Im a BIG fan of  Silverlight and also i really like the Model View ViewModel pattern. Nice part is that those two fit perfectly. The downside is that everyone is doing MVVM the way they like it or the way they think they like it. There is no strict guidance,  there are no ‘Bible’ kind of books written on the subject, there is only a enormous pile of information (of varying quality) on the web contributed by developers all over the world trying to (ab)use this pattern in their everyday practice.

(probably you can take this post as just another small contribution to that information pile on the MVVM).

Yes, there are many mature MVVM frameworks for Silverlight and WPF out there (Prism, MVVMLight, Caliburn etc) and they try to provide standardized ways of doing things but they either lack features or they are overly complex for my taste. Also problem with frameworks is that each of them solves some problems in good or excellent way yet other things are not covered or they are done in way that is not acceptable for each need.

So i guess if you could just merge them all to one solid framework it would be a hell of a tool.

One of the features that most of Silverlight MVVM frameworks are actively trying to solve is the process of wiring up Views and Viewmodels and related to this problem is so-called Blendability.  Blendability is just a fancy word for enabling designers to see the correct preview of their Views in Visual Studio and/or Expression Blend.

And this is not easy task that’s for sure.  Here is why:

As soon as your View Models become more complex and if you start using Dependency Injection with Constructor Injection – because IDE is trying to instantiate your ViewModel in designtime and fails – design surface will blow up in your face and instead of the preview you will just see some error message (if you are lucky enough and your IDE did not crashed already).

I do like how MVVMLight framework is preserving Blendability by using concept of the ViewModelLocator.

Basically this class is some sort of ViewModel factory that returns different ViewModel if its called in runtime or designtime. This keeps the IDE happy and it does not crash because it gets some dummy model when in designtime and when application is trully running it then gets realt ViewModel so everyone is happy :)

While this is pretty smart concept, i don’t like how it is implemented in practice.

In MVVMLight ViewModelLocator  is usaully defined in App.Xaml.cs as global resource (and therefore created on application startup).

Also designtime and runtime versions of ViewModel instances in ViewModelClass are defined as static properties and therefore also created when ViewModelLocator class is first used. Also those static ViewModels are then shared between multiple views which is in my opinion simply  wrong.

Every view should have its own ViewModel instance created for it in runtime.

For example what if i want to show same view multiple times on a single page? (lets say its case of multiple RSS feed widgets on one page each of them showing different RSS feed etc).

Each ViewModel should be instantiated at the same time as its View or at least i should have the freedom to choose when its created depending on the situation and also i should be able to choose if it is shared between multiple instances of same view or not.

So inspired by the concept of ViewModelLocator from MVVMLight i created my own variation of this class and i will be presenting this solution in this post.

So enough talk  lets first show some code. This is my current version of the ViewModelLocatorBase class:

ViewModelLocatorBase.cs:

using System.ComponentModel;
using Framework.Abstractions.Silverlight.ServiceLocation;

namespace Framework.Implementors.Silverlight.MVVM
{
  public abstract class ViewModelLocatorBase<TViewModel> : NotifyPropertyChangedEnabledBase where TViewModel : class
  {
    private static bool? isInDesignMode;
    private TViewModel runtimeViewModel;
    private TViewModel designtimeViewModel;

    /// <summary>
    /// Gets a value indicating whether the control is in design mode
    /// (running in Blend or Visual Studio).
    /// </summary>
    public static bool IsInDesignMode
    {
      get
      {
        if (!isInDesignMode.HasValue)
        {
          isInDesignMode = DesignerProperties.IsInDesignTool;
        }

        return isInDesignMode.Value;
      }
    }

    /// <summary>
    /// Holds the intance of the runtime version of the ViewModel that is instantiated only when application is really running by retrieving the instance from IOC container
    /// </summary>
    protected TViewModel RuntimeViewModel
    {
      get
      {
        if (this.runtimeViewModel == null)
        {
          this.RuntimeViewModel = SimpleServiceLocator.Instance.Get<TViewModel>();
        }
        return runtimeViewModel;
      }

      set
      {
        runtimeViewModel = value;
        this.OnPropertyChanged("ViewModel");
      }
    }

    /// <summary>
    /// Gets current ViewModel instance so if we are in designer its <see cref="DesigntimeViewModel"/> and if its runtime then its <see cref="RuntimeViewModel"/>.
    /// </summary>
    public TViewModel ViewModel
    {
      get
      {
        return IsInDesignMode ? this.DesigntimeViewModel : this.RuntimeViewModel;
      }
    }

    /// <summary>
    /// Holds the intance of the designtime version of the ViewModel that is instantiated only when application is opened in IDE designer (VisualStudio, Blend etc).
    /// </summary>
    public TViewModel DesigntimeViewModel
    {
      get
      {
        return designtimeViewModel;
      }

      set
      {
        designtimeViewModel = value;
        this.OnPropertyChanged("ViewModel");
      }
    }
  }
}

As you can see ViewModelLocatorBase is generic and abstract class that should be inherited by concrete ViewModelLocators, usually one for each View but this is not necessarily the case and you can have as many ViewModel locators for single view as you like (or you can share same ViewModelLocator between multiple instances of same View on one page if that is really what you need, also each view can have its own if needed).

