Subj : Re: CUJ Article: Atomic Reference Counting Pointers
To   : comp.programming.threads
From : Joseph Seigh
Date : Tue Jan 04 2005 04:49 pm

On Tue, 04 Jan 2005 22:29:25 +0100, Alexander Terekhov <terekhov@web.de>  
wrote:

>
> Joseph Seigh wrote:
>>
>> On Tue, 04 Jan 2005 19:47:00 +0100, Alexander Terekhov <terekhov@web.de>
>> wrote:
>>
>> >
>> > Joseph Seigh wrote:
>> > [...]
>> >> If you had real GC (I assume you're referring to Boehm style GC) and  
>> its
>> >> performance wasn't a problem, why would you use reference counting?
>> >
>> > You would use atomic<T *> without any reference counting. With real
>> > GC, strong thread-safety for shared_ptr<T>-like beasts comes almost
>> > "for free" (apart from extra msync which is not needed in the case
>> > of basic thread-safety).
>>
>> You're not making sense.  Why would you use one form of GC on top of
>> another?
>
> Replace "atomic<T *>" above by atomic<gc::uintptr_t> and read it
> again.
>
> Got it?
>

No.  Java has GC.  What would you use reference counting for in Java?

Joe Seigh

.