Jussi Piitulainen <xxxxxx@ling.helsinki.fi> writes:

> I've been thinking of this. I begin, slowly, to see that (1) there
> really is essentially no runtime cost for arrays, and (2) that R5RS
> vectors need not be redefined at all. So it seems like a win.
>
> Now, sharing a vector, or an array that shares a vector, can in some
> cases produce an array that essentially is that vector. Should we
> specify that the result of share-array is never a vector, or is a
> vector when it can be, or leave it unspecified?

I don't think that we want to specify that an array construction
procedure must return a vector in any case -- completely disjoint
vectors and arrays should be allowable as an implementation tactic.
SHARE-ARRAY should not be different than other array construction
procedures:  if a vector return value meets the specification, then it
should be allowed but not required.

This freedom of implementation should not be a burden on array-using
programs.  Objects produced by array construction procedures may
always be indexed using ARRAY-REF, objects produced by standard Scheme
vector construction procedures may always be indexed using VECTOR-REF.