[Numpy-discussion] Code for compatibility with gfortran, license issues

David Cournapeau cournape at gmail.com
Mon Jun 29 22:37:46 EDT 2009

On Tue, Jun 30, 2009 at 3:06 AM, Bruce Southey<bsouthey at gmail.com> wrote:

> I would think that you could just provide an appropriately licensed package
> that combines a separately downloaded numpy/scipy with the  separately
> downloaded/installed gfortran to install the new version of numpy/scipy.

That's exactly what I would like to avoid :) That's a lot of work,
just for a couple of functions.

> Essentially the same way you get non-free software like mp3 decoders for
> certain Linux distros. If that works, then perhaps a clean room
> implementation and rewrite of certain fortran code could be done to remove
> the gfortran dependencies.

As mentioned by Robert, the runtimes (both libgcc and libgfortran),
although under the GPL, have an exception so that linking against them
do not force you to release the resulting code under the GPL terms. Up
to now, numpy on windows and mac os x was built with gcc and have
dependency with those runtimes. What is not clear to me is whether
static or dynamic linking makes a difference (I did not think about
this when building recent mac os x binaries with libgfortran linked

IOW, the only real difference is that I would like to include some
sources under the GPLv3 (with the GCC exception) in numpy/scipy,
sources which would only be used in the cases where we would have to
link against code under the same license anyway.


More information about the NumPy-Discussion mailing list