From owner-freebsd-bugs  Sun Nov 12 20:14:05 1995
Return-Path: owner-bugs
Received: (from root@localhost)
          by freefall.freebsd.org (8.6.12/8.6.6) id UAA09493
          for bugs-outgoing; Sun, 12 Nov 1995 20:14:05 -0800
Received: from godzilla.zeta.org.au (godzilla.zeta.org.au [203.2.228.19])
          by freefall.freebsd.org (8.6.12/8.6.6) with ESMTP id UAA09478
          for <freebsd-bugs@freebsd.org>; Sun, 12 Nov 1995 20:13:59 -0800
Received: (from bde@localhost) by godzilla.zeta.org.au (8.6.9/8.6.9) id PAA04625; Mon, 13 Nov 1995 15:09:55 +1100
Date: Mon, 13 Nov 1995 15:09:55 +1100
From: Bruce Evans <bde@zeta.org.au>
Message-Id: <199511130409.PAA04625@godzilla.zeta.org.au>
To: freebsd-bugs@freebsd.org, tege@matematik.su.se
Subject: Re: strtouq broken in FreeBSD 2.0.5
Sender: owner-bugs@freebsd.org
Precedence: bulk

>This program demonstrates that strtouq doesn't work right.
>As a matter of fact, it seems to have more than one problem, since
>other things fail when a non-zero base argument is passed.

>The author of strtouq should learn to test his/her code.

>#include <sys/types.h>
>#include <stdlib.h>
>#include <limits.h>
>main ()
>{
>  unsigned long long rs = strtouq ("12345678910", 0, 0);
>  printf ("%qx\n", rs);
>}

This seems to be a bug in gcc.  Doesn't the author of gcc test his/her
code? :-)

main ()
{
  unsigned long qbase = 10;
  unsigned long long cutoff = ((unsigned long long)0 - 1)  / qbase;
  printf ("%qd\n", cutoff);
}

Output from `gcc -O -S z.c' under FreeBSD.

	.file	"z.c"
gcc2_compiled.:
___gnu_compiled_c:
.text
LC0:
	.ascii "%qd\12\0"
	.align 2
.globl _main
	.type	 _main,@function
_main:
	pushl %ebp
	movl %esp,%ebp
	call ___main
	pushl $0
	pushl $10
	pushl $0		// <== should be -1
	pushl $-1
	call ___udivdi3
	addl $16,%esp
	pushl %edx
	pushl %eax
	pushl $LC0
	call _printf
	leave
	ret
Lfe1:
	.size	 _main,Lfe1-_main

This is for gcc-2.6.3 under FreeBSD.  gcc-2.5.8 under Linux and gcc-2.7.0
under FreeBSD produce the correct code.

I think the complicated definition of UQUAD_MAX is to work around bugs
in older versions of gcc.  But <machine/limits.h> shouldn't depend on
typedefs or even use casts.

Bruce