Want to allow different applications to communicate over the Internet without requiring any service infrastructure of your own and without needing to learn and understand the complexities of WCF? There is now a solution that you might want to take a look at.

I've been working with a company called Linxter helping them build out a cloud messaging infrastructure that makes building cloud-connected applications even easier than building them using WCF. Basically they have set up the infrastructure and provided an easy to use SDK that encapsulates all the scary code required to really do reliable messaging over the internet, while supporting occasionally connected scenarios, security, reliability, and other aspects.

Basically all you do is download the SDK, add a reference to a single DLL, run a tool to generate a supporting database (file based, SQLite), and start making calls.

The API is really simple, you just call CreateMessage to send a message, subscribe to an event to know when you have received a message, and there is a one time step to register your program instance with the back end Linxter Internet Service Bus (ISB) infrastructure.

So a simple Hello World app could look something like this:

// The API is separated into three interfaces focused on 
// the three aspects of using the SDK 
ILinxterMessaging messaging = new LinxterSDK();
ILinxterRegistration registration = messaging as ILinxterRegistration;
ILinxterCommunicationChannel commChannels = messaging as ILinxterCommunicationChannel;

// Subscribe so received messages raise an event
messaging.MessageReceived += OnMessageReceived;

// One time step in the life of a program instance - register
registration.RegisterProgramInstance();

// Tell the SDK to start retrieving messages based on a configurable schedule
// On demand also supported
messaging.StartScheduledReceive();

//  Create a channel to another instance based on an ID or alias name
ProgramInstanceInfo responder = 
    commChannels.FindProgramInstances("HelloWorld Responder", 
    FindProgramInstanceType.ByAlias)[0];

commChannels.CreateRequest(responder, "Invitation message");
// Send on demand or on schedule supported
messaging.SendNow();

// Wait for channel to be opened to the receiver
while (!IsOpenCommunicaitonChannels(commChannels, responder))
{
    Console.WriteLine("Waiting for comm channel creation");
    Thread.Sleep(1000);
}

// Construct a message to send to the receiver, can be many
List<Guid> receivers = new List<Guid>();
receivers.Add(responder.ProgramInstanceId);
// Each message can have an activity ID to separate different message types
string activityId = "b9501157-be9a-4a98-aab6-1c1e17ac8d99";

// The message body is simply a string 
// (in which you can encode whatever data you want, however you want)
Guid msgID = messaging.CreateMessage(receivers, activityId, "Hello World");

// Messages are always queued in the client DB supporting offline scenarios well
messaging.SendNow();

Then of course you need the event handler that receives the messages:

static void OnMessageReceived(MessageReceivedEventArgs e)
{
    Console.WriteLine("\nNOTICE: Message Received: {0}", e.Body);
}

There is a lot more to it of course, but the point is that you can integrate Linxter into just abobut any kind of .NET program including smart client applications, web applications, and Windows services. You don't have to learn anything about how to get the distributed messaging done, they have done all the work for you.

Eventually there will be support for using Linxter in other platforms, so it can form the glue to integrate applications from different platforms.

Of course you could do these things yourself with WCF and .NET services, but to do it that way you would have to master a much more demanding set of skills and may have to set up some infrastructure of your own. With Linxter, you just need a reference to a library, a few method calls into that library, and an account and your applications can be connected really quickly.

If you want to learn more, you can visit the Linxter Developer site at http://linxterdeveloper.com. You can download the SDK and QuickStarts there as well.

Linxter should be releasing version 1.0 by early 2009, so that is another advantage for the near term compared to building it all yourself with WCF and .NET Servicces.

Check it out and you can give me any feedback, which I can pass along to the development team.

Friday, December 19, 2008 1:26:31 AM (GMT Standard Time, UTC+00:00)
Comments [1]  | 

I gave three talks at Visual Studio Connections in Las Vegas this week:

• Selecting the Right Client Technology
• This session summarized the capabilities, pros, cons, and influencing factors for deciding between smart client or web, WPF, Windows Forms, ASP.NET, Silverlight and Mobile. It also summarized the client software factories and guidance from Microsoft p&p at the end.
• Leverage Routed Events and Commands in WPF
• This session covered how routed events and commands work in WPF, how to hook them up, how to handle them, how to implement custom commands, and best practices for employing them.
• Developing Service Oriented Workflows
• This session covered building workflow services with WF and WCF. It covered the Send and Receive activities, context bindings, and Workflow Service Host. I showed using workflows as a business process orchestration that calls other services, multiple clients calling into the same workflow, implementing custom send activities to overcome the limitations of the built-in Send Activity, and calling between workflow instances.

