[r6rs-discuss] R5.95 questions

From: Thomas Lord <lord>
Date: Tue, 26 Jun 2007 12:20:28 -0700

R. Kent Dybvig wrote:
> This is precisely why I have always believed that integer? should return
> false for all inexact values.
>

That there should be controversy over such seems a side effect of
a pointless contention over a scarcity of names. The names are
scarce in the sense that they are insufficient in number to name the
concepts being modeled.

So, IS-CANONICALLY-INTEGER? might return #t
for flownums whose normalized exponent is non-negative (as well as
for exact integers). Leaving INTEGER? for (unambiguous)
membership in *I*. If we want an abstraction for intervals
considered more generally than flownums, we could also
have MIGHT-BE-INTEGER? which is true for exact
integers as well as intervals whose range includes integers.
In any event, TYPICAL-INTEGER might return an
exact integer, for a canonically integer inexact or for a
"might be" integer inexact.

There's good, parsimoious simplicity (like, "all of these flow of
control constructs can be reduced to uses of lambda
and procedure application"). Then there's a more dubious
collective-laziness simplicity (like "numbers in computing are
really hard to understand so let's just given an oversimplified
model of them in the language spec rather than trying to
explain").

The over-simplification is a good thing in a teaching language
where representational aspects of numerics aren't the main
thrust of the course but, for real world applications, its hard
to imagine doing harm by presenting a realistically complete
model, especially in the presence of a library system wherein
people might record the mostly commonly reached-for
simplifications insofar as they correspond to distinct APIs.

-t
Received on Tue Jun 26 2007 - 15:20:28 UTC