Xamarin.Android API Design Principles

In addition to the core Base Class Libraries that are part of Mono, Xamarin.Android ships with bindings for various Android APIs to allow developers to create native Android applications with Mono.

At the core of Xamarin.Android there is an interop engine that bridges the C# world with the Java world and provides developers with access to the Java APIs from C# or other .NET languages.

Design Principles

These are some of our design principles for the Xamarin.Android binding

  • Conform to the .NET Framework Design Guidelines.

  • Allow developers to subclass Java classes.

  • Subclass should work with C# standard constructs.

  • Derive from an existing class.

  • Call base constructor to chain.

  • Overriding methods should be done with C#'s override system.

  • Make common Java tasks easy, and hard Java tasks possible.

  • Expose JavaBean properties as C# properties.

  • Expose a strongly typed API:

    • Increase type-safety.

    • Minimize runtime errors.

    • Get IDE intellisense on return types.

    • Allows for IDE popup documentation.

  • Encourage in-IDE exploration of the APIs:

    • Utilize Framework Alternatives to Minimize Java Classlib exposure.

    • Expose C# delegates (lambdas, anonymous methods and System.Delegate) instead of single-method interfaces when appropriate and applicable.

    • Provide a mechanism to call arbitrary Java libraries ( Android.Runtime.JNIEnv).

Assemblies

Xamarin.Android includes a number of assemblies that constitute the MonoMobile Profile. The Assemblies page has more information.

The bindings to the Android platform are contained in the Mono.Android.dll assembly. This assembly contains the entire binding for consuming Android APIs and communicating with the Android runtime VM.

Binding Design

Collections

The Android APIs utilize the java.util collections extensively to provide lists, sets, and maps. We expose these elements using the System.Collections.Generic interfaces in our binding. The fundamental mappings are:

We have provided helper classes to facilitate faster copyless marshaling of these types. When possible, we recommend using these provided collections instead of the framework provided implementation, like List<T> or Dictionary<TKey, TValue>. The Android.Runtime implementations utilize a native Java collection internally and therefore do not require copying to and from a native collection when passing to an Android API member.

You can pass any interface implementation to an Android method accepting that interface, e.g. pass a List<int> to the ArrayAdapter<int>(Context, int, IList<int>) constructor. However, for all implementations except for the Android.Runtime implementations, this involves copying the list from the Mono VM into the Android runtime VM. If the list is later changed within the Android runtime (e.g. by invoking the ArrayAdapter<T>.Add(T) method), those changes will not be visible in managed code. If a JavaList<int> were used, those changes would be visible.

Rephrased, collections interface implementations that are not one of the above listed Helper Classes only marshal [In]:

// This fails:
var badSource  = new List<int> { 1, 2, 3 };
var badAdapter = new ArrayAdapter<int>(context, textViewResourceId, badSource);
badAdapter.Add (4);
if (badSource.Count != 4) // true
    throw new InvalidOperationException ("this is thrown");

// this works:
var goodSource  = new JavaList<int> { 1, 2, 3 };
var goodAdapter = new ArrayAdapter<int> (context, textViewResourceId, goodSource);
goodAdapter.Add (4);
if (goodSource.Count != 4) // false
    throw new InvalidOperationException ("should not be reached.");

Properties

Java methods are transformed into properties, when appropriate:

  • The Java method pair T getFoo() and void setFoo(T) are transformed into the Foo property. Example: Activity.Intent.

  • The Java method getFoo() is transformed into the read-only Foo property. Example: Context.PackageName.

  • Set-only properties are not generated.

  • Properties are not generated if the property type would be an array.

Events and Listeners

The Android APIs are built on top of Java and its components follow the Java pattern for hooking up event listeners. This pattern tends to be cumbersome as it requires the user to create an anonymous class and declare the methods to override, for example, this is how things would be done in Android with Java:

final android.widget.Button button = new android.widget.Button(context);

button.setText(this.count + " clicks!");
button.setOnClickListener (new View.OnClickListener() {
    public void onClick (View v) {
        button.setText(++this.count + " clicks!");
    }
});

The equivalent code in C# using events would be:

var button = new Android.Widget.Button (context) {
    Text = string.Format ("{0} clicks!", this.count),
};
button.Click += (sender, e) => {
    button.Text = string.Format ("{0} clicks!", ++this.count);
};

Note that both of the above mechanisms are available with Xamarin.Android. You can implement a listener interface and attach it with View.SetOnClickListener, or you can attach a delegate created via any of the usual C# paradigms to the Click event.

When the listener callback method has a void return, we create API elements based on an EventHandler<TEventArgs> delegate. We generate an event like the above example for these listener types. However, if the listener callback returns a non-void and non- boolean value, events and EventHandlers are not used. We instead generate a specific delegate for the signature of the callback and add properties instead of events. The reason is to deal with delegate invocation order and return handling. This approach mirrors what is done with the Xamarin.iOS API.

