Path: utzoo!utgpu!news-server.csri.toronto.edu!rpi!think.com!samsung!uunet!mcsun!hp4nl!charon!dik
From: dik@cwi.nl (Dik T. Winter)
Newsgroups: comp.lang.fortran
Subject: Re: matrix multiplication
Keywords: parameters, pass-by-value, pass-by-reference
Message-ID: <3488@charon.cwi.nl>
Date: 8 May 91 00:18:58 GMT
References: <1991May04.170203.22222@ariel.unm.edu> <23261@lanl.gov> <1991May07.155218.13741@ariel.unm.edu>
Sender: news@cwi.nl
Organization: CWI, Amsterdam
Lines: 25

In article <1991May07.155218.13741@ariel.unm.edu> scavo@cie.uoregon.edu (Tom Scavo) writes:
 > So what is the concensus on this issue?  Is it within the
 > standard to implement pass-by-value-result?  If so, what
 > compilers have gone this route?
I fail to see any difference between what Jim Giles said and what I said.
Pass-by-value-result is allowed, as is pass-by-reference.  The standard
allows both.  As I said, IBM compilers use pass-by-value-result, but CDC
compilers use pass-by-reference.
 > 
 > As an aside, how did it happen that two competing parameter
 > passing mechanisms got into the Fortran 77 standard?  (This
 > question assumes, of course, that this is indeed the case.) 
This is indeed *not* the case.  The standard simply leaves the parameter
passing mechanisms unspecified.  It only specifies the result for
non-aliased parameters.  Any program assuming one way or the other depends
on improper aliasing, and that makes the program illegal.

If you want a detour to Ada:  in Ada it *is* specified that simple parameters
are passed-by-value-return.  Non-simple parameters are to be passed by
either value-return or reference.  The reason is that if the subroutine
you are calling resides on a different processor you really *want*
value-return in most cases.
--
dik t. winter, cwi, amsterdam, nederland
dik@cwi.nl