[vlc-devel] [RFC PATCH 1/3] picture: add display orientation

[Rémi Denis-Courmont]

On mardi 2 janvier 2018 17:44:04 EET Thomas Guillem wrote:
> On Tue, Jan 2, 2018, at 17:41, Rémi Denis-Courmont wrote:
> > Le 2 janvier 2018 17:17:12 GMT+01:00, Thomas Guillem <thomas at gllm.fr> a 
écrit :
> > >On Fri, Dec 22, 2017, at 20:30, Rémi Denis-Courmont wrote:
> > >> Le torstaina 21. joulukuuta 2017, 18.08.53 EET Thomas Guillem a écrit
> > >> 
> > >> > In order to allow filters to modify the orientation without
> > >
> > >touching picture
> > >
> > >> > data.
> > >> 
> > >> You can already do that by changing the output format of the filter.
> > >
> > >Sorry for the late reply. If you change the filter fmt_out, the core
> > >will try to add a filter to compensate for format changes. So you'll
> > >have the first filter that will transform W x H to H x W and the adapt
> > >one that will try to fit H x W into W x H. It works for cpu formats,
> > >but do a lot of memcpy. I don't think we can adapt our usecase here.
> > >
> > >> There are likely still some broken assumptions in the filter chain
> > >
> > >and/or vout
> > >
> > >> code, either way.
> > >
> > >_______________________________________________
> > >vlc-devel mailing list
> > >To unsubscribe or modify your subscription options:
> > >https://mailman.videolan.org/listinfo/vlc-devel
> > 
> > There are no (legitimate) reasons why the picture would be rescaled.
> 
> if (!es_format_IsSimilar(p_fmt_current, &fmt_target)) {
>   msg_Dbg(vout, "Adding a filter to compensate for format changes");
>   ..
> }
> 
> I guess this was done because we didn't want to restart the vout.
> 
> > That would be a bug just as well on CPU surfaces. And there are no risks
> > of memcpy() penalties here; that problem only exists for CPU surfaces.

As noted earlier, there are bugs blocking this otherwise seemingly trivial 
change either way. No news here. The lack of transform with hardware decoding 
has been invented and discovered in 2.2 cycle already. The ugly and 
inefficient double rescaling even before that.

I am NOT putting up with well-known issues getting rebranded as critical 
blockers months into the code freeze. Not going through the "resume from 
whence you left" disaster again. Just no.

-- 
Rémi Denis-Courmont