Testing it would be padding; it should probably be deleted instead (left as-is here since that's a separate call).
• Backend cmd/movie-catalog/main.go's run() (44 stmts) blocks on a real net.Listener/OS signals; testing it meaningfully needs an injectable-listener refactor, which felt like a deliberate decision rather than a nightly one.
• Backend ui.go's "UI is built" branch — the existing ui_test.go already documents this as intentionally untested, since dist/ only ever holds a real build artifact in this repo, not a test fixture. Respected that convention rather than working around it.

Suggested next target

toast.tsx's dismiss/update/pause-on-hover branches and toaster.tsx's empty-state render need a test driving useToast() directly (5+1 stmts). App.tsx (2 stmts) just needs a smoke-render test of the real component. On the backend, main.go's run() is the largest remaining gap, pending the listener-injection refactor mentioned above.

Test plan

• cd backend && go test ./... — all pre-existing tests pass, unmodified
• cd ui && yarn test — 73/73 tests pass (8 pre-existing files + 5 new)
• cd ui && npx tsc -b --noEmit — type-checks clean

---

Generated by Claude Code

[github-actions]

Test coverage

Suite	Coverage
Backend (Go, statements)	77.2%
UI (lines)	89.1%