Beginning a WPF/MVVM application: Navigating between views
How do you go from one view to another without coupling views or view models together? Here is how I am currently doing it.
The Overall Application Design Layout
The layout we are after is an Outlook style navigation where there is a fixed layout template onto which all the application views will be rendered. It will therefore be possible for developers to create their own views and show them within the application's predetermined layout. The layout splits the views into two sections: The navigation section on the left and the main section on the right. These two sections form one atomic view and we want both to be bound to the same view model. The navigation content on the left allows us to change settings or select items that affect the display in the main content on the right.
Let's go into Expression Blend and create a new project. I'm going to create a simple stock trading application which initially contains just the following:
- MainWindow.xaml
- NavigationControl.xaml
- MainView.xaml
- PortfolioView.xaml
- StockQuoteView.xaml
I have created a view that consists of a Grid
with three rows and three columns, a status bar, a menu, a grid splitter and my own
NavigationControl
. I have placed the controls in the appropriate grid row and column positions.
Note the decision to create a separate control for the Navigation aspect of the main view. The behaviour of the navigation is clearly a separate concern to that of the main view and so this should be encapsulated into a separate control. This will enable the
navigation behaviour to be easily swapped for a more elaborate, more engaging navigation control later on for example. This also serves to limit the complexity of the main view XAML.
You may be asking yourself why I did not choose to make the menu also a separate control. The simple answer is: No reason at all. In fact it would probably be a better idea to do that, however for the purposes of simplicity I chose not to go down that route
because I will be defining each command that the menu is bound to as a separate member of the underlying view model and so there would be nothing much to gain in terms of simplicity or less lines of XAML overall. However if I decided that all the commands
would be served up by one IEnumerable<icommand>
property on the underlying view model from which I would dynamically create the menu items, then this would be a good reason to isolate that behaviour into a separate
DynamicMenuControl
for example. Another reason to create a separate control would be if I thought that the menu structure was going to be extensive and therefore it would benefit from having the menu defined in a separate file. The same goes for
the status bar and any other control for that matter. Either way, this is still obeying PMVVM and so the rest is just an implementation detail.
Here is the XAML for the main view:
<Grid x:Name="LayoutRoot">
<Grid.ColumnDefinitions>
<ColumnDefinition Width="0.2*"/>
<ColumnDefinition Width="Auto"/>
<ColumnDefinition Width="0.8*"/>
</Grid.ColumnDefinitions>
<Grid.RowDefinitions>
<RowDefinition Height="Auto"/>
<RowDefinition/>
<RowDefinition Height="Auto"/>
</Grid.RowDefinitions>
<GridSplitter HorizontalAlignment="Left" Width="5"
Grid.Column="1"
Grid.Row="1"/>
<StatusBar Margin="0"
VerticalAlignment="Top"
Height="23"
Grid.ColumnSpan="3"
Grid.Row="2"/>
<Menu Margin="0" Height="23" Grid.ColumnSpan="3">
<MenuItem Header="Views">
<MenuItem Header="Stock Quotes"
Command="{Binding Path=NavigateToStocksCommand}"/>
<MenuItem Header="Portfolio"
Command="{Binding Path=NavigateToPortfolioCommand}"/>
</MenuItem>
</Menu>
<local:NavigationControl
Margin="0,-23,0,0"
Grid.Row="1"
DataContext="{Binding}"/>
<ContentControl Margin="0"
Content="{Binding Path=CurrentView}"
d:LayoutOverrides="Width, Height"
Grid.Row="1"
Grid.Column="2"/>
</Grid>
The next thing I need to do is set-up the initial window of the application. It is a good idea to have as few windows defined as possible and define their content in separate user controls. Windows are expensive to maintain as they contain repeated code
and so a change to one will mean a change to all. For example, say your application goes into test and the testers discover that no parent has been set on any of the pop up windows or that the window style is not correct and there should not be a maximize
button. In this case, you would need to find all your pop-ups and change their properties. Inheritance is an option but in this case composition over inheritance allows you to define just one window into which you can dynamically set the content at run time.
