[FFmpeg-devel] [PATCH] avformat/mov: discard data streams with all zero sample_delta

[Gyan Doshi]

On 2022-07-06 08:23 am, "zhilizhao(赵志立)" wrote:
>
>> On Jul 5, 2022, at 10:33 PM, Gyan Doshi <ffmpeg at gyani.pro> wrote:
>>
>>
>>
>> On 2022-07-05 07:05 pm, "zhilizhao(赵志立)" wrote:
>>>> On Jul 5, 2022, at 8:07 PM, Gyan Doshi <ffmpeg at gyani.pro> wrote:
>>>>
>>>>
>>>>
>>>> On 2022-07-05 01:20 pm, Zhao Zhili wrote:
>>>>> From: Zhao Zhili <zhilizhao at tencent.com>
>>>>>
>>>>> Streams with all zero sample_delta in 'stts' have all zero dts.
>>>>> They have higher chance be chose by mov_find_next_sample(), which
>>>>> leads to seek again and again.
>>>>>
>>>>> For example, GoPro created a 'GoPro SOS' stream:
>>>>>    Stream #0:4[0x5](eng): Data: none (fdsc / 0x63736466), 13 kb/s (default)
>>>>>      Metadata:
>>>>>        creation_time   : 2022-06-21T08:49:19.000000Z
>>>>>        handler_name    : GoPro SOS
>>>>>
>>>>> With 'ffprobe -show_frames http://example.com/gopro.mp4', ffprobe
>>>>> blocks until all samples in 'GoPro SOS' stream are consumed first.
>>>>>
>>>>> Signed-off-by: Zhao Zhili <zhilizhao at tencent.com>
>>>>> ---
>>>>>   libavformat/mov.c | 14 ++++++++++++++
>>>>>   1 file changed, 14 insertions(+)
>>>>>
>>>>> diff --git a/libavformat/mov.c b/libavformat/mov.c
>>>>> index 88669faa70..2a4eb79f27 100644
>>>>> --- a/libavformat/mov.c
>>>>> +++ b/libavformat/mov.c
>>>>> @@ -3062,6 +3062,20 @@ static int mov_read_stts(MOVContext *c, AVIOContext *pb, MOVAtom atom)
>>>>>       st->nb_frames= total_sample_count;
>>>>>       if (duration)
>>>>>           st->duration= FFMIN(st->duration, duration);
>>>>> +
>>>>> +    // All samples have zero duration. They have higher chance be chose by
>>>>> +    // mov_find_next_sample, which leads to seek again and again.
>>>>> +    //
>>>>> +    // It's AVERROR_INVALIDDATA actually, but such files exist in the wild.
>>>>> +    // So only mark data stream as discarded for safety.
>>>>> +    if (!duration && sc->stts_count &&
>>>>> +            st->codecpar->codec_type == AVMEDIA_TYPE_DATA) {
>>>>> +        av_log(c->fc, AV_LOG_WARNING,
>>>>> +               "All samples in data stream index:id [%d:%d] have zero duration, "
>>>>> +               "discard the stream\n",
>>>>> +               st->index, st->id);
>>>>> +        st->discard = AVDISCARD_ALL;
>>>>> +    }
>>>>>       sc->track_end = duration;
>>>>>       return 0;
>>>>>   }
>>>> So this will allow audio and video streams to be demuxed, but not data?  That distinction seems arbitrary.
>>> Disable audio/video streams may create regression. It’s unlikely for random
>>> and broken data stream.
>>>
>>>> Print a warning and assign a duration to each sample. Either 1 or if not zero/Inf, st->duration/st->nb_frames.
>>> Set sample_duration to 1 doesn’t work. Dts still far behind other streams.
>>>
>>> Set sample_duration st->duration/st->nb_frames works for me, but I prefer
>>> current strategy for the following reasons:
>>>
>>> 1. AVDISCARD_ALL is more close to AVERROR_INVALIDDATA by giving up instead
>>> of trying correction and hope it works, which may not, e.g., st->duration
>>> is broken, or bad interleave even though we fixed sample_duration.
>> It's not about hoping that it works.  It's about not preventing the user from acquiring the stream payload.
>>
>> Can you test if setting -discard:d none -i INPUT allows reading the stream with your patch?
> Yes it does allow reading the stream. ’stts’ box is parsed during
> avformat_find_stream_info(), AVStream->discard flag can be modified
> after that. The patch has no effect if user changed AVStream->discard
> flag.

What's the duration of the demuxed stream?

Regards,
Gyan