[kaffe] frustrating news regarding the pthread instability

[Guilhem Lavaux]

Noa Resare wrote:
> Hello
> 
> Now I'm back from a few days away visiting my parents and I have
> continued my investigation on the pthread instability in kaffe.
> 
> What I have concluded is sadly not what I had hoped for. I was looking
> for a change in the pthread code somewhere between today and 1.1.4 that
> introduced the lockups that I have been experiencing. However, it seems
> like a simple test can "break" every version back to 1.1.4 given enough
> time. What I have been chasing has always turned out to be unrelated to
> threading in any other way than changing the timing and making the race
> turn up more or less likely.
> 
> The test I am using is the following:
> 
> (ThreadState is ThreadState.java from tests/regression in cvs HEAD with
> the timeout thread commented out.)
> 
> while true; do /tmp/kaffe-1.1.4/bin/kaffe  ThreadState || break;  date;
> sleep 0.1; done
> 
> head cvs lasts about 5 minutes of this. 1.1.4 an hour or two other
> versions in between probably about 10 minutes.
> 

I have identified and patched a race vulnerability which happens at 
thread destruction and another one (thanks to helmer) which happens at 
thread creation. I'll try to commit it this evening.

Thanks for looking for the problem.

Cheers,

Guilhem.

> I'm giving up on this approach now and instead I'll be trying to find
> the "fastest possible" way of producing a lockup in head cvs by
> artificially change the timing and when I have something that is
> reasonably fast (hopefully a few seconds) I'll try to analyze exactly
> what happens.
> 
> but that will be tomorrow.. any input is of course appreciated.
> 
> /noa
> 
> 
> _______________________________________________
> kaffe mailing list
> kaffe at kaffe.org
> http://kaffe.org/cgi-bin/mailman/listinfo/kaffe
>