RE: [boost] Formal Review for Interval Library beginning

From: Guillaume Melquiond (gmelquio_at_[hidden])
Date: 2002-09-06 15:07:16

On Fri, 6 Sep 2002, David Bergman wrote:

> Guillaume,
> 
> One often attaches semantics to the syntactical operators. The semantics
> one attaches to '<' is usually a relation, which has nothing to do with
> tribool comparisons. So, either a pair belongs to this relation or not.
> You want it to denote a non-relation.
> 
> I agree with Fernando in that if the user wants to fetch the
> indeterminate case, he/she cannot use a binary relation.
> 
> You want '<' to denote a function to a a 3-space, which is not obvious
> for a user, IMEHO.
> 
> So, we seem to be torn by two counter-intuitive extremes: (1) defining
> '<' not as a relation, but as a binary function into a 3-space and (2)
> using a "unnatural" total ordering (e.g., lexicographic) for '<'.
> 
> Which extreme do you prefer?
> 
> My bias is clear, (2), with a proper disclaimer.
> 
> /David

I personally don't have any opinion on this question, since I don't
use comparison operators in my code but only explicit comparison
functions. :-)

More seriously, this is one of the reason why there is a comparison
policy. Various users may have various needs. There is actually four
pre-defined comparison policies: compare_certainly and compare_possibly
define a real binary relation; compare_full allows to handle a 3-state
comparison (especially if the user could set the return type in the
policy); and compare_data is the lexicographic order. And obviously, there
are all the customized comparison policies an user could imagine.

It's why I think the current policy-based comparison system is a good
one. And maybe the main question is: what should be the default
policy?

Regards,

Guillaume
(happy to see that all the comparison problems seems to come to an end)