The QtWebKit Bridge

Overview

The technology

The QtWebKit bridge is a mechanism that extends WebKit's JavaScript environment to access native objects represented as QObjects. It takes advantage of the QObject introspection, a part of the Object Model, which makes it easy to integrate with the dynamic JavaScript environment. For example QObject properties map directly to JavaScript properties.

Use Cases

There are two main use cases for the QtWebKit bridge: web content in native applications and thin clients.

Web Content in Native Applications

This is a common use case in classic Qt application, and a design pattern used by several modern applications like an application that contains a media-player, playlist manager, and music store. The playlist manager is usually best authored as a classic desktop application, with the native-looking robust QWidgets as the application's backbone. The media-player control usually has a custom look and feel and is best written using the Graphics View framework or QtDeclarative. The music store, which shows dynamic content from the Internet and gets modified rapidly, is best authored in HTML and maintained on the server.

With the QtWebKit bridge, the music store component can interact with native parts of the application, for example, when a file needs to be saved to a specific location.

Thin Clients

The use case uses Qt as a native backend of a full web application, a so-called thin client. In this use case, the entire UI is driven by HTML, JavaScript and CSS. Additionally, it uses Qt-based components to access native features usually not exposed to the web, or to enable helper components that are best written in C++.

An example for such a client is a UI for a video-on-demand service on a TV. The entire content and UI can be kept on the server, served dynamically through HTTP and rendered with WebKit. Additional native components are used to access hardware-specific features like extracting a list of images out of a video stream.

Difference from Other Bridge Technologies

Of course, QtWebKit is not the only bridge technology out there. NPAPI, for example, is a long-time standard for web-native bridging. Due to Qt's meta-object system, full applications leveraging web technologies are much easier to develop with the QtWebKit bridge than with NPAPI. NPAPI, however, is better for cross-browser plugins, due to it being an accepted standard.

When developing a plugin for a browser, NPAPI is recommended. When developing a full application utilizing HTML-rendering, the QtWebKit bridge is recommended.

Relationship with QtScript

The QtWebKit bridge is similar to QtScript, especially for some of the features described in the Making Applications Scriptable page. However, Qt 4.7 does not provide the full QtScript API for web applications. Full support is planned for future versions. You might notice that some of the features described here are an exact copy of the ones described in the Making Applications Scriptable page. That is because the QtWebKit bridge is a subset of that functionality, and this page tries to capture the full capabilities available through the QtWebKit bridge specifically.

Accessing QObjects

Making QObjects known to JavaScript via QWebFrame

By default, no QObjects are accessible through the web environment, for security reasons. When a web application wants to access a native QObject, it must explicitly grant access to this QObject, using the following call:

 // ...
 QWebFrame *frame = myWebPage->mainFrame();
 frame->addToJavaScriptWindowObject("someNameForMyObject", myObject);
 // ...

See QWebFrame::addToJavaScriptWindowObject() for more information.

Using Signals and Slots

The QtWebKit bridge adapts Qt's central Signals and Slots feature for scripting. There are three principal ways to use signals and slots with the QtWebKit bridge:

  • Hybrid C++/script: C++ application code connects a signal to a script function. This approach is useful if you have a QObject but don't want to expose the object itself to the scripting environment. You just want to define how the script responds to a signal and leave it up to the C++ side of your application to establish the connection between the C++ signal and the JavaScript slot.
  • Hybrid script/C++: A script can connect signals and slots to establish connections between pre-defined objects that the application exposes to the scripting environment. In this scenario, the slots themselves are still written in C++, but the definition of the connections is fully dynamic (script-defined).
  • Purely script-defined: A script can both define signal handler functions (effectively "slots written in JavaScript"), and set up the connections that utilize those handlers. For example, a script can define a function that will handle the QLineEdit::returnPressed() signal, and then connect that signal to the script function.

Note that QtScript functions such as qScriptConnect are unavilable in the web environment.

Signal to Function Connections

 function myInterestingScriptFunction() { ... }
 ...
 myQObject.somethingChanged.connect(myInterestingScriptFunction);

The call to connect() establishes a connection between the signal somethingChanged and the slot myInterestingScriptFunction. Whenever the object myObject emits the signal somethingChanged, the slot myInterestingScriptFunction gets called automatically.

