On 4/6/2010 8:13 PM, David T. Lewis wrote:
> Using a Unix VM at version level 4.0.2-2172 (roughly equivalent to what
> I expect Ian to publish), I ran the unit tests in a 4.1 image updated
> to 9902. Aside from expected failures, the results are:

Thanks for running the tests, that is *very* encouraging!

> - AllocationTest>>testOneGigAllocation passes, but is marked as an
>    expected failure and shows up as an "unexpected pass" in the test
>    runner. This is due to workarounds in the test to avoid Unix VM
>    bugs, which will be corrected after the VMs are updated (but not
>    until then because the bugs lead to crashes).
>
> - WeakRegistryTest>>testFinalization failed once when running the
>    full test suite, but does not fail when run as an individual test.
>
> - MCChangeNotificationTest>>testCoreMethodModified failed when running
>    the full test suite (two out of three times that I tried the full
>    suite), but does not fail when run as an individual test.
>
> None of these seem to be show-stopper concerns with respect to the
> 4.1 release.

Yes, and I think I've taken care of the unreliable tests (the allocation 
test is just so that VM doesn't crash so I'll discount it :-) It turns 
out that three unreliable cases where all (more or less) easily fixed if 
you can just catch them in the act:

* StringSocketTestCase has an assumption about network I/O completing 
synchronously. A small wait does wonders.

* WeakRegistryTest similarly assumed synchronous completion although 
with better reasoning - it isn't clear to me why sometimes the 
finalization wouldn't happen synchronously (I have some suspicions but 
not enough time to verify them :-) but again a small wait fixes the 
problem 100%.

* MCChangeNotificationTest actually points to a real issue since it 
shows that the treatment of package names is not consistent, i.e., some 
places assume package names are case sensitive, others aren't. Since 
fixing that issue is way beyond scope I only fixed it in a minimally 
invasive and localized way.

As far as I know, all tests should now be reliable. If people still find 
occasional failures please point them out.

Cheers,
   - Andreas