Re: ProcessManager on Solaris

[David Daney <ddaney at avtrex dot com>]

Bryce McKinlay wrote:
> Eric Botcazou wrote:
>
> > As exhibited in 
> > http://gcc.gnu.org/ml/gcc-testresults/2005-02/msg00579.html ,
> > we have a problem with the ProcessManager on SPARC/Solaris: it 
> > randomly deadlocks.
> >
> > The proper solution would be to block SIGCHLD in the Boehm GC thread, 
> > but I don't know whether this is easily doable.
> >
> > What's the best approach to solving this problem?  Thanks in advance.
> >  
>
> It should be pretty trival to fix boehm-gc to block SIGCHLD. Basically 
> you should just need to copy the pthread_sigmask() code from 
> block_sigchld in posix-threads.cc into GC_pthread_create in 
> solaris_pthreads.c. I think it would be fine to make that change in 
> order to fix this problem, so please submit a patch.
>
> This problem does expose a underlying flaw in our java.lang.Process 
> implementation, however - If an application creates non-Java threads 
> independently of libgcj, then ProcessManager may break as the signals 
> could be delivered to the wrong thread. Unfortunately nobody has come up 
> with a better solution for this, yet.

I agree.  This is (somewhat) documented in gcj.texi (@subsection Posix 
signals)  But we should probably add that SIGCHLD must be blocked in all 
non-Java threads.

David Daney