WinRT demystified

by Miguel de Icaza

Windows 8 as introduced at Build is an exciting release as it has important updates to how Microsoft envisions users will interact with their computers, to a fresh new user interface to a new programming model and a lot more.

If you build software for end-users, you should watch Jensen Harris discuss the Metro principles in Windows 8. I find myself wanting to spend time using Windows 8.

But the purpose of this post is to share what I learned at the conference specifically about WinRT and .NET.

The Basics

Microsoft is using the launch of Windows 8 as an opportunity to fix long-standing problems with Windows, bring a new user interface, and enable a safe AppStore model for Windows.

To do this, they have created a third implementation of the XAML-based UI system. Unlike WPF which was exposed only to the .NET world and Silverlight which was only exposed to the browser, this new implementation is available to C++ developers, HTML/Javascript developers and also .NET developers.

.NET developers are very familiar with P/Invoke and COM Interop. Those are two technologies that allow a .NET developer to consume an external component, for example, this is how you would use the libc "system (const char *)" API from C#:

	[DllImport ("libc")]
	void system (string command);
	[...]

	system ("ls -l /");
	

We have used P/Invoke extensively in the Mono world to create bindings to native libraries. Gtk# binds the Gtk+ API, MonoMac binds the Cocoa API, Qyoto binds the Qt API and hundred other bindings wrap other libraries that are exposed to C# as object-oriented libraries.

COM Interop allows using C or C++ APIs directly from C# by importing the COM type libraries and having the runtime provide the necessary glue. This is how Mono talked with OpenOffice (which is based on COM), or how Mono talks to VirtualBox (which has an XPCOM based API).

There are many ways of creating bindings for a native library, but doing it by hand is bound to be both tedious and error prone. So everyone has adopted some form of "contract" that states what the API is, and the binding author uses this contract to create their language binding.

WinRT

WinRT is a new set of APIs that have the following properties:

  • It implements the new Metro look.
  • Has a simple UI programming model for Windows developers (You do not need to learn Win32, what an HDC, WndProc or LPARAM is).
  • It exposes the WPF/Silverlight XAML UI model to developers.
  • The APIs are all designed to be asynchronous.
  • It is a sandboxed API, designed for creating self-contained, AppStore-ready applications. You wont get everything you want to create for example Backup Software or Hard Disk Partitioning software.
  • The API definitions is exposed in the ECMA 335 metadata format (the same one that .NET uses, you can find those as ".winmd" files).

WinRT wraps both the new UI system as well as old Win32 APIs and it happens that this implementation is based on top of COM.

WinRT Projections

What we call "bindings" Microsoft now calls "projections". Projections are the process of exposing APIs to three environments: Native (C and C++), HTML/Javascript and .NET.

  • If you author a component in C++ or a .NET language, its API will be stored in a WinMD file and you will be able to consume it from all three environments (Native, JavaScript and .NET).

    Even in C++ you are not exposed to COM. The use of COM is hidden behind the C++ projection tools. You use what looks and feels like a C++ object oriented API.

    To support the various constructs of WinRT, the underlying platform defines a basic set of types and their mappings to various environment. In particular, collection objects in WinRT are mapped to constructs that are native to each environment.

    Asynchronous APIs

    Microsoft feels that when a developer is given the choice of a synchronous and an asynchronous API, developers will choose the simplicity of a synchronous API. The result usually works fine on the developer system, but is terrible when used in the wild.

    With WinRT, Microsoft has followed a simple rule: if an API is expected to take more than 50 milliseconds to run, the API is asynchronous.

    The idea of course is to ensure that every Metro application is designed to always respond to user input and to not hang, block or provide a poor user experience.

    Async programming has historically been a cumbersome process as callbacks and state must be cascaded over dozens of places and error handling (usually poor error handling) is sprinkled across multiple layers of code.

