From J.L.Schonfelder@liverpool.ac.uk Thu Feb 24 10:55:24 1994 Received: from mailhub.liverpool.ac.uk (mail.liv.ac.uk) by dkuug.dk with SMTP id AA24196 (5.65c8/IDA-1.4.4j for ); Thu, 24 Feb 1994 11:59:42 +0100 Received: from 138.253.31.129 by mailhub.liverpool.ac.uk with SMTP (PP) id <17281-0@mailhub.liverpool.ac.uk>; Thu, 24 Feb 1994 10:55:25 +0000 From: "Dr.J.L.Schonfelder" Message-Id: <9402241055.AA21386@uxd.liv.ac.uk> Subject: More on interp 125 To: SC22WG5@dkuug.dk (SC22/WG5 members) Date: Thu, 24 Feb 1994 10:55:24 +0000 (GMT) X-Mailer: ELM [version 2.4 PL23] Mime-Version: 1.0 Content-Type: text/plain; charset=US-ASCII Content-Transfer-Encoding: 7bit Content-Length: 3709 X-Charset: ASCII X-Char-Esc: 29 From jls Thu Feb 24 10:36:45 1994 I am sending this note in responce to a number of reactions to my original comment. I was well aware that this J3 meeting vote in favour of the distributed interpretation. I am also aware that at the Brussels meeting with a different subset present, it voted for an interpretation that would have required call by reference semantics. The reason I am making a fuss is that the final say on this issue is a WG5 one and I believe that a serious diservice to users is being done. That the meaning of the current standard is confused and essentially internally inconsistent is clear. Merely, legislating to clarify the lack of proper definition is a deriliction of responsibility in a standards committee (I am not apologising for strong language since I feel formalising areas of non portability is NOT what standards bodies should be about). The issue as I see it, and I thought I had made this clear before, is in one part of the standard target arguments are treated as if they like any ordinary variable could be passed either by reference or by copy. In an other it is recognised that such arguments are different because they may have pointers associated with them and hence they should be passed by reference/descripter. Like all other cases where pass by descripter is required matching attributes are needed and the interface must be explicit. This is an inconsistency in definition and intent. It needs to be resolved not by legalistic sophistry but by looking carefully at what provides the most utility and least surprise to the user. If it is decided that allowing copyin/copyout semantics is of overriding importance then not only should the interpretation say this, it should also remove the then unnecessary requirement that both actual and dummy be targets and the interface explicit. It should be spelt out that any association between a pointer and the actual argument is not reflected into the dummy argument and is not guaranteed to continue to exist after the call. This would have some utility (not much in my opinion) but it would be possible for a non target actual arguments to become associated with a target dummy which could be then associated with local pointers and manipulated via these. (Nag would like this see forwarded mail from JDC) Or we should define properly the rules which use the explicit interface and matching attributes to guarantee pass by reference for targets and hence preserve associations between pointer and target across the call. This I believe to be the better way of resolving the inconsistency. Programs using both pointer and target arguments become portable and predictable. Either way the proposed interpretation 125 is unacceptable. I would like to argue strongly that the call by reference interpretation is vastly to be preferred. The possibly reference possibly copy semantics at the behest of the implementation will if ratified lead to programs that are highly non-portable. Users on a call by reference machine will write software that depends on the maintenance of pointer target association. Such programs will fail in fairly enigmatic ways on machines that use copy semantics. This is a clear area where targets should be different from ordinary variables which cant have pointer associations to be preserved or otherwise. I believe WG5 should ask J3 to think again on this one. -- Dr.J.L.Schonfelder Director, Computing Services Dept. University of Liverpool, UK Phone: +44(51)794 3716 FAX : +44(51)794 3759 email: jls@liv.ac.uk -- Dr.J.L.Schonfelder Director, Computing Services Dept. University of Liverpool, UK Phone: +44(51)794 3716 FAX : +44(51)794 3759 email: jls@liv.ac.uk