Lvalues strongly prefer binding to lvalue references, and similarly rvalue references strongly prefer binding to rvalue references. Modifiable expressions weakly prefer binding to a non-const reference.
So when your compiler is doing overload resolution, it checks if there's an overload that takes an rvalue reference, because that's preferred strongly. In this case since the experssion is a modifiable rvalue, so the rvalue reference overload wins.
There is actually use for const rvalue references, they can be used to make sure something does not bind to an rvalue. Remember that an rvalue binds to a const lvalue reference, hence if you did:
template <typename T> void foo(const T& bar) { /* ... */ }
And called the function with:
foo(createVector());
It'd work fine. However sometimes it is desired to ensure that you can only pass lvalues to a function (this is the case for std::ref
for one). You can achieve this by adding an overload:
template <typename T> void foo(const T&&) = delete;
Remember, rvalues strongly prefer binding to rvalue references, and modifiable expressions prefer weakly binding to non-const references. Since we have a const rvalue-reference, it basically means that every single rvalue will bind to this, hence if you try to pass a rvalue to foo()
, your compiler will give an error. This is the only way to achieve such functionality, and thus is sometimes useful.
与恶龙缠斗过久,自身亦成为恶龙;凝视深渊过久,深渊将回以凝视…