My Windows Phone cohort and all-round good guy Peter Torr posted on exactly the same topic that I did at almost exactly the same time… and we didn’t co-ordinate it at all!

http://blogs.msdn.com/b/ptorr/archive/2010/08/16/virtualizing-data-in-windows-phone-7-silverlight-applications.aspx

Creepy, huh?

Peter has a great overview of virtualization and shows a different sample and given how tricky virtualization can be I highly recommend you also read his article.

There is a lot of data out there; on the the internet, tucked away in databases, sitting patiently on the other side of a REST web service just waiting to pounce on your unsuspecting Windows Phone application that just wants to display a little slice of it all so people can read it, touch it and generally make sense of it.  The problem is there is a lot of it, so what is a poor unsuspecting application to do, especially when it’s been crammed into a form factor that doesn’t allow ever expanding memory upgrades?

In this post and a few to come I’m going to explore the various ways you can work with the ListBox (the go-to control for displaying lists of data) in your application to keep it speedy and responsive.  Today we’re going to start with a feature added in Silverlight for Windows Phone: Data Virtualization.

Data Virtualization

Data virtualization is the concept of only loading “just enough” data to fill out your UI and be useful to interact with.  This is particularly useful if your data set is large and can’t all fit into memory at the same time.  Good examples are the 3,000 images you took of various pancakes shaped like childhood cartoon characters or all 23,083 remixes of The Postal Services “The District Sleeps Alone Tonight” you have loaded up on your phone.

Data virtualization in Silverlight is accomplished when you bind a custom IList implementation to a ListBox with a data template.  Let’s break that statement down by creating a very naïve virtualized list that simulates a list of 10,000 song objects.

Implement IList

This isn’t as daunting as it might seem, there are really only two methods you need to implement: Count and the indexer property.  Everything else can throw NotImplemented.  Count returns, well, the count of items while the indexer returns the actual item being requested by index.

Most of the magic happens in the indexer.  When an item is requested you now have a chance to go off and grab your data, fabricate it (such as using a WriteableBitmap), load it from IsolatedStorage, compute it based on index or generally return whatever makes sense.

Here I’m just creating a new Song class and setting its Title to a string that matches the requested index but in reality you’d be doing some parsing/loading/fetching here.

public class Song
{
    public string Title { get; set; }
    public string Length { get; set; }
}
public class VirtualSongList : IList<string>, IList
{
    /// <summary>
    /// Return the total number of items in your list.
    /// </summary>
    public int Count
    {
        get
        {
            return 10000;
        }
    }
    object IList.this[int index]
    {
        get
        {
            // here is where the magic happens, create/load your data on the fly.
            Debug.WriteLine("Requsted item " + index.ToString());
            return new Song() { Title = "Song " + index.ToString() };
        }
        set
        {
            throw new NotImplementedException();
        }
    }
    // everything else throws NotImplemented Exception.
    // .
    // .
    // .
}

NOTE: While the code is rather easy to implement it’s boring, mind numbingly boring so I’ve made the above implementation available for your source downloading pleasure.

Bind to ListBox

This is pretty straight-forward stuff, you’ll need to bind your new fancy list to the ListBox in question.

// Constructor
public MainPage()
{
    InitializeComponent();
    ItemList.ItemsSource = new VirtualSongList();
}

DataTemplate

One caveat is your ListBox needs to use a DataTemplate otherwise virtualization doesn’t kick in.  The virtualization code-path needs to make a few assumptions about your data and having a DataTemplate helps it down that path.  Without it all your hard work will go down the drain as the first time you bind your list every single item will be requested.

<ListBox x:Name="ItemList">
    <ListBox.ItemTemplate>
        <DataTemplate>
            <TextBlock Text="{Binding Title}" FontSize="32" />
	</DataTemplate>
    </ListBox.ItemTemplate>
</ListBox>

In Action

Let’s take this code for a spin.  An F5 and emulator launch later and we’re looking at this pretty screen:

As you can see we’re using the Song created during the indexer call.  To prove it’s actually virtualized I added a Debug.WriteLine to output every time an item was requested:

As you can see only 52 items were requested instead of the full 10,000.  Why 52 instead of just 13?  You always want a few items as buffer to give your virtualizing code a chance to actually do the work while items are being scrolled into view.  For this reason we request about three page worth of data so your UI is always responsive.

