PyPy [1] is an alternative implementation of a python interpreter written in python. It includes a JIT to speed up frequently run code [2]. The canonical python interpreter is written in c, hence cpython[3], numpy interfaces with this cpython interpreter through its capi [4]. PyPy has a c api, but it is "slow" and not JIT friendly. Some of us thought it would be fun to reimplement the core numpy ndarray as a first-class PyPy object, and allow the JIT to work its magic. We are 90% of the way there, which means we only have 90% of the effort left to do.

Most of the basic stuff works, significantly missing are SIMD optimizations and lazy expression evaluation.

Unfortunately, much scientific python software is written using the cpython capi, so it will require quite a bit of retooling of things like scipy [5] for the implementation of numpy on pypy to become popular.

[1] http://pypy.org [2] http://speed.pypy.org [3] http://python.org [4] http://docs.python.org/2/c-api [5] http://www.scipy.org/