Re: [xml] XPath fix

["Richard Jinks" <cyberthymia yahoo co uk>]

From: "Bjorn Reese" <breese mail1 stofanet dk>
> > Presumably, also functions to set the sign explicitly (i.e., so I can
set a
> > value to -0)?
>
> That is not necessary. You can assign -0.0 to a floating-point variable in
C.
> > As an aside, as you seem to have a good understanding of floating point
> > numbers, could you confirm that both ceiling(-0.1) and round(-0.1)
should
> > return -0 and not +0?
>
> That is correct. According to IEEE 754, the sign is preserved when
rounding
> numbers, including zero.
>
> In general, the sign rules that applies to normal numbers also applies to
> zero. The only exceptions are the following two cases:
>
>  x + x = -0 for x = -0
>        = +0 for x = +0
>        = +0 for x != 0
>
>  sqrt(-0) = -0

Thanks for the explanation.

Apologies if this is a hideous cheat, but...
The code in trionan always appears to work off the idea that a double is 64
bits long.
As the sign is the 64th bit (the one on the left), wouldn't a simple way of
checking the sign just be to right shift the value, i.e.

int trio_get_sign(double a) {
    return a>>63;
}

I suspect there might be a more scientific way, but this seems to be the
simplest, and avoids doing three different things depending on the #defines
/ platform / implementation.

Richard



_________________________________________________________
Do You Yahoo!?
Get your free @yahoo.com address at http://mail.yahoo.com