🚀 go-pugleaf

Well, aside from the miserable english in my first posting, maybe I didn't
make my point clear.  I agree that memo functions don't preserve the 
semantics of the language in the general case.  What I still fail to see is 
that it doesn't preserve the semantics in this case (i.e. either tak or fib).
The overwhelming majority of ``optimizing'' transformations don't preserve 
language semantics in the most general case (e.g. invariant hoisting fails 
with zero-trip loops, constant folding fails with finite precision arithmetic,
do to doall parallelization fails in the presence of pi-blocks, etc.).  A part
of all optimization is determining the legality of the transformation before 
it is applied.  Is there something obvious about this specific problem that 
I've missed?

David