Custom automation peers

Describes the concept of automation peers for Microsoft UI Automation, and how you can provide automation support for your own custom UI class.

UI Automation provides a framework that automation clients can use to examine or operate the user interfaces of a variety of UI platforms and frameworks. If you are writing a Windows app, the classes that you use for your UI already provide UI Automation support. You can derive from existing, non-sealed classes to define a new kind of UI control or support class. In the process of doing so, your class might add behavior that should have accessibility support but that the default UI Automation support does not cover. In this case, you should extend the existing UI Automation support by deriving from the AutomationPeer class that the base implementation used, adding any necessary support to your peer implementation, and informing the Windows app control infrastructure that it should create your new peer.

UI Automation enables not only accessibility applications and assistive technologies, such as screen readers, but also quality-assurance (test) code. In either scenario, UI Automation clients can examine user-interface elements and simulate user interaction with your app from other code outside your app. For info about UI Automation across all platforms and in its wider meaning, see UI Automation Overview.

There are two distinct audiences who use the UI Automation framework.

  • UI Automation clients call UI Automation APIs to learn about all of the UI that is currently displayed to the user. For example, an assistive technology such as a screen reader acts as a UI Automation client. The UI is presented as a tree of automation elements that are related. The UI Automation client might be interested in just one app at a time, or in the entire tree. The UI Automation client can use UI Automation APIs to navigate the tree and to read or change information in the automation elements.
  • UI Automation providers contribute information to the UI Automation tree, by implementing APIs that expose the elements in the UI that they introduced as part of their app. When you create a new control, you should now act as a participant in the UI Automation provider scenario. As a provider, you should ensure that all UI Automation clients can use the UI Automation framework to interact with your control for both accessibility and testing purposes.

Typically there are parallel APIs in the UI Automation framework: one API for UI Automation clients and another, similarly named API for UI Automation providers. For the most part, this topic covers the APIs for the UI Automation provider, and specifically the classes and interfaces that enable provider extensibility in that UI framework. Occasionally we mention UI Automation APIs that the UI Automation clients use, to provide some perspective, or provide a lookup table that correlates the client and provider APIs. For more info about the client perspective, see UI Automation Client Programmer's Guide.

Note

UI Automation clients don't typically use managed code and aren't typically implemented as a UWP app (they are usually desktop apps). UI Automation is based on a standard and not a specific implementation or framework. Many existing UI Automation clients, including assistive technology products such as screen readers, use Component Object Model (COM) interfaces to interact with UI Automation, the system, and the apps that run in child windows. For more info on the COM interfaces and how to write a UI Automation client using COM, see UI Automation Fundamentals.

Determining the existing state of UI Automation support for your custom UI class

Before you attempt to implement an automation peer for a custom control, you should test whether the base class and its automation peer already provides the accessibility or automation support that you need. In many cases, the combination of the FrameworkElementAutomationPeer implementations, specific peers, and the patterns they implement can provide a basic but satisfactory accessibility experience. Whether this is true depends on how many changes you made to the object model exposure to your control versus its base class. Also, this depends on whether your additions to base class functionality correlate to new UI elements in the template contract or to the visual appearance of the control. In some cases your changes might introduce new aspects of user experience that require additional accessibility support.

Even if using the existing base peer class provides the basic accessibility support, it is still a best practice to define a peer so that you can report precise ClassName information to UI Automation for automated testing scenarios. This consideration is especially important if you are writing a control that is intended for third-party consumption.

Automation peer classes

The UWP builds on existing UI Automation techniques and conventions used by previous managed-code UI frameworks such as Windows Forms, Windows Presentation Foundation (WPF) and Microsoft Silverlight. Many of the control classes and their function and purpose also have their origin in a previous UI framework.

By convention, peer class names begin with the control class name and end with "AutomationPeer". For example, ButtonAutomationPeer is the peer class for the Button control class.

Note

For purposes of this topic, we treat the properties that are related to accessibility as being more important when you implement a control peer. But for a more general concept of UI Automation support, you should implement a peer in accordance with recommendations as documented by the UI Automation Provider Programmer's Guide and UI Automation Fundamentals. Those topics don't cover the specific AutomationPeer APIs that you would use to provide the information in the UWP framework for UI Automation, but they do describe the properties that identify your class or provide other information or interaction.