    To simplify this process, C# and VB have been extended to support the F#-inspired await/async pattern, turning async programming into a joy. C++ got a setup that is as good as you can get with C++ lambdas and Javascript uses promises and "then ()".

    Is it .NET or Not?

    Some developers are confused as to whether .NET is there or not in the first place, as not all of the .NET APIs are present (File I/O, Sockets), many were moved and others were introduced to integrate with WinRT.

    When you use C# and VB, you are using the full .NET framework. But they have chosen to expose a smaller subset of the API to developers to push the new vision for Windows 8.

    And this new vision includes safety/sandboxed systems and asynchronous programming. This is why you do not get direct file system access or socket access and why synchronous APIs that you were used to consuming are not exposed.

    Now, you notice that I said "exposed" and not "gone".

    What they did was that they only exposed to the compiler a set of APIs when you target the Metro profile. So your application will not accidentally call File.Create for example. At runtime though, the CLR will load the full class library, the very one that contains File.Create, so internally, the CLR could call something like File.Create, it is just you that will have no access to it.

    This split is similar to what has been done in the past with Silverlight, where not every API was exposed, and where mscorlib was given rights that your application did not have to ensure the system safety.

    You might be thinking that you can use some trick (referencing the GAC library instead of the compiler reference or using reflection to get to private APIs, or P/Invoking into Win32). But all of those uses will be caught by AppStore review application and you wont be able to publish your app through Microsoft's store.

    You can still do whatever ugly hack you please on your system. It just wont be possible to publish that through the AppStore.

    Finally, the .NET team has taken this opportunity to do some spring cleaning. mscorlib.dll and System.dll have been split in various libraries and they have moved some types around.

    Creating WinRT Components

    Microsoft demoed creating new WinRT components on both C++ and .NET.

    In the .NET case, creating a WinRT component has been drastically simplified. The following is the full source code for a component that adds 2:

    
    	public sealed class AddTwo {
    		public int Add (int a, int b)
    		{
    			return a + b;
    		}
    
    		public async IAsyncOperation SubAsync (int a, int b)
    		{
    			return a - await (CountEveryBitByHand (b));
    		}
    	}
    	

    You will notice that there are no COM declarations of any kind. The only restriction is that your class must be sealed (unless you are creating a XAML UI component, in that case the restriction is lifted).

    There are also some limitations, you can not have private fields on structures, and there is not Task<T> for asynchronous APIs, instead you use the IAsyncOperation interface. Update to clarify: the no private fields rule is only limited to structs exposed to WinRT, and it does not apply to classes.

    UI Programming

    When it comes to your UI selection, you can either use HTML with CSS to style your app or you can use XAML UI.

    To make it easy for HTML apps to adhere to the Metro UI style and interaction model, Microsoft distributes Javascript and CSS files that you can consume from your project. Notice that this wont work on the public web. As soon as you use any WinRT APIs, your application is a Windows app, and wont run in a standalone web browser.

    .NET and C++ developers get to use XAML instead.

    There is clearly a gap to be filled in the story. It should be possible to use Microsoft's Razor formatting engine to style applications using HTML/CSS while using C#. Specially since they have shown the CLR running on their HTML/JS Metro engine.

    Right now HTML and CSS is limited to the Javascript use.

    In Short

    Microsoft has created a cool new UI library called WinRT and they have made it easy to consume from .NET, Javascript and C++ and if you adhere by their guidelines, they will publish the app on their appstore.

    Xamarin at BUILD

    If you are at build, come join us tonight at 6:30 at the Sheraton Park hotel, just after Meet the Experts. Come talk about Mono, Xamarin, MonoTouch, MonoDroid and MonoMac and discuss the finer points of this blog over an open bar.

    Comments

    There is a long list of comments in the moderation queue that are not directly related to WinRT, or bigger questions that are not directly related to WinRT, .NET and this post's topic, so I wont be approving those comments to keep things on focus. There are better forums to have discussions on Metro.

  • Posted on 15 Sep 2011