Re: uncounted_ptr

From: William Kempf (sirwillard_at_[hidden])
Date: 2001-01-10 09:47:10

--- In boost_at_[hidden], "Larry Evans" <jcampbell3_at_p...> wrote:
> The testdriver code and output from solutions to both the circular
> reference-counted pointer problem and to dangling references can be
> found from the home following page:
> 
>   http://sites.netscape.net/cppljevansusa/homepage
> 
> I've noticed there's some work being done in boost in this
> area (http://www.egroups.com/files/boost/circ_ptr.zip).  I
> think I could contribute here, and I'm finishing up html file
> describing the solution.  Email me if your interested in
> getting a copy when I'm done, or maybe I should just submit it
> for review by boost?

Again, from long experience with this problem in C++ I've grown the 
habit of just avoiding this problem by not using smart pointers in 
the cases where I've had circular references.  I realize that this 
isn't always possible and that others will have different experiences 
here.

That said, I would be very interested in a smart pointer that 
performed similarly to a full blown GC.  In fact, I'd prefer this 
over a real GC.  It would give me complete control when I needed it, 
including the use of RAII harnesses, while making it much simpler to 
program some constructs in which memory management is the hardest 
issue to deal with.  So my personal vote would be for such a 
submission.  I don't know if the circ_ptr in the vault qualifies, if 
what you have in mind should be considered instead, or if both have 
pros and cons and should both be considered/accepted.

Bill Kempf