Re: [boost] [Interval Comparisons] Proposed resolution

From: Sylvain Pion (pion_at_[hidden])
Date: 2002-09-08 19:40:03

On Fri, Sep 06, 2002 at 05:34:06PM -0400, Douglas Gregor wrote:
> I propose that we:
>  1) Remove comparison policies _in their current form_
>  2) Create a default comparison semantics that makes interval<T> a numeric 
> type that will work well in generic functions (it should throw if rounding 
> errors cause a comparison to be indeterminate)
>  3) Allow other comparison semantics to override the default semantics within 
>     a particular scope

I like your proposal.

I am also now convinced that the (lexicographic) std::less specialization is
probably a good thing to have.

> [Implementation Details]
> Here's a short sample program that describes the implementation scheme that 
> can be used. The crucial point is that the definition of operator< inside the 
> interval class is the default operator<. However, the second argument (not 
> 'this') requires a (no-op) implicit conversion via the interval_holder class 
> template. Thus, if any operator<(const interval<T>&, const interval<T>&) is 
> available (e.g., introduced via a 'using namespace ...'), that operator< will 
> be a better choice than the operator< inside the interval class, because the 
> second argument will only require cv-qualifier adjustments instead of a 
> user-defined conversion by constructor. The result is that 'using namespace 
> my_interval_semantics' brings in a set of operators whose semantics are 
> always better than the default semantics.

Nice trick !

-- 
Sylvain