Created attachment 39234 [details] [review]
Patch adding Vorbis to Matroska muxer.

This patch involves rewriting the tracks info after gst_matroska_mux_prepare_data. 
It seems like a bad hack to me, but the muxer did not work unless I have written
whole header before gst_matroska_mux_prepare_data in gst_matroska_mux_write_data.

Sidenote: the first gst_matroska_mux_prepare_data in _loop function always fails
in my setup, so I had to comment out the error fallback.

I will try to look deeper into these problems.

The failing prepare_data() implies a bug elsewhere. It should work (in theory).
As for the vorbis header data, I think you need to get the first three buffers
fromt he stream for that. It is not in the caps, as far as I know.

What is your testcase?

Created attachment 39372 [details]
Matroska muxer test

Here is my simple test code. I am using gstreamer 0.8.9 and gstreamer-plugins
0.8.8 from Debian Sarge (testing).

The problem with failing prepare_data() occurs here but also in any other setup
I have tried so far. 

Regarding the ogg vorbis setup headers: rawvorbis encoder sets them to
"streamheader" caps metadata when its _chain function is first called. (see
gst-plugins/ext/vorbis/vorbisenc.c)

It has always worked for me, so I'd assume you're doing something weird. Muxers
assume you link pads, and only then start iterating. Any other setup is not
supported.

I find your patch weird. First of all, the changes in data handling aren't
obvious, some even look wrong. The _mux_header_tracks() part, for example, is
very unobvious for me; The reason that we need data before writing the tracks is
*so that* we will have the data type of all contained streams. Other muxers work
this way, too. I'm willing to apply the vorbis-part of it and make that work,
but not the rest.

Your testcase has x-csrc as mimetype, but does not open in any of my editors.
Can you please re-attach it?

Created attachment 48141 [details]
Patch test case.

I reattach the testcase with mime type text/plain.

Created attachment 48152 [details] [review]
New version of matroska Vorbis support patch

I have investigated this problem and I have found that the reason why I was not
able to run gst_matroska_mux_prepare_data before writing header (what caused
that I had not correct vorbis headers) was a bug in
gst_matroska_mux_prepare_data (filled as bug #308633), this new patch need not
use the _mux_header_tracks() part.

Created attachment 48153 [details] [review]
Vorbis support to matroska

I am sorry, the last patch was in a bad format. I am resubmitting.

Thanks, looks perfect.