Disable explicit conversion for quantities of unrelated units

[kaihowl]

This fixes #36.

[kaihowl]

It seems that the windows builds are not running on a PR. It this on purpose? Should I look at them more closely?

[swatanabe]

AMDG This is a bit more complex than it needs to be. has_conversion_helper uses SFINAE to get a bool and then you pass the bool to enable_if to do more SFINAE. It's simpler to use SFINAE once: template<class Q1, class Q2> using require_conversion_factor = void_t<decltype(conversion_factor(Q1{}, Q2{}))>; template<class Q1, class Q2> using require_conversion_helper = void_t<decltype(conversion_helper<Q1, Q2>::convert(Q1{}))> This of course, only works in C++11, but I think you need to make it conditional anyway, to avoid the possibility of errors from over-eager evaluation of the SFINAE test. In Christ, Steven Watanabe

[kaihowl]

This solution was meant to work on C++03 as well. I do see that Travis does not like it anyway.

When I use the proposed solution by you, I assume that BOOST_NO_CXX11_DECLTYPE should be used to wrapped this then?

[swatanabe]

AMDG

On 2/25/19 3:06 AM, Kai Howelmeyer wrote: This solution was meant to work on C++03 as well. I do see that Travis does not like it anyway.

To make it work in C++03, you need to avoid SFINAE on expressions. // limit the scope of the extra overload namespace has_conversion_factor_impl { // a fallback overload that will be found only // if no other overload of conversion_factor exists. struct no_conversion_factor {}; no_conversion_factor conversion_factor(...); // identify whether the fallback was used. // conversion_factor should never return void, // so we don't need `operator,` magic. no& test(no_conversion_factor); yes& test(...); template<class Q1, class Q2> struct has_conversion_factor { enum { value = sizeof((test)(conversion_factor( declval<Q1>(), declval<Q2>()))) == sizeof(yes) }; }; } using has_conversion_factor_impl::has_conversion_factor; (Bleh.) has_conversion_helper can be handled by modifying the primary template of conversion_helper to make it possible to identify: template<class Q1, class Q2> struct conversion_helper { typedef void _is_primary_template; }; template<class Q1, class Q2, class = void> struct has_conversion_helper : true_type {}; template<class Q1, class Q2> struct has_conversion_helper<Q1, Q1, typename conversion_helper<Q1, Q2>::_is_primary_template> : false_type {};

When I use the proposed solution by you, I assume that `BOOST_NO_CXX11_DECLTYPE` should be used to wrapped this then?

and BOOST_NO_SFINAE_EXPR and BOOST_NO_CXX11_TEMPLATE_ALIASES. In Christ, Steven Watanabe

[kaihowl]

How would you attach the template type alias in the C++11 version for require_conversion_factor to the conversion_helper template? If I attach it as a default template parameter argument to the class, I get an ambiguity for absolute units. If I attach it to the convert function, I cannot do SFINAE anymore for the require_conversion_helper as this is not the immediate context anymore where the substitution fails. I am stuck. (Yet again.)

[swatanabe]

AMDG

On 3/7/19 8:17 AM, Kai Howelmeyer wrote: How would you attach the template type alias in the C++11 version for `require_conversion_factor` to the `conversion_helper` template? If I attach it as a default template parameter argument to the class, I get an ambiguity for absolute units. If I attach it to the `convert` function, I cannot do SFINAE anymore for the `require_conversion_helper` as this is not the immediate context anymore where the substitution fails. I am stuck. (Yet again.)

You need either C++17 void_t: template<class...> using void_t = void; (depends on CWG 1558) or extra indirection (make conversion_helper inherit from another template and put the SFINAE there.) In Christ, Steven Watanabe