Re: [boost] Re: Re: ptr<> renamed to placed_ptr<>

From: Larry Evans (jcampbell3_at_[hidden])
Date: 2002-08-30 18:08:38

Philippe A. Bouchard wrote:

>Larry Evans wrote:
>
>  
>
[snip]

>Reconstruction is less important but still is making the biggest difference 
>(doubled).  The other sectors are similar between shared_ptr & placed_ptr:
>http://groups.yahoo.com/group/boost/files/ptr/benchmark.3.txt
>
>Copying becomes a burden for shared_ptr on threaded (default) environments. 
> Swaps are null for placed_ptr:
>http://groups.yahoo.com/group/boost/files/ptr/benchmark.4.txt
>  
>
I'm convinced.  I tried the benchmark again and got different results a 
few moments
apart.  After several more runs,  placed_ptr, in the vast majority of 
runs, performed
better than shared_ptr.

I've tried to figure what David was talking about.  I've looked at 
boost::bind and ref,
but I've yet to figure how it can avoid the problems you objected to, 
i.e. the need to
forward CTOR arguments.  Maybe you can figure it out.

Thanks for the effort.  BTW,  there's a lively discussion about gc and 
reference
counting in the gcc compiler on gclist_at_iecc.com.  You might find it 
interesting.