[aspect-devel] aspect ctest fails?

[Wiederspahn, Mark]

This seems reasonable – one of the issues I was worried about was that the number of iterations in at least some cases was different, by at least 2. Some of the differences seemed to be related to the number of digits printed, which I had hoped than numdiff would deal with (much better than diff, but not perfect).

I haven’t seen any place where the “reference” system is described. That might be worth stating explicitly, since it should (in principle) be possible to install such a system locally and use that to achieve a perfect ctest score, leading to high confidence that no installation mishaps occurred. One could then walk out small deltas (different compiler, different debug, etc, etc, etc) from that known place.

I tried ctest on the Ubuntu virtualbox (which is a different aspect version and OS than the fedora run I posted) and that failed utterly. I’m not sure exactly why, and maybe this particular test wouldn’t be expected to succeed by those who know its internal details. I had high hopes that it could be used as a comparison benchmark (albeit glacially slow to run) to anchor expectations of the fedora results.

Thanks to all for the replies.

On 6/8/17, 4:56 PM, "Aspect-devel on behalf of Wolfgang Bangerth" <aspect-devel-bounces at geodynamics.org on behalf of bangerth at tamu.edu> wrote:

    
    Mark,
    I think the discussion already answered most questions, but here is a 
    summary I'm proposing for the manual:
    
    https://github.com/geodynamics/aspect/pull/1811
    
    Best
      W.
    
    -- 
    ------------------------------------------------------------------------
    Wolfgang Bangerth          email:                 bangerth at colostate.edu
                                www: http://www.math.colostate.edu/~bangerth/
    _______________________________________________
    Aspect-devel mailing list
    Aspect-devel at geodynamics.org
    http://lists.geodynamics.org/cgi-bin/mailman/listinfo/aspect-devel