testing-unit（检测单位）

Unit Testing Patterns

Focused patterns for writing isolated, fast, maintainable unit tests. Covers test structure (AAA), parametrization, fixture management, HTTP mocking (MSW/VCR), and test data generation with factories.

Each category has individual rule files in rules/ loaded on-demand, plus reference material, checklists, and scaffolding scripts.

Core Principles (ALWAYS apply)

1. AAA structure: Every test MUST follow Arrange-Act-Assert. Use // Arrange, // Act, // Assert comments for clarity.
2. Parametrize, don't duplicate: Use test.each (TypeScript) or @pytest.mark.parametrize (Python) when testing multiple inputs. Never copy-paste the same test body with different values.
3. Fixture scoping matters: Use scope="function" (default) for mutable data. Use scope="module" or scope="session" ONLY for expensive read-only resources (DB engines, ML models). Mutable data with shared scope causes flaky tests.
4. Speed target: Each unit test should run under 100ms. If it's slower, you're likely hitting I/O — mock it.
5. Mock at the network level: Use MSW (TypeScript) or VCR.py (Python) to intercept HTTP at the network layer. Never mock fetch/axios/requests directly.

Quick Reference

Total: 8 rules across 3 categories, 4 references, 3 checklists, 1 example set, 3 scripts

Unit Test Structure

Core patterns for structuring isolated unit tests with clear phases and efficient execution.

Reference: references/aaa-pattern.md — detailed AAA implementation with checklist

HTTP Mocking

Network-level request interception for deterministic tests without hitting real APIs.

References:

• references/msw-2x-api.md — full MSW 2.x API (handlers, GraphQL, WebSocket, passthrough)
• references/stateful-testing.md — Hypothesis RuleBasedStateMachine for stateful tests

Checklists:

• checklists/msw-setup-checklist.md — MSW installation, handler setup, test writing
• checklists/vcr-checklist.md — VCR configuration, sensitive data filtering, CI setup

Examples: examples/handler-patterns.md — CRUD, error simulation, auth flow, file upload handlers

Test Data Management

Factories, fixtures, and seeding patterns for isolated, realistic test data.

Reference: references/factory-patterns.md — advanced factory patterns (Sequence, SubFactory, Traits)

Checklist: checklists/test-data-checklist.md — data generation, cleanup, isolation verification

Quick Start

TypeScript (Vitest + MSW)

import { describe, test, expect, beforeAll, afterEach, afterAll } from 'vitest';
import { http, HttpResponse } from 'msw';
import { setupServer } from 'msw/node';
import { calculateDiscount } from './pricing';

// 1. Pure unit test with AAA pattern
describe('calculateDiscount', () => {
  test.each([
    [100, 0],
    [150, 15],
    [200, 20],
  ])('for order $%i returns $%i discount', (total, expected) => {
    // Arrange
    const order = { total };

    // Act
    const discount = calculateDiscount(order);

    // Assert
    expect(discount).toBe(expected);
  });
});

// 2. MSW mocked API test
const server = setupServer(
  http.get('/api/users/:id', ({ params }) => {
    return HttpResponse.json({ id: params.id, name: 'Test User' });
  })
);

beforeAll(() => server.listen({ onUnhandledRequest: 'error' }));
afterEach(() => server.resetHandlers());
afterAll(() => server.close());

test('fetches user from API', async () => {
  // Arrange — MSW handler set up above

  // Act
  const response = await fetch('/api/users/123');
  const data = await response.json();

  // Assert
  expect(data.name).toBe('Test User');
});

Python (pytest + FactoryBoy)

import pytest
from factory import Factory, Faker, SubFactory

class UserFactory(Factory):
    class Meta:
        model = dict
    email = Faker('email')
    name = Faker('name')

class TestUserService:
    @pytest.mark.parametrize("role,can_edit", [
        ("admin", True),
        ("viewer", False),
    ])
    def test_edit_permission(self, role, can_edit):
        # Arrange
        user = UserFactory(role=role)

        # Act
        result = user_can_edit(user)

        # Assert
        assert result == can_edit

Vitest 4.1 Features

aroundEach / aroundAll (preferred for DB transactions)

Wraps each test in setup/teardown — cleaner than separate beforeEach/afterEach for transactions:

test.aroundEach(async (runTest, { db }) => {
  await db.transaction(runTest)  // auto-rollback on test end
})

test('insert user', async ({ db }) => {
  await db.insert({ name: 'Alice' })
  // transaction auto-rolls back — no cleanup needed
})

aroundAll wraps entire suites the same way.

mockThrow / mockThrowOnce

Replaces the verbose mockImplementation(() => {throw err}) pattern:

const fn = vi.fn()
fn.mockThrow(new Error('connection lost'))  // always throws
fn.mockThrowOnce(new Error('timeout'))      // throws once, then normal

vi.defineHelper (clean stack traces)

Custom assertion helpers that point errors to the call site, not the helper internals:

const assertPair = vi.defineHelper((a, b) => {
  expect(a).toEqual(b)  // error points to where assertPair() was CALLED
})

Test Tags

Filter tests by tags in CLI — useful for CI fast paths:

// vitest.config.ts
test: {
  tags: {
    unit: { timeout: 5000 },
    flaky: { retry: 3 },
  }
}

vitest --tags-filter="unit and !flaky"
vitest --tags-filter="(unit or integration) and !slow"

Agent Reporter

Minimal output (failures only) — use in AI agent / CI contexts:

AI_AGENT=copilot vitest    # auto-detect agent mode

Key Decisions

Common Mistakes

1. Testing implementation details instead of public behavior (brittle tests)
2. Mocking fetch/axios directly instead of using MSW at network level (incomplete coverage)
3. Shared mutable state between tests via module-scoped fixtures (flaky tests)
4. Hard-coded test data with duplicate IDs (test conflicts in parallel runs)
5. No cleanup after database seeding (state leaks between tests)
6. Over-mocking — testing your mocks instead of your code (false confidence)
7. Verbose throw mocking — mockImplementation(() => {throw err}) instead of mockThrow(err) (Vitest 4.1+)

Scripts