Subj : Re: double-checked locking in C
To   : comp.programming.threads
From : Peter Dimov
Date : Sun Jul 10 2005 01:03 am

David Schwartz wrote:
> "Peter Dimov" <pdimov@gmail.com> wrote in message
> news:daoao9$13t$1@domitilla.aioe.org...
>
>> then we can talk about a sequence of load and store instructions. If
>> we don't, we can't. For the standard to have a meaning in a
>> particular context,
>> the implementation must specify what constitutes observable behavior
>> in this context. "Observable behavior" is the link between the formal
>> specification and the real world.
>
>    In other words, the standard isn't really saying anything (unless
> the implementation provides separate guarantees), which is exactly
> the position that I take.

Let's just say that I agree that "the semantics of volatile are 
implementation defined" is a workable approximation of what the standard 
actually says and that the differences are rarely, if ever, relevant.

.