Floating-point classification

Floating-point classification

[Olivier Andrieu]

Hello, 

Is there a way to test whether a float has value 'nan' or 'inf' ? I
didn't find in the standard library the equivalent of libc functions
isfinite() or isnan() (those are macros I think). 

I think that in C you can test if a float is nan by comparing it
with itself. It seems to work in  OCaml too :

        Objective Caml version 3.00

# let a = 0. /. 0. ;;
val a : float = nan
# a = a ;;
- : bool = false

Is it correct ? What about 'inf' ?

   Olivier

Re: Floating-point classification

[Xavier Leroy]

> Is there a way to test whether a float has value 'nan' or 'inf' ? I
> didn't find in the standard library the equivalent of libc functions
> isfinite() or isnan() (those are macros I think). 

Right.  Eventually, similar functions should be added to OCaml.
Currently, OCaml provides all math functions specified in ANSI C
(the lowest common denominator of all platforms it supports), but
isfinite(), isnan() and fpclassify() are from ISO C9X, not ANSI C.

> I think that in C you can test if a float is nan by comparing it
> with itself. It seems to work in  OCaml too :

Yes, it works provided the argument to = is statically known to be of
type "float" (so that the float-specific equality operation is
performed instead of generic equality).  As for infinities, you can
also test equality with +inf or -inf.  The following functions should
do the job:

let isnan (x: float) = x <> x
let posinf = 1.0 /. 0.0
let neginf = -. posinf
let isposinf x = x = posinf
let isneginf x = x = neginf
let isfinite x = not(isnan x || isposinf x || isneginf x)

etc, etc.

- Xavier Leroy