Re: [Boost-bugs] [Boost C++ Libraries] #3080: dynamic_cast returns 0 on polymorphic serialization with weak_ptr

Subject: Re: [Boost-bugs] [Boost C++ Libraries] #3080: dynamic_cast returns 0 on polymorphic serialization with weak_ptr
From: Boost C++ Libraries (noreply_at_[hidden])
Date: 2009-06-07 17:50:29

#3080: dynamic_cast returns 0 on polymorphic serialization with weak_ptr
----------------------------------------------------+-----------------------
  Reporter:  Takatoshi Kondo <kondo_at_[hidden]>  |       Owner:  ramey        
      Type:  Bugs                                   |      Status:  reopened     
 Milestone:  Boost 1.40.0                           |   Component:  serialization
   Version:  Boost 1.39.0                           |    Severity:  Problem      
Resolution:                                         |    Keywords:               
----------------------------------------------------+-----------------------

Comment(by ramey):

 I've been working on this problem and have found no way to fix it in a
 definitive manner. A definitive fix would require enhancements in the
 shared_ptr public
 interface which I don't expect to occur.

 The problem can be addressed in the test case:
 {{{
 struct Sub:public Base1, public Base2 {
     int m_x;
     boost::weak_ptr<Sub> m_wp;     //fails
     //boost::weak_ptr<Base2> m_wp; //OK
     Sub(){}
     Sub(int x) :
         Base1(x),
         Base2(x),
         m_x(x)
     {}
 }}}

 By using only one type of pointer to serialize the same object.
 This problem only occurs when using shared pointer and it's friends.

 I've updated the code to trap the condition and throw an exception
 when it is discovered.

 I've looked at your patch.  First, I'm very impressed that you've
 clearly understood how share_ptr_helper works and what it does. And
 this is without any explanation and documentation!!!.

 At first I was inclined to dismiss it as I had spent a lot of time
 only to conclude that it wasn't fixable.  Looking carefully at your
 code - which looks very similar to my own attempts, now I'm thinking
 you might be right.  So I'm still looking at it.  The problem for
 me is that the shared_ptr operations in the public interface
 are not all that transparent to me.

 As for testing, I would expect that a much simpler test could be made.


 {{{

 class Base {
 ...
 };
 class Derived : public Base {
 ...
 }

 smart_ptr dptr = new Derived;
 smart_ptr bptr = dptr;
 // now there is only one object with pointers of different types
 ...
 ar >> dptr;
 ar >> bptr; // creates problem.
             // dptr and bptr point to the same object


 }}}

 Anyway, I'm going to study your patch in more detail.

 Robert Ramey.

 PS - unfortunately, I had alread conclude that this was not fixable
 and had updated the manual and exception list accordingly.

-- 
Ticket URL: <https://svn.boost.org/trac/boost/ticket/3080#comment:11>
Boost C++ Libraries <http://www.boost.org/>
Boost provides free peer-reviewed portable C++ source libraries.