[FFmpeg-devel] [PATCH v2] avformat/hlsenc: fix hls_time not respected

[Ingo Oppermann]

> On 28 Jan 2025, at 15:10, Michael Niedermayer <michael at niedermayer.cc> wrote:
> 
> On Mon, Jan 27, 2025 at 10:09:30AM +0100, Ingo Oppermann wrote:
>> This fixes the criterion when to split the segments based on the
>> elapsed time for the current segment instead of using the theoretical
>> elapsed time since start based on hls_time and the number of written
>> segments.
>> 
>> hls_time is used to define the minimum length of a segment, however
>> this is not respected in all cases when a stream has variable GOP
>> sizes.
>> 
>> Imagine a stream starts with a key frame every 10 seconds for
>> e.g. 30 seconds. After that, key frames will come every second. This
>> will result in segments that are first 10 seconds (as expected), then
>> 1 second for some time (not as expected) and later 2 seconds (as
>> expected).
>> 
>> How to reproduce:
>> ffmpeg -t 30 -f lavfi -i testsrc2 -codec:v libx264 -g 250 part1.mp4
>> ffmpeg -t 30 -f lavfi -i testsrc2 -codec:v libx264 -g 25 part2.mp4
>> echo "file part1.mp4\nfile part2.mp4" > list.txt
>> ffmpeg -f concat -i list.txt -codec copy \
>>    -f hls -hls_time 2 -hls_list_size 0 parts.m3u8
>> cat parts.m3u8
>> 
>> Signed-off-by: Ingo Oppermann <ingo at datarhei.com>
>> ---
>> libavformat/hlsenc.c | 12 +++---------
>> 1 file changed, 3 insertions(+), 9 deletions(-)
> 
> this breaks / needs to update
> fate-hls-init-time
> 
> --- ./tests/ref/fate/hls-init-time 2025-01-28 13:58:16.733614814 +0100
> +++ tests/data/fate/hls-init-time 2025-01-28 15:09:03.869138579 +0100
> @@ -3,310 +3,308 @@
> #codec_id 0: pcm_s16le
> #sample_rate 0: 44100
> #channel_layout_name 0: mono
> -0,          0,          0,     1152,     2304, 0x28123557
> -0,       1152,       1152,     1152,     2304, 0x838c7e81
> -0,       2304,       2304,     1152,     2304, 0x4fb8704c
> -0,       3456,       3456,     1152,     2304, 0x5f787f9e
> ...
> 

As suggested by Gyan, I'll rework the patch in order to not change the default behaviour. Currently it overrules hls_init_time as well.

> [...]
> 
> -- 
> Michael     GnuPG fingerprint: 9FF2128B147EF6730BADF133611EC787040B0FAB
> 
> The misfortune of the wise is better than the prosperity of the fool.
> -- Epicurus
> _______________________________________________
> ffmpeg-devel mailing list
> ffmpeg-devel at ffmpeg.org
> https://ffmpeg.org/mailman/listinfo/ffmpeg-devel
> 
> To unsubscribe, visit link above, or email
> ffmpeg-devel-request at ffmpeg.org with subject "unsubscribe".