I still can't repro, but I committed some added verbosity (if you run w/ -Dtests.verbose=true) -- Robert can you try running verbose and see if you can still tickle the bug (it's llikely you can't... heisenbug... but worth a shot)?
Mike On Wed, Sep 8, 2010 at 5:27 AM, Michael McCandless <[email protected]> wrote: > This looks nasty!! > > But I can't repro :( I've got it running in a while(true) loop... which JRE? > > It must be a thread scheduling issue with CMS, since it intermittently > happens with a fixed seed. > > And it's very interesting that disabling deletions fixes it. This > points to the mixed bulk/non-bulk merging code for stored fields... > > Can you open an issue and post all the exceptions you've seen? > > Mike > > On Tue, Sep 7, 2010 at 4:56 PM, Robert Muir <[email protected]> wrote: >> Hello, >> I've tripped on this a few times lately, but never been able to reproduce >> it: it seems now i am able to reproduce it now semi-consistently with the >> below configuration. >> It would be great if someone else could try this out and see if its a real >> problem, or if its just my machine. >> occasionally i see a very nasty result >> from TestIndexWriter.testRandomStoredFields: >> either a read past EOF, IndexOutOfBounds, NegativeArraySizeException, or >> field X is wrong, expected <nonsense unicode> actual <different nonsense >> unicode> >> Here are my steps to reproduce: >> 1. edit line 87 of TestIndexWriter to plugin the seed: >> random = newRandom(3312389322103990899L); >> 2. run this command: >> ant clean test-core -Dtestcase=TestIndexWriter >> -Dtestmethod=testRandomStoredFields -Dtests.iter=10 >> -Dtests.codec="MockVariableIntBlock(29)" >> I used 10 iterations here, as it will usually fail with this seed and # of >> iterations for me. >> furthermore, if i comment out lines 5179 and 5180 from TestIndexWriter so >> that it no longer randomly deletes documents, the test will always pass: >> //w.deleteDocuments(new Term("id", delID)); >> //docs.remove(delID); >> -- >> Robert Muir >> [email protected] >> > --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
