[SciPy-dev] A stray pointer somwehere...

[Fernando Perez]

Christopher Hanley wrote:
> 
> Travis Oliphant wrote:
> 
> 
>>Can anybody else reproduce this or is it just on 64-bit boxes.  I can't 
>>get the problem to show up on 32-bit box.
>>
>>I suspect a reference count issue or an unitialized PyObject variable.   
>>It might help if where the segfault is occurring is tracked down (since 
>>you are running test(10,10) you should be able to see which tests are 
>>failing.  Is the segfault happening in the same place all the time?
>>
>>-Travis
>>
> 
> 
> I can reproduce this error on my 32-bit Redhat box.  It occurs in the 17 
> th iteration of scipy.test(10,10) with the failure occurring at:

Actually in my case it's different: all the tests pass _always_, it segfaults 
only when I exit the python shell (tested both with ipython and with the 
default interpreter):

... run several times test(10,10)

check_nd (scipy.base.index_tricks.test_index_tricks.test_grid) ... ok
check_1 (scipy.distutils.misc_util.test_misc_util.test_appendpath) ... ok
check_2 (scipy.distutils.misc_util.test_misc_util.test_appendpath) ... ok
check_3 (scipy.distutils.misc_util.test_misc_util.test_appendpath) ... ok

----------------------------------------------------------------------
Ran 140 tests in 0.291s

OK
Out[6]: <unittest.TextTestRunner object at 0x20000000040f7310>

In [7]:  ### EXIT HERE via Ctrl-D

Segmentation fault


So the tests actually pass, but they corrupt something inside the python 
interpreter's state, so it segfaults on exit, while cleaning up.

Cheers,

f