I finally tracked down the problem with nurbs rendering in PyOpenGL 3.x. Apparently passing a bald integer where a void-pointer is expected was causing ctypes to re-interpret the value. I went with returning a void_p and the nurbs started flowing again.

Added support for saving/viewing the depth buffer to one of the OpenGLContext tests. Added a few nurbs tests to the OpenGL tests too. I'm testing on Win32 now. Hint for PyOpenGL users... do not attempt to run OpenGL/OpenGLContext apps over an smb share... 14+ seconds on my Qemu emulator just to load the application. Runs fine once it's up, but the file access time is a killer.