$Id$
Major to-do items.

(1) DONE - In bitmatch operator, use patterns not expressions.

(2) DONE - Allow matching against strings.

(3) DONE - Change the syntax so { ... } surrounds match patterns.

(4) Provide UInt32 and UInt64 types.

(5) DONE - Allow for specific offsets and alignment.  Something like this:

    { start : 16;
      another : 16 : offset(256);   (* location would be 256 bits from start *)
    }

(6) and:

    { start : 16;
      another : 16 : align(32);     (* implicit 16 bit gap before this *)
    }

(7) Assertions:

    { start : 16 : assert (offset = 0); }

   (Q: Are these evaluated at compile time or at run time or selectable?)

(8) Named but unbound patterns to avoid "Warning Y: unused variable".

(9) DONE -
    Make the error locations fine-grained, particularly so they point to
    individual fields, not the whole match.

(10) Cross-module, persistent, named patterns, see:
  http://caml.inria.fr/pub/ml-archives/caml-list/2008/04/25992c9c9fa999fe1d35d961dd9917a2.en.html

(11) DONE -
     Runtime endiannness expressions.  The suggested syntax is:

    { field : len : endian (expr) }

    where expr would evaluate to something like BigEndian or
    LittleEndian.

    There are several protocols around where endianness is only
    determined at runtime, examples are libpcap and TIFF.

(12) DONE - More constant field lengths.

(13) Implement native endian functions.

(14) PARTLY DONE - A proper test suite.

(15) DONE - More examples:

    ELF binaries
    GIF images

(16) We now know the offset of the current field relative to the
     whole match.  This may allow more efficient aligned versions
     of functions to be called (at compile time).  However do note
     that the offset in the bitstring is usually not known.