Aggregate initialization basically performs element-wise copy-initialization. So this:
struct Foo {
int i;
std::string str;
};
Foo foo{1, std::string("Hello, world!")};
does the same initializations as:
int i = 1;
std::string str = std::string("Hello, world!");
And we have a new rule in C++17 that says that:
If the initializer expression is a prvalue and the cv-unqualified version of the source type is the same class as the class of the destination, the initializer expression is used to initialize the destination object. [ Example: T x = T(T(T()));
calls the T
default constructor to initialize x
. — end example ]
which means that the second initialization must behave as if you'd written:
std::string str("Hello, world!");
That is, zero copies.
A nice demonstration of the new rule is the following example:
struct X {
X(int ) { }
X(X&& ) = delete;
};
struct Y {
X x;
};
int main() {
Y y{X{4}}; // ill-formed in C++14 due to deleted move ctor
// ok in C++17, no move required
}