fortran: Fix MPI_ARGV(S)_NULL compilation error

[kawashima-fj]

Fortran constants MPI_ARGV_NULL and MPI_ARGVS_NULL are defined in MPI-3.1 p.680 as below.

MPI_ARGVS_NULL
2-dim. array of CHARACTER*(*)
MPI_ARGV_NULL
array of CHARACTER*(*)

MPI_ARGV_NULL and MPI_ARGVS_NULL are used as an argument of MPI_COMM_SPAWN and MPI_COMM_SPAWN_MULTIPLE respectively and their argument argv and array_of_argv are defined as below for USE mpi_f08 binding in MPI-3.1.

CHARACTER(LEN=*), INTENT(IN) :: argv(*)
CHARACTER(LEN=*), INTENT(IN) :: array_of_argv(count, *)

Defining them as INTEGER in mpi_f08 module will cause a compilation error of user programs like "There is no specific subroutine for the generic 'mpi_comm_spawn'".

@ggouaillardet @jsquyres I'm not so familiar with Fortran and I don't know whether they are intentionally defined as INTEGER in 9c77c6b. Could you confirm my change?

USE mpi and include 'mpif.h' bindings use ompi/include/mpif-sentinels.h and the two variables are defined correctly:

!     Making MPI_ARGV_NULL be the same type as the parameter that is
!     exepected in the F90 binding for MPI_COMM_SPAWN means that we
!     don't need another interface for MPI_COMM_SPAWN.
      character mpi_argv_null(1)
!     Ditto for MPI_ARGVS_NULL / MPI_COMM_SPAWN_MULTIPLE.
      character mpi_argvs_null(1, 1)

This commit may break ABI compatibility. But it affects only Fortran mpi_f08 module and MPI_ARGV_NULL and MPI_ARGVS_NULL with mpi_f08 have not been able to use for their purpose (argument of MPI_COMM_SPAWN(_MULTIPLE)) ever. So I'll create PRs for release branches.

[bwbarrett]

bot:ompi:retest

[ggouaillardet]

@kawashima-fj it looks like it's been there forever and undocumented.
imho, this is a bug and your fix is correct
@jsquyres any thoughts ?

[jsquyres]

@kawashima-fj Could you add a simple SPAWN test in the ompi-tests/ibm/dynamic directory that these this MPI_ARGV_NULL and MPI_ARGVS_NULL constants with all the Fortran interfaces? That would help prevent this from happening again.

Thanks!

[kawashima-fj]

Ok, I'll add.

[ggouaillardet]

it is probably best to add the test after all PRs are merged.
otherwise, building the test suite will fail and no test will be performed.