Re: Leaked refs. How to debug?
Jules Colding <[email protected]>
| Newsgroups | gmane.comp.gnome.orbit.general |
|---|---|
| Organization | OMC Denmark ApS |
| Message-ID | <[email protected]> |
Hi Michael, On Thu, 2005-02-10 at 17:32 +0000, Michael Meeks wrote: > On Sun, 2005-02-06 at 01:18 +0100, Jules Colding wrote: > > ** (process:30027): WARNING **: ORB: a total of 2 refs to ORB objects were leaked > > 1 refs to 'Object (type '')' > > 1 refs to 'ORB' > > 1 refs to 'Object (type '')' > > > > from my client/server application at exit. I have looked a million times > > and I am rather sure that no objects are ref'ed more than they are > > unref'ed. > > ;-) then quite possibly this is an ORB bug - in theory the ORB handles > its referencing perfectly. I think ORBit is off the hook too. The leaks are only present if I invoke ORB_init() and ORB_run() in separate threads. > > I have tried to Object_release() the orb just before the ORB_destroy() > > to see what would happen. I got: > > I would copy some of the samples that do it correctly; eg. test/poa/ > has a set of simple things like that I think, > > > So I am reasonably sure that I am releasing the orb as I should. Any way > > I can get more information of the remaining refs? Ref'ed line or address > > would be great. > > I guess, we could prolly easily add a walk of the Object handle hash > tables and print their type_ids that would prolly help :-) I guess you > want to walk the orb->objrefs hash and dump g_quark_to_string > (obj->type_qid) I guess. This is what is currently done, if I remember correctly. I suspect this to be more of a glib issue than an ORBit2 issue, but I have been wrong before ;-) I'll look some more into it and report any findings to the list. Best regards, jules