Re: GCJ + Roguewave Threads?

[Bryce McKinlay <mckinlay at redhat dot com>]

Johann Hibschman wrote:

> > I think we still have a problem with notifying the GC of a new
> > thread, if threads are created without going through GC_pthread_create.
> > I need to find time to work on that.
> >    
>
> Do you know of any documentation or examples of using
> GC_pthread_create?  I suppose I'll just look through the source
> tomorrow.
>
> To return this to gcj, the correct way to use multiple threads with
> one JVM is to call JvCreateJavaVM once in the main thread, then create
> subsequent threads via GC_pthread_create and call
> JvAttachCurrentThread from within those threads?
>  
GC_pthread_create() is just a wrapper around pthread_create(), so its 
arguments and usage are the same as for pthread_create().

> > If the Roguewave implementation is in header files, and you can
> > recompile the files that include it, you might be
> > able to work around the problem by defining GC_THREADS and including
> > gc.h before you include those header files.  That will redefine
> > pthread_create to the wrapper that notifies the GC.
> >    
>
> I don't believe the implementation is all in the headers, but do I
> have to define GC_THREADS in any gcj/c++ program that uses then? 
>  
No, you'd only need to do this in the code which is calling 
pthread_create().

> My hair-brained scheme for tomorrow is to see if I can launch a
> proper, GC-registered pthread from the main thread, and then use that
> thread for all subsequent gcj work, leaving the Rogue Wave threads to
> do their dirty business in the original thread.
Seems reasonable - just be sure to synchronize the threads correctly if 
you take this route.

Bryce