therning.org/ magnus

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:

1
2
3
4
5

class ReturnEvent : public QEvent
{
public:
    ReturnEvent() : QEvent(QEvent::Type(QEvent::User + 1)) {}
};

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

1
2
3
4
5
6
7
8
9
10
11
12

class ReturnTransition : public QAbstractTransition
{
public:
    ReturnTransition(QState *target=0) { if(target) setTargetState(target); }
 
protected:
    virtual bool eventTest(QEvent *e) {
        return(e->type() == QEvent::Type(QEvent::User + 1));
    }
 
    virtual void onTransition(QEvent *) {}
};

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

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19

class MButtons : public QWidget
{
    Q_OBJECT;
 
public:
    MButtons(QStateMachine &m)
        : _right("Right"), _left("Left"),
        _m(m), _delayed(0) {
        QBoxLayout *lo = new QBoxLayout(QBoxLayout::TopToBottom);
        lo->addWidget(&_right);
        lo->addWidget(&_left);
 
        setLayout(lo);
    }
    virtual ~MButtons() {}
 
    QPushButton _right,
                _left;
    QStateMachine &_m;

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

20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57

public slots:
    void sRestingEntered() {
        qDebug() << __PRETTY_FUNCTION__;
        if(_delayed) { _m.cancelDelayedEvent(_delayed); _delayed = 0; }
    }
 
    void sOneEntered() {
        qDebug() << __PRETTY_FUNCTION__;
        if(_delayed) { _m.cancelDelayedEvent(_delayed); _delayed = 0; }
        _delayed = _m.postDelayedEvent(new ReturnEvent, 1000);
    }
 
    void sTwoEntered() {
        qDebug() << __PRETTY_FUNCTION__;
        if(_delayed) { _m.cancelDelayedEvent(_delayed); _delayed = 0; }
        _delayed = _m.postDelayedEvent(new ReturnEvent, 1000);
    }
    void sThreeEntered() {
        qDebug() << __PRETTY_FUNCTION__;
        if(_delayed) { _m.cancelDelayedEvent(_delayed); _delayed = 0; }
        _delayed = _m.postDelayedEvent(new ReturnEvent, 1000);
    }
    void sFourEntered() {
        qDebug() << __PRETTY_FUNCTION__;
        if(_delayed) { _m.cancelDelayedEvent(_delayed); _delayed = 0; }
        _delayed = _m.postDelayedEvent(new ReturnEvent, 1000);
    }
    void sTurningEntered() {
        qDebug() << __PRETTY_FUNCTION__;
        if(_delayed) { _m.cancelDelayedEvent(_delayed); _delayed = 0; }
        _delayed = _m.postDelayedEvent(new ReturnEvent, 1000);
    }
    void sTurnRightEntered() {
        qDebug() << __PRETTY_FUNCTION__;
    }
    void sTurnLeftEntered() {
        qDebug() << __PRETTY_FUNCTION__;
    }

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:

58
59
60

private:
    int _delayed;
};

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

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57

int main(int argc, char **argv)
{
    QApplication app(argc, argv);
    QStateMachine m;
    MButtons b(m);
    b.show();
 
    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();
 
    m.addState(sResting);
    m.addState(sOne);
    m.addState(sTwo);
    m.addState(sThree);
    m.addState(sFour);
    m.addState(sTurning);
    m.addState(sTurnRight);
    m.addState(sTurnLeft);
    m.setInitialState(sResting);
 
    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));
    sFour->addTransition(sTurning);
    sTurning->addTransition(&b._right, SIGNAL(clicked()), sTurnRight);
    sTurning->addTransition(&b._left, SIGNAL(clicked()), sTurnLeft);
    sTurning->addTransition(new ReturnTransition(sResting));
    sTurnRight->addTransition(sTurning);
    sTurnLeft->addTransition(sTurning);
 
    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()));
 
    m.start();
 
    return(app.exec());
}

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.

Before the weekend I started looking at using Qt on Windows. More specifically I wanted to know whether this combination could be an option for a sub-project at work. We need to develop a program for the Windows desktop, and due to the overall context it would make sense to write it in C++ (that’s what we use for another part of the project). We already use both Eclipse and Visual Studio in the project, but I strongly prefer Eclipse, so I was hoping to be able to use it. However, it seems that the Qt developers strongly favour their own tool Qt Creator, though there are (outdated?) integrators for both Eclipse and Visual Studio. I’d rather avoid introducing a third IDE into a project—two is already one too many in my opinion. Anyway, I think I managed to find an acceptable configuration of Eclipse without using that old Qt integration plugin together with the MSVC (I was using the gratis version of MSVC for this).

Qt setup

I decided to install Qt into C:\QtSDK, and then I made the following permanent changes to the environment:

> set QTDIR=C:\QtSDK\Desktop\Qt\4.8.0\msvc2010
> set QMAKESPEC=%QTDIR%\mkspecs\win32-msvc2010
> set PATH=%PATH%;%QTDIR%\bin;C:\QtSDK\QtCreator\bin

Starting Eclipse so that it finds the compiler

It’s slightly disappointing that Eclipse happily lets one create MSVC project that isn’t buildable because it doesn’t know where the compiler is located. One easy way to remedy that seems to create a BAT file to create the proper environment for Eclipse:

@echo off
setlocal
call "C:\Program Files (x86)\Microsoft Visual Studio 10.0\VC\vcvarsall.bat"
start C:\Eclipse\Indigo\eclipse.exe
endlocal

Creating the project

Creating a “makefile” project in Eclipse is fairly straight forward; one needs a C/C++ project, of the makefile type, and make it empty too so that there isn’t any cruft in the way. Then add a single source file, e.g. main.cxx:

#include <iostream>
#include <Qt/QtGui>
 
int main(int argc, char **argv)
{
    std::cout << __FUNCTION__ << std::endl;
    QApplication app(argc, argv);
    return(app.exec());
}

And then a project file, e.g. Test.pro:

TEMPLATE = app
TARGET = 
DEPENDPATH += .
INCLUDEPATH += .
 
CONFIG += qt
 
HEADERS +=
SOURCES += main.cxx

After this use qmake to create the required makefile. I decided to use a subdirectory (_build) in the project, which qmake seems to have full support for:

> qmake ..\Test.pro

Setting up building from Eclipse

In the project properties modify the C/C++ Build settings for the Debug target. Instead of the default build command (which is make) one can use nmake, or even better jom:

• Build command: C:/QtSDK/QTCreator/bin/jom -f Makefile.Debug
• Build directory: ${workspace_loc:/Test}/_build

Then one can create a Release target, which differs only in that it builds using Makefile.Release.

Running qmake from inside Eclipse

It’s very convenient to be able to run qmake and re-generate the makefiles from inside Eclipse. One can set that up by adding an external tool:

• Location: C:\QtSDK\Desktop\Qt\4.8.0\msvc2010\bin\qmake.exe
• Working directory: ${workspace_loc:/Test}/_build
• Arguments: ../Test.pro

In closing

I plan to also have a look at the Qt Visual Studio Add-in, though I suspect we might be using the latest version of VS, which might cause trouble.

Suggestions for further integration with Eclipse would be most welcome, e.g. for forms and translations.