Re: [Boost-users] [statechart] Getting the current state from anasyncmachine

From: Johan Nilsson (r.johan.nilsson_at_[hidden])
Date: 2006-08-14 06:45:35

Andreas,

Andreas Huber wrote:
> Hi Johan
>
>> I need a way to get the current state from an asynchronous state
>> machine for
>> testing purposes - something analogous to state_cast.
>>
>> I'll only be using this in my unit tests, so I don't mind if it's a
>> bit
>> cumbersome to use. Thread-safety also isn't an issue. Is there any
>> way of
>> getting a reference to the actual Processor instance?
>
> No, there isn't. This is because any call to a member function of
> asynchronous_state_machine or simple_state from a different thread
> than the worker thread will *very* likely lead to a race condition
> (or a dead lock, if you try to use a mutex besides the one in
> fifo_scheduler).

I'm very well aware of that, and can understand why you didn't provide a way
for this from the beginning as it is generally unsafe. I'd just be using 
this under very specific
circumstances, as I mentioned above.

> Here's one way but it is not pretty:
> 1. Derive a new event, e.g. EvQueryState. As members, add a member
> stateType of type const std::type_info & and a boost::function object,
> accepting a bool as the only parameter.
> 2. Add a global handler for unconsumed events to your FSM
> <http://boost-consulting.com/boost/libs/statechart/doc/reference.html#unconsumed_event>
> 3. In the handler check whether the unconsumed event is EvQueryState.
> If so, call the embedded function object with
> typeid(*(this->state_begin())) == stateType. This of course assumes
> that the machine has exactly one innermost state at all times. If you
> also need to detect if the machine is terminated or has multiple
> innermost states you'd need to pass a collection of const
> std::type_info *.
> 4. Build a utility function that accepts a fifo_scheduler, a
> processor_handle object and a const type_info &. The function queues a
> EvQueryState, waits for the callback and then returns the bool that
> came with the callback.

Thank you! It works perfectly.

The only (minor) annoyance is that I'll have to provide the unconsumed_event 
inside the class for testing purposes only, but that could be conditionally 
enabled using preprocessor directives.

/ Johan