Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Path: utzoo!mnetor!uunet!seismo!mimsy!oddjob!gargoyle!jpusa1!news
From: news@jpusa1.UUCP (usenet)
Newsgroups: comp.unix.xenix
Subject: SCO Xenix 'C' pointer bug
Message-ID: <319@jpusa1.UUCP>
Date: Fri, 28-Aug-87 19:44:10 EDT
Article-I.D.: jpusa1.319
Posted: Fri Aug 28 19:44:10 1987
Date-Received: Sun, 30-Aug-87 06:05:39 EDT
References: <227@krebs.acc.virginia.edu>
Reply-To: stu@jpusa1.UUCP (Stu Heiss)
Followup-To: comp.unix.xenix
Organization: JPUSA - Chicago, IL
Lines: 46

Summary:

Expires:


In article <227@krebs.acc.virginia.edu> wrp@krebs.acc.virginia.edu (Wm Pearson) writes:
-
-	Attached is a test program which displays a bug I just found
-in the SCO 'C' development system 2.1.4 release G.  Compile the
-program and run it.  You will find the value of 'dpos' goes up to 3999
-just fine, but then becomes negative.  I would appreciate a work-around
-that does not require a division by sizeof(struct tstruct). Note
-that although the pointers are being treated as ints (intead of unsigned)
-in the expression (int)(dptr-diag), they are being treated properly in
-the 'dptr<&diag[5000].
-
-============= test.c
-Bill Pearson
-wrp@virginia.BITNET
-...!seismo!virginia!wrp
-
-
-#include <stdio.h>
-
-struct tstruct {
-	int one;
-	int two;
-	int three;
-	int four;
-	} *diag, *dptr;
-
-main()
-{
-	char *calloc();
-	unsigned int dpos;
-	
-	diag = (struct tstruct *)calloc(5000,sizeof(struct tstruct));
-	for (dptr=&diag[99]; dptr<&diag[5000]; dptr +=100) {
-		dpos = (unsigned int)(dptr-diag);
-		printf("%4d %4d\n",dpos,(int)(dptr-diag));
-	}
-}
-=============
The problem seems to be incorect pointer arithmetic and I don't see any
work around.  When the expression '(char *)dptr-(char *)diag' becomes
greater than 0x7fff, there is sign extension(!!!).  This is before the
scaling of sizeof(tstruct).  Use a macro like
#define pos(p,base,type) ((unsigned)((char *)p-(char *)base)/sizeof(type))
and 'dpos = pos(dptr,diag,struct tstruct);'

Stu Heiss {gargoyle,ihnp4}!jpusa1!stu