test-driven-development（测试驱动开发）

Test-Driven Development

Write test first. Watch it fail. Write minimal code to pass. Refactor.

Core principle: If you didn't watch the test fail, you don't know if it tests the right thing.

The Iron Law

NO BEHAVIOR-CHANGING PRODUCTION CODE WITHOUT A FAILING TEST FIRST

Wrote code before test? Delete it completely. Implement fresh from tests.

Refactoring is exempt: The refactor step changes structure, not behavior. Tests stay green throughout. No new failing test required.

Red-Green-Refactor Cycle

RED ──► Verify Fail ──► GREEN ──► Verify Pass ──► REFACTOR ──► Verify Pass ──► Next RED
         │                         │                            │
         ▼                         ▼                            ▼
      Wrong failure?           Still failing?              Broke tests?
      Fix test, retry          Fix code, retry             Fix, retry

RED - Write Failing Test

Write one minimal test for one behavior.

Good example:

test('retries failed operations 3 times', async () => {
  let attempts = 0;
  const operation = async () => {
    attempts++;
    if (attempts < 3) throw new Error('fail');
    return 'success';
  };

  const result = await retryOperation(operation);

  expect(result).toBe('success');
  expect(attempts).toBe(3);
});

*Clear name, tests real behavior, asserts observable outcome*

Bad example:

test('retry works', async () => {
  const mock = jest.fn()
    .mockRejectedValueOnce(new Error())
    .mockRejectedValueOnce(new Error())
    .mockResolvedValueOnce('success');
  await retryOperation(mock);
  expect(mock).toHaveBeenCalledTimes(3);
});

*Vague name, asserts only call count without verifying outcome, tests mock mechanics not behavior*

Requirements: One behavior. Clear name. Real code (mocks only if unavoidable).

Verify RED - Watch It Fail

MANDATORY. Never skip.

npm test path/to/test.test.ts

Test must go red for the right reason. Acceptable RED states:

• Assertion failure (expected behavior missing)
• Compile/type error (function doesn't exist yet)

Not acceptable: Runtime setup errors, import failures, environment issues.

Test passes immediately? You're testing existing behavior—fix test. Test errors for wrong reason? Fix error, re-run until it fails correctly.

GREEN - Minimal Code

Write simplest code to pass the test.

Good example:

async function retryOperation<T>(fn: () => Promise<T>): Promise<T> {
  for (let i = 0; i < 3; i++) {
    try {
      return await fn();
    } catch (e) {
      if (i === 2) throw e;
    }
  }
  throw new Error('unreachable');
}

*Just enough to pass*

Bad example:

async function retryOperation<T>(
  fn: () => Promise<T>,
  options?: { maxRetries?: number; backoff?: 'linear' | 'exponential'; }
): Promise<T> { /* YAGNI */ }

*Over-engineered beyond test requirements*

Write only what the test demands. No extra features, no "improvements."

Verify GREEN - Watch It Pass

MANDATORY.

npm test path/to/test.test.ts

Confirm: Test passes. All other tests still pass. Output pristine (no errors, warnings).

Test fails? Fix code, not test. Other tests fail? Fix now before continuing.

REFACTOR - Clean Up

After green only: Remove duplication. Improve names. Extract helpers.

Keep tests green throughout. Add no new behavior.

Repeat

Next failing test for next behavior.

Good Tests

Minimal: One thing per test. "and" in name? Split it. ❌ test('validates email and domain and whitespace')

Clear: Name describes behavior. ❌ test('test1')

Shows intent: Demonstrates desired API usage, not implementation details.

Example: Bug Fix

Bug: Empty email accepted

RED:

test('rejects empty email', async () => {
  const result = await submitForm({ email: '' });
  expect(result.error).toBe('Email required');
});

Verify RED:

$ npm test
FAIL: expected 'Email required', got undefined

GREEN:

function submitForm(data: FormData) {
  if (!data.email?.trim()) {
    return { error: 'Email required' };
  }
  // ...
}

Verify GREEN:

$ npm test
PASS

REFACTOR: Extract validation helper if pattern repeats.

Red Flags - STOP and Start Over

Any of these means delete code and restart with TDD:

• Code written before test
• Test passes immediately (testing existing behavior)
• Can't explain why test failed
• Rationalizing "just this once" or "this is different"
• Keeping code "as reference" while writing tests
• Claiming "tests after achieve the same purpose"

When Stuck

Legacy Code (No Existing Tests)

The Iron Law ("delete and restart") applies to new code you wrote without tests. For inherited code with no tests, use characterization tests:

1. Write tests that capture current behavior (even if "wrong")
2. Run tests, observe actual outputs
3. Update assertions to match reality (these are "golden masters")
4. Now you have a safety net for refactoring
5. Apply TDD for new behavior changes

Characterization tests lock down existing behavior so you can refactor safely. They're the on-ramp, not a permanent state.

Flakiness Rules

Tests must be deterministic. Ban these in unit tests:

• Real sleeps / delays → Use fake timers (vi.useFakeTimers(), jest.useFakeTimers())
• Wall clock time → Inject clock, assert against injected time
• Math.random() → Seed or inject RNG
• Network calls → Mock at boundary or use MSW
• Filesystem race conditions → Use temp dirs with unique names

Flaky test? Fix or delete. Flaky tests erode trust in the entire suite.

Debugging Integration

Bug found? Write failing test reproducing it first. Then follow TDD cycle. Test proves fix and prevents regression.

Planning: Test List

Before diving into the cycle, spend 2 minutes listing the next 3-10 tests you expect to write. This prevents local-optimum design where early tests paint you into a corner.

Example test list for a retry function:

• retries N times on failure
• returns result on success
• throws after max retries exhausted
• calls onRetry callback between attempts
• respects backoff delay

Work through the list in order. Add/remove tests as you learn.

Testing Anti-Patterns

When writing tests involving mocks, dependencies, or test utilities: See references/testing-anti-patterns.md for common pitfalls including testing mock behavior and adding test-only methods to production classes.

Philosophy and Rationalizations

For detailed rebuttals to common objections ("I'll test after", "deleting work is wasteful", "TDD is dogmatic"): See references/tdd-philosophy.md

Final Rule

Production code exists → test existed first and failed first
Otherwise → not TDD