[FFmpeg-devel] remove DEC Alpha DSP & support code

[Rémi Denis-Courmont]

Le 11 juin 2024 12:59:23 GMT+03:00, Michael Niedermayer <michael at niedermayer.cc> a écrit :
>On Mon, Jun 10, 2024 at 08:52:08PM -0400, Sean McGovern wrote:
>[...]
>> Are there any real concerns about the Alpha removal itself?
>> People still wanting to use FFmpeg for hardware that old can stick
>> with 7.0 (and fork it if they like -- that's the beauty of FOSS).
>
>Loosing security support, sounds not viable, so if alpha is removed
>the question what that would do to users (aka performance and does it
>work/build after the patchset) is still an open question ...

What supported distribution would people even be getting security support from? None of the mainstream distributions support Alpha anymore. So if (generic) you care about security support that architecture simply isn't viable, regardless of FFmpeg.

Also if security really is the concern, then using the supported plain C code of FFmpeg seems safer than using unmaintained SIMD optimisations.

>Also theres the question about how few people would be affected and what
>we gain from this?

Most likely zero people are affected. Unfortunately, I guess that the only way to know for sure is to remove the support and see if anyone complains about it then.

Another option is to post on the website news that Alpha will be removed if nobody steps up.  I don't think that's a very good use of the website's attention though.

>> It is worth mentioning that even Debian hasn't supported Alpha (along
>> with several other architectures) since release 8.0 in June 2018.
>
>I think debian dropped alpha from the officially supported architectures
>but it seems there is still some inofficial support:
>https://cdimage.debian.org/cdimage/ports/12.0/alpha/

To be precise, Alpha and HPPA were moved to Debian Ports in 2018 after they were dropped from the official archives. They have remained with only unstable support since then, which indicates that they are essentially moribund. Re-emerging from ports is infeasible for a purely legacy architecture.