[Insight-developers] Dashboard Clean Up : itk2.jouy.inra.fr SunOS-g++-i386 : KdTree test : Math Precision ?

[Luis Ibanez]

Hi Gaetan,


The build:

    itk2.jouy.inra.fr    SunOS-g++-i386


has 14 failing tests:

http://www.cdash.org/CDash/viewTest.php?onlyfailed&buildid=524655

They all seem to be related to the KdTree and
KdTree generator.


Looking at their output,
the errors seems to be related to
mathematical precision:

For example:
http://www.cdash.org/CDash/testDetails.php?test=42464662&build=524655

says:

Problem found
Query point [-0.977672, -1.03235]
Reported closest point [-0.902621, -1.07898] distance 0.0883604
Actual   closest point [-0.902621, -1.07898] distance 0.0883604

Test FAILED.
Problem found
Query point [-1.54991, 0.301798]
Reported closest point [-1.66722, 0.343659] distance 0.124557
Actual   closest point [-1.66722, 0.343659] distance 0.124557

Test FAILED.
Problem found
Query point [-0.553142, 1.44982]
Reported closest point [-0.739648, 1.45294] distance 0.186533
Actual   closest point [-0.739648, 1.45294] distance 0.186533

Test FAILED.
3 out of 100 points failed to find the correct closest point.


-------------


Options that come to mind:


    Is there a configuration flag that could
    improve the precision in this machine ?

    or... do you think that this indicates a
    real problem with the KdTree code ?

    or.... should we relax the tolerance in
    the comparisons of the test  ?



   Thanks for any hints,


         Luis