Re: [Boost-users] Query state on an asynchronous_state_machine

From: Andreas Huber (ahd6974-spamboostorgtrap_at_[hidden])
Date: 2008-03-07 15:01:46

Hi Jarl

> Can anyone tell me how to query a state on an
> asynchronous_state_machine?
>
> The obvious answer would be to say use the state_downcast method
> (which is inherited from state_machine. But in practise it seems
> harder than so.
>
> But opposed to state_machine, when using asynchronous_state_machine
> the instance is not created explicitely, it is created implicitely by
> the scheduler, like this: myMachine =
> myEventHandler.create_processor<AsyncMachine>(); , hence there is no
> handle to the state machine itself. All I have is a handle to the
> processor (sc::fifo_scheduler<>::processor_handle ).

This is intentional. If fifo_scheduler supported a state_downcast method 
then inevitably race-conditions would arise (the result of a state_downcast 
is a state object, which could have been destructed long before the caller 
gets hold of it).

> So is there a way to come from sc::fifo_scheduler<>::processor_handle
> to AsyncMachine (which inherits sc:asynchronous_state_machine)?

No, for the same reasons. A sc:asynchronous_state_machine<> subtype is 
serviced by the fifo_scheduler thread. Directly accessing it from a 
different thread would again open the door to race-conditions or dead-locks. 
The proper way to query an asynchronous state machine is by sending it a 
special event carrying a boost::function object. The state_machine typically 
defines an in_state_reaction for said event, which simply calls 
boost::function::operator(). This arrangement lets you execute arbitrary 
external code in an FSM. Of course, if said external code needs to send 
results to another thread, then you'd need to employ a condition variable or 
some such and make sure that you make copies of all heap-allocated data that 
you hand over.

This whole procedure could be implemented nicely with futures, but Boost 
does not (yet) sport such a library :-/.

HTH,

-- 
Andreas Huber
When replying by private email, please remove the words spam and trap
from the address shown in the header.