You can get the slides and demos for all these sessions here.

Thursday, November 13, 2008 7:27:38 PM (GMT Standard Time, UTC+00:00)
Comments [1]  | 

In my article on Routed Events and Commands in the September MSDN Magazine, I made a quick reference to a very subtle, confusing, and misunderstood aspect of how routed commands are actually routed. It was only a brief reference because it was only made clear to me through some excellent tech review feedback from the Product Team (thanks Dwayne and Varsha!) a couple days before going to print, so I couldn't add a big explanation without trashing the print flow for the magazine.

I had a question from a reader who is a well known WPF expert himself, so I figured I'd better expand my explanation.

The short simple answer if you don't want to understand all the complexities of what is going on under the covers is this: Always put your command bindings on the root element (Window or Page) and your command binding will always be consulted. You can do this explicitly in the XAML, or you can inject it there through CommandManager.RegisterClassCommandBinding.

But if you want to deviate from that guidance, you really have to understand the following.

With WPF routed commands, routed events are used under the covers to dispatch calls to the command handlers both to determine when the command should be enabled, and to invoke the handling for that command when the command is invoked. For example, if I have a Save button in a toolbar, and I have a command binding at my Window root level, a routed event will be used to ask the CanExecute method of that binding if the button should be enabled. And then when the user clicks it, a routed event will be used to call the Executed method of the command binding.

Where things get confusing is if you try to place that command binding somewhere deeper in the visual tree because you have to have a good understanding of where the routed events are started and where all they flow to.

There are a lot of people that believe (myself included until Dwayne and Varsha from the product team showed me the light) that the routed events for commands flow from the focused element to the root of the visual tree with tunneling and bubbling events. That is really only a part of the picture.

The fact is that the routed events for a routed command ALWAYS are initially raised on the command invoker or source control, not the focused element. Any control that implements the ICommandSource interface can be a command source. This includes buttons (and checkboxes... anything deriving from ButtonBase), menu items, hyperlinks, and input bindings (i.e. keybinding) out of the box in WPF. These events follow the standard pattern for routed events in WPF - a tunneling event is routed down the visual tree from the root  to the target and then a bubbling event is routed up the visual tree from the target back to the root.

The focused element in the UI only gets the events routed to it if the command source is contained in a focus scope (the Toolbar). When that is the case, as the bubbling event reaches the focus scope boundary, the event is re-raised on the focused element.

So what is a focus scope? A focus scope is any container element that sets the FocusManager.IsFocusScope attached property to true. Toolbars and menus have this property set, so those are the main focus scopes you will deal with other than a Window or Page which is always the root focus scope, but you could easily create your own by setting that property to true on a container control.

So in my Save button example above, the flow looks like this when the button is clicked:

• CommandManager.PreviewExecuted tunneling event raised on the Save button in the toolbar. This means it starts at the root element of the visual tree and tunnels down to the Save button through whatever element tree is in between.
• CommandManager.Executed bubbling event raised on the Save button in the toolbar. This means the bubbling event flows from the button up to the root element of the Toolbar.
• CommandManager.PreviewExecuted tunneling event raised on the focused element in the root UI.
• CommandManager.Executed bubbling event raised on the focused element in the root UI.

To make it even more concrete, I wrote a little form whose element tree looks like this with several sources of the Cut command:

<Window x:Class="CommandFlow.Window1"
    xmlns="http://schemas.microsoft.com/winfx/2006/xaml/presentation"
    xmlns:x="http://schemas.microsoft.com/winfx/2006/xaml"
    Title="Window1" Height="300" Width="300">
    <DockPanel>
        <ToolBar DockPanel.Dock="Top" Height="30">
            <Button Width="75" Height="25" Command="Cut">Cut</Button>
        </ToolBar>
        <StackPanel>
            <Button Width="75" Height="25" Command="Cut">Cut</Button>
            <TextBox Width="150" Height="25"/>
        </StackPanel>
    </DockPanel>
</Window>

I then added attached event handlers for the PreviewExecuted and Executed events to each element in the tree and did a Debug.WriteLine so it would be very clear where the Executed events were flowing. I then ran it, put the focus in the Textbox, and clicked the button in the toolbar. The result is below.

