dolfin team mailing list archive

[Anders Logg <logg@xxxxxxxxx>]

On Tue, Dec 05, 2006 at 12:33:51PM +0100, Garth N. Wells wrote:
> Anders Logg wrote:
> > Would it be possible to make the DofMap class not matrix/vector
> > specific?
> >
> 
> We could do something. If a Form knows whether it's bilinear, linear or 
> a functional, DofMapping could return suitable maps.

Soon, FFC will generate an UFC instead of a BilinearForm, LinearForm
or Functional. The UFC is independent of the rank of the operator it
represents. It does know its own rank, so we can check that.

> > We have a lot of code duplication right now in the assembly because we
> > need to think of tensors with different ranks. Maybe it's possible to
> > treat all cases uniformly?
> > 
> 
> Some of the functions in FEM can be removed once DofMapping is actually 
> used. I would have been disappointed if you didn't have any comments, 
> that's why I haven't incorporated DofMapping yet in FEM  :).

I expect to look more at the DofMapping later. One quick comment is
that I'd prefer "DofMap" instead of "DofMapping" since it's shorter.

Another thing we need to consider is how to make the connection
between a BilinearForm/LinearForm/Functional and a UFC form. Probably
the assembly should be based on just a UFC form, but then we can add
abstractions on top of it. A BilinearForm could have a UFC form member
etc.

/Anders