I ran kernel tests iteratively and saw intermittent failures of ThreadTest.testYield() on linux. Despite the test always passes on Windows, I think the test is not quite correct: it expects that the two threads run alternately but actually any other thread can run while some thread yields. This happens with high probability during the test run in batch mode. I created https://issues.apache.org/jira/browse/HARMONY-2343 describing the issue and suggested the fix.
What is interesting: I attached a test to JIRA issue which is an exact test case testYield() from the current ThreadTest. It never fails on drlvm (at least I have not seen a failure) but always fails on RI 1.5.0_08. Does anyone have some thoughts about Thread.yield() to share? Is the fix acceptable? -- Thanks, Elena
