[omniORB] Memory leak related to Proxy Object Factories

[Tuyen Chau]

I need some help tracking down what appears to be a memory leak deep in 
the proxy object factories (POF), or how they interact with other 
components in omniORB.  In our application, we derive from the default 
POFs to instantiate our own proxies.  This all works fine.  However, 
when we take another step in our proxy object factories and try to cache 
our proxies so that there is a unique object reference for each servant, 
we ran into a memory leak of about 600-700 bytes, each time our POF is 
called. 

Here's how we cache the object.  We use the key in the omniIdentity to 
identify object.  If an object has been created for that key, we 
increment the reference count on the object and return it.  If not, we 
create a new object and cache it.  Our newObjRef(() function looks 
approximately like this,

OurPofClass::newObjRef(omniIOR* ior, omniIdentity* ident) {

    unsigned char *key = (unsigned char *)ident->key();
    int keySize = ident->keysize();
    omniObjRef obj;

    obj = FIND_IN_CACHE(key, keySize);

    if (obj == NULL) {

        obj = new OUR PROXY();

        CACHE(obj, key, keySize);

    } else {

        ClassName::_duplicate(dynamic_cast<ClassName_ptr>(obj));
    }

    return obj;
}

As far we can tell, there is no leak in our cache implementation, which 
really is a hash table.  If we comment out just the line of code that 
puts the object in the cache, the memory leak is gone.  If we put the 
object in the cache, the memory leak is back.  It seems that omniORB 
checks on the return object and do something different if the object is 
not "brand new", and that something "different" is causing the memory 
leak.  But then we are puzzled as to how omniORB can even tell if the 
return object is "brand new" vs. cached.

Does this make sense to any of you?  Does anyone know how can we fix 
this memory leak?

Best regards,
Tuyen