Subj : Proxy GC
To   : comp.programming.threads
From : Joe Seigh
Date : Sun Sep 25 2005 12:30 pm

I think working on that RCU+SMR threw me off a little bit.
Even though you can only do acyclic traversals, not cyclic,
traversals with RCU+SMR, you can do cyclic traversals with
proxy based GC.  Provided your data structure logic supports
lock-free cyclic traversal.

Also, if you're using a mutex on the write side, you don't
need to hold it while calling the proxy defer routine as
long as the proxy object swap logic maintains FIFO order
(proxy objects are freed in same order as the swap order).
This means I can put the Kung Fu monkey through the trees
swap logic back into atomic pointer proxy collector defer
routine (appc::defer) if and when I ever release it again.
Could even do it for other proxy GC like RCU+SMR as well
though the latter is pretty much moot.

-- 
Joe Seigh

When you get lemons, you make lemonade.
When you get hardware, you make software.

.