Can't compile Fortran application with -std=f2008 using mpi_f08 module

[paulromano]

I have a Fortran application that uses MPI (via mpi_f08) and is normally compiled with gfortran -std=f2008. When trying to compile against OpenMPI 1.10.2, I get the following build errors:

/home/romano/openmc/src/message_passing.F90:5:6:

   use mpi_f08
      1
Error: GNU Extension: LOGICAL dummy argument ‘flag’ at (1) with non-C_Bool kind in BIND(C) procedure ‘mpi_comm_copy_attr_function’
/home/romano/openmc/src/message_passing.F90:5:6:

   use mpi_f08
      1
Error: GNU Extension: LOGICAL dummy argument ‘flag’ at (1) with non-C_Bool kind in BIND(C) procedure ‘mpi_type_copy_attr_function’
/home/romano/openmc/src/message_passing.F90:5:6:

   use mpi_f08
      1
Error: GNU Extension: LOGICAL dummy argument ‘flag’ at (1) with non-C_Bool kind in BIND(C) procedure ‘mpi_win_copy_attr_function’
/home/romano/openmc/src/message_passing.F90:5:6:

   use mpi_f08
      1
Error: GNU Extension: LOGICAL dummy argument ‘flag’ at (1) with non-C_Bool kind in BIND(C) procedure ‘mpi_comm_copy_attr_function’
/home/romano/openmc/src/message_passing.F90:5:6:

   use mpi_f08
      1
Error: GNU Extension: LOGICAL dummy argument ‘flag’ at (1) with non-C_Bool kind in BIND(C) procedure ‘mpi_type_copy_attr_function’
/home/romano/openmc/src/message_passing.F90:5:6:

   use mpi_f08
      1
Error: GNU Extension: LOGICAL dummy argument ‘flag’ at (1) with non-C_Bool kind in BIND(C) procedure ‘mpi_win_copy_attr_function’
/home/romano/openmc/src/message_passing.F90:5:6:

   use mpi_f08
      1
Error: GNU Extension: LOGICAL dummy argument ‘flag’ at (1) with non-C_Bool kind in BIND(C) procedure ‘mpi_comm_dup_fn’
/home/romano/openmc/src/message_passing.F90:5:6:

   use mpi_f08
      1
Error: GNU Extension: LOGICAL dummy argument ‘flag’ at (1) with non-C_Bool kind in BIND(C) procedure ‘mpi_comm_null_copy_fn’
/home/romano/openmc/src/message_passing.F90:5:6:

   use mpi_f08
      1
Error: GNU Extension: LOGICAL dummy argument ‘flag’ at (1) with non-C_Bool kind in BIND(C) procedure ‘mpi_type_dup_fn’
/home/romano/openmc/src/message_passing.F90:5:6:

   use mpi_f08
      1
Error: GNU Extension: LOGICAL dummy argument ‘flag’ at (1) with non-C_Bool kind in BIND(C) procedure ‘mpi_type_null_copy_fn’
/home/romano/openmc/src/message_passing.F90:5:6:

   use mpi_f08
      1
Error: GNU Extension: LOGICAL dummy argument ‘flag’ at (1) with non-C_Bool kind in BIND(C) procedure ‘mpi_win_dup_fn’
/home/romano/openmc/src/message_passing.F90:5:6:

   use mpi_f08
      1
Error: GNU Extension: LOGICAL dummy argument ‘flag’ at (1) with non-C_Bool kind in BIND(C) procedure ‘mpi_win_null_copy_fn’

I should note that the errors still persist with -std=f2008ts as well. I'm on Ubuntu 15.10 with gfortran 5.2.1 which I used to build OpenMPI from source. Not sure if there is a solution for this given the problems with logicals and C interoperability...

[ggouaillardet]

@paulromano thanks for the report
so gfortran 5.2.1 is more strict than previous versions when compiling with -std=f2008

at first glance, there is no need for C/Fortran interoperability here (though it is simple and convenient). i ll try to find a solution for that.

[jsquyres]

@ggouaillardet Looks like this is the fault of open-mpi/ompi-release#851 -- that PR introduced some BIND(C)'s to subroutines with LOGICAL parameters, which is not allowed.

[jsquyres]

PR to fix this issue: #1327

[jsquyres]

@paulromano Just curious: is there a reason you have to use the -std=f2008ts flag?

[paulromano]

@jsquyres No reason that I have to, per se. It's just part of the style guide for my code that any new additions should be F2008 compliant. When I get a pull request from someone with a change to the code, having -std=f2008 gives added assurance that no compiler-specific extensions have been used.

[jsquyres]

@paulromano I realized I never ACKed your response -- oops! Ok, thanks for the info. I don't think Github sends mails for the related PRs and issues opened on this ticket, but we've split the solution into 2 parts: 1) fix the accidental LOGICAL pollution of BIND(C) stuff, and 2) re-implement the MPI default attribute handling callbacks in the referenced Github PRs.

[paulromano]

@jsquyres @ggouaillardet Thanks for the quick action on this issue; look forward to the next version!