[SciPy-Dev] updating numpy scipy crash - numpy on pypi

[josef.pktd at gmail.com]

On Thu, Jan 13, 2011 at 9:37 AM, Ralf Gommers
<ralf.gommers at googlemail.com> wrote:
>
>
> On Thu, Jan 13, 2011 at 6:49 PM, <josef.pktd at gmail.com> wrote:
>>
>> On Thu, Jan 13, 2011 at 5:14 AM, Ralf Gommers
>> <ralf.gommers at googlemail.com> wrote:
>> >
>> >
>> > On Thu, Jan 13, 2011 at 1:29 AM, <josef.pktd at gmail.com> wrote:
>> >>
>> >> I *finally* updated numpy and scipy.
>> >>
>> >> For numpy 1.5.1, I used the files on pypi. When I installed scipy
>> >> 0.9b1 it crashed (segfaulted) in the testsuite.
>> >>
>> >> With the superpack installer for numpy from the sourceforge download,
>> >> scipy 0.9b1 passes all tests
>> >>
>> >> Ran 4806 tests in 287.843s
>> >>
>> >> OK (KNOWNFAIL=12, SKIP=35)
>> >> <nose.result.TextTestResult run=4806 errors=0 failures=0>
>> >>
>> >> It looks like the non-superpack numpy files on pypi are not compatible
>> >> with scipy in my setup (win32, sse2, python 2.5.2)
>> >
>> > Hmm, they should contain only a single SSE (but not SSE2/3) instruction.
>> > Can
>> > you check by running
>> >
>> > http://projects.scipy.org/scipy/attachment/ticket/1170/detect_cpu_extensions_wine.py
>> > (only need to modify hardcoded path to scipy)?
>>
>> I don't know if I can run a wine script. In either case, I have now
>> sse2 numpy and sse2 scipy, which both work without problems so far.
>>
>> My conclusion was that sse2 scipy (from the superpack) is not binary
>> compatible with the single SSE numpy from pypi.
>
> OK that makes sense. The reason to have the nosse binary on pypi is to make
> easy_install work on Windows, but this is the downside then.

However, I think this is pretty dangerous. If the incompatibility is
really present,
then everyone that downloads the binaries will end up with install problems when
they also want to install scipy.

I think there should be at least an indication on the pypi page, that
users should
go to sourceforge if they want to have a "proper" binary for their system.

Is it possible to rename the pypi binaries to xxx_nosse or put the
superpacks also on pypi?

On the other hand, I'm bit surprised that I'm the first one to report this.

Josef


>
> Ralf
>
>
> _______________________________________________
> SciPy-Dev mailing list
> SciPy-Dev at scipy.org
> http://mail.scipy.org/mailman/listinfo/scipy-dev
>
>