[boost] Re: Empty Base Optimization

From: Greg Colvin (gcolvin_at_[hidden])
Date: 1999-11-19 12:01:48

From: scleary_at_[hidden] 
> I have found a slightly different form to be useful in my programming.  I
> originally used Nathan's code, but added typedef's for the template
> parameters, the capability of letting Member be a reference type, and a
> specialization for swap (if Member has specialized swap for efficiency, that
> specialization is lost in the original, even if Base is empty).  Lastly, for
> convenience, I added three additional constructors.  I chose to use the name
> "empty_member".
> 
> I'm including the code I use below, and the docs that I made for it for my
> personal use.  These docs may serve as a useful starting point; feel free to
> use the docs or code in any way you see fit.
> 
> One thing I know aren't in my docs are requirements for Base and Member.
> 
> -Steve
> 
> P.S.  All of the code changes mentioned above (in the first paragraph) I
> would recommend incorporating, except possibly the additional constructors.
> I have used each of them at least once.

Looks interesting, and certainly the additional typedefs are keeping with
out style.  The T&, const T& trick is very clever, but I wonder if it is
actually needed with modern compilers.  That is, given an inlined constructor

  struct X { T t; X(T t):t(t){} };

is the compiler required to make a copy of the T parameter, or can it assign
the T argument directly to the T member?  That is, in the statement

   X x(t);

is the compiler allowed to copy t only once, or must it first push it on
the stack?