Comment 52 for bug 948788

(In reply to alexander :surkov from comment #39)
> (In reply to Trevor Saunders (:tbsaunde) from comment #38)
> > (In reply to alexander :surkov from comment #37)
> > > atk_object_set_name() called inside get_name is a wrong thing we try to
> >
> > I think the only thing we try to fix here is the infinite recurssion.
>
> and code madnness :)

why? the reason for the bug was crashes with a change to atk, so just fixing the way we interact with atk should be fine.

> > > remove here. It made us fire name change events which is ridiculous I think.
> > > If I get right then ATK implementation internals don't need that event as
> > > long as we override get_name. On the another hand I don't understand why the

I believe it needs event to keep cache in sync with reality.

> > > consumer might need name change event when it asks for the name.

what if there is more than one consumer, then it may be arguably less bad for other consumers to get the event late rather than never.

> > I absolutely agree what we do is crazy, but I know there is caching involved
> > and I'm atleast somewhat concerned not fireing an event in getName() could
> > break that even more than it is now.
>
> what kind of caching? And how does this caching is supposed to work if
> somebody asks us to calculate the name (bypassing that cache)? Or
> alternatively who uses that cache and why all consumers don't want to use it?

I think the way it works is that consumer processes have a local repreentation of atkobject in their process which keeps the name and atk updates the name based on name change event. So each consumer can choose for itself to use or not use cache as it likes, and might well want to not use the cache because it can easily become out of date due to us not always firing name change events.

> > If your offering to fix name change
> > events so they're fired whenever a name changes and never when it doesn't
> > then of course I'm happy to remove this madness.
>
> iirc AT needs this event every time when name is changed. In this sense our
> name change event might never work for this purpose.

its possible

in any case I'm not really willing to remove the madness we have now atleast until we talk to atk people about it, and I don't see a reason to make vd block on that.