Python Mock Isolation Design

Python Mock Isolation Design

Use this skill when replacing a dependency in a Python test changes what the test proves. Mocks are useful at slow, nondeterministic, or external boundaries; they become harmful when they replace the behavior the test is supposed to validate.

Source Traceability

Primary source: Harry Percival, Test-Driven Development with Python, 3rd ed. Guidance is transformed and paraphrased from chapters 20, 21, 27, and Appendix A, especially manual monkeypatching, unittest.mock.patch, mock coupling, call argument inspection, test isolation, and the architectural route out of mock-heavy suites.

Workflow

1. Name the boundary.

   • External service, email, clock, filesystem, network, framework adapter, or expensive side effect.
   • If the dependency is local domain logic, prefer real code.

2. Choose the replacement.

   • Use a fake when state and behavior are simple and meaningful.
   • Use patch when replacing a collaborator looked up by the module under test.
   • Use monkeypatching sparingly and restore state automatically.
   • Use autospec/spec when interface drift matters.

3. Keep behavior assertions primary.

   • Assert user-visible or domain-visible outcomes first.
   • Assert mock calls only when the interaction is the contract.
   • Inspect call arguments when they clarify the behavior, not as a substitute for it.

4. Watch for design feedback.

   • If setup requires many mocks, the code may have hidden dependencies.
   • If tests break on harmless refactors, the test is too coupled to implementation.
   • Consider introducing a boundary interface, service object, or functional core.

Read mock-isolation-patterns.md for replacement choices and smell handling.

Decision Rules

• Patch where the code under test looks up the dependency.
• Prefer fakes for small stable protocols.
• Use mock.return_value and side_effect deliberately; do not let default mocks create imaginary object graphs.
• Avoid broad MagicMock objects without specs for important interfaces.
• Keep at least one integration or contract test for each important boundary.