[SciPy-dev] Build problerms on various Windows versions

eric eric at scipy.org
Wed Feb 13 11:05:26 CST 2002


> The latest scipy in CVS should now build with Python 2.2.

Very good news!  A lot of people have asked about this, so its nice to have it
fixed.
>
> On the other hand, I find the issue mentioned in the setup_integrate.py
> comment:
>
>     #  Note that all extension modules will be linked against all c and
>     #  fortran libraries.  But it is a good idea to at least comment
>     #  the dependencies in the section for each subpackage.
>
> a bug in scipy_distutils.

Agreed.

> I would prefer that when defining an Extension,
> then it should _explicitly_ define also all libraries that this extension
> module depends on (this will avoid these strange dependency
> problems).

I can't remember why we did this in the first place, but I'm almost sure it was
work around.  Explicitly specifying the needed libraries is a good idea as far
as I can tell.

>
> Basically, scipy_distutils.setup should not use
> 'fortran_libraries' keyword as an extension of the 'libraries' keyword.
> In fact, both these keywords should not be used when a package defines
> more than one extension module with different set of libraries.
> If nobody will mind, I'll try to fix this in scipy_distutils.
> What do you think?

So your plan is to specify the static fortran libraries that need to be built in
fortran_libraries.  Then the extension modules have to list these libaries again
if they want to link the them.  Extension libraries *never* link against the
fortran_libraries setting.  It is only done against libraries settings.  Is this
right?  I say it is a reasonable plan.

see ya,
eric






More information about the Scipy-dev mailing list