I looked into this a while ago, bug 899697 [0] has some info. Summary: It's definitely Windows only. The only tests that fail are the ones that expect a return code != 0. Most of these tests expect return code 3 (because they have a "|jit-test| error" comment), but I think I've also seen tests fail that expect 6 (timeout). This happens also with very simple tests and with shell flags that disable the JITs.
When the tests fail, they do throw the error we expect. If you look at bug 916836 for instance: TEST-UNEXPECTED-FAIL | e:\builds\moz2_slave\m-in-w32-000000000000000000000\build\js\src\jit-test\tests\parallel\bug858582.js | --baseline-eager --no-ti --no-fpu: @evaluate:1:679 TypeError: cmpFunction is not a function The test that fails has "// |jit-test| error: TypeError", this TypeError is the one we expect and see on other platforms. So the test itself happily succeeds but the shell somehow returns 0 instead of 3 and the test fails. If we wait a few more months, the sheriffs will have filed bugs for all tests that have "|jit-test| error" or "|jit-test| exitstatus". That's only ~330 bugs... Even if I make the shell's main() always "return 3", we reach that return but the process exit code is still 0 every X thousand runs. Apparently this is possible on Windows in some cases [1]. I think it requires a thread-safe shell, it's probably something NSPR related. I also thought it didn't happen if I make main() return before JS_DestroyRuntime, but I may be wrong about that. The problem started after bug 776043 [2] landed, it made the jit-test harness check the return code. My suggestion in bug 899697 was to always allow return code == 0 on Windows to stop these intermittent failures. It's hiding the problem somehow, but unfortunately I don't have the time to get to the bottom of this.. It doesn't help that it can take > 30 minutes before a test fails, so it's hard to say whether a change "fixes" it.. Jan [0] https://bugzilla.mozilla.org/show_bug.cgi?id=899697 [1] http://blogs.msdn.com/b/oldnewthing/archive/2008/05/06/8461730.aspx [2] https://bugzilla.mozilla.org/show_bug.cgi?id=776043 On Sat, Sep 14, 2013 at 3:55 AM, Bill McCloskey <[email protected]>wrote: > Does anyone have any idea what's going on here? It seems like there are > maybe 5 or 10 a day getting filed. Examples: > > https://bugzilla.mozilla.org/show_bug.cgi?id=916388 > https://bugzilla.mozilla.org/show_bug.cgi?id=916386 > > They all have the form "X is undefined" or "X is not a function" or some > other JS error. Is this a new occurrence or were they just never filed > before? Is anyone looking into the problem? Would it make sense to group > them all into one big intermittent jit-tests bug? It's probably annoying > for the sheriffs to have to file a new bug for every jit-test. > > As far as I can tell, they all happen on Windows. Not that that helps much. > > -Bill > _______________________________________________ > dev-tech-js-engine-internals mailing list > [email protected] > https://lists.mozilla.org/listinfo/dev-tech-js-engine-internals > _______________________________________________ dev-tech-js-engine-internals mailing list [email protected] https://lists.mozilla.org/listinfo/dev-tech-js-engine-internals
