Re: Number/Algebraic concepts (was RE: [boost] RE: rational.hpp)

From: Doug Gregor (gregod_at_[hidden])
Date: 2001-01-14 22:30:22

On Sunday 14 January 2001 08:59, you wrote:
> That's a good question. I think the named conformance checking could work
> well for the semantic information, and could easy be added to the
> structural concept checks. However, I'd like to see something much simpler
> than what is in the Algebra library... something along the lines of
>
> is_associative<Domain, Op>::type   (true_type or false_type)
>
>                           ::value  (true or false)
>
> Then use this in a static assert in the appropriate concept checking
> classes.

Adding a simpler interface to Algebra is easy enough, but I do agree that its 
concept specification is complex. Oddly enough, the design is/was intended to 
make life easier for the user :).

In any case, for named conformance concept checking I would suggest 
representing a concept lattice as an inheritance hierarchy. Then, checking 
named conformance (taking subconcepts into account) reduces to checking a 
pointer conversion similarly to Boost's is_convertible. Unfortunately, the 
most complicated portion of the design is in handling relationships that 
aren't strictly refinement, or for which multiple refinement possibilities 
exist (e.g., the Abelian group/Abelian monoid relationship). Tecton lemmas 
handle this well, but introduce a lot of ugliness into the hierarchy.

> It would be nice if we could have a fairly portable system, works
> with VC++, Borland, etc. That means no partial spec. :(

As long as the is_convertible variants can be made to work, I think the 
system could be made portable for the simple checks (e.g., is_associative). 
Unfortunately, extracting informance from the hierarchy has proven 
complicated so far because it requires traversing the hierarchy using a lot 
of partial specialization... the kind that can't be simulated with member 
templates :(.

        Doug