watch make test
Via a changelog post I got this great idea. As long as your unit-tests run in under 2 seconds, you can use the watch(1) utility to run all your tests, all the time in a different screen shell; by the time you think you've fixed a bug or caused a problem and switched shells to re-run the tests, there they are.
Of course, you can always use -n 10 or something if your tests take too long. Not useful for my current job, where our tests take like 20 minutes...
Great trick when it works, though!