Xref: utzoo comp.lang.c:10299 comp.lang.fortran:683
Newsgroups: comp.lang.c,comp.lang.fortran
Path: utzoo!henry
From: henry@utzoo.uucp (Henry Spencer)
Subject: Re: no noalias not negligible - a difference between C and Fortran - long
Message-ID: <1988May23.000335.693@utzoo.uucp>
Organization: U of Toronto Zoology
References: <54080@sun.uucp>
Date: Mon, 23 May 88 00:03:35 GMT

Everyone agrees that the problem needs attention; unfortunately, there is
no well-proven solution to it for C, and thus it is inappropriate to try
to deal with it as part of standardization of the existing language.

> ...  Perhaps what will happen is that high-performance
> C compilers will adopt the questionable CDC/Cray Fortran practice 
> of providing "unsafe" optimization levels that, for instance,
> assume all pointers are unaliased.

High-performance compilers will have to have *SOME WAY* to find out about
aliasing or lack thereof.  The obvious way is to have some sort of magic
cookie in the source to selectively indicate when promises are made about
aliasing.  A compile-time option that covers a whole file isn't really a
good idea, considering how pervasive pointers are in C and how common
aliasing is.  The obvious type of cookie is #pragma.  (It is claimed,
probably correctly, that #pragma technically is not allowed to change
the semantics of the language, but that's not the way it will happen
in the real world.)
-- 
NASA is to spaceflight as            |  Henry Spencer @ U of Toronto Zoology
the Post Office is to mail.          | {ihnp4,decvax,uunet!mnetor}!utzoo!henry