ufl team mailing list archive

[Anders Logg <logg@xxxxxxxxx>]

On Mon, Mar 30, 2009 at 03:28:32PM +0200, Anders Logg wrote:
> On Mon, Mar 30, 2009 at 02:00:51PM +0200, UFL wrote:
> > One or more new changesets pushed to the primary ufl repository.
> > A short summary of the last three changesets is included below.
> > 
> > changeset:   780:bc9680be05bfd5048c3de1fbcc2f9281c3f2c5ef
> > tag:         tip
> > user:        "Martin Sandve Alnæs <martinal@xxxxxxxxx>"
> > date:        Mon Mar 30 14:02:24 2009 +0200
> > files:       ufl/algorithms/formfiles.py ufl/integral.py
> > description:
> > Implemented better version of tuple syntax:
> > 
> >   a = (u,v)*dx + (f,v)*dx
> > 
> > in this case a is a Form, so no changes are needed anywhere else.
> 
> Nice. I thought about this (implementing __rmul__ to handle tuples)
> but wanted the even shorter version (without dx).
> 
> Can we keep both options?
> 
> I haven't figured out how to make load_forms work though. The current
> implementation seems to extract *all* forms from a file, which may not
> always be appropriate. One might want to keep contributions in
> separate variables and then add up the results:
> 
>   CONV = (v, dot(beta, grad(u)))*dx
>   DIFF = (grad(v), grad(u))*dx
>   a = CONV + DIFF
> 
> Code should here be generated only for the form a, not CONV and DIFF.
> 
> FFC solves this by using reserved variable names for forms:
> 
>   a: bilinear form
>   L: linear form
>   M: functional
> 
>   element: single element

If this is not enough, load_forms could also look for a variable named
"forms" that lists the forms that should be compiled:

  forms = [a0, a1, L]

or

  forms = ["a0", "a1", "L"]

and if this is not found there could be a sensible default (like a, L,
M).

-- 
Anders

Attachment: signature.asc
Description: Digital signature