some ctest failure with MPIEXEC_MAX_NUMPROCS>9

[qbisi]

Describe the bug
some ctest failure with MPIEXEC_MAX_NUMPROCS>9

To Reproduce
these tests will fail running ctest with MPIEXEC_MAX_NUMPROCS>9

- Engine.BP.BPJoinedArray.MultiBlock.BP4.MPI
- Engine.BP.BPJoinedArray.MultiBlock.BP5.MPI
- Bindings.Fortran.BPWriteReadHeatMap6D.MPI

Possible Reason
insufficiently robust data generation and comparison for larger MPI sizes

see testing/adios2/engine/bp/TestBPJoinedArray.cpp

  mytable[row * Ncols + col] =
      static_cast<double>((step + 1) * 1.0 + rank * 0.1 + block * 0.01 +
                          row * 0.001 + col * 0.0001);

Solution
either limit MPIEXEC_MAX_NUMPROCS or modify data generation.

[eisenhauer]

Interesting... I'm running with MPIEXEC_MAX_NUMPROCS=12 and see no failures with these tests. It's true that the code line you cite doesn't produce unique table entries for the joined array if there are more than 9 ranks, but that shouldn't produce a test failure per se. Can you include output or other information?

[qbisi]

See NixOS/nixpkgs#466952

[eisenhauer]

Hmm. Not actual test output anywhere I can see there, just notes on what tests failed? Two sets where MultiBlock and HeatMap6D failed, then one set where more failed, including ZeroDataVar, etc. @vicentebolea , I'm thinking that the ZeroDataVar failures might actually be due to cmake 4 being used and therefore the tests not being actually run with MPI due to the problems noted in issue #4659. ZeroDataVar MPI tests will fail if not actually run with MPI because they require multiple ranks. WRT the MultiBlock failures, I'll try to duplicate this with 24 ranks, but without any actual indication of how the test failed we don't have any threads to pull to see what might have gone wrong.

[qbisi]

The detailed build log:

1. x86_64-linux
2. aarch64-darwin

I'm thinking that the ZeroDataVar failures might actually be due to cmake 4 being used and therefore the tests not being actually run with MPI due to the problems noted in issue

we resolve this problem with postPath, see https://github.com/NixOS/nixpkgs/blob/d2b635a26f0bc11715a2bdd595831c7cecb29aa1/pkgs/by-name/ad/adios2/package.nix#L70-L74

[eisenhauer]

I've reopened as all problems may not be fixed by #4758