[SciPy-dev] [MAJOR BUG] latest numpy leads to scikits fail [was: installing openopt]

David Cournapeau david@ar.media.kyoto-u.ac...
Mon Sep 22 07:12:55 CDT 2008

dmitrey wrote:
> Robert Kern wrote:
>> On Sun, Sep 21, 2008 at 14:02, dmitrey <dmitrey.kroshko@scipy.org> wrote:
>>> Hi all,
>>> after updating numpy I have troubles with openopt
>>> ImportError: No module named openopt
>>> and seems like it is related to timeseries as well: I have similar bug:
>>> ~/scikits/timeseries/scikits/timeseries/tests# python test_dates.py
>>> Traceback (most recent call last):
>>>  File "test_dates.py", line 21, in <module>
>>>    import scikits.timeseries as ts
>>> ImportError: No module named timeseries
>> In fact, I have fixed a bug related to the combination of setuptools
>> and numpy.distutils when using "python setup.py install". Please
>> remove all scikits and reinstall them.
>> If you are still having
>> problems, please tell me exactly what commands you are using to
>> install.
> Yes I have, openopt works incorrectly, you could check it by yourself
>  >>> from scikits.openopt import NLP
>  >>> NLP(lambda x:x**2,4).solve('ralg')

FWIW, this works for me (with last revision of both scikits and numpy),

> I guess now "python setup.py install" forms Python egg (previously there 
> was a simple copying of Python files to installation directory)

Yes, this is true unfortunately.

> , and 
> openopt can't find path to a solver. Dictionary with the paths to 
> *_oo.py files  is formed via os.walk() during start of openopt session, 
> and os.walk can't handle situation with eggs.

You need to change this, I guess, but for the time being, you can tell
setuptools install to avoid using eggs (I personally always install
scikits this way, since I want to avoid eggs on my system, and this
works well):

python setup.py install --prefix=foo --single-version-externally-managed

I have an alias which shall remain nameless in public to 'python
setup.py install --single-version-externally-managed --record=/dev/null'



More information about the Scipy-dev mailing list