Front page | perl.perl5.porters |
Postings from August 2010
On Thu, 29 Jul 2010, Andy Dougherty wrote: > On Wed, 28 Jul 2010, Jan Dubois wrote: > > > I'm more and more convinced that using a floating point number in > > boolean context is simply something you should not be doing for > > similar reasons that you generally are not comparing floating point > > numbers for equality. > > Even in floating point, 0 is special. When you know what you are doing, > comparing for exactly 0 is both reasonable and reliable. Yes, but -0 is typically not an explicit integral value, but the result of an arithmetic underflow, followed by rounding the small negative number (that cannot be represented with the current precision) to 0. The sign is only kept to indicate if you approached zero from above or from below. As such you are relying on the fact that your operation actually underflows and produces an exact -0 instead of a very small negative number. This depends very much on the actual representation of the floating point number, as well as the current rounding mode, and is generally not a portable and reliable assumption. Cheers, -JanThread Previous | Thread Next