Subj : Re: Memory visibility and the C compiler... ( a bit long )
To   : comp.programming.threads
From : SenderX
Date : Mon Jan 17 2005 10:35 am

>> paranoid. My library is intended to be used in production server code. I
>> figured that if all of my lock-free stuff was totally contained in an
>> externally assembled library, most of the issues wrt compiler reordering
>> would go away.
>
> Nothing would go away. Think of JIT inlining and reordering. You're
> still making assumption that may not hold.

Well, "at least" developers that decide to use my library as a framework for 
their lock-free algorithms will be able to read my paper ( which will 
contain links to this thread, and others ) to inform themselves of the 
possible trap they might be getting themselves into. I am thinking of 
production code and stability here, so all caveats must be thoroughly listed 
in my libraries documentation. Thanks for your helpful input Alex.




p.s.

I believe I am fighting an endless fight that the C compilers will always 
win. So the best I can do is heavily document the problems. That should be 
adequate?

:O

.