Материал из Wiki.crossplatform.ru

Версия 13:59, 30 октября 2008

Внимание: Актуальная версия перевода документации находится здесь

__NOTOC__

How to Create Qt Plugins

Qt provides two APIs for creating plugins:

• A higher-level API for writing extensions to Qt itself: custom database drivers, image formats, text codecs, custom styles, etc.
• A lower-lever API for extending Qt applications.

For example, if you want to write a custom QStyle subclass and have Qt applications load it dynamically, you would use the higher-level API.

Since the higher-level API is built on top of the lower-level API, some issues are common to both.

If you want to provide plugins for use with Qt Designer, see the QtDesigner module documentation.

The Higher-Level API: Writing Qt Extensions

Writing a plugin that extends Qt itself is achieved by subclassing the appropriate plugin base class, implementing a few functions, and adding a macro.

There are several plugin base classes. Derived plugins are stored by default in sub-directories of the standard plugin directory. Qt will not find plugins if they are not stored in the right directory.

But where is the plugins directory? When the application is run, Qt will first treat the application's executable directory as the pluginsbase. For example if the application is in C:\Program Files\MyApp and has a style plugin, Qt will look in C:\Program Files\MyApp\styles. (See QCoreApplication::applicationDirPath() for how to find out where the application's executable is.) Qt will also look in the directory specified by QLibraryInfo::location( QLibraryInfo::PluginsPath), which typically is located in QTDIR/plugins (where QTDIR is the directory where Qt is installed). If you want Qt to look in additional places you can add as many paths as you need with calls to QCoreApplication::addLibraryPath(). And if you want to set your own path or paths you can use QCoreApplication::setLibraryPaths(). You can also use a qt.conf file to override the hard-coded paths that are compiled into the Qt library. For more information, see the Using qt.conf documentation. Yet another possibility is to set the QT_PLUGIN_PATH environment variable before running the application. If set, Qt will look for plugins in the paths (separated by the system path separator) specified in the variable.

Suppose that you have a new style class called MyStyle that you want to make available as a plugin. The required code is straightforward, here is the class definition (mystyleplugin.h):

 class MyStylePlugin : public QStylePlugin
 {
 public:
     QStringList keys() const;
     QStyle *create(const QString &key);
 };

Ensure that the class implementation is located in a .cpp file (including the class definition):

 #include "mystyleplugin.h"
 
 QStringList MyStylePlugin::keys() const
 {
     return QStringList() << "MyStyle";
 }
 
 QStyle *MyStylePlugin::create(const QString &amp;key)
 {
     if (key.toLower() == "mystyle")
         return new MyStyle;
     return 0;
 }
 
 Q_EXPORT_PLUGIN2(pnp_mystyleplugin, MyStylePlugin)

(Note that QStylePlugin is case insensitive, and the lower-case version of the key is used in our create() implementation; most other plugins are case sensitive.)

For database drivers, image formats, text codecs, and most other plugin types, no explicit object creation is required. Qt will find and create them as required. Styles are an exception, since you might want to set a style explicitly in code. To apply a style, use code like this:

 QApplication::setStyle(QStyleFactory::create("MyStyle"));

Some plugin classes require additional functions to be implemented. See the class documentation for details of the virtual functions that must be reimplemented for each type of plugin.

Qt applications automatically know which plugins are available, because plugins are stored in the standard plugin subdirectories. Because of this applications don't require any code to find and load plugins, since Qt handles them automatically.

The default directory for plugins is QTDIR/plugins (where QTDIR is the directory where Qt is installed), with each type of plugin in a subdirectory for that type, e.g. styles. If you want your applications to use plugins and you don't want to use the standard plugins path, have your installation process determine the path you want to use for the plugins, and save the path, e.g. using QSettings, for the application to read when it runs. The application can then call QCoreApplication::addLibraryPath() with this path and your plugins will be available to the application. Note that the final part of the path (e.g., styles) cannot be changed.

