writing-tests（编写测试）

Writing Tests

How to write tests that catch bugs, document behavior, and remain maintainable.

Based on BugMagnet by Gojko Adzic. Adapted with attribution.

Critical Rules

🚨 Test names describe outcomes, not actions. "returns empty array when input is null" not "test null input". The name IS the specification.

🚨 Assertions must match test titles. If the test claims to verify "different IDs", assert on the actual ID values—not just count or existence.

🚨 Assert specific values, not types. expect(result).toEqual(['First.', ' Second.']) not expect(result).toBeDefined(). Specific assertions catch specific bugs.

🚨 One concept per test. Each test verifies one behavior. If you need "and" in your test name, split it.

🚨 Bugs cluster together. When you find one bug, test related scenarios. The same misunderstanding often causes multiple failures.

When This Applies

• Writing new tests
• Reviewing test quality
• During TDD RED phase (writing the failing test)
• Expanding test coverage
• Investigating discovered bugs

Test Naming

Pattern: [outcome] when [condition]

Good Names (Describe Outcomes)

returns empty array when input is null
throws ValidationError when email format invalid
calculates tax correctly for tax-exempt items
preserves original order when duplicates removed

Bad Names (Describe Actions)

test null input           // What about null input?
should work               // What does "work" mean?
handles edge cases        // Which edge cases?
email validation test     // What's being validated?

The Specification Test

Your test name should read like a specification. If someone reads ONLY the test names, they should understand the complete behavior of the system.

Assertion Best Practices

Assert Specific Values

// ❌ WEAK - passes even if completely wrong data
expect(result).toBeDefined()
expect(result.items).toHaveLength(2)
expect(user).toBeTruthy()

// ✅ STRONG - catches actual bugs
expect(result).toEqual({ status: 'success', items: ['a', 'b'] })
expect(user.email).toBe('test@example.com')

Match Assertions to Test Title

// ❌ TEST SAYS "different IDs" BUT ASSERTS COUNT
it('generates different IDs for each call', () => {
  const id1 = generateId()
  const id2 = generateId()
  expect([id1, id2]).toHaveLength(2)  // WRONG: doesn't check they're different!
})

// ✅ ACTUALLY VERIFIES DIFFERENT IDs
it('generates different IDs for each call', () => {
  const id1 = generateId()
  const id2 = generateId()
  expect(id1).not.toBe(id2)  // RIGHT: verifies the claim
})

Avoid Implementation Coupling

// ❌ BRITTLE - tests implementation details
expect(mockDatabase.query).toHaveBeenCalledWith('SELECT * FROM users WHERE id = 1')

// ✅ FLEXIBLE - tests behavior
expect(result.user.name).toBe('Alice')

Test Structure

Arrange-Act-Assert

it('calculates total with tax for non-exempt items', () => {
  // Arrange: Set up test data
  const item = { price: 100, taxExempt: false }
  const taxRate = 0.1

  // Act: Execute the behavior
  const total = calculateTotal(item, taxRate)

  // Assert: Verify the outcome
  expect(total).toBe(110)
})

One Concept Per Test

// ❌ MULTIPLE CONCEPTS - hard to diagnose failures
it('validates and processes order', () => {
  expect(validate(order)).toBe(true)
  expect(process(order).status).toBe('complete')
  expect(sendEmail).toHaveBeenCalled()
})

// ✅ SINGLE CONCEPT - clear failures
it('accepts valid orders', () => {
  expect(validate(validOrder)).toBe(true)
})

it('rejects orders with negative quantities', () => {
  expect(validate(negativeQuantityOrder)).toBe(false)
})

it('sends confirmation email after processing', () => {
  process(order)
  expect(sendEmail).toHaveBeenCalledWith(order.customerEmail)
})

Edge Case Checklists

When testing a function, systematically consider these edge cases based on input types.

