intermediate⏱55 minutesLesson 3 of 9

Unit Testing Best Practices

Master the Arrange-Act-Assert pattern, mocking strategies, test isolation, and writing maintainable unit tests

Unit Testing Best Practices

Writing tests that pass is easy. Writing tests that are maintainable, reliable, and trustworthy is a craft. This lesson covers the patterns and practices that separate great test suites from mediocre ones.

The AAA Pattern: Arrange-Act-Assert

Every test should follow three clear phases, separated by blank lines:

python

# AAA Pattern in action
def test_withdraw_reduces_balance():
    # Arrange
    account = BankAccount("Alice", 1000.0)

    # Act
    account.withdraw(300.0)

    # Assert
    assert account.balance == 700.0

# Poor test — phases are mixed together
def test_withdraw():
    account = BankAccount("Alice", 1000.0)
    account.withdraw(300.0)
    assert account.balance == 700.0
    account.withdraw(100.0)
    assert account.balance == 500.0

Why AAA Matters

Aspect	With AAA	Without AAA
Readability	Clear story: Given → When → Then	Multiple scenarios blurred together
Debugging	Pinpoint which phase failed	Unclear if Arrange or Act is wrong
Maintenance	Change one phase independently	Changing input forces rewriting the test
Reviewability	Reviewer understands intent quickly	Reviewer must parse the entire test

python

# Good: One clear scenario per test
def test_deposit_increases_balance():
    """Given an account with $0, when I deposit $100, balance becomes $100."""
    account = BankAccount("Bob")
    account.deposit(100.0)
    assert account.balance == 100.0

def test_deposit_negative_amount_raises_error():
    """Given an account, when I deposit a negative amount, an error is raised."""
    account = BankAccount("Bob", 500.0)
    with pytest.raises(ValueError, match="Amount must be positive"):
        account.deposit(-50.0)

def test_deposit_multiple_times_accumulates():
    """Given an account, when I deposit multiple times, balance accumulates."""
    account = BankAccount("Bob")
    account.deposit(100.0)
    account.deposit(50.0)
    account.deposit(25.0)
    assert account.balance == 175.0

💡Tip

If your test doesn't fit the AAA pattern naturally, it's a sign that the function under test is doing too much. Consider refactoring the production code.

One Assertion Per Test? Not Quite

A common myth: "one assertion per test." The real rule is one concept per test:

python

# BAD: Testing multiple unrelated behaviors
def test_account_operations():
    account = BankAccount("Alice", 1000.0)
    account.deposit(500.0)
    assert account.balance == 1500.0
    account.withdraw(200.0)
    assert account.balance == 1300.0
    assert account.owner == "Alice"

# GOOD: Group related assertions that test one behavior
def test_account_after_deposit():
    account = BankAccount("Alice", 1000.0)
    account.deposit(500.0)
    assert account.balance == 1500.0
    assert account.transaction_count == 1

# GOOD: Multiple assertions all confirming the same result
def test_user_creation_sets_all_fields():
    user = User(name="Alice", email="alice@test.com", age=30)
    assert user.name == "Alice"
    assert user.email == "alice@test.com"
    assert user.age == 30
    assert user.is_active is True

Test Naming: Tell a Story

python

# Descriptive test names
def test_given_empty_cart_when_add_item_then_item_count_is_one():
    cart = ShoppingCart()
    cart.add_item("apple", 1.50)
    assert cart.total_items() == 1

def test_given_item_in_cart_when_remove_item_then_item_count_is_zero():
    cart = ShoppingCart()
    cart.add_item("apple", 1.50)
    cart.remove_item("apple")
    assert cart.total_items() == 0

def test_given_expired_coupon_when_apply_discount_then_raises_error():
    cart = ShoppingCart()
    coupon = Coupon("SAVE10", expires_at=yesterday)
    with pytest.raises(CouponExpiredError):
        cart.apply_coupon(coupon)

Naming Style	Example
`test_[feature]`	`test_withdraw()`
`test_[scenario]_[expected]`	`test_overdraft_raises_error()`
`test_given_[context]_when_[action]_then_[result]`	`test_given_insufficient_funds_when_withdraw_then_error()`
`test_[method]_[condition]_[result]`	`test_divide_by_zero_raises_exception()`

✅Success

A good test name eliminates the need to read the test body. When a CI build fails, you should know roughly what broke just from the test name.

