dolfin team mailing list archive

["Martin Sandve Alnæs" <martinal@xxxxxxxxx>]

2008/7/8 Johan Hake <hake@xxxxxxxxx>:
> On Monday 07 July 2008 11:28:40 Martin Sandve Alnæs wrote:
> I had a general python-sequence to std::vector<number> typemap laying waiting
> to be used. I have added an in-typemap for any python sequence for Array<uin>
> and Array<real>. The typemap will take any python sequence and copying it to
> either an Array<uint> or an Array<real>. An exception is raised if any of
> the elements in the input sequence is not a number.
>
> The following is now possible:
>  f = Function(mesh, (1.0, 2.0, 3.0))
>  f = Function(mesh, [2,2], numpy.array([1.0, 2.0, 3.0, 4.0]))
>
> and if one do:
>  f = Function(mesh, (2,2), (1.0, 2.0, 3.0, "jada"))
>
> one will get
>  <type 'exceptions.ValueError'>: *** Error: Sequence elements must be numbers

Perfect!

> I have also out-typemaps for the same structures, then returning a numpy
> array. But as there could potentially be quite large such arrays, from e.g.
> some of the mesh data, and the data is copied, I was not sure of adding this.
> On the other hand an out-typemap will sure make the python interface a bit
> smoother and coping the underlaying data we ensure that no data is either
> lost or leaked.
>
> Johan

Not sure if that is necessary anywhere, we already can get e.g. Vector contents
or Mesh coordinates out as numpy arrays.

--
Martin