The main interface to PythonQt is the PythonQt singleton. PythonQt needs to be initialized via PythonQt::init() once. Afterwards you communicate with the singleton via PythonQt::self(). PythonQt offers a complete Qt binding, which needs to be enabled via PythonQt_QtAll::init().

Datatype Mapping

The following table shows the mapping between Python and Qt objects:

QStringunicode string
QByteArrayQByteArray wrapper
QStringListtuple of unicode strings
QVariantListtuple of objects
QVariantMapdict of objects
QVariantdepends on type, see below
QSize, QRect and all other standard Qt QVariantsvariant wrapper that supports complete API of the respective Qt classes
OwnRegisteredMetaTypeC++ wrapper, optionally with additional information/wrapping provided by registerCPPClass()
QList<AnyObject*>converts to a list of CPP wrappers
QVector<AnyObject*>converts to a list of CPP wrappers
EnumTypeEnum wrapper derived from python integer
QObject (and derived classes)QObject wrapper
C++ objectCPP wrapper, either wrapped via PythonQtCppWrapperFactory or just decorated with decorators

PyObject is passed as direct pointer, which allows to pass/return any Python object directly to/from a Qt slot that uses PyObject* as its argument/return value. QVariants are mapped recursively as given above, e.g. a dictionary can contain lists of dictionaries of doubles. All Qt QVariant types are implemented, PythonQt supports the complete Qt API for these object.

QObject Wrapping

All classes derived from QObject are automatically wrapped with a python wrapper class when they become visible to the Python interpreter. This can happen via

It is important that you call PythonQt::registerClass() for any QObject derived class that may become visible to Python, except when you add it via PythonQt::addObject(). This will register the complete parent hierachy of the registered class, so that when you register e.g. a QPushButton, QWidget will be registered as well (and all intermediate parents).

From Python, you can talk to the returned QObjects in a natural way by calling their slots and receiving the return values. You can also read/write all properties of the objects as if they where normal python properties.

In addition to this, the wrapped objects support

The below example shows how to connect signals in Python:

# define a signal handler function
def someFunction(flag):
print flag
# button1 is a QPushButton that has been added to Python via addObject()
# connect the clicked signal to a python function:
button1.connect("clicked(bool)", someFunction)

CPP Wrapping

You can create dedicated wrapper QObjects for any C++ class. This is done by deriving from PythonQtCppWrapperFactory and adding your factory via addWrapperFactory(). Whenever PythonQt encounters a CPP pointer (e.g. on a slot or signal) and it does not known it as a QObject derived class, it will create a generic CPP wrapper. So even unknown C++ objects can be passed through Python. If the wrapper factory supports the CPP class, a QObject wrapper will be created for each instance that enters Python. An alternative to a complete wrapper via the wrapper factory are decorators, see Decorator slots

Meta Object/Class access

For each known C++ class, PythonQt provides a Python class. These classes are visible inside of the "PythonQt" python module or in subpackages if a package is given when the class is registered.

A Meta class supports:

From within Python, you can import the module "PythonQt" to access these classes and the Qt namespace.

from PythonQt import QtCore
# namespace access:
print QtCore.Qt.AlignLeft
# constructors
a = QtCore.QSize(12,13)
b = QtCore.QFont()
# static method
# enum value

Decorator slots

PythonQt introduces a new generic approach to extend any wrapped QObject or CPP object with

The idea behind decorators is that we wanted to make it as easy as possible to extend wrapped objects. Since we already have an implementation for invoking any Qt Slot from Python, it looked promising to use this approach for the extension of wrapped objects as well. This avoids that the PythonQt user needs to care about how Python arguments are mapped from/to Qt when he wants to create static methods, constructors and additional member functions.

The basic idea about decorators is to create a QObject derived class that implements slots which take one of the above roles (e.g. constructor, destructor etc.) via a naming convention. These slots are then assigned to other classes via the naming convention.

The below example shows all kinds of decorators in action:

// an example CPP object
class YourCPPObject {
YourCPPObject(int arg1, float arg2) { a = arg1; b = arg2; }
float doSomething(int arg1) { return arg1*a*b; };
int a;
float b;
// an example decorator
class ExampleDecorator : public QObject
public slots:
// add a constructor to QSize that takes a QPoint
QSize* new_QSize(const QPoint& p) { return new QSize(p.x(), p.y()); }
// add a constructor for QPushButton that takes a text and a parent widget
QPushButton* new_QPushButton(const QString& text, QWidget* parent=NULL) { return new QPushButton(text, parent); }
// add a constructor for a CPP object
YourCPPObject* new_YourCPPObject(int arg1, float arg2) { return new YourCPPObject(arg1, arg2); }
// add a destructor for a CPP object
void delete_YourCPPObject(YourCPPObject* obj) { delete obj; }
// add a static method to QWidget
QWidget* static_QWidget_mouseGrabber() { return QWidget::mouseGrabber(); }
// add an additional slot to QWidget (make move() callable, which is not declared as a slot in QWidget)
void move(QWidget* w, const QPoint& p) { w->move(p); }
// add an additional slot to QWidget, overloading the above move method
void move(QWidget* w, int x, int y) { w->move(x,y); }
// add a method to your own CPP object
int doSomething(YourCPPObject* obj, int arg1) { return obj->doSomething(arg1); }
PythonQt::self()->addDecorators(new ExampleDecorator());

After you have registered an instance of the above ExampleDecorator, you can do the following from Python (all these calls are mapped to the above decorator slots):

from PythonQt import QtCore, QtGui, YourCPPObject
# call our new constructor of QSize
size = QtCore.QSize(QPoint(1,2));
# call our new QPushButton constructor
button = QtGui.QPushButton("sometext");
# call the move slot (overload1)
# call the move slot (overload2)
# call the static method
grabber = QtGui.QWidget.mouseWrapper();
# create a CPP object via constructor
yourCpp = YourCPPObject(1,11.5)
# call the wrapped method on CPP object
print yourCpp.doSomething(1);
# destructor will be called:
yourCpp = None

Ownership management

In PythonQt, each wrapped C++ object is either owned by Python or C++. When an object is created via a Python constructor, it is owned by Python by default. When an object is returned from a C++ API (e.g. a slot), it is owned by C++ by default. Since the Qt API contains various APIs that pass the ownership from/to other C++ objects, PythonQt needs to keep track of such API calls. This is archieved by annotating arguments and return values in wrapper slots with magic templates:

These annotation templates work for since C++ pointer types. In addition to that, they work for QList<AnyObject*>, to pass the ownership for each object in the list.


public slots:
PythonQtPassOwnershipToPython<QGraphicsItem*> createNewItemOwnedByPython();
void addItemToCPP(PythonQtPassOwnershipToPython<QList<QGraphicsItem*> > items);
void addItemParent(QGraphicsItem* wrappedObject, PythonQtNewOwnerOfThis<QGraphicsItem*> parent);