You could just define a generic style for all your windows which solved this problem but then you have written countless windows and repeated code for nothing! With this in mind, the
MainWindow
XAML looks like this:
<Window
xmlns="http://schemas.microsoft.com/winfx/2006/xaml/presentation"
xmlns:x="http://schemas.microsoft.com/winfx/2006/xaml"
x:Class="PMVVM.MainWindow"
x:Name="Window"
Title="MainWindow"
Width="640" Height="480">
<Grid x:Name="LayoutRoot">
<ContentControl Content="{Binding}"/>
</Grid>
</Window>
That’s it ! Ideally this is as much content as I would like to have in the main window.
At this point, the other views don’t contain much more than a text block to illustrate that the view is different from the next. We will come back to the views later.
Lastly the Navigation control needs to provide some means of allowing the user to invoke the relevant commands:
<Grid x:Name="LayoutRoot">
<Grid.RowDefinitions>
<RowDefinition Height="Auto"/>
<RowDefinition/>
<RowDefinition Height="Auto"/>
<RowDefinition Height="Auto"/>
</Grid.RowDefinitions>
<TextBlock
Text="{Binding Path=Name}"
TextWrapping="Wrap"
d:LayoutOverrides="Height"/>
<ContentControl
Content="{Binding Path=CurrentView}" Grid.Row="1"/>
<Button Grid.Row="2" Height="23"
Content="Stock Quotes"
Command="{Binding Path=NavigateToStocksCommand}"/>
<Button
Content="Portfolio" Grid.Row="3" Height="23"
Command="{Binding Path=NavigateToPortfolioCommand}"/>
</Grid>
Now I must temporarily take off my rather dusty designer hat and put on my developer hat. So I will open up the solution I created in Blend in Visual Studio.
Now I am about to introduce two very important aspects of PMVVM which, aside from the view models, are critical to its argument. They are: Commands and Dependency Injection. Commands allow us to interact with and communicate between view models in a decoupled way. Dependency injection facilitates this further by allowing us to deal with contracts (interfaces) rather than concrete implementations of our view models. Both these facilities enable the creation of an extremely flexible application.
Now first things first. Let's create the contracts for our view models :
using System.Windows.Input;
namespace PMVVM.ViewModels
{
public interface IMainViewModel : IViewModelBase{
object CurrentView { get; }
ICommand NavigateToStocksCommand { get; }
ICommand NavigateToPortfolioCommand { get; }
void NavigateToView(object viewToNavigate);
}
}
namespace PMVVM.ViewModels
{
public interface IStockQuotesViewModel :
IViewModelBase{
string Name { get; }
}
}
using System.Collections.Generic;
namespace PMVVM.ViewModels
{
public interface IPortofolioViewModel :
IViewModelBase{
string Name { get; }
IEnumerable<string> Portfolios { get; }
}
}
using System.ComponentModel;
namespace PMVVM.ViewModels
{
public interface IViewModelBase :
INotifyPropertyChanged{
}
}
Note the empty IViewModelBase
interface. I will be adding some members to that later on. But for now, it will serve as the default implementation of our
INotifyPropertyChanged
interface.
Let's also look at the implementation of the MainViewModel
which is the only implementation we are really interested in at this point:
using System.Windows.Input;
using PMVVM.Commands;
namespace PMVVM.ViewModels.Implementation
{
public class MainViewModel : ViewModelBase,IMainViewModel{
private object _currentView;
public object CurrentView{
get { return _currentView; }
private set
{_currentView = value;
OnPropertyChanged("CurrentView");
}
}
public ICommand NavigateToStocksCommand{
get { return new NavigateToViewCommand
(Container.Container.GetA<IStockQuotesViewModel>()); }
}
public ICommand NavigateToPortfolioCommand{
get { return new NavigateToViewCommand
(Container.Container.GetA<IPortofolioViewModel>()); }
}
public void NavigateToView(object viewToNavigate){
CurrentView = viewToNavigate;
}
}
}
Here I have implemented the behaviour of the NavigateToView
method and declared the navigation command properties that the designer was expecting would be present. Obviously we need to invoke
NotifyPropertyChanged
when the current view is changed.
So why have I created a method to set the current view and not exposed the setter in the interface? At the moment, this appears to be nothing more than a coding style however the behaviour of navigation is unlikely to remain as just a setter of a property and so if we are following best practices, we should not abuse the property setter by including behavioural code other than that associated with changing the state of that member. In our pursuit for clean code, we also want to be explicit in our invocation so that the intent is clear when developers are reading the code.
