Strange Move Assignment Operator Signature

Question

I came across an unfamiliar move assignment operator signature in Pytorch' tensor backend (ATen, source). Just out of curiosity, what does the && operator do at the end of

Tensor & Tensor::operator=(Tensor && rhs) &&

While I'm familiar with move semantics and the usual copy/move constructor and assignment operator signatures, I could not find any documentation online about the syntax above.

I would be grateful if someone could explain what this operator does, how it differs from the usual move assignment operation, and when it should be used.

Answer 1

Objects of a class used as expressions can be rvalues or lvalues. The move assignment operator is a member function of a class.

This declaration

Tensor & Tensor::operator=(Tensor && rhs) &&

means that this move assignment operator is called for rvalue object of the class.

Here is a demonstrative program.

#include <iostream>

struct A
{
    A & operator =( A && ) &
    {
        std::cout << "Calling the move assignment operator for an lvalue object\n";
        return *this;
    }

    A & operator =( A && ) &&
    {
        std::cout << "Calling the move assignment operator for an rvalue object\n";
        return *this;
    }
};

int main() 
{
    A a;

    a = A();

    A() = A();

    return 0;
} 

The program output is

Calling the move assignment operator for an lvalue object
Calling the move assignment operator for an rvalue object

That is in this statement

    a = A();

the left hand operand of the assignment is an lvalue.

In this statement

    A() = A();

the left hand operand of the assignment is rvalue (a temporary object).