Ui testing patterns

UI Testing Patterns Across Mobile

Instructions

UI tests are the most expensive layer to run and to maintain. Pick the lightest tool that still covers the risk, write them in a page-object / screen-robot style, and keep selectors stable.

1. Pick the Right Tool

Tool	Scope	Runs On	Best For
Espresso / Compose UI test	Android, in-process	Emulator / device	Single-screen behavior, fast iteration
Robolectric + Compose	Android, JVM	Host JVM	Compose component tests without an emulator
XCUITest	iOS, out-of-process	Simulator / device	End-to-end iOS flows
ViewInspector / SwiftUI Previews + snapshot	iOS, in-process	Simulator	SwiftUI component tests
flutter_test (widget tests)	Flutter	Host Dart VM	Widget-level, fast
integration_test	Flutter	Emulator / device	Real-device smoke
Detox	React Native	Emulator / device	Full RN E2E in JS
Maestro	Any mobile app	Emulator / device / real	Black-box YAML smoke flows, cross-platform

Rule of thumb: prefer in-process component tests (Compose, SwiftUI views, flutter_test, RN Testing Library) for 80 % of UI coverage; reserve out-of-process (XCUITest, Detox, Maestro) for smoke and critical paths.

2. Selector Strategy

Tests break most often because someone changed a string or a layout. Use stable, semantic selectors:

• Android Views: R.id.* resource IDs.
• Android Compose: Modifier.testTag("checkoutPayButton") + onNodeWithTag.
• iOS UIKit: accessibilityIdentifier.
• iOS SwiftUI: .accessibilityIdentifier("checkoutPayButton").
• Flutter: Key('checkoutPayButton') / ValueKey.
• React Native: testID="checkoutPayButton" + getByTestId.

Never select by visible text unless the text is the thing under test (e.g., a localization check).

3. Page-Object / Screen-Robot Pattern

Wrap each screen in a test-only API so tests read like business steps, not framework calls.

Kotlin + Compose:

class CheckoutRobot(private val rule: ComposeTestRule) {
    fun enterCard(number: String) = apply {
        rule.onNodeWithTag("cardNumber").performTextInput(number)
    }
    fun tapPay() = apply { rule.onNodeWithTag("pay").performClick() }
    fun assertSuccess() = apply { rule.onNodeWithTag("success").assertIsDisplayed() }
}

@Test fun happyPath() = composeRule.run {
    CheckoutRobot(this).enterCard("4242...").tapPay().assertSuccess()
}

Swift + XCUITest:

struct CheckoutScreen {
    let app: XCUIApplication
    @discardableResult func enterCard(_ s: String) -> Self {
        app.textFields["cardNumber"].tap()
        app.textFields["cardNumber"].typeText(s); return self
    }
    @discardableResult func tapPay() -> Self { app.buttons["pay"].tap(); return self }
    func assertSuccess() { XCTAssertTrue(app.staticTexts["success"].waitForExistence(timeout: 5)) }
}

Dart + flutter_test:

class CheckoutRobot {
  CheckoutRobot(this.t);
  final WidgetTester t;
  Future<void> enterCard(String s) async {
    await t.enterText(find.byKey(const Key('cardNumber')), s);
  }
  Future<void> tapPay() async => t.tap(find.byKey(const Key('pay')));
  void assertSuccess() => expect(find.byKey(const Key('success')), findsOneWidget);
}

4. Waits — No sleep

Use the framework's own "wait until condition" primitive:

• Espresso — IdlingResource plus sensible onView(...).check(matches(...)) which blocks on idling.
• Compose — waitForIdle, waitUntil { composeRule.onAllNodesWithTag(...).fetchSemanticsNodes().isNotEmpty() }.
• XCUITest — waitForExistence(timeout:), XCTNSPredicateExpectation.
• Flutter — pumpAndSettle() (scoped timeout) or an explicit pump loop with a deadline.
• Detox — waitFor(element(...)).toBeVisible().withTimeout(5000).

Never Thread.sleep / Task.sleep / await Future.delayed in UI tests. Waits must be condition-based.

5. Isolate from the World

Swap the network layer with a recorded or mocked server for the whole suite; swap the clock; seed a known user in local storage during @Before. UI tests must not depend on staging.

6. Cross-Platform Comparison

• Detox vs Maestro: Detox is JS-first and lives with the RN codebase; it covers complex flows with code. Maestro is YAML-first and works on any stack — prefer Maestro for smoke, Detox for deep RN-specific flows.
• Flutter integration_test vs Maestro: integration_test gives Dart-level access and test semantics; Maestro is more robust on CI device farms and produces cleaner videos.

7. Failure Artifacts

Every UI test failure must produce: screenshot, view hierarchy / semantics dump, logs (logcat / os_log / Flutter debugDumpApp). Make that automatic in a @AfterEach / tearDown hook and upload from CI.

8. Checklist

• Selectors are test IDs / keys / accessibility identifiers — never raw visible text.
• Screens are wrapped in page objects / robots.
• No sleep-based waits anywhere.
• Network, time, and storage are isolated from real services.
• Failure produces screenshot + hierarchy + logs as CI artifacts.
• Each flow runs in under ~60 s; if longer, split it.
• Smoke vs regression suites are tagged and run on different cadences.