After talking to Andrew, I realized I was off-base here.
> This looks a bit suspicious. Should we be also comparing the
> pointer, not just the instance pointer here?
Assuming that only one object can have the instance it shouldn't
matter. When the object or the instance are removed, the property
change record is also removed so it won't accidentally point at a
I for some reason was thinking we're tracking l_dbus_interface objects,
not interface_instance. So after being being reminded of that, the code
is just fine.