For each test that asserts the behavior of either "direct" or "indirect" eval (but not both), introduce an equivalent test for the opposite case.