[boost] Re: assert in inline function w/o ODR violation

From: Beman Dawes (beman_at_[hidden])
Date: 1999-09-02 10:27:38

Kevlin Henney wrote:

>     However, as it probably now stands (I'm looking at code from
>     polymorphic_cast as was), it is not actually particularly safe.

Yes, and we should change the name to down_cast, or whatever, unless
there is some magic way to make it totally safe.

>     If
>     NDEBUG is set, I lose all of the compile time checking as well
as the
>     runtime checking -- baby has been thrown out w/ the bath water!
This
>     can be fixed, and at no extra execution cost, by adding a dummy
>     dynamic_cast that is parsed and checked, but is not executed
and -- on
>     a good compiler -- will not have code gen'd for it either:
>
>        template<class Derived, class Base>
>        inline Derived safe_downcast(Base* x)
>        {
>            assert(dynamic_cast<Derived>(x) != 0);
>            if(false)
>                dynamic_cast<Derived>(x);
>            return static_cast<Derived>(x);
>        }
>
>     Thoughts?

Kevlin, after wading throught the standard's description of
dynamic_cast twice, I am still having trouble knowing which errors
your added code will detect, and which will still get through.  Could
you enlighten me?

--Beman