[vlc-devel] [PATCH] demux: mkv: fix seek

[Steve Lhomme]

Hi


On 2018-06-26 11:48, Zhao Zhili wrote:
> This fixes a regression from 6b10c2e6. Seek failed due to empty
> _tracks_seekpoints[video_track].
> ---
>   modules/demux/mkv/matroska_segment.cpp | 9 +++++++++
>   1 file changed, 9 insertions(+)
>
> diff --git a/modules/demux/mkv/matroska_segment.cpp b/modules/demux/mkv/matroska_segment.cpp
> index 148e66e..844558c 100644
> --- a/modules/demux/mkv/matroska_segment.cpp
> +++ b/modules/demux/mkv/matroska_segment.cpp
> @@ -635,6 +635,15 @@ bool matroska_segment_c::Preload( )
>   
>   
>               cluster = kc_ptr;
> +
> +            // add first cluster as trusted seekpoint for all tracks

Not exactly trusted, it just kickstarts the seeking with a known position.

> +            for( tracks_map_t::const_iterator it = tracks.begin();
> +                 it != tracks.end(); ++it )
> +            {
> +                _seeker.add_seekpoint( it->first,
> +                    SegmentSeeker::Seekpoint( cluster->GetElementPosition(), 0 ) );

It marks the Seekpoint as TRUSTED, ie sure to be a seek point for that 
track because we encountered a keyframe. Thie may not be the case at all.
SegmentSeeker::Seekpoint::TrustLevel::QUESTIONABLE would be better here.

Also a pts of 0 may not be correct. A Segment can start with a high PTS 
if it's the continuation of a previous Segment. So using -1 here is 
better (technically it should be VLC_TS_INVALID but that's not what the 
code expects).

After these 2 changes the fix still works but it's cleaner.

> +            }
> +
>               /* stop pre-parsing the stream */
>               break;
>           }
> -- 
> 2.9.5
>
> _______________________________________________
> vlc-devel mailing list
> To unsubscribe or modify your subscription options:
> https://mailman.videolan.org/listinfo/vlc-devel