C# events or properties are only automatically generated if the Android event-registration method:

  1. Has a set prefix, e.g. setOnClickListener.

  2. Has a void return type.

  3. Accepts only one parameter, the parameter type is an interface, the interface has only one method, and the interface name ends in Listener , e.g. View.OnClick Listener.

Furthermore, if the Listener interface method has a return type of boolean instead of void, then the generated EventArgs subclass will contain a Handled property. The value of the Handled property is used as the return value for the Listener method, and it defaults to true.

For example, the Android View.setOnKeyListener() method accepts the View.OnKeyListener interface, and the View.OnKeyListener.onKey(View, int, KeyEvent) method has a boolean return type. Xamarin.Android generates a corresponding View.KeyPress event, which is an EventHandler<View.KeyEventArgs>. The KeyEventArgs class in turn has a View.KeyEventArgs.Handled property, which is used as the return value for the View.OnKeyListener.onKey() method.

We intend to add overloads for other methods and ctors to expose the delegate-based connection. Also, listeners with multiple callbacks require some additional inspection to determine if implementing individual callbacks is reasonable, so we are converting these as they are identified. If there is no corresponding event, listeners must be used in C#, but please bring any that you think could have delegate usage to our attention. We have also done some conversions of interfaces without the "Listener" suffix when it was clear they would benefit from a delegate alternative.

All of the listeners interfaces implement the Android.Runtime.IJavaObject interface, because of the implementation details of the binding, so listener classes must implement this interface. This can be done by implementing the listener interface on a subclass of Java.Lang.Object or any other wrapped Java object, such as an Android activity.

Runnables

Java utilizes the java.lang.Runnable interface to provide a delegation mechanism. The java.lang.Thread class is a notable consumer of this interface. Android has employed the interface in the API as well. Activity.runOnUiThread() and View.post() are notable examples.

The Runnable interface contains a single void method, run(). It therefore lends itself to binding in C# as a System.Action delegate. We have provided overloads in the binding which accept an Action parameter for all API members which consume a Runnable in the native API, e.g. Activity.RunOnUiThread() and View.Post().

We left the IRunnable overloads in place instead of replacing them since several types implement the interface and can therefore be passed as runnables directly.

Inner Classes

Java has two different types of nested classes: static nested classes and non-static classes.

Java static nested classes are identical to C# nested types.

Non-static nested classes, also called inner classes, are significantly different. They contain an implicit reference to an instance of their enclosing type and cannot contain static members (among other differences outside the scope of this overview).

When it comes to binding and C# use, static nested classes are treated as normal nested types. Inner classes, meanwhile, have two significant differences:

  1. The implicit reference to the containing type must be provided explicitly as a constructor parameter.

  2. When inheriting from an inner class, the inner class must be nested within a type that inherits from the containing type of the base inner class, and the derived type must provide a constructor of the same type as the C# containing type.

For example, consider the Android.Service.Wallpaper.WallpaperService.Engine inner class. Since it's an inner class, the WallpaperService.Engine() constructor takes a reference to a WallpaperService instance (compare and contrast to the Java WallpaperService.Engine() constructor, which takes no parameters).

An example derivation of an inner class is CubeWallpaper.CubeEngine:

class CubeWallpaper : WallpaperService {
    public override WallpaperService.Engine OnCreateEngine ()
    {
        return new CubeEngine (this);
    }

    class CubeEngine : WallpaperService.Engine {
        public CubeEngine (CubeWallpaper s)
                : base (s)
        {
        }
    }
}

Note how CubeWallpaper.CubeEngine is nested within CubeWallpaper, CubeWallpaper inherits from the containing class of WallpaperService.Engine, and CubeWallpaper.CubeEngine has a constructor which takes the declaring type -- CubeWallpaper in this case -- all as specified above.

Interfaces

Java interfaces can contain three sets of members, two of which cause problems from C#:

  1. Methods

  2. Types

  3. Fields

Java interfaces are translated into two types:

  1. An (optional) interface containing method declarations. This interface has the same name as the Java interface, except it also has an ' I ' prefix.

  2. An (optional) static class containing any fields declared within the Java interface.

Nested types are "relocated" to be siblings of the enclosing interface instead of nested types, with the enclosing interface name as a prefix.

For example, consider the android.os.Parcelable interface. The Parcelable interface contains methods, nested types, and constants. The Parcelable interface methods are placed into the Android.OS.IParcelable interface. The Parcelable interface constants are placed into the Android.OS.ParcelableConsts type. The nested android.os.Parcelable.ClassLoaderCreator<T> and android.os.Parcelable.Creator<T> types are currently not bound due to limitations in our generics support; if they were supported, they would be present as the Android.OS.IParcelableClassLoaderCreator and Android.OS.IParcelableCreator interfaces. For example, the nested android.os.IBinder.DeathRecipient interface is bound as the Android.OS.IBinderDeathRecipient interface.

