Subj : Re: Lock Free Programming with Boost::shared_ptr instead of Hazard Pointers To : comp.programming.threads From : Joe Seigh Date : Sun Mar 20 2005 07:40 pm On 20 Mar 2005 18:50:13 -0500, Peter Dimov wrote: > "Andrei Alexandrescu (See Website For Email)" wrote in message news:<423D309D.4080700@moderncppdesign.com>... >> >> Peter Dimov's post as of two days ago suggests that this problem has >> been fixed on Windows. Other known SP designs that cater for >> multithreading used atomic increment and decrement since time >> immemorial, so I'm not sure why this delay. > > shared_ptr is unique in that in the general case it needs to maintain > two reference counts. It took some time for Alexander Terekhov to > discover a lock-free algorithm. The algorithm is overhead-free when > shared_ptr is used as an ordinary strong smart pointer (i.e. it only > touches one count in this case). The delay in incorporating his > algorithm in the Boost code base is entirely my fault, of course. What "overhead-free" algorithm is that? -- Joe Seigh .