Realized today as I wrote the presentation for PyOpenGL that one of the bullet points "calculate the transform matrix yourself" is going to be a considerable challenge for new users of (Py)OpenGL. OpenGLContext/PyVRML97 has code to do this already for its hierarchy of nodes, but it's a C-coded extension to have reasonable performance.
May need to look into having those calculations available as a C accelerator module in OpenGL_accelerate so that generating (numpy) arrays for regular rotate/translate/scale matrices is efficient... actually, translate and scale are pretty easy, it's rotate that sort of needs a C implementation.
OpenGLContext doesn't have a glFrustum calculator yet, will need to implement that for OpenGLContext, so maybe add it to the accelerators? Might want to have glOrtho calculator available as well... but how far down the road do you go? OpenGL is deprecating these functions, and they're easily looked up wrt the matrix required.
The dividing line likely needs to be something like "these would be too slow to implement in Python and are needed in almost every PyOpenGL script".