Boost logo

Geometry :

Subject: [ggl] space partitioning
From: Hartmut Kaiser (hartmut.kaiser)
Date: 2010-09-11 13:50:17


> On Fri, Sep 10, 2010 at 9:11 PM, Hartmut Kaiser <hartmut.kaiser_at_[hidden]>
> wrote:
> > Since all these examples uses EMPTY classes, codes as following will
> > complaint nothing and event run as people expects -- the tree::foo()
> > will be invoked:
> >
> > struct tree : public node
> > {
> > void foo()
> > {
> > do_something_with_node(static_cast<node*>(this));
> > }
> > /* no atributes */
> > }
> >
> > internal_node *in = new internal_node(); tree *t =
> > static_cast<tree*>(static_cast<node*>(in));
> > t->foo();
> >
> > As a member function needs a object pointer to be invoked, in facts,
> > it's not matter what's the pointer really pointed to, if your class
> > hierarchy is empty classes, no CRASH will complaint. But if you try to
> > insert some member fields in 'tree', and intend to access any fields,
> > the whole world will CRASH -- the 'this' pointer is in fact pointed to
> > something totally different.
> Even with empty classes it's still undefined behavior, so all bets are
> off.
>
> Still undefined of cause, unless people uses virtual on 'foo()'. In fact
> examples like that tree-node-foo is nothing object oriented, that empty-
> class trick is just a strange thing in object oriented, hides real object
> type to C++, and cause more strange things happened while doing typeinfo
> on that 'this' pointer of tree. Think of the following segment:
>
> #include <iostream>
> #include <typeinfo>
>
> struct base
> {
> };
>
> struct derived : base
> {
> void blah() { std::cout<<"derived"<<std::endl; }
> void pointer_type() { std::cout<<typeid(this).name()<<std::endl; }
> void object_type() { std::cout<<typeid(*this).name()<<std::endl; } };
>
> int main()
> {
> base obj;
> base *o = &obj;
> derived *p = (derived*)o;
> p->blah();
> p->pointer_type();
> p->object_type();
> }
>
> Well the typeid operator is in fact static binded, so people may gains
> output as one expected, but we all knows that the C++ compiler has been
> tricked. How to make C++ compiler smarter? And how could C++ people be
> more aware of the real type without so many virtuals?

Well, I'm certainly not sure what you're aiming at here, but if you want to avoid virtual functions, templates are still your friend. Everything you can do with virtual functions at runtime can be done with templates at compile time.

Regards Hartmut
---------------
http://boost-spirit.com


Geometry list run by mateusz at loskot.net