DEV Community

Sandor Dargo
Sandor Dargo

Posted on • Edited on • Originally published at sandordargo.com

The concept behind C++ concepts

The idea of concepts is one of the major new features added to C++20. Concepts are an extension for templates. They can be used to perform compile-time validation of template arguments through boolean predicates. They can also be used to perform function dispatch based on properties of types.

With concepts, you can require both syntactic and semantic conditions. In terms of syntactic requirements, imagine that you can impose the existence of certain functions in the API of any class. For example, you can create a concept Car that requires the existence of an accelerate function:

#include <concepts>

template <typename C>
concept Car = requires (C car) {
  car.accelerate()
};
Enter fullscreen mode Exit fullscreen mode

Don't worry about the syntax, we'll get there next week.

Semantic requirements are more related to mathematical axioms, for example, you can think about associativy or commutativity:

a + b == b + a // commutativity
(a + b) + c == a + (b + c) // associativity
Enter fullscreen mode Exit fullscreen mode

There are concepts in the standard library expressing semantic requirements. Take for example std::equality_comparable.

It requires that

  • the two equality comparison between the passed in types are commutative,
  • == is symmetric, transitive and reflexive,
  • and equality_comparable_with<T, U> is modeled only if, given any lvalue t of type const std::remove_reference_t<T> and any lvalue u of type const std::remove_reference_t<U>, and let C be std::common_reference_t<const std::remove_reference_t<T>&, const std::remove_reference_t<U>&>, bool(t == u) == bool(C(t) == C(u)).

Though this latter one is probably a bit more difficult to decipher. Anyway, if you are looking for a thorough article dedicated to semantic requirements, read this one by Andrzej Krzemieński.

The motivation behind concepts

We have briefly seen from a very high-level what we can express with concepts. But why do we need them in the first place?

For the sake of example, let's say you want to write a function that adds up two numbers. You want to accept both integral and floating-point numbers. What are you going to do?

You could accept doubles, maybe even long doubles and return a value of the same type.

#include <iostream>

long double add(long double a, long double b) {
    return a+b;
}

int main() {
  int a{42};
  int b{66};
  std::cout << add(a, b) << '\n';
}
Enter fullscreen mode Exit fullscreen mode

The problem is that when you call add() with two ints, they will be cast to long double. You might want a smaller memory footprint, or maybe you'd like to take into account the maximum or minimum limits of a type. And anyway, it's not the best idea to rely on implicit conversions.

Implicit conversions might allow code to compile that was not at all in your intentions. It's not bad by definition, but implicit conversions should be intentional and not accidental.

In this case, I don't think that an intentional cast is justified.

Defining overloads for the different types is another way to take, but it is definitely tedious.

#include <iostream>

long double add(long double a, long double b) {
  return a+b;
}

int add(int a, int b) {
  return a+b;
}

int main() {
  int a{42};
  int b{66};
  std::cout << add(a, b) << '\n';
}
Enter fullscreen mode Exit fullscreen mode

Imagine that you want to do this for all the different numeric types. Should we also do it for combinations of long doubles and shorts? Eh... Thanks, but no thanks.

Another option is to define a template!

#include <iostream>

template <typename T>
T add(T a, T b) {
    return a+b;
}

int main() {
  int a{42};
  int b{66};
  std::cout << add(a, b) << '\n';
  long double x{42.42L};
  long double y{66.6L};
  std::cout << add(x, y) << '\n';

}
Enter fullscreen mode Exit fullscreen mode

If you have a look at CPP Insights you will see that code was generated both for an int and for a long double overload. There is no static cast taking place at any point.

Are we good yet?

Unfortunately, no.

What happens if you try to call add(true, false)? You'll get a 1 as true is promoted to an integer, summed up with false promoted to an integer and then they will be turned back (by static_cast) into a boolean.

What if you add up two string? They will be concatenated. But is that really what you want? Maybe you don't want that to be a valid operation and you prefer a compilation failure.

So you might have to forbid that template specialization. And for how many types do you want to do the same?

What if you could simply say that you only want to add up integral or floating-point types. In brief, rational numbers. And here come concepts into the picture.

With concepts, you can easily express such requirements on template parameters.

You can precise requirements on

  • the validity of expressions (that certain functions should exist in the class' API)
  • the return types of certain functions
  • the existence of inner types, of template specializations
  • the type-traits of the accepted types

How? That's what we are going to explore in this series on C++ concepts.

What's next?

During the next couple of weeks we are going to discuss:

  • the 4 ways to use a concept
  • what kind of predefined concepts the standard library introduced
  • how to write our own concepts

Stay tuned!

If you want to learn more details about C++ concepts, check out my book on Leanpub!

Connect deeper

If you found interesting this article, please subscribe to my personal blog and let's connect on Twitter!

Top comments (12)

Collapse
 
pgradot profile image
Pierre Gradot

Finally discovering a link to one of my articles here XD

Collapse
 
sandordargo profile image
Sandor Dargo

It's not your only article I linked during the past months! ;)

Collapse
 
pgradot profile image
Pierre Gradot

The first I notice recently ^^

Collapse
 
_ruturaj7 profile image
Ruturaj Porlekar

Could help me for dev c++ gui program ?

Collapse
 
sandordargo profile image
Sandor Dargo

Sorry, that's something I never worked with.

Collapse
 
bartosz profile image
Bartosz Wójcik

Why did you never work on GUI in C++? How can you teach C++ if you never touched real-life C++ developer problems like GUI?

Thread Thread
 
sandordargo profile image
Sandor Dargo

I beg your pardon, Sire! Developers responsible for servers dealing with millions of transactions per second and also developers who work on embedded systems should not speak about C++. Probably game developers neither. Please help us, which mineshaft should we choose to hide?

Collapse
 
_ruturaj7 profile image
Ruturaj Porlekar

Ok

Collapse
 
albertopdrf profile image
Alberto Pérez de Rada Fiol

That's such an interesting topic Sandor, I'm looking forward to the next post! 😃

Collapse
 
sandordargo profile image
Sandor Dargo

Thanks, for your comment Alberto! The next part comes next Wednesday morning!

Collapse
 
bartosz profile image
Bartosz Wójcik

I'll pass. I choose JavaScript. I choose C#. I choose PHP. I choose clean C. This is useless stuff for maybe 10 C++ experts, nobody in their right mind will ever use it.

Collapse
 
0916dhkim profile image
Danny Kim

A big portion of the video game industry heavily depends on C++. IMO C++ is a neat language. It is just not made for web dev & it cannot stop developers from writing crappy code.