Test data

Test data quietly determines how good a test suite is. Data that is too clean produces tests that pass on inputs no real user generates. Data that is shared produces tests that fail depending on what ran first. Both problems are invisible until they cost a day.

Each test should build what it needs and care about nothing else.

1. Prefer factories to fixtures

A fixture is a fixed blob of data everything shares. A factory is a function that builds an object with sensible defaults and lets the test override only what it cares about.

Factories win because the test states its own preconditions:

user = make_user(plan="enterprise")     # this test cares about the plan, nothing else