Mocking: Isolating the Unit Under Test

Mocking replaces real dependencies with controlled substitutes:

python

from unittest.mock import Mock, patch, MagicMock
import pytest

# Example: Payment processing
class PaymentProcessor:
    def __init__(self, gateway):
        self.gateway = gateway

    def process_payment(self, user_id: int, amount: float) -> dict:
        if amount <= 0:
            raise ValueError("Amount must be positive")

        user = self.gateway.get_user(user_id)
        if not user["active"]:
            raise RuntimeError("User is not active")

        charge = self.gateway.charge(user["card_token"], amount)
        return {"status": "success", "transaction_id": charge["id"]}

# Test with Mock
def test_process_payment_success():
    # Arrange
    mock_gateway = Mock()
    mock_gateway.get_user.return_value = {
        "id": 1, "active": True, "card_token": "tok_123"
    }
    mock_gateway.charge.return_value = {
        "id": "txn_abc", "amount": 50.0, "status": "captured"
    }

    processor = PaymentProcessor(mock_gateway)

    # Act
    result = processor.process_payment(1, 50.0)

    # Assert
    assert result["status"] == "success"
    assert result["transaction_id"] == "txn_abc"
    mock_gateway.get_user.assert_called_once_with(1)
    mock_gateway.charge.assert_called_once_with("tok_123", 50.0)

Mock vs Patch vs MagicMock

python

from unittest.mock import Mock, patch, MagicMock, PropertyMock

# Mock — basic mock object
mock_obj = Mock()
mock_obj.some_method.return_value = 42
mock_obj.some_method(1, 2, 3)
mock_obj.some_method.assert_called_once_with(1, 2, 3)

# MagicMock — Mock with "magic methods" pre-created
mm = MagicMock()
mm.__len__.return_value = 5
assert len(mm) == 5
mm.__iter__.return_value = iter([1, 2, 3])
assert list(mm) == [1, 2, 3]

# patch — context manager for replacing attributes
with patch("module.ClassName") as MockClass:
    MockClass.return_value.some_method.return_value = 42
    instance = MockClass()
    result = instance.some_method()
    assert result == 42

# patch as decorator
@patch("module.ClassName")
def test_something(MockClass):
    MockClass.return_value.method.return_value = "mocked"
    instance = MockClass()
    assert instance.method() == "mocked"

💡 Tip: Use Ctrl + Mouse Wheel to zoom, or click the fullscreen icon

When (and When Not) to Mock

Mock When...	Don't Mock When...
External API calls	Simple data transformations
Database operations	Pure functions (same input → same output)
File system access	Business logic validation
Network requests	String manipulation
Third-party SDKs	Math/algorithmic operations
Time-dependent code	Internal helper methods

python

# APPROPRIATE mock: external HTTP call
def test_get_weather():
    mock_response = Mock()
    mock_response.json.return_value = {"temp": 22.5}
    mock_response.raise_for_status.return_value = None

    with patch("requests.get", return_value=mock_response):
        service = WeatherService("fake_key")
        temp = service.get_temperature("London")
        assert temp == 22.5

# INAPPROPRIATE mock: simple math
def test_discount_calculation():
    # Don't mock this! It's a pure function
    result = calculate_discount(100.0, 0.1)
    assert result == 90.0

# BETTER: Don't mock pure logic, test it directly
def test_is_valid_email():
    assert is_valid_email("user@example.com") is True
    assert is_valid_email("not-an-email") is False
    assert is_valid_email("") is False

⚠️Warning

Over-mocking is a common antipattern. If you mock everything, your tests become brittle — they break when implementation changes, not when behavior breaks. Mock boundaries, not internals.

