Developing a Robust Data Driven UI Using WPF - The DataModel

.NET, Development, WPF March 24th, 2008

imageIn the first post in the series I gave an overview of the pattern we’ll be using.
This post will go deeper into the DataModel, as defined in the previous post:

The DataModel is defined exactly as the Model in MVC; it is the data or business logic that stores the state and does processing of the problem domain.
The DataModel abstracts expensive operations such as data fetching without blocking the UI thread. It can keep data “alive” fetching it periodically from source (example: stock ticket), merge information from several sources etc.
The DataModel is completely UI independent and pretty much straightforward to unit test.

The DataModel exposes data in a way that makes it easily consumable by WPF. As such, all if its public APIs, called by WPF for data-biding, must be called on the UI thread only. It must not block the UI thread because we want a robust functional UI so it usually performs operations on a background thread using the Dispatcher to send results back to the UI thread.

Therefore, the simplest DataModel implementation exposes several public Properties that expose data, implements INotifyPropertyChanged and/or INotifyCollectionChanged, and it abstracts the way information is fetched (using background threads to avoid blocking the UI thread when fetching the data is an expensive operation).

For two-way binding a commit and rollback mechanism, a dirty flag, etc.  We’ll get to that later on…

As the DataModel implementation needs to abstract expensive data fetching operations and work with multiple threads we need some basic understanding of WPF’s threading model before we look at the DataModel implementation…

WPF Threading Model - A Quick Overview

A typical WPF uses two threads:

  • Rendering thread - runs in the background and handles rendering
  • UI thread - Receive inputs, handles events, paints the screen and runs application code.

The UI thread queues work items in a Dispatcher object. The Dispatcher object selects work items on a priority basis and runs each one to completion.
Every UI thread must have at least one Dispatcher, and each Dispatcher can only use one thread to execute work items.

Therefore, in order to build responsive UI that doesn’t block the UI thread, the application has to maximize the Dispatcher’s throughput by keeping work items small as to minimize the time the Dispatcher spends on processing them - which keeps other work items waiting causing the UI to lag.

In order to perform expensive operations without blocking the UI thread we can use a separate thread that will run in the background, leaving the UI thread free to process items in the Dispatcher queue. When the background thread is done processing it can report results back to the UI thread for display.
Doing this isn’t trivial as Windows only allows UI elements to be accessed by the thread that created them. This means that the background thread we used for some long-running task cannot access and update our UI when it is finished (or during work to show progress) - a background thread updating a control (such as a list box) during its rendering can cause strange UI behaviors that this limitation is there to prevent.

WPF uses the following design to enforce this kind of coordination between the UI thread and other threads:
Most of the classes in WPF derive from DispatcherObject. During construction, a DispatcherObject stores a reference to the Dispatcher linked with the current running thread - creating an association between itself and the thread that created it.
At the beginning of every method in the DispatcherObject, it calls VerifyAccess which compares the Dispatcher associated with the current thread with the Dispatcher stored during the object’s construction - if they do not match it throws an exception.

If only the creator of a DispatcherObject can access it, how can a background thread interact with the user?
The background thread does not access the UI directly but it can ask the UI thread to perform a task on its behalf by registering work items to its Dispatcher using it’s Invoke (for a synchronous call that returns when the UI thread finished executing the delegate) or BeginInvoke methods (which runs asynchronously)

The DataModel Class

So now, after the brief discussion on the use of the Dispatcher we can start coding our base DataModel class.
We’ll start with the simple class and constructor definition:

public abstract class DataModel : DispatcherObject, INotifyPropertyChanged
{
    public DataModel()
    {
    }


We’re deriving from DispatcherObject because we need to have the Dispatcher available so that we can run background jobs that dispatch results to the UI thread.

As discussed earlier, each call to the DataModel should be made on the UI thread. Therefore we would like to enforce that limitation at the beginning of each publicly exposed API. The DispatcherObject class that we derived from contains a VerifyAccess() method that does just that. The method is public but unfortunately marked with the [EditorBrowsable(EditorBrowsableState.Never)] attributes which will make it hard to find for developers using driving their data model from our class.