The argument of connect() can be any JavaScript function as in the above example or a slot of a QObject as in the following example:

 myQObject.somethingChanged.connect(myOtherQObject.doSomething);

When the argument is a slot of a QObject, the argument types of the signal and the slot do not have to be compatible. If possible, the QtWebKit bridge converts the signal arguments such that they match the slot argument.

To disconnect a slot from a signal, you call the signal's disconnect() function with the slot as its argument:

 myQObject.somethingChanged.disconnect(myInterestingFunction);
 myQObject.somethingChanged.disconnect(myOtherQObject.doSomething);

When a script function is invoked in response to a signal, the this object will be the Global Object.

Signal to Member Function Connections

 myQObject.somethingChanged.connect(thisObject, function)

The call to connect() establishes a connection between the signal somethingChanged and the slot function. Whenever the object myObject emits the signal somethingChanged, the slot function of the object thisObject gets called automatically. Let's illustrate this with an example.

If you have a push button in a form, you typically want the form to do something in response to the button's clicked signal. The call to connect() makes sure that the function onClicked() is called whenever you click on the push button, that is, whenever the the signal clicked() is emitted by myButton. The slot onClicked() prints the value of x as stored in the form.

 var form = { x: 123 };
 var onClicked = function() { print(this.x); };
 myButton.clicked.connect(form, onClicked);

To disconnect a slot from a signal, you pass the same arguments to disconnect() as you passed to connect(). In general, this looks as follows:

 myQObject.somethingChanged.disconnect(thisObject, function);

Signal to Named Member Function Connections

 myQObject.somethingChanged.connect(thisObject, "functionName")

This form of the connect() function requires that the first argument thisObject is the object that will be bound to this when the function functionName is invoked in response to the signal somethingChanged. The second argument functionName specifies the name of a function that is connected to the signal. It refers to a member function of the object thisObject.

Note that the function is resolved when the connection is made, not when the signal is emitted.

 var obj = { x: 123, fun: function() { print(this.x); } };
 myQObject.somethingChanged.connect(obj, "fun");

To disconnect from the signal, pass the same arguments to disconnect() as you passed to connect:

 myQObject.somethingChanged.disconnect(thisObject, "functionName");

Error Handling

When connect() or disconnect() succeeds, the function will return undefined; otherwise, it will throw a script exception. You can obtain an error message from the resulting Error object. Example:

 try {
     myQObject.somethingChanged.connect(myQObject, "slotThatDoesntExist");
 } catch (e) {
     print(e);
 }

Emitting Signals from Scripts

To emit a signal from script code, you simply invoke the signal function, passing the relevant arguments:

 myQObject.somethingChanged("hello");

It is currently not possible to define a new signal in a script; i.e., all signals must be defined by C++ classes.

Overloaded Signals and Slots

When a signal or slot is overloaded, the QtWebKit bridge will attempt to pick the right overload based on the actual types of the QScriptValue arguments involved in the function invocation. For example, if your class has slots myOverloadedSlot(int) and myOverloadedSlot(QString), the following script code will behave reasonably:

 myQObject.myOverloadedSlot(10);   // will call the int overload
 myQObject.myOverloadedSlot("10"); // will call the QString overload

You can specify a particular overload by using array-style property access with the normalized signature of the C++ function as the property name:

 myQObject['myOverloadedSlot(int)']("10");   // call int overload; the argument is converted to an int
 myQObject['myOverloadedSlot(QString)'](10); // call QString overload; the argument is converted to a string

If the overloads have different number of arguments, the QtWebKit bridge will pick the overload with the argument count that best matches the actual number of arguments passed to the slot.

For overloaded signals, JavaScript will throw an error if you try to connect to the signal by name; you have to refer to the signal with the full normalized signature of the particular overload you want to connect to.

