Probably over 70% - 80% of code generation, from .idx and .xml, to .i is related to the smart pointer wrapping. This is a serious performance issue when they are not actually wrapped. Each language must have the choice to turn this feature off completely.

Smart pointer wrapers define all the inherited methods and that's why they occupy most of the code generation performance.

Probably out of context, but I am working on updating the wrapitk mandriva package, to rebuild it with gcc 4.5.1 and python 2.7. Now it fais to build due to errors like:

/home/mandrake/rpm/BUILD/wrapitk-0.3.0/build/Typedefs/Base.includes:985:9: error: 'itk::LightObject::Pointer::SmartPointer' names the constructor, not the type

I am not a C++ programmer (only C) or have good understanding of itk, I plan to try two builds with the proper alternate change of the initial patch attached, that is, one build with 's|Pointer::SmartPointer|Pointer|' and another with 's|Pointer::SmartPointer|SmartPointer|', and if both build, choose one... But I think it should be obvious to wrapitk developers what is the correct approach.