Subj : Re: A question about atomic_ptr
To   : comp.programming.threads
From : Chris Thomasson
Date : Mon Apr 18 2005 02:11 am

> Hm. How about the following DWCASless algorithm:

OK, this is similar to logic that I used a while back:

http://groups.google.ca/groups?hl=en&lr=&selm=R6fmb.18865%24Fm2.9642%40attbi_s04

You could run into a similar race-condition described. Your algorithm could 
cause a memory leak:

Think if the following scenario occurred:




static obj_t *sptr = new objA;


Threads A and B - ( both writers )
----------

obj_t *lptr;

1: inc gate
2: point lptr to new obj_t
3: xchg sptr with lptr
4: dec gate, if > 0 goto 6
5: dec lptr
6: done




And they happen to execute like this:


A1: gate = 1         |
                     |  B1: gate = 2
A2: lptr = new objB  |
                     |  B2: lptr = objC
                     |  B3: sptr = objC && lptr = objA
                     |  B4: gate = 1
                     |  B6: done

A3: sptr = objB && lptr = objC
A4: gate = 0
A5: dec objC
A6: done




Notice how Thread B leaked objA?

.