[concurrency-interest] Fences.keepAlive

Matthias Ernst matthias at mernst.org
Wed Jan 14 12:32:14 EST 2009


On Wed, Jan 14, 2009 at 5:05 PM, Gregg Wonderly <gergg at cox.net> wrote:
> Doug Lea wrote:
>>
>> [Same cross-post conventions as the last one.]
>>
>> As another make-good for a long-standing problem, the
>> Fences class should include what amounts to a fence with
>> respect to the garbage collector. This addresses a problem that
>> Hans Boehm has written and talked about for years.
>> (Google finds his JavaOne 2005 slides at
>> http://gceclub.sun.com.cn/java_one_online/2005/TS-3281/ts-3281.pdf)
>>
>> This turns out not to be expressible using other kinds of
>> fences. It tells the compiler/runtime that the ref in
>> question must be considered as being used even if nothing in
>> the program itself indicates it. (The suggested JVM implementation
>> of this is an "opaque fence" that bypasses/evades any
>> mechanics or optimizations that might cause ref to vanish.)
>
> After reading Hans paper from J1, I'm not sure I understand why this fence
> is needed.  Phantom reference solves the issues that I see with trying to
> "clean up" due to GC finalization happening.  The ReferenceTracker class
> that I posted here works to defeat all the issues that Hans brought up I
> believe.
>
> Would it make sense to include something like ReferenceTracker in lieu of
> this kind of fence?  Perhaps I'm not understanding how the problem Hans
> discusses turns out to be an issue for developers or JVM implementations?
>  Is it a problem if you use PhantomReference instead of finalize()?

I think the issue would be the same. The "Holder" objects you propose
may go out of scope while a call to the referent is in-flight and so
the referent may be released concurrently. The fence makes sure that
you can reliably extend the holder's scope beyond the call into the
referent.

>
> Gregg Wonderly
> _______________________________________________
> Concurrency-interest mailing list
> Concurrency-interest at cs.oswego.edu
> http://cs.oswego.edu/mailman/listinfo/concurrency-interest
>


More information about the Concurrency-interest mailing list