[FFmpeg-devel] [PATCH 1/2] avformat/movenc: fix sample size being zero in pcmC

[Raphaël Zumer]

On 7/26/23 11:22, Zhao Zhili wrote:
>> From: ffmpeg-devel <ffmpeg-devel-bounces at ffmpeg.org> On Behalf Of Raphaël Zumer
>> Sent: 2023年7月26日 21:38
>> To: ffmpeg-devel at ffmpeg.org
>> Subject: Re: [FFmpeg-devel] [PATCH 1/2] avformat/movenc: fix sample size being zero in pcmC
>>
>> On 7/25/23 23:45, Zhao Zhili wrote:
>>> From: Zhao Zhili <zhilizhao at tencent.com>
>>>
>>> -    avio_w8(pb, track->par->bits_per_raw_sample);
>>> +    sample_size = track->par->bits_per_raw_sample;
>>> +    if (!sample_size)
>>> +        sample_size = av_get_exact_bits_per_sample(track->par->codec_id);
>>> +    av_assert0(sample_size);
>>> +    avio_w8(pb, sample_size);
>> Why not just replace bits_per_raw_sample completely with av_get_exact_bits_per_sample()?
>>
>> Are there edge cases, like incorrect codec ID, or sample sizes smaller than specified by the codec ID (and if there is ever a mismatch,
>> which is expected in pcmC?)
>>
>> If not, there should be no need to prioritize bits_per_raw_sample in the first place and add a fallback - going straight to
>> av_get_exact_bits_per_sample() should suffice and keep the code simpler.
> Check bits_per_raw_sample first for a little bit of performance.
> Don't use bits_per_coded_sample because it make less sense in this case and to keep the code simpler.
I doubt it is worth the trouble, but my opinion is not important if no one else cares.
>> As noted in another thread, bits_per_raw_sample seems to always be 0 when muxing PCM to MP4 with FFmpeg, whereas
>> bits_per_coded_sample has the correct value from the samples that I tested.
> That's only one usecase. bits_per_coded_sample has been set while bit_per_raw_sample wasn't only
> because the implementation of wav demux and there's no encoding process. (You can try -c:a pcm_s16le)
>
> There are more usecases to support, e.g., only use mp4 muxer without any demuxer or encoder.
> User can create and set a AVCodecParameters manually, and let bits_per_coded_sample be zero.

That is fair, I just wanted to point out the mismatch as a potential (separate) issue.

RZ