ufl team mailing list archive

[Anders Logg <logg@xxxxxxxxx>]

On Sat, Mar 12, 2011 at 11:49:10AM +0100, Martin Sandve Alnæs wrote:
> I don't have time now either, but I think it can be done easier:
>
> Find all arguments and coefficients in form or expression.
>
> Make new ones and make a dict mapping from old to new.
>
> Use the existing function replace().

ok, sounds good.

--
Anders


> Martin
>
> Den 12. mars 2011 11.44 skrev "Anders Logg" <logg@xxxxxxxxx> f lgende:
> > On Sat, Mar 12, 2011 at 11:41:23AM +0100, Martin Sandve Aln s wrote:
> >> They should be removed. If used on an element used by an expression, subtle
> >> incorrectness issues will appear, sometimes... Cache issues are only a
> symptom
> >> here, the real problems will be wrong results with no crash.
> >
> > Yes I agree, no objections. The only problem is someone needs to write
> > a transformer that replaces the element degree/shape throughout a form
> > and I won't have time to dig into that just now.
> >
> >
> >
> >> Martin
> >>
> >> Den 12. mars 2011 11.36 skrev "Anders Logg" <logg@xxxxxxxxx> f lgende:
> >> > On Sat, Mar 12, 2011 at 11:28:15AM +0100, Martin Sandve Aln s wrote:
> >> >> Changing repr is incorrect, and the bugs you're finding will never end.
> >> >>
> >> >> It is a fundamental design choice in UFL to make everything
> >> >> immutable, and the repr strings can never change. This is a
> >> >> critical property for lots of things to work, and it is assumed
> >> >> all over the place. It cannot ever be changed.
> >> >>
> >> >> Instead of modifying an UFL object, always create a
> >> >> new one with the changes you want. This is how every
> >> >> algorithm in UFL works, and this is how every future
> >> >> algorithm working on UFL objects must work.
> >> >
> >> > ok, good point.
> >> >
> >> > The reason for these changes are the two functions set_cell and
> >> > set_degree in the finite element classes, which are used to set the
> >> > cell and degree of an element when they are unspecified ("?"). They
> >> > have been around for a long time.
> >> >