From -178717130743094222
X-Google-Language: ENGLISH,ASCII-7-bit
X-Google-Thread: f78e5,e94695845965b249
X-Google-Attributes: gidf78e5,public
X-Google-ArrivalTime: 1992-12-22 00:07:10 PST
Newsgroups: comp.std.c++
Path: sparky!uunet!munnari.oz.au!metro!extro.ucc.su.OZ.AU!maxtal
From: maxtal@extro.ucc.su.OZ.AU (John MAX Skaller)
Subject: Re: Zero-length structures and pointer comparisons
Message-ID: <1992Dec22.002957.25597@ucc.su.OZ.AU>
Sender: news@ucc.su.OZ.AU
Nntp-Posting-Host: extro.ucc.su.oz.au
Organization: MAXTAL P/L C/- University Computing Centre, Sydney
References: <1992Dec16.201044.2968@microsoft.com> <1992Dec18.193554.18588@ucc.su.OZ.AU> <1gtblgINN72@darkstar.UCSC.EDU>
Date: Tue, 22 Dec 1992 00:29:57 GMT
Lines: 71

In article <1gtblgINN72@darkstar.UCSC.EDU> daniel@cse.ucsc.edu (Daniel R. Edelson) writes:
>
>>	For this reason I think the ARM requirement is unworkable,
>>and must be modified. The rule should state that pointers
>>derived from the same object directly in the program must
>>compare equal. Pointers obtained from the OS need not follow
>>this rule, in that case it is implementation defined.
>
>Neither the C standard nor the C++ standard discusses any OS.
>(As far as I recall.)

	Precisely, nor should they, instead, they ought discuss
the conditions under which the rule is true, otherwise
it is implementation defined.
>
>A call to the operating system should be viewed as
>a call to an extension supplied by the language provider. 
>A program that uses an extension can be conforming but
>not strictly conforming, as per the X3J16 definitions. 

	Sure. But my interpretation of the p==q rule
is that any implementation in which it isnt true
is *non*-conforming. That is, there is no scope in the rule
for non-strict conformance.

>
>Thus, I think a better rule is:
>	``In a strictly conforming program, two pointers
>	to the same object compare equal.''
>Thus, this property need not be true in any program that
>calls the operating system.

	That is in the spirit of what I wanted, but I'm
not sure if a standard can say it in those words. Besides,
I wanted to be more explicit: even if I do an operting
system call somewhere, I want the system
to guarrantee

	T* p=new T;
	T* q=p;
	if(!(p==q))cout<"Non conforming program";

But I want that

	q=alias(p); // OS call to alias pointer
	if(!(p==q))cout<"Not a strictly conforming program";

(Given that p and q point to the same object after aliasing).
In particular I dont want the above code to render the whole
program non-conforming.	
>
>I think actually that ramifications of multiple inheritance 
>and multiple object-addresses should be explicitly addressed
>in the rule, such as by saying that a pointer to a base-class
>subobject is not a pointer to the same object as a pointer
>to the most-derived class object.
>
	The rule only applies when p and q are the same type.
If a conversion from a derived to base type occurs a comparison
would quite correctly yield 1 if the base subobject was a
sub-object of the complete object.

	The other issue is whether empty subobjects are generated
or not, if they can be, pointers to different subobjects of the
same type might compare equal.

-- 
;----------------------------------------------------------------------
        JOHN (MAX) SKALLER,         maxtal@extro.ucc.su.oz.au
	Maxtal Pty Ltd, 6 MacKay St ASHFIELD, NSW 2131, AUSTRALIA
;--------------- SCIENTIFIC AND ENGINEERING SOFTWARE ------------------