[vlc-devel] commit: Implement access_GetParentInput and demux_GetParentInput and use. ( Pierre d'Herbemont )
pdherbemont at gmail.com
Wed Aug 26 21:11:19 CEST 2009
On Aug 26, 2009, at 8:01 PM, Rémi Denis-Courmont wrote:
>> Are we not supposed to do that? Fine, but don't let it compile
> How do you know that the caller is right or wrong?
Just don't allow vlc_object_hold() on it. Then you are safe. Or make
sure that you allow p_demux->psz_path to be NULL, and don't set a
random value to it.
What I am trying to point out, is that our API doesn't reflect the
reality at all. You can vlc_object_hold() an object, but that won't
actually let you live with a valid reference of it, with apparently no
way to know it.
Either choose, we fix demux_Delete() or we make sure vlc_object_hold()
doesn't work on that object.
> If you have solved the halting problem, I'm all ears.
>> So, if they don't need to be refcounted and doesn't support
>> refcounting, vlc_object_hold() should be disabled on those objects
> The private data does not support reference counting. It is completely
> impossible. We need to attach an object *before* we run
> module_need() on it,
> so that variables can be propagated in pf_open. At the same time, as
> soon as
> an object is attached, it becomes visible from vlc_object_find(),
> vlc_object_find_name() and vlc_list_children().
Well, I thought we were going away from this. ie, we simply let owner
decide to expose the object when it is ready, and try to remove those
Refcounting is only mostly useful in term of *memory data*.
module_need() and module_unneed() seems to be start() and stop()
counter part data that will actually perform some work. You wouldn't
want to refcount them.
> So evidently, anything that pf_open initializes or modifies is
> outside the
> scope of reference counting. From access_t, p_module, p_sys,
> pf_read, p_block, psz_access, psz_path and info are all in that case
> instance. Besides, we really don't want to make all object callbacks
> entrant. It would be too complicated, error-prone and inefficient.
Err. Creating something outside refcounting is just fine, provided
that access to the data is thread safe and that you don't let a random
pointer value when you free it. This is the problem I am stating above.
>> They are multiple owners when you do refcounting.
> I stay by my statement that pf_destructor is almost completely
> useless. It's
> only useful for those few object types with type-specific thread-
> safe data.
It's useful if you want a set of data to be freed when it is not
needed. ie - when you expect to do refcounting on some memory object.
More information about the vlc-devel