I'd like to clarify what exactly we mean by exposing a C API. Do we have in mind that our random number generators can be used from standalone C code, or via Cython `cimport` like with the current numpy.pxd?
It sounds like we want to expose the highest level generators; do we also want to provide access to the bit streams?
What do you mean by standalone C? A Python extension is written in C (but not Cython)? Or a C application that doesn't include Python.h? The former is pretty easy since you can use a few PyObjects to simplify initializing the bit generator, and the rest of the code can be directly used in C without any more Python objects. The latter is also doable although the low-level functions needed to initialize the bit generators (which are just C structs) have no standardization. I think the only component in a standalone C application that would need some non-trivial work is SeedSequence (i.e., more than changing function names or reorganizing files).
Like Robert, I suspect that Cython users would be the largest immediate beneficiaries of a lower-level API.
numba end-users can already consume the bit generators through the exposed CFFI/ctypes interface they provide. These can then be used with the higher-leverl generators, although end users have to build a shared lib/DLL first. Getting the C API in shape to be used directly by numba is probably a bigger task.
Kevin