[petsc-dev] testing: with parmetis, without 64-bit, without complex

[Barry Smith]

   Once the release is out the door (any moment now) the next big project is to redo the test infrastructure to make keeping clean accurate tests a simple reality. I am open to all kinds of proposals to do this properly (except any proposal that reuses the current approach :-)).

    Barry



On May 13, 2013, at 1:08 PM, Jed Brown <jedbrown at mcs.anl.gov> wrote:

> Matthew Knepley <knepley at gmail.com> writes:
> 
>>  a) Switch the nightlies from make to Python
> 
> Are you talking about compilation or running tests?  Compilation should
> be irrelevant because all systems should produce the same result.  I
> think testing is higher priority.
> 
>>  b) Change the categories to tags
> 
> Yeah, just can't do this within the current make-based running system.
> 
>>  c) Simply parse parts of the output for numerical comparison
> 
> It's useful to compare everything because otherwise, new clutter can
> show up unintentionally and won't be caught by testing.