Re: [boost] Re: Future of share_ptr

From: Dave Harris (brangdon_at_[hidden])
Date: 2002-06-19 17:28:22

In-Reply-To: <aeoir4$lu4$1_at_[hidden]>
On Tue, 18 Jun 2002 19:24:10 -0500 David B. Held 
(dheld_at_[hidden]) wrote:
> [intrusive versus non-intrusive shared_ptr]
> What were the problems?

The main issue was the lack of a safe way to get a counted pointer from a 
raw pointer. It's not a problem if you can avoid raw pointers everywhere, 
but that isn't always feasible. In particular, "this" is a raw pointer and 
there was no good way for an object to get a counted pointer to itself. 
Eg you couldn't say:

    void Item::append_to( Container &c ) {
        c.push_back( shared_ptr( this ) );
    }

because this would create a new reference count each time it was called. 

A second issue is efficiency. Intrusive pointers can be more efficient to 
create as they don't need extra memory allocations for the count.

The advantage of non-intrusive pointers is that they can point to 
anything. The current design is a hybrid intended to give the best of both 
worlds.

There are also some issues over whether smart pointers should be 
policy-based, and if so how those policies should be specified for 
concrete uses. As far as I know, the boost community itself is not yet 
decided about this, let alone the standards committee.

-- Dave Harris