Logo Questions Linux Laravel Mysql Ubuntu Git Menu
 

Boost MSM only processing internal transitions

I'm using the new Boost 1.44.0 MSM library to produce a state machine. In this state machine there are two classes of events class1 and class2. class1 events can be processed by either states S1 or S2 while class2 events can only be processed by state S2.

A special class1 event upgrade_req requests an upgrade from state S1 to state S2.

I've implemented that in Boost::MSM as follows:

// State S1 and S2 allow any class1 events
struct class1 {};
// Only state S2 allows class2 events
struct class2 {};

// an upgrade request is a class1 event that requests an upgrade to state 2
struct upgrade_req : public class1 {};

struct MyFSM : public msm::front::state_machine_def< MyFSM >
{
    /// State 1. Allows any class1 event
    struct S1 : public msm::front::state<>
    {
        /// functor says "processing event in State 1"
        struct ProcessEvent { /* ... */ };

        struct internal_transition_table : mpl::vector<
            //        Event   Action        Guard
            //       +-------+-------------+------------+
            Internal< class1, ProcessEvent, none >
        > {};
    }; // S1

    /// State 2. Allows any class1 or class2 events
    struct S2 : public msm::front::state<>
    {
        /// functor says "processing event in State 2"
        struct ProcessEvent { /* ... */ };

        struct internal_transition_table : mpl::vector<
            //        Event   Action        Guard
            //       +-------+-------------+------------+
            Internal< class1, ProcessEvent, none >,
            Internal< class2, ProcessEvent, none >
        > {};
    }; // S2

    /// everybody starts in state 1
    typedef S1 initial_state;

    /// send an error if a class2 event was received for state1
    struct SendError { /* ... */ };

    /// Send a response to the upgrade request
    struct SendUpgradeRsp { /* ... */ };

    /// functor returns true if the request to upgrade to state 2 is OK.
    struct VerifyUpgradeReq { /* ... */ };

    struct transition_table : mpl::vector<
        //  Start  Event         Next   Action           Guard
        // +------+-------------+------+----------------+------------------+
        Row< S1,   class1,       none,  none,            none,
        Row< S1,   class2,       S1,    SendError,       none >,
        Row< S1,   upgrade_req,  S2,    SendUpgradRsp,   VerifyUpgradeReq >,

        Row< S2,   class1,       none,  none,            none,
        Row< S2,   class2,       none,  none,            none >
    > {};
}; // MyFSM

My problem is that when I use this as it is, the upgrade_req event is never processed by the main MyFSM::transition_table. It is only processed by the S1::internal_transition_table.

For example:

int main( int argc, char* argv[] )
{
    msm::back::state_machine< MyFSM > sm;
    sm.start();
    sm.process_event( class1() );
    sm.process_event( upgrade_req() ); 
    sm.process_event( class2() );
    return 0;
}

I would desire the output of this to be:

processing event in State 1.
Upgrade Request OK.
processing event in State 2.

But, what I get is this:

processing event in State 1.
processing event in State 1.
Error. Received class 2 event in State 1.

Does anybody have a suggestion on how I can fix this issue?

Thanks, PaulH

like image 274
PaulH Avatar asked Oct 14 '10 17:10

PaulH


1 Answers

Your problem is that the priority of internal transitions is higher than those defined in the transition table. And update_req being a class1, the internal transiton fires. This is actually conform to the UML standard. MSM offers you a second solution, you can define S1's internal transition with a Row with none as target inside transition_table instead of using an internal_transition_table. If you define it BEFORE the transition S1 + upgrade_reg -> S2, it will have a lesser prio and will be tried only if the other one cannot be considered.

If you absolutely need an internal_transition_table then you can only provide a guard to reject class1 if it's not an update_req.

HTH, Christophe Henry

PS: I only found this post by luck. Posting to the boost user list will guarantee you a much faster answer.

like image 108
Christophe Henry Avatar answered Nov 16 '22 23:11

Christophe Henry