Luca,
Thanks for preparing a new draft.
After various discussions, particularly the surrounding the oggless[1]
proposal from the ffmpeg/nut/mplayer people, I've reversed my opinion
on the value of a general packed headers format, instead of the current
codec-specific versions, so I'd like to ask that we redefine that.
Requirements:
* general mechanism for framing bitstream headers in a single blob
(this means we have to include the metadata header)
* support arbitrary number of headers
* support for arbitrary length (maybe)
The simplest extension would be to ammend the packed header definition
so:
[24 bit] id
[8 bit] count
for each header in count:
[16 bit] length
[...] length bytes of header
That would probablly work for most codecs (but we'll need to include
text about truncating or rejecting the stream if any of the packets are
too big)
More flexible would be to define a variable-length encoding scheme, for
example, uft-style high-bit-marks-continuation, or jpeg2k-style where
the length includes the size of the length marker, which reserves
lengths of 2 to mean 'read 4 more bytes as the length' in which a value
of 3 means 'read 8 more bytes as the length'.
I prefer the utf-style because it can be used for the count field as
well, and saves a few bytes in the current common cases.
Comments?
-r
[1] http://svn.mplayerhq.hu/nut.merged/docs/oggless-xiph-codecs.txt?view=markup