On 9/28/06, Alexey Varlamov <[EMAIL PROTECTED]> wrote:

Apparently there are some threading issues, which affect almost all
suites (except may be smoke tests). I observe quite many sporadic
crashes in c-unit, kernel and classlib tests. For example, the
following assert fails often:
drlvm/vm/thread/src/thread_native_suspend.c:351:
hythread_set_safepoint_callback: Assertion `status == (0)' failed.
Another incomprehensible error is described in HARMONY-1614.
These defects most probably point to races in TM or VM internals, and
are hard enough to insulate and root-cause.

Aside from that, there are "known" stable failures, like kernel
ThreadTest and ClassLoaderTest, or classlib's


ThreadTest is a kind of "always failing test" unfortunately. Now its failure
is related to launcher related issue mentioned in another thread as well as
the HARMONY-1537 issue. All these issues are fixed with HARMONY-1582 patch.
On the other hand, HARMONY-1582 reports that ThreadTest fails showing up new
problems possibly related to H-1519 patch. So ThreadTest is very helpful in
catching regressions and should be paid a special attention to each time it
fails :)

As for general statistics of kernel test failures, ThreadTest fails now on
all platforms in all modes (jet, opt, int) and ClassLoaderTest and
ClassAnnotationTest fail on opt (win, linux), which is under Egor's
investigation.

Thanks,
Elena

org.apache.harmony.archive.tests.java.util.jar.JarOutputStreamTest and
org.apache.harmony.luni.tests.java.lang.ThreadGroupTest. These are
good candidates to fix first - especially Thread related - maybe this
would help to get closer to more subtle defects.


28 Sep 2006 14:47:27 +0700, Egor Pasko <[EMAIL PROTECTED]>:
> On the 0x1F2 day of Apache Harmony Geir Magnusson, Jr. wrote:
> > Right now, we have lots of tests that are skipped in smoke tests, and
> > some that fail in kernel, and I believe that we're still failing the
> > classlib test suite.
> >
> > We should drive this to goodness and health.  How shall we start?  :)
>
> I am working on the ClassLoaderTest :)
> "memopt" optimization pass makes it fail, it even removes some
> array-bounds-checks (surprize, surprize), which looks like, ehm,
> unfair. For now I could not reduce the test to a small one. On small
> pieces of code "memopt" works less dangerously.
>
> Disabling "memopt" helps, but that would drop performance down. You
> can disable it temporarily if stability is very critical. Just edit
> the opt.emconf file (and server.emconf, etc.)
>
> --
> Egor Pasko, Intel Managed Runtime Division
>
>
> ---------------------------------------------------------------------
> Terms of use : http://incubator.apache.org/harmony/mailing.html
> To unsubscribe, e-mail: [EMAIL PROTECTED]
> For additional commands, e-mail: [EMAIL PROTECTED]
>
>

---------------------------------------------------------------------
Terms of use : http://incubator.apache.org/harmony/mailing.html
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]




--
Thanks,
Elena

Reply via email to