[Swig-devel] Directors and C++ subclasses in C#/D/Java vs. Python

[David N. <co...@kl...>]

Consider the following piece of code (ignoring the memory leak in 
createBarAndGetId for the sake of simplicity):

---
%module(directors="1") director_weird;

%feature("director") Foo;

%inline %{
struct Foo {
   virtual ~Foo() {}
   virtual int id() {
     return 0;
   }
};

struct Bar : Foo {};

struct Baz : Bar {
   virtual int id() {
     return 2;
   }
};

Bar* createBar() {
   return new Baz;
}

int createBarAndGetId() {
   return createBar()->id();
}
%}
---

Note that createBar really creates and returns a Baz (which overrides 
Foo::id()), and that the director feature is enabled for Foo.

When called from C++, createBar()->id() and createBarAndGetId() 
obviously return the same value. This assumption also holds when using 
the code from Python: ( director_weird.createBar().id() == 
director_weird.createBarAndGetId() ) yields True.

However, when createBar().id() is called from Java (or C# and D which 
share the same basic mechanism for implementing directors), it 
»resolves« to Foo::id() and yields 0, while createBarAndGetId() still 
returns 2, of course. This happens because Foo.id() calls 
Foo_idSwigExplicitFoo in this case, which in turn specifically calls 
(arg1)->Foo::id().

This is certainly not what I'd expect, and I also got a bug report for 
the D module related to this, but did this really stay unnoticed in the 
C# and Java modules for so long? Or am I missing something?


David