Peers, patterns and control types

A control pattern is an interface implementation that exposes a particular aspect of a control's functionality to a UI Automation client. UI Automation clients use the properties and methods exposed through a control pattern to retrieve information about capabilities of the control, or to manipulate the control's behavior at run time.

Control patterns provide a way to categorize and expose a control's functionality independent of the control type or the appearance of the control. For example, a control that presents a tabular interface uses the Grid control pattern to expose the number of rows and columns in the table, and to enable a UI Automation client to retrieve items from the table. As other examples, the UI Automation client can use the Invoke control pattern for controls that can be invoked, such as buttons, and the Scroll control pattern for controls that have scroll bars, such as list boxes, list views, or combo boxes. Each control pattern represents a separate type of functionality, and control patterns can be combined to describe the full set of functionality supported by a particular control.

Control patterns relate to UI as interfaces relate to COM objects. In COM, you can query an object to ask what interfaces it supports and then use those interfaces to access functionality. In UI Automation, UI Automation clients can query a UI Automation element to find out which control patterns it supports, and then interact with the element and its peered control through the properties, methods, events, and structures exposed by the supported control patterns.

One of the main purposes of an automation peer is to report to a UI Automation client which control patterns the UI element can support through its peer. To do this, UI Automation providers implement new peers that change the GetPattern method behavior by overriding the GetPatternCore method. UI Automation clients make calls that the UI Automation provider maps to calling GetPattern. UI Automation clients query for each specific pattern that they want to interact with. If the peer supports the pattern, it returns an object reference to itself; otherwise it returns null. If the return is not null, the UI Automation client expects that it can call APIs of the pattern interface as a client, in order to interact with that control pattern.

A control type is a way to broadly define the functionality of a control that the peer represents. This is a different concept than a control pattern because while a pattern informs UI Automation what info it can get or what actions it can perform through a particular interface, the control type exists one level above that. Each control type has guidance about these aspects of UI Automation:

  • UI Automation control patterns: A control type might support more than one pattern, each of which represents a different classification of info or interaction. Each control type has a set of control patterns that the control must support, a set that is optional, and a set that the control must not support.
  • UI Automation property values: Each control type has a set of properties that the control must support. These are the general properties, as described in UI Automation Properties Overview, not the ones that are pattern-specific.
  • UI Automation events: Each control type has a set of events that the control must support. Again these are general, not pattern-specific, as described in UI Automation Events Overview.
  • UI Automation tree structure: Each control type defines how the control must appear in the UI Automation tree structure.

Regardless of how automation peers for the framework are implemented, UI Automation client functionality isn't tied to the UWP, and in fact it's likely that existing UI Automation clients such as assistive technologies will use other programming models, such as COM. In COM, clients can QueryInterface for the COM control pattern interface that implements the requested pattern or the general UI Automation framework for properties, events or tree examination. For the patterns, the UI Automation framework marshals that interface code across into UWP code running against the app's UI Automation provider and the relevant peer.

When you implement control patterns for a managed-code framework such as a UWP app using C# or Microsoft Visual Basic, you can use .NET Framework interfaces to represent these patterns instead of using the COM interface representation. For example, the UI Automation pattern interface for a Microsoft .NET provider implementation of the Invoke pattern is IInvokeProvider.

For a list of control patterns, provider interfaces, and their purpose, see Control patterns and interfaces. For the list of the control types, see UI Automation Control Types Overview.

Guidance for how to implement control patterns

The control patterns and what they're intended for are part of a larger definition of the UI Automation framework, and don't just apply to the accessibility support for a UWP app. When you implement a control pattern you should make sure you're implementing it in a way that matches the guidance as documented in these docs and also in the UI Automation specification. If you're looking for guidance, you can generally use the Microsoft documentation and won't need to refer to the specification. Guidance for each pattern is documented here: Implementing UI Automation Control Patterns. You'll notice that each topic under this area has an "Implementation Guidelines and Conventions" section and "Required Members" section. The guidance usually refers to specific APIs of the relevant control pattern interface in the Control Pattern Interfaces for Providers reference. Those interfaces are the native/COM interfaces (and their APIs use COM-style syntax). But everything you see there has an equivalent in the Windows.UI.Xaml.Automation.Provider namespace.

