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

SciPy Trac scipy-tickets@scipy....
Tue Sep 25 07:18:14 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    |  
--------------------------+-------------------------------------------------
 I've been trying for a week or so to get a fully working scipy
 installation on Mountain Lion. It's been a long process! I've compiled my
 own (x86_64/i386) gfortran, using the source code from
 opensource.apple.com, patched with GNU's gcc-4.2.4 libgfortran and
 gcc/fortran updates.

 With this, I compiled a full x86_64 ATLAS (3.10) w/ Netlib lapack, due to
 the Accelerate framework's rounding issues that cause test failures,
 reported on Lion also. As others have reported, using llvm-gcc causes an
 EXC_BAD_ACCESS in TestDoubleIFFT. Recompiling scipy with the
 NSZombieEnabled=YES environment variable set, the error seems to come from
 line 60 in fftpack/src/zfft.c. A compiler flag might fix that issue, but I
 don't know what to look for in gcc's man page, to solve it.

 Compiling with clang/clang++, the test suite finishes, but I still get
 several failed tests (18 errors and 1 failure), which are mostly similar.
 I attach a report of the test failures, which were run with the command:-
     gdb --args /usr/local/bin/python -c 'import scipy; scipy.test("full")'

 Any help fixing those failed test cases would be greatly appreciated!

 The build command I used, and site.cfg contents are below...

 Some questions: What overflow and floating point precision semantics does
 scipy expect? Does scipy expect precise IEEE semantics, or a "twos-
 complement representation" of overflow, as provided by the '-fwrapv' flag?

 Also, should scipy work with threaded builds of Lapack, Blas and ATLAS?
 I'm building with the serial version atm, as haven't had much joy with the
 threaded builds, but would like to use the parallel dylibs, if possible..
 Should it be?

 CC=/usr/bin/clang \
     LD=/usr/bin/ld \
     CXX=/usr/bin/clang++ \
     FFLAGS="-g -O0 -arch x86_64" \
     LDFLAGS="-arch x86_64 -undefined dynamic_lookup
 -L/usr/local/atlas/lib" \
     CFLAGS="-arch x86_64 -O0 -g" \
     CPPFLAGS="-I/usr/local/atlas/include" \
     F77BLAS="f77blas" \
     BLAS="f77blas" CBLAS="cblas" ATLAS="satlas" \
     CLAPACK="lapack" FLAPACK="lapack" \
     /usr/local/bin/python setup.py config --fcompiler=gnu95  build_ext
 build_clib build install --user


 $ grep -v '^\(\s*\|#.*\)$' site.cfg
 [DEFAULT]
 include_dirs = /usr/local/atlas/include:/usr/local/include
 library_dirs = /usr/local/atlas/lib:/usr/local/lib
 [blas_opt]
 libraries    = f77blas, cblas, satlas
 [lapack_opt]
 libraries    = lapack, f77blas, cblas, satlas
 [ptatlas]
 libraries    = satlas
 [atlas]
 atlas_libs   = f77blas, cblas, satlas
 lapack_libs  = lapack
 [lapack_atlas]
 libraries    = lapack

 If I can provide any further info, please let me know.
 Kind regards,
 Alex
 p.s. Using scipy from git...

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


More information about the Scipy-tickets mailing list