Microsoft's high-risk Windows 8 .NET switch

Third parties not welcome

Only a small, safe subset of the Win32 API is available to Metro-style apps. The runtime throws an error if you try to call a disallowed API. You can use your own native code DLLs and call them from .NET using platform invocation (P/Invoke), but if that native code DLL calls a disallowed API again an error will be thrown.

This is a locked-down platform then, with hints that it will be especially locked-down in its ARM incarnation. One tap gets you back to desktop Windows with all its chaotic freedom, but Microsoft is determined to make the Metro-style side a safe platform and on ARM it is possible that even the desktop side will be restricted.

The WinRT framework is not really extensible, other than with code that you install into your app's own directory. There is provision for shared framework extensions, which can be downloaded on demand from the forthcoming Windows Store, but Microsoft says that third parties will not be able to create these shared extensions, other than in corporate roll-outs.

Within those deliberate limitations, Microsoft has built what looks like a programmer-friendly, high-performance platform in WinRT. There is no message loop and no Graphics Device Interface (GDI), to mention just two fundamentals of the old Win32 API. The user interface (UI) is always rendered by DirectX.

Another notable feature is the number of asynchronous APIs. Microsoft claims to have made every call that may take more than 50ms asynchronous, and Redmond has not provided synchronous alternatives. "Developers call them even when they shouldn't," we were told.

The pain of asynchronous programming has been largely removed by new language features, such as the "await" keyword in C#, which enables developers to code in a synchronous manner even though callbacks are used behind the scenes. The UI is on a single thread.

What then are the implications for .NET on the client? It was obvious at BUILD that it is C# developers that dominate on Microsoft's platform – sessions on the language were packed to overflowing – and it is likely that more WinRT code will be written in C# than in any other language. It is also obvious that Metro-style apps are not capable of everything Windows developers want to code, so desktop development will remain strong and largely .NET-based.

Looking into the future as Microsoft envisages it, though, desktop Windows seems unlikely to evolve much beyond what we already have in Windows 7, which is set to become the new Windows XP in terms of its life on corporate desktops.

Silverlight 5.0 is due to be released soon, but we heard little about it at BUILD and it is no longer strategic for Microsoft. Microsoft's energy is all going into WinRT, into trying to catch up with Apple on tablets and into promoting cloud services for consumption by Metro-style apps.

Porting most existing .NET applications to WinRT is not easy. The UI will need to be reworked, the code ported to allow for all the parts of the framework that are no longer available, and the architecture revised to work with the WinRT cloud services model.

Fear and programming

In one sense then, the fears of .NET developers were justified and COM has its revenge. In mitigation, C# and Visual-Basic programming is alive and well, XAML remains, and WinRT works seamlessly with the CLR. The mood at BUILD was generally upbeat, as developers have enjoyed the responsive and touch-friendly Metro-style Windows – even though when you talk to them many admit that their apps will not work in Metro and they are not rushing to support it with apps. One factor is that corporate developers dominate at BUILD, but Metro has so far been perceived as more of a consumer than a professional platform.

Windows 8 has made a good impression at BUILD, but it is a high-risk strategy for Microsoft. ®