On 2020-08-27 19:05 -0400, John Cowan wrote:
> I'm now wondering if we shouldn't just get rid of inexact number support in
> numeric-range.  It'll still be possible in a non-numeric range, to be sure,
> but a warning about the dangers of that should be enough.  I remember
> reading warnings against inexact-number DO/FOR loops a very long time ago,
> and I wondered what had happened since.

I'd rather keep inexacts in numeric-range.  Removing them is mainly a
change of emphasis--we still have to handle ranges over inexact
numbers, but it's slightly harder to create them.

Part of SRFI 196's shtick is also that "Roundoff error ... is greatly
reduced compared to accumulated error by repeated adding".  This
suggests that ranges over inexacts are useful, and not as Evil and
Wrong as inexact-number DO/FOR loops.

> If we do this, some of the considerations go away at least.  Do we still
> need a constraint on indexers to eliminate non-monotonic indexer functions,
> or is that no longer necessary either?

Yes, unfortunately.  range-drop, etc. still need to do the right thing
with ranges with non-monotonic indexers.

--
Wolfgang Corcoran-Mathe  <xxxxxx@sigwinch.xyz>

"I started out as a BASIC programmer.  Some people would say that I'm
permanently damaged.  Some people are undoubtedly right." --Larry Wall