"Peter Dimov" <pdimov@mmltd.net> wrote in message news:03d401c54a75$d3a71a20$6401a8c0@pdimov2...
Gennadiy Rozental wrote:
By PBSP, do you mean policy-based smart pointers in general, or a Loki-derived smart pointer in particular?
You may say in general or Loki-derived - depend on point of view. I have many objections to Loki solution (and some of it's derivatives). But what I have in mind is similar. Do you have any other alternative model?
If you ask me for my personal opinion: I have little use for the Conversion and the Checking policies. However I do need an easier way of specifying a
Checking policy could be very convenient for implementing Debug only extra checking. Similar to STLPort debug mode. Conversion is just a matter of satisfying legacy interfaces.
deleter (ala Howard's unique_ptr) and a ConstPropagation policy (with the default being No for shallow copy and Yes for deep copy).
In addition, I'm not happy with the integration of shared/weak/intrusive_ptr in the current model. To be really flexible, a policy-based incarnation of a reference counted pointer needs to leave its aspects configurable, covering the whole spectrum from a simple counted_ptr to full-featured shared_ptr (with or without weak_ptr support in both cases) and possibly covering the unsafe/basic/strong thread safety axis as well. This might imply a policy controlling whether conversions from smart_ptr<T> to smart_ptr<U> are allowed (because disallowing these allows alternative implementations or omitting one pointer from the control block.)
My needs may not be typical, of course.
I hope David is listening to this and can comment. I am a bit out of date with details of PBSP design/implementation to comment on that. Gennadiy