Subj : Re: Implementing lock-free data structures
To   : comp.programming.threads
From : David Hopwood
Date : Thu Mar 10 2005 03:13 am

Joe Seigh wrote:
> I was going to do a write up of general instructions on how to
> implement lock-free data structures using RCU, atomic_ptr,
> hazard pointers, Boehm GC, etc... since it doesn't appear that
> anyone else has, except for COW objects and linked lists, and
> it's clear that lock-free won't take off until there's better
> documentation.

Why do you expect it to, or think that it should "take off"? In my view
lock/wait-free techniques are best suited to being used *by language
and operating system implementations* to provide carefully chosen sets
of higher-level concurrency abstractions (message queues, pipes, promises,
M-structures, etc.)

Lock-free stuff is (and always will be) too complicated, error-prone,
and platform-specific to use directly in applications. And applications
shouldn't care whether the abstractions they use are implemented using
lock-free techniques or otherwise.

-- 
David Hopwood <david.nospam.hopwood@blueyonder.co.uk>

.