Stream layer split proposition.

Hello,

Here comes my recent idea: to split current stream layer to two layers:
physical stream and logical stream. Here is what I mean:

In most of cases, input streams are very simple: file over something
(over filesystem, over ftp, over some other network protocol, etc).

But even DVD streams are much worse: DVD stream can exist over raw-device
(read with bunch of ioct's for decoding), over any fd (simple reading & seeking)
and over bunch-of-VOBs-and-IFOs. All these are _physical_ streams.
But, from player's POV, DVD can be seen as (at least)
two different _logical_ streams: simple per-title reading (as in current
MPlayer & G2) and as those fancy navigation menus, events, etc.

Another example: VCD, CD-DA and other non-plain CD's can exist as normal
CD/CD-R/etc or as .cue+.bin, as .cue+bunch-of-tracks (maybe it's the
same. I'm not sure). Or as some metadata(track offsets, sector sizes,
etc) and raw byte stream (fd).

So here (again) is the proposition: to split stream code to two
independant layers with caching between them. (I mean, only physical
stream must be cached. Logical stream should be directly feeded to the
demuxers.)

I can't propose any API (yet), but I think, that they can be as simple
as static metadata (i.e. file name, size, etc. for file-based phys.streams,
cuesheet for cd-based, etc).

I can see only two problems with this approach:
1) device-based DVD phys stream must be intellectual to determine
   if a block is encrypted or not without the help from logical stream
   layer (it should duplicate/incorporate some part of current libdvdread)

2) again DVD, but now bunch-of-VOBS problem. to provide VOBs as
   unificated DVD stream, we should incorporate some type of 
   'mkisofs -dvd-video ...' into phys stream.

There is another solution for this -- use stream-of-VOBs-and-IFOs as
a basic DVD stream and convert all other DVD streams to this, but I
don't know if it's better or worse.

-- 
With best wishes
Dmitry Baryshkov