[Icecast-dev] Icecast WebM Support Patch Second Edition

[Krad Radio]

Howdy,

tl;dr:  Nothing new or interesting to non-developers

Attached is a newer Icecast WebM support patch for Icecast SVN, there is
simply aesthetic changes. Whitespace has been altered to match Icecast
project style,
some functions have been renamed and moved around.

Some discussion.

The format_ebml.c file lines 0-296 operates much the same as the
format_ogg.c file does. It uses a parsing library api to feed in and pull
out bytes via
the _get_buffer function. In the case of Ogg this is of course libogg, in
the case of ebml, the "parsing library" consists of the functions below
line 296.
This mini-included library comes from some debugging code that I wrote
during my work on a much more robust EBML muxing library, originally I had
thought
icecast would need to do some manipulation of the EBML segment header in
all cases, but it turns out in most cases this is better done by the source
client.
This mini-included library implements the same function calls that Icecast
would need to use if it was using this EBML muxing library, but "parse" is
a very
generous descriptor of whats actually happening. If you are familiar with
the internals of Ogg streaming, you know that the stream starts out with a
few header pages
that are then followed by pages containing the acutual video and audio
packets. EBML doctype WebM and MKV (The only currently known) work in
a similar way.
There is a "Segment Header" followed by "Clusters" (Its actually more
nuanced than this, but this oversimplification will suffice for the
moment). Clusters are comparable to
Ogg pages. The connecting client needs to be sent the header, and then it
can start on any cluster as long as it starts exactly on a cluster. (This
is actually not a format requirement per se but all media players I am
aware of lack the ability to re-sync themselves if started on a random
byte, something I intend not to be a limitation of my own work). So, in a
live WebM stream, everything before the first cluster is the "Header" and
then the rest is the clusters, of which the boundary between them is marked
with 4 specific bytes. So at every cluster the refbuf is marked as a sync
point. The size of clusters could vary significantly during a stream or
from stream to stream, but on a properly constructed one would indicate a
keyframe in the case of a stream with video. The 'mini-library' doesn't
actually 'parse' the stream at all, all it does is look for the four byte
sequence indicating a cluster boundary and informs the format functions
appropriately. I am a fraudulent mathematician at best, but I calculate
that there is a 1 in 4.2 billion chance of this happening for any given 4
bytes, and 1 in 4294 per megabyte, and likely once per 4.2 gigabytes. This
is not a problem once the stream has started for the client however, it
only matters when the source connects and the header is stored, and when a
client connects and needs a proper starting point. I suppose that makes it
very unlikely to cause a problem, even though its technically wrong. It
also means that each byte is being compared, whilst when properly parsing
most would be skipped, but computers are so damn fast that its moot. At any
rate proper parsing could be added this this mini-library or provided by
the external library as it matures.

Enjoy,

David
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.xiph.org/pipermail/icecast-dev/attachments/20120208/ac2eddbb/attachment-0001.htm 
-------------- next part --------------
A non-text attachment was scrubbed...
Name: icecast-webm-support-clean1.patch
Type: text/x-patch
Size: 20519 bytes
Desc: not available
Url : http://lists.xiph.org/pipermail/icecast-dev/attachments/20120208/ac2eddbb/attachment-0001.bin