WPF and Win32 Interoperation
This topic provides an overview of how to interoperate WPF and Win32 code. Windows Presentation Foundation (WPF) provides a rich environment for creating applications. However, when you have a substantial investment in Win32 code, it might be more effective to reuse some of that code.
There are two basic techniques for interoperation between WPF and Win32 code.
Host WPF content in a Win32 window. With this technique, you can use the advanced graphics capabilities of WPF within the framework of a standard Win32 window and application.
Host a Win32 window in WPF content. With this technique, you can use an existing custom Win32 control in the context of other WPF content, and pass data across the boundaries.
Each of these techniques is conceptually introduced in this topic. For a more code-oriented illustration of hosting WPF in Win32, see Walkthrough: Hosting WPF Content in Win32. For a more code-oriented illustration of hosting Win32 in WPF, see Walkthrough: Hosting a Win32 Control in WPF.
WPF APIs are managed code, but most existing Win32 programs are written in unmanaged C++. You cannot call WPF APIs from a true unmanaged program. However, by using the /clr option with the Microsoft Visual C++ compiler, you can create a mixed managed-unmanaged program where you can seamlessly mix managed and unmanaged API calls.
One project-level complication is that you cannot compile Extensible Application Markup Language (XAML) files into a C++ project. There are several project division techniques to compensate for this.
Create a C# DLL that contains all your XAML pages as a compiled assembly, and then have your C++ executable include that DLL as a reference.
Create a C# executable for the WPF content, and have it reference a C++ DLL that contains the Win32 content.
Use Load to load any XAML at run time, instead of compiling your XAML.
Do not use XAML at all, and write all your WPF in code, building up the element tree from Application.
Use whatever approach works best for you.
If you have not used C++/CLI before, you might notice some "new" keywords such as gcnew and nullptr in the interoperation code examples. These keywords supersede the older double-underscore syntax (__gc) and provide a more natural syntax for managed code in C++. To learn more about the C++/CLI managed features, see Component Extensions for Runtime Platforms and Hello, C++/CLI.
To make the most of WPF "HWND interop", you need to understand how WPF uses HWNDs. For any HWND, you cannot mix WPF rendering with DirectX rendering or GDI / GDI+ rendering. This has a number of implications. Primarily, in order to mix these rendering models at all, you must create an interoperation solution, and use designated segments of interoperation for each rendering model that you choose to use. Also, the rendering behavior creates an "airspace" restriction for what your interoperation solution can accomplish. The "airspace" concept is explained in greater detail in the topic Technology Regions Overview.
All WPF elements on the screen are ultimately backed by a HWND. When you create a WPF Window, WPF creates a top-level HWND, and uses an HwndSource to put the Window and its WPF content inside the HWND. The rest of your WPF content in the application shares that singular HWND. An exception is menus, combo box drop downs, and other pop-ups. These elements create their own top-level window, which is why a WPF menu can potentially go past the edge of the window HWND that contains it. When you use HwndHost to put an HWND inside WPF, WPF informs Win32 how to position the new child HWND relative to the WPF Window HWND.
A related concept to HWND is transparency within and between each HWND. This is also discussed in the topic Technology Regions Overview.
The key to hosting a WPF on a Win32 window is the HwndSource class. This class wraps the WPF content in a Win32 window, so that the WPF content can be incorporated into your user interface (UI) as a child window. The following approach combines the Win32 and WPF in a single application.
Implement your WPF content (the content root element) as a managed class. Typically, the class inherits from one of the classes that can contain multiple child elements and/or used as a root element, such as DockPanel or Page. In subsequent steps, this class is referred to as the WPF content class, and instances of the class are referred to as WPF content objects.
Implement a Win32 application with C++/CLI. If you are starting with an existing unmanaged C++ application, you can usually enable it to call managed code by changing your project settings to include the /clr compiler flag (the full scope of what might be necessary to support /clr compilation is not described in this topic).
Set the threading model to Single Threaded Apartment (STA). WPF uses this threading model.
Handle the WM_CREATE notification in your window procedure.
Within the handler (or a function that the handler calls), do the following:
Create a new HwndSource object with the parent window HWND as its parent parameter.
Create an instance of your WPF content class.
Implement a managed class that contains a static field that holds a reference to your WPF content object. This class allows you to get a reference to the WPF content object from your Win32 code, but more importantly it prevents your HwndSource from being inadvertently garbage collected.
Receive notifications from the WPF content object by attaching a handler to one or more of the WPF content object events.
