discuss ABI implementation of mpi_subarrays_supported

[jeffhammond]

mpi_async_protects_nonblocking is, as best I can tell, strictly dependent on the Fortran compiler, so it can be implemented in a compiler-dependent but MPI library-agnostic way.

mpi_subarrays_supported is largely dependent on the MPI library itself, in how it handles CFI_cdesc_t.

How do people want to handle this in the ABI? I am thinking it can be returned from the Fortran ABI info object, although that API was designed for the Fortran compiler type information. Can we mix them, or should there be a new API for this?

[jeffhammond]

I never figured out how to address this, and the state of the Fortran ABI is complete in other respects.

Some day, we will just have to require subarrays, I guess.