[C++-sig] Ownership of a C++ object extended in Python using wrapper

[Alexandre Hamez]

On 16 mars 2010, at 22:21, Jim Bosch wrote:

> On Tue, 2010-03-16 at 20:56 +0100, Alexandre Hamez wrote:
>> On 16 mars 2010, at 13:12, Neal Becker wrote:
>> 
>>> Alexandre Hamez wrote:
>>> 
>>>> Hi all,
>>>> 
>>>> I have a C++ library that needs to store in a hash map user's code, that
>>>> is, Python-extended objects. To make the C++ interface available to
>>>> Python, I use boost::python::wrapper, as stated in the tutorial
>>>> 
>>> (http://www.boost.org/doc/libs/1_42_0/libs/python/doc/tutorial/doc/html/python/exposing.html#python.class_virtual_functions).
>>>> Because I store this extend objects, I need to tell Python that I've got a
>>>> reference on these objects.
>>>> A thread from 2007
>>>> (http://mail.python.org/pipermail/cplusplus-sig/2007-March/011790.html)
>>>> states there is a problem for this particular use case, but in the end, no
>>>> solution is given. So my question is the following: as of today, what is
>>>> the correct way to  tell Python we hold a reference to Python extend
>>>> objects of C++ classes, using the boost::python::wrapper mechanism?
>>>> 
>>>> Thanks,
>>>> ---------------------
>>>> Alexandre Hamez
>>> 
>> 
>> Thanks for answering so quickly.
>>> I've use 2 methods:
>>> 
>>> 1) with_custodian_and_ward
>> OK, I didn't think of this method. Seems a good way to do it.
>> 
>>> 2) boost::shared_ptr
>> 
>> For this point, do you mean you just take the objects that should seen they references increased by a shared_ptr? Just after having asked this question, I tried with this simple solution, and it seems to work. I hope there is no hidden potential crash :-)
>> 
> 
> If you get a boost::shared_ptr to a Python object (either by wrapping a
> function taking a shared_ptr or by using boost::python::extract), that
> shared_ptr will properly manage the Python reference count via a custom
> deleter that decref's the PyObject* when it goes out of scope.  The only
> catch is that shared_ptr doesn't do garbage collection, so you can get
> cycles that don't get deleted if your shared_ptrs own other shared_ptrs.
> Other than that, you're should be safe if you just use shared_ptr
> everywhere.
> 
> This will work differently internally if you declare your wrapped
> classes to have shared_ptr storage (that's generally what I do), but the
> effects should be mostly the same.  Note that you can also declare your
> storage to be shared_ptr to a wrapper class.
> 
>> Do you use these two methods conjointly in the same project? Or did you identify some use cases which favor a method rather the other one?
>> 
> 
> I generally use shared_ptr storage for everything, but I often mix that
> with with_custodian_and_ward and its cousins (return_internal_reference,
> in particular) when I don't have access to a shared_ptr on the C++ side.
> 
> 
> Jim Bosch
> 
> _______________________________________________
> Cplusplus-sig mailing list
> Cplusplus-sig at python.org
> http://mail.python.org/mailman/listinfo/cplusplus-sig


OK, thank you very much for your answer!

---------------------
Alexandre Hamez