From: Larry Evans (jcampbell3_at_[hidden])
Date: 2002-01-18 17:18:09
David Abrahams wrote:
> The current shared_ptr "handles" MI already, so would you please spell out
> for us what your submission is supposed to do which shared_ptr does not? For
> myself, I don't have time to figure out what your code does; I have too much
> other code to figure out!
The zip file:
contains 2 test drivers and their output (.run files) illustrating the problem. The problem is the delete in
checked_delete used a pointer that's pointing inside of instead of to the start of the object.
I believe the boost postings with subject="[boost] Re: shared_ptr: unsafe, or am I crazy?"
have discussed the problem. I think if the destructors were virtual, the problem would disappear.
I think that's been mentioned in the "am I crazy" posts.
I also thought it would be important if ever intrusive refcounts were used because then, every
change of the refcount would require the smart pointer access the start of the object. Since
intrusive counts were talked about elsewhere (I think in the "Loki SmartPtr questions" posts),
I thought those people might be interested.
Boost list run by bdawes at acm.org, gregod at cs.rpi.edu, cpdaniel at pacbell.net, john at johnmaddock.co.uk