[FFmpeg-devel] [PATCH 1/7 v4] avcodec: add an Immersive Audio Model and Formats frame split bsf

[Andreas Rheinhardt]

James Almer:
> On 2/12/2024 2:28 PM, Andreas Rheinhardt wrote:
>> James Almer:
>>>
>>>
>>> On 2/12/2024 2:04 PM, Andreas Rheinhardt wrote:
>>>> James Almer:
>>>>> On 2/12/2024 1:40 PM, Andreas Rheinhardt wrote:
>>>>>> James Almer:
>>>>>>> On 2/6/2024 10:05 AM, James Almer wrote:
>>>>>>>> Signed-off-by: James Almer <jamrial at gmail.com>
>>>>>>>> ---
>>>>>>>> Now reading decriptors from extradata, plus a setting to ensure any
>>>>>>>> descriptors
>>>>>>>> present inband are omitted has been added.
>>>>>>>>
>>>>>>>>      doc/bitstream_filters.texi            |  16 +
>>>>>>>>      libavcodec/bitstream_filters.c        |   1 +
>>>>>>>>      libavcodec/bsf/Makefile               |   1 +
>>>>>>>>      libavcodec/bsf/iamf_frame_split_bsf.c | 887
>>>>>>>> ++++++++++++++++++++++++++
>>>>>>>>      4 files changed, 905 insertions(+)
>>>>>>>>      create mode 100644 libavcodec/bsf/iamf_frame_split_bsf.c
>>>>>>>
>>>>>>> Will apply the set soon if there are no objections.
>>>>>>
>>>>>> I still object to the BSF in #1 existing as it just duplicates
>>>>>> parsing
>>>>>> code into lavc and lavf. And the issue with creating new framings for
>>>>>> stuff for which no framing except raw data can't exist is still
>>>>>> there.
>>>>>
>>>>> I insist on using the split bsf, but i can try to remove the merge one
>>>>> and do that within lavf, to avoid creating packets with OBU framing.
>>>>
>>>> Why is splitting not simply done inside lavf (and inside the demuxer,
>>>> not the generic code in demux.c)? What is the advantage of that?
>>>
>>> Not making a mess in mov.c's read_packet() from reiterated calls because
>>> one Track Sample has packets for several AVStreams.
>>
>> Mess? All you would need to do is add a check at the beginning of
>> read_packet whether there are any more packets buffered. If so, return
>> them, if not, read new data.
> 
> Buffered where? Internally in the demuxer?
> 

That's it: We add a new function (may be an inline wrapper for
avpriv_packet_list_put()) for demuxers to append multiple packets to the
raw_packet_buffer queue in a single read_packet call. That way the
ordinary (already existing) logic in ff_read_frame() will take care of
everything.

- Andreas