dolfin team mailing list archive

[Anders Logg <logg@xxxxxxxxx>]

On Sun, May 17, 2009 at 04:41:13PM +0100, Garth N. Wells wrote:
> 
> 
> Anders Logg wrote:
> > On Sun, May 17, 2009 at 03:43:48PM +0100, Garth N. Wells wrote:
> >>
> >> Johan Hake wrote:
> >>> On Saturday 16 May 2009 21:21:46 Garth N. Wells wrote:
> >>>> The Python demos need to be looked at now - I get a seg fault for Python
> >>>> demos but I don't know what the problem is.
> >>> I have fixed this now. I haven't checked all python demos. The bug was 
> >>> introduced in an esthetic (I think) cleanup of DofMap.h,
> > 
> > Yes...
> > 
> >> where the reference 
> >>> construct where moved before the shared_ptr constructor. Swig encoutners this 
> >>> and choose this instead of the right shared_ptr one.
> >>>
> >>> I have no ignored all reference constructors in, DofMap, FiniteElement and 
> >>> FunctionSpace, so this will not happen again. Are these used anywhere (the 
> >>> reference constructors) or can we remove them from the interface? 
> >>>
> >> They're there to provide a simple interface for constructing objects, 
> >> and partly to simplify wrapping for the Python interface (before 
> >> wrapping shared_ptr was sorted out). Since construction of DofMaps will 
> >> be done primarily via the automatically generated FunctionSpace 
> >> wrappers, I think we can get rid of the reference constructors. Anyone 
> >> who creates a DofMap directly will be a more advanced user and should be 
> >> able to understand the shared_ptr syntax.
> > 
> > I agree. They're not needed anymore to make user code look pretty.
> > 
> > Johan, can you remove them as you know what corresponding changes to
> > make (ignores to remove) in the Python interface?
> > 
> 
> I've half done it already so I'll push in a moment.
> 
> Garth

ok, nice.

-- 
Anders

Attachment: signature.asc
Description: Digital signature