Re: [boost] Problem with shared_ptr

From: Greg Colvin (gcolvin_at_[hidden])
Date: 2001-05-16 10:53:17

From: Beman Dawes <bdawes_at_[hidden]>
> At 03:29 AM 5/16/2001, Thomas Maeder wrote:
> 
>  >This problem report is a result of a discussion in
>  >news://newsgroups.borland.com/borland.public.cppbuilder.language
>  >
>  >The following program demonstrates it, using boost 1.21.1 and gcc 2.95.3:
>  >
>  >#include <iostream>
>  >
>  >#include "boost/smart_ptr.hpp"
>  >
>  >struct S
>  >{
>  >    ~S()
>  >    {
>  >      std::cout << "~S()\n";
>  >    }
>  >
>  >    boost::shared_ptr<S> next;
>  >};
>  >
>  >int main()
>  >{
>  >  boost::shared_ptr<S> head(new S);
>  >  head->next.reset(new S);
>  >  std::cout << "before assignment\n";
>  >  head = head->next;
>  >  std::cout << "after assignment\n";
>  >}
>  >
>  >The output is
>  >
>  >before assginment
>  >~S()
>  >~S()
>  >after assginment
> 
> Ouch!

Amen.

>  >while one would expect it to be
>  >
>  >before assignment
>  >~S()
>  >after assignment
>  >~S()
>  >
>  >The reason is that the shared_ptr copy-assignment operator first reduces
>  >the reference count of the previously owned object and only then 
> increases
>  >that of the newly owned object; if both objects are the same, strange
>  >things happen.
> 
> It isn't really a case of both objects being the same, but that the right 
> hand object is transitively dependent on the left hand object.
> 
> The code causing the problem is:
> 
>     void share(T* rpx, long* rpn) {
>        if (pn != rpn) { // Q: why not px != rpx? A: fails when both == 0
>           dispose();
>           px = rpx;
>           ++*(pn = rpn);
>        }
> 
> My first reaction is to change it to:
> 
>     void share(T* rpx, long* rpn) {
>        if (pn != rpn) { // Q: why not px != rpx? A: fails when both == 0
>           ++*rpn; // done before dispose() in case rpn transitively
>                   // dependent on *this (bug reported by Thomas Maeder)
>           dispose();
>           px = rpx;
>           pn = rpn;
>        }

This looks right to me.

 
> I'll have give it a bit more thought.  Greg?  Also do some testing.  The 
> test program needs updating anyhow.
>
> Thanks for the bug report.
> 
> --Beman
> 
> 
> To unsubscribe, send email to: <mailto:boost-unsubscribe_at_[hidden]>
>  
> 
> Your use of Yahoo! Groups is subject to http://docs.yahoo.com/info/terms/ 
> 
> 
>