e2e-test-conventions（e2e 测试约定）

E2E Test Conventions

These conventions apply to all Playwright E2E test code. Read and follow them whenever generating, modifying, or reviewing tests.

Technology Stack

• Playwright for browser automation and test running
• TypeScript for all test code (no plain JavaScript)
• Framework-agnostic — adapt selectors and helpers to whatever UI framework the project uses

Project Structure

The e2e/ directory follows this layout:

e2e/
├── auth/                    # Authentication setup (runs before all tests)
│   └── auth.setup.ts        # Logs in and saves session state
├── fixtures/                # Custom Playwright fixtures
│   └── base.ts              # Extends Playwright's test with custom fixtures
├── helpers/                 # Shared utility functions (NOT page objects)
│   └── env-config.ts        # Environment resolution
├── poms/                    # Page Object Models (one file per page)
│   └── base.page.ts         # Abstract base class — all POMs extend this
├── test-data/               # External test data (JSON files)
├── tests/                   # Test specs organised by suite
│   ├── handover/            # Ticket-driven handover tests (temporary)
│   ├── regression/          # Full regression tests (permanent)
│   └── smoke/               # Quick critical-path checks
├── playwright.config.ts
├── tsconfig.json
└── .env.example             # Template for environment variables

Rules:

• Do NOT create files outside this structure
• Do NOT move existing files without explicit permission

See references/folder-structure.md for details on each directory.

Browser × Suite Configuration

The Playwright config defines projects using {browser}:{suite} naming:

Browsers: chromium, firefox, webkit, mobile-chrome, mobile-safari

Suites:

Every project depends on the setup project which handles authentication.

Running Tests

# From e2e/ directory
npx playwright test                                    # All browsers × all suites
npx playwright test --project="chromium:regression"    # Single project
npx playwright test --project="*:smoke"                # One suite, all browsers
npx playwright test --project="firefox:*"              # One browser, all suites

Naming Conventions

The {feature} name must match across POM, spec, and test-data files.

Test Independence and Parallelism

Every Test Must Be Fully Independent

• Each test runs in isolation and in any order
• A test must never depend on state created by a previous test
• Each test navigates to its page via the POM's navigation method

Parallel Safety

All tests run in parallel across multiple workers. To avoid collisions:

• Append unique suffixes (timestamp + random ID) to all test data values
• Never share mutable state between tests
• Each test creates its own data, verifies it, and cleans it up

Authentication

Authentication is performed once in a setup project (auth/auth.setup.ts) that runs before all test projects. The session state is saved to a file and reused via Playwright's storageState configuration.

NEVER write login logic in specs, POMs, beforeEach, or beforeAll. Authentication is already handled.

If the application stores auth tokens in IndexedDB (Firebase Auth, Supabase, AWS Amplify, etc.), use the indexedDB option:

await page.context().storageState({ path: authFile, indexedDB: true });

See references/auth-setup.md for the full pattern.

Navigation

• Always use direct URL navigation (page.goto('/dashboard'))
• Do NOT click through menus or sidebars — menu state and animations cause flaky tests
• After navigating, assert the URL and a key heading/element are visible

BasePage vs Derived POM Methods

BasePage is the single home for reusable helper methods that are useful across multiple pages. Derived POMs should stay focused on page-specific behavior only.

When a method belongs in BasePage

Move a helper to BasePage when it is:

• Reusable across multiple POMs — e.g., dismissing a modal, waiting for a toast notification, or checking a loading spinner
• Not tightly coupled to a single page — it works the same regardless of which page is active
• Generic enough to be inherited cleanly — no page-specific selectors or assumptions
• Likely to be duplicated if left in a feature-specific POM

Examples of BasePage helpers:

• waitForToast(message) — waits for and verifies a toast notification
• dismissModal() — closes any open modal dialog
• waitForLoadingComplete() — waits for a loading spinner to disappear
• getTableRowCount() — counts rows in a data table present on many pages

When a method belongs in a derived POM

Keep a helper in the specific POM when it is:

• Unique to one page — e.g., filling a page-specific form
• Dependent on page-specific structure — uses selectors that only exist on that page
• Not expected to be reused elsewhere

Why this matters

• Reduces duplication — shared logic lives in one place instead of being copied across POMs
• Centralises shared behavior — updates to a common helper propagate to all POMs automatically
• Keeps derived POMs small and focused — each POM only contains what is specific to its page
• Improves discoverability — developers know to look at BasePage for shared utilities

Rule

If you find yourself writing the same helper in a second POM, promote it to BasePage immediately. Do not leave duplicate helpers scattered across feature POMs.

Selectors and Locator Strategy

Use this priority order:

1. getByRole() — buttons, headings, dialogs (most resilient)
2. getByLabel() — form fields
3. getByText() — visible text content
4. getByPlaceholder() — input placeholders
5. locator() with CSS / filter() — last resort

Tips:

• Dynamically rendered attributes (tooltips, popovers) may not be in the DOM at query time — use filter() to match child content
• Icon buttons often lack visible text — match by aria-label or child icon content
• Prefer role-based and label-based selectors over CSS classes (brittle and framework-specific)

See references/selector-priority.md for examples.

Environment Configuration

Each environment has its own .env.{env} file in the e2e/ directory:

Every file uses the same variable names — only the values differ:

BASE_URL="https://your-app.example.com"
LOGIN_EMAIL="test-user@example.com"
LOGIN_PASSWORD="your-password"
AUTH_FILE="e2e/.auth/user.json"

Selecting the Active Environment

TEST_ENV is required. If it is not set, env-config.ts throws immediately — the test run will not start. This prevents accidental E2E runs against production when someone forgets to set the variable.

TEST_ENV=dev npx playwright test          # loads .env then .env.dev
TEST_ENV=production npx playwright test    # loads .env then .env.production
npx playwright test                        # ❌ ERROR — TEST_ENV is not set

Two-Layer Loading

helpers/env-config.ts reads TEST_ENV and loads environment variables in two layers via dotenv:

1. e2e/.env — base file (can hold TEST_ENV and all variables)
2. e2e/.env.{env} — optional environment-specific override

Both files are optional. If neither exists the process relies on variables already present in the environment (e.g. injected by CI or a container). dotenv never overwrites a variable that is already set, so CLI exports and CI-injected values always win.

The module exports helper functions — getEnvConfig(), getBaseUrl(), getCredentials(), and getAuthFilePath() — instead of a static constant. Loading runs once per process; subsequent calls are no-ops.

Rules

• TEST_ENV is required — missing it throws an error so E2E runs never silently target production
• All variables (BASE_URL, LOGIN_EMAIL, LOGIN_PASSWORD, AUTH_FILE) are required — missing ones throw an error
• Never fall back to hardcoded defaults for required environment values; throw an error if they are missing
• .env.* files contain secrets and are git-ignored
• .env.example is the only env file committed — it serves as the template

Test Data

• All test data lives in e2e/test-data/{feature}.json
• Never hardcode data in spec or POM files
• Use a custom fixture to load and stamp test data with unique suffixes
• Always import test from your custom fixtures, not from @playwright/test

Spec File Rules

• All page interaction goes through POMs — never call page.getByRole(), page.locator(), etc. directly in specs
• The only acceptable uses of page in a spec: passing to POM constructor, or creating contexts in beforeAll
• Use test.beforeAll with a manually created browser context to call POM setUp() for cleanup
• Each test navigates independently via POM navigation methods