[Numpy-discussion] Bus error for Debian / SPARC on current trunk

[Matthew Brett]

Hi,

On Sun, Mar 4, 2012 at 8:32 PM, Mark Wiebe <mwwiebe at gmail.com> wrote:
> On Sun, Mar 4, 2012 at 10:08 PM, Matthew Brett <matthew.brett at gmail.com>
> wrote:
>>
>> Hi,
>>
>> On Sun, Mar 4, 2012 at 11:41 AM, Mark Wiebe <mwwiebe at gmail.com> wrote:
>> > On Sun, Mar 4, 2012 at 11:27 AM, Matthew Brett <matthew.brett at gmail.com>
>> > wrote:
>> >>
>> >> Hi,
>> >>
>> >> On Sat, Mar 3, 2012 at 12:07 AM, Matthew Brett
>> >> <matthew.brett at gmail.com>
>> >> wrote:
>> >> > Hi,
>> >> >
>> >> > On Fri, Mar 2, 2012 at 9:05 PM, Charles R Harris
>> >> > <charlesr.harris at gmail.com> wrote:
>> >> >>
>> >> >>
>> >> >> On Fri, Mar 2, 2012 at 4:36 PM, Matthew Brett
>> >> >> <matthew.brett at gmail.com>
>> >> >> wrote:
>> >> >>>
>> >> >>> Hi,
>> >> >>>
>> >> >>> Sorry that this report is not complete, I don't have full access to
>> >> >>> this box but, on a Debian squeeze machine running linux
>> >> >>> 2.6.32-5-sparc64-smp:
>> >> >>>
>> >> >>> nosetests
>> >> >>>
>> >> >>>
>> >> >>> ~/usr/local/lib/python2.6/site-packages/numpy/lib/tests/test_io.py:TestFromTxt.test_user_missing_values
>> >> >>>
>> >> >>> test_user_missing_values (test_io.TestFromTxt) ... Bus error
>> >> >>>
>> >> >>> This on current master : 1.7.0.dev-b9872b4
>> >> >>>
>> >> >>
>> >> >> Hmm, some tests might have been recently enabled. Any chance of
>> >> >> doing a
>> >> >> bisection?
>> >>
>> >> Struggling because compilation is very slow and there are lots of
>> >> untestable commits.  df907e6 is the first known bad.  Here's the
>> >> output from a log:
>> >>
>> >> * df907e6 - (HEAD, refs/bisect/bad) BLD: Failure in single file build
>> >> mode because of a static function in two separate files (6 months ago)
>> >> [Mark Wiebe]
>> >> * 01b200b - (refs/bisect/skip-01b200b10149312f51234448e44b230b1b548046)
>> >> BUG: nditer: The nditer was reusing the reduce loop inappropriately
>> >> (#1938) (6 months ago) [Mark Wiebe]
>> >> * f45fd67 - (refs/bisect/skip-f45fd67fe8eefc8fd2e4b914ab4e376ab5226887)
>> >> DOC: Small tweak to release notes (6 months ago) [Mark Wiebe]
>> >> * 73be11d - (refs/bisect/skip-73be11db794d115a7d9bd2e822c0d8008bc14a28)
>> >> BUG: Some bugs in squeeze and concatenate found by testing SciPy (6
>> >> months ago) [Mark Wiebe]
>> >> * c873295 - (refs/bisect/skip-c8732958c8e07f2306029dfde2178faf9c01d049)
>> >> TST: missingdata: Finish up NA mask tests for np.std and np.var (6
>> >> months ago) [Mark Wiebe]
>> >> * e15712c - (refs/bisect/skip-e15712cf5df41806980f040606744040a433b331)
>> >> BUG: nditer: NA masks in arrays with leading 1 dimensions had an issue
>> >> (6 months ago) [Mark Wiebe]
>> >> * ded81ae - (refs/bisect/skip-ded81ae7d529ac0fba641b7e5e3ecf52e120700f)
>> >> ENH: missingdata: Implement tests for np.std, add skipna= and
>> >> keepdims= parameters to more functions (6 months ago) [Mark Wiebe]
>> >> * a112fc4 - (refs/bisect/skip-a112fc4a6b28fbb85e1b0c6d423095d13cf7b226)
>> >> ENH: missingdata: Implement skipna= support for np.std and np.var (6
>> >> months ago) [Mark Wiebe]
>> >> * 0fa4f22 - (refs/bisect/skip-0fa4f22fec4b19e2a8c1d93e5a1f955167c9addd)
>> >> ENH: missingdata: Support 'skipna=' parameter in np.mean (6 months
