[FFmpeg-devel] [PATCH 2/2] avcodec: remove sonic lossy/lossless audio

[Lynne]

Mar 1, 2024, 21:33 by anton at khirnov.net:

> Quoting Lynne (2024-03-01 20:41:53)
>
>> Mar 1, 2024, 19:59 by anton at khirnov.net:
>>
>> > Quoting Lynne (2024-02-29 14:30:55)
>> >
>> >> We offer support for it. Someone must have used it.
>> >> I'm sure there are less used game codecs we support than Sonic.
>> >>
>> >
>> > [citation needed]
>> >
>> > Is there any evidence that it was used for anything other than
>> > development and testing? That anyone has any actual samples in it they'd
>> > like to play?
>> >
>>
>> There's always the possibility that someone encoded samples years
>> ago and deleted the originals. Same as with Snow.
>>
>> My only concern is that this sets a sensitive precedent to get rid of
>> old codec code with low weight under unclear conditions. Where do
>> we draw the line?
>> SMPTE? ISO/IEC standardization? IETF? AOM?
>> Having *a* spec (Sonic doesn't, but neither did FFv1 until a few years
>> ago)?
>> A single company pushing for it many years ago with no spec,
>> like SpeedHQ, or the many niche lossless/intra-only codecs?
>>
>> Or simply having one single implementation in FFmpeg, and no spec at all,
>> and having an experimental flag, and no activity for a very long amount of time?
>>
>
> I'd draw the line somewhere around what I said above - some evidence
> that it was used "seriously".
>

That is something I cannot agree with - it cuts off 70% of all codecs we support,
and zeroes out any potential chance of any codecs being developed by us.


>> If it's the last case, or simply being a one-off, I can agree with deprecating
>> and remove it next bump. But if the developer thinks they will have time and
>> have motivation to work on it in the future, I think we should leave it at just
>> deprecation plus disabling  building the encoder by default, until the issue is
>> brought up again at say, after the next version bump.
>> After all, CELT started out more than 10 years before Opus was standardized,
>> and its EC code has made it all the way to AV2 despite its core remaining
>> visibly similar over the years. And FFv1 was used as an inspiration for FLIC,
>> which went on to become JPEG-XL.
>>
>
> Sure, but there was active interest in all of these. I see no active
> interest in Sonic.
>

No one had any interest in CELT for years until it was used in Mumble,
or Speex initially, or FLIC, as it they were not competitive enough on its own,
and Vorbis, or Opus for the first few years.

The developer has said there's interest in continuing to work on it
in the future.
I don't see why we shouldn't give him a chance to work on it for
a bit longer before we bring it up again for removal. It's a simple
codec with no oddities like self-modifying assembly, codec to
demuxer to codec avpriv interaction or special API user considerations.