The commands here are just subclasses of a variation of the RelayCommand
that Microsoft endorses in the context of MVVM (online ref #1 Josh Smith on MVVM). For illustration purposes, here it is:
using System;
using System.Collections.Generic;
using System.Linq;
using System.Windows.Input;
namespace PMVVM.Commands
{
public abstract class WpfCommand : ICommand{
private readonly string _verb;
protected WpfCommand(string verb){
_verb = verb;
}
public string Verb{
get { return _verb; }
}
public void Execute(object parameter){
RunCommand(parameter);
}
protected abstract void RunCommand(object parameter);
protected abstract IEnumerable<string> GetPreconditions(object parameter);
public bool CanExecute(object parameter){
return GetPreconditions(parameter).Count() < 1;
}
public event EventHandler CanExecuteChanged{
add { CommandManager.RequerySuggested += value; }
remove { CommandManager.RequerySuggested -= value; }
}
}
}
The verb parameter is not significant at this point but it will prove invaluable later on when we want to dynamically link commands to short-cuts or dynamically control the display text of commands in menu items for example.
Here is the navigation command subclass:
using System.Collections.Generic;
using PMVVM.ViewModels;
namespace PMVVM.Commands
{
public class NavigateToViewCommand : WpfCommand{
private readonly object _viewToNavigate;
public NavigateToViewCommand(object viewToNavigate) : base("Navigate"){
_viewToNavigate = viewToNavigate;
}
protected override void RunCommand(object parameter){
Container.Container.GetA<IMainViewModel>().NavigateToView(_viewToNavigate);
}
protected override IEnumerable<string> GetPreconditions(object parameter){
yield break;
}
}
}
The implementation of the container you see here is not important but you should be familiar with the concepts of dependency injection and inversion of control and perhaps look up the Windsor Container framework. Please refer to the source code for the implementation detail of the container framework I have implemented.
So what happens when we compile and run the solution? When we click the portfolio view button, we see our portfolio view in both left and right panels. The same with the stock quote view. But this is not quite what we want, is it? On the left, we want to have a different view showing some controls the user can interact with and invoke changes that are displayed on the right.
So how do we do that? We create another view model with an elaborate system for communicating between the panels, right? Wrong! I mean you can do that if you like but it is not necessary! This is where we leverage the wonders of WPF and data templates. As far as the logical tree is concerned, the view models for both left and right panels are the same and so they should be. Which brings me onto another argument of separating concerns: The navigation control needs to be told that its visual representation of the same view model must be different because it has a different purpose.
In the navigation control, we need to add the following XAML:
<UserControl.Resources>
<DataTemplate DataType="{x:Type Implementation:StockQuotesViewModel}">
<PMVVM:StocksQuoteNavigationView/>
</DataTemplate>
<DataTemplate DataType="{x:Type Implementation:PortofolioViewModel}">
<PMVVM:PortfolioNavigationView/>
</DataTemplate>
</UserControl.Resources>
Here I am overriding the existing data templates for the view models with a different visual representation.
I now have the requirement for two more views that control the navigation aspect of each atomic view. It's OK for me to go ahead and create these, however I will pass them on to the designer to actually implement them:
For illustration purposes, I am only going to change the portfolio navigation view by adding a couple of
TextBlock
s and a ListBox
so the user can select his/her portfolio. I will bind the list box to the list of portfolios on the underlying view model.
<Grid x:Name="LayoutRoot">
<Grid.RowDefinitions>
<RowDefinition Height="Auto"/>
<RowDefinition Height="Auto"/>
<RowDefinition/>
</Grid.RowDefinitions>
<TextBlock
Text="{Binding Path=Name}" TextWrapping="Wrap"/>
<ListBox Grid.Row="2"
ItemsSource="{Binding Path=Portfolios}"/>
<TextBlock HorizontalAlignment="Left"
Text="Select a portfolio"
TextWrapping="Wrap"
d:LayoutOverrides="Height" Grid.Row="1"/>
</Grid>
So now what happens when I compile and run the application?