They are useful for both providing safety and optimizations.
For member functions returning a pointer to something they own (either directly or via view types like std::string_view
or std::span
), disabling the rvalue overloads can prevent errors:
struct foo {
int* take_ptr_bad() { return &x; }
int* take_ptr() & { return &x; }
int x;
};
foo get_foo();
void bar() {
auto ptr = get_foo().take_ptr_bad();
// ptr is dangling
auto ptr = get_foo().take_ptr();
// does not compile
}
The other is to provide some optimizations. For instance, you might overload a getter function to return an rvalue reference if this
is an rvalue to prevent unnecessary copies:
struct foo {
const std::string& get_str() const & {
return s;
}
std::string&& get_str() && {
return std::move(s);
}
std::string s;
};
void sink(std::string);
foo get_foo();
void bar() {
sink(get_foo().get_str());
// moves the string only if the r-value overload is provided.
// otherwise a copy has to be made, even though the foo object
// and transitively the string is temporary.
}
These are how I use the feature, and I'm sure there are more use cases.