I am quite confused with the dynamic_cast
keyword in C++.
struct A {
virtual void f() { }
};
struct B : public A { };
struct C { };
void f () {
A a;
B b;
A* ap = &b;
B* b1 = dynamic_cast<B*> (&a); // NULL, because 'a' is not a 'B'
B* b2 = dynamic_cast<B*> (ap); // 'b'
C* c = dynamic_cast<C*> (ap); // NULL.
A& ar = dynamic_cast<A&> (*ap); // Ok.
B& br = dynamic_cast<B&> (*ap); // Ok.
C& cr = dynamic_cast<C&> (*ap); // std::bad_cast
}
the definition says:
The dynamic_cast keyword casts a datum from one pointer or reference type to another, performing a runtime check to ensure the validity of the cast
Can we write an equivalent of dynamic_cast
of C++ in C so that I could better understand things?
dynamic_cast<>
works behind the scenes (or how much of C++ works), a good book (that's also pretty easy to read for something so technical) is Lippman's "Inside the C++ Object Model". Also Stroustrup's "Design and Evolution of C++" and "The C++ Programming Language" books are good resources, but Lippman's book is dedicated to how C++ works 'behind the scenes'.
B* b2 = dynamic_cast<B*> (ap) // 'b'
means? b2 is pointer to b
or what?
Here's a rundown on static_cast<>
and dynamic_cast<>
specifically as they pertain to pointers. This is just a 101-level rundown, it does not cover all the intricacies.
static_cast< Type* >(ptr)
This takes the pointer in ptr
and tries to safely cast it to a pointer of type Type*
. This cast is done at compile time. It will only perform the cast if the types are related. If the types are not related, you will get a compiler error. For example:
class B {};
class D : public B {};
class X {};
int main()
{
D* d = new D;
B* b = static_cast<B*>(d); // this works
X* x = static_cast<X*>(d); // ERROR - Won't compile
return 0;
}
dynamic_cast< Type* >(ptr)
This again tries to take the pointer in ptr
and safely cast it to a pointer of type Type*
. But this cast is executed at runtime, not compile time. Because this is a run-time cast, it is useful especially when combined with polymorphic classes. In fact, in certian cases the classes must be polymorphic in order for the cast to be legal.
Casts can go in one of two directions: from base to derived (B2D) or from derived to base (D2B). It's simple enough to see how D2B casts would work at runtime. Either ptr
was derived from Type
or it wasn't. In the case of D2B dynamic_cast<>s, the rules are simple. You can try to cast anything to anything else, and if ptr
was in fact derived from Type
, you'll get a Type*
pointer back from dynamic_cast
. Otherwise, you'll get a NULL pointer.
But B2D casts are a little more complicated. Consider the following code:
#include <iostream>
using namespace std;
class Base
{
public:
virtual void DoIt() = 0; // pure virtual
virtual ~Base() {};
};
class Foo : public Base
{
public:
virtual void DoIt() { cout << "Foo"; };
void FooIt() { cout << "Fooing It..."; }
};
class Bar : public Base
{
public :
virtual void DoIt() { cout << "Bar"; }
void BarIt() { cout << "baring It..."; }
};
Base* CreateRandom()
{
if( (rand()%2) == 0 )
return new Foo;
else
return new Bar;
}
int main()
{
for( int n = 0; n < 10; ++n )
{
Base* base = CreateRandom();
base->DoIt();
Bar* bar = (Bar*)base;
bar->BarIt();
}
return 0;
}
main()
can't tell what kind of object CreateRandom()
will return, so the C-style cast Bar* bar = (Bar*)base;
is decidedly not type-safe. How could you fix this? One way would be to add a function like bool AreYouABar() const = 0;
to the base class and return true
from Bar
and false
from Foo
. But there is another way: use dynamic_cast<>
:
int main()
{
for( int n = 0; n < 10; ++n )
{
Base* base = CreateRandom();
base->DoIt();
Bar* bar = dynamic_cast<Bar*>(base);
Foo* foo = dynamic_cast<Foo*>(base);
if( bar )
bar->BarIt();
if( foo )
foo->FooIt();
}
return 0;
}
The casts execute at runtime, and work by querying the object (no need to worry about how for now), asking it if it the type we're looking for. If it is, dynamic_cast<Type*>
returns a pointer; otherwise it returns NULL.
In order for this base-to-derived casting to work using dynamic_cast<>
, Base, Foo and Bar must be what the Standard calls polymorphic types. In order to be a polymorphic type, your class must have at least one virtual
function. If your classes are not polymorphic types, the base-to-derived use of dynamic_cast
will not compile. Example:
class Base {};
class Der : public Base {};
int main()
{
Base* base = new Der;
Der* der = dynamic_cast<Der*>(base); // ERROR - Won't compile
return 0;
}
Adding a virtual function to base, such as a virtual dtor, will make both Base and Der polymorphic types:
class Base
{
public:
virtual ~Base(){};
};
class Der : public Base {};
int main()
{
Base* base = new Der;
Der* der = dynamic_cast<Der*>(base); // OK
return 0;
}
Unless you're implementing your own hand-rolled RTTI (and bypassing the system one), it's not possible to implement dynamic_cast
directly in C++ user-level code. dynamic_cast
is very much tied into the C++ implementation's RTTI system.
