[boost] Re: Yet Another Pointer Policy Proposal... Fast One

From: Philippe A. Bouchard (philippeb_at_[hidden])
Date: 2002-07-28 19:20:44

"Gennadiy Rozental" <rogeeff_at_[hidden]> wrote in message
news:ai1ujf$ngu$1_at_main.gmane.org...
> > We do you think about those different possibilities:
> > 1) new type();
> > 2) new (xmm) type();
> > 3) make_rc(type());
> > 4) make_rc(new (xmm) type());
> > 5) new (new (xmm) type()) rc<type>;
>
> Are you kidding me??!! Can't you explain without SINGLE C++ statement,
what
> is your idea and why it could not be implemented the conventional way and
> without overridding any operators new?

Don't you have an idea of what is going on in the backstage here?!?  Again,
my main goal is to add 1 integer length when the object is allocated; one
malloc call not two (for counters); copying is faster when
sizeof(shared_ptr<type>) == sizeof(type *); no need for policy based
functions in your class to handle reference counts; works with typenames
like: int, float, char *, char (*)(), char (* volatile const)(int, float,
double (*)(int, long, double), char (*) [10][20][30], ...); the integer
(counter) can be replaced by a virtual table, size_t, etc.; thus is more
extensible for memory management or dynamic information.  It is impossible
to do so with the current standards or policies easily in an efficient way
because I _have_ to modify the operator new when objects are allocated, not
after.  It will be too late for performance gains.

Thank you.

Philippe A. Bouchard