Consider this test:

# The function we're testing def foo(a: int, b: int) -> int: return a * 3 + b @pytest.mark.parametrize(['a', 'b', 'result'], [  [1, 2, 5],  [2, 3, 8],  [5, 3, 18], ]) def test_foo(a, b, result): return foo(a, b) == result

Did you spot the error? The second parametrization has a typo, 2 * 3 + 3 is 9, not 8. But this test actually doesn't test anything, because it returns the assertion rather than asserting it. This is a common enough mistake, and it wouldn't normally be a problem except that it can silently cause false positives in test suites.

I propose that test functions that return anything except None fail with a message that cues users that they probably meant to assert rather than return. This feature could be disabled (or enabled, if there are backwards-compatibility issues) via a config flag if necessary.