flutter-widget-tests

name	flutter-widget-tests
description	Use when writing or modifying Flutter widget tests under frontend/test/ in this project - covers BDD naming, pump helpers, fake-over-mock, interaction-over-static-render, router testing, i18n handling, and when to prefer a widget test over a controller/notifier test.

Flutter Widget Tests (this project)

How this project writes widget tests. Follow these when adding or changing tests under frontend/test/.

Core principle

Test behavior through the UI, not controller internals. A widget test that taps "delete" and asserts the repo no longer contains the item is worth more than a controller test asserting the same state transition — and it covers the same lines of controller code for free.

BDD naming

Every test name reads given X, when Y, then Z. Helpers mirror the structure:

testWidgets(
  'given a trip with entries, when the detail screen loads, '
  'then the trip name and each timeline entry are rendered',
  (tester) async {
    await _givenTripDetailScreen(tester, ...);
    _thenTripNameIsVisible('Kyoto');
    _thenAtLeastOneTimelineEntryIsShown();
  },
);

Private helpers: _givenXScreen(), _whenUserTapsY(), _thenZIsVisible().

Merge tests sharing same Given/When

If two sibling tests have identical Given and When and only differ in Then, merge them into one test with multiple assertions. Three static-render tests that all skip interaction and check different parts of the initial view should be one test.

Use project helpers, not raw pumpWidget

• pumpScreen(tester, child: ..., overrides: [...]) — simple screen, no routing
• pumpRouterApp(tester, routes: [...], overrides: [...]) — whenever the screen calls context.push / pushNamed / context.pop

Both live in test/helpers/pump_app.dart and already bootstrap EasyLocalization, SharedPreferences mocks, and ProviderScope.

Fakes, not mocks

Never use mocktail for repositories or services. Use the existing fakes under test/fakes/:

• In-memory repos: InMemoryJourneyRepository, InMemoryTripRepository, InMemoryUsageRepository, InMemoryQuickGuideRepository, InMemorySavedLocationsRepository, ...
• Fake services: FakeImagePickerService, FakeNarrationService, FakeQuickGuideAiService, FakeImageAnalysisService, FakeLocationService, FakePlacesRepository, FakeRewardedAdService

If you need to assert "was this called?" — extend the fake with a counter or captured-args field (see FakePlacesRepository.nearbyCallCount, FakeImagePickerService.pickCount). Don't reach for a mock.

Cover interactions, not just static render

Interaction paths that screen tests commonly miss:

• Dialogs and bottom sheets (default hidden, need user action to open)
• FloatingActionButton callbacks (rendered but onPressed never fires)
• Multi-select / edit mode entry + batch actions
• Error / retry / empty / loading states
• Off-screen-render widgets used by services (capture/export flows)

Checklist for a screen test: tap, onPressed, onTap, showDialog, showModalBottomSheet, error state, empty state. If none appear in the file, interaction coverage is missing.

Prefer widget test over controller/notifier test

Controller tests usually cover the same lines widget tests already hit via the UI. Delete the controller test unless it covers one of these non-UI contracts:

• SharedPreferences or other persistence
• Error-type mapping branches with no visible difference
• State machine transitions the UI can't observe

For everything else, cover the behavior with a widget test that asserts the UI outcome (or the underlying repo's state after the interaction).

Router tests

Replace destination routes with stub Scaffolds that capture state.extra — assert on what was pushed rather than navigating through real screens:

final extras = <Object?>[];
await pumpRouterApp(tester, routes: [
  GoRoute(path: '/', builder: (_, __) => const MyScreen()),
  GoRoute(name: 'player', path: '/player', builder: (_, state) {
    extras.add(state.extra);
    return const Scaffold(key: Key('player-screen'));
  }),
], overrides: [...]);
// ... trigger action ...
expect(extras.single, isA<Map<String, dynamic>>());

If the test exercises context.pop(), start at / and push to the screen under test — popping from the initialLocation leaves an empty stack.

i18n inside tests

pumpScreen wires _EmptyAssetLoader, so tr() returns the raw key — assert on 'passport.title', 'camera.take_photo', etc. If the raw key causes a RenderFlex overflowed error in a fixed-width widget, write a local _MapAssetLoader that returns short Chinese/English strings (see journey_sharing_card_test.dart).

Image.memory needs valid bytes

Invalid bytes trigger a codec exception that marks the test failed even if assertions pass. Copy the 1x1 transparent PNG helper from journey_sharing_card_test.dart / select_narration_aspect_screen_test.dart when seeding imageBytes.

Common gotchas

• Chips in SingleChildScrollView: call tester.ensureVisible(chipFinder) before tap — they may be clipped off-screen on the default 800×600 test surface.
• pump vs pumpAndSettle: use pumpAndSettle() after interactions that chain async work (provider invalidation, navigation, dialog animations). Use pump(Duration) to step through controlled animations.
• Static test state: if a test uses static capture fields (_Host.lastSelection), reset them in setUp() — not just at the end of the test body.
• Search-field clear suffix: on fields where the suffix appears conditionally on controller.text.isNotEmpty, the widget only rebuilds on onChanged / submit. You may need to submit once before the clear icon exists.
• Current-trip / prefs tests: SharedPreferences.setMockInitialValues({...}) BEFORE pumpScreen to seed persisted state, then don't override the notifier — test the real hydration path.

When in doubt

Look at an existing test in the same feature. The patterns above were extracted from the actual test corpus; consistency with neighbors is usually the right call.