[Python-Dev] GC and ExtensionClass - a summary of the problem and a workaround

[M.-A. Lemburg]

skip@pobox.com wrote:
> 
>     mal> I wonder how the GC collector could "see" the type object before it
>     mal> has been initialized... since PyGtkTreeSortable_Type is a static C
>     mal> array and not a known PyObject until you add it to some Python
>     mal> dictionary as type object or use it for creating instances, it
>     mal> seems strange that the GC collector can reach out for it and get
>     mal> hit by the fact that it is not yet properly initialized.
> 
> It is actually PyGtkWidget_Type that is not yet initialized when it is
> placed in the bases tuple for one of its subclasses.  GC traverses that
> tuple, then dives into each element.  It hits the PyGtkWidget_Type object,
> whose ob_type field has not yet been initialized.  The actual object whose
> bases tuple is being traversed is (in all the crashes I encountered),
> GdkDragContext.  The ordering of the registration calls could perhaps be
> reordered.  Currently GdkDragContext is patched up before GtkWidget, its
> base class.  This code is generated by James Henstridge's wrapper code
> generator, so perhaps he can maintain the necessary class hierarchy
> relationships and insure that base classes are initialized before their
> subclasses.

Wouldn't it be easier to simply set the ob_type fields right at the
start of the initGtk() function ? This is what I do for all
my extensions and I've never seen any problems with it.

-- 
Marc-Andre Lemburg
______________________________________________________________________
Company & Consulting:                           http://www.egenix.com/
Python Software:                        http://www.lemburg.com/python/