Default template parameter & lambda in unevaluated context: bug or feature?

We consider the goal of creating two different types, using the exact same syntax. This can be easily done with lambdas:

auto x = []{};
auto y = []{};
static_assert(!std::is_same_v<decltype(x), decltype(y)>);

But instead of using lambdas, we are looking for another, more elegant syntax. Here are some tests. We start by defining some tools:

#include <iostream>
#include <type_traits>
#define macro object<decltype([]{})>
#define singleton object<decltype([]{})>

constexpr auto function() noexcept
{
    return []{};
}

template <class T = decltype([]{})>
constexpr auto defaulted(T arg = {}) noexcept
{
    return arg;
}

template <class T = decltype([]{})>
struct object
{
    constexpr object() noexcept {}
};

template <class T>
struct ctad
{
    template <class... Args>
    constexpr ctad(const Args&...) noexcept {}
};

template <class... Args>
ctad(const Args&...) -> ctad<decltype([]{})>;

and the following variables:

// Lambdas
constexpr auto x0 = []{};
constexpr auto y0 = []{};
constexpr bool ok0 = !std::is_same_v<decltype(x0), decltype(y0)>;

// Function
constexpr auto x1 = function();
constexpr auto y1 = function();
constexpr bool ok1 = !std::is_same_v<decltype(x1), decltype(y1)>;

// Defaulted
constexpr auto x2 = defaulted();
constexpr auto y2 = defaulted();
constexpr bool ok2 = !std::is_same_v<decltype(x2), decltype(y2)>;

// Object
constexpr auto x3 = object();
constexpr auto y3 = object();
constexpr bool ok3 = !std::is_same_v<decltype(x3), decltype(y3)>;

// Ctad
constexpr auto x4 = ctad();
constexpr auto y4 = ctad();
constexpr bool ok4 = !std::is_same_v<decltype(x4), decltype(y4)>;

// Macro
constexpr auto x5 = macro();
constexpr auto y5 = macro();
constexpr bool ok5 = !std::is_same_v<decltype(x5), decltype(y5)>;

// Singleton
constexpr singleton x6;
constexpr singleton y6;
constexpr bool ok6 = !std::is_same_v<decltype(x6), decltype(y6)>;

and the following test:

int main(int argc, char* argv[])
{
    // Assertions
    static_assert(ok0); // lambdas
    //static_assert(ok1); // function
    static_assert(ok2); // defaulted function
    static_assert(ok3); // defaulted class
    //static_assert(ok4); // CTAD
    static_assert(ok5); // macro
    static_assert(ok6); // singleton (macro also)

    // Display
    std::cout << ok1 << std::endl;
    std::cout << ok2 << std::endl;
    std::cout << ok3 << std::endl;
    std::cout << ok4 << std::endl;
    std::cout << ok5 << std::endl;
    std::cout << ok6 << std::endl;

    // Return
    return 0;
}

this is compiled with the current trunk version of GCC, with options -std=c++2a. See result here in compiler explorer.

The fact that ok0, ok5 and ok6 work are not really a surprise. However, the fact that ok2 and ok3 are true while ok4 is not is very surprising to me.

• Could someone provide an explanation of the rules that make ok3 true but ok4 false?
• Is it really how it should work, or this is a compiler bug concerning an experimental feature (lambdas in unevaluated contexts)? (reference to the standard or to C++ proposals are very welcomed)

Note: I really hope this is a feature and not a bug, but just because it makes some crazy ideas implementable