Re: [boost] limited form of type deduction

From: Peter Dimov (pdimov_at_[hidden])
Date: 2003-10-11 13:23:48

Brian McNamara wrote:
> On Sat, Oct 11, 2003 at 08:37:24PM +0300, Peter Dimov wrote:
>> Joel de Guzman wrote:
>>
>>>         typedef result_of_plus<int, double>::type result;
>>
>> You should name that
>>
>>     typedef result_of<plus(int, double)>::type result;
>>
>> instead. :-)
>
> There is a smiley, but seriously, if you did
>
>    struct op_plus {
>       template <class R> struct result;
>       template <class F, class A1, class A2> struct result<F(A1,A2)> {
>          typedef typename result_of_plus<A1,A2>::type type;
>       };
>    };
>    // etc.
>
> then
>
>    typedef result_of< op_plus(int, double) >::type result;
>
> would truly work.  (Names like op_plus are just used as tags which
> stand for the corresponding C++ operators.)

Or even:

struct plus;

template<class A1, class A2> struct result_of<plus(A1, A2)>
{
    // result_of_plus<A1, A2> goes here
};

struct plus
{
    template<class A1, class A2> typename result_of<plus(A1, A2)>::type
        operator()(A1 const & a1, A2 const & a2) { return a1 + a2; }
};

Which one is more "correct" is not clear. It's better to leave result_of
unspecialized if your wonderful TR1 implementation always does the right
thing. It's better to specialize it otherwise, since the nested result<>
forwarding sometimes "breaks" SFINAE.

Let's just hope that one day the proper decltype() will arrive and render
the issue moot.