Albert Strasheim wrote:
Hello
Firstly, great work. I always thought SCons was the way to go for NumPy and SciPy, and you've pulled it off.
So basically, I believe most of the things planned in http://projects.scipy.org/scipy/numpy/wiki/DistutilsRevamp are now available because they are available in scons, if numpy developers decide to follow the scons route. Before being usable, I need to finish fortran support and blas/lapack/atlas detection: once this is done, numpy should be able to support some platforms not supported yet (intel compiler on windows, sun compiler with performance libraries, etc...), which was the reason I started this work in the first place.
I don't think you should make the autodetection of BLAS and LAPACK too auto. A .cfg file like mpi4py's mpi.cfg (almost like NumPy's site.cfg, but simpler) would be great. This way, you can provide a few default build configurations for common platforms, but still make it easy for people to configure their build exactly the way they want it, if they so choose.
I agree it should be overridable (I have implemented a NumpyCheckLib, which is just a CheckLib overridable with a site.cfg; I still have to implement the others checkers). But I think it should work out of the box, that is if you have a usable blas and lapack in standard directories, then you should not have to do anything; for this, I really think setting default configuration is NOT the way to go. The autoconf way is much better IMHO, because it would be able to use other compilers without trouble. It would also detect configuration problems before compilation in most cases, cheers, David