Qt state machines and automatic timed transitions

In the interest of full disclosure: this post is related to what I do for a living, development of and for embedded systems. I work for Semcon, but they don’t make me wear a suit and tie so these are my words, and mine alone.

A bit of background info

In a recent project we had a system where the turning of the wheels were controlled by a simple dial. It emitted pulses as it was turned and the pulse train was shifted slightly depending on the direction of the turn. In software this was mapped onto two signals, one for each direction, with one signal emitted for each pulse in the train. All very straight forward so far.

To avoid accidental change of direction we decided that

  1. only start turning the wheels after having received four initial signals, and
  2. if a full second without receiving any signal meant that the turning had stopped.

The solution

The application was to be implemented using Qt, so using the Qt state machine framework was an obvious choice. The full state machine wouldn’t have to be large, only 8 states. The initial state (sResting) would indicate that the system was in a steady state (no turning), from there any received signal would advance into a successive state (sOne, sTwo, sThree, sFour) to indicate the number of received signals. From the fourth state the machine would advance directly to a state (sTurning) where a received signal would initiate an actual turn of the wheels. The turning would happen upon the entry into two separate states (sTurnRight and sTurnLeft), each of these states would instantly return to sTurning. All of this is simple and straight forward, what wasn’t so clear was to implement the automatic return to the initial state after 1s of inactivity.

The implementation

As I like to do, I first experimented a little to find a suitable solution to the problem. What follows is the resulting code of that experiment. The final code used in the project ended up being very similar. It’s all based around the method postDelayedEvent() found in QStateMachine.

First off a new type of event is nedded, a ReturnEvent:

There is also a need for a new type of transition, ReturnTransition:

For the experiment I decided to use a simple widget containing two buttons, it would also hold the state machine:

The widget also holds the slots for all the state entry functions:

Sure, several of the entry functions could be folded into one, but in order to validate the idea it’s easier to make separate ones for each state. The pattern is easy to spot, on entry a delayed return event is registered (if there’s a previous one its replaced with a new), except for the steady state (sResting) where any delayed event is removed, and the turning states (sTurnRight and sTurnLeft) since those states immediately return to sTurning anyway.

Finally it also holds the handle for the delayed event:

Now the main function for setting it all up is simple:

int main(int argc, char **argv)
    QApplication app(argc, argv);
    QStateMachine m;
    MButtons b(m);

    QState *sResting = new QState(),
           *sOne = new QState(),
           *sTwo = new QState(),
           *sThree = new QState(),
           *sFour = new QState(),
           *sTurning = new QState(),
           *sTurnRight = new QState(),
           *sTurnLeft = new QState();


    sResting->addTransition(&b._right, SIGNAL(clicked()), sOne);
    sResting->addTransition(&b._left, SIGNAL(clicked()), sOne);
    sOne->addTransition(&b._right, SIGNAL(clicked()), sTwo);
    sOne->addTransition(&b._left, SIGNAL(clicked()), sTwo);
    sOne->addTransition(new ReturnTransition(sResting));
    sTwo->addTransition(&b._right, SIGNAL(clicked()), sThree);
    sTwo->addTransition(&b._left, SIGNAL(clicked()), sThree);
    sTwo->addTransition(new ReturnTransition(sResting));
    sThree->addTransition(&b._right, SIGNAL(clicked()), sFour);
    sThree->addTransition(&b._left, SIGNAL(clicked()), sFour);
    sThree->addTransition(new ReturnTransition(sResting));
    sTurning->addTransition(&b._right, SIGNAL(clicked()), sTurnRight);
    sTurning->addTransition(&b._left, SIGNAL(clicked()), sTurnLeft);
    sTurning->addTransition(new ReturnTransition(sResting));

    QObject::connect(sResting, SIGNAL(entered()), &b, SLOT(sRestingEntered()));
    QObject::connect(sOne, SIGNAL(entered()), &b, SLOT(sOneEntered()));
    QObject::connect(sTwo, SIGNAL(entered()), &b, SLOT(sTwoEntered()));
    QObject::connect(sThree, SIGNAL(entered()), &b, SLOT(sThreeEntered()));
    QObject::connect(sFour, SIGNAL(entered()), &b, SLOT(sFourEntered()));
    QObject::connect(sTurning, SIGNAL(entered()), &b, SLOT(sTurningEntered()));
    QObject::connect(sTurnRight, SIGNAL(entered()), &b, SLOT(sTurnRightEntered()));
    QObject::connect(sTurnLeft, SIGNAL(entered()), &b, SLOT(sTurnLeftEntered()));



Conclusion and open questions

I’m fairly happy with the solution, but I’d be curious how other people, people more skilled in using Qt, would have solved the problem.

For a while I considered solving the skipping of four initial signals using a single state and counter, but I saw no obvious easy way to implement that, so I instead opted to use separate states. Slightly wasteful of resources, but not too bad, and simplicity is important. I’m very curious to find out if there’s a simply way to implement it using a single state.

⟸ Manual setup of Qt+Eclipse on Windows Compiling boost for Windows, with MinGW on Linux ⟹
Leave a comment