[Libav-user] HLS and accurate seek

[Gabriele Greco]

>From what I read in the documentation input seeking in modern version of
ffmpeg should be almost as accurate as output seeking, but this is not true
if the input is an HLS file, I tried both with 4.2.x release and master,
and this is the case, for instance in a file like the following one with 1
keyframe every 6 seconds:

ffmpeg-4.2 -y -ss 861 -i
http://wycutters.wyscout.com/box/recchiavari201/hls-30s/hls-30s.m3u8 -f
image2 -q:v 1 -vframes 1 ~/Desktop/+1.png
ffmpeg-4.2 -y -ss 859 -i
http://wycutters.wyscout.com/box/recchiavari201/hls-30s/hls-30s.m3u8 -f
image2 -q:v 1 -vframes 1 ~/Desktop/0.png
ffmpeg-4.2 -y -ss 859 -i
http://wycutters.wyscout.com/box/recchiavari201/hls-30s/hls-30s.m3u8 -f
image2 -q:v 1 -vframes 1 ~/Desktop/-1.png

... all gives the same image.

This has been noticed before also in this list:

http://www.ffmpeg-archive.org/HLS-accurate-seeking-vs-MP4-td4685570.html

A ticket has been written:

https://trac.ffmpeg.org/ticket/7485

.... and more than a year and half ago a patch has been made that fixes
this issue:

https://patchwork.ffmpeg.org/project/ffmpeg/patch/20181024125015.65290-2-nick.paul.ryan@gmail.com/

I've tried to apply this patch to 4.2.x branch and it works as expected
(the three images have the clock hud with different seconds).

My question is:

Why this patch has not been applied to the official version?

It is somewhat broken i a way I can not see?

-- 
Bye,
 Gabry
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://ffmpeg.org/pipermail/libav-user/attachments/20200404/a866f549/attachment.html>