But, to help you understand RTTI (and thus dynamic_cast
) more, you should read up on the <typeinfo>
header, and the typeid
operator. This returns the type info corresponding to the object you have at hand, and you can inquire various (limited) things from these type info objects.
dynamic_cast
are very skimpy. :-P Just play with it yourself until you get the hang of it. :-)
More than code in C, I think that an english definition could be enough:
Given a class Base of which there is a derived class Derived, dynamic_cast
will convert a Base pointer to a Derived pointer if and only if the actual object pointed at is in fact a Derived object.
class Base { virtual ~Base() {} };
class Derived : public Base {};
class Derived2 : public Base {};
class ReDerived : public Derived {};
void test( Base & base )
{
dynamic_cast<Derived&>(base);
}
int main() {
Base b;
Derived d;
Derived2 d2;
ReDerived rd;
test( b ); // throw: b is not a Derived object
test( d ); // ok
test( d2 ); // throw: d2 is not a Derived object
test( rd ); // ok: rd is a ReDerived, and thus a derived object
}
In the example, the call to test
binds different objects to a reference to Base
. Internally the reference is downcasted to a reference to Derived
in a typesafe way: the downcast will succeed only for those cases where the referenced object is indeed an instance of Derived
.
The following is not really close to what you get from C++'s dynamic_cast
in terms of type checking but maybe it will help you understand its purpose a little bit better:
struct Animal // Would be a base class in C++
{
enum Type { Dog, Cat };
Type type;
};
Animal * make_dog()
{
Animal * dog = new Animal;
dog->type = Animal::Dog;
return dog;
}
Animal * make_cat()
{
Animal * cat = new Animal;
cat->type = Animal::Cat;
return cat;
}
Animal * dyn_cast(AnimalType type, Animal * animal)
{
if(animal->type == type)
return animal;
return 0;
}
void bark(Animal * dog)
{
assert(dog->type == Animal::Dog);
// make "dog" bark
}
int main()
{
Animal * animal;
if(rand() % 2)
animal = make_dog();
else
animal = make_cat();
// At this point we have no idea what kind of animal we have
// so we use dyn_cast to see if it's a dog
if(dyn_cast(Animal::Dog, animal))
{
bark(animal); // we are sure the call is safe
}
delete animal;
}
First, to describe dynamic cast in C terms, we have to represent classes in C. Classes with virtual functions use a "VTABLE" of pointers to the virtual functions. Comments are C++. Feel free to reformat and fix compile errors...
// class A { public: int data; virtual int GetData(){return data;} };
typedef struct A { void**vtable; int data;} A;
int AGetData(A*this){ return this->data; }
void * Avtable[] = { (void*)AGetData };
A * newA() { A*res = malloc(sizeof(A)); res->vtable = Avtable; return res; }
// class B : public class A { public: int moredata; virtual int GetData(){return data+1;} }
typedef struct B { void**vtable; int data; int moredata; } B;
int BGetData(B*this){ return this->data + 1; }
void * Bvtable[] = { (void*)BGetData };
B * newB() { B*res = malloc(sizeof(B)); res->vtable = Bvtable; return res; }
// int temp = ptr->GetData();
int temp = ((int(*)())ptr->vtable[0])();
Then a dynamic cast is something like:
// A * ptr = new B();
A * ptr = (A*) newB();
// B * aB = dynamic_cast<B>(ptr);
B * aB = ( ptr->vtable == Bvtable ? (B*) aB : (B*) 0 );
A dynamic_cast
performs a type checking using RTTI. If it fails it'll throw you an exception (if you gave it a reference) or NULL if you gave it a pointer.
There are no classes in C, so it's impossible to to write dynamic_cast in that language. C structures don't have methods (as a result, they don't have virtual methods), so there is nothing "dynamic" in it.
