Path: utzoo!mnetor!uunet!husc6!uwvax!rutgers!clyde!watmath!rbutterworth
From: rbutterworth@watmath.waterloo.edu (Ray Butterworth)
Newsgroups: comp.lang.c
Subject: Re: You win.  (I now dis-like NOALIAS)
Message-ID: <16042@watmath.waterloo.edu>
Date: 24 Dec 87 15:03:07 GMT
References: <10989@brl-adm.ARPA>
Organization: U of Waterloo, Ontario
Lines: 31

In article <10989@brl-adm.ARPA>, TLIMONCE%DREW.BITNET@CUNYVM.CUNY.EDU writes:
> After reading all the recent posts about noalias I have come to the following
> conclusions:
>         - Leave it out.  It's not worth all this debating.
>         - ...

A small number of compiler writers seem to think that it is
necessary to have noalias as a part of the language (e.g. those
that share a common code generator with fortran or want to use
vector hardware).

On most compilers, noalias would probably have no effect,
and it is fairly simple to add to the grammar.

But why bother making things more complicated than they need to be?
There isn't really any reason to make noalias part of the language.

Those compiler writers that think they need it are already allowed
to do something like adding a "_noalias" symbol to their compiler.
They simply have to put a warning into their manuals indicating
that any code written for this compiler should have something like
#if !defined(_XYZ_COMPILER)
#   define _noalias /*nothing*/
#endif
at the top if it is to be ported to any other vendor's compiler.

This is only one suggestion.  There are many variations on this.
The important point is that it can be done by those that need it
without requiring any additions to the language by the ANSI standard
while still allowing code that takes advantage of their compiler's
optimizations to be portable to other ANSI compilers.