Hi, I've just finished the first part of cleaning fftpack (only zfft for now; wanted to get dev feedback before implementing the scheme for everything). The patch is available as the ticket 408: http://projects.scipy.org/scipy/scipy/ticket/408 The patch consists in two part: - one part split each implementation out of zfft.c, to put it in zfft_name.c files, where name is the name of the backend (mkl, fftw, fftw3, etc...). The corresponding files have no #ifdef, etc... and should be easy to read/improve now. - the above source files are then used to generate a zfft.c file, through a python script. The main reason why I used python to generate the file instead of C macro is because I find this solution easier to add backend: adding a backend only consists in creating a new zfft_name.c file, and adding name to a python list in the python script. Also, as the zfft.c is human readable and do not use new macros, it should be easy to read it now. - Regenerating the zfft.c file is only necessary when one of the backend source is changed, or when a new backend is added. Otherwise, it should work exactly as before; any change of behaviour with respect to configuring/compiling/running is a bug :) I've made no attempt for the python script to be pretty, too cheers, David