I encountered some divergent behavior between clang and gcc recently regarding void_t property detection and protected/private class information. Consider a type trait defined as follows:
#include <type_traits>
template<typename T, typename = void>
constexpr const bool has_nested_type_v = false;
template<typename T>
constexpr const bool has_nested_type_v
<T, std::void_t<typename T::type>> = true;
Given example types with protected or private nested type classes and a simple program
#include "has_nested_type.hpp"
#include <iostream>
struct Protected {
protected:
struct type{};
};
struct Private {
private:
struct type{};
};
int main() {
std::cout << "Protected: "
<< (has_nested_type_v<Protected> ? "detected" : "not detected")
<< std::endl;
std::cout << "Private: "
<< (has_nested_type_v<Private> ? "detected" : "not detected")
<< std::endl;
}
clang compiles successfully with a failed detection (as expected). The program, compilation, and output are reproduced on wandbox here.
gcc fails to compile, issuing a diagnostic. This error can be reproduced on wandbox here.
GCC issues the following error for this program.
prog.cc:16:21: error: 'struct Protected::type' is protected within this context
<< (has_nested_type_v<Protected> ? "detected" : "not detected")
^~~~~~~~~~~~~~~~~~~~~~~~~~~~
prog.cc:6:14: note: declared protected here
struct type{};
^~~~
prog.cc:19:21: error: 'struct Private::type' is private within this context
<< (has_nested_type_v<Private> ? "detected" : "not detected")
^~~~~~~~~~~~~~~~~~~~~~~~~~
prog.cc:11:14: note: declared private here
struct type{};
^~~~
My question is which behavior is standard conformant? Should clang error here and issue a diagnostic or is GCC being overly eager?