Invokable Methods

Both slots and signals are invokable from scripts by default. In addition, it is also possible to define a method that is invokable from scripts, although the method is neither a signal nor a slot. This is especially useful for functions with return types, as slots normally do not return anything (it would be meaningless to return a value from a slot, as the connected signals cannot handle return values). To make a non-slot method invokable, simply add the Q_INVOKABLE macro before its definition:

 class MyObject : public QObject
 {
     Q_OBJECT

 public:
     Q_INVOKABLE void thisMethodIsInvokableInJavaScript();
     void thisMethodIsNotInvokableInJavaScript();

     ...
 };

Accessing Properties

The properties of a QObject are available as properties of the corresponding JavaScript object. When you manipulate a property in script code, the C++ get/set method for that property will automatically be invoked. For example, if your C++ class has a property declared as follows:

     Q_PROPERTY(bool enabled READ enabled WRITE setEnabled)

then script code can do things like the following:

 myQObject.enabled = true;

 ...

 myQObject.enabled = !myQObject.enabled;

Accessing Child QObjects

Every named child of a QObject (that is, every child for which QObject::objectName() does not return the empty string) is by default available as a property of the JavaScript wrapper object. For example, if you have a QDialog with a child widget whose objectName property is "okButton", you can access this object in script code through the expression

 myDialog.okButton

Because objectName is itself a Q_PROPERTY, you can manipulate the name in script code to rename an object. For example:

 myDialog.okButton
 myDialog.okButton.objectName = "cancelButton";
 // from now on, myDialog.cancelButton references the button

Data types

When calling slots, receiving signals or accessing properties, usually some payload is involved. For example, a property "text" might return a QString parameter. The QtWebKit bridge does the job of converting between a given JavaScript data-type, and the expected or given Qt type. Each Qt type has a coresponding set of rules of how JavaScript treats it.

The data type conversions are also applicable for the data returned from non-void invokable methods.

Numbers

All Qt numeric data types are converted to or from a JavaScript number. These include int, short, float, double, and the portable Qt types (qreal, qint etc). A special case is QChar. If a slot expects a QChar, the QtWebKit bridge uses the Unicode value in case of a number and the first character in case of a string.

Note that non-standard (typedef'ed) number types are not automatically converted to or from a JavaScript number - we suggest to use standard number types for signals, slots and properties.

When a non-number is passed as an argument to a method or property that expects a number, the appropriate JavaScript conversion function (parseInt / parseFloat) is used.

Strings

When JavaScript accesses methods or properties that expect a QString, the QtWebKit bridge will automatically convert the value to a string (if it is not already a string), using the built-in JavaScript toString method.

When a QString is passed to JavaScript from a signal or a property, the QtWebKit bridge converts it into a JavaScript string.

Date & Time

Both QDate, QTime and QDateTime are automatically translated to or from the JavaScript Date object. If a number is passed as an argument to a method that expects one of the date/time types, the QtWebKit bridge treats it as a timestamp. If a sting is passed, QtWebKit tries the different Qt date parsing functions to perform the right translation.

Regular Expressions

The QtWebKit bridge automatically converts a JavaScript RegEx object to a QRegExp. If a string is passed to a method expecting a QRegExp, the string is converted to a QRegExp.

Lists

The QtWebKit bridge treats several types of lists in a special way: QVariantList, QStringList, QObjectList and QList<int>. When a slot or property expects one of those list types, the QtWebKit bridge tries to convert a JavaScript array into that type, converting each of the array's elements to the single-element type of the list.

The most useful type of list is QVariantList, which can be converted to and from any JavaScript array.

Compound (JSON) objects

JavaScript compound objects, also known as JSON objects, are variables that hold a list of key-value pairs, where all the keys are strings and the values can have any type. This translates very well to QVariantMap, which is nothing more than a QMap from QString to QVariant.

The seamless conversion between JSON objects and QVariantMap allows for a very convenient way of passing arbitrary structured data between C++ and the JavaScript environment. If the native QObject makes sure that compound values are converted to QVariantMaps and QVariantLists, JavaScript is guaranteed to receive them in a meaningful way.

Note that types that are not supported by JSON, such as JavaScript functions and getters/setters, are not converted.

QVariants

When a slot or property accepts a QVariant, the QtWebKit bridge creates a QVariant that best matches the argument passed by JavaScript. A string, for example, becomes a QVariant holding a QString, a normal JSON object becomes a QVariantMap, and a JavaScript array becomes a QVariantList.

Using QVariants generously in C++ in that way makes C++ programming feel a bit more like JavaScript programming, as it adds another level of indirection. Passing QVariants around like this is very flexible. The program can figure out the type of argument at runtime just like JavaScript would do. But doing so also takes away the type safety and robustness of C++. We recommended to use QVariants only for high-level functions, and to keep most of your QObjects type-safe.

