Le 23/07/2015 13:34, Boris Yakobowski a écrit :
> Hi Sébastien,
>
> I feel obligated to point out that the _semantics_ of floating-point
> comparison is a bit tricky. IEEE 754 mandates that NaN == NaN should
> return false (as well as NaN != NaN), breaking all algebraic laws
> known to mankind :-).

Beaware:

Nan <> Nan  -> true
Nan = Nan -> false
Nan != Nan and Nan == Nan : depends on the memory layout.

> OCaml's operators '=' and '!=' follow this convention, but  'compare
> nan nan' returns 0, which is usually the desired behavior. However,
> 'compare 0. (-0.)' also returns 0, while you might want to distinguish
> those two values.
>
> HTH,
>
> On Thu, Jul 23, 2015 at 10:35 AM, Sébastien Hinderer
> <Sebastien.Hinderer@inria.fr <mailto:Sebastien.Hinderer@inria.fr>> wrote:
>
>     Dear all,
>
>     What's the most efficient way to compare floats, please?
>     Is it the polymorphic compare function, or is there a more specialized
>     version of it?
>
>     I saw Float.compare mentionned on the web but that does not seem
>     to exist
>     any longer?
>
>     Thanks,
>
>     Sébastien.
>
>     --
>     Caml-list mailing list.  Subscription management and archives:
>     https://sympa.inria.fr/sympa/arc/caml-list
>     Beginner's list: http://groups.yahoo.com/group/ocaml_beginners
>     Bug reports: http://caml.inria.fr/bin/caml-bugs
>
>
>
>
> -- 
> Boris