Re: [Swig-user] Reference counting

[Josh C. <jc...@nc...>]

On Tue, 14 Aug 2007, David Piepgrass wrote:

> There have been a couple of threads on this before but surprisingly it
> hasn't gone anywhere.
>
> "Questions about Smart Pointer support in SWIG"
> http://article.gmane.org/gmane.comp.programming.swig/3832/match=referenc
> e+count
> "Back to the reference counting problem... and the "derivability"
> option"
> http://article.gmane.org/gmane.comp.programming.swig/276/match=back+refe
> rence+counting
>
> I need some help to get reference counting working. In my C++ code I use
> smart pointers, but I don't want SWIG to produce bloated results by
> duplicating the original classes in the smart pointer classes.
>
> First of all, SWIG has a built-in memory management system involving the
> concept of "ownership". I don't know how that would interact with
> reference counting, so how do I disable it for a ref-counted class (so
> that all instances of the wrapper act like they are the owner?)
>
> I'd like to do reference counting in a "generic" way if possible,
> without rewriting typemaps for every reference-counted class. So, let's
> say I want to wrap this class:

I have a system for dealing with this.  It took some effort to set up, but 
it works.  What I do for the actual objects (and regular pointers and 
references to them) is described at 
http://article.gmane.org/gmane.comp.programming.swig/7851.  For smart 
pointers, I provide "out" typemaps that make a regular pointer proxy and 
adjust the reference count appropriately.  I also provide "in" typemaps 
and typechecks that accept a regular pointer.  These typemaps go in a 
smart pointer template declaration, so they only need to appear once.  I 
use %template() with empty parentheses to apply the typemaps without 
making full wrappers for the smart pointer classes.  Users see only 
objects (not smart pointers), but the objects magically have 
reference-counted semantics.

Josh