Re: [boost] [FSM] Review library version update

From: Steven Watanabe (steven_at_[hidden])
Date: 2007-02-03 18:09:37

AMDG

Andrey Semashev <andysem <at> mail.ru> writes:

> But it is volatile as it's being passed to the init_states_info
> function which does the initialization. It's argument type is
> volatile.

My mistake

> 
> > line 912
> >     template< typename EventT >
> >     static state_dispatcher< EventT > const& get_state_dispatcher()
> >     {
> >       static const state_dispatcher< EventT > dispatcher;
> >       return dispatcher;
> >     }
> > function static is not thread safe.
> 
> It doesn't need to be. The state_machine class is intended to be used
> in single-thread context. Once you have to access the machine from
> multiple threads you should use locking_state_machine which
> synchronizes on "process" function calls which is the only way
> "get_state_dispatcher" gets called.

Unfortunately, it isn't that simple.  The dispatcher
is shared by all state machines of the same type.
Therefore even apparently unrelated machines can
attempt to initialize it concurrently.

> 
> > locking_state_machine.hpp
> > dynamic_locker is unnecessary
> > you can use
> >     scoped_lock this_lock(false);
> > and
> >     this_lock.lock();
> > to do the same thing.
> 
> Actually, I can't - the lightweight mutex doesn't have this. The only
> thing the Boost.FSM requires from the locker class is the
> construction/destruction locking.

I see.

In Christ,
Steven Watanabe