Recently, in the binary sizes series, we discussed how run-time type information affects RTTI. I also mentioned that in my opinion the lack of RTTI leads to better practices and you'll end up with a more readable, more maintainable code.
It's time to delve into this topic and see why.
What is RTTI again?
But first of all, let's quickly recap what is run-time type information. RTTI let us have information on the dynamic type of reference/pointer types. It lets us use dyanamic_cast
s and also call the typeid()
function and query the returned instance of std::type_info
.
I think that not having access to these tools will let you write better code. Why so?
Let's start with probably the less controversial tool.
Don't rely on typeid()
Inexperienced programmers might want to automatically use typeid().name()
to see what's the dynamic type of something during development and testing. It's usually not kept because often the output is not that readable. I mean for int
, you might simply get i
. That's something easily recognizable, but not something you'd want to see in the logs.
The bigger problem is that the output is implementation-defined, so you really shouldn't count on it if you want code that is portable. (Even though there is a fair chance that they will be the same on GCC and clang).
Let's see a couple of examples.
#include <iostream>
int main() {
std::cout << typeid(5).name() << ' ';
std::cout << typeid(5u).name() << ' ';
std::cout << typeid(5.0).name() << ' ';
std::cout << typeid(true).name() << ' ';
std::cout << typeid('5').name() << '\n';
return 0;
}
/*
gcc: i j d b c
clang: i j d b c
*/
Different outputs among compilers do not mean that you cannot use them consistently. You can still use typeid().name()
or typeid.hash()
to compare types against each other in your code and branch the execution based on those comparisons...
On the other hand, you don't have many reasons to do that. If you have to deal with, if you have to compare unrelated types, probably you should extract the type-dependent code and use different overloads.
#include <memory>
#include <iostream>
#include <vector>
class Wine {
public:
virtual ~Wine() = default;
virtual void print() { std::cout << "this is wine\n"; }
};
class WhiteWine {
public:
virtual ~WhiteWine() = default;
virtual void print() { std::cout << "this is white wine\n"; }
};
class RedWine {
public:
virtual ~RedWine() = default;
virtual void print() { std::cout << "this is red wine\n"; }
};
class RoseWine {
public:
virtual ~RoseWine() = default;
virtual void print() { std::cout << "this is rose wine\n"; }
};
int main() {
std::vector<std::unique_ptr<Wine>> wines;
auto wine1 = std::make_unique<WhiteWine>();
auto wine2 = std::make_unique<RedWine>();
auto wine3 = std::make_unique<RoseWine>();
}
If the types are from the same inheritance tree, then you should either use dynamic_cast
s or even better, you should benefit from dynamic dispatching.
dynamic_cast
is slightly better, but still should be avoided
dynamic_cast
safely converts pointers and references to classes up, down and sideways along the inheritance hierarchy - according to CppReference. Often when you have a collection of pointers to the base class, you'll try to cast it to different derived classes and if the case is successful you do whatever you want with that type.
In other words, when you have no idea what types you have, you can start casting things. You might even have a series of casts like this:
OffRoader* offroader = dynamic_cast<OffRoader*>(car);
if (offroader) {
offroader->turnOnAllWheelDrive();
}
Van* van = dynamic_cast<Van*>(car);
if (van) {
van->attachThirdSeatRow();
}
Roadster* roadster = dynamic_cast<Roadster*>(car);
if (roadster) {
roadster->removeRoof();
}
Many would say that this is a code smell. I'm among them. Many would go further and say that using dynamic_cast
in general is a code smell.
As mentioned earlier, this is almost always better than relying on typeid()
, but behind dyanimc_cast
s you'll only find bad inheritance trees and messed up APIs.
Rather than having different public interfaces, we should have a unified one.
There is a convenient reason for that. If the interface is unified, we don't need to cast our objects, we can simply rely on runtime dispatching of our function calls.
#include <iostream>
#include <memory>
#include <vector>
class Car {
public:
virtual ~Car() = default;
virtual void doSomeFun() = 0;
};
class OffRoader: public Car {
public:
void doSomeFun() override {
std::cout << "use all wheel drive on OffRoader\n";
}
};
class Van: public Car {
public:
void doSomeFun() override {
std::cout << "attach third seatrow in a van\n";
}
};
class Roadster: public Car {
public:
void doSomeFun() override {
std::cout << "remove roadster's roof\n";
}
};
int main() {
std::vector<std::unique_ptr<Car>> myCars;
myCars.push_back(std::make_unique<OffRoader>());
myCars.push_back(std::make_unique<Van>());
myCars.push_back(std::make_unique<Roadster>());
for (auto& car: myCars) {
car->doSomeFun();
}
return 0;
}
When typeid
is still better?
