Was just reading through the GLEWpy source-code for the ARB vertex-buffer-object extension. It's basically about the same level as the auto-generated SWIG, or ctypes interfaces. It should work fine, but it doesn't feel natural. Brian's interface looks a lot more natural to me, but it's also a lot more code.

I just have this nagging suspicion that the proper way to move the project forward (and keep it going easily) is to teach these tools how to naturally deal with Numpy arrays when auto-wrapping, rather than creating more and more code to handle the corner cases. That said, I'm not sure VBO as an extension is a good thing to consider as an exemplar. It appears to be one of the "messiest" extensions from a conceptual standpoint.

Or maybe manually wrapping everything is the way to go. Write the code once and write it so well that no-one will ever want to change it again... nah. That wouldn't be any fun.