(j3.2006) edit for interp F08/0144
Fri Jan 8 02:52:55 EST 2016
On 01/07/16 20:55, Cohen Malcolm wrote:
> The edit provided for F08/0144 does not say anything about child data
> transfer statements, nor should it, considering that is not the question
> being asked by F08/0144.
The question asked in the title of F08/0144 is
Is nonadvancing I/O allowed during execution of DO CONCURRENT?
The answer given is
It was intended that nonadvancing input/output not be
permitted within a DO CONCURRENT construct.
The final sentence of Subclause 220.127.116.11 states
A formatted child input/output statement is a
nonadvancing input/output statement, and any
ADVANCE= specifier is ignored.
Therefore, either the answer is wrong or the edit is wrong.
If the answer were
It was intended that nonchild nonadvancing
input/output not be permitted within a
DO CONCURRENT construct.
then the proposed edit would be adequate.
I recognize that there are other problems with child data
transfer statements and DO CONCURRENT that are not addressed
> This would seem to be an additional question.
> Also, since one cannot tell at compile time whether a data transfer
> statement is a child, any thought towards the prohibition of this would need
> a different kind of fix to the one in F08/0144. We could mandate raising an
> error condition, or we could simply say it is not allowed and leave it up to
> the implementation to make mincemeat of the user's program. Sounds like an
> interesting topic for the next meeting.
> -----Original Message-----
> From: Robert Corbett
> Sent: Tuesday, January 5, 2016 12:13 PM
> To: j3
> Subject: (j3.2006) edit for interp F08/0144
> The edit provided for interp F08/0144 appears to allow child
> data transfer statements to be executed within a DO CONCURRENT
> construct. Did the committee intend to allow such usage?
> An example would be if a procedure invoked by user-defined
> derived-type editing contained a DO CONCURRENT loop that
> executed a child data transfer statement on some or all of
> its iterations.
> User-defined derived-type editing hurts my brain enough
> without the added complexity of interactions with
> DO CONCURRENT.
> Robert Corbett
> J3 mailing list
> J3 at mailman.j3-fortran.org
> This e-mail has been scanned for all viruses by Star.
More information about the J3