Stefan Behnel wrote:
> Dag Sverre Seljebotn, 20.05.2010 12:26:
>
>> The first thing I'd try is to set OPT and CFLAGS to -O0, and possibly
>> remove -march=nocona as well.
>>
>
> I had already removed the "nocona" option and used -O2, that didn't change
> anything.
>
>
>
>> Alternatively, if you just make Hudson's python2.6-bin.tar.gz available
>> to me
>>
>
> You can download it directly from Hudson. Just click on a test job that
> failed, e.g.
>
> https://sage.math.washington.edu:8091/hudson/view/cython-devel/job/cython-devel-tests-py26-c/330/
>
> Then, follow the links under "changed dependencies", i.e. build "#11" of
> "py26-ext-svn" in this case, and click on the build artifacts. That will
> give you exactly the artifact that was used in the failed build. Same for
> the Cython build.
>
Sweet :-)
OK, simply untarring that, and running runtests.py on a trunk Cython,
gives segfaults on NumPy import.
However, if I build NumPy myself and install it ("../py26-ext/bin/python
setup.py install"): No segfaults.
Can you think of anything in the build environments that is different?
NumPy + distutils is (naturally) a rather entagled mess. How is the
build launched? I'd make sure that "/the/python setup.py install" is
launched from the shell, and not through any fancy distribute/setuptools
features.
Dag Sverre
_______________________________________________
Cython-dev mailing list
[email protected]
http://codespeak.net/mailman/listinfo/cython-dev
