[kaffe] cvs kaffe flakey (i386, recent glibc)

[Dalibor Topic]

Noa Resare wrote:
> With today's cvs kaffe i have experienced some serious instability on my
> Fedora Core 2 test 2 box (Athlon). When trying the TheadState regression
> test 10 times i got the following:
> 
> 6 times no problem
> 
> 2 times:
> Internal error: caught an unexpected exception.
> Please check your CLASSPATH and your installation.
> java/lang/NullPointerException
> Avbruten (SIGABRT)
> 
> 1 time, silent lockup. No Success messages, no cpu utilization.
> 1 time, lockup after a 5 Success messages, no cpu utilization
> 
> I'll do some binary searching for the 'latest good' cvs state now, but
> if anyone has an idea of what to look for I'm interseted.

Hi Noa,

pthreads have been flakey for me for a while, it was just the recent 
changes that made the flakeyness occur more regularly. The two tests 
that give me irregular breakage are GCTest & WaitTest. WaitTest uses 
ocasionally to either hang in the second thread for more than 300sec, or 
to quit the second thread with a bad value. GCTest ocassionally seems to 
miss a line of output. I've noticed the frequency of flakey output 
increased after my change from

2004-10-12  Dalibor Topic <robilad at kaffe.org>

         * kaffe/kaffevm/systems/unix-pthreads/thread-internal.h
         (jthread_on_current_stack): De-inlined to fix compiler
         warnings.

         * kaffe/kaffevm/systems/unix-pthreads/thread-impl.c
         (jthread_on_current_stack): Moved over here.

but I'd be surprised if deinlining it skewed the calculations as they 
use stackMin/Max which are set when a thread is created.

cheers,
dalibor topic