QObjects

Pointers to a QObject or a QWidget can be used in signals, slots and properties. This object can then be used like an object that is exposed directly. Its slots can be invoked, its signals connected to, etc. However, this functionality is fairly limited - the type used has to be QObject* or QWidget*. If the type specified is a pointer to a non-QWidget subclass of QObject, the QtWebKit bridge does not recognize it as a QObject.

In general its advised to use care when passing QObjects as arguments, as those objects don't become owned by the JavaScript engine; That means that the application developer has to be extra careful not to try to access QObjects that have already been deleted by the native environment.

Pixmaps and Images

The QtWebKit bridge handles QPixmaps and QImages in a special way. Since QtWebKit stores QPixmaps to represent HTML images, QPixmaps coming from the native environment can be used directly inside WebKit. A QImage or a QPixmap coming from Qt is converted to an intermediate JavaScript object, which can be represented like this:

 {
     width: ...,
     height: ...,
     toDataURL: function() { ... },
     assignToHTMLImageElement: function(element) { ... }
 }

The JavaScript environment can then use the pixmap from Qt and display it inside the HTML environment, by assigning it to an existing <img> element with assignToHTMLImageElement(). It can also use the toDataURL() function, which allows using the pixmap as the src attribute of an image or as a background-image URL. Note that the toDataURL() function is costly and should be used with caution.

Example code:

C++:

 class MyObject : QObject {
     Q_OBJECT
     Q_PROPERTY(QPixmap myPixmap READ getPixmap)

 public:
     QPixmap getPixmap() const;
 };

 /* ... */

 MyObject myObject;
 myWebPage.mainFrame()->addToJavaScriptWindowObject("myObject", &myObject);

HTML:

 <html>
     <head>
         <script>
             function loadImage()
             {
                 myObject.myPixmap.assignToHTMLImageElement(document.getElementById("imageElement"));
             }
         </script>
     </head>
     <body onload="loadImage()">
         <img id="imageElement" width="300" height="200" />
     </body>
 </html>

When a Qt object expects a QImage or a QPixmap as input, and the argument passed is an HTML image element, the QtWebKit bridge would convert the pixmap assigned to that image element into a QPixmap or a QImage.

QWebElement

A signal, slot or property that expects or returns a QWebElement can work seamlessly with JavaScript references to DOM elements. The JavaScript environment can select DOM elements, keep them in variables, then pass them to Qt as a QWebElement, and receive them back. Example:

C++:

 class MyObject : QObject {
         Q_OBJECT

     public slots:
         void doSomethingWithWebElement(const QWebElement&);
     };

     /* ... */

     MyObject myObject;
     myWebPage.mainFrame()->addToJavaScriptWindowObject("myObject", &myObject);

HTML:

 <html>
      <head>
          <script>
              function runExample() {
                 myObject.doSomethingWithWebElement(document.getElementById("someElement"));
              }
         </script>
      </head>
      <body onload="runExample()">
          <span id="someElement">Text</span>
      </body>
  </html>

This is specifically useful to create custom renderers or extensions to the web environment. Instead of forcing Qt to select the element, the web environment selects the element and then sends the selected element directly to Qt.

Note that QWebElements are not thread safe - an object handling them has to live in the UI thread.

Architecture Issues

Limiting the Scope of the Hybrid Layer

When using QtWebKit's hybrid features, it is a common pitfall to make the API exposed to JavaScript very rich and use all its features. This, however, leads to complexity and can create bugs that are hard to find. Instead, it is advisable to keep the hybrid layer small and manageable: create a gate only when there's an actual need for it, i.e. there's a new native enabler that requires a direct interface to the application layer. Sometimes new functionality is better handled internally in the native layer or in the web layer; simplicity is your friend.

This usually becomes more apparent when the hybrid layer can create or destroy objects, or uses signals, slots or properties with a QObject* argument. It is advised to be very careful and to treat an exposed QObject as a system - with careful attention to memory management and object ownership.

Internet Security

When exposing native objects to an open web environment, it is important to understand the security implications. Think whether the exposed object enables the web environment access things that shouldn't be open, and whether the web content loaded by that web page comes from a trusted source. In general, when exposing native QObjects that give the web environment access to private information or to functionality that's potentially harmful to the client, such exposure should be balanced by limiting the web page's access to trusted URLs only with HTTPS, and by utilizing other measures as part of a security strategy.

Notes provided by the Qt Community

No notes