[SciPy-Dev] ANN: SciPy 0.9.0 beta 1

[Christoph Gohlke]

On 12/13/2010 1:43 PM, Pauli Virtanen wrote:
> On Mon, 13 Dec 2010 10:18:13 -0800, Christoph Gohlke wrote:
> [clip]
>> I ran into the following issues with scipy 0.9.0b1 on Windows when
>> building with msvc 9 and intel 11.1 compilers. Numpy is version 1.5.1,
>> linked against MKL.
>>
>> 1) after patching some source files and setup scripts (see msvc9.diff
>> attachment), "python setup.py bdist_winist" succeeds on 32- and 64-bit
>> Python 2.6, 2.7, and 3.1. The Visual C compiler reproducibly crashes
>> while compiling scipy\special\cephes\ndtr.c. Running the build command
>> again (without cleaning the build directory) seems to work.
>
> Some of that patch should end up in the tree? I'm not sure which parts,
> though.

The patch can not be applied as is. It is a quick hack to get scipy 
compiling with msvc9/ifort again and likely breaks things in other build 
environments. I can open tickets later.

>
> [clip]
>> 2) the build process fails on Python 3.2b1 with the following errors:
>>
>> scipy\sparse\sparsetools\csr_wrap.cxx(2451) : error C3861:
>> 'PyCObject_Import': identifier not found
>> scipy\sparse\sparsetools\csr_wrap.cxx(2521) : error C3861:
>> 'PyCObject_FromVoidPtr': identifier not found
>> scipy\sparse\sparsetools\csr_wrap.cxx(2544) : error C3861:
>> 'PyCObject_AsVoidPtr': identifier not found
>
> It seems that the version of SWIG used there is not compatible with
> Python 3.2.

Yes, see SWIG bug #3057804 "[Python] wrapper does not compile with 
python 3.2"
<http://sourceforge.net/tracker/?func=detail&aid=3057804&group_id=1645&atid=101645>

>
>> 3) 32-bit Python 2.6 is the only platform where scipy.test() finishes;
>> albeit with 32 failures. Besides the Correlate complex192 and TestODR
>> errors (http://projects.scipy.org/scipy/ticket/678) I get
> [clip]
>> FAIL: Real-valued Bessel domains
> [clip]
>> AssertionError: (nan, nan, nan, nan)
>
> This indicates some issue with AMOS and ifort. What do you get from
>
> import warnings, scipy.special as sc
> warnings.simplefilter('always', sc.SpecialFunctionWarning)
> sc.errprint(1)

0

> sc.airye(-1)

__main__:1: SpecialFunctionWarning: airye:: unknown error
__main__:1: SpecialFunctionWarning: airye:: unknown error
(nan, nan, 0.10399738949694459, 0.59237562642279229)

>
> It's possible that the test is too strict.
>
>> 4) scipy.test() on Python 2.7 and 3.1 crashes as reported in ticket
>> #1210<http://projects.scipy.org/scipy/ticket/1210>
>
> This smells like something triggering a DeprecationWarning when GIL is
> released. If so, it should be compiler independent and reproducible on
> wine+mingw. Or, some PyCObject/PyCapsule issue.
>
>> 5) On 64 bit platforms, scipy.test() quits at the following test
>>
>> test_interpnd.TestCloughTocher2DInterpolator.test_dense ... QH6084 qhull
>> internal error (qh_meminit): sizeof(void*) 8>  sizeof(ptr_intT) 4.
>> Change ptr_intT in mem.h to 'long long'
>
> Should be fixed in r7013

Thanks.