On 11/22/06, Geir Magnusson Jr. <[EMAIL PROTECTED]> wrote:



Elena Semukhina wrote:
> On 11/22/06, Geir Magnusson Jr. <[EMAIL PROTECTED]> wrote:
>>
>>
>>
>> Elena Semukhina wrote:
>> > On 11/22/06, Geir Magnusson Jr. <[EMAIL PROTECTED]> wrote:
>> >>
>> >> As an interim step to track where we are as the build-test and
>> >> harmonytest.org keep evolving, I've created a wiki page :
>> >>
>> >>   http://wiki.apache.org/harmony/DRLVMTestTracking
>> >>
>> >> that's reachable from the front page to keep a log.  I know we have
>> >> classlib unit tests tracked somewhere else, so we should at least
link
>> >> the two pages.
>> >>
>> >> I have linux x86 and linux x86_64 logged for today, and will do
>> windows
>> >> unless someone beats me to it :)
>> >
>> >
>> > I updated the page with Windows data for kernel tests. Windows XP opt
>> hangs
>> > on ThreadTest in "once" mode while passes in "perTest".
>>
>> Boy oh boy.
>>
>> 1) I don't understand your numbers.   How can you have 2x as many
>> excluded?
>
>
> exclude.drlvm_kernel.windows.x86.* contain two more tests excluded
> comparing
> to  exclude.drlvm_kernel.linux.x86.*:
> java/lang/SystemExtensionTest.java
> java/lang/ObjectTest.java
>
> They are 37 test cases. I meant that 44 is for the tests in
> exclude.drlvm_kernel.linux.x86.*.
> And then 37 + 44 = 81.

Ah - I'm just counting classes, not individual tests.

>
> 2) Yes, I saw the hang yesterday (only in int, btw), but hadn't figured
>> out which test it was.
>
>
> I see it only for opt on Windows XP in "once" mode.

Odd.  Scary.  And Odd - I was able to get winXP, debug build to run
through all opt and jet in once mode, and then hang in int. (I think.
I'll recheck)

>
> So do we exclude ThreadTest?  I think so - I also suspect it's time to
>> call "full stop" and run these thread problems into the ground.  As
>> amusing as the discussion around HARMONY-1951 is, I couldn't sleep
>> worrying about it :)
>
>
> Excluding ThreadTest helped tests pass. Let's exclude the test for a
while
> until the problem is identified and fixed.
> I'll file a JURA issue.
> Actually this test helped to catch a lot of bugs in the past so we
cannot
> afford to exclude it for a long time :)
> Besides, this test is known to fail intermittently on join() and
yield().
> I'll do iterative runs for it to gather statistics meanwhile.

Right - my point is that excluding our thread tests is a big red flag.
I know that you and others have been working very hard to get this fixed.


I managed to run kernel tests iteratively for 40 times on linux in jet, opt,
int (debug build, "once" mode) and have not seen a failure! ThreadTest was
included. Tried the same on Windows 2003 and saw 1 failure of ThreadTest in
jet mode. It failed on interrupting a waiting thread.
So it is worth to exclude ThreadTest on Windows. Should we do this on linux?
I'll continue my experiments tomorrow.

Thanks,
Elena

geir




--
Thanks,
Elena

Reply via email to