Mike Matrigali wrote: > New tests that break regression suite should be backed out of suite.
This case is interesting because the test does "pass" on Sun's jdk 1.4.2, that I think is reasonable for test that truly passes to be an addition to the derbyall regression suite. [ Assuming of course the test is not intentionally written to only run on a single jvm. ] Now if a valid test fails on IBM's 1.4.2, that's a problem for whoever has the itch for derby to run on that vm, same as if the test failed on Jrockit. But in this case, the test is actually failing, but running clean since the failures are in the master file. Since the failures are not expected in the long run, I assume not a lot of work has gone into making the failure output consistent across jvms. I'm not sure what should be done about this. Seems to be wait for the patch that implements the functionality or remove it from the suites. I know that it wastes my time, having this consistent failure in my runs, also conditions me to ignore that test in the future. If I'm breaking that in some way, and don't realize it's meant to run clean, then I'm going to ignore it's error and checkin a bad fix. I just added a test that I knew showed an error (AggregateClassLoading.java) but didn't add it to the suites until I had the fix, seemed more natural to me. Makes the test available for others to use, doesn't break anyone. Dan.