Test Isolation: No Shared State

Tests must never depend on each other:

python

# BAD: Tests share state through global variable
items = []

def test_add_item():
    items.append("apple")
    assert len(items) == 1

def test_add_another_item():
    items.append("banana")
    assert len(items) == 1  # Fails! items already has "apple"

# GOOD: Each test creates its own state
def test_add_item():
    cart = ShoppingCart()
    cart.add_item("apple")
    assert cart.total_items() == 1

def test_add_another_item():
    cart = ShoppingCart()
    cart.add_item("banana")
    assert cart.total_items() == 1  # Passes — fresh cart

Isolation Rules

Rule	Why
No shared mutable state	Tests become order-dependent
No database records shared	Test A's data corrupts Test B
No global variables	One test pollutes state for another
No filesystem side effects	Test A's file creation breaks Test B
No time-dependent logic	Tests break at midnight or on Mondays
Fresh fixtures per test	`scope="function"` is your default

python

# BAD: Shared mutable fixture
test_users = []

@pytest.fixture
def user():
    test_users.append(User("Alice"))
    return test_users[-1]

# GOOD: Fresh state per test
@pytest.fixture
def user():
    return User("Alice")  # Fresh object every time

# BAD: Tests affect each other through DB
def test_create_user(db):
    db.execute("INSERT INTO users VALUES ('Alice')")
    assert db.query("SELECT count(*) FROM users")[0][0] == 1

def test_another_user(db):
    # This will FAIL if test_create_user ran first
    assert db.query("SELECT count(*) FROM users")[0][0] == 0

Testing Edge Cases

The most valuable tests are often the edge cases:

python

# Edge case patterns
def test_divide_by_zero():
    with pytest.raises(ZeroDivisionError):
        compute(10, 0)

def test_empty_input():
    assert process_list([]) == []

def test_single_element():
    assert process_list([42]) == [42]

def test_none_input():
    with pytest.raises(TypeError):
        process_list(None)

def test_boundary_values():
    assert validate_age(0) is False   # Boundary
    assert validate_age(1) is True    # Just above
    assert validate_age(17) is True   # Below threshold
    assert validate_age(18) is True   # At threshold
    assert validate_age(120) is True  # Upper limit
    assert validate_age(121) is False # Exceeded

def test_negative_values():
    assert process_transaction(-1) == "invalid"

def test_large_values():
    large_text = "a" * 10_000
    result = process_text(large_text)
    assert len(result) == 10_000

def test_special_characters():
    assert sanitize_input("<script>alert('xss')</script>") == ""

def test_unicode():
    assert process_name("José 💻") is not None

The FIRST Principles of Good Unit Tests

Principle	Meaning	How to Apply
Fast	Tests execute quickly	Millisecond-range. No I/O in unit tests.
Isolated	Tests don't depend on each other	Fresh fixtures, no shared state
Repeatable	Same result every time	No random values, no time dependence
Self-validating	Pass or fail automatically	No manual output inspection
Thorough	Cover edge cases, not just happy path	Boundary values, errors, nulls

python

import random
from datetime import datetime

# BAD: Non-repeatable test
def test_random_choice():
    result = random.choice([1, 2, 3])
    assert result > 0  # Weak assertion — always passes

# BAD: Time-dependent test
def test_good_morning_greeting():
    hour = datetime.now().hour
    greeting = get_greeting()
    if 5 <= hour < 12:
        assert greeting == "Good morning"
    else:
        assert greeting != "Good morning"

# GOOD: Controlled inputs
def test_random_choice_with_seed():
    random.seed(42)
    result = random.choice([1, 2, 3])
    assert result == 2  # Deterministic with seed

