unit-testing

Structure every test in three phases: Arrange sets up inputs and dependencies, Act calls the unit under test exactly once, and Assert verifies the outcome. Keep one logical assertion per test and name tests to describe the scenario and expected result.

Python uses unittest.mock or pytest-mock's mocker.patch, TypeScript uses jest.fn or vi.spyOn, and Go defines an interface for the dependency with a test struct implementing it. Mock only external boundaries like databases, HTTP, and filesystems, never the unit under test.

A stub returns fixed values, a mock records calls for interaction verification, a spy wraps the real implementation while recording calls, and a fake is a lightweight working implementation like an in-memory repository. Use fakes for stateful dependencies and mocks for fire-and-forget side effects.

Target 80% line and 70% branch coverage for greenfield projects, and 95% line with 90% branch for critical paths like payments and auth. Exclude generated code, migrations, and configuration loaders, and treat coverage as a floor rather than a goal.

Flakiness usually comes from shared mutable state between tests, execution-order dependencies, or time-based sleeps. Give each test its own fixtures, clean up with teardown hooks or t.Cleanup, and inject a clock abstraction instead of calling sleep.

TDD pays off most for new features with clear acceptance criteria, complex business logic, and regression prevention. It adds less value for exploratory spike code, UI layout work, and third-party integration spikes where integration tests are more appropriate.