Software and UX

… And saving on utility bills while you’re at it! What I’m about to show you is a way to wake up your machine remotely with nothing more than a Twitter client and a small utility running on one of the machines at your office/campus.

Abridged version

To get the utility that allows you to wake up a machine from a Twitter direct message, you need to go to the project’s home page at CodePlex. The source code is also available.

The problem…

Back in 2004, when it started becoming practical to use Remote Desktop to access my office machine, I started taking advantage of that to get more stuff done from home. Unfortunately, that required leaving the office machine turned on the whole time, 24/7, as I never knew when I might need to access it again. As you can imagine, this is terribly inefficient energetically (although arguably efficient in economic terms, as you could reason that the work done from home more than pays for the cost of the electricity bill).

Over the years, I have toyed with the idea of being able to remotely wake up my machine upon demand, leaving it in sleep mode or in hibernation mode during off-hours. I tried Wake-on-LAN, and although I could easily set it up within two machines in the office, our VPN box seemed to interfere with the communications required to wake up the machine. I needed a way to bypass the VPN…

My solution (and my first practical use for Twitter!)

Fast forward to 2010: with the advent of social networking platforms with open APIs, and the proliferation of free client software and libraries to access them, I now had all the tools I needed to quickly build a small mash up application that would sit on the server and react on Twitter direct messages. I started devising the requirements for such a tool:

• Have low impact on the machine (no installation or 3rd party products to install);
• Must be usable by everyone at the office, not just me;
• Will take requests, and reply back once your machine is online. I immediately thought of Hanselman’s toy project, Tweet Sandwich. That was pretty much what I needed as far as the basic logic. However, instead of printing orders, my utility would run a piece of networking code that would take care of waking up a machine in the local network.
  

  How does it work?

  The simple answer is that you will be creating a file with an entry on each line for each machine you want to be able to wake up (normally one for each co-worker). Each line contains:

• the Twitter screen name of the person;
• the MAC address of the machine’s network card (you can type ipconfig /all at the Command Prompt to find out the address); and
• the machine name or IP address.

When the utility app receives a special message from a given befriended Twitter account, it broadcasts a Magic Packet to the network, containing the MAC address of the specific machine, which will cause that machine to wake up. Twitter Buttler then checks whether the machine is now online and will fire back a Direct Message to the Twitter user letting him now that his machine is now turned on. At this point, the user can connect to his machine and carry on working.

Note: for you to be able to send direct messages to the utility app, the accounts listed in the config file need to follow the account used in Twitter Buttler, and vice-versa.

The are two commands you can currently send to Twitter Buttler: “wake up”, and “who’s online?”. They are pretty self-explanatory, so I won’t go into them here.

Give it a try if you normally leave your office machine running for no good reason other than to be able to access it from home. Every little step counts in the quest to reduce our carbon footprint, and as the data suggests, there is quite a lot to be gained by just leaving our machines turned off.

Yesterday, I spent the whole morning wrestling with WPF because an object I was re-implementing was appearing too blurry when compared to the existing native version. This was a basic label object, containing a picture and some text. Both the picture and the text were quite fuzzy, and don’t get me wrong, I am aware that WPF 3.5 SP1 still has issues with the text layout mechanism (the ideal layout compared to pixel-aligned layout). But this was much worse, as you can see in the screen grab:

You may notice that I am applying a drop shadow effect to the label, for some eye candy. And believe it or not, after a few hours of tweaking the XAML, and selectively disabling Visuals, I found the culprit to be the DropShadowEffect! As a designer, I would expect the DropShadowEffect to create a shadow behind my object, with the same shape as the object boundaries. Apparently this is not what the effect does, but rather overlays (or underlays?) a version of the source Visual, causing the blur effect. The solution to these types of problems is very simple: you just need to embed my Visual inside a Grid, so that you can add a Border with the same shape under my Visual. That Border will then have the effect applied to it, so basically I have moved any blurriness behind my label:

compared to

You can easily test this behavior in Kaxaml, by entering the following XAML, and removing the Effect from the Border element:

<Border Background="Yellow" BorderBrush="Black" BorderThickness="1" Width="100" Height="26">
    <Border.Effect>
        <DropShadowEffect/>
    </Border.Effect>
    <TextBlock Text="Label"/>
</Border>

I would still like to understand the reason for this behavior of the DropShadowEffect, so if you have any explanation for it, do let me know.

Today I decided to encapsulate the circular progress indicator I use in Scrum Sprint Monitor in a reusable behavior. This served as a learning experience as well, since I had not yet had the chance to play with attached behaviors. My requirements were the following:

• Have a behavior that displays an animation during long running operations. It should dim out a defined region in the UI, optionally preventing input into that region. All state transitions must be animated.
• Have minimal impact on the logical tree.
• Allow configuration of parameters, such as dim brush, transition duration, etc.

This problem turned out to be a perfect candidate to writing a behavior, and I needed only a few hours to crank it out. Here is a short video demoing it:

This behavior makes use of the simple and great Circular Progress Bar control by Sasha Barber, up on CodeProject. You can easily replace the control used for the animation though, should you require a different one.

Usage

The behavior has a simple requirement, that it can only be attached to a Grid element (I believe that is a requirement that can easily be satisfied, in most projects). The only required property is the BusyState property, which tells the behavior when to do its work:

<Grid DataContext="{Binding Path=ConfigurationViewModel, Source={StaticResource serviceLocator}}"
      Behaviors:BusyIndicatorBehavior.BusyState="{Binding IsBusy}"
      Behaviors:BusyIndicatorBehavior.TargetVisual="{Binding RelativeSource={RelativeSource FindAncestor, AncestorType={x:Type Grid}}}">

The behavior works by adding a Visual Tree at runtime under the attached DependencyObject (the Grid). You connect your Busy state flag by providing a Binding to the BusyState attached property. There are a few other optional properties you can use, to customize things such as the dimming brush, the transition duration, margins, etc.

