On Thu, 2 Feb 2017 10:47:52 -0800 Vignesh Venkatasubramanian <vigneshv-at-google....@ffmpeg.org> wrote:
> On Tue, Jan 31, 2017 at 10:18 PM, wm4 <nfx...@googlemail.com> wrote: > > > > On Tue, 31 Jan 2017 12:02:01 -0800 > > Chris Cunningham <chcunning...@chromium.org> wrote: > > > > > Thanks for taking a look. > > > > > > Definitely missing a "break;" - will fix in subsequent patch. > > > > > > Agree timestamps should be relative (didn't realize this). Vignesh points > > > out that "0" in the test file is due to a bug in ffmpeg (and probably > > > other > > > muxers) where this value is not written as a relative timestamp, but > > > instead as the timestamp of the previous frame. https://github.com/FFmp > > > eg/FFmpeg/blob/master/libavformat/matroskaenc.c#L2053 > > > <https://www.google.com/url?q=https%3A%2F%2Fgithub.com%2FFFmpeg%2FFFmpeg%2Fblob%2Fmaster%2Flibavformat%2Fmatroskaenc.c%23L2053&sa=D&sntz=1&usg=AFQjCNGs8m6GsWbhTvCZl0Q_juGAldQblA> > > > > > > > Just a few lines below this reads > > > > mkv->last_track_timestamp[track_number - 1] = ts - mkv->cluster_pts; > > > > which looks like it intends to write a relative value. Though "ts" can > > be a DTS, while the other value is always a PTS. > > Just to clarify: This line makes the timestamp relative to the > cluster's timestamp. Not relative to the block its referencing (which > is what the spec says if i understand it correctly). Yeah, the current spec just says "Timestamp of another frame used as a reference (ie: B or P frame). The timestamp is relative to the block it's attached to." Is this a bug? Did FFmpeg always mux this incorrectly? Is there even an implementation that uses the value written to block reference elements? (And not in a trivial way like FFmpeg.) _______________________________________________ ffmpeg-devel mailing list ffmpeg-devel@ffmpeg.org http://ffmpeg.org/mailman/listinfo/ffmpeg-devel
