On that topic, it is possible to do cool things with macros with GCC extensions. A simple example of generic macros was discussed about a week ago: https://news.ycombinator.com/item?id=7896280
Also, I'm not sure I get this "type erasure" business. In ML languages, there is no such erasure. I guess you need this erasure because the language doesn't support parametric polymorphism to begin with? I'm thinking about Java, which hacked generics after the fact, and maintained backward compatibility through type erasure.
In C++, there are few type erasures. Generally, when we want a function to receive a closure, we write it like this (minus dark corner I have missed):
template <typename T, typename E>
for_each(T functor, vector<E>);
No base class in sight there, just lots of duplicated object code. (Which is better is left as an exercise to the reader.)By the way, why do we need std::function at all? Every time I saw it, it was to work around some dark corner of C++ type system or template dark magic.
class A { void foo(); };
A a;
There is no such construct that lets you store a call to foo specialized for the instance 'a': ??? thing = a.foo; // doesn't exist
The only option in C++ is to take the address of the member function: &A::foo
But even this is not sufficient as it is equivalent to: void (*)(A * __this)
That is, the "this" pointer is not stored. Along comes mem_fun: struct mem_fun{ A *a; void(A::*ptr)(); };
Now you can store a call to foo for the instance a: mem_fun fun;
fun.a = &a;
fun.ptr = &A::foo;
Adding an operator() to mem_fun lets you treat a mem_fun instance as a function: void operator()(){ a->*ptr(); }
...
fun()
Now, lets say you need to store this as a generic "callback". Well, how would you do that? Sure, you can store mem_fun instances but what about just plain function pointers? Now you can't store them. In comes std::function. With a little bit of VERY simple to understand magic, you can do: std::function<void()> foo = fun; // mem_fun
std::function<void()> foo2 = global_fun; // simple member pointer
Hey, that sounds cool. Do you have a link? A quick search doesn't seem to work.
The vtables on the other hand I will steer clear from. Closures are way more useful than class based polymorphism. Combine it with parametric polymorphism (templates), and you won't need to subclass anything ever again —okay, I'm exaggerating a little.