testing-guide

Writing a Single Good Test

How to structure, name, populate, and verify one test — and how to pick the right test double — independent of any framework. It governs the shape of an individual test, not the rhythm in which you write tests.

Essentials

• Structure every test as Arrange-Act-Assert - set up, run one action, verify the outcome, see references/test-structure-and-first.md
• Make each test meet FIRST - Fast, Independent, Repeatable, Self-validating, Timely, see references/test-structure-and-first.md
• Name the right double for the job - Dummy / Stub / Spy / Mock / Fake are distinct, not synonyms, see references/test-double-taxonomy.md
• Prefer state verification; mock only true outputs - assert outcomes, reserve call-checking for external commands, see references/state-vs-behaviour-and-what-to-mock.md
• Mock at architectural seams, not internals - double the payment gateway, never value objects or the system under test, see references/state-vs-behaviour-and-what-to-mock.md
• Read a failing test as a smell report - fragile, obscure, eager, erratic tests each name a fix, see references/test-smells-and-fixtures.md

Gotchas

• "Mock" is not a synonym for "test double". A Mock holds pre-programmed call expectations and is verified; most things people call mocks are actually Stubs that answer queries with canned state.
• A Stub answers queries (state); a Mock verifies commands (behaviour). Asserting on calls when state would do couples the test to the implementation and makes it Fragile.
• AAA is three phases, not four. Explicit Teardown belongs to the Four-Phase Test and is usually unnecessary for a true unit test.
• A Spy is not a Mock: a Spy passively records calls for the test to assert later; a Mock owns the expectations itself and fails on an unexpected or missing call.
• A Fake is not a Stub: a Fake has a real, working, shortcut implementation (an in-memory store); a Stub only returns hard-coded answers with no logic.
• A test that needs many doubles or asserts on internal calls is a coupling signal — fix the production design, and take the coupling diagnosis to connascence-guide, not a bigger fixture.
• FIRST's "I" (Independent / Isolated) and "S" (Self-validating / Self-verifying) both circulate; same intent, neither spelling is wrong.

Example

// Arrange — one Dummy, one Stub, one Spy, one Mock, one Fake
const auditId = "ignored"; // Dummy: fills the signature, never read
const fares = stubFare({cityCenter: 1250}); // Stub: canned answer (Responder)
const receipts = spyReceipts(); // Spy: records each send(...)
const gateway = mockGateway().expect("charge", 1250, "EUR"); // Mock: command expectation
const cards = inMemoryCardStore(); // Fake: real but shortcut storage

// Act — exactly one action
const confirmation = chargeRide(trip, gateway, receipts, fares, cards, auditId);

// Assert — state first, then the one behaviour expectation
expect(confirmation.status).toBe("paid"); // state verification
expect(receipts.sent).toHaveLength(1); // spy: outcome, not call order
gateway.verify(); // behaviour verification of the command

Progressive Disclosure

• Read references/test-double-taxonomy.md - Load when choosing or naming a double, or settling mock vs stub vs fake vs spy
• Read references/state-vs-behaviour-and-what-to-mock.md - Load when deciding what to mock, whether to assert on state or calls, or diagnosing over-mocking
• Read references/test-structure-and-first.md - Load when structuring a test (AAA / Four-Phase), naming it, or checking it against FIRST
• Read references/test-smells-and-fixtures.md - Load when a test is fragile/obscure/eager/erratic/tautological, or when building fixtures with a Builder or Object Mother