[NumPy-Tickets] [NumPy] #1632: test_structured_arrays_with_objects2 fails with too high refcounts due to gcc 4.5.x bug

[NumPy Trac]

#1632: test_structured_arrays_with_objects2 fails with too high refcounts due to
gcc 4.5.x bug
--------------------+-------------------------------------------------------
 Reporter:  nics    |       Owner:  somebody
     Type:  defect  |      Status:  new     
 Priority:  normal  |   Milestone:  1.5.1   
Component:  Other   |     Version:  1.5.0   
 Keywords:          |  
--------------------+-------------------------------------------------------

Old description:

> Hi everybody,
>
> this isn't really a numpy bug but numpy is affected by a gcc bug (that is
> UNCONFIRMED right now) I've found while debugging the following numpy
> testsuite failer:
>
> ======================================================================
> FAIL: Ticket #1299 second test
> ----------------------------------------------------------------------
> Traceback (most recent call last):
>   File
> "/pf/m/m222086/xas/solaris10/python2/python-2.7-ve0-gcc/lib/python2.7/sit
> e-packages/numpy/core/tests/test_regression.py", line 1255, in
> test_structured_a
> rrays_with_objects2
>     assert sys.getrefcount(strb) == numb
> AssertionError:
>     7 = <module 'sys' (built-in)>.getrefcount('aaaa')
>     array([[(0L, 'aaaa'), (0L, 'aaaa')]],
>           dtype=[('f0', '>i8'), ('f1', '|O4')]) = <module 'numpy' from
> '/pf/m/m2
> 22086/xas/solaris10/python2/python-2.7-ve0-gcc/lib/python2.7/site-
> packages/numpy
> /__init__.pyc'>.array([[(0,'aaaa'),(1,'bbbb')]], 'i8,O')
> >>  array([[(0L, 'aaaa'), (0L, 'aaaa')]],
>           dtype=[('f0', '>i8'), ('f1', '|O4')])[array([[(0L, 'aaaa'),
> (0L, 'aaaa
> ')]],
>           dtype=[('f0', '>i8'), ('f1', '|O4')]).nonzero()] = array([[(0L,
> 'aaaa'
> ), (0L, 'aaaa')]],
>           dtype=[('f0', '>i8'), ('f1', '|O4')]).ravel()[:1]
>     assert <module 'sys' (built-in)>.getrefcount('bbbb') == 7
>     assert <module 'sys' (built-in)>.getrefcount('aaaa') == 7 + 2
>
> You'll find the gcc bug here:
> http://gcc.gnu.org/bugzilla/show_bug.cgi?id=45967
>
> Maybe we just wait for what the gcc people say about this issue; in the
> meanwhile you can use the attached patch against numpy-1.5.0 that fixes
> that issue with a workaround. The workaround isn't very clear code (and
> not commented ;) and thus, it either shouldn't be applied to the main
> branch at all or at least it should be removed again as soon as the issue
> has been fixed in gcc (again: if it is really a gcc bug).
>
> Wishes
>
> Nicolai

New description:

 Hi everybody,

 this isn't really a numpy bug but numpy is affected by a gcc bug (that is
 UNCONFIRMED right now) I've found while debugging the following numpy
 testsuite failer:
 {{{
 ======================================================================
 FAIL: Ticket #1299 second test
 ----------------------------------------------------------------------
 Traceback (most recent call last):
   File
 "/pf/m/m222086/xas/solaris10/python2/python-2.7-ve0-gcc/lib/python2.7/sit
 e-packages/numpy/core/tests/test_regression.py", line 1255, in
 test_structured_a
 rrays_with_objects2
     assert sys.getrefcount(strb) == numb
 AssertionError:
     7 = <module 'sys' (built-in)>.getrefcount('aaaa')
     array([[(0L, 'aaaa'), (0L, 'aaaa')]],
           dtype=[('f0', '>i8'), ('f1', '|O4')]) = <module 'numpy' from
 '/pf/m/m2
 22086/xas/solaris10/python2/python-2.7-ve0-gcc/lib/python2.7/site-
 packages/numpy
 /__init__.pyc'>.array([[(0,'aaaa'),(1,'bbbb')]], 'i8,O')
 >>  array([[(0L, 'aaaa'), (0L, 'aaaa')]],
           dtype=[('f0', '>i8'), ('f1', '|O4')])[array([[(0L, 'aaaa'), (0L,
 'aaaa
 ')]],
           dtype=[('f0', '>i8'), ('f1', '|O4')]).nonzero()] = array([[(0L,
 'aaaa'
 ), (0L, 'aaaa')]],
           dtype=[('f0', '>i8'), ('f1', '|O4')]).ravel()[:1]
     assert <module 'sys' (built-in)>.getrefcount('bbbb') == 7
     assert <module 'sys' (built-in)>.getrefcount('aaaa') == 7 + 2
 }}}
 You'll find the gcc bug here:
 http://gcc.gnu.org/bugzilla/show_bug.cgi?id=45967

 Maybe we just wait for what the gcc people say about this issue; in the
 meanwhile you can use the attached patch against numpy-1.5.0 that fixes
 that issue with a workaround. The workaround isn't very clear code (and
 not commented ;) and thus, it either shouldn't be applied to the main
 branch at all or at least it should be removed again as soon as the issue
 has been fixed in gcc (again: if it is really a gcc bug).

 Wishes

 Nicolai

--

Comment(by pv):

 Thanks for the sleuthing!

 I think it might be best to not include workarounds for compiler bugs,
 especially ones that affect only non-X86 platforms. But I'm not 100% sure
 whether this is the correct choice, as PPC (which IIRC also requires
 pointer alignment) is semi-common thanks to Apple computers.

 We probably should nevertheless mention this GCC bug in the 1.5.1 release
 notes.

-- 
Ticket URL: <http://projects.scipy.org/numpy/ticket/1632#comment:1>
NumPy <http://projects.scipy.org/numpy>
My example project