Hi!
marc fleury wrote:
> > > So what's the problem ? Is it something which deals with the hack which
> > > optimizes intra-VM calls ?
> >
> > Yes. The interface is loaded twice by different classloaders. The fix is
> > to not only detect in-VM, but also see if it is the proper classloader.
> > In all other cases a copy must be made.
> >
> > I will look into this.
> >
> > You can make this work right now by simply turning off local
> > optimizations in the container configurations (I think, let me know if
> > it doesn't work anyway).
>
> well part of the reason of why we do this integration with Tomcat is that we
> want to see the InVM stack invocation from servlet to bean (so we can adress
> issues like "how come Orion is in VM and not Tomcat+jboss") so that is
> coming up and we will release a binary for distribution very soon. The
> native pointer passage is an important speed up option we need to turn on
> for these tests.
Currently this would be impossible. Tomcat uses one classloader and
jBoss/EJB another. In order to pass stuff between these two we *have to*
do serialization copying.
The workarounds are:
* Make Tomcat and jBoss use the same classloader (very difficult)
* Put bean interfaces in server classloader (takes away hot deployment)
Pick one :-) I think both are bad.
/Rickard
--
Rickard �berg
@home: +46 13 177937
Email: [EMAIL PROTECTED]
http://www.telkel.com
http://www.jboss.org
http://www.dreambean.com
