[SciPy-dev] Another segfault

[Nils Wagner]

Ed Schofield wrote:
> Robert Cimrman wrote:
>   
>> Ed Schofield wrote:
>>   
>>     
>>> Nils Wagner wrote:
>>>
>>>     
>>>       
>>>> Thank you for your advice. This note should be available on the Wiki as a
>>>> warning for other users. I have spend more than one day on this - pure 
>>>> waste of time.
>>>> Anyway the segfault w.r.t. sparse_test.py persists independent from 
>>>> g77/gfortran.
>>>> Can you reproduce this segfault ?
>>>>  
>>>>       
>>>>         
>>> I can't.  I'm using g77 v3.4.6 with a default distutils config.  Does it
>>> work if you install scipy from scratch this way?  It may be a bug,
>>> rather than a configuration error, but we need to reproduce it to fix it.
>>>     
>>>       
>> The sparse_test.py works ok for me, but if one runs sparse.py module as 
>> a script, a segfault occurs in solve() when using LU. (UMFPACK works ok, 
>> if present, that is.) Can you reproduce this, Ed?
>>   
>>     
> No, I can't.  (I only have a 32-bit machine.)  I've added a unit test
> that runs Nils' code.  If others start reporting segfaults we're in
> business ;)
>
> I've fixed some bugs in the handling of data type arguments and added a
> small unit test for tocsc() conversion with complex data types.  This is
> unlikely to fix the segfault problem though ...
>
> -- Ed
>
> _______________________________________________
> Scipy-dev mailing list
> Scipy-dev at scipy.net
> http://www.scipy.net/mailman/listinfo/scipy-dev
>   
Hi Ed,

Thanks for adding the unit test. I have installed numpy/scipy from 
scratch using a 32 and a 64 bit
machine. The segfault is definitely a 64 bit problem.

Nils

On a 32 bit bit machine
scipy.test(1,10)
Ran 1111 tests in 7.764s

OK
<unittest.TextTestRunner object at 0xb209016c>
 >>> scipy.__version__
'0.4.7.1631'
 >>>                                       

On a 64 bit machine

scipy.test(1,10)
check_rmatvec (scipy.sparse.tests.test_sparse.test_csc) ... ok
check_setelement (scipy.sparse.tests.test_sparse.test_csc) ... ok
Test whether the lu_solve command segfaults, as reported by NilsUse 
minimum degree ordering on A'+A.
Segmentation fault