Generic type TViewModel is usually the interface that your ViewModel is implementing (you could use concrete class type there but i strongly disagree with that – Views should bind to interfaces and later you can easily mock them etc).
ViewModelLocatorBase inside its static property IsInDesignMode determines if its instantiated inside aome IDE designer or inside the real running application.

I have tested it in Blend 3 and Visual Studio 2010 Beta 2 and it works fine.

Based on that fact (runtime or designtime) ViewModelLocatorBase.ViewModel property returns appropriate ViewModel instance.
If its designtime it returns value of DesignTimeViewModel property and if its runtime it returns RuntimeViewModel property value.
Now as you can see in the getter of RuntimeViewModel property, when its first time accessed it retrieves the concrete instance of our generic TViewModel type (it uses my home grown IOC adapter called SimpleServiceLocator but you can plugin any IOC you want behind it).

This instance is then saved to private property so its created only once.

So this way on runtime you will get there whichever concrete implementation you set up in your IOC for the TViewModel type.

Here is how i setup IOC in Bootstrapper class:

  public class Bootstrapper
  {
    public static void InitializeIoc()
    {
      SimpleServiceLocator.SetServiceLocatorProvider(new UnityServiceLocator());
      SimpleServiceLocator.Instance.Register<IMessageBoxService, MessageBoxService>();
      SimpleServiceLocator.Instance.Register<IEditUsersViewModel, EditUsersViewModel>();
    }
  }

To have something returned from the getter of the ViewModel in designtime we need to set the DesigntimeViewModel property.
We do that in the concrete ViewModelLocator class that is inheriting from ViewModelLocatorBase so that in design time IDE will get this instance.

Here is one simple concrete ViewModelLocator implementation:

  public class EditUsersViewModelLocator : ViewModelLocatorBase<IEditUsersViewModel>
  {
    public EditUsersViewModelLocator()
    {
      this.DesigntimeViewModel = new DummyEditUsersViewModel();
    }
  }

So the concrete EditUsersViewModelLocator class simply inherits the ViewModelLocatorBase and for the generic type it specifies the interface IEditUsersViewModel that ViewModels for the EditUsersView should implement and after this it sets the value of the DesigntimeViewModel property to a instance of DummyEditUsersViewModel which is a (as its name clearly states) just some dummy implementation of the IEditUsersViewModel with some test data to be displayed on the screen when editing this view in Blend or VisualStudio designer.

Sounds simple? well that was the original goal :)

Lets paste some more code. This is how the IEditUsersViewModel interface looks like:

  public interface IEditUsersViewModel
  {
    ObservableCollection<User> Users { get; }
    ICommand DeleteUserCommand { get; }
  }

Here is the dummy ViewModel for the design time that has test data to be displayed in designtime:
(it uses static properties to prevent VS2010 with r# from crashing if you open the views without first compiling the whole solution)

  public class DummyEditUsersViewModel : ViewModel, IEditUsersViewModel
  {
    private static ObservableCollection<User> users = new ObservableCollection<User>();

    static DummyEditUsersViewModel()
    {
      users = new ObservableCollection<User>
                     {
                       new User {Username = "Designtime User a", IsAdmin = true},
                       new User {Username = "Designtime User b", IsAdmin = true},
                       new User {Username = "Designtime Admin User c", IsAdmin = true},
                       new User {Username = "Designtime Admin User d", IsAdmin = true}
                     };
       deleteUserCommand = new DelegateCommand<object>(p => {});
    }

      private static ICommand deleteUserCommand;
      public ICommand DeleteUserCommand
      {
          get { return deleteUserCommand; }
          private set { deleteUserCommand = value; }
      }

      public ObservableCollection<User> Users
    {
      get { return users; }
      set { users = value;}
    }
  }

And here is how we would use this concrete ViewModelLocator in our view:

<UserControl x:Class="MvvmBlendability.Views.EditUsersView"
    xmlns="http://schemas.microsoft.com/winfx/2006/xaml/presentation"
    xmlns:x="http://schemas.microsoft.com/winfx/2006/xaml"
    xmlns:d="http://schemas.microsoft.com/expression/blend/2008"
    xmlns:mc="http://schemas.openxmlformats.org/markup-compatibility/2006" xmlns:ViewModels="clr-namespace:MvvmBlendability.ViewModels" mc:Ignorable="d"
    d:DesignHeight="300" d:DesignWidth="400"  d:DataContext="{Binding Source={StaticResource viewModelLocator}, Path=ViewModel}"  >
    <UserControl.Resources>
        <ViewModels:EditUsersViewModelLocator x:Key="viewModelLocator"  />
    </UserControl.Resources>

    <UserControl.DataContext>
        <Binding Source="{StaticResource viewModelLocator}" Path="ViewModel" />
    </UserControl.DataContext>

    <Grid x:Name="LayoutRoot" Background="Silver" HorizontalAlignment="Center" VerticalAlignment="Center" Margin="10">
        <Grid.RowDefinitions>
            <RowDefinition Height="Auto"></RowDefinition>

