Re: [boost] Interesting variadic template trick

Subject: Re: [boost] Interesting variadic template trick
From: Andrey Semashev (andrey.semashev_at_[hidden])
Date: 2013-11-18 00:55:24

On Mon, Nov 18, 2013 at 8:42 AM, Louis Dionne <ldionne.2_at_[hidden]> wrote:
> Louis Dionne <ldionne.2 <at> gmail.com> writes:
>
>>     template <typename ...T> void allow_expansion(T&&...);
>>
>>     template <typename ...T>
>>     struct or_ {
>>         static constexpr bool value = !noexcept(
>>             allow_expansion((T::value ? throw : 0)...)
>>         );
>>     };
>
> How embarrassing; my unit test was very poor and the above trick
> does not seem to work. However, the following was tested more rigorously
> and works (with similar performance improvements):
>
>     template <typename ...T>
>     constexpr bool all_pointers(T*...) { return true; }
>
>     template <typename ...T>
>     constexpr bool all_pointers(T...) { return false; }
>
>     template <typename ...T>
>     struct or_ {
>         static constexpr bool value = !all_pointers(
>             ((typename std::conditional<T::value, int, void*>::type)0)...
>         );
>     };
>
>     template <>
>     struct or_<> {
>         static constexpr bool value = false;
>     };

It requires N instantiations of std::conditional, doesn't it? In that
case is it actually faster than the naive recursive instantiation of
or_?