Window Preview  Executed Called
DockPanel Preview  Executed Called
Toolbar Preview  Executed Called
Toolbar button Preview  Executed Called
Toolbar button  Executed Called
Window Preview  Executed Called
DockPanel Preview  Executed Called
StackPanel Preview  Executed Called
Textbox Preview  Executed Called

As you can see, a tunneling (Preview) event works its way down from the root element to the Toolbar button that is the command source, followed by a bubbling event back up. The important thing to notice here though, which is where the confusion about how  the focused element is involved, is that the bubbling event seems to have stopped after firing on the toolbar button, and then a new tunneling event is raised which tunnels down to the Textbox and ends.

What is happening is that the Toolbar is actually handling the bubbling Executed event (Handled = true), which stops the bubbling process. The CommandManager then re-raises the event on the focused element (Textbox) and a Preview event tunnels down to it. The Textbox has an internal handler for the Cut command, so that is why nothing bubbles back up from there.

If that wasn't confusing enough, you then also have to look at what goes on for command enabling. The CanExecute part of ICommand has a similar pair of tunneling and bubbling events that fire. They too always get raised on command source controls first. These events are triggered by things that change the focus of the UI. If there is more than one command source, each one gets consulted in turn. Then the path to the focused element in the UI gets consulted. Below is the sequence of events after the UI is up and idle, then I click in the Textbox to set the focus there.

Window Preview Can Execute Called
DockPanel Preview Can Execute Called
StackPanel Preview Can Execute Called
Window button Preview Can Execute Called
Window button Can Execute Called
StackPanel Can Execute Called
DockPanel Can Execute Called
Window Can Execute Called
Window Preview Can Execute Called
DockPanel Preview Can Execute Called
Toolbar Preview Can Execute Called
Toolbar button Preview Can Execute Called
Toolbar button Can Execute Called
Window Preview Can Execute Called
DockPanel Preview Can Execute Called
StackPanel Preview Can Execute Called
Textbox Preview Can Execute Called

So the bottom line to me is this: it is very difficult to explain and understand where the events are flowing in even a simple UI, let alone a very complex one. Trying to rationalize that in your development process to put command bindings anywhere except on the root element is the path to command routing hell. You are better off just always placing your command bindings on the root element, or if you don't want to do that, consider using Prism Commands which get the handlers decoupled from the visual tree and the focus. If you are a control developer, and you need to handle commands (such as Cut) that are inherently coupled to whether your control has the focus, then putting command bindings in your control will also make sense.

I hope that helps to clarify the magic that is happening under the covers of routed command routing.

Sunday, November 09, 2008 3:40:57 PM (GMT Standard Time, UTC+00:00)
Comments [2]  | 

Have you by chance looked at the cool stuff we produced for Composite Application Guidance for WPF (aka Prism), and wished you could have the same capabilities for Windows Forms, but without needing to adopt CAB/SCSF? Now you can get at least part of those capabilities.

When we were drawing to a close on Prism 1, there were certain things that we would have liked to do, but had to skip to stay within scope and schedule.

One of those things was that we realized that a lot of what we had come up with could be used in Windows Forms as well. Specifically, our module loading, events, and UI Composition patterns could all be used in the context of a Windows Forms application with just a little more effort. For those things that we knew were more general, we at least factored them out to a separate library, Microsoft.Practices.Composite. Alas, that is as far as it went because our charter was for WPF, not Windows Forms.

So I have been trying to find time to complete the work needed to use the non-WPA capabilities of CAL in Windows Forms and finally found a chunk of time at 37K feet on a trip to get it done.

