[omniORB] How to get rid of the POA?

[Andreas Bille]

Hello Duncan,

Duncan Grisby wrote:

>On Monday 15 September, Andreas Bille wrote:
>
>  
>
>>spinning around a little bit, I found no solution to the
>>(beginners?) problem to peform several run/destroy-cylces with omniORB,
>>if during such a cycle the POA is resolved.
>>    
>>
>
>What are you really trying to achieve?  The test program you posted is
>pretty weird. What is the use case you are experimenting with?
>
well, I want to write some program which can be configured to run as a 
CORBA-server in addition to being a client. There is some GUI to set the 
port of that server and start/stop it, which poses the problem to react 
to changes of the port. I see four options:
1. Don't allow changes at all ( not an option, despite the fact that 
clients must update their 'server-knowledge').
2.  Any change is effectiv only after restart of the whole program.
3. Any change is effectiv after restart of the thread which runs the orb 
( marvellously, running comniORB in a secondary thread works, though it 
is non-portable. )
4. There is no need to restart at all, because there is an API to add 
and remove endpoints during orb-runtime.

 I have tried to implement option 3, which would enable reconfiguration 
of the port during runtime of my program at least, though of course 4 
would be nicer. But I have found it to be impossible due to the 
described problem.
So much to my usecase ( and the reason for the small testprogram )

>
>I think the problem you're seeing is that the OS hasn't caught up with
>the closure of the listening socket by the time you try to reopen it,
>but I fail to see why you're wanting to do that anyway.
>
Hm, don't know the internals of omniOrb, but thought that indeed the 
problem is that some static variables are not (yet?) cleaned up. Putting 
some time interval between the first call of shutdown and the second 
call of the constructor of the 'Shutdown'-class had not made the problem 
disappear, so it might be the case that it is not an OS problem.

Greetings,
Andreas

>
>Cheers,
>
>Duncan.
>
>  
>