[FFmpeg-devel] [PATCH 001/279] Add a new channel layout API

[Anton Khirnov]

Quoting Marton Balint (2021-12-08 11:55:37)
> 
> 
> On Wed, 8 Dec 2021, Lynne wrote:
> 
> > 8 Dec 2021, 11:06 by cus at passwd.hu:
> >
> >>
> >>
> >> On Wed, 8 Dec 2021, Lynne wrote:
> >>
> >>> 8 Dec 2021, 10:22 by h.leppkes at gmail.com:
> >>>
> >>>> On Wed, Dec 8, 2021 at 10:14 AM Lynne <dev at lynne.ee> wrote:
> >>>>
> >>>>>
> >>>>> 8 Dec 2021, 02:06 by jamrial at gmail.com:
> >>>>>
> >>>>>>
> >>>>>> +enum AVChannel {
> >>>>>> +    ///< Invalid channel index
> >>>>>> +    AV_CHAN_NONE = -1,
> >>>>>> +    AV_CHAN_FRONT_LEFT,
> >>>>>>
> >>>>>
> >>>>> No, not the pixfmt mistake again. Set AV_CHAN_NONE to 0,
> >>>>> the rest can follow. Or keep AV_CHAN_NONE to -1
> >>>>> and add a new AV_CHAN_UNSPECIFIED as 0.
> >>>>>
> >>>>
> >>>> Care to elaborate on the reasons of this opinion? Using -1 as invalid
> >>>> and 0...x as valid entries seems quite reasonable to me.
> >>>>
> >>>
> >>> Zero-initialization. I've had issues in the past telling
> >>> YUV420P from <uninitialized>.
> >>>
> >>
> >> It is convenient to be able to set the channel layout mask by a simple byte shift of the ID-s.
> >>
> >> AV_CH_FRONT_LEFT = 1 << AV_CHAN_FRONT_LEFT.
> >>
> >> So I'd say that is the reason that it is designed that way, because this way existing channel layout masks can be kept without breaking ABI.
> >>
> >
> > Those can be set with offsets, e.g.
> > AV_CH_FRONT_LEFT = 1 << (AV_CHAN_FRONT_LEFT - 2).
> 
> Well, I find this less fortunate then the need to initialize NONE to -1...
> 
> > I'd like to not have weird values in enums because the old
> > API, just being deprecated, must have its way and can't deal
> > with an offset.
> 
> AV_CH_FRONT_LEFT and similar is not deprecated, these are still used for 
> the native channel order which is still using a mask.

Yes, the direct mapping between channel indices and WAVEFORMATEXTENSIBLE
values is a design goal here, just as it was in the old API.

I also belive that the initialization issue of pixel/sample formats is
less of a problem here, since you rarely need to store actual channel
ids. So IMO retaining channel index compatibility is more important.

-- 
Anton Khirnov