[omniORB] Possible deadlock / hang in wait_for_detached_objects()

[Harri Pasanen]

Hi,

I run into situation where an application sometimes hung at exit, and 
traced it to the following call sequence:

omniOrbPOA::deactivate_object(const PortableServer::ObjectId& oid)
  -> calls  detached_object();
omniOrbPOA::lastInvocationHasCompleted(omniLocalIdentity* id)
  -> calls  servant->_remove_ref();
  -> calls met_detached_object();

The servant->_remove_ref(); can throw an exception, in my case it was 
CORBA::OBJECT_NOT_EXIST, and as a result met_detached_object() was 
never called, and pd_nDetachedObjecs was not decremented.

The app would then hang until kingdom come in:

void
omniObjAdapter::wait_for_detached_objects()
{
 ...
  while( pd_nDetachedObjects )  sd_detachedObjectSignal.wait();
  ...
}

In my case I could catch the exception in my servants _remove_ref, but 
as this is a recurring pattern in our code, it would make life easier 
if the OmniORB code would catch the possible exception.  I would also 
think it to be conceptually the correct thing to do, as the uncaught 
exception corrupts OmniORB's internal state.

Duncan, what is your opinion?

Regards,

Harri