direct initialization

From cppreference.com
< cpp‎ | language

Initializes an object from explicit set of constructor arguments.

Syntax

T object ( arg );

T object ( arg1, arg2, ... );

(1)
T object { arg }; (2) (since C++11)
T ( other )

T ( arg1, arg2, ... )

(3)
static_cast< T >( other ) (4)
new T(args, ...) (5)
Class::Class() : member(args, ...) { ... } (6)
[arg](){ ... } (7) (since C++11)

Explanation

Direct initialization is performed in the following situations:

1) initialization with a nonempty parenthesized list of expressions or braced-init-lists (since C++11)
2) initialization of an object of non-class type with a single brace-enclosed initializer (note: for class types and other uses of braced-init-list, see list-initialization)
3) initialization of a prvalue temporary by functional cast or with a parenthesized expression list
4) initialization of a prvalue temporary by a static_cast expression
5) initialization of an object with dynamic storage duration by a new-expression with a non-empty initializer
6) initialization of a base or a non-static member by constructor initializer list
7) initialization of closure object members from the variables caught by copy in a lambda-expression

The effects of direct initialization are:

  • If T is an array type,
  • The program is ill-formed
(until C++20)
struct A { explicit A(int i = 0) {} };
A a[2](A(1)); // OK: initializes a[0] with A(1) and a[1] with A()
A b[2]{A(1)}; // error; implicit copy-list-initialization of a[1]
              //        from {} selected explicit constructor
(since C++20)
  • If T is a class type,
  • if the initializer is a prvalue expression whose type is the same class as T (ignoring cv-qualification), the initializer expression itself, rather than a temporary materialized from it, is used to initialize the destination object: see copy elision
(since C++17)
  • the constructors of T are examined and the best match is selected by overload resolution. The constructor is then called to initialize the object.
  • otherwise, if the destination type is a (possibly cv-qualified) aggregate class, it is initialized as described in aggregate initialization except that narrowing conversions are permitted, designated initializers are not allowed, a temporary bound to a reference does not have its lifetime extended, there is no brace elision, and any elements without an initializer are value-initialized.
struct B {
  int a;
  int&& r;
};
 
int f();
int n = 10;
 
B b1{1, f()};               // OK, lifetime is extended
B b2(1, f());               // well-formed, but dangling reference
B b3{1.0, 1};               // error: narrowing conversion
B b4(1.0, 1);               // well-formed, but dangling reference
B b5(1.0, std::move(n));    // OK
(since C++20)
  • Otherwise, if T is a non-class type but the source type is a class type, the conversion functions of the source type and its base classes, if any, are examined and the best match is selected by overload resolution. The selected user-defined conversion is then used to convert the initializer expression into the object being initialized.
  • Otherwise, if T is bool and the source type is std::nullptr_t, the value of the initialized object is false.
  • Otherwise, standard conversions are used, if necessary, to convert the value of other to the cv-unqualified version of T, and the initial value of the object being initialized is the (possibly converted) value.

Notes

Direct-initialization is more permissive than copy-initialization: copy-initialization only considers non-explicit constructors and non-explicit user-defined conversion functions, while direct-initialization considers all constructors and all user-defined conversion functions.

In case of ambiguity between a variable declaration using the direct-initialization syntax (1) (with round parentheses) and a function declaration, the compiler always chooses function declaration. This disambiguation rule is sometimes counter-intuitive and has been called the most vexing parse.

#include <iterator>
#include <string>
#include <fstream>
int main()
{
    std::ifstream file("data.txt");
    // the following is a function declaration:
    std::string str(std::istreambuf_iterator<char>(file),
                    std::istreambuf_iterator<char>());
    // it declares a function called str, whose return type is std::string,
    // first parameter has type std::istreambuf_iterator<char> and the name "file"
    // second parameter has no name and has type std::istreambuf_iterator<char>(),
    // which is rewritten to function pointer type std::istreambuf_iterator<char>(*)()
 
    // pre-c++11 fix: extra parentheses around one of the arguments
    std::string str( (std::istreambuf_iterator<char>(file) ),
                      std::istreambuf_iterator<char>());  
    // post-C++11 fix: list-initialization for any of the arguments
    std::string str(std::istreambuf_iterator<char>{file}, {});
}

Similarly, in the case of an ambiguity between a expression statement with a function-style cast expression (3) as its leftmost subexpression and a declaration statement, the ambiguity is resolved by treating it as a declaration. This disambiguation is purely syntactic: it doesn't consider the meaning of names occurring in the statement other than whether they are type names.

struct M { };
struct L { L(M&); };
 
M n;
void f() {
    M(m); // declaration, equivalent to M m;
    L(n); // ill-formed declaration
    L(l)(m); // still a declaration
}

Example

#include <string>
#include <iostream>
#include <memory>
 
struct Foo {
    int mem;
    explicit Foo(int n) : mem(n) {}
};
 
int main()
{
    std::string s1("test"); // constructor from const char*
    std::string s2(10, 'a');
 
    std::unique_ptr<int> p(new int(1)); // OK: explicit constructors allowed
//  std::unique_ptr<int> p = new int(1); // error: constructor is explicit
 
    Foo f(2); // f is direct-initialized:
              // constructor parameter n is copy-initialized from the rvalue 2
              // f.mem is direct-initialized from the parameter n
//  Foo f2 = 2; // error: constructor is explicit
 
    std::cout << s1 << ' ' << s2 << ' ' << *p << ' ' << f.mem  << '\n';
}

Output:

test aaaaaaaaaa 1 2

See also