From: David B. Held (dheld_at_[hidden])
Date: 2002-07-23 11:50:32
"Philippe A. Bouchard" <philippeb_at_[hidden]> wrote in message
news:ahk0in$gk6$1_at_main.gmane.org...
> It seems you cannot use it with typenames that easily; it depends on
> virtual tables and you'll have to prepare your pointee type similarly to
> detail::counted_base derivatives. I'm still not convinced this
> complexity / benefits ratio is better.
I'm not sure what your first objection means, but it seems that you
don't quite understand how the policy-based smart pointers work.
First of all, I'm not aware of any smart pointer type that calls through
a virtual table. That would seem like a rather extravagant way to go
that most people would be unwilling to pay for. Secondly, while you
can use a counted_base type intrusive pointer with most of the
policy-based implementations, the whole point of policy-based
design is *choice*. You can just as easily choose an external
count or a wrapped count. What Gennadiy was suggesting is:
See if you can't write your squad_ptr as a policy set for one of the
policy-based pointers. Right now, that's a little tricky, since they
aren't all well-documented and thoroughly tested. But the advantage
of writing it as a policy set is that it makes it easier to define
interoperability conversions between similar but different pointers.
Also, it reduces the proliferation of smart pointer headers (since a
good deal of your code is more or less identical to the various
smart pointers in Boost currently).
Dave