[Numpy-discussion] Lots of 32bits specific errors

[David Cournapeau]

On Wed, Jan 6, 2010 at 8:17 AM, Xavier Gnata <xavier.gnata at gmail.com> wrote:
>
>> On Wed, Jan 6, 2010 at 6:48 AM, Xavier Gnata <xavier.gnata at gmail.com> wrote:
>>
>>> Hi,
>>>
>>> I have compiled numpy 1.5.0.dev8039 both on a 32 and a 64bits ubuntu
>>> machine.
>>>
>>> On the 64bits one, everything is fine:
>>> numpy.test get a perfect score:
>>> <nose.result.TextTestResult run=2504 errors=0 failures=0>
>>>
>>> On the 32bits ubuntu, the story is not that nice:
>>> <nose.result.TextTestResult run=2504 errors=1 failures=16>
>>>
>> Almost all your errors are in linalg - most likely an atlas problem.
>> Atlas with sse2 is buggy on some Ubuntu versions. You should check
>> Ubuntu bug tracker to see if it affects the version you are using.
>>
>>
> Thanks, you got it! atlas sse2 in fully buggy in karmic (at least on
> 32bits machines).

That's just incompetence at that point. This bug is known for > one
year, and they still have not fixed it. The least they could do is
removing the package so that people has slower, but accurate version.

David