Floating point arithmetics

[mauro]

I'd like a part of the standard covering the IEEE representations of nan (not-a-number), +inf, -inf (inifinity).

For example, SICStus Prolog has got an execution mode that supports them.
Why not to make it a standard?

Cheers,

[gotlieb]

Yes, I agree that the standard should conform the IEEE 754 standard for
floating-point computations that include infinities and NaNs. This is crucial for many applications that computes over floating-point values as the existence of symbolic values allows the interpreter/compiler taking care of overflows and exceptions.

Note that the sicstus mode is no more available in the last version of SICStus (4.0). As a consequence, floating-point variable cannot be unified to +inf, -inf or NaN. According to the SICStus Prolog developpers, this is due to their will to conform the ISO standard. They seems to be ready to get back to a mode where symbolic values would be available.

Cheers,
Arnaud Gotlieb

[Joachim Schimpf]

I had a look at floating point related issues again, in the context of the
suggested new prolog flags, but i've also discussed infinity and nan.
My proposal is at http://eclipse-clp.org/Specs/core_float_20091008.txt
(i tried to attach it, but the forum software doesn't allow .txt extensions...)

Best regards,
Joachim