The normal way to include a plugin with an application is either to compile it in with the application or to compile it into a dynamic library and use it like any other library. If you want the plugin to be loadable then one approach is to create a subdirectory under the application and place the plugin in that directory. For more information about deployment, see the Deploying Qt Applications documentation.

The Style Plugin Example shows how to implement a plugin that extends the QStylePlugin base class.

The Lower-Level API: Extending Qt Applications

Not only Qt itself but also Qt application can be extended through plugins. This requires the application to detect and load plugins using QPluginLoader. In that context, plugins may provide arbitrary functionality and are not limited to database drivers, image formats, text codecs, styles, and the other types of plugin that extend Qt's functionality.

Making an application extensible through plugins involves the following steps:

1. Define a set of interfaces (classes with only pure virtual functions) used to talk to the plugins.
2. Use the Q_DECLARE_INTERFACE() macro to tell Qt's meta-object system about the interface.
3. Use QPluginLoader in the application to load the plugins.
4. Use qobject_cast() to test whether a plugin implements a given interface.

Writing a plugin involves these steps:

1. Declare a plugin class that inherits from QObject and from the interfaces that the plugin wants to provide.
2. Use the Q_INTERFACES() macro to tell Qt's meta-object system about the interfaces.
3. Export the plugin using the Q_EXPORT_PLUGIN2() macro.
4. Build the plugin using an suitable .pro file.

For example, here's the definition of an interface class:

 class FilterInterface
 {
 public:
     virtual ~FilterInterface() {}
 
     virtual QStringList filters() const = 0;
     virtual QImage filterImage(const QString &filter, const QImage &image,
                                QWidget *parent) = 0;
 };
 
 Q_DECLARE_INTERFACE(FilterInterface,
                     "com.trolltech.PlugAndPaint.FilterInterface/1.0")

Here's the definition of a plugin class that implements that interface:

 #include <QObject>
 #include <QStringList>
 #include <QImage>
 
 #include <plugandpaint/interfaces.h>
 
 class ExtraFiltersPlugin : public QObject, public FilterInterface
 {
     Q_OBJECT
     Q_INTERFACES(FilterInterface)
 
 public:
     QStringList filters() const;
     QImage filterImage(const QString &amp;filter, const QImage &amp;image,
                        QWidget *parent);
 };

The Plug & Paint example documentation explains this process in detail. See also Creating Custom Widgets for Qt Designer for information about issues that are specific to Qt Designer. You can also take a look at the Echo Plugin Example is a more trivial example on how to implement a plugin that extends Qt applications.

Loading and Verifying Plugins Dynamically

When loading plugins, the Qt library does some sanity checking to determine whether or not the plugin can be loaded and used. This provides the ability to have multiple versions and configurations of the Qt library installed side by side.

• Plugins linked with a Qt library that has a higher major and/or minor version number will not be loaded by a library with a lower major and/or minor version number.
  Rationale: A plugin linked against a newer Qt library may use new features that are not available in older versions. Trolltech has a policy of adding new features and APIs only between minor releases, which is why this test only looks at the major and minor version numbers, and not at the patch version number.
  
• The Qt library and all plugins are built using a build key. The build key in the Qt library is examined against the build key in the plugin, and if they match, the plugin is loaded. If the build keys do not match, then the Qt library refuses to load the plugin.
  Rationale: See the Rationale for the build key below.
  

When building plugins to extend an application, it is important to ensure that the plugin is configured in the same way as the application. This means that if the application was built in release mode, plugins should be built in release mode, too.

If you configure Qt to be built in both debug and release modes, but only build applications in release mode, you need to ensure that your plugins are also built in release mode. By default, if a debug build of Qt is available, plugins will only be built in debug mode. To force the plugins to be built in release mode, add the following line to the plugin's project file:

 CONFIG += release

This will ensure that the plugin is compatible with the version of the library used in the application.

The Build Key

When loading plugins, Qt checks the build key of each plugin against its own configuration to ensure that only compatible plugins are loaded; any plugins that are configured differently are not loaded.