No, not easily. The compiler assigns a unique identity to every class, that information is referenced by every object instance, and that is what gets inspected at runtime to determine if a dynamic cast is legal. You could create a standard base class with this information and operators to do the runtime inspection on that base class, then any derived class would inform the base class of its place in the class hierarchy and any instances of those classes would be runtime-castable via your operations.
edit
Here's an implementation that demonstrates one technique. I'm not claiming the compiler uses anything like this, but I think it demonstrates the concepts:
class SafeCastableBase
{
public:
typedef long TypeID;
static TypeID s_nextTypeID;
static TypeID GetNextTypeID()
{
return s_nextTypeID++;
}
static TypeID GetTypeID()
{
return 0;
}
virtual bool CanCastTo(TypeID id)
{
if (GetTypeID() != id) { return false; }
return true;
}
template <class Target>
static Target *SafeCast(SafeCastableBase *pSource)
{
if (pSource->CanCastTo(Target::GetTypeID()))
{
return (Target*)pSource;
}
return NULL;
}
};
SafeCastableBase::TypeID SafeCastableBase::s_nextTypeID = 1;
class TypeIDInitializer
{
public:
TypeIDInitializer(SafeCastableBase::TypeID *pTypeID)
{
*pTypeID = SafeCastableBase::GetNextTypeID();
}
};
class ChildCastable : public SafeCastableBase
{
public:
static TypeID s_typeID;
static TypeID GetTypeID()
{
return s_typeID;
}
virtual bool CanCastTo(TypeID id)
{
if (GetTypeID() != id) { return SafeCastableBase::CanCastTo(id); }
return true;
}
};
SafeCastableBase::TypeID ChildCastable::s_typeID;
TypeIDInitializer ChildCastableInitializer(&ChildCastable::s_typeID);
class PeerChildCastable : public SafeCastableBase
{
public:
static TypeID s_typeID;
static TypeID GetTypeID()
{
return s_typeID;
}
virtual bool CanCastTo(TypeID id)
{
if (GetTypeID() != id) { return SafeCastableBase::CanCastTo(id); }
return true;
}
};
SafeCastableBase::TypeID PeerChildCastable::s_typeID;
TypeIDInitializer PeerChildCastableInitializer(&PeerChildCastable::s_typeID);
int _tmain(int argc, _TCHAR* argv[])
{
ChildCastable *pChild = new ChildCastable();
SafeCastableBase *pBase = new SafeCastableBase();
PeerChildCastable *pPeerChild = new PeerChildCastable();
ChildCastable *pSameChild = SafeCastableBase::SafeCast<ChildCastable>(pChild);
SafeCastableBase *pBaseToChild = SafeCastableBase::SafeCast<SafeCastableBase>(pChild);
ChildCastable *pNullDownCast = SafeCastableBase::SafeCast<ChildCastable>(pBase);
SafeCastableBase *pBaseToPeerChild = SafeCastableBase::SafeCast<SafeCastableBase>(pPeerChild);
ChildCastable *pNullCrossCast = SafeCastableBase::SafeCast<ChildCastable>(pPeerChild);
return 0;
}
static_cast< Type* >(ptr)
static_cast in C++ can be used in scenarios where all type casting can be verified at compile time.
dynamic_cast< Type* >(ptr)
dynamic_cast in C++ can be used to perform type safe down casting. dynamic_cast is run time polymorphism. The dynamic_cast operator, which safely converts from a pointer (or reference) to a base type to a pointer (or reference) to a derived type.
eg 1:
#include <iostream>
using namespace std;
class A
{
public:
virtual void f(){cout << "A::f()" << endl;}
};
class B : public A
{
public:
void f(){cout << "B::f()" << endl;}
};
int main()
{
A a;
B b;
a.f(); // A::f()
b.f(); // B::f()
A *pA = &a;
B *pB = &b;
pA->f(); // A::f()
pB->f(); // B::f()
pA = &b;
// pB = &a; // not allowed
pB = dynamic_cast<B*>(&a); // allowed but it returns NULL
return 0;
}
For more information click here
eg 2:
#include <iostream>
using namespace std;
class A {
public:
virtual void print()const {cout << " A\n";}
};
class B {
public:
virtual void print()const {cout << " B\n";}
};
class C: public A, public B {
public:
void print()const {cout << " C\n";}
};
int main()
{
A* a = new A;
B* b = new B;
C* c = new C;
a -> print(); b -> print(); c -> print();
b = dynamic_cast< B*>(a); //fails
if (b)
b -> print();
else
cout << "no B\n";
a = c;
a -> print(); //C prints
b = dynamic_cast< B*>(a); //succeeds
if (b)
b -> print();
else
cout << "no B\n";
}
dynamic_cast uses RTTI. It can slow down your application, you can use modification of the visitor design pattern to achieve downcasting without RTTI http://arturx64.github.io/programming-world/2016/02/06/lazy-visitor.html
Success story sharing
Base* base = new Base;
,dynamic_cast<Foo*>(base)
will beNULL
.base
is not aFoo
. ABase
pointer can point to aFoo
, but it's still aFoo
, so a dynamic cast will work. If you doBase* base = new Base
,base
is aBase
, not aFoo
, so you cannot dynamically cast it to aFoo
.bar->BarIt();
will printbaring It...
, even for theFoo
class. Although the answer is very dated now, it might make for a nice comment regarding why. Namely, the coincidental overlapping vtable entry forBarIt
andFooIt
in the class definitions.