For now I've created these as a stand alone set of libraries that you can add into your own solutions. Feel free to take the code, change the namespaces, etc. Eventually I'll integrate the capabilities into the CompositeWPFContrib libraries so it will be accessible there as well. What I added includes the following:

• A SimpleUnityBootstrapper base class that removes the dependence on WPF and which removes the parts that create the Shell and Region related stuff. Regions in Prism were implemented in a way specific to WPF, and the CreateShell override in the UnityBootstrapper that comes with Prism expects you to return a DependencyObject, which is also a WPF base class. The SimpleUnityBootstrapper makes it so you can follow the same bootstrapping process as a WPF Prism application to get the container, module, and event services registered and ready to go.
• A CompositeEvent class and supporting classes that allows you to have the same programming model and infrastructure as Prism loosely coupled Pub / Sub events, but this class supports UI thread dispatching to either Windows Forms, WPF, or any SynchronizationContext aware subscriber. If you are not familiar with the SynchronizationContext class, it came in with .NET 2.0 and provides a standard thread dispatching mechanism that is used by both WPF and Windows Forms under the covers. You can also create custom synchronization contexts for other places that you have thread affinity. If you want to see an example of a custom SynchronizationContext, check out the unit tests. I needed to write a mock SynchronizationContext to verify the dispatching worked correctly.
• A sample application that shows a way to get a Prism Region-like capability to inject views into the shell.
• Unit tests for all the supporting library code, most of which were blantantly adapted (mostly direct copy/paste) from the unit tests of Prism since I was trying to create something with pretty much the same API and the same behavior as the Prism counterparts.

I'm not going to pick apart the implementation here unless demand is great enough, but you are welcome to dig into the code in the source.

However, the point was to make it easy to have modular loading and loosely coupled events in a Windows Forms application. So let me show you the use of these capabilities in a sample application. The app itself is very simple, just a drop down list of products and a save selection button to simulate a save operation.

The app, in all its stunning UI glory, looks like this:

The combobox and button are part of a view loaded up by a module, which is loaded through the standard module enumeration and loading services in Prism. When the button is pressed in the view, its ViewModel fires a pub/sub event, which the shell subscribes to and updates the status bar. The view is a Windows Forms user control that I was able to add dynamically to the shell using a simple form of region based on resolution through the container.

So the derived bootstrapper class from the SimpleUnityBootstrapper base class looks like this:

internal class Bootstrapper: SimpleUnityBootstrapper
{
    protected override void ConfigureContainer()
    {
        base.ConfigureContainer();
        Shell = Container.Resolve<Shell>();
    }
    protected override IModuleEnumerator GetModuleEnumerator()
    {
        return new StaticModuleEnumerator().AddModule(typeof(CompositeWFModule));
    }

    public Shell Shell { get; private set; }
 
}

I used the override of ConfigureContainer (same purpose as the UnityBootstrapper ConfigureContainer method) as a convenient place to construct the shell through the container, so that it can participate in dependency injection. This is an alternative approach to the CreateShell override that is required in the UnityBootstrapper of CAL, which is needed in CAL because the base class uses WPF Dependency Properties to add the RegionManager reference as an attached property to the shell after its construction. Since we are in Windows Forms land, that approach won't work, so regions don't exist in my modified libraries, but my sample app shows a way to achieve a lightweight form of dependency injection based on the shell being available through the container.

For module enumeration, you can see that I am just using the StaticModuleEnumerator that ships with CAL. All that is really needed to take advantage of the modular loading capabilities of CAL is a suitable bootstrapper base class that doesn't assume you are in WPF land. Thus my SimpleUnityBootstrapper class.

The Shell property is exposed to make it easy to get the shell to pass to Application.Run in the program's main method:

[STAThread]
static void Main()
{
    Application.EnableVisualStyles();
    Application.SetCompatibleTextRenderingDefault(false);
    Bootstrapper bootstrapper = new Bootstrapper();
    bootstrapper.Run();
    Application.Run(bootstrapper.Shell);
}

To create a similar capability to Regions for view injection in the Shell in Windows Forms, I simply have the shell register its containers that it wants to act as a region with the container so they can be resolved by the modules that load up:

public partial class Shell : Form
{
    int m_UIThreadId = Thread.CurrentThread.ManagedThreadId;

    public Shell(IUnityContainer container,IEventAggregator evtAggregator)
    {
        InitializeComponent();
        container.RegisterInstance<Panel>("MainRegion",m_MainRegionPanel);
        evtAggregator.GetEvent<ProductSavedEvent>().Subscribe(OnProductSaved,
         ThreadOption.SubscriberAffinityThread);
    }

    void OnProductSaved(Product p)
    {
        Debug.Assert(Thread.CurrentThread.ManagedThreadId == m_UIThreadId);
        m_StatusLabel.Text = p.Name + " saved"; 
    }
}

Notice that the shell is able to have the container and the CAL event aggregator service injected as dependencies because it was constructed through the container itself (in the bootstrapper shown earlier). It then registers the existing instance of a container control with the container as a named instance (MainRegion in this case) so that a loading module can resolve that named instance to know where to add its views as children.

