Mathemagix

[Joris van der Hoeven <address@concealed>]

On Mon, Dec 10, 2007 at 05:17:22PM +0100, Bernard Mourrain wrote:
> In the first case, an interest is that the code can be reused more easily.
> For instance, if one want to manipulate polynomials in the bernstein 
> basis on a interval [a,b]
> the container will be = [array of coeff. + size + a,b]
> and the operations on this representation will be linked to the 
> bernstein basic operations at the interface class level.
> (In the other approach a specific class has to be rewritten).
> This gives a way to use general functions on all dense polynomials, 
> whereas in the other case, helpers have to be developed, case by case.

I don't agree. In the case of algebramix, it suffices to template
over the polynomial type. You may then instantiate by polynomials
of whatever representation you wish.

> What about multivariate polynomials or sparse structures: should the 
> type of the monomials be attached to the class ?

Yes (or as a template parameter, or in the variant),
because it is part of the implementation.

> The path that you propose requires the current class names to be more 
> explicit:
> 
> vector -> dense_vector
> matrix -> dense_matrix
> structured_matrix
> sparse_matrix  (with many possible formats)

Yes.