How do you write reliable widget tests?

A widget test runs the real framework in a test environment. It can build, lay out and interact with widgets without a device. The tester controls time through pump and can enter text, tap and scroll. Anything that reaches the network, platform channels or real storage must be replaced by a fake.

testWidgets('shows error for invalid email', (tester) async {
  await tester.pumpWidget(const MaterialApp(home: SignupForm()));
  await tester.enterText(find.byKey(const Key('email')), 'not-an-email');
  await tester.tap(find.byKey(const Key('submit')));
  await tester.pump();

  expect(find.text('Enter a valid email'), findsOneWidget);
});

• Use keys for interactive targets so tests do not depend on visual text
• pumpAndSettle waits for animations, but it can time out on infinite animations
• Use pump with an explicit duration when testing a specific animation frame
• Test loading, empty, error and success states, not only the happy path
• Check semantics with SemanticsTester for accessibility-critical flows

Common flakiness

Tests become flaky when they depend on real timers, unawaited futures or plugin calls. FakeAsync and pump control time. If a widget starts a request in initState, inject a fake repository through the widget or an inherited provider. Avoid asserting on transient frames that a later rebuild replaces.