dolfin team mailing list archive

["Marie E. Rognes" <meg@xxxxxxxxx>]

On 05/30/2011 11:26 AM, Martin Sandve Alnæs wrote:
> Since this feature implementation relies on modifying immutable objects,
> I'm not the least surprised you're getting problems. The bug is not that
> dolfin subdomains are not passed with forms, but that they are allowed
> to be attached in the first place on an existing and assumed immutable
> form object.


Yes...


> The short term solution to this bug is to revert pydolfin back to
> providing subdomains as arguments to assemble and variationalproblem
> where they belong, instead of attaching them to forms. I think this
> should be done for fenics 1.0 if this bug is a problem.
>
> Improvements to the language for expressing subdomains of various kinds
> is in the design stage, but that won't happen before the summer.

Specifications of subdomain does not belong as arguments to assemble and 
variational problem. If you have a form

	L = g*v*dG

where G is a part of a boundary (In semi-math, semi-UFL notation), G 
should be related to the form. Not to the matrix resulting from the 
assembly of the form.

(cc to DOLFIN since the below involves DOLFIN mainly)

The interface to VariationalProblem

	VariationalProblem(., ., bcs, exterior_facet_domains, 
interior_facet_domains, cell_facet_domains)

was rather suboptimal because it assumed implicitly that the bilinear 
and the linear form were defined over the same subdomains. That in, 
combination with dx = dx(0) etc, is highly bugprone.

I care of course because if you want to use the same patent for an 
variational problem with automatic adaptivity, and take care of the 
above, the required  input will look something like this

	VariationalProblem(., ., bcs,
			primal_bilinear_exterior_facet_domains,
			primal_bilinear_interior_facet_domains,
			primal_bilinear_cell_domains,
			primal_linear_exterior_facet_domains,
			primal_linear_interior_facet_domains,
			primal_linear_cell_domains,
			goal_exterior_facet_domains,
			goal_interior_facet_domains,
			goal_cell_domains)

which I can't live with.

The Coefficient/Function magic must involve some of the same issues as 
this. I imagine that a similar way of fixing it should be possible.

--
Marie
	




> ** Changed in: ufl
>         Status: New =>  Won't Fix
>
> ** Changed in: ufl
>     Importance: Undecided =>  Low