Test driven development

Test-Driven Development

Purpose

Use tests to drive design. TDD's value is not primarily the tests it produces — it is the pressure it puts on the design, because code that is hard to test is hard to use.

When to Use

• Implementing logic with clear inputs and outputs.
• Fixing a bug (the failing test comes first, always).
• Working in a codebase where you do not fully understand the existing behavior.
• Designing an API you will have to live with.

Capabilities

• The red-green-refactor cycle.
• Test selection: choosing the next test that moves the design forward.
• Outside-in (from the API) and inside-out (from the core) approaches.
• Using tests to characterize existing behavior before changing it.

Inputs

• The requirement, small enough to express as one test.
• The existing code, if any.

Outputs

• Code that exists because a test demanded it.
• A suite where every test failed before it passed.
• A design that is testable because it was built under test.

Workflow

1. Red — Write the smallest test that fails for the right reason. Run it. Watch it fail. A test you have not seen fail is a test that may not be testing anything.
2. Green — Write the least code that makes it pass. Hardcoding the answer is legitimate at this stage; the next test will force generalization.
3. Refactor — With the test green, improve the structure. The test is your safety net, and this is the only step where you may change design without changing behavior.
4. Choose the next test deliberately — The next test should either force a generalization or cover a new case. A test that already passes teaches nothing.
5. Repeat in small cycles — Minutes, not hours. If you have been red for twenty minutes, the step was too big — revert and take a smaller one.

Best Practices

• Watch the test fail. A test that passes before you have written the code is testing nothing, and this happens more often than anyone admits.
• If a test is hard to write, the design is wrong. That signal is TDD's main product; do not suppress it by mocking your way around the difficulty.
• Do not write five tests and then the implementation. That is not TDD; it is speculation followed by a large step.
• Triangulate: the second test is what forces you to replace return 42 with an actual calculation. One test permits a hardcoded answer.
• TDD is poor at discovering an algorithm you do not yet understand. Spike it in a scratch file, throw the spike away, then TDD the real thing knowing where you are going.
• Never skip the refactor step. TDD without refactoring produces working code with a poor design and a suite that pins it in place.

Examples

The cycle, in full, on a small requirement:

# RED — the simplest failing test. Run it; watch it fail with a NameError.
def test_no_items_is_free():
    assert price(items=[]) == 0

# GREEN — the least code that passes. Yes, this is a hardcoded return.
def price(items): return 0

# RED — this test forces the hardcoding out.
def test_single_item_costs_its_price():
    assert price(items=[Item("widget", cents=250)]) == 250

# GREEN
def price(items): return sum(i.cents for i in items)

# RED — the requirement that actually has business logic in it.
def test_bulk_discount_applies_over_ten_units():
    items = [Item("widget", cents=250)] * 12
    assert price(items) == int(250 * 12 * 0.9)

# GREEN
def price(items):
    total = sum(i.cents for i in items)
    return int(total * 0.9) if len(items) > 10 else total

# REFACTOR — now that it is green, make it express the domain.
BULK_THRESHOLD = 10
BULK_MULTIPLIER = 0.9

def price(items: list[Item]) -> int:
    subtotal = sum(i.cents for i in items)
    return _apply_bulk_discount(subtotal, len(items))

def _apply_bulk_discount(subtotal: int, count: int) -> int:
    return int(subtotal * BULK_MULTIPLIER) if count > BULK_THRESHOLD else subtotal

Each test failed before it passed. Each piece of production code exists because a test demanded it. The design emerged from the pressure of writing the tests, not from a diagram.

Notes

• The most valuable use of TDD is bug fixing. A failing test that reproduces the bug proves you understood it, and it becomes the regression test for free.
• TDD does not mean "test first" in the trivial sense of writing all the tests up front. The loop is minutes long, and the design feedback arrives continuously.
• TDD is a poor fit for exploratory work, UI layout, and performance tuning. Use it where correctness is the constraint, not where the shape of the answer is unknown.