SRFI-4 says:

> [T]he external representation of instances of the datatype TAGvector is
> #TAG(... elements ...). ... Note that the syntax for float vectors
> conflicts with Standard Scheme which parses #f32() as 3 objects: #f, 32,
> and ().  For this reason, conformance to this SRFI implies this minor
> nonconformance to Standard Scheme.

        So why choose this syntax for float vectors?  Introducing the
ambiguity makes the parser a lot more complicated without any compensating
advantage whatever.  Given that we have #a, #c, #g, #h, #j, #k, #l, #m, #n,
#p, #q, #r, #v, #w, #y, #z to choose from (since none of these is used in
Standard Scheme or elsewhere in SRFI-4), surely we can find a somewhat
memorable prefix that doesn't overload one that is already in use.  Why not
#r32 and #r64, with `r' for `real'?