1. 9
  1. 2

    I often define setup functions and call them from each test that needs them, as suggested here as a middle ground. I find that it works reasonably well. The tests are each shorter, but it’s still very easy to identify and inspect the setup.