[C++-sig] Bug and patch for boost.python withenable_shared_from_this

[Nicolas Lelong]

Hi,

> Could you at least ask Peter Dimov if he thinks the
> dont_enable_shared_from_this patch could be generally useful?
> Maybe he's OK adding it, then it wouldn't be a big deal
> fixing Boost.Python, which would probably prevent a lot of
> confusion and lost time in the future.

I stumbled once again on this problem, so I took some time to bother Peter Dimov with the patch suggestion. Here, is a copy of our exchange :
  Hello Peter,

  I wanted to contact you about a problem that was raised, on february, on c++-sig mailing list about a problem happening when wrapping classes derived from 'enable_shared_from_this' using boost.python library.

  The problem is briefly described in this post - including a patch to boost/shared_ptr.hpp that would allow to fix the problem.
  http://mail.python.org/pipermail/cplusplus-sig/2008-February/012973.html

  The original poster was encouraged by Ralf W. Grosse-Kunstleve to contact you about this, but we never heard from him on the list (http://mail.python.org/pipermail/cplusplus-sig/2008-February/013003.html).

  To make it short, boost.python creates shared_ptr objects, holding the wrapped c++ objects, with a custom deleter managing the python object reference count. This leads to something like that :

  #include <boost/enable_shared_from_this.hpp>
  namespace {
  class A : public boost::enable_shared_from_this<A>
    {
    public:
      ~A() {};
    };
    void my_deleter(void*)
    {
    }
  };
  BOOST_AUTO_TEST_CASE( test_enable_shared_from_this )
  {
    boost::shared_ptr<A> a( new A );
    {
      boost::shared_ptr<A> ater = boost::shared_ptr<A>( a.get(), my_deleter );
      // OP patch proposal :: boost::shared_ptr<A> ater = boost::shared_ptr<A>( a.get(), my_deleter
      //                        , boost::dont_enable_shared_from_this() );
      BOOST_CHECK( a == ater );
    }
    boost::shared_ptr<A> abug = a->shared_from_this(); // this throws bad_weak_ptr
  }

  Did you hear about this problem ? Do you think this patch is viable and could be applied in boost trunk ? Do you happen to have another idea of workaround that could be applied in client code and/or in shared_ptr ?

  Thanks for your time,
  best regards,

  Nicolas Lelong
  mgdesign
Followed by Peter's answer that lead to Trac ticket #2584 :
  Yes, I am aware of this problem. It is possible today to emulate the 
  proposed functionality by using the aliasing constructor, but I think that a 
  more proper fix for the issue is to make enable_shared_from_this more 
  intelligent so that it doesn't reinitialize itself in such scenarios.

  I'll appreciate it if you file a Trac ticket for this and assign it to me. 
  I've a few pending enable_shared_from_this issues anyway, and I'll try to 
  find the time to address them before the next release ships.

  --
  Peter Dimov
  http://www.pdplayer.com 
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.python.org/pipermail/cplusplus-sig/attachments/20081212/4c26663e/attachment-0001.htm>