Re: [g-a-devel]Accessible_unref() and leaks

[Bill Haneman <bill haneman sun com>]

On Tue, 2002-09-17 at 12:00, Michael Meeks wrote:
> Hi Padraig,
> 
> On Tue, 2002-09-17 at 07:58, Padraig O'Briain wrote:
> > While sweeping at-poke for leaks I found that I needed to call 
> > Accessible_unref() on the return value from Accessible_getChildAtIndex() but not 
> > from the return value from Accessible_getParent().
> 
> 	That sounds _extraordinarily_ odd;
> 
> 	If you look at at-spi/libspi/accessible.c (..._get_parent) you see it
> calls spi_accessible_new_return with a FALSE release_ref - such that
> this doesn't release the ref on the parent - I imagine because I checked
> that atk_object_get_parent returned a ref on it's parent;

I take it you mean that you saw that "atk_object_get_parent () did not
increment the parent ref-count", thus the spi_accessible_new_return()
call didn't need to unref() the source object after the call; 

the reason for spi_accessible_new_return()'s "unref" of the source
AtkObject being that either a pre-existing SpiAccessible is being
reused, or a new one created (with a consequential incrementing of the
AtkObject's ref count); in both cases an unref() would be appropriate to
balance the ATK ref-count.

[That is my recollection of the logic anyway...]

> 
> 	It seems to do the CORBA ref counting just fine.

Yes, it seems so to me too, but it can get a little confusing, tracing
the ATK refcounts with respect to the Bonobo refcounts and the separate
issue of SpiAccessible creation/reuse.
 
So, worth digging a little deeper I guess.

-Bill

> > Is this what we expect to happen?
> 
> 	No; but I don't see how the CORBA ref count can be wrong off hand;
> which ref count is not getting set correctly ?
> 
> > If not, should this be fixed in at-spi or in gail?
> 
> 	Hard to say.
> 
> 	Incidentally, if you're beggining to notice performance issues with
> un-reffing stuff; there are 2 strategies I think we could take:
> 
> 	a) Idle unreffing of the cspi reference cache;
> 	b) Async unrefs.
> 
> 	b) Is pretty easy to implement, I plan to do it in libbonobo for 2.2;
> such that a bonobo_release_unref (obj, NULL); will never block /
> re-enter.
> 
> 	HTH,
> 
> 		Michael.
> 
> -- 
>  mmeeks gnu org  <><, Pseudo Engineer, itinerant idiot
> 
> _______________________________________________
> Gnome-accessibility-devel mailing list
> Gnome-accessibility-devel gnome org
> http://mail.gnome.org/mailman/listinfo/gnome-accessibility-devel