If you're using the default automation peers and expanding on their behavior, those peers have already been written in conformance to UI Automation guidelines. If they support control patterns, you can rely on that pattern support conforming with guidance at Implementing UI Automation Control Patterns. If a control peer reports that it's representative of a control type defined by UI Automation, then the guidance documented at Supporting UI Automation Control Types has been followed by that peer.

Nevertheless you might need additional guidance for control patterns or control types in order to follow the UI Automation recommendations in your peer implementation. That would be particularly true if you're implementing pattern or control type support that doesn't yet exist as a default implementation in a UWP control. For example, the pattern for annotations isn't implemented in any of the default XAML controls. But you might have an app that uses annotations extensively and therefore you want to surface that functionality to be accessible. For this scenario, your peer should implement IAnnotationProvider and should probably report itself as the Document control type with appropriate properties to indicate that your documents support annotation.

We recommend that you use the guidance that you see for the patterns under Implementing UI Automation Control Patterns or control types under Supporting UI Automation Control Types as orientation and general guidance. You might even try following some of the API links for descriptions and remarks as to the purpose of the APIs. But for syntax specifics that are needed for UWP app programming, find the equivalent API within the Windows.UI.Xaml.Automation.Provider namespace and use those reference pages for more info.

Built-in automation peer classes

In general, elements implement an automation peer class if they accept UI activity from the user, or if they contain information needed by users of assistive technologies that represent the interactive or meaningful UI of apps. Not all UWP visual elements have automation peers. Examples of classes that implement automation peers are Button and TextBox. Examples of classes that do not implement automation peers are Border and classes based on Panel, such as Grid and Canvas. A Panel has no peer because it is providing a layout behavior that is visual only. There is no accessibility-relevant way for the user to interact with a Panel. Whatever child elements a Panel contains are instead reported to UI Automation trees as child elements of the next available parent in the tree that has a peer or element representation.

UI Automation and UWP process boundaries

Typically, UI Automation client code that accesses a UWP app runs out-of-process. The UI Automation framework infrastructure enables information to get across the process boundary. This concept is explained in more detail in UI Automation Fundamentals.

OnCreateAutomationPeer

All classes that derive from UIElement contain the protected virtual method OnCreateAutomationPeer. The object initialization sequence for automation peers calls OnCreateAutomationPeer to get the automation peer object for each control and thus to construct a UI Automation tree for run-time use. UI Automation code can use the peer to get information about a control’s characteristics and features and to simulate interactive use by means of its control patterns. A custom control that supports automation must override OnCreateAutomationPeer and return an instance of a class that derives from AutomationPeer. For example, if a custom control derives from the ButtonBase class, the object returned by OnCreateAutomationPeer should derive from ButtonBaseAutomationPeer.

If you're writing a custom control class and intend to also supply a new automation peer, you should override the OnCreateAutomationPeer method for your custom control so that it returns a new instance of your peer. Your peer class must derive directly or indirectly from AutomationPeer.

For example, the following code declares that the custom control NumericUpDown should use the peer NumericUpDownPeer for UI Automation purposes.

using Windows.UI.Xaml.Automation.Peers;
...
public class NumericUpDown : RangeBase {
    public NumericUpDown() {
    // other initialization; DefaultStyleKey etc.
    }
    ...
    protected override AutomationPeer OnCreateAutomationPeer()
    {
        return new NumericUpDownAutomationPeer(this);
    }
}
Public Class NumericUpDown
    Inherits RangeBase
    ' other initialization; DefaultStyleKey etc.
       Public Sub New()
       End Sub
       Protected Overrides Function OnCreateAutomationPeer() As AutomationPeer
              Return New NumericUpDownAutomationPeer(Me)
       End Function
End Class
// NumericUpDown.idl
namespace MyNamespace
{
    runtimeclass NumericUpDown : Windows.UI.Xaml.Controls.Primitives.RangeBase
    {
        NumericUpDown();
        Int32 MyProperty;
    }
}

// NumericUpDown.h
...
struct NumericUpDown : NumericUpDownT<NumericUpDown>
{
	...
    Windows::UI::Xaml::Automation::Peers::AutomationPeer OnCreateAutomationPeer()
    {
        return winrt::make<MyNamespace::implementation::NumericUpDownAutomationPeer>(*this);
    }
};
//.h
public ref class NumericUpDown sealed : Windows::UI::Xaml::Controls::Primitives::RangeBase
{
// other initialization not shown
protected:
    virtual AutomationPeer^ OnCreateAutomationPeer() override
    {
         return ref new NumericUpDownAutomationPeer(this);
    }
};

Note

The OnCreateAutomationPeer implementation should do nothing more than initialize a new instance of your custom automation peer, passing the calling control as owner, and return that instance. Do not attempt additional logic in this method. In particular, any logic that could potentially lead to destruction of the AutomationPeer within the same call may result in unexpected runtime behavior.