The build key contains the following information:

• Architecture, operating system and compiler.
  Rationale: In cases where different versions of the same compiler do not produce binary compatible code, the version of the compiler is also present in the build key.
  
• Configuration of the Qt library. The configuration is a list of the missing features that affect the available API in the library.
  Rationale: Two different configurations of the same version of the Qt library are not binary compatible. The Qt library that loads the plugin uses the list of (missing) features to determine if the plugin is binary compatible.
  
  Note: There are cases where a plugin can use features that are available in two different configurations. However, the developer writing plugins would need to know which features are in use, both in their plugin and internally by the utility classes in Qt. The Qt library would require complex feature and dependency queries and verification when loading plugins. Requiring this would place an unnecessary burden on the developer, and increase the overhead of loading a plugin. To reduce both development time and application runtime costs, a simple string comparision of the build keys is used.
  
• Optionally, an extra string may be specified on the configure script command line.
  Rationale: When distributing binaries of the Qt library with an application, this provides a way for developers to write plugins that can only be loaded by the library with which the plugins were linked.
  

For debugging purposes, it is possible to override the run-time build key checks by setting the QT_NO_PLUGIN_CHECK environment variable to a non-zero value in the environment from which your application is launched.

Static Plugins

Plugins can be linked statically against your application. If you build the static version of Qt, this is the only option for including Qt's predefined plugins.

When compiled as a static library, Qt provides the following static plugins:

To link statically against those plugins, you need to use the Q_IMPORT_PLUGIN() macro in your application and you need to add the required plugins to your build using QTPLUGIN. For example, in your main.cpp:

 #include <QApplication>
 #include <QtPlugin>
 
 Q_IMPORT_PLUGIN(qjpeg)
 Q_IMPORT_PLUGIN(qgif)
 Q_IMPORT_PLUGIN(qkrcodecs)
 
 int main(int argc, char *argv[])
 {
     QApplication app(argc, argv);
     ...
     return app.exec();
 }

In the .pro file for your application, you need the following entry:

 QTPLUGIN     += qjpeg \
                 qgif \
                 qkrcodecs

It is also possible to create your own static plugins, by following these steps:

1. Add CONFIG += static to your plugin's .pro file.
2. Use the Q_IMPORT_PLUGIN() macro in your application.
3. Link your application with your plugin library using LIBS in the .pro file.

See the Plug & Paint example and the associated Basic Tools plugin for details on how to do this.

The Plugin Cache

In order to speed up loading and validation of plugins, some of the information that is collected when plugins are loaded is cached through QSettings. This includes information about whether or not a plugin was successfully loaded, so that subsequent load operations don't try to load an invalid plugin. However, if the "last modified" timestamp of a plugin has changed, the plugin's cache entry is invalidated and the plugin is reloaded regardless of the values in the cache entry, and the cache entry itself is updated with the new result.

This also means that the timestamp must be updated each time the plugin or any dependent resources (such as a shared library) is updated, since the dependent resources might influence the result of loading a plugin.

Sometimes, when developing plugins, it is necessary to remove entries from the plugin cache. Since Qt uses QSettings to manage the plugin cache, the locations of plugins are platform-dependent; see the QSettings documentation for more information about each platform.

For example, on Windows the entries are stored in the registry, and the paths for each plugin will typically begin with either of these two strings:

 HKEY_CURRENT_USER\Software\Trolltech\OrganizationDefaults\Qt Plugin Cache 4.2.debug
 HKEY_CURRENT_USER\Software\Trolltech\OrganizationDefaults\Qt Plugin Cache 4.2.false

Debugging Plugins

There are a number of issues that may prevent correctly-written plugins from working with the applications that are designed to use them. Many of these are related to differences in the way that plugins and applications have been built, often arising from separate build systems and processes.

The following table contains descriptions of the common causes of problems developers experience when creating plugins:

You can also use the QT_DEBUG_PLUGINS environment variable to obtain diagnostic information from Qt about each plugin it tries to load. Set this variable to a non-zero value in the environment from which your application is launched.

See also QPluginLoader, QLibrary, and Plug & Paint Example.