[FFmpeg-devel] [RFC PATCH 3/3] HACK: avformat: rawenc: allow to output a raw PRFT

[Andreas Rheinhardt]

Clément Péron:
> Hi Andreas,
> 
> On Fri, 22 Sept 2023 at 09:58, Andreas Rheinhardt
> <andreas.rheinhardt at outlook.com> wrote:
>>
>> Clément Péron:
>>> Hi Michael, Andreas,
>>>
>>> On Thu, 21 Sept 2023 at 22:50, Andreas Rheinhardt
>>> <andreas.rheinhardt at outlook.com> wrote:
>>>>
>>>> Michael Niedermayer:
>>>>> On Thu, Sep 21, 2023 at 02:17:00PM +0200, Clément Péron wrote:
>>>>>> Output the producer reference time to a dirty raw output.
>>>>>>
>>>>>> Signed-off-by: Clément Péron <peron.clem at gmail.com>
>>>>>> ---
>>>>>>  libavformat/rawenc.c | 122 +++++++++++++++++++++++++++++++++++++++++++
>>>>>>  1 file changed, 122 insertions(+)
>>>>>
>>>>> this breaks fate-filter-volume and others
>>>>> (Segmentation fault)
>>>>>
>>>>> I can rerun it with debug symbols and provide peoper gdb output
>>>>> but i suspect given this has "HACK" in the title you are aware of this
>>>
>>> The "HACK" tag meaning was not supposed to be: "it's ok if it
>>> segfaults", but more to trigger a discussion is it possible to
>>> properly support an output timestamp in the raw video demux, and if
>>> yes how to do it :)
>>
>> If you need a timestamp for raw video, then use a proper container and
>> not raw video. In fact, this patch basically creates new formats
>> different from all the raw formats.
> 
> Yes I agree, but I do not want to add too much overhead nor
> computation processing or memory copy to my pipeline just to mux and
> demux between ffmpeg and my python script.
> 
> The idea is to have a very light structure to easily pipe it.
> 

Our libraries are meant to be used by API users and are designed for
that. The ffmpeg command line tool is just one user among many and
adding code to a library to circumvent a limitation of ffmpeg (or
another user of the libraries) is not appropriate. We would end up with
a ton of hacks.

> I'm not familiar with audio/video container but it seems to me that
> parsing containers are not very light no?

For certain formats the overhead of parsing containers can be negative
when compared to the raw format, because the containers provide a length
field for the packet whereas one has to search for the packet boundaries
in case of a truely raw stream. But this is not true for raw video which
is fixed size.

- Andreas