[SciPy-User] [build] visual studio 2008 and intel fortran 11.1 compiler

[David Cournapeau]

On Wed, Dec 16, 2009 at 1:51 PM, Eloi Gaudry <eg at fft.be> wrote:

> An
> alternative to distutils is a good thing, especially if it's easily
> maintainable and configurable (i.e. adding support for newer compilers
> or changing whatever compilers flags, macro, etc. should be something easy).

Yes, indeed. Numscons was a first tentative in that direction.

> Does this means that all future versions of numpy and scipy will be
> built (default behavior) using numscons ?

I myself use numscons pretty much exclusively, but I still use
distutils to build the official binaries. I think numscons works well
for developers, but it often fails for end-users who want to build,
and this cannot be easily fixed if at all while being used within
distutils (numscons still depends on distutils for packaging, etc...).

Currently, I am working on a full, written from scratch packaging
solution. I have officially announced it at Scipy India:

git://github.com/cournape/toydist.git

The main idea is to have a declarative file which can fully describe
all that's needed for most packages, and have a stable and specified
interface with real build systems like make/scons/waf for the others.

David