[v3] ext/type_traits.h

[Paolo Carlini]

Benjamin Kosnik wrote:

>>I want to say that whereas, per your citation, I also called the trait 
>>to_* when I "invented" ;) it, the new naming scheme may have advantages: 
>>we are directing attention to the syntactic notion of adding the 
>>""unsigned qualification"" or removing it, not to the actual signedness 
>>of the types at issue in input and output (I hope I'm sufficiently 
>>clear).
>>    
>>
>I think so.
>
Ok...

> FYI this is the reason I think __add_unsigned is better than __add_signed.
>  
>
... but isn't __add_signed necessary anyway, from out point of view? 
Imean, we would have,

 template<typename _Value>
   struct __add_signed
   { typedef _Value __type; };

 template<>
   struct __add_signed<char>
   { typedef signed char __type; };

   ...

 template<typename _Value>
   struct __remove_signed
   { typedef _Value __type; };

 template<>
   struct __remove_signed<signed char>
   { typedef char __type; };

   ...

Indeed, for short, int, and so on, the ""signed qualifier"" is implicit 
(I thnink that was your point), from a semantic point of view, but 
syntactically, our route in fact, only adding both appear to lead to a 
consistent and complete set. About the latter, I was thinking yesterday, 
that, if I'm not mistaken, we could also have:

 template<typename _Value>
   struct __to_signed
   {
        typedef typename
          __add_signed<typename 
__remove_unsigned<_Value>::__type>::__type __type;
    };

 template<typename _Value>
   struct __to_unsigned
   {
        typedef typename
          __add_unsigned<typename 
__remove_signed<_Value>::__type>::__type __type;
    };

Paolo.