UFL state should contain Coefficients rather than FiniteElements

[gmarkall]

Change semantics of state.scalar_fields[] etc. to return a coefficient instead of an element. If you want the element, use T.element() (where T is the field from state) instead. That way, you can use T in a form.

[kynan]

I see a potential problem here:

• we need T to be a TrialFunction if it is a field we want to solve for (e.g. the field we define the UFL equation for)
• we need T to be a Coefficient otherwise

Have I got this right?

[gmarkall]

I think the idea is that it should be used like this (for a Poisson's equation example):

F = state.scalar_fields["SourceTerm"] # T is a Coefficient instance E = F.element() v = TestFunction(E) u = TrialFunction(E)

a = dot(grad(v),grad(u))_dx L = v_F*dx

soln = solve(a,L) state.scalar_fields["Solution"] = soln

I think this avoids the issue of whether it is a TrialFunction or a Coefficient, by explicitly getting the element of the field and using that to define Test and Trial Functions.