In typical implementations of OnCreateAutomationPeer, the owner is specified as this or Me because the method override is in the same scope as the rest of the control class definition.

The actual peer class definition can be done in the same code file as the control or in a separate code file. The peer definitions all exist in the Windows.UI.Xaml.Automation.Peers namespace that is a separate namespace from the controls that they provide peers for. You can choose to declare your peers in separate namespaces also, as long as you reference the necessary namespaces for the OnCreateAutomationPeer method call.

Choosing the correct peer base class

Make sure that your AutomationPeer is derived from a base class that gives you the best match for the existing peer logic of the control class you are deriving from. In the case of the previous example, because NumericUpDown derives from RangeBase, there is a RangeBaseAutomationPeer class available that you should base your peer on. By using the closest matching peer class in parallel to how you derive the control itself, you can avoid overriding at least some of the IRangeValueProvider functionality because the base peer class already implements it.

The base Control class does not have a corresponding peer class. If you need a peer class to correspond to a custom control that derives from Control, derive the custom peer class from FrameworkElementAutomationPeer.

If you derive from ContentControl directly, that class has no default automation peer behavior because there is no OnCreateAutomationPeer implementation that references a peer class. So make sure either to implement OnCreateAutomationPeer to use your own peer, or to use FrameworkElementAutomationPeer as the peer if that level of accessibility support is adequate for your control.

Note

You don't typically derive from AutomationPeer rather than FrameworkElementAutomationPeer. If you did derive directly from AutomationPeer you'll need to duplicate a lot of basic accessibility support that would otherwise come from FrameworkElementAutomationPeer.

Initialization of a custom peer class

The automation peer should define a type-safe constructor that uses an instance of the owner control for base initialization. In the next example, the implementation passes the owner value on to the RangeBaseAutomationPeer base, and ultimately it is the FrameworkElementAutomationPeer that actually uses owner to set FrameworkElementAutomationPeer.Owner.

public NumericUpDownAutomationPeer(NumericUpDown owner): base(owner)
{}
Public Sub New(owner As NumericUpDown)
    MyBase.New(owner)
End Sub
// NumericUpDownAutomationPeer.idl
import "NumericUpDown.idl";
namespace MyNamespace
{
    runtimeclass NumericUpDownAutomationPeer : Windows.UI.Xaml.Automation.Peers.AutomationPeer
    {
        NumericUpDownAutomationPeer(NumericUpDown owner);
        Int32 MyProperty;
    }
}

// NumericUpDownAutomationPeer.h
...
struct NumericUpDownAutomationPeer : NumericUpDownAutomationPeerT<NumericUpDownAutomationPeer>
{
    ...
    NumericUpDownAutomationPeer(MyNamespace::NumericUpDown const& owner);
};
//.h
public ref class NumericUpDownAutomationPeer sealed :  Windows::UI::Xaml::Automation::Peers::RangeBaseAutomationPeer
//.cpp
public:    NumericUpDownAutomationPeer(NumericUpDown^ owner);

Core methods of AutomationPeer

For UWP infrastructure reasons, the overridable methods of an automation peer are part of a pair of methods: the public access method that the UI Automation provider uses as a forwarding point for UI Automation clients, and the protected "Core" customization method that a UWP class can override to influence the behavior. The method pair is wired together by default in such a way that the call to the access method always invokes the parallel "Core" method that has the provider implementation, or as a fallback, invokes a default implementation from the base classes.

When implementing a peer for a custom control, override any of the "Core" methods from the base automation peer class where you want to expose behavior that is unique to your custom control. UI Automation code gets information about your control by calling public methods of the peer class. To provide information about your control, override each method with a name that ends with "Core" when your control implementation and design creates accessibility scenarios or other UI Automation scenarios that differ from what's supported by the base automation peer class.