Ignore the event subscription code for now, I'll get back to that.

The solution has a module library with a module class in it following the patterns of Prism and using the CAL IModule interface definition.

public class CompositeWFModule : IModule
{
    IUnityContainer m_Container;
    public CompositeWFModule(IUnityContainer container)
    {
        m_Container = container;
    }

    public void Initialize()
    {
        Panel mainRegion = m_Container.Resolve<Panel>("MainRegion");
        ModuleView view = m_Container.Resolve<ModuleView>();
        view.Dock = DockStyle.Fill;
        mainRegion.Controls.Add(view);
    }
}

Here you can see how the module injects the view into the pseudo-region. It just resolves the region out of the container with a known container type (yes, you do have to live with some type coupling here, at least to the minimal common denominator base class Panel), and once it has that, it just adds the view as a child control. For  more advanced scenarios such as adding tabs to a tab control you would need more type information to add the child, activate it, set the tab text, etc., but you could still be reasonably loosely coupled to where that control exists in the UI and how it gets constructed.

That is about it for the module loading and a lightweight substitute approach for regions in Windows Forms.

Then the real thing I wanted to get working for Windows Forms is Prism's loosely coupled pub/sub events model. This took a little more work simply because we tied ourselves a little too tightly (in my opinion) to WPF when implementing the UI Thread dispatching option for events in Prism. Both WPF and Windows Forms actually share a common approach to UI thread dispatching under the covers - SynchronizationContext. They both use it a little differently and encapsulate it so you can't directly get to it through their API, but since Windows Forms 2.0, that is available under the covers and in WPF that is what the DispatcherObject base class of UIElements is using to dispatch things to the UI thread.

So what I wrote was a more generic approach to UI thread dispatching under the covers of the CompositeEvent class (following the patterns of the CompositeWpfEvent class in CAL). The dispatching approach checks to see if the subscriber has a synchronization context, and if so, captures it and uses it to dispatch the event back to the UI thread through the SynchronizationContext when the event is fired if the subscriber asked for the event to be dispatched on the "SubscriberAffinityThread" (the name of the ThreadOption enum member I came up with.

I didn't want to call it UIThread because the pattern is actually more general than that. You can create your own SynchronizationContexts with a little work (in fact, I had to do just that to write a mock for the SynchronizationContext to test the code I was writing), and you might do that for specialized scenarios. For example, say you were going to be doing some background work in your app, you wanted to have a custom thread pool of 5 threads, and those threads had thread affinity (i.e. using ThreadLocalStorage). In that case, you could implement a thread pool synchronizer (see the example Thread Pool Attribute in our downloads at IDesign) that would ensure that there was a SynchronizationContext to dispatch to the appropriate thread. In that case, the CompositeEvent class I came up with will properly dispatch based on that SynchronizationContext instead of just a UI one.

So anyway, it is way more complicated to try to explain it than it is to use it. It follows the same simple but powerful patterns and capabilities as the CAL event.

To define an event type, you just derive from my CompositeEvent class instead of the CAL CompositeWpfEvent class, specifying the strongly typed event argument or payload as the generic type parameter:

public class ProductSavedEvent : CompositeEvent<Product>
{
}

To subscribe, you have the single line of code shown earlier in the shell constructor which points to the handling method (strongly typed based on the event argument type:

evtAggregator.GetEvent<ProductSavedEvent>().Subscribe(OnProductSaved,           
  ThreadOption.SubscriberAffinityThread);

In this case I show the overload that just specifies the threading option, but just like the CAL events, this one supports the same three options in addition to just loosely coupled pub/sub:

• Weak references or not for the subscriber object to avoid holding the subscriber object alive if the subscription is the only thing keeping it alive
• Thread dispatching option for the notification
• Filter delegate based on Predicate<T> to allow strongly typed decision based on the event payload to determine if your subscriber should be called or not.

