Re: Bug 84894 - [F2018] provide iso_fortran_binding.h

[Jeff Hammond <jeff.science@gmail.com>]

Sorry, I meant J3 not J5 but in any case,
http://mpi-forum.org/docs/mpi-3.1/mpi31-report/node409.htm contains this:

===================

*Rationale.*

The features in TS 29113 on further interoperability with C were decided on
by ISO/IEC JTC1/SC22/WG5 and designed by PL22.3 (formerly J3) to support a
higher level of integration between Fortran-specific features and C than
was provided in the Fortran 2008 standard; part of this design is based on
requirements from the MPI Forum to support MPI-3.0. According to [41
<http://mpi-forum.org/docs/mpi-3.1/mpi31-report/node522.htm#-Bib41>], ``an
ISO/IEC TS is reviewed after three years in order to decide whether it will
be confirmed for a further three years, revised to become an International
Standard, or withdrawn. If the ISO/IEC TS is confirmed, it is reviewed
again after a further three years, at which time it must either be
transformed into an International Standard or be withdrawn.''

The TS 29113 contains the following language features that are needed for
the MPI bindings in the mpi_f08 module: assumed-type and assumed-rank. It
is important that any possible actual argument can be used for such dummy
arguments, e.g., scalars, arrays, assumed-shape arrays, assumed-size
arrays, allocatable arrays, and with any element type, e.g., REAL,
CHARACTER*5, CHARACTER*(*), sequence derived types, or BIND(C) derived
types. Especially for backward compatibility reasons, it is important that
any possible actual argument in an implicit interface implementation of a
choice buffer dummy argument (e.g., with mpif.h without argument-checking)
can be used in an implementation with assumed-type and assumed-rank
argument in an explicit interface (e.g., with the mpi_f08 module).

A further feature useful for MPI is the extension of the semantics of the
ASYNCHRONOUSattribute: In F2003 and F2008, this attribute could be used
only to protect buffers of Fortran asynchronous I/O. With TS 29113, this
attribute now also covers asynchronous communication occurring within
library routines written in C.

The MPI Forum hereby wishes to acknowledge this important effort by the
Fortran PL22.3 and WG5 committee. (* End of rationale.*)

===================

When we voted this into MPI-3, we thought it was going to appear in Fortran
much sooner.

In any case, all of the Fortran compiler implementors that attend the MPI
Forum supported this.  IBM, Intel, and Cray implemented it rather quickly.

Jeff


On Thu, Mar 15, 2018 at 7:45 PM Jeff Hammond <jeff.science@gmail.com> wrote:

>
>
> > On Mar 15, 2018, at 7:07 PM, Steve Kargl <
> sgk@troutmask.apl.washington.edu> wrote:
> >
> >> On Thu, Mar 15, 2018 at 06:50:11PM -0700, Damian Rouson wrote:
> >>
> >>> On March 15, 2018 at 5:43:03 PM, Jeff Hammond (jeff.science@gmail.com
> (mailto:jeff.science@gmail.com)) wrote:
> >>>
> >>> It’s in TS29113. Look at
> https://gcc.gnu.org/onlinedocs/gfortran/Further-Interoperability-of-Fortran-with-C.html
> for details.
> >>>
> >>
> >> And the features described in TS29113 have been incorporated into the
> Fortran 2018 Draft International Standard available at
> http://isotc.iso.org/livelink/livelink?func=ll&objId=19442438&objAction=Open&viewType=1
> .
> >>
> >
> > The module name is mpi_f08 not mpi_f08_ts29113 or mpi_f18.
> > Seems strange to name a module with a moniker that identifies
> > a specific standard, which does not include the required
> > information.
>
> Please talk to one of the J5 members who was part of both MPI-3 and
> TS29113 efforts if you want the history.
>
> Is it your intent to filibuster this feature and prevent gfortran from
> supporting MPI-3 or do you want to have a constructive conversation where
> we talk about things that benefit users?
>
> Jeff

-- 
Jeff Hammond
jeff.science@gmail.com
http://jeffhammond.github.io/