At a minimum, whenever you define a new peer class, implement the GetClassNameCore method, as shown in the next example.

protected override string GetClassNameCore()
{
    return "NumericUpDown";
}

Note

You might want to store the strings as constants rather than directly in the method body, but that is up to you. For GetClassNameCore, you won't need to localize this string. The LocalizedControlType property is used any time a localized string is needed by a UI Automation client, not ClassName.

GetAutomationControlType

Some assistive technologies use the GetAutomationControlType value directly when reporting characteristics of the items in a UI Automation tree, as additional information beyond the UI Automation Name. If your control is significantly different from the control you are deriving from and you want to report a different control type from what is reported by the base peer class used by the control, you must implement a peer and override GetAutomationControlTypeCore in your peer implementation. This is particularly important if you derive from a generalized base class such as ItemsControl or ContentControl, where the base peer doesn't provide precise information about control type.

Your implementation of GetAutomationControlTypeCore describes your control by returning an AutomationControlType value. Although you can return AutomationControlType.Custom, you should return one of the more specific control types if it accurately describes your control's main scenarios. Here's an example.

protected override AutomationControlType GetAutomationControlTypeCore()
{
    return AutomationControlType.Spinner;
}

Note

Unless you specify AutomationControlType.Custom, you don't have to implement GetLocalizedControlTypeCore to provide a LocalizedControlType property value to clients. UI Automation common infrastructure provides translated strings for every possible AutomationControlType value other than AutomationControlType.Custom.

GetPattern and GetPatternCore

A peer's implementation of GetPatternCore returns the object that supports the pattern that is requested in the input parameter. Specifically, a UI Automation client calls a method that is forwarded to the provider's GetPattern method, and specifies a PatternInterface enumeration value that names the requested pattern. Your override of GetPatternCore should return the object that implements the specified pattern. That object is the peer itself, because the peer should implement the corresponding pattern interface any time that it reports that it supports a pattern. If your peer does not have a custom implementation of a pattern, but you know that the peer's base does implement the pattern, you can call the base type's implementation of GetPatternCore from your GetPatternCore. A peer's GetPatternCore should return null if a pattern is not supported by the peer. However, instead of returning null directly from your implementation, you would usually rely on the call to the base implementation to return null for any unsupported pattern.

When a pattern is supported, the GetPatternCore implementation can return this or Me. The expectation is that the UI Automation client will cast the GetPattern return value to the requested pattern interface whenever it is not null.

If a peer class inherits from another peer, and all necessary support and pattern reporting is already handled by the base class, implementing GetPatternCore isn't necessary. For example, if you are implementing a range control that derives from RangeBase, and your peer derives from RangeBaseAutomationPeer, that peer returns itself for PatternInterface.RangeValue and has working implementations of the IRangeValueProvider interface that supports the pattern.

Although it is not the literal code, this example approximates the implementation of GetPatternCore already present in RangeBaseAutomationPeer.

protected override object GetPatternCore(PatternInterface patternInterface)
{
    if (patternInterface == PatternInterface.RangeValue)
    {
        return this;
    }
    return base.GetPattern(patternInterface);
}

If you are implementing a peer where you don't have all the support you need from a base peer class, or you want to change or add to the set of base-inherited patterns that your peer can support, then you should override GetPatternCore to enable UI Automation clients to use the patterns.

For a list of the provider patterns that are available in the UWP implementation of UI Automation support, see Windows.UI.Xaml.Automation.Provider. Each such pattern has a corresponding value of the PatternInterface enumeration, which is how UI Automation clients request the pattern in a GetPattern call.

A peer can report that it supports more than one pattern. If so, the override should include return path logic for each supported PatternInterface value and return the peer in each matching case. It is expected that the caller will request only one interface at a time, and it is up to the caller to cast to the expected interface.

