From owner-freebsd-hackers  Sat Oct 19 05:18:26 1996
Return-Path: owner-hackers
Received: (from root@localhost)
          by freefall.freebsd.org (8.7.5/8.7.3) id FAA18794
          for hackers-outgoing; Sat, 19 Oct 1996 05:18:26 -0700 (PDT)
Received: from hps.sso.wdl.lmco.com (hps.sso.wdl.lmco.com [158.186.22.100])
          by freefall.freebsd.org (8.7.5/8.7.3) with SMTP id FAA18784
          for <hackers@FreeBSD.org>; Sat, 19 Oct 1996 05:18:22 -0700 (PDT)
Received: from miles.sso.wdl.lmco.com by hps.sso.wdl.lmco.com (4.1/SSO-4.01-LMCO)
	id AA10760; Sat, 19 Oct 96 08:17:04 EDT
Received: by miles.sso.wdl.lmco.com (4.1/SSO-SUN-2.04)
	id AA14931; Sat, 19 Oct 96 08:15:02 EDT
Date: Sat, 19 Oct 1996 08:15:01 -0400 (EDT)
From: Richard Toren <rpt@sso.wdl.lmco.com>
X-Sender: rpt@miles
To: hackers <hackers@FreeBSD.org>
Subject: Re: C++ question
In-Reply-To: <199610172119.PAA20617@rocky.mt.sri.com>
Message-Id: <Pine.SUN.3.91.961019081145.14917A-100000@miles>
Mime-Version: 1.0
Content-Type: TEXT/PLAIN; charset=US-ASCII
Sender: owner-hackers@FreeBSD.org
X-Loop: FreeBSD.org
Precedence: bulk

Here is another anomoly between g++ and Sun C++ 4.1

Here is one from the obscure pages of the ARM (I hope).

What action is to be taken when an object is passed by value to a
function that uses 'varargs'. Is the copy ctor called? is the dtor
also supposed to be called?

Who did it correctly ??

The following situation was noticed due to a type (rather common) when
attempting to print the address of an object, and the '&' was left
off. Unfortunately, the real object was a reference counted pointer. I
noticed that the Sun 4.1 C++ compiler used the supplied copy ctor, but
never called the dtor. This left the reference count one too high.

Some experimenting with G++ and an old CenterLine compiler showed that
both of these made copies in the stack without calling the copy ctor
(probably just the compilers byte-wise copy) and then passed the
adress rather than the object itself.

Sun C++ 4.1 gave no warnings (even with +w).
CenterLine gave no warnings, but did not call the user copy ctor.
G++ gave a warning 
     (missingDtor.C:39: warning: cannot pass objects 
                        of type `Apple' through `...'

G++ output looks like:
  copyCtorbug - called with 1 args
  empty ctor - refCount=1
  The real address of myfruit is 0xefbfd6b4
  Addr of myfruit is 0xefbfd6ac
  refCount = 1
  dtor - refCount=0

  copyCtorbug - called with 2 args
  empty ctor - refCount=1
  The real address of myfruit is 0xefbfd6ac
  Addr of myfruit is 0xefbfd6ac
  refCount = 1
  dtor - refCount=0

CenterLine gave essentialy the same result.

Sun compiler gave:
  copyCtorbug - called with 1 args
  empty ctor - refCount=1
  The real address of myfruit is 0xeffff1e0
  copy ctor - refCount = 2                   // !!!!!!!!!!
  Addr of myfruit is 0xeffff1d8
  refCount = 2
  dtor - refCount=1                          // wrong ???

  copyCtorbug - called with 2 args
  empty ctor - refCount=1
  The real address of myfruit is 0xeffff1d8
  Addr of myfruit is 0xeffff1d8
  refCount = 1
  dtor - refCount=0


//
// test missing dtor with call to vararg
//

#include <stdio.h>

 int refCount = 0;


class Apple {
    int   somevalue;
public:
    Apple ( ):somevalue(100) {
        refCount ++;
        printf("empty ctor - refCount=%d\n",refCount);
    }
    Apple(const Apple& rs) {
        refCount++;
        printf("copy ctor - refCount=%d\n",refCount);
        somevalue = rs.somevalue;
    }
    ~Apple ( ) {
        refCount--;
        printf("dtor - refCount=%d\n",refCount);
    }
};


main (int argc, char **argv )
{
    printf("\n\n\ncopyCtorbug - called with %d args\n",argc);
    Apple myfruit;
    Apple *mfp;
    
    mfp = &myfruit;
    printf("The real address of myfruit is %#x\n",mfp);

    if (argc==1) {
        printf("Addr of myfruit is %#x\n",  myfruit);
    }else {
        printf("Addr of myfruit is %#x\n", &myfruit);
    }
    printf("refCount = %d\n",refCount);
    return 0;
}



                         ====================================================
Rip Toren               | The bad news is that C++ is not an object-oriented |
rpt@sso.wdl.lmco.com    | programming language. .... The good news is that   |
                        | C++ supports object-oriented programming.          |
                        |    C++ Programming & Fundamental Concepts          |
                        |     by Anderson & Heinze                           |
                         ====================================================