[omniORB] trying to interpret omniORB debug traces

[Stefan Seefeld]

Stefan Seefeld wrote:

> Where is the error ? What causes the 'WouldDeadLock' exception
> to be thrown ?

sorry, some clarification may be in order:

I create a servant temporarily inside the server application
from within a request. I create a new POA and make it a member
of that servant. No objects are ever activated with that POA.

Finally, just before that request is over, I deactivate (and
thus implicitely delete) my servant, resulting in the

KitImpl::~KitImpl destructor to be called, which calls

_poa->destroy(true, true);

Again, that POA doesn't hold any objects, so it can be
deactivated and etherealised immediately (well, the _poa
member is a proxy still holding a reference, so the etherealising
only happens when my KitImpl destructor quits and '_poa' gets
'out of scope'.

That's all. Now I'm reading the following comment in poa.cc:

       // In the context of an operation invocation.

       // This is interesting. It would be sufficient (but harder) to
       // only complain if in the context of an invocation in _this_
       // POA. However, the 2.4 spec says "...some POA belonging to the
       // same ORB as this POA...". Since we only ever support one ORB,
       // life is easy.

       OMNIORB_THROW(BAD_INV_ORDER,
		    BAD_INV_ORDER_WouldDeadLock,
		    CORBA::COMPLETED_NO);

which seems to suggest that an 'BAD_INV_ORDER' is thrown if *ever*
a poa is (attempted to be) destroyed while inside a request. Why ?
The request doesn't even involve the poa I'm trying to destroy,
so there isn't any deadlock in sight !

What am I doing wrong ? When am I allowed to call 'poa->destroy(true, true)' ?

Thanks for your help !

Stefan