Re: [Swig-user] using %extend to implement virtual functions

[William S F. <ws...@fu...>]

Josh Cherry wrote:
> 
> 
> On Wed, 28 Mar 2007, William S Fulton wrote:
> 
>> Your only solution right now is to use RTTI as Josh mentioned:
>>
>> %extend Base {
>>  virtual const char *__str__() const
>>  {
>>    if (dynamic_cast<const Derived2 *>($self))
>>      return "I'm Derived2 class";
>>    else if (dynamic_cast<const Derived1 *>($self))
>>      return "I'm Derived1 class";
>>    else if (dynamic_cast<const Base *>($self))
>>      return "I'm Base class";
>>  }
>> }
> 
> I had in mind grabbing typeid($self).name(), which, on the platforms
> I've looked at (Linux/gcc and Windows/msvc), has the class name embedded
> in it (in different ways).  That way there's no need even to know the
> classes in advance.  In fact I think it should be possible to wrap the
> rtti stuff so that this can be applied in Python to any polymorphic class.
> 

Many moons ago when I used that approach, I found different manglings in
the names returned by each different compiler.

I just checked the C++ standard and it says that std::type_info::name()
returns an implementation defined NTBS (null terminated byte string). So
the standard allows for the inconsistent behaviour I've noticed across
compilers.

William