To resolve this I simply defined a protected method as follows:

/// <summary>
/// Makes sure the call is in the correct thread (the UI thread) by comparing the current dispatcher
/// object with the dispatcher we got when the DataModel was created.
/// </summary>
[System.Diagnostics.Conditional("Debug")]
protected void VerifyCalledOnUIThread()
{
    this.VerifyAccess();
}

This method will be visible to anyone deriving from our class and it simply calls VerifyAccess to make sure code is made from the UI thread.
The Conditional attribute makes this code execute only in debug bits avoiding this kind of assertion on retail bits - some performance gain.

In order to support asynchronous data fetching the DataModel should encapsulate the information about its state - valid (data fetched), invalid (error fetching data), fetching (processing).

public enum DataModelState
{
    /// <summary>
    /// The model is fetching data
    /// </summary>
    Fetching,
    /// <summary>
    /// The model is in an invalid state
    /// </summary>
    Invalid,
    /// <summary>
    /// The model has fetched its data
    /// </summary>
    Active
}

The data model’s state is exposed using a property:

public DataModelState State
{
    get
    {
        VerifyCalledOnUIThread();
        return _state;
    }
    set
    {
        VerifyCalledOnUIThread();
        if (value != _state)
        {
            _state = value;
            OnPropertyChanged("State");
        }
    }
}

We also implement INotifyPropertyChanged to allow the model to communicate changes in its values.
Since adding\removing event handlers to the PropertyChanged event is a public API exposed by the DataModel, it also requires verification that calls to it are made from the UI thread. We’ll define our own add\remove handlers in order to perform this verification:

protected virtual void OnPropertyChanged(string propertyName)
{
    VerifyCalledOnUIThread();

    if (_propertyChangedEvent != null)
    {
        _propertyChangedEvent(this, new PropertyChangedEventArgs(propertyName));
    }
}

#region INotifyPropertyChanged Members
public event PropertyChangedEventHandler PropertyChanged
{
    add
    {
        VerifyCalledOnUIThread();
        _propertyChangedEvent += value;
    }
    remove
    {
        VerifyCalledOnUIThread();
        _propertyChangedEvent -= value;
    }
}
#endregion

Any property that we’ll add to our data model will call OnPropertyChanged on its setter in order to notify it has changed.

It’s Alive!

One more ability we’d like to add to our DataModel class is the ability to enable\disable it.
As defined earlier, the DataModel encapsulates the logic of fetching data and keeping it “alive” and up to date. To do that, it’ll need to keep some internal timer for updating information or register to some change notification event on its source.
This will keep the DataModel alive and can result in memory leaks, which is why we need some way to turn the DataModel on and off, allowing it to unregister from its data sources when that connection is no longer required:

public bool Enabled
{
    get
    {
        VerifyCalledOnUIThread();
        return _isEnabled;
    }
    set
    {
        VerifyCalledOnUIThread();
        if (value != _isEnabled)
        {
            _isEnabled = value;
            OnPropertyChanged("Enabled");
        }
    }
}

public void Enable()
{
    VerifyCalledOnUIThread();

    if (!_isEnabled)
    {
        this.Enabled = true;
        OnEnabled();
    }
}

public void Disable()
{
    VerifyCalledOnUIThread();

    if (_isEnabled)
    {
        this.Enabled = false;
        OnDisabled();
    }
}
protected virtual void OnEnabled()
{
}
protected virtual void OnDisabled()
{
}

When binding UI elements to the DataModel we’ll need some mechanism to enable the DataModel when the element is loaded and disable it when the element is unloaded. There’s an elegant way to implement this behavior which we’ll implement in a future post.

That’s it! We’ve got a basic class to derive out data models from. Note that we’re only addressing one-way data binding for the moment. We’ll address a two-way data model (which requires the ability to commit\rollback data etc.) in future post.