            <RowDefinition Height="*"></RowDefinition>
        </Grid.RowDefinitions>
        <TextBlock HorizontalAlignment="Right" Text="Edit Users"  FontSize="18" Margin="0,10,10,5" FontFamily="Arial Black" />
        <ListBox Grid.Row="1" ItemsSource="{Binding Path=Users}" Background="AliceBlue" HorizontalAlignment="Stretch" >

            <ListBox.ItemTemplate>
                <DataTemplate>
                    <Grid>
                        <Grid.ColumnDefinitions>
                            <ColumnDefinition MinWidth="300"></ColumnDefinition>
                            <ColumnDefinition Width="21*"></ColumnDefinition>
                            <ColumnDefinition Width="21*"></ColumnDefinition>
                        </Grid.ColumnDefinitions>
                        <TextBlock Text="{Binding Path=Username}" />
                        <Button Grid.Column="2" Content="Del" Command="{Binding Source={StaticResource viewModelLocator}, Path=ViewModel.DeleteUserCommand}" CommandParameter="{Binding}" />

                    </Grid>
                </DataTemplate>
            </ListBox.ItemTemplate>
        </ListBox>
    </Grid>

</UserControl>

In the Xaml its very important that you notice this part of the code in the UserControl definition:

d:DataContext="{Binding Source={StaticResource viewModelLocator}, Path=ViewModel}"

Its the very important part and if you miss this out you will not see much in the designer. This actually tells the designer what to use as DataContext of the UserControl (this is our view) in designtime and we will point it to the ViewModel property of our ViewModelLocator instance we defined in the UserControl.Resources section on top of the view.

Yet this code sets the DataContext binding for view for runtime:

    <UserControl.Resources>
        <ViewModels:EditUsersViewModelLocator x:Key="viewModelLocator"  />
    </UserControl.Resources>

    <UserControl.DataContext>
        <Binding Source="{StaticResource viewModelLocator}" Path="ViewModel" />
    </UserControl.DataContext>

We first create instance of out ViewModelLocator in resources of the UserControl (View) and then we set this as DataContext.
And in the design time in VisualStudio when we open EditUsersView we will get this:

As you can see design surface is working and we can see our View showing data from our  DummyEditUsersViewModel.

I have created one more View called ShellView and placed in its Xaml two instances of EditUsersView so that we can be sure it will work in this scenario also.

As i mentioned each instance of the view will get its own ViewModel created and wired up (in this case since its designtime each view will get its own instance of designtime view mode – DummyEditUsersViewModel).

First lets look at the xaml of the ShellView:

<UserControl x:Class="MvvmBlendability.Views.ShellView"
    xmlns="http://schemas.microsoft.com/winfx/2006/xaml/presentation"
    xmlns:x="http://schemas.microsoft.com/winfx/2006/xaml"
    xmlns:d="http://schemas.microsoft.com/expression/blend/2008"
    xmlns:mc="http://schemas.openxmlformats.org/markup-compatibility/2006" xmlns:Views="clr-namespace:MvvmBlendability.Views" mc:Ignorable="d"
    d:DesignHeight="400" d:DesignWidth="400">

    <Grid x:Name="LayoutRoot" Background="White">
        <Grid.RowDefinitions>
            <RowDefinition></RowDefinition>
            <RowDefinition></RowDefinition>
        </Grid.RowDefinitions>
        <Views:EditUsersView Grid.Row="0" />
        <Views:EditUsersView Grid.Row="1" />
    </Grid>
</UserControl>

In the ShellView we have two instances of the EditUsersView and both of  them will be assigned with separate ViewModel instance as we declared it in the Xaml of those views via ViewModelLocator so we dont need any ViewModelLocator for ShellView.

Basically EditUsersViews are here used as a component that we dropped on another view and this component takes care of its own DataContext without having to worry about it again. Off course we could also create view that does not have ViewModel assigned via ViewModel locator and place it on ShellView and set its DataContext directly via DataBinding.

All possibilities  are there we just need to choose the appropriate one.

So lets see how ShellView looks in the VS2010 Beta 2 designer:

And its working in Blend also (but be sure to use Microsoft Expression Blend Preview for .NET 4 version to be able to open this solution. Its not working in plain Blend 3 because this version is not supporting VS 2010 projects/solutions).

So here is the ShellView in Blend:

On the other hand, when we actually run our application in browser then ViewModelLocator sets ViewModel property to the real ViewModel instance (runtime version) and view is populated with real data from EditUsersViewModel just as we would expect:
Here is how it looks when we start our application in the browser:

As you can see each instance of the View has different data shown that proves that each has different instance of ViewModel set to its DataContext.

So we have accomplished our three main design goals:

  1. Simple yet powerful View to ViewModel wireup
  2. Blendability
  3. simplicity :)

Im attaching the VisualStudio 2010 solution with full source code of project used in this post.
You can see the live demo of this app here.

Hopefully this has given you some ideas on how to start with MVVM without having to learn frameworks before you event start programming. Later when you realize that you cannot accomplish all your ideas you can always switch to some framework and use it.

Unitl then: Happy DataBinding!