Sat Nov 7 11:09:10 CET 2009

tokenized tree parser -> postfix stream

Got the numbers working.

Now, instead of using a prefix format.  What about using a postfix
one, where bytes are pushed and constructors executed afterwards?  An
extra bit could encode whether the byte string is quoted, or whether
it is an operation.

This way the parser doesn't need procedure recursion, but can use an
explicit (more efficient?) reduction stack.

However, to make the format partially parsable, arity still needs to
be specified.

So, it looks like the tradeof is to use bottom up or top down formats.

The conclusion seems to be that really, parsing isn't the problem.
The hairy one is tokenizing, and representing a tokenized stream so it
can be trivially reconstructed.

Another thing: tokenizing produces a stream of things.  What things?
There needs to be a distinction between code/constructors and

Ascii rep constraints:
  - no '"' and '\' characters so it fits in C strings
  - 3 modes: prefix numbers: last digit contains quotation bit