NetNews Offline 2/NetNews Offline Volume 2.iso/news/comp/lang/c-part2/14254

Internet Message Format | 1996-08-05 | 1.8 KB

Path: mail2news.demon.co.uk!genesis.demon.co.uk
From: Lawrence Kirby <[e-mail]>
Newsgroups: comp.lang.c
Subject: Re: fastest code
Date: Fri, 12 Apr 96 16:31:32 GMT
Organization: none
Message-ID: <[e-mail]>
References: <[e-mail]> <4kghs7$[e-mail]> <4kgu7g$[e-mail]> <[e-mail]> <4kljb4$[e-mail]>
Reply-To: [e-mail]
X-NNTP-Posting-Host: genesis.demon.co.uk
X-Newsreader: Demon Internet Simple News v1.27
X-Mail2News-Path: genesis.demon.co.uk

In article <4kljb4$[e-mail]>
           [e-mail] "Peter Seebach" writes:

>In article <[e-mail]>,
>Gareth Rees <[e-mail]> wrote:
>>Surely, under the `as if' rule, code that invokes undefined behaviour
>>*at any point* need not produce the results you expect even for those
>>portions that don't invoke undefined behaviour?
>
>True, although in practice it doesn't come up.
>
>>So in the case
>
>>    int x[17], i;
>>    for (i=0; i<16; i++) x[i] = x[i+i];
>
>>the compiler is perfectly at liberty to optimise away the loop body
>>because of the undefined behaviour of the whole program.

Yes, it is undefined by definition from ANSI/ISO 3.16 because the code uses
the value of 'indeterminately valued objects' i.e. the elements of x
(assuming x is defined in block scope). If x was, say, static this would be
well defined.

>Right.  As it turns out, x had dimension 32 in the original post, but this
>was not posted along with the loop description.

In the test program posted recently the elements of x were suitably
initialised.

--
-----------------------------------------
Lawrence Kirby | [e-mail]
Wilts, England | [e-mail]
-----------------------------------------