Testing: Address case when setting COMM serial and NUM_MPI_PROCS N for N > 1

[bartlettroscoe]

Description

As described in https://github.com/trilinos/Trilinos/pull/11322#issuecomment-1334063481, the updated behavior of TriBITS is to not add a test in an non-MPI SERIAL build when the test has an explicit NUM_MPI_PROCS N for N > 1. But test will never be added if COMM serial is passed into tribits_add[_advanced]_test() in these cases.

This Issue is to update tribits_add[_advanced]_test() to address the case where COMM serial and NUM_MPI_PROCS N for N > 1 are passed in.

Proposed solution

Many different proposed solutions are discussed in comments after https://github.com/trilinos/Trilinos/pull/11322#issuecomment-1333904958. What we arrived at is to implement the following:

Case 1: NUM_MPI_PROCS N for N > 1 but no explicit COMM ... argument passed in:

• MPI builds: Test will be added with N procs (which is the current TriBITS behavior)
• Serial builds: Test will be slipped (which is the current TriBITS behavior)

Case 2: NUM_MPI_PROCS N for N > 1 with explicit COMM serial mpi passed in:

• MPI builds: Two tests will be added: one with N procs and one with 1 proc (currently just the N proc test is added)
• Serial builds: One test with 1 proc will be added (currently no test is added)

Case 3: NUM_MPI_PROCS N for N > 1 with explicit COMM serial (but not mpi) passed in:

• Configure error for MPI and Serial builds with error message stating that the test will never be added and provide suggestions for how to fix this to do something logical (currently the test is skipped with a message why printed to the script when tracing add_test)

[bartlettroscoe]

@rppawlo and @etphipp, can you please look above and add a thumbs up if those behaviors are consistent with what we have agreed to in https://github.com/trilinos/Trilinos/pull/11322#issuecomment-1334530762? If not, please comment on what should be fixed.

[bartlettroscoe]

@rppawlo and @etphipp, now the question is, how urgent is this compared to #367? But this is something I might get help with from Kitware.

[etphipp]

I can't speak to the urgency #367, but I think this is somewhat urgent just because we have a bunch of tests that aren't being run in serial builds.