[boost] Re: RFC: Inheritance alternatives for policy-based smart_ptr

From: Andrei Alexandrescu (andrewalex_at_[hidden])
Date: 2002-05-02 02:46:19

"Phil Nash" <phil.nash.lists_at_[hidden]> wrote in message
news:000201c1f1ab$33f52510$700a0c0a_at_TimeMachine...
> Good point. That's reassuring. But in the context of the smart_ptr
> discussion we have to contend with an existing design that uses
straight MI
> with virtually no complexity.
> Don't worry I haven't dropped the bastion of policy_mixers just yet
;-) ...
> just trying to be realistic too...

By the way, we need to amend Gennadiy's excellent RFC with the fact
that the MI-based solution does not incur size overhead anymore.

> It could be me that is mistaken, but I was under the impression that
we
> wanted to move quickly to get a policy-based smart_ptr into boost as
the LWG
> have specifically requested it. It seems to me that we need
something fairly
> stable in boost before the october meeting, and we have a lot of
things to
> get through yet...
> I would like to be persuaded that we have a chance of getting a
policy_mixer
> proposal for smart_ptr taken seriously. If so it would likely be a
simple,
> specialised, variant for the purposes of smart_ptr (but in that case
I would
> hope that it could be done in such a way that a more general
implementation
> could be substituted at a later date).

I would also like to experiment with chained policies (it's just too
attractive to have policies "decorate" each other, I think that might
lead somewhere cool). Just like Phil, I still think it would be best
to first get the straight MI implementation together. The main
building blocks (policies) should conceptually be the same in both
cases. An implementation would certainly provide insights into what
the issues are and how the design's limitations can be solved.

Andrei