> Tests that test that we simply don't crash seem inadequate - could we test > that we have the correct behavior rather than just "don't crash" (probably > an IRGen test)? OK. Expanded the test in r194050. Cheers, Rafael