Here's an example of a GetPatternCore override for a custom peer. It reports the support for two patterns, IRangeValueProvider and IToggleProvider. The control here is a media display control that can display as full-screen (the toggle mode) and that has a progress bar within which users can select a position (the range control). This code came from the XAML accessibility sample.

protected override object GetPatternCore(PatternInterface patternInterface)
{
    if (patternInterface == PatternInterface.RangeValue)
    {
        return this;
    }
    else if (patternInterface == PatternInterface.Toggle)
    {
        return this;
    }
    return null;
}

Forwarding patterns from sub-elements

A GetPatternCore method implementation can also specify a sub-element or part as a pattern provider for its host. This example mimics how ItemsControl transfers scroll-pattern handling to the peer of its internal ScrollViewer control. To specify a sub-element for pattern handling, this code gets the sub-element object, creates a peer for the sub-element by using the FrameworkElementAutomationPeer.CreatePeerForElement method, and returns the new peer.

protected override object GetPatternCore(PatternInterface patternInterface)
{
    if (patternInterface == PatternInterface.Scroll)
    {
        ItemsControl owner = (ItemsControl) base.Owner;
        UIElement itemsHost = owner.ItemsHost;
        ScrollViewer element = null;
        while (itemsHost != owner)
        {
            itemsHost = VisualTreeHelper.GetParent(itemsHost) as UIElement;
            element = itemsHost as ScrollViewer;
            if (element != null)
            {
                break;
            }
        }
        if (element != null)
        {
            AutomationPeer peer = FrameworkElementAutomationPeer.CreatePeerForElement(element);
            if ((peer != null) && (peer is IScrollProvider))
            {
                return (IScrollProvider) peer;
            }
        }
    }
    return base.GetPatternCore(patternInterface);
}

Other Core methods

Your control may need to support keyboard equivalents for primary scenarios; for more info about why this might be necessary, see Keyboard accessibility. Implementing the key support is necessarily part of the control code and not the peer code because that is part of a control's logic, but your peer class should override the GetAcceleratorKeyCore and GetAccessKeyCore methods to report to UI Automation clients which keys are used. Consider that the strings that report key information might need to be localized, and should therefore come from resources, not hard-coded strings.

If you are providing a peer for a class that supports a collection, it's best to derive from both functional classes and peer classes that already have that kind of collection support. If you can't do so, peers for controls that maintain child collections may have to override the collection-related peer method GetChildrenCore to properly report the parent-child relationships to the UI Automation tree.

Implement the IsContentElementCore and IsControlElementCore methods to indicate whether your control contains data content or fulfills an interactive role in the user interface (or both). By default, both methods return true. These settings improve the usability of assistive technologies such as screen readers, which may use these methods to filter the automation tree. If your GetPatternCore method transfers pattern handling to a sub-element peer, the sub-element peer's IsControlElementCore method can return false to hide the sub-element peer from the automation tree.

Some controls may support labeling scenarios, where a text label part supplies information for a non-text part, or a control is intended to be in a known labeling relationship with another control in the UI. If it's possible to provide a useful class-based behavior, you can override GetLabeledByCore to provide this behavior.

GetBoundingRectangleCore and GetClickablePointCore are used mainly for automated testing scenarios. If you want to support automated testing for your control, you might want to override these methods. This might be desired for range-type controls, where you can't suggest just a single point because where the user clicks in coordinate space has a different effect on a range. For example, the default ScrollBar automation peer overrides GetClickablePointCore to return a "not a number" Point value.

GetLiveSettingCore influences the control default for the LiveSetting value for UI Automation. You might want to override this if you want your control to return a value other than AutomationLiveSetting.Off. For more info on what LiveSetting represents, see AutomationProperties.LiveSetting.

You might override GetOrientationCore if your control has a settable orientation property that can map to AutomationOrientation. The ScrollBarAutomationPeer and SliderAutomationPeer classes do this.

Base implementation in FrameworkElementAutomationPeer

The base implementation of FrameworkElementAutomationPeer provides some UI Automation information that can be interpreted from various layout and behavior properties that are defined at the framework level.

Note

Default UWP peers implement a behavior by using internal native code that implements the UWP, not necessarily by using actual UWP code. You won't be able to see the code or logic of the implementation through common language runtime (CLR) reflection or other techniques. You also won't see distinct reference pages for subclass-specific overrides of base peer behavior. For example, there might be additional behavior for GetNameCore of a TextBoxAutomationPeer, which won't be described on the AutomationPeer.GetNameCore reference page, and there is no reference page for TextBoxAutomationPeer.GetNameCore. There isn't even a TextBoxAutomationPeer.GetNameCore reference page. Instead, read the reference topic for the most immediate peer class, and look for implementation notes in the Remarks section.