On the next post we’ll look into a concrete DataModel implementation for our Stocky application.

You can download the code for this post from here:

 

Further Reading

kick it on DotNetKicks.com

Tags: , , , ,

Developing a Robust Data Driven UI Using WPF - Introduction

.NET, Architecture, Development, WPF March 18th, 2008

WPF, Microsoft’s not-so-new-anymore UI technology offers new capabilities allowing both developers and designers to work together to achieve a stunning experience for their applications.

Power, however, does not come without complexity, and WPF does not provide a framework or a model to solve many of the problems faced by developers and designer when building an application:

1. Handling Rich Data Forms. Many applications, especially when it comes to enterprise applications, rely heavily on displaying and manipulating data. Fetching the data while keeping the UI alive and responsive is a complicated task that’s also hard to debug and requires an experienced developer doing the work.
Can we come up with a framework that will simplify data fetching?

2. Testability is a Requirement for Software Development Framework. Development organizations are no longer satisfied with simple reduction of costs for initial development and there’s a growing demand for frameworks and tool to facilitate a sustainable and agile development process.
Can we come up with a model that will allow writing tests for the application’s UI and behavior?

3. Metadata Driven User-Interface. WPF provides XAML as a meta-model for UI definitions. However there is no clear separation between metadata and code which is a mess when it comes to designer and developers working together.
Can we come up with a model to allow developers provide all the UI logic as closed building blocks that designer can just use in a plug-and-play manner?

Providing a Framework for Building Robust, Data-Driven UIs

The Model\View\Controller (MVC) architectural pattern has long been used by complex applications to present large amount of data to the user.
The pattern allows developers to separate the actual data (Model) from the user interface (View) and the business logic manipulating the data (Controller).

In the following set of articles I will present a variation of the MVC pattern tailored for modern UI development (in WPF) where we’d like the View to be the responsibility of a designer rather than a classic developer writing code.

I’ll be using the DataModel\ViewModel\View terminology to describe the pattern (although you may find the same pattern described using various other terminologies when browsing the net).

Introducing the DataModel\ViewModel\View Pattern

As mentioned earlier, the DataModel\ViewModel\View pattern is a variation of the MVC pattern. Its focus is on making the View, which is the actual UI presented to the user, the responsibility of a designer - a person who is generally more oriented towards graphics, art and interaction than to classic coding.

The design of the view should be done in a declarative form (XAML) using a WYSIWYG tool (Expression Blend).
In short, the actual UI is developed using different tools and languages by a person with a different skills set than business logic and data backend.

In order to understand the meaning behind the DataModel\ViewModel\View terminology lets look at the following diagram describing
typical architecture for our application’s presentation using this pattern:

image

The DataModel

The DataModel is defined exactly as the Model in MVC; it is the data or business logic that stores the state and does processing of the problem domain.
The DataModel abstracts expensive operations such as data fetching without blocking the UI thread. It can keep data “alive” fetching it periodically from source (example: stock ticket), merge information from several sources etc.
The DataModel is completely UI independent and pretty much straightforward to unit test.

The View

The View consists of visual elements and represents the actual user interface presented to the users (buttons, windows, graphics, etc.). It also defines interaction for keyboard shortcuts and other input devices .

The View is defined declaratively in XAML by the designer (usually using a tool such as Expression Blend).
Using such a declarative model makes it to harder to represent some state that the original  View from the MVC pattern was meant to deal with - this includes dealing with multiple modes of interaction (such as “view mode” and “edit mode”) that change the visuals and behavior of the controls.

This is where we make use of WPF’s advanced data binding mechanism. In a simple scenario we can simply bind the View to the DataModel and use binding expressions to perform one-way binding for display only values or two-way binding to allow editing values in the DataModel.

