Please report new issues athttps://github.com/DOCGroup
Smart proxy objects are never _non_existent(), with the result that you can test for existence, get a 'still exists' response then get OBJECT_NOT_EXIST thrown when you make a subsequent remote invocation. Also, object_to_string() throws a MARSHAL exception for a smart proxy object. Both problems may stem from the fact that the smart proxy object is a 'local' object, i.e. attribute 'is_local_' is set to true. Is this the expected behaviour? PS My TAO installation includes bug# 2245 fix (Smart proxy destructor does not run). Thanks for your help, Andrew
To be more precise: I find that a smart proxy always returns false to _non_existent() even when a dumb proxy for the same object returns true (e.g. when the associated CORBA object has been deactivated). Also I should say the smart proxy is a 'locality-constrained' object rather than 'local'.
can you provide an automated regression to reproduce this?
because things are now derived from local object besides the is_local_ several more operations do behave different, see tao/Local_Object.cpp
Created attachment 486 [details] Trivial test zipped: IDL, server/client & smart proxy source (C++), README
added depends on 2245, this seems to change the behaviour. I have my doubts about the fix of 2245, adding this extra base does much more then just a simple refcount.
Created attachment 487 [details] README.txt missing from previous attachment
I have committed a change that allows users to override the default IOR conversion for Smart Proxies. Users can provide an alternate implementation for "virtual bool can_convert_to_ior (void) const" in their Smart Proxy implementation to allow the object reference to be converted to an IOR. Mon Nov 27 12:50:48 UTC 2006 Chad Elliott <elliott_c@ociweb.com> * tao/Object.h: * tao/Object.cpp: Provide hooks to enable custom Object to IOR conversion or allowing local objects (such as Smart Proxies) to be converted to an IOR string. * tao/ORB.cpp: Utilize the new hooks when performing the Object to IOR string conversion. * tests/Smart_Proxies/Smart_Proxy_Impl.h: * tests/Smart_Proxies/Smart_Proxy_Impl.cpp: * tests/Smart_Proxies/client.cpp: * tests/Smart_Proxies/run_test.pl: This test has been modified to implement both aforementioned hooks and perform comparisons to ensure that the right thing has been done.
Disadvantage is that the addition of this virtual method seems to increase footprint between 1 to 2% for a lot of files in the core libs.
Now localobject are refcounted by default, do we need to have the mentioned change, the can_convert_to_ior method is now the only real user of the local flag in object.
(In reply to comment #9) > Now localobject are refcounted by default, do we need to have the mentioned > change, the can_convert_to_ior method is now the only real user of the local > flag in object. > I may have missed your point, but I don't see what having the local object being ref counted has to do with the _is_local() method. Prior to the addition of the can_convert_to_ior() method, the ORB called _is_local() on the Object directly.
I am looking if we still need the is_local flag in object, it was used in the past in object itself to control whether we have refcounting or not. Now we always have refcounting I wonder if we still need this flag or whether there are other ways to control this.
(In reply to comment #11) > I am looking if we still need the is_local flag in object, it was used in the > past in object itself to control whether we have refcounting or not. Now we > always have refcounting I wonder if we still need this flag or whether there > are other ways to control this. > Oh, ok. Well, then I would think that it is still needed. I don't know of any other way to tell if an Object is local or not.