[vlc-devel] Status of Win32 branch nightly

[Laurent Aimar]

On Wed, Sep 10, 2008, Rémi Denis-Courmont wrote:
> 
> On Wed, 10 Sep 2008 11:05:37 +0200, Marian ??urkovi?? <md at bts.sk> wrote:
> > On Wed, Sep 10, 2008 at 10:30:33AM +0200, Derk-Jan Hartman wrote:
> >> On 10 sep 2008, at 08:04, Marian ??urkovi?? wrote:
> >> > - we still have crash on AR change
> >>
> >> Did we have a crashlog of this ?
> >> We should really see if there is not a way to fix this.
> > 
> > Yes, there is - vout shouldn't be destroyed and recreated due to AR
> > change.
> 
> I am not an expert, but I would expect that this is impossible from the
> video output drivers. Is it not the video codec responsibility to create a
> new vout when the AR changes anyway?
 Decoders are responsible for fmt_out modifications and as soon as they do
(width/height/chroma/rgb masks/ar) they know that the vout will be
recreated the next time they call pf_vout_buffer_new.

 With that knowledge, they MUST release all pictures they are referencing
(if any) (they probably need to copy them into new vout pictures to avoid
breaking decodings on AR changes).

 A decoder keeps a reference by calling pf_picture_link. I am pretty sure that
all of our decoders using pf_picture_link while modifying fmt_out are buggy...
- libmpeg2 vlc wrapper is one of them.
- avcodec with direct rendering is probably buggy too.

-- 
fenrir