In most scenarios, however, only a small subset of the application’s UI can be bounded directly to the DataModel. This can be the case when the DataModel is a pre-existing class or data schema over which the application developer has no control. The values exposed by the DataModel are likely to require some processing in order to allow binding to UI elements. There may also be several complex operations that require code implementation and do not fit into the strict declarative-only definition for a View but are too application specific to be part of the DataModel (which we might not have control over).
We may also want to save some view state such as view mode (view\edit\etc.) or item selection etc.

To bridge this gap between the declarative View and the DataModel we define the ViewModel…

The ViewModel

The ViewModel bridges between the DataModel and the View and performs all the tasks mentioned in the previous paragraph.
The terms is meant to describe a “Model of a View” which basically means that the ViewModel abstracts all the behavior logic behind a specific screen (View) in the application.
The ViewModel include converters that can transform DataModel types into View types, Commands that can be executed the the View’s control and interact with the DataModel and general behaviors that can be attached to UI elements in the View.

Summary and Next Steps

stockyscreen

The DataModel\ViewModel\View defines a simple yet powerful pattern allowing developers and designers to collaborate on building a robust, data-driver WPF UIs.

It allows separating the data layer from the view layer and the UI to support easier development of granular components that are also unit-testable.

To demonstrate how the various pattern components are developed and used we’ll be going over the development process of a stock ticker widget-like application dubbed Stocky (screenshot on the right) and see how this development pattern simplifies the creation of an otherwise quite complicated little application.

References:

kick it on DotNetKicks.com

Comments (6) from www.ekampf.com/blog/:

Tuesday, March 18, 2008 4:09:58 PM (GMT Standard Time, UTC+00:00)

In my company We developing a very big medical system with UI based on WPF.
We used a combination of the Model-View-Presenter and the DataModel-View-ViewModel introduced by Den Crevier’s.
looking forward to see your implementation.

Ran Trifon

Tuesday, March 18, 2008 4:52:31 PM (GMT Standard Time, UTC+00:00)

Well it’s pretty much the same…
The goal here is to summarize all the information into one place. Dan’s post are pretty short and straightforwards aimed at experienced developers and these post are meant to be more detailed.
I am going to post about topics he didn’t mention though…

Eran Kampf

Tuesday, March 25, 2008 3:41:31 PM (GMT Standard Time, UTC+00:00)

Nice post - I’m really looking forward to seeing where you go with this. I’ve just recently being trying to find some guidance on setting up an MVC/MVP framework in WPF. Dan’s series is great but I must admit that I really didn’t understand it all until I began my own implementation and things began to “gel”. Will be great to see another perspective on it.

Nigel Spencer

Tuesday, March 25, 2008 3:53:55 PM (GMT Standard Time, UTC+00:00)

Thanks Nigel,
Next post in the series is already available at http://www.ekampf.com/blog/2008/03/24/DevelopingARobustDataDrivenUIUsingWPFTheDataModel.aspx

Eran Kampf

Tuesday, March 25, 2008 6:38:44 PM (GMT Standard Time, UTC+00:00)

Eran:
Just wanted to say “keep up the good work”. between your work and Dan’s series of articles, I think I’m starting to get a handle on this. My one request is that I’d like to see how your DataModel interacts with the DataAccess Layer against SQL Server. Maybe just something against Northwind. I realize this might be outside the main scope but I think it would be interesting.
Sincerely,
Dale Williams

Dale Williams

Tuesday, March 25, 2008 7:34:38 PM (GMT Standard Time, UTC+00:00)

Hey Dale,
Thanks for the feedback :)
The 3rd post in the series will show a concrete DataModel example.
Since I was aiming to show how I build a Yahoo finance widget clone I was building the DataModel on that - keeping a stock data up to date (kind of like in Dan’s article).
However, once you see how the DataModel fetching is implemented it doesn’t really matter if the actual data is fetched via SOAP call, http, or a DB access so you’ll be able to implement a one-way binding to a data source of your choice.
While the current implementation only deals with one-way binding (only fetching the data from the without the ability to update data on the source) I do plan to show how to implement two-way binding and support comitting and rolling back data in future posts.
Thanks,
Eran

Eran Kampf

Tags: , , ,