Re: [boost] dangerous_cast<>

From: David Abrahams (dave_at_[hidden])
Date: 2002-12-04 11:23:30

Gabriel Dos Reis <gdr_at_[hidden]> writes:

> David Abrahams <dave_at_[hidden]> writes:
>
> | Gabriel Dos Reis <gdr_at_[hidden]> writes:
> | 
> | > "David B. Held" <dheld_at_[hidden]> writes:
> | >
> | > | Does this have any merit whatsoever?
> | > | 
> | > | template <typename T, typename U>
> | > | T* dangerous_cast(U* p)
> | > | {
> | > |     return static_cast<T*>(static_cast<void*>(p));
> | > | }
> | >
> | > Is it any different from reinterpret_cast<T*>(p) ?
> | 
> | It might be, depending on your compiler. The behavior of
> | reinterpret_cast<T*> is implementation-defined.
>
> The mapping used by reinterpret_cast<> is implementation defined,
> thus an implementation is required to document it.
>
>  1) Do you have any corresponding guarantee for converting a void* to
>     T* when the source is *not* a T* nor the address obtained from a T?
>
>  2) Isn't the mapping U* -> void* -> T* at best
>     implementation-defined if not undefined behaviour? 

template <class T>
struct holder
{
   union {
      aligned_POD<T>::type aligner;
      char storage[sizeof(T)];
   };
};

holder<Foo> h;
new (h.storage) Foo;
   ...
T* u = dangerous_cast<T*>(h.storage); // can't do this with reinterpret_cast

HTH,
-- 
                       David Abrahams
   dave_at_[hidden] * http://www.boost-consulting.com
Boost support, enhancements, training, and commercial distribution