Numbers

• Zero
• Negative numbers
• Very large numbers (near MAX_SAFE_INTEGER)
• Very small numbers (near MIN_SAFE_INTEGER)
• Decimal precision (0.1 + 0.2)
• NaN
• Infinity / -Infinity
• Boundary values (off-by-one at limits)

Strings

• Empty string ""
• Whitespace only " "
• Very long strings (10K+ characters)
• Unicode: emojis 👨‍👩‍👧‍👦, RTL text, combining characters
• Special characters: quotes, backslashes, null bytes
• SQL/HTML/script injection patterns
• Leading/trailing whitespace
• Mixed case sensitivity

Collections (Arrays, Objects, Maps)

• Empty collection [], {}
• Single element
• Duplicates
• Nested structures
• Circular references
• Very large collections (performance)
• Sparse arrays
• Mixed types in arrays

Dates and Times

• Leap years (Feb 29)
• Daylight saving transitions
• Timezone boundaries
• Midnight (00:00:00)
• End of day (23:59:59)
• Year boundaries (Dec 31 → Jan 1)
• Invalid dates (Feb 30, Month 13)
• Unix epoch edge cases
• Far future/past dates

Null and Undefined

• null input
• undefined input
• Missing optional properties
• Explicit undefined vs missing key

Domain-Specific

• Email: valid formats, edge cases (plus signs, subdomains)
• URLs: protocols, ports, special characters, relative paths
• Phone numbers: international formats, extensions
• Addresses: Unicode, multi-line, missing components
• Currency: rounding, different currencies, zero amounts
• Percentages: 0%, 100%, over 100%

Violated Domain Constraints

These test implicit assumptions in your domain:

• Uniqueness violations (duplicate IDs, emails)
• Missing required relationships (orphaned records)
• Ordering violations (events out of sequence)
• Range breaches (age -1, quantity 1000000)
• State inconsistencies (shipped but not paid)
• Format mismatches (expected JSON, got XML)
• Temporal ordering (end before start)

Typed Property Validation

When testing code that validates properties against type constraints (e.g., validating route: string in an interface):

Wrong-type literals:

• Numeric literal when string expected (route = 123)
• Boolean literal when string expected (route = true)
• String literal when number expected (count = 'five')
• String literal when boolean expected (enabled = 'yes')

Non-literal expressions:

• Template literal (` route = /path/${id} `)
• Variable reference (route = someVariable)
• Function call (route = getRoute())
• Computed property (route = config.path)

Correct type:

• Valid literal of correct type (route = '/orders')
• Edge values (empty string '', zero 0, false)

Why this matters: A common bug pattern is validating "is this a literal?" without checking "is this the RIGHT TYPE of literal?"

• hasLiteralValue() returns true for 123, true, and 'string'
• hasStringLiteralValue() returns true only for 'string'

When an interface specifies property: string, validation must reject numeric and boolean literals, not just non-literal expressions.

Bug Clustering

When you discover a bug, don't stop—explore related scenarios:

1. Same function, similar inputs - If null fails, test undefined, empty string
2. Same pattern, different locations - If one endpoint mishandles auth, check others
3. Same developer assumption - If off-by-one here, check other boundaries
4. Same data type - If dates fail at DST, check other time edge cases

When Tempted to Cut Corners

• If your test name says "test" or "should work": STOP. What outcome are you actually verifying? Name it specifically.
• If you're asserting toBeDefined() or toBeTruthy(): STOP. What value do you actually expect? Assert that instead.
• If your assertion doesn't match your test title: STOP. Either fix the assertion or rename the test. They must agree.
• If you're testing multiple concepts in one test: STOP. Split it. Future you debugging a failure will thank you.
• If you found a bug and wrote one test: STOP. Bugs cluster. What related scenarios might have the same problem?
• If you're skipping edge cases because "that won't happen": STOP. It will happen. In production. At 3 AM.

Integration with Other Skills

With TDD Process: This skill guides the RED phase—how to write the failing test well.

With Software Design Principles: Testable code follows design principles. Hard-to-test code often has design problems.