Test bench instability

[Thrameos]

One of the unittests fails, but only in combination with random testing sequence. Somewhere in the code we are getting into a state where we can exist this code without setting an exception. Unfortunately, I can't find the correct sequence as it requires running to a fail, deleting tests, rinse, lather, repeat until only the repeater remains.

        class q:
            def __float__(self):
                raise SystemError("nope")
        ja = JArray(JDouble)(5)
        a = [1, -1, q(), 3, 4]
        with self.assertRaises(SystemError):
            ja[:] = a

[marscher]

pytest-randomly spits out the random number generator seed it is using to make the sequence of execution reproducible after running into an failure. Maybe that's useful in this situation.

$ py.test test -x
====================================================================================== test session starts =======================================================================================
platform linux -- Python 3.14.4, pytest-9.0.3, pluggy-1.6.0 -- /usr/local/miniforge3/envs/jpype/bin/python3.14
cachedir: .pytest_cache
Using --randomly-seed=774890949

[Thrameos]

I think the difficulty here is that to create a replicator we have to remove the stray tests. When it was a single test bench that had an issue, I just added the disable until all I have left was A before B pass, B before A fail. The current one appears to be two test benches. So to narrow it down we must run until fail. Then delete all the test bench after it, repeat until we get the order.

I suspect one of the fault tests is the culprit as they can actually leave things in bad state. Which would explain how we get a "exeception not set" SystemError. But copying the whole test bench to jpypetest2 and the trimming is very tedious. It took me 20 runs to get it to repeat and sadly it was 2/3 down the bench meaning not too useful for elimination.