The driver design and code could certainly do with some more work.  Here are
some ideas:

1.  Convert vpb.rtf (originally input file to Windows help compiler) to HTML.
2.  Most files are DOS text files - convert to unix format text files.
3.  David Sugar has written a nice new COFF loader (loader.cpp) that replaces
    coff.cpp but I havent had a chance to try it out yet.
4.  When 'verbose' mode is on, the driver spews all sorts of messages to the
    console.  It would be nice to redirect these to syslog, perhaps as an 
    option.
5.  Some more design documention, such as a class diagram that relates the
    various modules would be useful.
6.  Many of the classes jump straight to another class without actually doig
    anything!  This was done to provide better hiding of internal class data 
    in header files but makes the code hard to follow.  This doubling up of 
    classes should be removed.
7.  The original driver eused C++ exception handling.  This cased problems
    across thread contexts and has been removed.  Error handling is now
    done by bombing out with asserts!  This is a bit primitive, and may need
    enhancement.
8.  The comp.cpp module is redundant and should be removed, as bit packing
    is now performed in the DSP firmware.
9.  All references to "relay" FIFOs relate to the NT driver and can be removed.
10. The code is a mixture of C++ & C, but only has a C library interface.  It
    would be fairly easy to make the whole thing pure C, which may extend it's
    appeal.
11. I am sure playrec.cpp is overcomplicated, and could be redesigned to be
    much simpler.

David Rowe <voicet@bigpond.com.au>