[boost] Re: Proposal : Interval arithmetric ...

From: Valentin Bonnard (Bonnard.V_at_[hidden])
Date: 2000-03-14 12:22:41

Jens Maurer wrote:
> 
> jsiek_at_[hidden] wrote:
> > Looking at the interval proposal there's some missing functionality.
> 
> I've added some of the missing functionality.
> 
> >  Also it includes the assembly code
> > for Solaris rounding modes.
> 
> Since it is for non-commercial use only, it is incompatible with boost
> copyright rules.
> 
> > There are some issues that need carefull attention, the behaviour of
> > divide by zero,
> 
> This currently gives [-inf, inf] with my class.

(assuming 0 means exact 0)

This doesn't seem correct to me. Dividing by a very small value should 
cause overflow (infinity of the sign of the product). Dividing by 0 
makes no sens.

With fp numbers, 0. may means very small. Interval is expected to be 
exact: [0, 0] means 0.

Two behaviours are possible:
- throw an exception
- x/0 returns { r | r*0 \in x } (the empty interval if x doesn't contain 
  0, [-inf, inf] otherwise)

BTW, I saw something funny in the doc 
(at http://www.egroups.com/docvault/boost/interval/interval.html):

interval multiplicative_inverse(const interval&)

Returns: 1/*this 

-- 
Valentin Bonnard