Fork from Bug 969290 comment 2. Jim points out the data size of a cluster is not correct. Should fix in EbmlComposer::FinishCluster(). (Now we only compute the first block's size) reference: http://matroska.org/technical/specs/index.html search "Data size"

Comment on attachment 8410899 [details] element lengths in test2.webm Hi Jim: After apply my patch, I use the script you provide to generate the result. Would you please give some feedback?

As far as I can tell, the element lengths are fixed, at least for this file. I took a closer look at the file with another script (output attached) to see if I could spot any other obvious problems, and I think I have. The duration given in the segment information is zero, as are all the timecodes of the blocks in the audio stream.

dependent on https://raw.githubusercontent.com/ccd0/containerchan/master/matroska.php https://raw.githubusercontent.com/ccd0/containerchan/master/matroska-elements.txt

(In reply to Jim Black from comment #5) > Created attachment 8410957 [details] > more detailed look at test2.webm > > As far as I can tell, the element lengths are fixed, at least for this file. > > I took a closer look at the file with another script (output attached) to > see if I could spot any other obvious problems, and I think I have. The > duration given in the segment information is zero, as are all the timecodes > of the blocks in the audio stream. Yeah, we don't output the timestamp for vorbis audio in VorbisTrackEncoder.c. It's not difficult to get the timestamp and duration for decoder by counting the audio samples

r=rillian try server: https://tbpl.mozilla.org/?tree=Try&rev=0e92bd257623