ISO/IEC JTC1/SC22/WG5 N1710

                Moving the co-array feature into a TR

                    David Muxworthy and John Reid


There is a significant minority view within WG5 that the 
co-array feature should be moved to a TR. This paper explores  
this idea. It would

1. Reduce the extent of the changes and make Fortran 2008 
more the minor revision that was the original intent.

2. Avoid committing co-array syntax to be included in Fortran
until we are certain that it is the best approach.

3. Sustain the encouragement to vendors to provide implementations,
since the syntax and semantics will be defined.

We already have draft text to hand to achieve this:

1. The Editor has a "remove co-arrays" switch built into the Latex 
source for the draft standard. He has constructed a version without 
co-arrays and this is visible as N1709. 

2. John has already constructed a technical summary of co-arrays in 
N1708, which may be used as a first draft for the technical summary 
in the TR. 

3. In these days of electronic publication, we see no problem with 
using the complete Fortran 2008 plus co-arrays text for the detailed 
specification. It would require that the Editor continues to 
maintain a single source for both versions. 

For timing, we see no problem with simultaneous construction of 
the Final CD for Fortran 2008 and the final draft TR. Our schedule
(N1693) aims for this to happen in July 2009.

The wording that we have put in each of our recent TRs promising to 
include its contents in the next revision is not quite right for 
this case. We suggest:

"It is the intention of ISO/IEC JTC1/SC22/WG5 that if co-arrays are
included in a future revision of ISO/IEC 1539-1:2004 (Fortran), the
semantics and syntax described in this Technical Report shall be 
incorporated therein exactly as they are specified here unless
experience in the implementation and use of this feature has
identified any errors which need to be corrected."

We will need to make a case to SC22 for creating another TR. The
essence of this case is as follows:

"It was the intention of WG5 that the revision of Fortran 2003
should be minor, given that Fortran 2003 is a major revision of
Fortran 95.  WG5 is concerned that three years after the publication
of Fortran 2003, there are still no full implementations (though
several are near completion). Following revisions made at the 2007
meeting in London, there are only two major new features in the
draft. The smaller of these is submodules, which was promised to be
part of the revision in  ISO/IEC TR 19767. The larger is co-arrays.
Drastic reduction of the extent of this feature would make it
insufficiently powerful to be adopted widely. We therefore propose
to move co-arrays to a TR. This would leave a draft revision that is
a minor extension while retaining a precise definition of the
co-array feature."