[FFmpeg-trac] #8578(avformat:new): FFmpeg 4.2 breaks Matroska streaming

[FFmpeg]

#8578: FFmpeg 4.2 breaks Matroska streaming
-------------------------------------+-------------------------------------
             Reporter:  Sesse        |                    Owner:
                 Type:  defect       |                   Status:  new
             Priority:  normal       |                Component:  avformat
              Version:  git-master   |               Resolution:
             Keywords:  mkv          |               Blocked By:
  regression                         |
             Blocking:               |  Reproduced by developer:  0
Analyzed by developer:  0            |
-------------------------------------+-------------------------------------

Comment (by Sesse):

 Martin Storsjö helped me debug this. The gist of it is: Earlier, there
 would be only one write callback per chunk. Now, there is two; one for the
 EBML ID (4 bytes) and one for the actual data (50–60 kB). Since the
 Matroska muxer doesn't use data markers (type in the callback, after the
 header has been written, is always AVIO_DATA_MARKER_UNKNOWN), my code
 assumes that the point right after the EBML ID is a valid starting point
 for the stream, whereas it isn't.

 Worse yet, since there's a sizeable delay in wall clock time after the
 EBML ID is written, it's practically guaranteed that the client will
 indeed hit that point. If I add a hack to my code saying that

 {{{
         if (buf_size == 4) type = AVIO_DATA_MARKER_SYNC_POINT;
 }}}

 then things instantly start working again.

--
Ticket URL: <https://trac.ffmpeg.org/ticket/8578#comment:3>
FFmpeg <https://ffmpeg.org>
FFmpeg issue tracker