dolfin team mailing list archive

[Anders Logg <logg@xxxxxxxxx>]

On Mon, Mar 10, 2008 at 07:24:29PM +0100, Ola Skavhaug wrote:
> Anders Logg skrev den 10/03-2008 følgende:
> > On Mon, Mar 10, 2008 at 05:57:15PM +0100, Martin Sandve Alnæs wrote:
> > > 2008/3/10, Anders Logg <logg@xxxxxxxxx>:
> > > > On Mon, Mar 10, 2008 at 04:02:33PM +0100, Martin Sandve Alnæs wrote:
> > > >  > 2008/3/3, Ola Skavhaug <skavhaug@xxxxxxxxx>:
> > > >  > >  so perhaps the assembler doesn't know what to do. Is there a way of sending
> > > >  > >  the correct functions (concrete cpp_Functions, for instance) to the assembler?
> > > >  >
> > > >  > assemble(..., coefficients=[...])
> > > >  > but the coefficients list must then include all coefficient functions
> > > >  > (in order).
> > > >  >
> > > >  > Or use the Assembler class directly.
> > > >
> > > >
> > > > It looks to me like you are using FFC forms. Then you don't need to
> > > >  send any coefficients. All functions that appear in the form will be
> > > >  extracted and then sent to the (C++) assembler automatically.
> > > 
> > > But have you implemented this for Functions(mixed_element)?
> > 
> > No.
> > 
> > > Looks to me like that's what didn't work in Olas code.
> > 
> > I don't think Functions() is needed at all. If element = A + B, then
> > it's enough to do
> > 
> > f = Function(A, mesh, ...)
> > g = Function(B, mesh, ...)
> 
> Which is exactly what I did...

I thought you just did

  f = Function(A)
  g = Function(A)

(no mesh and no Vector/eval specified)

This means you can define the form, but then the assembler will have
trouble finding the data.

-- 
Anders