Peers and AutomationProperties

Your automation peer should provide appropriate default values for your control's accessibility-related information. Note that any app code that uses the control can override some of that behavior by including AutomationProperties attached-property values on control instances. Callers can do this either for the default controls or for custom controls. For example, the following XAML creates a button that has two customized UI Automation properties: <Button AutomationProperties.Name="Special" AutomationProperties.HelpText="This is a special button."/>

For more info about AutomationProperties attached properties, see Basic accessibility information.

Some of the AutomationPeer methods exist because of the general contract of how UI Automation providers are expected to report information, but these methods are not typically implemented in control peers. This is because that info is expected to be provided by AutomationProperties values applied to the app code that uses the controls in a specific UI. For example, most apps would define the labeling relationship between two different controls in the UI by applying a AutomationProperties.LabeledBy value. However, LabeledByCore is implemented in certain peers that represent data or item relationships in a control, such as using a header part to label a data-field part, labeling items with their containers, or similar scenarios.

Implementing patterns

Let's look at how to write a peer for a control that implements an expand-collapse behavior by implementing the control pattern interface for expand-collapse. The peer should enable the accessibility for the expand-collapse behavior by returning itself whenever GetPattern is called with a value of PatternInterface.ExpandCollapse. The peer should then inherit the provider interface for that pattern (IExpandCollapseProvider) and provide implementations for each of the members of that provider interface. In this case the interface has three members to override: Expand, Collapse, ExpandCollapseState.

It's helpful to plan ahead for accessibility in the API design of the class itself. Whenever you have a behavior that is potentially requested either as a result of typical interactions with a user who is working in the UI or through an automation provider pattern, provide a single method that either the UI response or the automation pattern can call. For example, if your control has button parts that have wired event handlers that can expand or collapse the control, and has keyboard equivalents for those actions, have these event handlers call the same method that you call from within the body of the Expand or Collapse implementations for IExpandCollapseProvider in the peer. Using a common logic method can also be a useful way to make sure that your control's visual states are updated to show logical state in a uniform way, regardless of how the behavior was invoked.

A typical implementation is that the provider APIs first call Owner for access to the control instance at run time. Then the necessary behavior methods can be called on that object.

public class IndexCardAutomationPeer : FrameworkElementAutomationPeer, IExpandCollapseProvider {
    private IndexCard ownerIndexCard;
    public IndexCardAutomationPeer(IndexCard owner) : base(owner)
    {
         ownerIndexCard = owner;
    }
}

An alternate implementation is that the control itself can reference its peer. This is a common pattern if you are raising automation events from the control, because the RaiseAutomationEvent method is a peer method.

UI Automation events

UI Automation events fall into the following categories.

Event Description
Property change Fires when a property on a UI Automation element or control pattern changes. For example, if a client needs to monitor an app's check box control, it can register to listen for a property change event on the ToggleState property. When the check box control is checked or unchecked, the provider fires the event and the client can act as necessary.
Element action Fires when a change in the UI results from user or programmatic activity; for example, when a button is clicked or invoked through the Invoke pattern.
Structure change Fires when the structure of the UI Automation tree changes. The structure changes when new UI items become visible, hidden, or removed on the desktop.
Global change Fires when actions of global interest to the client occur, such as when the focus shifts from one element to another, or when a child window closes. Some events do not necessarily mean that the state of the UI has changed. For example, if the user tabs to a text-entry field and then clicks a button to update the field, a TextChanged event fires even if the user did not actually change the text. When processing an event, it may be necessary for a client application to check whether anything has actually changed before taking action.

AutomationEvents identifiers

UI Automation events are identified by AutomationEvents values. The values of the enumeration uniquely identify the kind of event.

Raising events

UI Automation clients can subscribe to automation events. In the automation peer model, peers for custom controls must report changes to control state that are relevant to accessibility by calling the RaiseAutomationEvent method. Similarly, when a key UI Automation property value changes, custom control peers should call the RaisePropertyChangedEvent method.

