[Scipy-tickets] [SciPy] #1737: OS X Mountain Lion test failures

SciPy Trac scipy-tickets@scipy....
Fri Sep 28 05:53:00 CDT 2012


#1737: OS X Mountain Lion test failures
--------------------------+-------------------------------------------------
 Reporter:  aleach        |       Owner:  cdavid     
     Type:  defect        |      Status:  new        
 Priority:  normal        |   Milestone:  Unscheduled
Component:  Build issues  |     Version:  0.10.0     
 Keywords:  OSX, ATLAS    |  
--------------------------+-------------------------------------------------
Changes (by aleach):

 * cc: beamesleach@… (added)


Comment:

 Thanks for the response and info. Just noticed one Overflow warning was
 generated from the quadmath module, but I don't have GNU's libquadmath. I
 guess that would account for that 1 {{RuntimeWarning}}, at least. I saw
 later that a lot of the other warnings were generated from numpy, when I
 ran {{ numpy.seterr(all="raise")}} before running the scipy tests.

 I did see the r.research.att.com link and figured I'd be pointed in that
 direction if I asked, but I was quite keen for a Mountain Lion build, so
 thought I'd give building it a shot. Patching it was necessary in just a
 few places. Apple's GCC raised errors about missing "%s" string formatting
 in various fprintf and printf functions. e.g. (1 necessary patch in
 gfortranspec.c):

 -      fprintf (stderr, _("Driving:"));
 +      fprintf (stderr, "%s", _("Driving:"));


 Otherwise, I left Apple's own patches in, which are wrapped in quite
 obvious comments, containing {{APPLE}} or {{LLVM}}. Again, very minor
 changes from the GNU release. I couldn't say whether updating the fortran
 bits to 4.2.4 was necessary to get my test results, but it wasn't working
 perfectly before; the changes between versions 4.2.1-4 looked like very
 minor fixes. Fixes that Apple won't implement though, because of the
 GPLv2->3 license change between versions. All the patches I made are up on
 [https://github.com/alexleach/gfortran-mlion/blob/master/patch github].

 I did have a look at the zfft.c code which was causing the TestDoubleIFFT
 error. It looks like there's some pointer incrementing going on. I thought
 that might disagree with Apple's enforced PIE binaries, but I couldn't say
 for sure; I'm no expert.

 Yea, I should probably start a blog or something to document my builds. I
 know here isn't exactly the best place to write this, but... I've shared
 the gfortran build-scripts I used at [http://github.com/alexleach
 /gfortran-mlion/ github], and have a
 [http://bioltfws1.york.ac.uk/~albl500/Downloads/gfortran-mlion.tar.gz
 direct link] to the gfortran binaries and libraries I made, which aren't
 exactly easy to find.

 Atlas was probably the longest build step; played quite a bit with the
 different compilers and options, clang, gcc, openmp, and netlib lapack.
 I've still got about 6 builds - the clang ones took hours - but have ended
 up using fairly standard gnu compilers and options: {{CFLAGS="-fPIC -O2
 -g"}}, and {{FFLAGS="-ff2c -fno-second-underscore -g -O2"}}. With scipy, I
 also used FFLAGS="-ff2c -fno-second-underscore".

-- 
Ticket URL: <http://projects.scipy.org/scipy/ticket/1737#comment:2>
SciPy <http://www.scipy.org>
SciPy is open-source software for mathematics, science, and engineering.


More information about the Scipy-tickets mailing list