Note

Beginning with Xamarin.Android 1.9, Java interface constants are duplicated in an effort to simplify porting Java code. This helps to improve porting Java code that relies on android provider interface constants.

In addition to the above types, there are four further changes:

  1. A type with the same name as the Java interface is generated to contain constants.

  2. Types containing interface constants also contain all constants that come from implemented Java interfaces.

  3. All classes that implement a Java interface containing constants get a new nested InterfaceConsts type which contains constants from all implemented interfaces.

  4. The Consts type is now obsolete.

For the android.os.Parcelable interface, this means that there will now be an Android.OS.Parcelable type to contain the constants. For example, the Parcelable.CONTENTS_FILE_DESCRIPTOR constant will be bound as the Parcelable.ContentsFileDescriptor constant, instead of as the ParcelableConsts.ContentsFileDescriptor constant.

For interfaces containing constants which implement other interfaces containing yet more constants, the union of all constants is now generated. For example, the android.provider.MediaStore.Video.VideoColumns interface implements the android.provider.MediaStore.MediaColumns interface. However, prior to 1.9, the Android.Provider.MediaStore.Video.VideoColumnsConsts type has no way of accessing the constants declared on Android.Provider.MediaStore.MediaColumnsConsts. As a result, the Java expression MediaStore.Video.VideoColumns.TITLE needs to be bound to the C# expression MediaStore.Video.MediaColumnsConsts.Title which is hard to discover without reading lots of Java documentation. In 1.9, the equivalent C# expression will be MediaStore.Video.VideoColumns.Title.

Furthermore, consider the android.os.Bundle type, which implements the Java Parcelable interface. Since it implements the interface, all constants on that interface are accessible "through" the Bundle type, e.g. Bundle.CONTENTS_FILE_DESCRIPTOR is a perfectly valid Java expression. Previously, to port this expression to C# you would need to look at all the interfaces which are implemented to see from which type the CONTENTS_FILE_DESCRIPTOR came from. Starting in Xamarin.Android 1.9, classes implementing Java interfaces which contain constants will have a nested InterfaceConsts type, which will contain all the inherited interface constants. This will allow translating Bundle.CONTENTS_FILE_DESCRIPTOR to Bundle.InterfaceConsts.ContentsFileDescriptor.

Finally, types with a Consts suffix such as Android.OS.ParcelableConsts are now Obsolete, other than the newly introduced InterfaceConsts nested types. They will be removed in Xamarin.Android 3.0.

Resources

Images, layout descriptions, binary blobs and string dictionaries can be included in your application as resource files. Various Android APIs are designed to operate on the resource IDs instead of dealing with images, strings or binary blobs directly.

For example, a sample Android app that contains a user interface layout ( main.axml), an internationalization table string ( strings.xml) and some icons ( drawable-*/icon.png) would keep its resources in the "Resources" directory of the application:

Resources/
    drawable-hdpi/
        icon.png

    drawable-ldpi/
        icon.png

    drawable-mdpi/
        icon.png

    layout/
        main.axml

    values/
        strings.xml

The native Android APIs do not operate directly with filenames, but instead operate on resource IDs. When you compile an Android application that uses resources, the build system will package the resources for distribution and generate a class called Resource that contains the tokens for each one of the resources included. For example, for the above Resources layout, this is what the R class would expose:

public class Resource {
    public class Drawable {
        public const int icon = 0x123;
    }

    public class Layout {
        public const int main = 0x456;
    }

    public class String {
        public const int first_string = 0xabc;
        public const int second_string = 0xbcd;
    }
}

You would then use Resource.Drawable.icon to reference the drawable/icon.png file, or Resource.Layout.main to reference the layout/main.xml file, or Resource.String.first_string to reference the first string in the dictionary file values/strings.xml.

Constants and Enumerations

The native Android APIs have many methods that take or return an int that must be mapped to a constant field to determine what the int means. To use these methods, the user is required to consult the documentation to see which constants are appropriate values, which is less than ideal.

For example, consider Activity.requestWindowFeature(int featureID).

In these cases, we endeavor to group related constants together into a .NET enumeration, and remap the method to take the enumeration instead. By doing this, we are able to offer IntelliSense selection of the potential values.

The above example becomes: Activity.RequestWindowFeature(WindowFeatures featureId).

Note that this is a very manual process to figure out which constants belong together, and which APIs consume these constants. Please file bugs for any constants used in the API that would be better expressed as an enumeration.