RE: Is enq(Node node) safe?

Sorry, I do not notice that. 
Then how about compareAndSetTail part of this enq method, then my test
scenario does apply, doesn't it?


-----Original Message-----
From: Doug Lea [mailto:dl@xxxxxxxxxxxxx] 
Sent: Wednesday, February 09, 2005 3:36 PM
To: Hu, Jinsong
Cc: concurrency-interest@xxxxxxxxxxxxxxxxxxxx
Subject: Re: [concurrency-interest] Is enq(Node node) safe?

Hu, Jinsong wrote:
> My original posting does not assume that both threads need to successfully
> compareAndSetHead at the same time, they do have orders, one at a time,
the
> point is, the first succeeded thread was forced to sleep after it
> successfully compareAndSetHead, since the first thread already finished
> this, the second thread should not fail on compareAndSetHead.

No. compareAndSetHead(h) expands to a compareAndSet(null, h). Only
the first thread can succeed in CAS'ing head to non-null. The other 
might spin for a while until the successful one wakes up, but since this
is a race in one-time initialization code, the potential for an 
unbounded spin is not worth specially handling. (Preliminary versions
of this class had some spin control here, but in testing even on big
multiprocessors, it was found not to have any effect so was simplified
away.)

-Doug