IBM Z and LinuxONE - Languages - Group home

Introduction to the C++11 feature: trailing return types

  

In C++03, the return type of a function template cannot be generalized if the return type relies on those of the template arguments. Here is an example, mul(T a, T b) is a function template that calculates the product of a and b. Arguments a and b are of an arbitrary type T, which is not decided until the template instantiation. Thus, we cannot declare a return type for mul to generalize all the cases for a*b. If we must define such a function template, we have to introduce another template parameter as follows:


template<class T,class U>

U mul(T a, T b){

return a*b;

}


We have some trouble to instantiate this function template because we have to explicitly provide type U in the template instantiation. Can we use feature decltype to let the compiler deduce the result type automatically? See the following example:


template<class T>

decltype(a*b) mul(T a, T b){

return a*b;

}


This program lets the compiler deduce the return type of function template mul. Unfortunately, it doesn't work. The compiler is parsing codes from left to right, so it will issue an error message to indicate that variables a and b are used before their declarations.

 

To solve this problem, C++11 introduced a feature called trailing return types. Using this feature, the previous program can be rewritten as follows:


template<class T>

auto mul(T a, T b) -> decltype(a*b){

return a*b;

}


We specify the function return type after the declaration of parameter declarations. Composite symbol ->decltype(t1+t2) is called a trailing return type. The auto keyword is placed before the function identifier, which is the placeholder of the return type specifier. When a trailing return type is used, the placeholder return type must be auto. Meanwhile, the auto type specifier cannot be used in a function declaration without a trailing return type.

 

The biggest difference between ordinary functions and functions using trailing return types is whether to postpose the return types. See the following example:


auto max(int a, int b) -> int{}


Function max is using a trailing return type, which is equal to int max(int a, int b). This example shows a valid scenario of trailing return types, but it doesn't reflect the benefits of this feature. We can fully enjoy the convenience of generic programming by using trailing return types. See the following example:


#include <iostream>

using namespace std;

 

template<typename T1, typename T2>

auto sum(T1 & t1, T2 & t2) -> decltype(t1 + t2){

return t1 + t2;

}

int main(){

auto i = 1;

auto j = 1.3;

auto k = sum(a,b);

cout << c << endl;

}


This program doesn't contain any explicitly specified types. All the types are deduced by the compiler using auto type deductions and trailing return types, which saves a lot of programming efforts.

 

Another benefit of using trailing return types is the improvement of readability and maintainability of programs. See the following example:


template <class T> class tmp{

public:

int i;

};

tmp<int> (*(*foo())())() {

return 0;

}


Do you feel terrible after reading this program? Actually, foo is a function whose return type is a function pointer. The function pointer points to a function that returns a function pointer. Using trailing return types, the previous program can be rewritten as follows:


template <class T> class tmp{

public:

int i;

};

auto foo()->auto(*)()->tmp<int>(*)(){

return 0;

}


Do you see the magic of trailing return types in this example?

 

Besides the scenarios described above, trailing return types can also be used in function pointers, function references, member functions in classes/structures/class templates.