[Bug c++/81589] New: Possible False-Positive with decltype

["hunter at openrobotics dot org" <gcc-bugzilla at gcc dot gnu dot org>]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=81589

            Bug ID: 81589
           Summary: Possible False-Positive with decltype
           Product: gcc
           Version: 7.1.1
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: c++
          Assignee: unassigned at gcc dot gnu.org
          Reporter: hunter at openrobotics dot org
  Target Milestone: ---

I can confirm that this effects GCC-8.0 svn as well (have not tested with
GCC-6).

There's a header file which compiles just fine with GCC-5.4.0 (and GCC-4.9.4),
but fails to compile with GCC-7.1.1 and GCC-8.0.

Please note that I'm only reporting this because of the inconsistency.

error: decltype cannot resolve address of overloaded function

The problematic line is reported as this one:

using arguments = typename tuple_tail<
  typename function_traits<decltype< &
FuncitionT::operator())>::arguments>::type;

In this context:

#include <functional>
#include <memory>
#include <tuple>

namespace rclcpp
{

namespace runction_traits
{

template<typename T>
struct tuple_tail;

template<typename Head, typename ... Tail>
struct tuple_tail<std::tuple<Head, Tail ...>>
{
  using type = std::tuple<Tail ...>;
};

// std::function
template<typename FunctionT>
struct function_traits
{
  using arguments = typename tuple_tail<
      typename function_traits<decltype( &
FunctionT::operator())>::arguments>::type;

  static constexpr std::size_t arity = std::tuple_size<arguments>::value;

  template<std::size_t N>
  using argument_type = typename std::tuple_element<N, arguments>::type;

  using return_type = typename function_traits<decltype( &
FunctionT::operator())>::return_type;
};

/* other structs follow */
}

}

I'll attach the precompiler output from my computer with GCC 8 on it.

I'll also note that our library, rclcpp, does compile with GCC 8, but, when we
try to compile our tests or use the library elsewhere, the compiler throws
errors.

I'm super hesitant in reporting this, but I just can't see anything wrong in
the code, and the inconsistency is certainly an issue. It should error in GCC
5.4 if it's "bad" code (also note that the code is being compiled with the same
standard, C++-14).