Il mar 29 apr 2025, 16:37 Marc-André Lureau <marcandre.lur...@redhat.com>
ha scritto:
> Hi
>
> On Tue, Apr 29, 2025 at 6:23 PM Paolo Bonzini <pbonz...@redhat.com> wrote:
> >
> >
> >
> > Il mar 29 apr 2025, 16:03 <marcandre.lur...@redhat.com> ha scritto:
> >>
> >> From: Marc-André Lureau <marcandre.lur...@redhat.com>
> >>
> >> Since commit 9859fac ("object: release all props"), the code was changed
> >> to tracks the already released properties in a hash table. I am not sure
> >> why this was done, perhaps to prevent from potential crashes if
> >> properties are being added dynamically during release. I am not sure if
> >> it's a valid concern though.
> >
> >
> > You always need object_property_iter_init in case prop->release deletes
> a property, thus invalidating the GHashTable iterator. The hash table
> instead is needed in case prop->release does *not* delete a property,
> because then the property reappears on subsequent recreations of the
> iterator.
>
> Yes, changing the properties while they are being removed is I think
> the reason I added the HashTable/Set. But is this a valid concern?
> (the penalty is a bit sad)
>
I think deleting is a possibility... For example releasing a property could
trigger an object_unparent(). Since there is no better way to notice I
think it's better to keep it.
Maybe we could add a generation count to the Object struct?
Paolo
>
