Test MueLu_UnitTestsEpetra_MPI_1 failing on ATDM builds on shiller

Created by: bartlettroscoe

CC: @trilinos/muelu

Next Action Status

This was resolved by fixing the tests

Description

The test MueLu_UnitTestsEpetra_MPI_1 is showing several failed ATDM build of Trilinos as shown at:

https://testing.sandia.gov/cdash/queryTests.php?project=Trilinos&date=2018-02-22&filtercount=2&showfilters=1&filtercombine=and&field1=testname&compare1=65&value1=MueLu_UnitTestsEpetra&field2=status&compare2=61&value2=Failed

which currently shows:

Site	Build Name	Test Name	Status	Time	Details	Build Time
hansen/shiller	Trilinos-atdm-hansen-shiller-gnu-debug-serial	MueLu_UnitTestsEpetra_MPI_1	Failed	15.75	Completed (Failed)	2018-02-22T08:06:46 UTC
ascic144	Linux-intel-17.0.1-MPI_Release_intel_17.0.1_openmpi_1.8.7_DEV	MueLu_UnitTestsEpetra_MPI_1	Failed	6.22	Completed (Failed)	2018-02-22T08:42:53 UTC
hansen/shiller	Trilinos-atdm-hansen-shiller-gnu-opt-serial	MueLu_UnitTestsEpetra_MPI_1	Failed	5.91	Completed (Failed)	2018-02-22T09:41:15 UTC
hansen/shiller	Trilinos-atdm-hansen-shiller-intel-debug-serial	MueLu_UnitTestsEpetra_MPI_1	Failed	21.81	Completed (Failed)	2018-02-22T11:36:23 UTC

Looking at these failing tests it looks like they are due to a bad floating-point equality comparison like shown at:

https://testing.sandia.gov/cdash/testDetails.php?test=44832397&build=3400210

which shows:

 diagVec->norm1() = 2.50000000000000071e+01 == diagVec->getGlobalLength() = 25 : FAILED ==> /home/jenkins/hansen/workspace/Trilinos-atdm-hansen-shiller-gnu-debug-serial/SRC_AND_BUILD/Trilinos/packages/muelu/test/unit_tests/StructuredAggregationFactory.cpp:339

In general, you can't do an equality comparison of computed floating-point numbers. You can't expect every digit to match. Instead please use TEST_FLOATING_EQUALITY.

Steps to Reproduce

Instructions to reproduce this on hansen or shiller can be found at:

https://github.com/trilinos/Trilinos/blob/develop/cmake/std/atdm/README.md