On Thu, May 22, 2014 at 12:05 AM, Paolo Bonzini <pbonz...@redhat.com> wrote:
> Il 21/05/2014 22:22, Stefan Hajnoczi ha scritto:
>> +void object_property_add_alias(Object *obj, const char *name,
>> + Object *target_obj, const char
>> *target_name,
>> + Error **errp)
>> +{
>> + AliasProperty *prop;
>> + ObjectProperty *target_prop;
>> +
>> + target_prop = object_property_find(target_obj, target_name, errp);
>> + if (!target_prop) {
>> + return;
>> + }
>> +
>> + prop = g_malloc(sizeof(*prop));
>> + prop->target_obj = target_obj;
>> + prop->target_name = target_name;
>> +
>> + object_property_add(obj, name, target_prop->type,
>> + property_get_alias,
>> + property_set_alias,
>> + property_release_alias,
>> + prop, errp);
>> +}
>
>
> ref target_obj here, and unref in property_release_alias?
No. I originally did this but realized it is probably not a good idea.
1. If you want to alias a property on the same object instance (foo.a
-> foo.b) this would create a refcount leak (unless someone explicitly
deletes this property to bring the refcount back to 1).
2. The virtio callers already have a child reference and I suspect
other callers would too. Therefore we don't need to do extra
refcounting.
I had an intermediate version of this patch with a flag so you could
tell object_property_add_alias() whether or not to ref the target
object. But in the end it seems like overengineering since the
refcount case is rare or non-existent. The refcount is a convenience
feature just for the case where you want to alias to a random object
that you don't otherwise hold a refcount on (I couldn't think of an
example).
Do you see what I mean?
Stefan
