From: David B. Held (dheld_at_[hidden])
Date: 2002-08-12 14:22:34
"Philippe A. Bouchard" <philippeb_at_[hidden]> wrote in message
news:aj8m3k$oce$1_at_main.gmane.org...
> > I'm not sure if this ptr<> type under discussion works this way, but in
> > general you can't conformingly use char* arithmetic with offsets to find
> > the sub-parts of non-POD types.
>
> The offsets are what static_cast<>s return. I cheat in the syntax when I
> say ptr_base<T, __false_type>::m_ptr is of type T *; it's just to make the
> code cleaner.
Umm...my understanding is that what Dave is talking about is exactly
what you are doing. After all, what is the business with
reinterpret_cast<char*> to do offset arithmetic?
> [...]
> template <typename T>
> struct ptr_block
> {
> int m_counter;
> int m_typeid;
>
> union
> {
> int;
> T;
> } m_object;
> };
>
> But operator new (size_t, rc_type const &) won't be easy to implement
> because size_t is not a compile-time constant.
I spent a while trying to see if it was possible to make this "invisible
count
wrapper", but ran into the same problem: you need to know the type,
because operator new only provides the size of the type as a runtime
value. I don't see how you can make a safe, conforming, portable count
wrapper without explicitly naming the type somewhere (like your original
placement new operator which passed in the type).
Dave