Right you are. With regards to my original snippet: template<typename Attrib> struct typed_attrib_topython { static PyObject* convert(const Attrib& a) { return bp::incref(bp::object(a.value()).ptr()); // return POD type } }; template<typename T> struct TypedAttribBind { typedef TypedAttribute<T> attrib_t; TypedAttribBind() { boost::python::to_python_converter<attrib_t, typed_attrib_topython<attrib_t> >(); } }; TypedAttribute<T> derives from Attribute, which derives from Object (the common polymorphic base class). TypedAttribute<T>.value() is a function which returns the POD type associated with the attribute. TypedAttribBind<T> is instantiated for a range of POD types T, in order to register the to-python converters. This doesn't behave as expected - if I call any bound function which returns a shared_ptr<Object> from python (where the object is actually a TypedAttribute<T>), I just get a wrapped Object back, rather than a POD type. On Mon, Dec 13, 2010 at 10:04 AM, Dave Abrahams <dave@boostpro.com> wrote:
At Mon, 13 Dec 2010 09:57:59 +1100, Allan Johns wrote:
Sorry I don't think I was being clear enough.
I want to have the derived classes to-python convert into basic types
(float
,string etc). In other words, I don't actually want TypedAttribute<T> to be directly bound at all.
Oh, I see.
Background: I have a 'table' class that can hold various types of attributes (all derived from the same base class), a subset of which are just wrapper classes for POD types. I want these POD-based attributes to automatically to-python convert to the relevant types.
I actually have this working at the moment but not by way of boost.python converters... I have a hand-rolled dispatcher going on (think std::map keyed on std::type_info), that my table class's python binding uses directly. But I'd rather not be duplicating boost.python functionality, if that's what I'm doing (it would seem more elegant, and more flexible, to do this with boost.python itself).
You might start by answering the question of in which way what you did fails to work.
-- Dave Abrahams BoostPro Computing http://www.boostpro.com
_______________________________________________ Unsubscribe & other changes: http://lists.boost.org/mailman/listinfo.cgi/boost