fenics team mailing list archive

[Anders Logg <logg@xxxxxxxxx>]

On Thu, Feb 24, 2011 at 07:54:35PM +0000, Garth N. Wells wrote:
>
>
> On 24/02/11 19:20, Anders Logg wrote:
> > On Thu, Feb 24, 2011 at 04:23:36PM +0000, Garth N. Wells wrote:
> >> We've just added 'create' functions to UFC, but it would also be useful
> >> to add 'copy' functions. It'd like to get a copy a ufc::dofmap, but I
> >> can't do it at present. 'create' doesn't work since I don't have the
> >> mesh available in order to initialise the ufc::dofmap that it returned
> >> by 'create'.
> >
> > Implementing copy might be problematic for form compilers since it
> > would involve copying data that may have been generated as part of the
> > call to init_mesh and init_cell. It's no big deal in the case of FFC
> > since only _global_dimension needs to be copied, but it might be
> > problematic in general. The more functions we add to the interface,
> > the more work it is to implement the UFC interface.
> >
> > It would be better to store the Mesh as part of the DofMap so that a
> > copy can be created using the create function.
> >
>
> I'm reluctant to add the Mesh to DofMap, since unless we store a copy of
> the Mesh the Mesh could change which would make it inconsistent with the
> dof map. Also, I've tried to simplify DofMap and minimise the private
> data because it can get pretty complex with views, parallel ownership, etc.

Good.

> For now I've added GenericDofMap::copy(const Mesh& mesh), which is ok.
> Internally, DofMap creates some UFCMesh objects which could be avoided
> if ufc::dofmaps could be copied.

I was going to suggest something like that.

--
Anders