From: "NAKAMURA, Hiroshi" <nakahiro@...>
Date: 2003-12-25T10:47:41+09:00
Subject: Re: Ruby 1.8.1 preview4

Hi, Dennis,

> From: "Dennis Ranke" <exoticorn@epost.de>
> Newsgroups: comp.lang.ruby
> Sent: Wednesday, December 24, 2003 10:01 PM

> > I can use "-1.0 / (1.0 / 0.0)" instead of "-0.0" for soap4r test.
> > How does this dump?
> >
> > $ ruby -e 'p(-1.0 / (1.0 / 0.0))'
> 
> -0.0

Thanks.  Tests in 1.8.1 use this workaround.

> Now, since the double is indeed converted to a single-precision float, this 
> is a known limitation of the Playstation 2 hardware which doesn't support 
> denormalized floats. The smallest (positive, non-zero) normalized float is 
> 2**-126, and all results smaller than that just (silently) underflow to 
> zero.

Hmm.

> The smallest positive non-zero single-precision float is exactly 2**-149, 
> so you could change your code to:
> f = (d.abs < 2**-149) ? 0.0 : d

Sure.  It's intuitive.

Regards,
// NaHi