En effet!. I very much appreciate that you folded that work into yours btw.
If I remember correctly, in my design (and yours), the attribute function returns an instance of the type directly,
I think what Cleiton is proposing is that the compiler would return a reflection of the synthetized value

Probably, what I was thinking is "where would be the point of instantiation of the attribute" ?

Other questions:

1 - If you ask the std::meta::attribute n times it will be copied n times ?

2 - constexpr auto y = std::meta::attribute<Y&>(reflexpr(W)) // can I get a reference to Y ? could it be copy free ?

3 - Can I call functions in the attribute that returns consteval literal types ?

4 - Can attributes be template instantiations ?

5 - Attributes in templates will have different instantiations for every instance of the template ?

struct A{};

struct B{ constexpr i; }

[[A(int)]] // <- this is one for all template instantiations ?

template<int B>

[[B{N}]] // <- one per instantiation ? template dependent parameter in attributes ?

struct Z{};

Thanks