Is there no way to specify non-NaN flonum types?

[Aeva]

What version of Racket are you using?

8.0

What program did you run?

(:type #:verbose Positive-Float) prints (U Float-Nan Positive-Float-No-NaN), which is odd because (positive? +nan.0) returns #f.

(:type Positive-Float-No-NaN) throws an error claiming Positive-Float-No-NaN is unbound. As far as I can tell, under no circumstance does Positive-Float-No-NaN appear to be a valid type that I can use 😭 .

What should have happened?

It seems like Positive-Float-No-NaN should be a type? I've been digging through the docs for hours trying to work out a way to specify a Non-NaN float as a type, and it just doesn't seem to be possible???????

Also the mismatch between Positive-Float and positive? is weird, but it looks like this is a known issue (see #750).

If you got an error message, please include it here.

. Type Checker: parse error in type;
 type name `Positive-Float-No-NaN' is unbound in: Positive-Float-No-NaN

[samth]

I agree that the printing/input difference is a bug. And #750 should probably be fixed.

However, non-NaN types are basically useless in our experience, and the fact that there's a distinction in the type system is a historical legacy. Effectively no operations can be shown to preserve lack-of-NaN without much fancier reasoning than TR does. When we had specifications of what could produce NaN, they were always buggy, and that's why we got rid of them.