[racket-dev] nan?, infinite? and regular-real? [Was: Latest Plot Package]

From: Matthew Flatt (mflatt at cs.utah.edu)
Date: Fri Nov 18 22:21:35 EST 2011

Except for `finite?' (which seems to be covered by `rational?'), these
additions sound fine to me.

I'm not sure whether they should go in `racket/math' or `racket/base',
though. Although it feels wrong to keep adding to `racket/base', two
thoughts make me lean in that direction:

 * It will be confusing to have bindings such as
   `exact-nonnegative-integer?' in `racket/base', while binding such as
   `nonnegative-real?' are in `racket/math'. Putting all of those
   functions in the same place seems more consistent.

 * Adding things to `racket/base' probably leads to fewer compatibility
   issues. That's because `racket/math' could be explicitly imported
   into a module that defines `nonnegative-real?', leading to an
   import--definition conflict. If `racket/base' as a module language
   provides nonnegative-real?', then a module's own definition of
   `nonnegative-real?' can shadow it without conflict.

At Fri, 18 Nov 2011 12:47:16 -0700, Ryan Culpepper wrote:
> On 11/18/2011 12:34 PM, Neil Toronto wrote:
> > On 11/18/2011 12:22 PM, Ryan Culpepper wrote:
> >> On 11/18/2011 12:13 PM, Neil Toronto wrote:
> >>> I've moved this to dev from a private discussion with Doug, who has just
> >>> tried the nightly build.
> >>>
> >>> On 11/18/2011 11:43 AM, Williams, Doug wrote:
> >>>> 2) The plot/utils has nan? and infinite?, which are also exported from
> >>>> the science collection and I had to explicitly exclude them in the
> >>>> require. Can we talk to the powers that be and see about getting them
> >>>> (and finite?) moved to racket/math? Then we can both remove them from
> >>>> our code. [I'm not sure about any interaction with the r6rs functions.]
> >>>
> >>> Powers That Be, Doug and I would like `nan?' and `infinite?' to be in
> >>> `racket/math'.
> >>>
> >>> I made an exact equivalent of Doug's `finite?', but called it
> >>> `regular-real?'. I use it in all kinds of function contracts (it returns
> >>> #f for non-reals). We would also like that one, with either name, in
> >>> `racket/math'.
> >>
> >> How does 'finite?' differ from 'rational?'?
> >
> > Whoa! It doesn't!
> >
> >> If we're making a wishlist, I'd really like
> >> '{positive,nonnegative}-{real,rational}?'.
> >
> > For use or for contracts? For contracts, we already have `>/c', `</c',
> > `>=/c' and `<=/c', which can do the same as `positive-real?',
> > `negative-real?', `nonnegative-real?' and `nonpositive-real?' if applied
> > to 0, and are arguably easier to read.
> 
> I'd argue that they're less easy to read. Also, despite their 
> availability, I've still seen contracts that say 'real?' even when they 
> mean '(>=/c 0)', etc. (Or possibly 'nonnegative-rational?'.)
> 
> > For the `*-rational?' ones, would they identify only exact rationals?
> > Because that would be inconsistent with `rational?'.
> 
> No, they should be consistent with 'rational?' and accept inexacts.
> 
> > If we *are* making a wishlist, I'd like to add exact-round, exact-floor,
> > exact-truncate and exact-ceiling. I rarely need to chop off fractional
> > parts without also making the result exact.
> 
> +1
> 
> Ryan



Posted on the dev mailing list.