Reputation: 4453
Consider the following example struct
:
struct A {
int i;
void init(const A &a)
{
this->i = a.i;
}
A(int i)
{
this->i = i;
}
A(const A &a) = delete;
A &operator=(const A &a) = delete;
A(A &&a)
{
init(a); // Is this allowed?
}
A &operator=(A &&a)
{
init(a); // Is this allowed?
return *this;
}
};
The rvalue reference A &&a
is passed on to a function accepting a const A &a
, i.e. a constant lvalue reference. Is this allowed and resulting in well defined behaviour in C++?
Upvotes: 1
Views: 308
Reputation: 3258
You need to cast a
to an rvalue in your example to get the intended effect, because variable names themselves are lvalues (this is a tricky detail of lvalues and rvalues in C++). So as written, it's correct C++ but not doing what you think it's doing.
If you cast it, using std::move(a)
instead of just a
, the code now does what you want and is still correct. This is because of a special rule in C++ that temporaries can be bound to const lvalues, a more detailed discussion of which can be found here. This features comes as a great convenience when you have code like:
void ProcessData(const std::vector<int>& input_vector);
and then you want to test it with the following:
ProcessData(std::vector<int>{1, 2, 3, 4, 5});
This saves you from having to explicitly create the object before passing it on as a const lvalue reference. Note the const here is critical, without it the code is incorrect. There is a more detailed discussion of this rationale for this choice here.
Upvotes: 1
Reputation: 29962
Yes, it is allowed.
Note, that value category of the expression a
is lvalue, even though a
's declared type is rvalue reference.
Furthermore, if you create an rvalue from a
with std::move
, your code is still well-formed, as an rvalue can be bound to a const lvalue reference:
init(std::move(a)); // std::move(a) is an rvalue (to be precise, xvalue), but still OK
Upvotes: 1