# GOOD: Inject the time
def test_greeting_at_specific_time():
    assert get_greeting_at(hour=9) == "Good morning"
    assert get_greeting_at(hour=14) == "Good afternoon"
    assert get_greeting_at(hour=20) == "Good evening"

Refactoring Production Code for Testability

Sometimes code isn't testable. Here's how to fix it:

python

# BEFORE: Untestable — hardcoded dependency
def send_welcome_email(user_email: str) -> None:
    import smtplib
    server = smtplib.SMTP("smtp.gmail.com", 587)
    server.login("user@gmail.com", "password")
    server.sendmail("from@test.com", user_email, "Welcome!")
    server.quit()

# AFTER: Testable — dependency injection
class EmailService:
    def __init__(self, smtp_client=None):
        self.smtp = smtp_client or self._default_smtp()

    def send_welcome(self, user_email: str) -> None:
        self.smtp.sendmail("from@test.com", user_email, "Welcome!")

    def _default_smtp(self):
        import smtplib
        server = smtplib.SMTP("smtp.gmail.com", 587)
        server.login("user@gmail.com", "password")
        return server

# Test
def test_send_welcome():
    mock_smtp = Mock()
    service = EmailService(smtp_client=mock_smtp)
    service.send_welcome("alice@test.com")
    mock_smtp.sendmail.assert_called_once_with(
        "from@test.com", "alice@test.com", "Welcome!"
    )

💡 Tip: Use Ctrl + Mouse Wheel to zoom, or click the fullscreen icon

Testing Exceptions and Error Paths

python

import pytest

class Validator:
    def validate_user(self, user_data: dict) -> None:
        if not isinstance(user_data, dict):
            raise TypeError("user_data must be a dict")
        if "email" not in user_data:
            raise ValueError("email is required")
        if "@" not in user_data["email"]:
            raise ValueError("Invalid email format")
        if user_data.get("age", 0) < 0:
            raise ValueError("Age cannot be negative")
        if user_data.get("age", 0) > 150:
            raise ValueError("Age seems unrealistic")

class TestValidatorExceptions:
    def test_rejects_non_dict(self):
        with pytest.raises(TypeError, match="must be a dict"):
            Validator().validate_user("not a dict")

    def test_rejects_missing_email(self):
        with pytest.raises(ValueError, match="email is required"):
            Validator().validate_user({"name": "Alice"})

    def test_rejects_invalid_email(self):
        with pytest.raises(ValueError, match="Invalid email"):
            Validator().validate_user({"email": "not-an-email"})

    def test_rejects_negative_age(self):
        with pytest.raises(ValueError, match="negative"):
            Validator().validate_user({"email": "a@b.com", "age": -5})

    def test_rejects_unrealistic_age(self):
        with pytest.raises(ValueError, match="unrealistic"):
            Validator().validate_user({"email": "a@b.com", "age": 200})

    def test_accepts_valid_user(self):
        """Happy path — no exception raised."""
        result = Validator().validate_user({
            "email": "alice@example.com",
            "age": 30,
        })
        assert result is None

Testing Asynchronous Code

python

import pytest

# Async function to test
async def fetch_user_data(user_id: int) -> dict:
    await asyncio.sleep(0.1)
    return {"id": user_id, "name": "Alice"}

# Async test
@pytest.mark.asyncio
async def test_fetch_user_data():
    result = await fetch_user_data(1)
    assert result["id"] == 1
    assert result["name"] == "Alice"

# Async with mocking
@pytest.mark.asyncio
async def test_async_service():
    mock_db = AsyncMock()
    mock_db.get_user.return_value = {"id": 1, "name": "Alice"}

    service = UserService(mock_db)
    result = await service.get_user(1)
    assert result["name"] == "Alice"
    mock_db.get_user.assert_called_once_with(1)

