boost::future_errc not usable

[steffenb7333]

Compiler: gcc 9.3.0, clang++ 9.0.1 / c++17 mode
Boost: 1.73

#include <boost/thread.hpp>
auto future_errc_compile_error(const boost::future_error& e)
{
	return e.code() == boost::future_errc::broken_promise;
}
...

This fails with

/usr/include/boost/system/error_code.hpp:595:32: error: 'make_error_code' was not declared in this scope, and no declarations were found by argument-dependent lookup at the point of instantiation [-fpermissive]
/usr/include/boost/thread/futures/future_error_code.hpp:47:5: note: 'boost::system::error_code boost::system::make_error_code(boost::future_errc)' declared here, later in the translation unit
47 | make_error_code(future_errc e) BOOST_NOEXCEPT

[kuzkry]

Hi, I investigated this a little bit and discovered that it can be simplified to such code:

namespace boost {
namespace system {

class error_code
{
public:
  constexpr error_code() noexcept = default;

  template<class ErrorCodeEnum>
  constexpr error_code( ErrorCodeEnum e ) noexcept
  {
    *this = make_error_code( e );
  }
};
} // namespace system

enum class future_errc
{
  broken_promise = 1
};

namespace system
{
inline error_code make_error_code(future_errc e) noexcept
{
  return error_code{};
}
} // namespace system
} // namespace boost

int main()
{
  boost::system::error_code(boost::future_errc::broken_promise);
}

My understanding is that ADL cannot find future_errc inside boost::system and a potentially matching declaration of make_error_code is found later than used.
I don't think we can really fix the problem by adding a declaration above its use because error_code is defined inside boost::system, which is already a dependency for boost::thread - so boost::system code should contain no references to boost::thread unfortunately.

I figured out we could trick ADL a little bit by introducing a helper namespace and I'm trying to deliver a PR #390 based on that idea.

[pdimov]

make_error_code should be defined in the namespace of the enum, and not in boost::system. There's no need for a helper namespace.

[kuzkry]

Hi @pdimov, thanks your interest and sorry for the long wait.

Yes, you're right that make_error_code should be there (at least originally), but I didn't want to break existing code that someone might rely on, thus I implemented this helper namespace for backward compatibility.

So... I'm just pointing that out but I'm open for making an exception here and do it like you proposed so the code is more simplified. Please let me know what you think.