[SciPy-user] Bus error on Intel mac pro w/Leopard, gcc 4.2, numpy from svn head

Anand Patil anand.prabhakar.patil@gmail....
Tue Jun 17 06:24:49 CDT 2008


On 17 Jun 2008, at 09:34, Anand Patil wrote:

> On 16 Jun 2008, at 20:10, Robert Kern wrote:
>
>> On Mon, Jun 16, 2008 at 12:06, Anand Patil
>> <anand.prabhakar.patil@gmail.com> wrote:
>>>
>>> On 16 Jun 2008, at 17:49, Robert Kern wrote:
>>>
>>>> On Mon, Jun 16, 2008 at 10:58, Anand Patil
>>>>
>>>> Does your program crash using the www.python.org binary built with
>>>> gcc 4.0.1?
>>>
>>> Actually, I wasn't able to get my Python environment set up with the
>>> python.org binary . I need OpenMP, which means I need gcc 4.2 ...  
>>> but
>>> the python.org binary has no-cpp-precomp and Wno-long-double baked  
>>> in.
>>
>> Good enough reason.
>
> Can I be pretty sure that the problem is caused by a reference  
> cycle, and how can I:
> - Reproduce the problem with a simple program, assuming it were in  
> numpy itself;
> - Look for problematic reference cycles in my program?



I found the bit of PyMC that was making views of views & changed it,  
and the problem seems to be gone (now that I have announced that I'm  
sure it'll come back this afternoon)... so this isn't a personal  
emergency anymore. So thanks for the tip about views.


I tried reproducing the bug as follows:

In [1]: from numpy import *

In [2]: A = zeros(10)

In [3]: for i in xrange(100000000):
    ...:     B = A.view(ndarray)

but no errors happened. Any other tests I can try? Would it be  
productive to submit the crash report as a bug, or is it too vague?


Anand


More information about the SciPy-user mailing list