Re: [boost] Re: Variant implementation change (was re: unions) final summary

From: Douglas Gregor (gregod_at_[hidden])
Date: 2003-10-09 15:45:53

On Thursday 09 October 2003 04:17 pm, E. Gladyshev wrote:
> --- Douglas Gregor <gregod_at_[hidden]> wrote:
> > Please do note that you are refuting something that is considered to have
> > already been proven in prior discussions, which have been cited. In this
> > circumstance, the burden of proof falls on you.
>
> I read his message. I don't see any prove there just a statement from
> the standard that doesn't prove anything.
[snip]
> I repeat it again all my code is doing is copying one char[]
> to another char[]. Both variables have been allocated
> as char[] originally there is not any casting going on.

I'll spell it out, then. Your code starts with:

        variant_copy( b1, data_, size );
        
        try
        {
            //place new object in _data
            T* obj = new(data_) T;
            // ...

At the line

        T* obj = new (data_) T;

you've ended the lifetime of the object that used to be in data_. The clauses 
David A. quotes state that specifically. It doesn't matter that you saved the 
bits at that location, because the object's lifetime has ended when the 
storage was reused (by constructing a new object there): you can't bring it 
back to life by copying the bits back where the object should be. It's okay 
for POD types, but not for non-POD types.

        Doug