The next code example shows how to get the peer object from within the control definition code and call a method to fire an event from that peer. As an optimization, the code determines whether there are any listeners for this event type. Firing the event and creating the peer object only when there are listeners avoids unnecessary overhead and helps the control remain responsive.

if (AutomationPeer.ListenerExists(AutomationEvents.PropertyChanged))
{
    NumericUpDownAutomationPeer peer =
        FrameworkElementAutomationPeer.FromElement(nudCtrl) as NumericUpDownAutomationPeer;
    if (peer != null)
    {
        peer.RaisePropertyChangedEvent(
            RangeValuePatternIdentifiers.ValueProperty,
            (double)oldValue,
            (double)newValue);
    }
}

Peer navigation

After locating an automation peer, a UI Automation client can navigate the peer structure of an app by calling the peer object's GetChildren and GetParent methods. Navigation among UI elements within a control is supported by the peer's implementation of the GetChildrenCore method. The UI Automation system calls this method to build up a tree of sub-elements contained within a control; for example, list items in a list box. The default GetChildrenCore method in FrameworkElementAutomationPeer traverses the visual tree of elements to build the tree of automation peers. Custom controls can override this method to expose a different representation of child elements to automation clients, returning the automation peers of elements that convey information or allow user interaction.

Native automation support for text patterns

Some of the default UWP app automation peers provide control pattern support for the text pattern (PatternInterface.Text). But they provide this support through native methods, and the peers involved won't note the ITextProvider interface in the (managed) inheritance. Still, if a managed or non-managed UI Automation client queries the peer for patterns, it will report support for the text pattern, and provide behavior for parts of the pattern when client APIs are called.

If you intend to derive from one of the UWP app text controls and also create a custom peer that derives from one of the text-related peers, check the Remarks sections for the peer to learn more about any native-level support for patterns. You can access the native base behavior in your custom peer if you call the base implementation from your managed provider interface implementations, but it's difficult to modify what the base implementation does because the native interfaces on both the peer and its owner control aren't exposed. Generally you should either use the base implementations as-is (call base only) or completely replace the functionality with your own managed code and don't call the base implementation. The latter is an advanced scenario, you'll need good familiarity with the text services framework being used by your control in order to support the accessibility requirements when using that framework.

AutomationProperties.AccessibilityView

In addition to providing a custom peer, you can also adjust the tree view representation for any control instance, by setting AutomationProperties.AccessibilityView in XAML. This isn't implemented as part of a peer class, but we'll mention it here because it's germane to overall accessibility support either for custom controls or for templates you customize.

The main scenario for using AutomationProperties.AccessibilityView is to deliberately omit certain controls in a template from the UI Automation views, because they don't meaningfully contribute to the accessibility view of the entire control. To prevent this, set AutomationProperties.AccessibilityView to "Raw".

Throwing exceptions from automation peers

The APIs that you are implementing for your automation peer support are permitted to throw exceptions. It's expected any UI Automation clients that are listening are robust enough to continue on after most exceptions are thrown. In all likelihood that listener is looking at an all-up automation tree that includes apps other than your own, and it's an unacceptable client design to bring down the entire client just because one area of the tree threw a peer-based exception when the client called its APIs.

For parameters that are passed in to your peer, it's acceptable to validate the input, and for example throw ArgumentNullException if it was passed null and that's not a valid value for your implementation. However, if there are subsequent operations performed by your peer, remember that the peer's interactions with the hosting control have something of an asynchronous character to them. Anything a peer does won't necessarily block the UI thread in the control (and it probably shouldn't). So you could have situations where an object was available or had certain properties when the peer was created or when an automation peer method was first called, but in the meantime the control state has changed. For these cases, there are two dedicated exceptions that a provider can throw:

  • Throw ElementNotAvailableException if you're unable to access either the peer's owner or a related peer element based on the original info your API was passed. For example, you might have a peer that's trying to run its methods but the owner has since been removed from the UI, such as a modal dialog that's been closed. For a non-.NET client, this maps to UIA_E_ELEMENTNOTAVAILABLE.
  • Throw ElementNotEnabledException if there still is an owner, but that owner is in a mode such as IsEnabled=false that's blocking some of the specific programmatic changes that your peer is trying to accomplish. For a non-.NET client, this maps to UIA_E_ELEMENTNOTENABLED.

Beyond this, peers should be relatively conservative regarding exceptions that they throw from their peer support. Most clients won't be able to handle exceptions from peers and turn these into actionable choices that their users can make when interacting with the client. So sometimes a no-op, and catching exceptions without rethrowing within your peer implementations, is a better strategy than is throwing exceptions every time something the peer tries to do doesn't work. Consider also that most UI Automation clients aren't written in managed code. Most are written in COM and are just checking for S_OK in an HRESULT whenever they call a UI Automation client method that ends up accessing your peer.