Digging in a bit into the slow nightlies, I think this explains a lot (see my other email: question about seed)
I opened https://issues.apache.org/jira/browse/LUCENE-3995, but honestly i would really prefer if we didnt have to change our tests here, if somehow, when we do getRandom() the class's hashcode or something like that is xored with the seed to mix it up a bit, I think it would be ok. I can easily fix LuceneTestCase/SolrTestCaseJ4/etc but there are many tests that do stuff in beforeClass (e.g. build a directory), so if all tests are 'doing the same thing' we get less predictable test times, and less test efficiency. This is because there is redundancy in our tests (of course), which I think is ok, actually good, as long as these defaults are varying per-test-class such that we get wide coverage within a single 'ant test'. On Tue, Apr 17, 2012 at 3:53 PM, Robert Muir <[email protected]> wrote: > I haven't looked, but i seriously doubt thats responsible at all. > Thats just an example "crazy" one. > > More likely it got SimpleText codec :) > > On Tue, Apr 17, 2012 at 3:49 PM, Dawid Weiss > <[email protected]> wrote: >>> Like this? :) >>> >>> https://builds.apache.org/job/Lucene-trunk/1896/testReport/org.apache.lucene.index/TestIndexWriterReader/history/ >> >> If it's correlated with the commit of RandomizedRunner then I'd check >> if it's not that "asserting" Random instance or random() being >> repeatedly called gazillion times. Like I said -- there is an extra >> cost for these assertions so in tight loops (where there is no >> possibility of Random escaping to another thread, etc.), I'd just >> create a simple Random rnd = new Random(random().nextLong()). >> >> Dawid >> >> --------------------------------------------------------------------- >> To unsubscribe, e-mail: [email protected] >> For additional commands, e-mail: [email protected] >> > > > > -- > lucidimagination.com -- lucidimagination.com --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