The publisher code, in this case in the view being loaded up from the module, is just a single line of code like in CAL:

 evtAggregator.GetEvent<ProductSavedEvent>().Publish(product);

So there it is. If you have no exposure to Prism, you really need to start there, because I am not going to try to duplicate all the documentation on the capabilities of Prism. But once you are comfortable with the capabilities of Prism for WPF, now you have some code that will let you directly repeat the use of modules and events from Prism in a Windows Forms application with the exact same patterns, and you also have an example of how to pull off a "Region-like" capability using the IoC container and container controls in Windows Forms.

Feedback highly welcomed. I want to get any needed changes in there before I add it to the trunk for CompositeWPFContrib.

Here's the goods.

Monday, October 13, 2008 2:31:58 AM (GMT Standard Time, UTC+00:00)
Comments [4]  | 

Gave two sessions on the second day at the Software Developers Network conference (aka SDC) in Netherlands.

First was on Prism (Composite Application Guidance for WPF). I covered the architecture of a Prism app and each of the features of Prism (Modules, Regions, Composite Commands, and Composite Events).

You can grab the slides and demos for Developing Composite Applications with Prism here:  Slides   Demos

Second session was on WPF Data Binding, where I showed the many capabilities and how to use WPF data binding.

You can get the slides and demos for WPF Data Binding here:  Slides   Demos

Saturday, October 11, 2008 10:58:00 PM (GMT Standard Time, UTC+00:00)
Comments [0]  | 

Gave two sessions yesterday at the Software Developers Network conference (aka SDC) in Netherlands.

First was on using WF as a service technology. I talked about where WF fits into a SO architecture, as well as SO tenets, WF capabilities for services, and then of course, several demos of how to use it. I showed simple use of Send/Receive activities, how to split out the context instance ID from the proxy so it can be handed off to multiple clients, and how to make callsbacks from one workflow to another with service calls.

You can grab the slides and demos for Developing Service Oriented Workflows here:  Slides   Demos

Second session was on Windows Forms and WPF interop, where I showed how easy it is to embed controls from one technology into the other, and then covered some of the limitations and things to be aware of.

You can get the slides and demos for WPF in Windows Forms and Vice Versa here:  Slides   Demos

Tuesday, October 07, 2008 7:30:13 AM (GMT Standard Time, UTC+00:00)
Comments [0]  | 

I gave a talk on WCF last night at Roanoke Valley .NET Users Group. Thanks to Robin and crew for inviting me and for running the meeting. You've got a great group and facility there!

The talk covered the fundamentals of WCF along with some brief coverage of the advanced capabilities including security, transactions, queued calls, callbacks, reliability, interoperability, and other aspects. A lot of material to pack into an hour and a half along with an end to end demo coded from scratch in terms of the WCF aspects (had an existing business layer and a client that I tacked on to the back and front ends respectively, but did all the service contract, data contracts, config, proxies, etc. on the fly just to show it could be done in an hour easy.

If you are interested in the slides and demos from that talk, you can find them below.

Slides

Demos

For those who attended, the little glitch towards the end of the live coding demo was simply a matter of having two connectionStrings elements in my web.config after pasting in the EF connection string for the database.

This demo, like many of mine, use a music library database I generate from my music collection. You can either download and attach a DB backup from here or you can read about how to generate one of your own from your own collection here.

Friday, October 03, 2008 7:45:09 PM (GMT Standard Time, UTC+00:00)
Comments [0]  | 

Looks like my episode on Composite WPF (Prism) went live on DNRTV. You can find it here. didn't get through as much as I would have liked due to good questions, so will definitely try to follow up with another one soon covering events and commands.

Friday, October 03, 2008 1:50:09 PM (GMT Standard Time, UTC+00:00)
Comments [2]  | 

Glenn Block (Microsoft) and I did a video interview on buidling composite WPF applications with WPF at TechEd this year, and that interview has now gone live. You can find it in various formats off the TechEd Online landing page or here is a direct link to the low res version.

Friday, September 12, 2008 2:43:45 PM (GMT Standard Time, UTC+00:00)
Comments [0]  |