data directed programming and types (was Re: learning how to use MrSpidey)

[Marco Maggesi <maggesi@math.unifi.it>]

Matthias Felleisen <matthias@cs.rice.edu> writes:

> The answers are no, no, and no.
>
> Two reasons:
>
> Spidey implements a naive set-based interpretation of programs. In
> particular, all arguments that flow into the first position of a binary
> function are co-mingled with all arguments that flow into the second
> position (well more or less).
>
> Furthermore, Spidey's if-splitting works only when you have a single
> predicate in a test. A conjunction pollutes the analysis.
>
> -- Matthias


So, please, let me ask for a suggestion for the original problem I had
in mind.  I would like to write a computer algebra package that
implements Buchberger's algorithm for Groebner basis (useful for
elimination in polynomial equations and many other applications in
commutative algebra).

My original idea was to use a data directed programming style very
much like in the example of the generic algebra package illustrated in
SICP.  As far as I understand, I cannot have userful support for static
type debugging in this way.  So, are there better strategies?

Thanks again, MM