dolfin team mailing list archive

[Johan Hake <johan.hake@xxxxxxxxx>]

On Tuesday 09 March 2010 09:46:25 Garth N. Wells wrote:
> On 09/03/10 16:02, Johan Hake wrote:
> > On Tuesday 09 March 2010 00:43:10 Garth N. Wells wrote:
> >> Johan Hake wrote:
> >>> Hello!
> >>>
> >>> I am trying to close the Array typemap Blueprint.
> >>>
> >>> Is it worth changing the double* x/y arguments in SubDomain methods to
> >>> Array? The pros would be to have a consistent Python interface, but it
> >>> makes things more difficult in the C++ implementation.
> >>
> >> What makes it difficult on the C++ side?
> >
> > Well, difficult might not be the right word. My point was that we add
> > more lines in the C++ code to pass Array instead of double* than we
> > remove by making the python wrapper simpler.
> >
> >>> We could maybe add an Array version of the different methods of
> >>> SubDomain
> >>>
> >>>    inside, map, snap
> >>>
> >>> Then can the double * version of the same methods call the Array
> >>> version by default. Then we only expose the Array version to Python.
> >>
> >> I would prefer to have dolfin::Array versions only, if possible.
> >>
> >>> This is not a big deal as it is working fine as it is, but would be
> >>> good to get rid of most of the double * typemaps.
> >>
> >> Yes. It would be good to remove all plain array arguments in the public
> >> interface in favour of safer types.

Should we do this for the GenericFoo::{set,get,set_local} too?

> > Ok, then I try to fix this. I assume we just change all double * to
> > Array?
> 
> Have you made a start on this on the C++ side? I can make the change for
> SubDomain, but I don't want to get in your way.

I haven't, so please have a go! Once the C++ stuff is in place, I can update 
the Python part, if you have not got it to work. I think it is only to make 
the director typemaps in array_typemap greedy (non argument specific).

Johan