ISO/IEC JTC1/SC22/WG5 N1752

              Draft response to N1745 (VOLATILE coarrays)

                              John Reid


The primary purpose of allowing coarrays to be volatile is to support the
spin loop, illustrated in NOTE 8.38 of N1723 (the currect draft for the next
Fortran standard). The presence of the spin loop allows the programmer to
construct needed synchronizations that are not provided directly in the
language. 

On behalf of the UK, Nick Maclaren points out that programmers are likely to
use volatile coarrays in other ways and will get results that they did not
expect. Futhermore, these unexpected results may occur quite infrequently,
leading to failures of programs that were thought to be tested. I am
convinced by these arguments.

The US proposes that locks should be added to the draft (see J3/08-256r2), to
provide an efficient mechanism for programmers to avoid race conditions
when they require data to be altered by more than one executing image. This
is an example of a synchronization that could be coded with the help of a
volatile coarray, but I agree that it is of sufficient importance to
justify its inclusion as an intrinsic feature. 

It is my belief that this feature provides an adequate replacement for the
volatile coarray to support the spin loop. For example, the code in NOTE
8.38 of N1723:

LOGICAL,VOLATILE :: LOCKED[*] = .TRUE. 
INTEGER :: IAM, P, Q 

IAM = THIS_IMAGE() 
IF (IAM == P) THEN
   ! Preceding segment
   SYNC MEMORY               ! A 
   LOCKED[Q] = .FALSE.       ! segment Pi
   SYNC MEMORY               ! B
ELSE IF (IAM == Q) THEN
   DO WHILE (LOCKED); END DO ! segment Qj
   SYNC MEMORY               ! C
   ! Subsequent segment 
END IF

may be replaced by 

USE, INTRINSIC :: ISO_FORTRAN_ENV

TYPE (LOCK_TYPE) :: SpinLock[*]
INTEGER :: IAM, P, Q 

LOCK(SpinLock) ! Initialize to LOCKED
     :
IAM = THIS_IMAGE() 
IF (IAM == P) THEN
   ! Preceding segment
   UNLOCK(SpinLock[Q])    
ELSE IF (IAM == Q) THEN
   LOCK (SpinLock) ! The image spins here
   ! Subsequent segment 
END IF

I see the replacement as much clearer. 

For these reasons, I suggest to WG5 that coarrays should not be permitted
to have the VOLATILE attribute and that the feature for locks proposed by
the US be added.