14.13 — Temporary class objects

Consider the following example:

#include <iostream>

int add(int x, int y)
{
    int sum{ x + y }; // stores x + y in a variable
    return sum;       // returns value of that variable
}

int main()
{
    std::cout << add(5, 3) << '\n';

    return 0;
}

In the add() function, the variable sum is used to store the result of the expression x + y. This variable is then evaluated in the return statement to produce the value to be returned. While this might be occasionally useful for debugging (so we can inspect the value of sum if desired), it actually makes the function more complex than it needs to be by defining an object that is then only used one time.

In most cases where a variable is used only once, we actually don’t need a variable. Instead, we can substitute in the expression used to initialize the variable where the variable would have been used. Here is the add() function rewritten in this manner:

#include <iostream>

int add(int x, int y)
{
    return x + y; // just return x + y directly
}

int main()
{
    std::cout << add(5, 3) << '\n';

    return 0;
}

This works not only with return values, but also with most function arguments. For example, instead of this:

#include <iostream>

void printValue(int value)
{
    std::cout << value;
}

int main()
{
    int sum{ 5 + 3 };
    printValue(sum);

    return 0;
}

We can write this:

#include <iostream>

void printValue(int value)
{
    std::cout << value;
}

int main()
{
    printValue(5 + 3);

    return 0;
}

Note how much cleaner this keeps our code. We don’t have to define and give a name to a variable. And we don’t have to scan through the entire function to determine whether that variable is actually used elsewhere. Because 5 + 3 is an expression, we know it is only used on that one line.

Do note that this only works in cases where an rvalue expression is accepted. In cases where an lvalue expression is required, we must have an object:

#include <iostream>

void addOne(int& value) // pass by non-const references requires lvalue
{
    ++value;
}

int main()
{
    int sum { 5 + 3 };
    addOne(sum);   // okay, sum is an lvalue

    addOne(5 + 3); // compile error: not an lvalue

    return 0;
}

Temporary class objects

The same issue applies in the context of class types.

The following example is similar to the ones above, but uses program-defined class type IntPair instead of int:

#include <iostream>

class IntPair
{
private:
    int m_x{};
    int m_y{};

public:
    IntPair(int x, int y)
        : m_x { x }, m_y { y }
    {}

    int x() const { return m_x; }
    int y() const { return m_y; }
};

void print(IntPair p)
{
    std::cout << "(" << p.x() << ", " << p.y() << ")\n";        
}
        
int main()
{
    // Case 1: Pass variable
    IntPair p { 3, 4 };
    print(p); // prints (3, 4)
    
    return 0;
}

In case 1, we’re instantiating variable IntPair p and then passing p to function print().

However, p is only used once, and function print() will accept rvalues, so there is really no reason to define a variable here. So let’s get rid of p.

We can do that by passing a temporary object instead of a named variable. A temporary object (sometimes called an anonymous object or an unnamed object) is an object that has no name and exists only for the duration of a single expression.

There are two common ways to create temporary class type objects:

#include <iostream>

class IntPair
{
private:
    int m_x{};
    int m_y{};

public:
    IntPair(int x, int y)
        : m_x { x }, m_y { y }
    {}

    int x() const { return m_x; }
    int y() const{ return m_y; }
};

void print(IntPair p)
{
    std::cout << "(" << p.x() << ", " << p.y() << ")\n";        
}
        
int main()
{
    // Case 1: Pass variable
    IntPair p { 3, 4 };
    print(p);

    // Case 2: Construct temporary IntPair and pass to function
    print(IntPair { 5, 6 } );

    // Case 3: Implicitly convert { 7, 8 } to a temporary Intpair and pass to function
    print( { 7, 8 } );
    
    return 0;
}

In case 2, we’re telling the compiler to construct an IntPair object, and initializing it with { 5, 6 }. Because this object has no name, it is a temporary. The temporary object is then passed to parameter p of function print(). When the function call returns, the temporary object is destroyed.

In case 3, we’re also creating a temporary IntPair object to pass to function print(). However, because we have not explicitly specified what type to construct, the compiler will deduce the necessary type (IntPair) from the function parameter, and then implicitly convert { 7, 8 } to an IntPair object.

To summarize:

IntPair p { 1, 2 }; // create named object p initialized with { 1, 2 }
IntPair { 1, 2 };   // create temporary object initialized with { 1, 2 }
{ 1, 2 };           // compiler will try to convert { 1, 2 } to temporary object matching expected type (typically a parameter or return type)

We’ll discuss this last case in more detail in lesson 14.16 -- Converting constructors and the explicit keyword.

A few more examples:

std::string { "Hello" }; // create a temporary std::string initialized with "Hello"
std::string {};          // create a temporary std::string using value initialization / default constructor

Creating temporary objects via direct initialization Optional

Since we can create temporary objects via direct-list-initialization, you might be wondering whether you can create temporary objects via the other initialization forms. There is no syntax to create temporary objects using copy initialization.

However, you can create temporary objects using direct initialization. For example:

Foo (1, 2); //  temporary Foo, direct-initialized with (1, 2) (similar to `Foo { 1, 2 }`)

Putting aside the fact that it looks like a function call at first glance, this produces the same result as Foo { 1, 2 } (just with no narrowing conversion prevention). Pretty normal right?

