Re: [FFmpeg-devel] [PATCH 1/7] avformat/apvenc: Only allow APV

[Andreas Rheinhardt <andreas.rheinhardt@outlook.com>]

Mark Thompson:
> On 27/04/2025 19:42, Andreas Rheinhardt wrote:
>> Patches attached.
>>
>> - Andreas
> 
>> [PATCH 1/7] avformat/apvenc: Only allow APV
> 
> Oops, sure.
> 
>> [PATCH 2/7] avformat/apvenc: Remove unused header
> 
> Sure.
> 
>> [PATCH 3/7] avformat/apvdec: Use ffio_read_size()
> 
> Is there a cleaner way to discard very short files without reading them?  The minimum size of a valid is significantly larger than this, and an early failure would be nice rather than it get packetised and something later has to realise it's bad.

There is avio_size(). But I know of no demuxer that uses it to discard
small files.
(What is the minimum size of a valid file?)

> 
>> [PATCH 4/7] avformat/apvdec: Remove inappropriate flags
> 
> What does the GENERIC_INDEX flag affect here?
> 

With the flag set, the demux code adds index entries for keyframes based
on AVPacket.pos. For this demuxer, AVPacket.pos points to the position
of the signature, not of the access unit size field which is where the
read_packet() code expects to be when called. So with the current code,
seeking to an already encountered place will lead to an APV_SIGNATURE
being mistaken for an access unit size. You can see this for yourself with
ffmpeg -stream_loop 50 -i $(FATE_SUITE)/apv/profile_422-10.apv -c copy
-f null -
An alternative would be to decrement AVPacket.pos by four.

> Currently mpv does successfully seek in a raw file (to already-seen places), but this may or may not be related.

IIRC mpv caches a certain amount of data (in the form of AVPackets IIRC)
in order to enable short seeks backwards and forwards from the current
position. I don't think it calls the avformat seek functions for your seeks.

- Andreas

_______________________________________________
ffmpeg-devel mailing list
ffmpeg-devel@ffmpeg.org
https://ffmpeg.org/mailman/listinfo/ffmpeg-devel

To unsubscribe, visit link above, or email
ffmpeg-devel-request@ffmpeg.org with subject "unsubscribe".