Extending And Testing

A virtualized list that only returns 10,000 items is a bit limited.  A better, more testable pattern would be to create a separate repository or “creator” class that actually creates, counts and manages the objects.  Your list should be nothing but that, a list.

Caching

Another thing to be aware of, and this is important to pay attention to, virtualization doesn’t do any caching for you.  This means if you scroll down and then back up your virtualized list will ask for index 0 again and if you have a naïve implementation you will recreate item 0 over and over again.   Depending on what you’re returning you may want your repository to have some kind of object cache based on either time or current index.

When To Virtualize

If you attempt to do a lot in your indexer’s get, where you actually make the “virtual” data real, you’ll soon discover that your UI is now *worse* than it was before. That’s because as you’re scrolling it’s getting hit over and over again so if you have an expensive creation you’ll block your UI thread and everything will come to a screaming to a halt (some people say screeching halt but if I can’t interact with my UI you will hear screaming).

The general rule of thumb on when to use data virtualization is for large lists or medium lists whose items are heavy weight, such as images and expensive XAML objects.  The bottom line is you just can’t have all 3,000 images loaded up at once so you need some way to give the user a feeling of a nice, continuous smooth scroll.

So how do you minimize the delay from heavy weight objects?  Stay tuned for my next blog post when I talk about using proxy (some people like to say broker) objects to push as much of the real work into the background thread.

I’ve noticed a few questions in the forums around why the Blur and DropShadow effects aren’t showing up in the Windows Phone 7 Series emulator and the simple answer is you have to set CacheMode to BitmapCache.

<TextBlock Text="DropShadow" Foreground="Black" FontSize="48" CacheMode="BitmapCache">
	<TextBlock.Effect>
		<DropShadowEffect/>
	</TextBlock.Effect>
</TextBlock>
<TextBlock Text="Blur" Foreground="Black" FontSize="48" CacheMode="BitmapCache">
	<TextBlock.Effect>
    	<BlurEffect/>
	</TextBlock.Effect>
</TextBlock>

We are working on setting this automatically so you don’t have to scratch your head each time you apply an effect and wonder why it looks fine on the design surface yet doesn’t show up in the emulator.

UPDATE [7/20/2011] – I was looking through my blog and realized this needs to be updated. For anyone still looking for this information all the effects such as Blur & DropShadow were intentionally disabled from the product for both 7.0 and the upcoming Mango.  The performance hit that applications took from using these effects put too much of a strain on the system and it was decided that if we couldn’t deliver a perfomant feature we would disable until such a time as we could.

Given that the main way users enter text in your Windows Phone 7 Series application is via a tiny onscreen keyboard (a “Software Input Panel” or SIP) it’s in everyone’s best interest to make sure the right keys are available to the user when they need them.

A good example is when you go to enter a PIN number you really only want to see number-related keys, when composing a SMS you’d like a quick way to insert an emoticon and if writing something longer you probably want auto-correction and replacement suggestions. The good news for Windows Phone developers is you can control which SIP layout is used and which auto-correct settings are enabled by setting the InputScope property of your TextBox.

The concept of InputScope isn’t new to XAML by the way, this concept already exists in WPF and was a nice fit with the native SIP behavior so if you’ve used InputScope before you should be right at home.

Setting InputScope

There are really three ways to set InputScope, two of which are cumbersome yet provide Intellisense and a third that is easy but requires you to know exactly which InputScope you’re after.

