dolfin team mailing list archive

[Johan Hake <hake@xxxxxxxxx>]

On Wednesday 07 October 2009 23:03:41 Anders Logg wrote:
> On Wed, Oct 07, 2009 at 10:59:28PM +0200, Johan Hake wrote:
> > On Wednesday 07 October 2009 22:33:06 Anders Logg wrote:
> > > On Wed, Oct 07, 2009 at 10:30:42PM +0200, Johan Hake wrote:
> > > > On Wednesday 07 October 2009 21:58:36 Garth N. Wells wrote:
> > > > > The Python mixed-poisson demo is failing in parallel because the
> > > > > extraction of sub-functions via split() is making a deep copy, even
> > > > > though it should be a 'shallow' copy. Any ideas on how to fix it?
> > > >
> > > > Yes I see now whats wrong... I thought that the assignment operator
> > > > somewhat kept the shallow copy.
> > > >
> > > > I think we need to add the hack we had previously in DiscreteFunction
> > > > to make it work again. Should we add this in Function directly or
> > > > should I try to subclass Function and implement that particular
> > > > constructor? I would prefer the first.
> > >
> > > What is the hack? Yes, it would be good to avoid creating a special
> > > subclass.
> >
> > It is this constructor we need.
> >
> >    /// Sub-function constructor (shallow copy of the vector)
> >    DiscreteFunction(Function& v, uint i)
> >    {
> >      // Check that vector exists
> >      if (!v.has_vector())
> >        error("Unable to extract sub function, missing coefficients (user-
> > defined function).");
> >
> >      // Get sub-function (Function will store pointer to sub-Function)
> >      Function& sub_function = v[i];
> >
> >      // Copy function space pointer
> >      this->_function_space = sub_function._function_space;
> >
> >      // Copy vector pointer
> >      this->_vector = sub_function._vector;
> >    }
> 
> Looks ok to me (except for the has_vector thing which has been removed).

Ok, I'll add it and fix it on the Python side.

johan

> --
> Anders
>