Finally, I've achieved one of my long-term goals for the VTK python wrappers: automatic wrapping of templated types! And if I had know how much work it would entail, I probably never would have started, but still, I'm glad that I did it.
The first part, parsing the templates from the C++ header files, is something that I did about a year ago. Templates are one thing that my own parser handles very differently from gccxml. Whereas gccxml spits out information only for the templates that are instantiated in the code, my parser spits out descriptions of the generic templates prior to instantiation. So the biggest part of my current project was creating subroutines that could take the info about generic templates, and generate info for instantiated templates that could then be wrapped. A big advantage of doing things this way is that the wrappers can produce any instantiations that are desired, whether or not they appeared in the original header files. Another advantage is that my wrappers can provide documentation for both the generic and the instantiated class templates.
In python, the templates are a class generator of sorts. They look kind of like a class, and kind of like a dict. They are used as follows:
v = vtkVector[float,3]([1.0, 2.5, 10.2])
The template args are provided in square brackets. Given the template args, the template "vtkVector" instantiates the template to produce a class. The class is then given arguments in parentheses, which it uses to construct an object. Other than the use of square brackets instead of angle brackets, the main difference from C++ template instantiation is that only a limited selection of template args are supported. Calling help( ) on the template will list all the available args.