So there's a bit of code in PyOpenGL that is now producing deprecation warnings. In essence, we have a type that is a "run-time lookup" integer. So when you have an element that is of unknown size, you lookup that value via some function. That function is often OpenGL-context-dependent. As of latest Numpy that's now producing deprecation warnings, as you're no longer allowed to pass in an object-with-__int__ as a dimension on a zeros() call.

Update: and it turns out that I already dealt with this in the accelerate module, it's only the testing of the non-accelerated release (thanks tox) that revealed the bug.