@boegel thanks for the PR!
Going over OpenMPI in foss-2019b with --module-only --include-easyblocks-from-pr 2582 seemed to work in my debug build (designated local buildpath, --debug flag and set tmp dirs). To double check I also ran the build process in foss-2021a without --module-only, building it in a "production setting" and this is giving me the same error as before:

FAILED: Installation ended unsuccessfully (build directory: /tmp/easybuild/rome/OpenMPI/4.1.1/GCC-10.3.0): build failed (first 300 chars): Sanity check failed: sanity check command mpirun -n 8 /tmp/easybuild/rome/OpenMPI/4.1.1/GCC-10.3.0/mpi_test_hello_mpifh exited with code 127 (output: /tmp/easybuild/rome/OpenMPI/4.1.1/GCC-10.3.0/mpi_test_hello_mpifh: error while loading shared libraries: libgfortran.so.5: cannot open shared object f...

I'm a bit confused by this and think the only relevant difference should be running with/without --module-only.
Could this be affected by possibly left-over files from a previous attempt to build OpenMPI in 2021a?

In any case, I'm trying to apply this again in a clean build with 2021a and get back to you. Just wanted to give an update on this

Building OpenMPI again in 2021a still fails. It also fails in 2019b with --rebuild. Running again with --module-only with the eb from this PR, seems to make the build happy, but my feeling is that the part resulting in the error is skipped (although the sanity check step is run). Not sure why module-only makes a difference.

Building OpenMPI again in 2021a still fails. It also fails in 2019b with --rebuild. Running again with --module-only with the eb from this PR, seems to make the build happy, but my feeling is that the part resulting in the error is skipped (although the sanity check step is run). Not sure why module-only makes a difference.

Indeed, the example progress are not run during the sanity check under --module-only, because the example programs are compiled from the unpacked source directory (which is not there when using --module-only). That's a bug in how the sanity check is done, it shouldn't make a difference.

Test report by @boegel

Overview of tested easyconfigs (in order)

• SUCCESS OpenMPI-4.0.3-GCC-9.3.0.eb

Build succeeded for 1 out of 1 (1 easyconfigs in total)
login1 - Linux UNKNOWN UNKNOWN, x86_64, Intel(R) Xeon(R) CPU E5-2690 v3 @ 2.60GHz (haswell), Python 3.9.5
See https://gist.github.com/1eccc898554fcc6badfb7be6fbf95de9 for a full test report.

edit: this was tested in the EESSI environment, where --rpath is enabled