[SciPy-dev] scipy_distutils/*_info.py files
pearu at cens.ioc.ee
Mon Feb 18 14:46:58 CST 2002
On Mon, 18 Feb 2002, Fernando Pérez wrote:
> > > Agreed. As long as we don't get in the business of trying to
> write autoconf. :-)
> Have you ever looked at scons (http://www.scons.org)? I know it looks like
> more of a make than autoconf replacement, but it might be worth a look. It
> would be great if here we could concentrate on the Sci part of SciPy and
> leverage other's work for the 'software engineering' part. I realize you
> guys have already been forced to do a ton of work around distutils
> limitations (frankly it's a pretty primitive system, and I'd argue one of
> Python's Achilles' heels).
> I recently also saw QMTest for testing at
> http://www.codesourcery.com/qm/qmtest, also Python-based.
> I may be completely off-mark here, I'm just thinking of how we can best use
> other's work for some areas. The task ahead for scipy is big enough as it is.
> But you guys may have already gone over this, so feel free to shoot me down
> if I'm just blabbering nonsense.
No, we have not discussed about SCons. To me it looks as a great
replacement for distutils. Though it supports currently only C
compilations, one could easily add Fortran support.
However, I don't think that we should move to using SCons right now for
the following reasons:
1) Current scipy_distutils serves scipy needs regarding building extension
modules quite well, including Fortran support. scipy_distutils may need
some polishing when using it with other platforms than Linux,Win32 and
other Fortran compilers than g77. Using current SCons would not ease this
task as it lacks the same things as distutils. And scipy_distutils already
has most of the required hooks implemented.
2) SCons does not solve our autoconf problem. As you said, it's more like
a make replacement. And I think your proposal with Eric's modifications
will be quite adequate for scipy needs.
May be Eric has more to say about QMTest. I have not really looked at
Eric's testing framework yet and therefore I have very little to say
about testing except it is 'must to have'.
More information about the Scipy-dev