From JLS@liverpool.ac.uk Fri Nov 29 18:37:39 1991 Received: from danpost2.uni-c.dk by dkuug.dk via EUnet with SMTP (5.64+/8+bit/IDA-1.2.8) id AA00808; Fri, 29 Nov 91 18:37:39 +0100 Received: from vm.uni-c.dk by danpost2.uni-c.dk (5.65/1.34) id AA20579; Fri, 29 Nov 91 17:37:45 GMT Message-Id: <9111291737.AA20579@danpost2.uni-c.dk> Received: from vm.uni-c.dk by vm.uni-c.dk (IBM VM SMTP V2R1) with BSMTP id 0703; Fri, 29 Nov 91 18:37:34 DNT Received: from UKACRL.BITNET by vm.uni-c.dk (Mailer R2.07) with BSMTP id 2406; Fri, 29 Nov 91 18:37:34 DNT Received: from RL.IB by UKACRL.BITNET (Mailer R2.07) with BSMTP id 1433; Fri, 29 Nov 91 16:23:13 GMT Received: from RL.IB by UK.AC.RL.IB (Mailer R2.07) with BSMTP id 7209; Fri, 29 Nov 91 16:20:54 GMT Via: UK.AC.LIV.IBM; 29 NOV 91 16:19:49 GMT Received: from JLS@UK.AC.LIVERPOOL by MAILER(4.2.a); 29 Nov 1991 16:19:05 GM Date: Fri, 29 Nov 91 15:58:46 GMT From: Lawrie Schonfelder Subject: PCF Parallel extensions To: SC22/WG5 members X-Charset: ASCII X-Char-Esc: 29 I ahve just got round to reading John Reid's critique of the PCF proposals. I would like to strongly support his general points. Extension to F77 language is absolutely unacceptable. (I would go further any extension to the syntax of Fortran, even to F90 is unacceptable, unless it is done by the mechanism of proposal to WG5 and is processed vis the proper channels for Language revision or ammendment). I have very considerable reservations about the nature of the approach being proposed. There is a a strong embedding of a particular architectural model into the thinking of most of these suggestions, viz. a shared-memory multi-processor. I think it highly undesirable to design language of this type into Fortran. We should be trying to reduce the degree of architecture lock-in in the language. The good bits of F90 were an attempt to reduce the degree to which the single proceesor linear memory Von Neuman machine model was built into the language. I would be totally opposed to adding extensions which built the shared memory model in for parallel computing. System implementation languages should and need to be architecture aware. In principle they have limited portability and are intended for the developement of machine specific code like OS kernals and Compiler back-ends etc. Applications languages which I consider Fortran to be should be system architecture neutral but be expressive of the problem, the what rather than the detailed how. These parallel constructs are excessively bothered by expressing the how. Lawrie