[cpp-threads] D2335 (sequential consistency proof) revision

[Herb Sutter]

Hans wrote:
> > Here's another possible way out of this.  I'm not 100% sure this is
> > correct yet, but it seems like it on first glance:
> >
> > We allow trylock and timedlock to fail spuriously, even if the lock
> > was never held.
[...]
> > If we do this, I believe we no longer need acquire semantics for a
> > failed trylock.

Howard wrote:
> It sounds good to me.  I'm not an expert in the visibility area so I'd
> like to double check my understanding.
>
> This code which takes two locks/mutexes and locks them without
> deadlock, in part by using try_lock, is not affected by Hans'
> proposal, right?

Right.

> template <class L1, class L2>
> void
> lock(L1& l1, L2& l2)
> {
>      while (true)
>      {
>          l1.lock();
>          if (l2.try_lock())
>              break;
>          l1.unlock();
>          l2.lock();
>          if (l1.try_lock())
>              break;
>          l2.unlock();
>      }
> }

The only change Hans is proposing is that a failed try_lock doesn't have ordering guarantees, which doesn't affect this example because the try_lock's have nowhere to be reordered to anyway because they can't move out of the tightly enclosing lock/unlock critical sections.

Herb