From owner-freebsd-bugs  Thu May 17  5:34:19 2001
Delivered-To: freebsd-bugs@freebsd.org
Received: from core.inf.ethz.ch (core.inf.ethz.ch [129.132.178.196])
	by hub.freebsd.org (Postfix) with ESMTP id 9EAC837B440
	for <bugs@freebsd.org>; Thu, 17 May 2001 05:34:09 -0700 (PDT)
	(envelope-from cartho@silmaril.org)
Received: from lillian-ics.inf.ethz.ch (root@lillian-ics.inf.ethz.ch [129.132.134.2])
	by core.inf.ethz.ch (8.9.3/8.9.3) with ESMTP id OAA27721;
	Thu, 17 May 2001 14:34:04 +0200 (MET DST)
Received: from silmaril.org (babylon8.inf.ethz.ch [129.132.134.131])
	by lillian-ics.inf.ethz.ch (8.9.3/8.9.3) with ESMTP id OAA04001;
	Thu, 17 May 2001 14:34:06 +0200 (MET DST)
Message-ID: <3B03C53C.349F9491@silmaril.org>
Date: Thu, 17 May 2001 14:34:04 +0200
From: Cyrille Artho <cartho@silmaril.org>
Organization: ETH Zurich
X-Mailer: Mozilla 4.76 [en] (X11; U; Linux 2.2.19pre17 i686)
X-Accept-Language: en
MIME-Version: 1.0
To: bugs@freebsd.org
Cc: Ernst de Haan <ernst@jollem.com>
Subject: Insufficient type casts in stdargs.h
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit
Sender: owner-freebsd-bugs@FreeBSD.ORG
Precedence: bulk
X-Loop: FreeBSD.org

Hi,
in FreeBSD 4.3's stdargs.h, the macro va_arg does not contain enough
type casts. As a result, gcc 2.95.3 rejects using va_arg with void*
arguments.
The GNU stdargs.h contains extra type casts, allowing this. I rewrote
the BSD to include extra type casts, fixing now problems with void*
va_arg arguments:
Original macro in /usr/include/stdarg.h, line 53:
#define va_arg(ap, type) \
        (*(type *)((ap) += __va_size(type), (ap) - __va_size(type)))

My suggestion:
#define va_arg(ap, type) \
        (*(type *)((ap) = (type *)((char *)ap + __va_size(type)), \
	(type *)(((char *)ap) - __va_size(type))))

This adds three extra type casts, two intermediate ones to char* and one
more to type*. Moreover, I had to write out the += operator as it is not
allowed with void* anymore.
Please review this patch carefully, as it is a critical header file, and
not very easy to read - so I could have made an error, or there may be a
better way to fix the problem.

Originally, this compiler problem was discovered by Ernst de Haan, when
he tried to compile my Java program checker Jlint
(http://artho.com/jlint/). I include the full information about the line
of code that did not compile below:

Original source code:

parameter[n_parameters++] = va_arg(ap, void*);

-------------------------------------------------------------------------

Linux little endian systems:

#define va_arg(AP, TYPE)                                               
\
 (AP = (__gnuc_va_list) ((char *) (AP) + __va_rounded_size (TYPE)),    
\
  *((TYPE *) (void *) ((char *) (AP) - __va_rounded_size (TYPE))))

Expanded source code:
parameter[n_parameters++] = ( ap  = (__gnuc_va_list) ((char *) ( ap ) +
(((sizeof (   void*  ) + sizeof (int) - 1) / sizeof (int)) * sizeof
(int)) ),       *((  void*  *) (void *) ((char *) ( ap ) - (((sizeof (  
void*  ) + sizeof (int) - 1) / sizeof (int)) * sizeof (int)) ))) ;

-------------------------------------------------------------------------

FreeBSD:

#define va_arg(ap, type) \
        (*(type *)((ap) += __va_size(type), (ap) - __va_size(type)))

Expanded source code:
parameter[n_parameters++] = (*(  void*  *)(( ap ) += (((sizeof(   void* 
) + sizeof(int) - 1) /  sizeof(int)) * sizeof(int)) , ( ap ) -
(((sizeof(   void*  ) + sizeof(int) - 1) / sizeof(int)) * sizeof(int))
))

jlint.cc:141: ANSI C++ forbids using pointer of type `void *' in
arithmetic
jlint.cc:141: ANSI C++ forbids using pointer of type `void *' in
arithmetic

Provisional fix:

#define va_arg(ap, type) \
        (*(type *)((ap) = (type *)((char *)ap + __va_size(type)), \
	(type *)(((char *)ap) - __va_size(type))))

-- 
Regards,
Cyrille Artho - http://artho.com/ - Tel. +41 - [0]1 - 632 09 88
If you go on with this nuclear arms race, all you are going to do is
make the rubble bounce.         
                -- Winston Churchill

To Unsubscribe: send mail to majordomo@FreeBSD.org
with "unsubscribe freebsd-bugs" in the body of the message