[FFmpeg-devel] [PATCH] avformat: add vapoursynth wrapper

[James Almer]

On 5/4/2018 9:19 PM, Michael Niedermayer wrote:
> On Fri, May 04, 2018 at 02:02:02PM -0300, James Almer wrote:
>> On 5/4/2018 1:51 PM, wm4 wrote:
>>> On Fri, 4 May 2018 13:30:38 -0300
>>> James Almer <jamrial at gmail.com> wrote:
>>>
>>>> On 5/4/2018 12:58 PM, wm4 wrote:
>>>>> On Sat, 28 Apr 2018 19:05:29 +0200
>>>>> wm4 <nfxjfg at googlemail.com> wrote:
>>>>>   
>>>>>> This can "demux" .vpy files.
>>>>>>
>>>>>> Some minor code copied from other LGPL parts of FFmpeg.
>>>>>>
>>>>>> I did not found a good way to test a few of the more obscure features,
>>>>>> like VFR nodes, compat pixel formats, or nodes with dynamic size/format
>>>>>> changes. These can be easily implemented on demand.
>>>>>> ---
>>>>>>  configure                 |   5 +
>>>>>>  libavformat/Makefile      |   1 +
>>>>>>  libavformat/allformats.c  |   1 +
>>>>>>  libavformat/vapoursynth.c | 421 ++++++++++++++++++++++++++++++++++++++++++++++
>>>>>>  4 files changed, 428 insertions(+)
>>>>>>  create mode 100644 libavformat/vapoursynth.c
>>>>>>  
>>>>>
>>>>> Pushed, with some minor changes, and zero-copy frame passing.  
>>>>
>>>> The first step to fix something (in this case, usage sizeof(AVFrame)
>>>> outside libavutil) is not adding even more cases of the issue in question.
>>>> You still can replace this with rawvideo. Someone even already wrote it
>>>> for you.
>>>>
>>>> Lets try to abide our own ABI rules...
>>>
>>> That's requires a frame copy and is not what I went through all the
>>> effort for.
>>>
>>> Why didn't you say anything when the kmsgrab code did the same thing?
>>> Or when the unwrapped frame stuff was added in the first place?
>>
>> I did the other day on IRC when you asked me why i was against this, if
>> you recall, because it was then when i found out this has been the case
>> for a long while, and why I'm now saying adding even more cases is going
>> in the opposite direction of an actual solution.
>>
>> In any case, i explicitly didn't block this, and no one else seems to
>> care, so whatever.
> 
> I just now realized this, no, use of sizeof(AVFrame) outside libavutil is not ok,
> thats a ABI/API breakage.
> 
> This must be removed/reverted. In all cases that are relevant.

Fixed/adapted, not reverted. For vapoursynth and kmsgrab it should be a
matter of making them output rawvideo packets.

> (there can be exceptions if they are irrelevant to API/ABI compatibility)
> 
> For example the "if (pkt->size < sizeof(AVFrame))" in libavcodec/wrapped_avframe.c
> should be ok

How is that one ok? Is it because sizeof(AVFrame) can grow but never
shrink without a soname bump so the check is safe?

Still, that's only the wrapped_avframe decoder. What about the encoder?
It's used by the null and yuv4mpeg muxers, plus decklink and
libndi_newtek outdevs.