what does 'equivalent to Fortran' mean

[jeffhammond]

Problem

Regarding MPI 4.0 page 869, the following seem to be missing a word:

"C type: MPI_Fint* equivalent to Fortran"
"C type: MPI_F08_status* equivalent to Fortran"

I know for a fact that MPI_Fint and MPI_F08_status are, by themselves, not equivalent to Fortran.

Proposal

Say the thing in Fortran that is actually actually equivalent to MPI_Fint and MPI_F08_status.

Changes to the Text

TBD

Impact on Implementations

Implementaters will be less confused when reading the spec.

Impact on Users

Users will be less confused when reading the spec.

References and Pull Requests

PR 713

[jeffhammond]

@RolfRabenseifner This is an erratum, right?

[RolfRabenseifner]

@RolfRabenseifner This is an erratum, right?

I expect that the text on MPI-4.0 pages 844 and 845 clearly tells to implementors and users the meaning of these constants.
But I understand your problem with this table.

Do you want to change the text "equivalent to Fortran" into

• "equivalent to Fortran constants:"

or

• "Equivalent to Fortran constants:"

that it gets absolutely clear that this column header is the column header of the two values below, rather than being a some continuation of the wording "C type: MPI_Fint*" and "C type: MPI_F08_status*"?

[RolfRabenseifner]

@jeffhammond Which correction would you prefer? One of the two corrections proposed above, or a third one?

[bkmgit]

There have been related complications in implementations:

• In obscure situations, sizeof(MPI_Fint ) != sizeof (int) is a problem for us open-mpi/ompi#76
• mpi4: add Fortran constants to mpi.h pmodels/mpich#5057

[jeffhammond]

@RolfRabenseifner "equivalent in Fortran" seems like the smallest correct change

[bkmgit]

Another related issue open-mpi/ompi#7918

OpenMPI MPI_Fint is the same as ompi_fortran_INTEGER_t
Check sizes during configuration of OpenMPI

MPICH Definitions that are determined by configure

In both cases, the type is the default Fortran integer type specified when compiling the library. Programs using the compiled library need to use the same default Fortran integer type.

Maybe "equivalent Fortran constant" is better?

[bkmgit]

@jeffhammond @RolfRabenseifner Feedback appreciated

[RolfRabenseifner]

Yes, I also needed some time to fully understand what this table should say.

I would recommend to substitute

C type: MPI_Fint* ___________ equivalent to Fortran

by

C constant (type: MPI_Fint*) ______ is equivalent to the Fortran constant