>> >> ago) [Mark Wiebe]
>> >> * bfda229 - (refs/bisect/skip-bfda229ec93d37b1ee2cdd8b9443ec4e34536bbf)
>> >> ENH: missingdata: Create count_reduce_items function (6 months ago)
>> >> [Mark Wiebe]
>> >> * d9b3f90 - (refs/bisect/skip-d9b3f90de3213ece9a78b77088fdec17910e81d9)
>> >> ENH: missingdata: Move the Reduce boilerplate into a function
>> >> PyArray_ReduceWrapper (6 months ago) [Mark Wiebe]
>> >> * 67ece6b - (refs/bisect/skip-67ece6bdd2b35d011893e78154dbff6ab51c7d35)
>> >> ENH: missingdata: Finish count_nonzero as a full-fledged reduction
>> >> operation (6 months ago) [Mark Wiebe]
>> >> * 6bfd819 - (refs/bisect/skip-6bfd819a0897caf6e6db244930c40ed0d17b9e62)
>> >> ENH: missingdata: Towards making count_nonzero a full-featured
>> >> reduction operation (6 months ago) [Mark Wiebe]
>> >> * a1faa1b - (refs/bisect/skip-a1faa1b6883c47333508a0476c1304b0a8a3f64e)
>> >> ENH: missingdata: Move some of the refactored reduction code into the
>> >> API (6 months ago) [Mark Wiebe]
>> >> * f597374 - (refs/bisect/skip-f597374edc298810083799e8539c99fc0a93b319)
>> >> ENH: missingdata: Change default to create NA-mask when NAs are in
>> >> lists (6 months ago) [Mark Wiebe]
>> >> * 965e4cf - (refs/bisect/skip-965e4cff5c4c50e8ff051a3363adc6cf6aa640cd)
>> >> ENH: missingdata: trying some more functions to see how they treat NAs
>> >> (6 months ago) [Mark Wiebe]
>> >> * b1cb211 - (refs/bisect/skip-b1cb211d159c617ee4ebd16266d6f1042417ef75)
>> >> ENH: missingdata: Add nastr= parameter to np.set_printoptions() (6
>> >> months ago) [Mark Wiebe]
>> >> * ba4d116 - (refs/bisect/skip-ba4d1161fe4943cb720f35c0abfd0581628255d6)
>> >> BUG: missingdata: Fix mask usage in PyArray_TakeFrom, add tests for it
>> >> (6 months ago) [Mark Wiebe]
>> >> * a3a0ee8 - (refs/bisect/skip-a3a0ee8c72fdd55ffacb96bbb1fa9c3569cfb3e9)
>> >> BUG: missingdata: The ndmin parameter to np.array wasn't respecting NA
>> >> masks (6 months ago) [Mark Wiebe]
>> >> * 9194b3a - (refs/bisect/skip-9194b3af704df71aa9b1ff2f53f169848d0f9dc7)
>> >> ENH: missingdata: Rewrite PyArray_Concatenate to work with NA masks (6
>> >> months ago) [Mark Wiebe]
>> >> * 99a21ef - (refs/bisect/good-99a21efff4b1f2292dc370c7c9c7c58f10385f2a)
>> >> ENH: missingdata: Add NA support to np.diagonal, change np.diagonal to
>> >> always return a view (6 months ago) [Mark Wiebe]
>> >>
>> >> So - the problem arises somewhere between 99a21ef (good) and  df907e6
>> >> (bad)
>> >>
>> >> There seems to be a compilation error for the skipped commits - here's
>> >> the one I tested, 9194b3a:
>> >
>> >
>> > If you enable separate compilation by setting an environment variable,
>> > these
>> > commits should build as well.
>> >
>> > $ export ENABLE_SEPARATE_COMPILATION=1
>> > <proceed with builds>
>>
>> I might be doing something wrong but:
>
>
> I made a mistake, sorry! I even copy/pasted the variable name to make sure I
> didn't misspell it, but I didn't notice that setup.py uses a different but
> nearly identical name internally as the environment variable. That should
> have been
>
> $ export NPY_SEPARATE_COMPILATION=1

Thanks, that did it:

9194b3af704df71aa9b1ff2f53f169848d0f9dc7 is the first bad commit

Let me know if I can debug further,

See you,

Matthew