Re: [Swig-user] Reference counting in Python callbacks?

[Magnus L. H. <ma...@he...>]

On Apr 18, 2006, at 22:27, Marcelo Matus wrote:

> Magnus Lie Hetland wrote:
[snip]
>> The problem isn't that Python does a GC on it -- the problem is  
>> that  the C++ code removes a reference (which *should* be  
>> removed), and  thereby it's deallocated while it's still available  
>> in Python.
>
>
> well, that is what the DISOWN typemap tells to python, "don't try  
> to deallocate the object since you don't own it anymore",

But... My point is that what I want is exactly the opposite: I *want*  
Python to own it (i.e., to add a reference to it), because my C++  
data structure will be deallocated and therefore will remove *its*  
references. If Python hasn't added one, the objects will be  
deallocated by C++, and Python will try to keep using them.

I'm curious: Shouldn't the existence of a proxy (and/or a director)  
object ensure that the refcount is at least 1 (as long as the C++  
doesn't mess with it)?

> and C++ will take care of it.
>
> You can also use the %exception trick, ie;
>
> %exception Result::add() {
>     AddRef(arg1);  // or so
>     $action
> }

Hm. Yes.

But when are the _wrap_* functions called, then -- i.e. those that  
call the add_ref stuff? When I return objects? (With or without % 
newobject?)

> or %extend your method as:
>
> %ignore Result::add;
> %extend {
>     void add(Class *value) {
>             AddRef(value);
>             self->add(value);
>     }
> }

Sure. Extending the method probably isn't all that hard. I was just  
hoping it wouldn't be necessary. I guess I don't understand how SWIG  
handles the reference counting yet (i.e. when the ref-part of the  
feature is actually used).

[snip]
>> BTW: Where is the authoritative documentation for the DISOWN  
>> typemap?  I can just find brief mentions of it here and there  
>> (and, e.g., an  explanation in the Chicken section).
>
> hmm, try Customization.html, if it is not there, we welcome new docs!.

Yeah, I've looked there before. All I found was:

"This is similar to use the DISOWN typemap in the first method  
argument, and in fact, it also depends on the target language on  
implementing the 'disown' mechanism properly."

I guess I could contribute some docs -- if/when I understand this  
stuff ;)

--
Magnus Lie Hetland
http://hetland.org