[libc++] Are some standard overloads of `std::pow` missing?

[frederick-vs-ja]

I'm referring to C++20/WG21 N4868 here, because the changes of additional overloads of std::pow introduced in C++23 by P1467R9 look defective and libc++ hasn't implemented P1467R9 yet.

In N4868 [complex.syn] there were 3 explicitly specified standard overloads of pow:

  template<class T> complex<T> pow  (const complex<T>&, const T&);
  template<class T> complex<T> pow  (const complex<T>&, const complex<T>&);
  template<class T> complex<T> pow  (const T&, const complex<T>&);

And then, several additional overloads were specified in N4868 [cmplx.over]/3, but their exact forms were not specified.

Currently, libc++ provides 4 overloads:

llvm-project/libcxx/include/complex

Lines 1095 to 1116 in 1c26e2b

	template <class _Tp>
	inline _LIBCPP_HIDE_FROM_ABI complex<_Tp> pow(const complex<_Tp>& __x, const complex<_Tp>& __y) {
	return std::exp(__y * std::log(__x));
	}
	template <class _Tp, class _Up>
	inline _LIBCPP_HIDE_FROM_ABI complex<typename __promote<_Tp, _Up>::type>
	pow(const complex<_Tp>& __x, const complex<_Up>& __y) {
	typedef complex<typename __promote<_Tp, _Up>::type> result_type;
	return std::pow(result_type(__x), result_type(__y));
	}
	template <class _Tp, class _Up, __enable_if_t<is_arithmetic<_Up>::value, int> = 0>
	inline _LIBCPP_HIDE_FROM_ABI complex<typename __promote<_Tp, _Up>::type> pow(const complex<_Tp>& __x, const _Up& __y) {
	typedef complex<typename __promote<_Tp, _Up>::type> result_type;
	return std::pow(result_type(__x), result_type(__y));
	}
	template <class _Tp, class _Up, __enable_if_t<is_arithmetic<_Tp>::value, int> = 0>
	inline _LIBCPP_HIDE_FROM_ABI complex<typename __promote<_Tp, _Up>::type> pow(const _Tp& __x, const complex<_Up>& __y) {
	typedef complex<typename __promote<_Tp, _Up>::type> result_type;
	return std::pow(result_type(__x), result_type(__y));

If I understood correctly, only the first one is an explicitly specified overload, and the rest are additional overloads. So it seems that the remaining 2 overloads are missing.

However, when I attempted to added them during handling #110235, I found that it's very hard, if not impossible, to portably verify that these overloads are added. Because instantiating std::complex<NonFP> is unspecified and in some cases std::complex<std::complex<T>> would be instantiated when using braced-init-list due to the form of these overloads.
E.g. according to the current standard overload set, std::pow(std::complex<float>(1.0f), {1l, 1l}) would cause pow(const T&, const complex<T>&) (where T is complex<float>) to be detected, which in turn instantiates (probably nonsense) std::complex<std::complex<float>>.

It's not clear to me

1. whether libc++ is effectively conforming now, i.e. lacking of these 2 overloads isn't observable in any conforming program, and
2. whether there's any defect in the specification of explicitly overloads (effectively unchanged since C++20, see [complex.syn]) as they possibly cause std::complex<std::complex<T>> to be instantiated.

Edit: LWG3693 is somehow related.

[ldionne]

I think our overloads that use __promote might have been an undocumented extension we provided by mistake. I'd be curious to try removing them and replace them by the Standard ones.

I don't think it's possible to observe that we don't provide the overloads exactly as specified in the Standard, because those overloads are only enabled for arithmetic types and std::complex<T> can only be instantiated for T = {float, double, long double}. However, it might still be worth getting back in line with what the spec says.

[frederick-vs-ja]

Closing as there at least shouldn't be observable differences. I remembered that I've submitted LWG3693 which is somehow related.