Sat Mar 12 14:12:39 EST 2011

Moving forth parsing words to signatures.

I don't see an immediate starting point to do this incrementally.
Maybe that's actually good; maybe the real solution is to throw it all
away and start over?

Man, this is hard..

Let's start at the beginning.

A significant roadblock is the non-composability of define-signature.
Can that be fixed transparently?

I think it really needs a rewrite, starting bottom up, on top of
label^, and taking into account all the problems that are associated
with the flatness.

Starting from scratch also allows preprocessing macros to be
represented differently.  They should not be syntax, because they do
not behave as scheme macros.

The basic idea is that the gizmos that implement the separate
behaviour of the syntax-juggling code are themselves implemented as
signature-specified _values_ to avoid the problem of having to define
them as macros.

Can this be done in a straightforward way?  I.e. there should be just
another stage.

The essential element seems to be to "split" the identifiers into two
classes: those that are part of the transformer stage and those that
are part of the code stage.