Also, please add the newly added components to stack_custom.yaml and stack_mac.yaml for CI.

Also, please add the newly added components to stack_custom.yaml and stack_mac.yaml for CI.

Done in c75b290.

The CI is failing because MAPL can't find ESMF when built without modules. I think you need to set ESMFMKFILE in build_mapl.sh

The CI is failing because MAPL can't find ESMF when built without modules. I think you need to set ESMFMKFILE in build_mapl.sh

Would it be better to set ESMFMKFILE in the ESMF modulefile?

This is when building without modules. On the else side of if $MODULES.

I think $ESMF_ROOT/lib/esmf.mk should do the trick.

This is when building without modules. On the else side of if $MODULES.

I think $ESMF_ROOT/lib/esmf.mk should do the trick.

Done. Thank you!

I can't figure out why the build is failing on Ubuntu. HDF5 has an error linking to pthreads, but I'm not sure why. I'm pretty sure pthreads is a built-in library. It was working previously.

Looks like everything is resolved. I am trying to figure out the CI failure, and then this can be merged.

Changing the OS to Ubuntu 18.04 works

Thanks @rmontuoro. I anticipate the builds will pass now.

The build gets a lot further on Ubuntu now, and fails in MAPL.

/home/runner/work/hpc-stack/hpc-stack/mpich/pkg/mapl-v2.7.1/shared/Interp/Interp_implementation.F90:1:25:
1 | submodule (MAPL_InterpMod) Interp_implementation
|                         1
2 |   use, intrinsic :: iso_fortran_env, only: REAL32
 |                                                 2
Error: Symbol ‘real32’ at (1) conflicts with the symbol at (2)
/home/runner/work/hpc-stack/hpc-stack/mpich/pkg/mapl-v2.7.1/shared/Interp/Interp_implementation.F90:1:25:
1 | submodule (MAPL_InterpMod) Interp_implementation
|                         1
2 |   use, intrinsic :: iso_fortran_env, only: REAL32
3 |   use, intrinsic :: iso_fortran_env, only: REAL64
|                                                 2
Error: Symbol ‘real64’ at (1) conflicts with the symbol at (2)

Seems like a compiler error. It works on macOS which is using gfortran 9.4.0, but Ubuntu is using gfortran 9.3.0. I suspect that's the difference.

/home/runner/work/hpc-stack/hpc-stack/mpich/pkg/mapl-v2.7.1/shared/Interp/Interp_implementation.F90:1:25:
1 | submodule (MAPL_InterpMod) Interp_implementation
| 1
2 | use, intrinsic :: iso_fortran_env, only: REAL32
| 2
Error: Symbol ‘real32’ at (1) conflicts with the symbol at (2)
/home/runner/work/hpc-stack/hpc-stack/mpich/pkg/mapl-v2.7.1/shared/Interp/Interp_implementation.F90:1:25:
1 | submodule (MAPL_InterpMod) Interp_implementation
| 1
2 | use, intrinsic :: iso_fortran_env, only: REAL32
3 | use, intrinsic :: iso_fortran_env, only: REAL64
| 2
Error: Symbol ‘real64’ at (1) conflicts with the symbol at (2)

I've opened an issue with NASA.

I think I just need to update GCC. I found a bug on the GCC forums from this year that seems related.

I think I just need to update GCC. I found a bug on the GCC forums from this year that seems related.

From which version to which? We need to be able to support GNU 9.x for the moment.

It appears MAPL doesn't build with gfortran 9.3.0, but works with 9.4.0. I couldn't find a way to install 9.4.0 in the CI, so I'm trying 10.2.0.

It appears MAPL doesn't build with gfortran 9.3.0, but works with 9.4.0. I couldn't find a way to install 9.4.0 in the CI, so I'm trying 10.2.0.

Means we need a patch so that it works with 9.1.0+, otherwise it's a showstopper or at least will cause a longer delay, because we would need to update the GNU compilers on Hera and Cheyenne and possibly for the CI tests.

I'll investigate a little more and let you know

@kgerheiser - see comment in MAPL issue #892

@climbfuji we have a fix that allows MAPL to build with gfortran 9.x

@climbfuji we have a fix that allows MAPL to build with gfortran 9.x

Thanks, that's great news.

A new MAPL release including the fix is being assembled and should be available soon.

@kgerheiser - Could you please look into the openmpi build failure? We may need to specify proper linking flags for C++ bindings.

I thought that might happen. I have a fix.

I put in a PR to your branch. If you build OpenMPI yourself then the C++ library is linked in properly when using the Fortran wrapper.

All builds pass now. @aerorahul any more comments?