Via XAML

Doing it this way you’ll get full Intellisense yet it’s a lot of work to set a simple property:

<TextBox>
    <TextBox.InputScope>
        <InputScope>
            <InputScopeName NameValue="Text" />
        </InputScope>
    </TextBox.InputScope>
</TextBox>

Via Code

If you’re more the code-behind sort here it is in C#:

textBox1.InputScope = new InputScope()
{
    Names = { new InputScopeName() { NameValue = InputScopeNameValue.Text } }
};

Via XAML with a TypeConvertor

Remember how in high school they’d always teach you the hard way before showing you the easy version? Guilty as charged. If you already know the exact InputScopeNameValue you want to use, for example ‘Text’, then you can take advantage of the built-in TypeConvertor that is already wired up to the property and write this much easier XAML:

<TextBox InputScope="Text" />

InputScope Example Application

When we were bringing InputScopes from WPF to Silverlight for Windows Phone I wrote a quick app to show all the available InputScopes and automatically set the selected one on a TextBox. This gives you a feel for how many there are and what layout comes up when set. Here is what it looks like along with a link to the source:

Download the source.

Common InputScopes

Now that you know how to set them here are some of the more useful InputScopes:

Default	This is the default InputScope when no input scope is specified. Auto-capitalize first letter of sentence. The app can show app specific text suggestions. Layout: Standard QWERTY layout.
Number	When all you’re looking for is basic number entry. All features like auto-capitalize are turned off. Layout: the first symbol page of the standard QWERTY layout.
Text	When the user is typing standard text and can benefit from the full range of typing intelligence features: Text suggestions (while typing and when tapping on a word) Auto-correction Auto-Apostrophe (English) Auto-Accent Auto-capitalize first letter of sentence Layout: Text layout and access to letters, numbers, symbols and ASCII based emoticons + text suggestions. Example fields: email subject and body, OneNote notes, appointment subject and notes, Word document, etc.
Chat	The user is expected to type text using standard words as well as slang and abbreviations and can benefit from some of the typing intelligence features: Text suggestions (while typing and when tapping on a word) Auto-Apostrophe (English) Auto-Accent Auto-capitalize first letter of sentence Layout: Chat layout and access to letters, numbers, symbols and rich MSN like emoticons + text suggestions. Example fields: SMS, IM, Communicator, Twitter client, Facebook client, etc.
Url	The user is expected to type a URL. All auto-correct features are turned off. Layout: Web layout with “.com” and “go” key

For a complete list of InputScopes supported in Windows Phone 7 check out this MSDN link: InputScopeNameValue Enumeration.

In Conclusion

As you’re writing your applications don’t forget about InputScope, it’ll give it just that much more polish and can really make a difference in usability.

I promised someone at the end of my talk that I’d post my code and slides and while
I’m lagging a little behind I’ve finally bundled them up and made them available
for download here:

Slides + Code

Also, as a small note I’m slowly working this blog over to have a “Windows Phone
7 Design Series” look and feel, not sure how well the dark background and light
text is going to carry over or how well my design kung fu will stack up but here
goes.

UPDATE #1: I’m also including the actual session itself here for
your viewing pleasure.  You can either watch it below or
directly on the MIX website.

UPDATE #2: Several people have asked me for sample code to the
solution that has a Windows Phone and Silverlight desktop application both consuming
the same Class Library.  Well, here it is for your downloading pleasure.

A quick heads up for anyone that saw my upcoming talk at MIX10. I’m super excited (why do normal excited when you can be *super* excited!) to show everyone the basics of Silverlight so people can begin creating rockstar web and Windows Phone applications. Speaking of the basics I want to share for anyone that did a Bing search wondering who this Shawn Oster fellow is and what his talk is really about.

My talk, An Introduction to Developing Applications for Microsoft Silverlight, really is an introduction, a 101. Maybe you are a HTML/JavaScript ninja, perhaps you’re just starting out, maybe you decided hand tuning assembly had lost its luster, whatever reason if you’re a blank Silverlight slate then this is the talk for you. On the other hand if you know about this crazy concept called XAML or have the basics of wiring up event handlers and doing a little data binding then this talk is definitely not for you.

If you’re hoping for some Windows Phone love then you don’t have to wait too much longer, Mike Harsh and Peter Torr lay down all the Windows Phone specific goods for you in the talks following mine. Think of my talk as a primer for those that need a quick intro to some of the concepts they’ll be seeing later.

MIX is an awesome, information-packed developer/designer lovefest and I’d hate for you to miss out on another track having to sit through 101 information.