[SciPy-User] numpy core failure
Tue Nov 2 02:28:56 CDT 2010
Hmm, I'll have to research 'ABI compatibility', but this is really
surprising. I would think this is commonly done. Aren't a lot of folks
out there working on clusters?
I think I'm seeing the result of what you discuss especially in the
behavior of PyNIO. On some machines we have the Ubuntu grib
distribution, but on others I've manually installed it on a networked
drive. It seems this may create a hopeless case! I guess one really
needs a sys-admin who's also quite familiar with Python to accomplish
I don't think at this point I'll try what you suggest, only because
I've been pushing our IT rather aggressively to just bring all our
boxes to 10.04, I hope this will solve the problem! But I do
appreciate the information (and something new to google).
On Tue, Nov 2, 2010 at 2:12 AM, David <email@example.com> wrote:
> On 11/02/2010 05:47 AM, John wrote:
>> Folks, solved it partially. On the machine I compiled numpy with, we
>> had softlinked g77 to another compiler. Numpy used this, but of course
>> it's libraries were only available on the one machine. That solves the
>> libaf90math.so problem.
>> However, on the 9.10 machines I'm still getting this error:
>> ImportError: /lib/libc.so.6: version `GLIBC_2.11' not found (required
>> by /x64/site-packages/numpy/core/multiarray.so)
> It means that you are depending on some symbols from the libc which are
> not available on your machine. If you need to use the same binary on
> multiple machines with different versions, you need to build it on the
> oldest machine you have so that you don't depend on features only
> available in some subsets of your configurations.
> Note however that this is mostly hopeless - ABI compatibility is very
> hard to obtain with python extensions, especially on Linux.
> SciPy-User mailing list
Zope (Zope 2.9.7-final, python 2.4.4, linux2),
Procmail v3.22 2001/09/10
More information about the SciPy-User