Path: utzoo!mnetor!uunet!lll-winken!lll-tis!ames!amdahl!nsc!taux01!orr
From: orr@taux01.UUCP (Orr Michael )
Newsgroups: comp.lang.pascal
Subject: Re:  Global variables
Message-ID: <539@taux01.UUCP>
Date: 30 Mar 88 16:02:49 GMT
References: <12654@brl-adm.ARPA> <529@taux01.UUCP> <4770@aw.sei.cmu.edu>
Reply-To: orr@taux01.UUCP (Orr Michael )
Organization: National Semiconductor (Israel) Ltd.
Lines: 46
Keywords: gloabl optimizing compilers aliasing

In article <4770@aw.sei.cmu.edu> firth@bd.sei.cmu.edu.UUCP (Robert Firth) writes:
>In article <529@taux01.UUCP> orr@taux01.UUCP (Orr Michael ) writes:
>>As one of a team of compiler-writers, including a pascal compiler,
>>let me note that global variables reduce the compiler's room for
>>optimization. this is because any time you erite through a pointer,
>>potentially you may be changing the value of any global var...
>
>I don't think that's right.  For example, in Pascal, an explicit
>pointer (^Thing) cannot point anywhere except into the heap, ie
>at an object created by NEW.  It cannot alias any global variable.
>
>That leaves only VAR parameters as potential aliases.  Even here,
>a VAR parameter of one type (VAR p: ThingA) cannot alias a global
>variable of any other type (VAR g: ThingB).
>
>The extent to which this matters depends on people's coding style,
>but with good Pascal programs (lots of different types) the amount
>of possible aliasing is less than 5% (ie for any pointer you know
>it cannot alias 95% or more of globals in scope).
	
	You are right, but not entirely right. While it's true that
    in STANDARD pascal a pointer can only point at objects on the heap,
    (i.e. objects you get by NEW), in the CTP compilers, there exists
    an extension allowing you to use 'addr_of(object)'. Moreover, 
    some programmers (and let's not name names ;-) sometimes
    use a variant-record to explicitly manipulate the value of a
    pointer, making it possible to get the pointer to point
    ANYWHERE at all. since this practice is hard to guard against,
    usually you must be pessimistic.
    I would like to make two additional points :
    1. I was talking about global variables in general, not only in
       PASCAL. (in the CTP compiler suite, by the time the optimizer
       sees the code, it has no way of telling that it was originally
       written in PASCAL, as the front-end has transformed it into
       an intermediate representation. Ditto for the code-generator)
    2. There are other good reasons to avoid global variables, all well
       known. (mostly SE reasons). I was only trying to point out a
       relatively unknown disadvantage of global vars. I would guess
       that the reasons for the origianl objections cited, is not the
       one I raised, at least not mainly.


-- 
orr@nsc.NSC.COM		  	      IBM's motto: Machines should work,
{amdhal,decwrl,hplabs,pyramid,sun}!nsc!orr          People should think.
<CTP project, NSTA, Israel (972)52-522259>     Orr's remark: Neither do.
Disclaimer: Opinions, come home. All is forgiven. Papa.