[FFmpeg-devel] [PATCH 01/24] avcodec: add color_range to AVCodec struct

[wm4]

On Wed, 2 May 2018 13:44:48 +0200
Nicolas George <george at nsup.org> wrote:

> Vittorio Giovara (2018-05-01):
> > Well no, let's step back a little.
> > 
> > First of all there is no stigma about adding fields to AVCodecContext,
> > in fact, the more, the merrier, right?
> > 
> > Secondly, there _is_ concern about adding a field to AVCodec (not avctx),
> > since is a delicate point of entry, and very sensitive to ABI (which in fact
> > was broken in the first iteration of this patch).
> > 
> > Finally, I immensely disagree that this is a good API at all, just throwing
> > fields to structure is rarely a good approach to anything. On a separate note
> > I also disagree that color_range (despite its importance) is an "essential
> > property of images", at least not any more than the other color propeties:
> > it is simply being noticed because there are special purpose enum values, but
> > that doesn't mean that the other properties will work "just fine". Because
> > they don't.
> > The transfer characteristics for example is a much more impactful and
> > delicate property that can affect the the color of the image in a much
> > more profund way than color range.
> > Similar arguments could be made for primaries and matrix.
> > 
> > As much as you may hate hear it, this proposed filtering system does *not*
> > belong to neither lavfi, lavf or lavc, but rather to the scaling library.
> > Unfortunately the one used by default, swscale, is a monster spaghetti code
> > and too difficult to work around. That however should not affect the design
> > of APIs used in the other libraries: if swscale is not suited anymore for
> > its job it should be replaced, you shouldn't be adding workarounds to AVCodec.  
> 
> Thanks for taking the time to express the reservations. I fully agree
> with what you wrote.

Please don't prevent us from fixing libavfilter.