[Numpy-discussion] numpy 1.3.0 eggs with python2.6 seem broken on osx, and linux

René Dudfield renesd at gmail.com
Wed Dec 2 05:53:47 EST 2009


On Sat, Nov 28, 2009 at 2:56 PM, David Cournapeau <cournape at gmail.com>wrote:

> On Sat, Nov 28, 2009 at 10:42 PM, René Dudfield <renesd at gmail.com> wrote:
>
> >
> > yeah, I completely understand the unfortunate packaging situation (eg,
> some
> > of my packages do not work with this install method).
> >
> > Here is a simple package requiring numpy.  It uses buildout
> > (http://www.buildout.org/).  To help easily reproduce the problem, here
> are
> > the commands to reproduce.
>
> I have not tried it, but from your explanation, I would suspect that
> buildout does some monkeypatching which is not exactly the same as
> setuptools itself, and that would break numpy.distutils.
>
> The distutils architecture unfortunately requires to take into account
> and special case any distutils extensions, because of the "extension
> as command inheritance" approach. I know I prefer spending my time on
> new packaging solution, but I would take patches.
>
> David
>
>
>

Package/module semantics changed during 2.6/3.0.  I think it might have
something to do with that.  Hopefully some of the py3 work will fix these
problems too.

Actually, I think it might also be a buildout bug.  It seems buildout uses
the sys.path variable to add paths to the python interpreter.  Whereas the
'proper' way is to use the PYTHONPATH environment imho.  Since modifying
sys.path does not affect sub processes paths.  It seems numpy uses a sub
process in the install?  Or maybe buildout/python2.6 use a subprocess.  Or
with 2.5 already having a numpy installed (on osx), then perhaps it was
using the old numpy.distutils already installed to do the install.

Anyway, that's my findings so far.  Not sure if I'll look into it more.


cu,
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.python.org/pipermail/numpy-discussion/attachments/20091202/4a737b52/attachment.html>


More information about the NumPy-Discussion mailing list