Re: implicit_cast (was: Re: [boost] Digest Number 365)

From: Beman Dawes (bdawes_at_[hidden])
Date: 2001-01-13 08:48:31

At 11:46 AM 1/13/2001 +0000, Stephen Silver wrote:
 >Beman Dawes wrote:
 >
 >>At 04:05 PM 1/12/2001 -0500, George A. Heintzelman wrote:
 >
 >> >template <class T> rational(T n) :
 >> > num(boost::implicit_cast<IntType>(n)),
 >> > den(IntType(1)) {} // Or whatever you finally decide to use for 1.
 >> >
 >> >I seem to remember some discussion about exactly this sort of need, 
and
 >> >had thought that it was in the BCL, but it doesn't seem to be. Perhaps
 >> >this is an example which suggests it needs to be added?
 >>
 >>We removed implicit_cast<> because it had a number of problems.  The
 >>comment in the header says:
 >>
 >>// Removed due to uncertain purpose. Use either numeric_cast (see below)
 >>// or static_cast according to the need.
 >
 >This comment makes no sense to me.  static_cast can make use of an
 >explicit constructor (ISO 5.2.9 / 2), so it can never be used in any
 >situation for which implicit_cast is intended.  And numeric_cast is
 >something quite different.
 >
 >In any case, neither static_cast nor numeric_cast is suitable in the
 >above situation, but implicit_cast is.  What were the problems with
 >implicit_cast?

Search the archive for "implicit cast" and you will find a bunch of 
messages.  Read the thread starting about message 889, for example.

IIRC, the best shot at getting it to work correctly was Darin Adler's:

template<typename T> struct mirror_type { typedef T type; }; 
template<typename T>
inline T implicit_cast( typename mirror_type<T>::type t ) { return t; }

--Beman