[Numpy-discussion] Three bugs fixed
Michael Droettboom
mdroe at stsci.edu
Fri Dec 4 10:45:53 EST 2009
Unfortunately, I can't reproduce it on Solaris/SPARC (SVN r7878,
trunk). Could it be Linux/SPARC-specific? (We don't have a Linux/SPARC
machine lying around, I don't think).
Mike
Michael Droettboom wrote:
> Charles R Harris wrote:
>
>> On Thu, Dec 3, 2009 at 4:25 PM, Travis Oliphant
>> <oliphant at enthought.com <mailto:oliphant at enthought.com>> wrote:
>>
>>
>> I'm not sure which fix that was. The bus errors started when I
>> added the test for #1299. I also fixed the test with the reduceat
>> bug in that same checkin, but the SPARC build machines are
>> reporting a bus error in testing #1299.
>>
>> I'm not sure where the bus error is occurring in that code exactly
>> (i.e. the test wasn't being run before my fix so it could have
>> exposed a bus error earlier --- obviously I need to play better
>> with the build-bots than I am doing).
>>
>>
>> The error could certainly been lurking all these years, I don't think
>> reduceat gets used much.
>>
>>
>> A likely candidate though is the PyArray_Item_INCREF call (though
>> it uses the new copy-object-before-refcount-changes semantics).
>>
>>
>> Would that cause a bus error? That looks like an alignment issue.
>> There was another buffer alignment issue a while back that I fixed,
>> I'll try to track it down. Maybe we can get Michael Droettboom to help
>> here, he has access to SPARC.
>>
> Sure. I have some time to look into it this morning.
>
> Mike
>
>
--
Michael Droettboom
Science Software Branch
Operations and Engineering Division
Space Telescope Science Institute
Operated by AURA for NASA
More information about the NumPy-Discussion
mailing list