Common Test Code Smells

python

# SMELL 1: Testing implementation details
def test_sort_uses_quicksort():
    """BAD: Tests how sort works, not what it does."""
    arr = [3, 1, 2]
    result = sort(arr)
    # Checking internal algorithm is fragile

# SMELL 2: Multiple logic paths in one test
def test_complex():
    """BAD: Tests too many things."""
    if some_condition:
        assert x == 1
    else:
        assert y == 2

# SMELL 3: Test uses production data
def test_with_real_data():
    """BAD: Depends on external, changing data."""
    data = fetch_from_production_api()
    assert len(data) > 0

# SMELL 4: Conditionals in tests
def test_conditional():
    """BAD: Tests should be deterministic."""
    if os.name == "nt":
        assert windows_specific()
    else:
        assert unix_specific()

Smell	Symptom	Fix
Fragment Test	Tests many unrelated behaviors	Split into focused tests
Lazy Test	Asserts too little	Test the actual output
Subordinate Test	Tests helper, not main	Identify the real unit
Indecent Test	Accesses private methods	Test via public interface
Mystery Guest	Uses global state	Inject dependencies
Resource Optimist	Tests don't clean up	Use fixtures with teardown
Local Hero	Tests depend on locale	Set locale explicitly

Practice Exercises

Refactor to AAA: Take this poorly structured test and refactor it into three separate AAA tests:

python
def test_cart(): cart = ShoppingCart() cart.add_item("apple", 1.0) assert cart.total() == 1.0 cart.add_item("banana", 2.0) assert cart.total() == 3.0 cart.remove_item("apple") assert cart.total() == 2.0
Write Mock Tests: Create a NewsletterService that calls an email API. Write tests that mock the API call. Verify the correct data is sent.
Identify Isolation Violations: Find and fix the shared-state problems in this test suite:

python
users = [] def test_create_user(): users.append(User("Alice")) assert len(users) == 1 def test_delete_user(): users.clear() assert len(users) == 0
Edge Case Coverage: Write comprehensive edge-case tests for a function parse_int(s: str) -> int that handles regular numbers, negative numbers, zeros, leading zeros, empty strings, non-numeric strings, and whitespace.
FIRST Audit: Review a test suite from a real or sample project. For each test, evaluate it against the FIRST principles. Which tests violate which principles?
Dependency Injection Refactor: Take this untestable code and refactor it for testability, then write tests:

python
class ReportGenerator: def generate(self): import datetime now = datetime.datetime.now() with open("/var/log/app.log") as f: data = f.read() return f"Report generated at {now}: {len(data)} bytes"
Test a Fibonacci Function: Write tests for a fibonacci(n) function. Cover n=0, n=1, n=2, n=10, negative n, non-integer n, and large n. Use parametrize.
Mocking a Payment Gateway: A PaymentService has a process_payment(user_id, amount) method that calls a payment gateway. Write tests for success, declined card, insufficient funds, and timeout scenarios using mocks.

Summary

AAA Pattern: Arrange → Act → Assert — one clear story per test
One concept per test: Not one assertion, but one behavior
Descriptive names: test_given_X_when_Y_then_Z tells the full story
Mock boundaries: Mock external dependencies, not internal logic
Isolation: Fresh state per test, no sharing, no order dependence
Edge cases: Empty, null, negative, boundary — test them all
FIRST: Fast, Isolated, Repeatable, Self-validating, Thorough
Testability: Design code so it can be tested (dependency injection)

✅Success

Great testing is a design discipline. The AAA pattern, proper mocking, and strict isolation are not overhead — they are investments that make every future refactoring safe and every bug report solvable.

Progress33%

Writing Your First Tests

Test Coverage & Quality

Course Content

1. Introduction to TDD 2. Writing Your First Tests 3. Unit Testing Best Practices 4. Test Coverage & Quality 5. Pre-Commit Hooks 6. Linting & Code Quality 7. Code Formatting 8. Static Analysis & Code Scanning 9. Complete CI/CD Pipeline