[Numpy-discussion] Controlling the way a module is imported

[David Cournapeau]

Robert Kern wrote:
>
> Can you be more specific?

Sure: in my branch to refactor fftpack,  every non default backend (that 
is everything but fftpack) is a separate python module, which implements 
some fft functions, and is 'importable'.

So in scipy.fftpack, I have a function which:
    - tries to find one of the backend from a list (by using __import__)
    - if no backend is found, just use the default backend
    - if one is found, pull the functions from the found backend, and 
use the default backend as a fallback for non available functions (each 
backend can implement a different subset of fftpack).

This is done in the following prototype:

http://projects.scipy.org/scipy/scipy/browser/branches/refactor_fft/scipy/fftpack/common.py

I would like to add the possibility to  control which backend is used: 
the problem is how to communicate the name of the backend to 
scipy.fftpack in a persistent way (such as if you want to use mkl, it 
will remember it at the next python session), and in a dynamic way (such 
as you can decide which one to use before importing scipy.fftpack, or 
even changing the backend on the fly).

thanks,

David