Are there any performance or robustness reasons to prefer one over the other?
#include <iostream>
#include <typeinfo>
struct B
{
virtual bool IsType(B const * b) const { return IsType2nd(b) && b->IsType2nd(this); }
virtual bool IsType2nd(B const * b) const { return dynamic_cast<decltype(this)>(b) != nullptr; }
};
struct D0 : B
{
virtual bool IsType(B const * b) const { return IsType2nd(b) && b->IsType2nd(this); }
virtual bool IsType2nd(B const * b) const { return dynamic_cast<decltype(this)>(b) != nullptr; }
};
struct D1 : B
{
virtual bool IsType(B const * b) const { return IsType2nd(b) && b->IsType2nd(this); }
virtual bool IsType2nd(B const * b) const { return dynamic_cast<decltype(this)>(b) != nullptr; }
};
int main()
{
using namespace std;
B b, bb;
D0 d0, dd0;
D1 d1, dd1;
cout << "type B == type B : " << (b.IsType(&bb) ? "true " : "false") << endl;
cout << "type B == type D0 : " << (b.IsType(&dd0) ? "true " : "false") << endl;
cout << "type B == type D1 : " << (b.IsType(&dd1) ? "true " : "false") << endl;
cout << "type D0 == type B : " << (d0.IsType(&bb) ? "true " : "false") << endl;
cout << "type D0 == type D0 : " << (d0.IsType(&dd0) ? "true " : "false") << endl;
cout << "type D0 == type D1 : " << (d0.IsType(&dd1) ? "true " : "false") << endl;
cout << "type D1 == type B : " << (d1.IsType(&bb) ? "true " : "false") << endl;
cout << "type D1 == type D0 : " << (d1.IsType(&dd0) ? "true " : "false") << endl;
cout << "type D1 == type D1 : " << (d1.IsType(&dd1) ? "true " : "false") << endl;
cout << endl;
cout << "type B == type B : " << (typeid(b) == typeid(bb) ? "true " : "false") << endl;
cout << "type B == type D0 : " << (typeid(b) == typeid(dd0) ? "true " : "false") << endl;
cout << "type B == type D1 : " << (typeid(b) == typeid(dd1) ? "true " : "false") << endl;
cout << "type D0 == type B : " << (typeid(d0) == typeid(&bb) ? "true " : "false") << endl;
cout << "type D0 == type D0 : " << (typeid(d0) == typeid(dd0) ? "true " : "false") << endl;
cout << "type D0 == type D1 : " << (typeid(d0) == typeid(dd1) ? "true " : "false") << endl;
cout << "type D1 == type B : " << (typeid(d1) == typeid(bb) ? "true " : "false") << endl;
cout << "type D1 == type D0 : " << (typeid(d1) == typeid(dd0) ? "true " : "false") << endl;
cout << "type D1 == type D1 : " << (typeid(d1) == typeid(dd1) ? "true " : "false") << endl;
}
output:
type B == type B : true
type B == type D0 : false
type B == type D1 : false
type D0 == type B : false
type D0 == type D0 : true
type D0 == type D1 : false
type D1 == type B : false
type D1 == type D0 : false
type D1 == type D1 : true
type B == type B : true
type B == type D0 : false
type B == type D1 : false
type D0 == type B : false
type D0 == type D0 : true
type D0 == type D1 : false
type D1 == type B : false
type D1 == type D0 : false
type D1 == type D1 : true
From the design perspective, the double dispatch is far more flexible:
Currently you check for strict egality between the type with
IsType2nd(b) && b->IsType2nd(this)
. But may be at some time you'd like to derivate furtherBut one day you may want to derivate further D1, but still want to consider it as if it where a D1 object when comparing types. This kind of special case is easily done with double dispatch.
This flexibility comes at a cost: the assembler code would use 2 indirect calls via a vtable, plus a dynamic cast.
The direct type information is not the greatest design, as Sergey pointed out: It will always be a strict type comparison, with no special case possible.
This inflexibility comes with advantage of simplicity in code generation: the code has just to get the dynamic type information at the start of the vtable (and the compiler can easily optimise this fetch away for object where the type is known at compile time.
For the sake of curiosity, here some code generated : he typeid is optimised away at compile time, whereas the double displatch still relies on indirect calls.