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

From: E. Gladyshev (egladysh_at_[hidden])
Date: 2003-10-10 15:16:53

--- Douglas Gregor <gregod_at_[hidden]> wrote:
[...]

> ones that prove ways in which the problem can't be solved. You have 
> categorically refused to do your share to understand the old arguments, and 
> in doing so have closed our minds.

Here is another consideration that shows (I believe)
that your conclusions about undefined behavior and 
object lifetime are inconsistent.

struct my_type {...};

/*
5.3.4/8
[...]
[Note:  since  allocation  functions are assumed to return pointers to
  storage that is appropriately aligned for objects of  any  type,  this
  constraint  on  array  allocation overhead permits the common idiom of
  allocating character arrays into which objects  of  other  types  will
  later be placed.  ]
*/

char buf[sizeof(my_type)];  //5.3.4/8 -> can safely allocate my_type in buf

my_type* p1 = new( buf ) my_type;

/* 
according to 5.3.4/9
  "The address of the created object will not necessarily be the same as that
  of the block of storage if the object is an array."

We are not allocating an array, so p == buf
*/

p1->f();  //ok

my_type* p2 = new( buf ) my_type;  //reusing *p1 storage

// again p2 == buf

p2->f(); //ok
p1->f();  //undefined -- lifetime of *p1 is ended

Note: p1 is equal p2 and they pointing to the same type, 
but p1->f() is undefined while p2->f() is fine.

It is nonsense!

What else can I do to open your minds? :)

Eugene

__________________________________
Do you Yahoo!?
The New Yahoo! Shopping - with improved product search
http://shopping.yahoo.com