[petsc-dev] Regression tests

[Matthew Knepley]

On Thu, Jul 26, 2012 at 2:31 PM, Jed Brown <jedbrown at mcs.anl.gov> wrote:

> On Thu, Jul 26, 2012 at 2:17 PM, Matthew Knepley <knepley at gmail.com>wrote:
>>
>> You can tell from the primary key what it is. If the primary key is
>> wrong, it does not help you in your scenario as it could be
>> misspelled, or intended to be put in another directory. This is dubious
>> at best.
>>
>
> I don't need the primary key to be "right" to have good diagnostics, I
> just need to know that the Test() declaration (which executes code at the
> time the test is created) is inconsistent. My error message tells you the
> line that is bad. This is important.
>
>
>>  We should have a discussion about whether we want to support this after
>> 3.3.
>>
>>
>
> What's the problem? Is builder.py having trouble figuring out how to link
> PETSc in general? Maybe it should use "make getlinklibs". ;-D
>
> Note that people occasionally ask for minimal builds for low-level
> functionality. Barry was even asking if it would be feasible to have a way
> to build only some KSPs, PCs, and matrix formats. I like that the option
> keeps us honest.
>
>
>>
>>
>  No no. Your original make log is in PETSC_ARCH/conf/make.log. It should
>> never be corrupted. You
>> remove the link at make.log.bkp.
>>
>
> Well, that's an interesting (and confusing) idea for how it *could* work
> some day, but it's not how it currently works.
>

How the fuck is that confusing? Is this just for the sake of argument. This
is how we DESIGNED configure to work
and treat logs. Were you absent that day? Do you have a note?

    Matt


>
>
>>
>>
>
>> Same way we do in the makefiles now. It has an attribute "complex",
>> "superlu", etc.
>>
>
> So what you're saying is that your system could be extended to do this?
> I'm not disagreeing, but I think my suggestion is more maintainable and
> extensible. As I said in my first reply, the differences are mostly
> cosmetic, but I think there is good reason to prefer the form I suggest.
> The important thing in either case is to have a database of tests so we can
> program behavior.
>
>
>>  Some have these.
>
>
> Held separately?
>
> regressionRequirements = {'src/vec/vec/examples/tests/ex31':
>  set(['Matlab'])
>                           }
>
>


-- 
What most experimenters take for granted before they begin their
experiments is infinitely more interesting than any results to which their
experiments lead.
-- Norbert Wiener
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/petsc-dev/attachments/20120726/2e88fa80/attachment.html>