Communicate with the WPF content object by using the reference that you stored in the static field to set properties, call methods, etc.
You can do some or all of the WPF content class definition for Step One in XAML using the default partial class of the content class, if you produce a separate assembly and then reference it. Although you typically include an Application object as part of compiling the XAML into an assembly, you do not end up using that Application as part of the interoperation, you just use one or more of the root classes for XAML files referred to by the application and reference their partial classes. The remainder of the procedure is essentially similar to that outlined above.
Each of these steps is illustrated through code in the topic Walkthrough: Hosting WPF Content in Win32.
The key to hosting a Win32 window within other WPF content is the HwndHost class. This class wraps the window in a WPF element which can be added to a WPF element tree. HwndHost also supports APIs that allow you to do such tasks as process messages for the hosted window. The basic procedure is:
Create an element tree for a WPF application (can be through code or markup). Find an appropriate and permissible point in the element tree where the HwndHost implementation can be added as a child element. In the remainder of these steps, this element is referred to as the reserving element.
Derive from HwndHost to create an object that holds your Win32 content.
In that host class, override the HwndHost method BuildWindowCore. Return the HWND of the hosted window. You might want to wrap the actual control(s) as a child window of the returned window; wrapping the controls in a host window provides a simple way for your WPF content to receive notifications from the controls. This technique helps correct for some Win32 issues regarding message handling at the hosted control boundary.
Override the HwndHost methods DestroyWindowCore and WndProc. The intention here is to process cleanup and remove references to the hosted content, particularly if you created references to unmanaged objects.
In your code-behind file, create an instance of the control hosting class and make it a child of the reserving element. Typically you would use an event handler such as Loaded, or use the partial class constructor. But you could also add the interoperation content through a runtime behavior.
Process selected window messages, such as control notifications. There are two approaches. Both provide identical access to the message stream, so your choice is largely a matter of programming convenience.
Have the hosting WPF element process the messages by handling the MessageHook event. This event is raised for every message that is sent to the main window procedure of the hosted window.
You cannot process messages from windows that are out of process using WndProc.
Communicate with the hosted window by using platform invoke to call the unmanaged SendMessage function.
Following these steps creates an application that works with mouse input. You can add tabbing support for your hosted window by implementing the IKeyboardInputSink interface.
Each of these steps is illustrated through code in the topic Walkthrough: Hosting a Win32 Control in WPF.
You can think of HwndHost as a special control. (Technically, HwndHost is a FrameworkElement derived class, not a Control derived class, but it can be considered a control for purposes of interoperation.) HwndHost abstracts the underlying Win32 nature of the hosted content such that the remainder of WPF considers the hosted content to be another control-like object, which should render and process input. HwndHost generally behaves like any other WPF FrameworkElement, although there are some important differences around output (drawing and graphics) and input (mouse and keyboard) based on limitations of what the underlying HWNDs can support.
FrameworkElement, which is the HwndHost base class, has quite a few properties that imply changes to the UI. These include properties such as FrameworkElement.FlowDirection, which changes the layout of elements within that element as a parent. However, most of these properties are not mapped to possible Win32 equivalents, even if such equivalents might exist. Too many of these properties and their meanings are too rendering-technology specific for mappings to be practical. Therefore, setting properties such as FlowDirection on HwndHost has no effect.
HwndHost cannot be rotated, scaled, skewed, or otherwise affected by a Transform.
HwndHost does not support the Opacity property (alpha blending). If content inside the HwndHost performs System.Drawing operations that include alpha information, that is itself not a violation, but the HwndHost as a whole only supports Opacity = 1.0 (100%).
HwndHost will appear on top of other WPF elements in the same top-level window. However, a ToolTip or ContextMenu generated menu is a separate top-level window, and so will behave correctly with HwndHost.
In general, while input devices are scoped within the HwndHost hosted Win32 region, input events go directly to Win32.
Related stylus properties and events are analogous, and do not report information while the stylus is over HwndHost.
Tabbing between Win32 and WPF components
Mnemonics and accelerators that work both when focus is within a Win32 component and when it is within a WPF component.
The HwndHost and HwndSource classes both provide implementations of IKeyboardInputSink, but they may not handle all the input messages that you want for more advanced scenarios. Override the appropriate methods to get the keyboard behavior you want.
The interfaces only provide support for what happens on the transition between the WPF and Win32 regions. Within the Win32 region, tabbing behavior is entirely controlled by the Win32 implemented logic for tabbing, if any.