[SciPy-user] scipy 0.7.0.dev4373 + atlas FAILED (failures=2, errors=12)

Xavier Gnata xavier.gnata@gmail....
Sun May 18 07:02:24 CDT 2008


Indeed, scipy is very easy to compile when the lib are installed.

well atlas using the package is ok.
umfpack is not on my ubuntu.
Yes I'm able to compile scipy but it complains that umfpack (or 
libsparesuite...) is missing when I run the scipy.test().

Maybe the scipy.test() should clearly split the errors in two groups : 
One group including "real" bugs and one reporting that libs a missing 
but that it is *not* a problem if you do not plan to use this part of scipy

I do agree that the doc should tell the user to use the .deb and I think 
all this stuff be just vanish when gfortran will be the default compiler.

To tell you the story : I have installed gcc-4.3 on my hardy to be able 
to use openmp with scipy.weave.inline (it works just fine for instance 
to implement a TOTAL idl like function. A bug in gcc-4.2 prevent this to 
work.).
I do hope that the next release of ubuntu will be based gcc-4.3 and on 
gfortran *and* that they will provide us this nice atlas packages.

I must admit that I'm a user working "only" on large arrays so it is a 
kind of a corner case ;)

Cheers,
Xavier
ps : I'm not sure but I don't think you can compile the kernel using icc 
;). the kernel is written in gcc qnd not in C :)


> Xavier Gnata wrote:
>   
>> Well from my point of view, it is much harder compile scipy than to 
>> compile gcc or a kernel because it looks like black magic :(
>>   
>>     
>
> No, it's not. Really. On a new debian/Ubuntu machine:
>
> sudo apt-get install g++ gcc g77 atlas3-base-dev python-dev
>
> And then, you can build numpy/scipy, and it will work 100 % of the time. 
> What is difficult is to build the dependencies, and people try to build 
> them by themselves, or in "weird" configurations.
>
> Try building the kernel with ICC, or try building gcc with a libc which 
> is not glibc. For that matter, try to build any software from scratch, 
> and you will see it is not easier.
>
> I think we should make it clearer on scipy install pages that you should 
> really use the packaged libraries when available. Before, major 
> distribution had broken blas/lapack, but those days are mostly over,
>
> cheers,
>
> David
> _______________________________________________
> SciPy-user mailing list
> SciPy-user@scipy.org
> http://projects.scipy.org/mailman/listinfo/scipy-user
>   



More information about the SciPy-user mailing list