[petsc-dev] elemental stuff still broken

[Jack Poulson]

On 12/20/2013 05:30 PM, Satish Balay wrote:
> On Fri, 20 Dec 2013, Satish Balay wrote:
> 
>> On Fri, 20 Dec 2013, Jed Brown wrote:
>>
>>> Barry Smith <bsmith at mcs.anl.gov> writes:
>>>
>>>> On Dec 20, 2013, at 1:45 PM, Jed Brown <jedbrown at mcs.anl.gov> wrote:
>>>>
>>>>> Barry Smith <bsmith at mcs.anl.gov> writes:
>>>>>
>>>>>>  Since Jack has fixed these issues why is this still broken in
>>>>>>  nightly tests of next?
>>>>>
>>>>> He fixed the issues in his repository, but we never told PETSc to use
>>>>> his new version.
>>>>
>>>>    Should we be using his repository? Should we be figuring out better
>>>>    ways of handling updates to other software from our stuff? It is
>>>>    not productive to have this kind of stuff happen, how can we
>>>>    prevent it in the future?
>>>
>>> When he updates, we should be making a one-line commit in
>>> config/PETSc/packages/elemental.py to use his new version.  It's not
>>> hard, but someone has to do it.  I'm about to step onto a plane, but I
>>> can do it when I land.
>>
>> Sorry didn't update this before. Will test and commit the fix.
> 
> I tried using latest elemental commit - and got errors. Then I noticed
> the fix was the following - so retried with this snapshot - and I still get build errors.
> 
> https://github.com/elemental/Elemental/commit/08daf5a157bd3addc8b1a6649f99e2f2e65ec03d
> 
> Attaching the elemental build log. This build is with clang:
> 
> clang version 3.3 (tags/RELEASE_33/final)
> Target: x86_64-unknown-linux-gnu
> Thread model: posix
> 
> 
> Satish
> 
>>
>> BTW: this is in jed/elemental-int64 branch [which hasn't been merged
>> to master yet]
>>
>> And I'll also remove all self.download usage when self.gitcommit [as
>> this is inconsistant code]
>>
>> [And I need to complete balay/giturl-fixes so that tarball urls are
>> automatically constructed from giturls - for usage without git]
>>
>> Satish
>>
>>>
>>> I don't want to "automatically" use the most recent version because then
>>> PETSc builds will break when he uses a new language feature or changes
>>> an interface, and we won't be able to reliably rewind to an old state.
>>>
>>
>>

I'm on a plane that's about to leave, but I will look into this when I
land. FYI, I just committed support for high-performance pseudospectrum
computation into Elemental; that might be a very useful addition to
PETSc. I will try to extend the current PETSc interface to add support
for Elemental's Bunch-Kaufman, SVD, Hermitian EVD, Schur decomposition,
and the new pseudospectrum support.

Jack