Re: [boost] Problem solved: EBO pessimization in smart_ptr not an issue anymore!

From: Howard Hinnant (hinnant_at_[hidden])
Date: 2002-05-02 08:58:23

On Wednesday, May 1, 2002, at 11:46  PM, Andrei Alexandrescu wrote:

>  Also, I saw a very similar trick,
> if not the very same trick (can't check now, I'm on a very slow
> connection) in compressed_pair.
>
> The idea is to use derivation only for those policies that are
> nonempty, and
> to simply not derive at all from policies that are empty.

Dave already pointed this out, but I wanted to confirm:  compressed_pair 
doesn't work this way.  compressed_pair derives from empty classes and 
contains non-empty types as members.

We thought about building compressed_pair the way you describe, but 
because of the lack of running the empty class constructors and 
destructors decided it was not a good idea.  A constructor or destructor 
of an empty class could still effect the state of a program (class 
static data or a global variable) and thus needs to behave as if it is a 
member.

This is not to say that the technique you outlined would not be suitable 
for a smart pointer.  But the interface (docs) should state:  
constructors/destructors of empty policies are not allowed to do 
*anything*.

-Howard