Re: Factories and other functions returning pointers

Hi Kris,

Could your factory function simply return a std::shared_ptr? I pasted a
code example showing specifics in the github issue you raised.  I tested it
with your example code and all worked well. It has a couple of benefits
since you'd no longer need to use the %factory and %newobject directives
and using std::make_shared is more efficient than other methods since
allocation of the smart pointer control block can occur with the allocation
of the object.

Burlen

On Fri, Dec 3, 2021 at 1:20 AM <[email protected]> wrote:

> Hi all
>
>
>
> Seems I mailed to soon. This morning my strategy does seem to work if I
> add 2 factory declarations:
>
>
>
> %factory(A *, A1, A2);
>
> %factory(A const*, A1, A2);
>
>
>
> I also mentioned worries about ownership, but I now think that this is
> independent, and I just have to use %newobject for the appropriate cases.
>
>
>
> So I think I have resolved my problems for bare pointers. Still struggling
> with shared_ptr, but I'll document that in the GitHub issue %factory
> fails with %shared_ptr · Issue #2110 · swig/swig (github.com)
> <https://github.com/swig/swig/issues/2110>
>
>
>
> Kris
>
>
>
> *From:* Kris Thielemans>
> *Sent:* 02 December 2021 22:21
>
>
>
> Hi all
>
>
>
>
>
>
>
> I'm trying to make our Python interface to STIR friendlier w.r.t. derived
> classes. Many of our C++ functions return a bare pointer to the base class.
>
>
>
>
>
>
>
> class A;
>
>
>
> class A1: public A;
>
>
>
> class A2: public A;
>
>
>
>
>
>
>
> class B
>
>
>
> {
>
>
>
>   A const * some_function() const;
>
>
>
> };
>
>
>
>
>
>
>
> It'd be great if in Python, the result of B::some_function() would be
> actually of the appropriate type (e.g. A1), as there is no way to downcast
> in Python.
>
>
>
>
>
>
>
> I looked in the use of %factory. This is intended for "factory" functions
> that create a new object. It then occurred to me that nothing in the
> %factory mechanism seems to require that the result is a new object.
> Indeed, using %factory for some_function() seems to work perfectly fine
> (although I'm wondering about ownership).
>
>
>
>
>
>
>
> %factory(A const* some_function, A1, A2);
>
>
>
>
>
>
>
> This generates
>
>
>
>
>
>
>
> typemap(out) A const* some_function{ … }
>
>
>
>
>
>
>
> which handles the downcasting. Great! So then I hoped to just do this for
> all functions that return an A* (or A const*).
>
>
>
>
>
>
>
> %factory(A const*, A1, A2);
>
>
>
>
>
>
>
> Generating
>
>
>
>
>
>
>
> typemap(out) A const* { … }
>
>
>
>
>
>
>
> However, this typemap doesn't match the signature of the
> B::some_function(). Am I doing something stupid? (This is my first foray
> into typemaps, so the answer is probably "yes").
>
>
>
>
>
>
>
> Any help appreciated.
>
>
>
>
>
>
>
> Thanks!
>
>
>
>
>
>
>
> Kris
>
>
>
>
>
>
>
> PS: in fact, most of our functions return a shared_ptr<const A> or even
> unique_ptr, but I have to work around those as SWIG doesn't handle
> unique_ptr yet unfortunately. See %factory fails with %shared_ptr · Issue
> #2110 · swig/swig (github.com) for more information on my struggles with
> shared_ptr.
>
>
>
>
>
>
>
>
> _______________________________________________
> Swig-user mailing list
> [email protected]
> https://lists.sourceforge.net/lists/listinfo/swig-user
>

_______________________________________________
Swig-user mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/swig-user