As I mentioned earlier, while dynamic_cast
is not a good solution, it's almost always better than using typeid()
. Why not always?
There is a not-so-subtle difference between these 2 RTTI tools. dynamic_cast
checks whether an object "is kind of" another class. But typeid()
will look for an exact match.
#include <iostream>
#include <memory>
#include <vector>
class A {
public:
virtual ~A() = default;
};
class B: public A {
};
class C: public B {
};
int main() {
std::unique_ptr<A> p1 = std::make_unique<C>();
B* p2 = dynamic_cast<B*>(p1.get());
C* p3 = dynamic_cast<C*>(p1.get());
if (p2) {
std::cout << "p1 is like B*. it's convertible to B\n";
} else if (p3) {
// we won't reach this branch due to bad ordering of if/else branches
std::cout << "p1 is like C*. it's convertible to C\n";
}
if (typeid(*p1) == typeid(B{})) {
std::cout << "*p1 is B\n";
} else if (typeid(*p1) == typeid(C{})) {
std::cout << "*p1 is C\n";
}
return 0;
}
/*
p1 is like B*. it's convertible to B
*p1 is C
*/
As such, typeid()
is both simpler and faster. Even though we had to construct an extra instance of B
and C
in the above example to be able to compare the type information. If that's costly and/or we must do this several times, we could create a map of std::type_index
objects.
// ...
const std::map<std::string, std::type_index> typeMap = {
{"A", std::type_index(typeid(A{}))},
{"B", std::type_index(typeid(B{}))},
{"C", std::type_index(typeid(C{}))}
};
if (p2) {
std::cout << "p1 is like B*. it's convertible to B\n";
} else if (p3) {
// we won't reach this branch due to bad ordering of if/else branches
std::cout << "p1 is like C*. it's convertible to C\n";
}
if (typeid(*p1) == typeMap.at("B")) {
std::cout << "*p1 is B\n";
} else if (typeid(*p1) == typeMap.at("C")) {
std::cout << "*p1 is C\n";
}
// ...
There is a couple of things to note:
- typeMap
is const
. It's necessary, because std::type_index
is a copyable wrapper around the non-copyable std::type_info
. It is not default constructible. In other words, if the map
is not const
, then the code wouldn't compile as a mutable map's value type must be default constructible.
-
std::type_index
is a copyable wrapper around the non-copyablestd::type_info
. Asstd::type_index
does not have a default constructor, you cannot useoperator=()
to add new items to the map. You either initialize it at declaration (which is preferable), or you use theinsert()
method. - We use
map::at()
and cannot usemap::operator[]
. The reason is thatmap::operator[]
is notconst
whereasmap::at()
has aconst
overload.
Overall, the best is still avoiding both using typeid
and dynamic_cast
.
When to still use dyanmic_cast
?
The C++ core guidelines explain well the differences between casting to a pointer or a reference type and when you should refer which (C.147 and C.148), but it doesn't mean that you should use any.
In fact, even C.153 says that you should prefer using virtual functions to casting. Casting is error-prone and you can make mess it up easily. Virtual functions are safe and when you call virtual functions it's guaranteed that you'll reach the most derived function. On the other hand, we saw with dynamic_cast
that you might end up calling an intermediary function.
As we saw earlier, with virtual functions your code will be cleaner and speed is not an issue because dynamic_cast
is (also) slow.
Conclusion
With Run-time Type Information, we get access to typeid()
and dynamic_cast
. They help us query the dynamic types of polymorphic objects. In other words, they help us identify which derived objects are held by base class pointers.
With great power, we also get great responsibility. Sadly, experience shows that responsibility is often abused. These tools are often overused and result in messy code. Using virtual
functions and relying on dynamic dispatching is almost always better.
Besides having cleaner code, you can also benefit from shorter compile times and smaller binaries as the type information doesn't have to be generated and stored. To me, turning RTTI off is a great option to consider.
Do you rely on it?
Connect deeper
If you liked this article, please
- hit on the like button,
- subscribe to my newsletter
- and let's connect on Twitter!