[[PageOutline]]

This page explains how to build numpy with NumScons on different platforms, as well as advanced customization possible. All those instructions assume you have successfully installed numscons (python -c "import numscons" should be successful).

= General =

The basic method is the same that with numpy.distutils, except you should use a different setup.py script

{{{
python setupscons.py install
}}}

= Working on the C code =

Numscons is especially practical when working on C code. Numscons only builds changed source code (and dependencies), which makes the compile-debug much shorter. In the case of numpy, the additional feature of build the core as a set of separate files is particularly useful:

{{{
# in place build
NPY_SEPARATE_COMPILATION=1 python setupscons.py scons -i
}}}

Note that you can build without optimization to make the compile cycle even faster, or even adding a full set of warnings:

{{{
# in place build
NPY_SEPARATE_COMPILATION=1 CFLAGS="-DDEBUG -Wall -Wextra" python setupscons.py scons -i
}}}

Or adding parallel builds:

{{{
# parallel builds
NPY_SEPARATE_COMPILATION=1 CFLAGS="-DDEBUG -Wall -Wextra" python setupscons.py scons -i --jobs=3
}}}

Building only one subpackage: this is mainly useful with scipy:

{{{
python setupscons.py scons --package-list=scipy.signal
}}}