[omniORB] Exceptions in constructors

[Duncan Grisby]

On Wednesday 28 November, David Hyde wrote:

> A CORBA object that I am working on is written in such a way that the
> constructor is liable to throw an exception in certain circumstances.  (This
> is legacy code that I avoid making fundamental architectural changes to
> unless there is a good reason).  I have converted this code from using a
> proprietry ORB to use OmniORB and it used to work.

[...]
> I've a feeling that I am going to have to rewrite the code so that the
> constructor doesn't throw an exception, but is this right???  Presumably the
> ORB thinks that it has a reference to an object even though it hasn't been
> activated in the POA and it is crashing when it is erroneously trying to
> destroy it.  What happens if an exception that is out of my control is
> called while I'm in my constructor?  Do I have to put try - catches in all
> of my constructors?

The problem is that the POA thinks the object _has_ been activated. It
is therefore trying to deactivate it, which involves accessing a data
member inherited from omniServant. Since the object doesn't actually
exist, it understandably blows up.

So, the question is why the POA thinks the object was successfully
activated. I suspect your constructor calls _this(), which implicitly
activates the object, before the exception is thrown. That way, the
object is never fully constructed, but it finds its way into the POA's
active object map. The solution to that is to avoid activating the
object in its constructor.

If that isn't the problem, can you post a minimal example program that
exhibits the problem?

Cheers,

Duncan.

-- 
 -- Duncan Grisby  \  Research Engineer  --
  -- AT&T Laboratories Cambridge          --
   -- http://www.uk.research.att.com/~dpg1 --