Subj : Re: Can C++ local static objects be made thread safe?
To   : comp.programming.threads
From : Bruce Visscher
Date : Fri Feb 11 2005 11:49 am

>  Usually "__thread" makes a variable thread-local.

Thanks for explaining that.

> Whereas 'volatile' is used when a variable might be modified from
> a signal handler.

Yes, but doesn't it also inhibit certain compiler optimizations that
might be causing the code in question to fail as I (think I) have
observed?

>   If you're trying to access a variable in a thread-safe way under
> UNIX, it's usually thread pthread_mutex functions that you want.

Sorry.  I guess I wasn't clear.

The problem is: what I think I saw code similar to Gianni Mariani's
MTSafeStatic macro (but without the __thread keyword!) fail inspite of
the fact that it *has* pthread_mutex locking already(*).  Which, from
studying this thread is starting to make some sense.

What I am asking is whether I can fix this using volatile instead of
__thread.  There will be a pthread mutex involved in either case.  If
not then maybe I need to rely on atomic operations instead.

I actually think the VMS platform would be better off with the more
naive looking:

> Item &cofu() {
>         MTSafeStatic( Item, the_item, );
>         return the_item;
> }

which IIRC, CXX on VMS will make threadsafe like GCC 4.0 does (maybe in
a different manner).  I am seeking a more portable solution.

(*) I have a SingletonLock<Owner,Resouce> class that I use for this
which utlizes pthread_once to avoid a race condition with the
initialization of the mutex.

.