std::bad_cast pointer vs reference situation

Question

I've noticed with regard to the std::bad_cast exception that references and pointers don't seem to act the same way. For example:

class A { public: ~A() {} };
class B : public A {};

//Case #1
int main()
{
    A a;
    B& b = dynamic_cast<B&>(a);  //Would throw std::bad_cast.
}

//Case #2
int main()
{
    A* a = new A;
    B* b = dynamic_cast<B*>(a);  //Would not throw std::bad_cast.
}

In the first case, an exception of std::bad_cast is generated, and in the second case no exception is generated - instead, the b pointer just is assigned the value NULL.

Can someone explain to me why only the former throws an exception when both are bad_cast examples? I figure there's a good motive behind the decision, and that I'm misusing something as I don't understand that motivation.

Answer 1

Can someone explain to me why only the former throws an exception?

That is how dynamic_cast is specified to behave: a bad dynamic_cast involving pointers yields a null pointer, but there are no null references, so a bad dynamic_cast involving references throws a bad_cast.

The fact that a failed dynamic_cast involving pointers yields a null pointer is useful because it allows for cleaner, simpler type checking and allows for the following idiom:

if (B* b = dynamic_cast<B*>(a))
{
    // The dynamic_cast succeeded and 'b' is non-null.
}

With this idiom, b is in scope and usable if and only if it is non-null.