Some fortran fixes

[awlauria]

fortran module: assumed size arg must match actual array arg of any rank

Here's an example code that fails to compile prior to this checkin:

     use mpi
     integer, contiguous, pointer :: requests(:, :)
     integer :: ierr
     allocate(requests(2,2))
     requests(1,1) = 1
     requests(2,1) = 2
     requests(1,2) = 3
     requests(2,2) = 4
     call MPI_Waitall(size(requests),requests,MPI_STATUSES_IGNORE,ierr)
     end

But according to the standard under "A.4 Fortran Bindings with mpif.h or the mpi
Module", waitall is supposed to take

    INTEGER COUNT, ARRAY_OF_REQUESTS(*)

and under fortran that assumed size arg should match the above actual arg.

The fortran module previously had

    integer, dimension(count), intent(inout) :: array_of_requests

and this checkin changes it to

    integer, dimension(*), intent(inout) :: array_of_requests

According to Rafik Zurob the rules for assumed-size arrays are such that we know
the incoming array_of_requests is contiguous so nothing changes in the fundamental
operation of the waitall call.

Signed-off-by: Austen Lauria awlauria@us.ibm.com

[awlauria]

I will take a look at the mailmap faiure.

[awlauria]

mailmap issue should be resolved.

[jsquyres]

I think the corresponding changes are also needed in the use-mpi-ignore-tkr directory, too, right?

[awlauria]

@markalle can you check to see if there's any additional changes needed?

[awlauria]

@jsquyres I think @markalle addressed this in a seperate PR here:
#9367

[jsquyres]

Upon further review, I do not believe that the 2nd commit on this PR is correct.

I'm fine with the first commit, but the 2nd one needs to be either removed or revised per #9367 (review).

[awlauria]

I seperated out the first commit to a seperate PR: #9405

And removed it from here.

[awlauria]

I converted this to a draft until we decide what to do with it.