[SciPy-Dev] scipy error compiling csr_wrap under Python 2.7

[Ralf Gommers]

On Thu, Jul 8, 2010 at 9:50 PM, Vincent Davis <vincent at vincentdavis.net>wrote:

> On Thu, Jul 8, 2010 at 7:40 AM, Ralf Gommers
> <ralf.gommers at googlemail.com> wrote:
> >
> >
> > On Thu, Jul 8, 2010 at 9:05 PM, Vincent Davis <vincent at vincentdavis.net>
> > wrote:
> >>
> >> On Thu, Jul 8, 2010 at 5:33 AM, Ralf Gommers
> >> <ralf.gommers at googlemail.com> wrote:
> >> >
> >> >
> >> > On Wed, Jul 7, 2010 at 5:27 AM, Bruce Southey <bsouthey at gmail.com>
> >> > wrote:
> >> >>
> >> >> Hi,
> >> >> I failed to get the scipy 0.8 rc1 and SVN to build under Python2.7
> with
> >> >> numpy '2.0.0.dev8469' and gcc version 4.4.4 20100503 (Red Hat
> 4.4.4-2)
> >> >> (GCC) . But scipy does compile with Python 2.6.
> >> >>
> >> > If this is easy to fix on linux (and someone comes up with that fix)
> it
> >> > makes sense to do so. On OS X the problems are more serious and it
> looks
> >> > like they require changes to numpy.distutils first, so it's not going
> to
> >> > work with numpy 1.4.1 in any case. So full support for 2.7 on all
> >> > platforms
> >> > is not going to happen for the 0.8.0 release.
> >>
> >> Not that I know what I am doing but I do have recent numpy and scipy
> >> installed on oxs running py2.7. I did use numscon and installed 64bit,
> >> I get 4 errors and 2 failures. Lots of warnings, "Warning: invalid
> >> value encountered in absolute" not sure what those are about.
> >>
> > Those errors are all known, so all seems fine. How did you install Python
> > 2.7? I see your python says it's compiled with gcc 4.2, while I remember
> my
> > scipy builds picking up 4.0 by default.
>
> Well it was not skill or knowledge :-) pure blind trial and error but
> it seems simple now.
> for python 2.7 from source
> ./configure --with-universal-archs=64-bit
> --enable-universalsdk=/Developer/SDKs/MacOSX10.5.sdk
> --enable-framework
> then make, make install
>
> I have been taking notes so I can repeat it. More details here (numpy,
> scipy)
> http://vincentdavis.net/installing-python
>

Interesting. When I try the same build command for numpy with either of the
2.7 binaries from python.org (both built with gcc-4.0), numscons aborts
after failing this test:
# We check declaration AND type because that's how distutils does it.
if config.CheckDeclaration('PY_LONG_LONG', includes = '#include
<Python.h>\n'):
    st = config.CheckTypeSize('PY_LONG_LONG',
                              includes = '#include <Python.h>\n')
    assert not st == 0

And setup.py install fails because it complains about missing x86_64
architecture.


> By the way what are "Warning: invalid value encountered in absolute"
> warnings I tried to find out more but didn't find much.
> Vincent
>
> Those are from ufuncs operating on masked arrays - can't remember why it
was hard to fix, but it's a known (and annoying) issue.

Cheers,
Ralf
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.python.org/pipermail/scipy-dev/attachments/20100708/17317623/attachment.html>