Subj : Re: double-checked locking in C
To   : comp.programming.threads
From : Laurent Deniau
Date : Fri Jul 08 2005 03:09 pm

David Schwartz wrote:
> "Laurent Deniau" <Laurent.Deniau@cern.ch> wrote in message 
> news:dalle2$nlu$1@sunnews.cern.ch...
> 
> 
>>The as-if rule is an implementation-defined problem. The as-if-safe rule 
>>becomes an as-if-unsafe rule because of some changes in the context of 
>>execution which is not considered by the implementation (e.g. switching 
>>off the computer). Unless specified, this is leading to an undefined 
>>behavior. Imagine that the implementation of volatile includes memory 
>>barrier or some cache synchronisation mechanism (nothing in the C says to 
>>not do so), it would still be conform with and without thread. But at the 
>>price of hidden penalties that C programmers do not want to hear about.
> 
> 
>     Since no standards compliant C code can tell the difference, the C 
> standard cannot require the barriers. Some other standard (for example, a 
> threading standard) might specify that volatile provides ordering guarantees 
> across threads, in which case failure to provide barriers might lead to 
> violating that standard on that platform.

This is exactly what I claim. Standards rely on architecture, they do 
not force architecture (including CPUs).

ld.

.