[vlc-devel] [vlc-commits] aout: do not use input resource for visualization

[Rémi Denis-Courmont]

Le maanantaina 21. tammikuuta 2019, 13.50.58 EET Thomas Guillem a écrit :
> On Mon, Jan 21, 2019, at 12:11, Rémi Denis-Courmont wrote:
> > Hi,
> > 
> > I don't see why filters should need to pass by decoders, or access the
> > clock (which might not even exist - unpaced sout), as they are much
> > better off with original timestamps.
> > 
> > 
> > But more prosaically, I don't see the link with this patch, or even the
> > whole patch series, which affects only visualisations, not audio filters.
> OK, audio filters will generally only use original timestamps.
> Visualization filters are audio filters, right ?

That's oversimplifying. Filters do not have a request_vout while 
visualizations are not allowed to change the format or payload. It's just 
historical convenience to treat them in the same filter chain.

> They will need to access
> the main clock in order to synchronize with it. If the user change the
> audio delay, the visualization filter need to know it.
> 
> If I remember correctly, we discussed about it during last VDD. You first
> disagreed with me, then when agreed that timestamps should not be touched
> until output. The problematic point, that you mentioned, was how to
> synchronize visualization filters.

> This is what my patch was trying to solve.

I don't see how the request_vout would have helped here given that most 
visualizations already did not use it, but use the vlc_gl_* API instead. It 
does not make practical sense to standardize on the non-GL visualizations at 
this point. I might have liked it, but it is ostensibly never going to become 
the norm back again.

And having said that, I do not have any good idea how to solve the problem of 
synchronizing OpenGL with audio. I am not familiar with OpenGL, and it was not 
my idea to use it for visualizations, so I feel neither competent nor 
responsible here.

-- 
Реми Дёни-Курмон
http://www.remlab.net/