This is a summary of the discussion at the Fortran Specialist Group meeting on 12^th June 2008. It followed presentations by John Reid and David Muxworthy on the revision of the current (2003) Fortran standard.

The discussion was introduced with a reference to the corresponding discourse at the previous year's AGM. Then the conclusion had been that the first Working Draft of Fortran 2008 was too much, too soon because of the unanticipated delay in vendors producing Fortran 2003 processors and because J3 had completed more second priority items than expected. The UK position at the forthcoming WG5 meeting was recommended to be that the schedule should be slipped by two years, three major items (BITS, coarrays and intelligent macros) should be removed and that consideration should be given to removal of some of the medium-sized items.

The net effect at the two WG5 meetings in the interim had been that the schedule was slipped by ten months and BITS and intelligent macros were deleted. It had not been considered worth the trouble of deleting medium-sized items since they had been already been incorporated into the document. That had left coarrays as the most contentious item. At the WG5 meeting in February 2008 the UK had proposed a compromise of moving all coarray features to a separate part of the standard; this would allow vendors to regard coarrays as optional. The US had proposed a different compromise of retaining the basic features of coarrays in the base language and putting the remainder into a Technical Report, which vendors could then implement as they wished; the US position had prevailed. The Japanese delegation had had instructions not to support coarrays in the base language at all as their vendors preferred to pursue HPF.

Discussion now centred on this point. Nick Maclaren said that experience with Open MP had not been happy and that most users had gone back to MPI. He feared that the coarray model was analogous to that of Open MP. It was more difficult to debug Open MP programs. Coarrays were telling the compiler how to solve the problem: Fortran should leave that to the compiler. John Reid reported on a program transferred from using MPI to using coarrays. There had been little change in execution speed but the coarray code was much simpler. Coarray expression was more elegant. To test the mood of the meeting there was a straw vote on the principle of the relation of coarrays to the base standard.
The result was:

• coarrays as currently proposed (US compromise): 7
• coarrays only in an optional part of the standard: 4
• remove coarrays totally: 0
• undecided: 1

There was a further short discussion on whether the commonly used keywords real*4 and real*8 could be standardised. It was quickly decided that they could not.

David Muxworthy
14^th June 2008

---

Comments on this or any other of the Group's pages should be sent by email to the FSG Web Editor.

Last modified: Wed 2 Nov 2011 09:30:20