dolfin team mailing list archive

[Anders Logg <logg@xxxxxxxxx>]

On Sat, Nov 28, 2009 at 10:24:43AM -0800, Johan Hake wrote:
> On Saturday 28 November 2009 00:53:51 Anders Logg wrote:
> > On Fri, Nov 27, 2009 at 11:59:52PM -0800, Johan Hake wrote:
> > > Hello!
> > >
> > > I have not pushed anything, but Expressions should be up to shape now,
> > > and I am working on simplifying the logic in other places of
> > > expression.py
> > >
> > > I do not like the implementation of the dim method. It is used to read a
> > > user defined value dimension and then discarded. I suggest that we add a
> > > value_shape attribute to the dolfin.Expression class. If not set by the
> > > user it is by default (), i.e., a scalar Expression.
> > >
> > >   class MyExpression(Expression):
> > >       value_shape = (2,)
> > >       def eval(self, values, x):
> > >          values[0] = 1
> > >          values[1] = 1
> > >
> > > The dim name is also only making sense for Vector Expressions, where I
> > > think a value_shape attribute makes sense for all occasions. A
> > > value_shape for scalar Expression will still be the default.
> >
> > I think a dim method makes more sense than adding a variable since it
> > is consistent with eval: Then all data a subclassed Expression needs
> > to give is defined by a member function.
>
> Ok, but then I think we should keep it and not just pop it from the
> dictionary, so a user can use it after the class is created.

I just tried follow your lead. I didn't even consider popping it would
remove it.

> Is it very intuitive that dim stands for value dimension? Should we call it
> value_dim or value_dimension?

Maybe not, but dim is nice and short, and it's less formal than
value_rank and value_dimension. So I'd say it should be named dim.

> > If we need value_shape and value_dimension (which is good to have),
> > can't we just store those values as _value_shape, _value_dimension
> > after we have figured out the shape and provide member functions that
> > return those values?
>
> We already have value_rank and value_dimension in the cpp interface. So it
> might be overkill to have these for no other sake.

ok.

--
Anders


> > Also note the clever/simplified interface for dim() allowing (1) not
> > to overload it for scalars and (2) to return the just the vector
> > dimension for vectors: n instead of (n,)).
>
> Yes, I notices.
>
> Johan
>

Attachment: signature.asc
Description: Digital signature