[libc++][ranges] ranges::to can have non-class return type

[frederick-vs-ja]

Currently, this ill-formed program which violates [range.utility.conv.to]/1 and [range.utility.conv.adaptors]/1 isn't rejected when using libc++ (Godbolt link):

#include <concepts>
#include <ranges>

int main() {
  struct R {
    int* begin() const;
    int* end() const;

    operator int() const { return 0; }
  };

  std::same_as<int> auto m [[maybe_unused]] = std::ranges::to<int>(R{});
  std::same_as<int> auto n [[maybe_unused]] = R{} | std::ranges::to<int>();
}

It's weird that the "class" part is mentioned in comments but not actually implemented.

llvm-project/libcxx/include/__ranges/to.h

Lines 80 to 83 in ead2724

	// Mandates: C is a cv-unqualified class type.
	static_assert(!is_const_v<_Container>, "The target container cannot be const-qualified, please remove the const");
	static_assert(
	!is_volatile_v<_Container>, "The target container cannot be volatile-qualified, please remove the volatile");

llvm-project/libcxx/include/__ranges/to.h

Lines 207 to 210 in ead2724

	// Mandates: C is a cv-unqualified class type.
	static_assert(!is_const_v<_Container>, "The target container cannot be const-qualified, please remove the const");
	static_assert(
	!is_volatile_v<_Container>, "The target container cannot be volatile-qualified, please remove the volatile");

Note that libstdc++ and MSVC STL also reject union types by only using is_class_v, despite that the term "class type" in the C++ core language specification also covers union types. Perhaps we should also only accept non-union class types here and submit an LWG issue for this.

Edit: LWG4229

[philnik777]

I don't think we should reject unions. There is no fundamental difference between struct S { union {} } and union S {}, but we would reject the latter while accepting the former.