We’ll now spend the remainder of this section showing you why you probably shouldn’t do this.

Now let’s look at the case where we don’t have any arguments:

Foo();     // temporary Foo, value-initialized (identical to `Foo {}`)

You probably didn’t expect that Foo() would create a value-initialized temporary just like Foo {} does. And that’s probably because this syntax has a completely different meaning when used with a named variable!

Foo bar{}; // definition of variable bar, value-initialized
Foo bar(); // declaration of function bar that has no parameters and returns a Foo (inconsistent with `Foo bar{}` and `Foo()`)

Ready to get real weird?!?

Foo(1);    // Function-style cast of literal 1, returns temporary Foo (similar to `Foo { 1 }`)
Foo(bar);  // Defines variable bar of type Foo (inconsistent with `Foo { bar }` and `Foo(1)`)

Wait, what?

• The version with literal 1 in parentheses behaves consistently with all the other versions of this syntax that create temporary objects.
• The version with identifier bar in parentheses defines a variable named bar (identical to Foo bar;). If bar is already defined, this will cause a redefinition compile-error.

The compiler knows that literals can’t be used as identifiers for variables, so it’s able to treat that case consistently with the others.

Okay, fun over. Back to the boring stuff.

Temporary objects and return by value

When a function returns by value, the object that is returned is a temporary object (initialized using the value or object identified in the return statement).

Here are some examples:

#include <iostream>

class IntPair
{
private:
    int m_x{};
    int m_y{};

public:
    IntPair(int x, int y)
        : m_x { x }, m_y { y }
    {}

    int x() const { return m_x; }
    int y() const { return m_y; }
};

void print(IntPair p)
{
    std::cout << "(" << p.x() << ", " << p.y() << ")\n";        
}

// Case 1: Create named variable and return
IntPair ret1()
{
    IntPair p { 3, 4 };
    return p; // returns temporary object (initialized using p)
}

// Case 2: Create temporary IntPair and return
IntPair ret2()
{
    return IntPair { 5, 6 }; // returns temporary object (initialized using another temporary object)
}

// Case 3: implicitly convert { 7, 8 } to IntPair and return
IntPair ret3()
{
    return { 7, 8 }; // returns temporary object (initialized using another temporary object)
}
     
int main()
{
    print(ret1());
    print(ret2());
    print(ret3());

    return 0;
}

In case 1, when we return p, a temporary object is created and initialized using p.

The cases in this example are analogous to the cases in the prior example.

A few notes

First, just as in the case of an int, when used in an expression, a temporary class object is an rvalue. Thus, such objects can only be used where rvalue expressions are accepted.

Second, temporary objects are created at the point of definition, and destroyed at the end of the full expression in which they are defined . A full expression is an expression that is not a subexpression.

static_cast vs explicit instantiation of a temporary object

In cases where we need to convert a value from one type to another but narrowing conversions aren’t involved, we often have the option to use either static_cast or explicit instantiation of a temporary object.

For example:

#include <iostream>

int main()
{
    char c { 'a' };

    std::cout << static_cast<int>( c ) << '\n'; // static_cast returns a temporary int direct-initialized with value of c
    std::cout << int { c } << '\n';             // explicitly creates a temporary int list-initialized with value c

    return 0;
}

static_cast<int>(c) returns a temporary int that is direct-initialized with the value of c. int { c } creates a temporary int that is list-initialized with the value of c. Either way, we get a temporary int initialized with the value of c, which is what we want.

Let’s show a slightly more complex example:

printString.h:

#include <string>
void printString(const std::string &s)
{
    std::cout << s << '\n';
}

main.cpp:

#include "printString.h"
#include <iostream>
#include <string>
#include <string_view>

int main()
{
    std::string_view sv { "Hello" };

    // We want to print sv using the printString() function
    
//    printString(sv); // compile error: a std::string_view won't implicitly convert to a std::string

    printString( static_cast<std::string>(sv) ); // Case 1: static_cast returns a temporary std::string direct-initialized with sv
    printString( std::string { sv } );           // Case 2: explicitly creates a temporary std::string list-initialized with sv
    printString( std::string ( sv ) );           // Case 3: C-style cast returns temporary std::string direct-initialized with sv (avoid this one!)

    return 0;
}

Let’s say the code in header file printString.h isn’t code we can modify (e.g. because it’s distributed with some 3rd party library we’re using, and has been written to be compatible with C++11, which doesn’t support std::string_view). So how do we call printString() with sv? Because a std::string_view won’t implicitly convert to a std::string (for efficiency reasons), we can’t just use sv as an argument. We must use some explicit form of conversion.

In case 1, static_cast<std::string>(sv) invokes the static_cast operator to cast sv to a std::string. This returns a temporary std::string that has been direct-initialized using sv, which is then used as the argument for the function call.

In case 2, std::string { sv } creates a temporary std::string that is list-initialized using sv. Since this is an explicit construction, the conversion is allowed. This temporary is then used as the argument for the function call.

In case 3, std::string ( sv ) use a C-style cast to cast sv to a std::string. Although this works here, C-style casting can be dangerous in general and should be avoided. Notice how similar this looks to the prior case!