[cpp-threads] Causality on more than two processors, write atomicity

[Bill Pugh]

On Sep 5, 2005, at 4:42 PM, Herb Sutter wrote:
>
> Does that mean I can't always reliably maintain invariants, if I'm  
> using
> unsynchronized updates (no locks) and the variables involved in the
> invariant can be updated by more than one thread over the life of the
> program?

Correct.

If your program is not correctly synchronized, you get very weird  
behaviors, and you may
have to understand the details of JSR-133 in order to understand the  
possible behaviors.
The number of people who are capabile of designing correct algorithms  
with data races
(conflicting non-volatile accesses not ordered by happens-before) is  
tiny (less than a
dozen by my estimate). Doug and I spent hours going over a place in  
ConcurrentHashMap
where Doug was deliberating undersynchronizing.


You have to use some form of synchronization: whether it be built in  
locks, volatile
variables, or things introduced by jsr-166. No one who writes  
undersynchronized
code should be expecting any hand holding in helping them make their  
software reliable.

     Bill