Re: signed NaNs, printf and signbit

[yyyyyyyyyy@xxxxxxx]

In a message dated 2003-06-11 6:00:43 AM Eastern Daylight Time, gwc@opengroup.org writes:

> A related issue is whether NaNs always have a sign bit if they are
> supported.  The signbit() description says "NaNs, zeros, and
> infinities have a sign bit", but this is wrong for zeros so maybe
> it is also wrong for NaNs.  (Perhaps that text is meant to be MX
> shaded?)
>
> I would be grateful if anyone can shed any light on these issues.


Most floating point schemes use only zero exponents and mantissas to designate zero, and ignore the sign, and thus while of limited use, setting the sign bit is legal and nominally should be accounted for...

Just a FYI,
Mark