Sorry, brainfart, I was wrong.
My mental model broke down when I considered cv-qualified types.

// Robin

On Sun, Apr 13, 2025, 19:44 Jonathan Wakely <cxx_at_[hidden]> wrote:

>
>
> On Sun, 13 Apr 2025, 17:22 Jonathan Wakely, <cxx_at_[hidden]> wrote:
>
>>
>>
>> On Sun, 13 Apr 2025, 16:46 Robin Savonen Söderholm via Std-Proposals, <
>> std-proposals_at_[hidden]> wrote:
>>
>>> But '&&' does not mean 'forwarding reference'. It means 'rvalue
>>> reference'. It's just that an rvalue reference of an lvalue reference is an
>>> lvalue reference.
>>>
>>
>> That's not really true. It does have special magic rules in the context
>> of a deduced template parameter. It's not just reference collapsing.
>>
>
> To expand on this, you cannot generally bind an rvalue reference to an
> lvalue:
>
> X x;
> X&& r = x; // ill-formed
>
> But you can with a T&& function parameter when T is a deduced type:
>
> template<typename T> void f(T&&);
> f(x);
>
> Or simply:
>
> auto&& r = x;
>
> If the && in these deduction cases was just an rvalue reference then they
> wouldn't work. It requires a special language rule that deduces either an
> lvalue or rvalue type. So there really is a special thing that we call a
> forwarding reference, which is a reference to a deduced type.
>
>