>From: Conrad Schneiker <schneik / austin.ibm.com>
>Subject: [ruby-talk:5113] Re: Proposal: Add constants to Math

> On this and related topics of what results should be produced under
> various conditions, does anyone know what Python does?

By searching with grep, I can't see any NaN-generating code
in either the Python or its Numerical Extension.
GSL-0.6 defines it from NAN macro or (0.0/0.0),
but they will "sort out" it, according to their TODO.

IMHO, A practical way is, to check IEEE 754 support at the
configure time, and then define Float::NAN with 0.0/0.0 or
the bit pattern of NaN.

Masa Tanaka