[FFmpeg-devel] [PATCH 1/4] swr: convert resample_common/linear_int16_mmx2/sse2 to yasm.
andreas.cadhalpun at googlemail.com
Sat Jul 12 22:20:30 CEST 2014
On 12.07.2014 22:01, Hendrik Leppkes wrote:
> On Sat, Jul 12, 2014 at 9:56 PM, Andreas Cadhalpun
> <andreas.cadhalpun at googlemail.com> wrote:
>> On 12.07.2014 21:40, Hendrik Leppkes wrote:
>>> On Sat, Jul 12, 2014 at 9:26 PM, Andreas Cadhalpun
>>> <andreas.cadhalpun at googlemail.com> wrote:
>>>> I'm not sure what you mean with 'officially' here. Every symbol exported
>>>> a shared library can be used by another program.
>>>> Even if only other av* libraries use the avpriv_* symbols they can't just
>>>> removed, because that would break ABI compatibility between the old
>>>> of the library using such a symbol with the new version of the other
>>>> not exporting this symbol anymore.
>>> Unrelated to any ABI or API problems, a use-case of mixing different
>>> versions of the libraries strongly advised against ever using or a
>>> distribution enabling such use, because its hardly ever tested, and
>>> results may be rather unpredictable, especially with a lot of
>>> applications sadly still using some of the API "incorrectly".
>> There shouldn't be any problem, as long as libraries with the same major
>> SOVERSION are ABI backward-compatible.
>> If there would be a problem, I would call it a (very strange) bug.
>> Have you ever heard of such a problem?
>> If you don't want different versions of the libraries used together, you
>> should either bundle all these libraries into a big one or bump all major
>> SOVERSIONs in every release. Neither of those make much sense, I think.
> There have been a lot of those bugs, which get fixed when found, but
> its something that easily breaks and is hard to find, so its just
> something you should avoid instead of encourage.
I don't intend to encourage such use, but I can't really prevent it
either. So I'm trying to make sure it doesn't break in obvious ways.
More information about the ffmpeg-devel