The source code is available for download from the Microsoft Expression Community Gallery. Do leave some feedback if you have any suggestions, or if you just find it useful.

If you want to take a look at source code that uses this behavior, head to the Scrum Sprint Monitor project at CodePlex, and search for Behaviors:BusyIndicatorBehavior.

This post is sort of a future reference for myself. I needed to animate the Visibility property as part of a larger Storyboard, and it wasn’t immediately apparent how to do it. It took a bit of effort to find the solution on Google, so I am posting the solution in hope that it will make someone else’s life easier as well.

If you have something like the following, for an object that is usually hidden:

<Setter Property="Visibility" Value="Visible" />

and you want to fade in the object instead of suddenly showing it, you will need to convert the Setter into an Animation action in a Storyboard. You are probably aware of the DoubleAnimation and BooleanAnimationUsingKeyFrames classes already. To animate the Visibility property, however – which is an enum –, the solution is not immediately apparent: you need to use ObjectAnimationUsingKeyFrames:

<Storyboard>
    <ObjectAnimationUsingKeyFrames Storyboard.TargetProperty="Visibility">
        <DiscreteObjectKeyFrame KeyTime="00:00:00">
            <DiscreteObjectKeyFrame.Value>
                <Visibility>Visible</Visibility>
            </DiscreteObjectKeyFrame.Value>
        </DiscreteObjectKeyFrame>
    </ObjectAnimationUsingKeyFrames>

    <DoubleAnimation Duration="00:00:01" Storyboard.TargetProperty="Opacity" To="1" />
</Storyboard>

There you go. The animation will start by making the object visible, and then animating it to 100% opaque. Hope this helps you!

As I was going through the UI of the Scrum Sprint Monitor to introduce animations to the larger UI elements, one type of property was proving difficult to animate: the dynamic background color of some elements which were bound to Brush (or Color) properties of the ViewModel.

Picture this: you have a Color property that is currently Green. At some point it changes to Red. You would like the UI to update smothly, so you would not see the color jump from Green to Red. You are not able to do this in XAML with triggers and storyboards directly. The problem lies in the fact is that the Animation objects in XAML cannot be bound to dynamic values (see instances here and here), as they must be frozen for performance and thread safety reasons.

<DataTemplate DataType="{x:Type ViewModels:BuildStatusBackgroundViewModel}">
    <Grid Background="{Binding Path=BuildStatusColor}" />
</DataTemplate>

Therefore there is no point in trying to introduce triggers (be it EventTriggers or DataTriggers) in this Grid element. What I ended up doing is deriving a class from Grid and spawning the animation from code behind.

<DataTemplate DataType="{x:Type ViewModels:BuildStatusBackgroundViewModel}">
    <Controls:BuildStatusBorder />
</DataTemplate>

See the code-behind example here. I did encounter a pitfall while developing this approach. I ran into a “Cannot freeze this Storyboard timeline tree for use across threads” exception, when the animation was kicked off. I fixed this by cloning the existing property value, essentially removing any data binding from the property.

Hope this helps someone who is running into the same problem of animating transitions between databound properties!

UPDATE: The same functionality is now available out of the box using the Blend 3 SDK, through the FluidMoveBehavior. Just drag that behavior to your container panel, and set the Duration and AppliesTo properties! The only difference is that easing functions are not yet supported in WPF. We will need to wait until WPF 4 is released.

Last evening I finally implemented a long awaited feature (by me!) in the Scrum Sprint Monitor: animated WPF panels. As the team members in the Sprint move up and down the list (an ObservableCollection<>), get added or removed, I always wished that change could be animated. My current knowledge of the WPF layout mechanism wasn’t sufficient to finish on that endeavor within a few hours, though.

I finally found a blog post that set me on the right path, on Ed Foh’s blog. Ed in turn was inspired by Kevin Moore’s WPF Bag of Tricks, which also includes an Animating Tile Panel.

Here is a video demonstrating the enhanced behavior of my AnimatedUniformGrid:

Why those two solutions didn’t work for me

Both of the aforementioned solutions contained hardwired positioning logic, though. I simply needed to add the animation behavior to StackPanel, WrapPanel and UniformGrid, not a completely custom panel. Ideally, the solution would be a behavior that could be added on top of those containers. That particular aspect wasn’t realized, and I ended up deriving classes for each of these panels, prefixing the new classes with “Animated”. It is still a pretty acceptable solution. UPDATE: I eventually found this was the same approach taken by at least one commercial offer.

The following is the code required to instantiate the uniform grid (note how you simple need to prefix the panel class with Wpf:Animated):

<ItemsControl ItemsSource="{Binding TeamMembersIncludingUnassigned}">
    <ItemsControl.ItemsPanel>
      <ItemsPanelTemplate>
         <Wpf:AnimatedUniformGrid Duration="00:00:01" />
      </ItemsPanelTemplate>
   </ItemsControl.ItemsPanel>
</ItemsControl>

Adding animation to an existing panel control class

The workflow for converting an existing Panel to an AnimatedPanel is pretty easy:

1. derive a new class from the desired panel class (prefix it with Animated);
2. override the ArrangeOverride method in the new class (don’t call the base class implementation);
3. using .NET Reflector, extract the ArrangeOverride method contents from the base class implementation, and simply substitute the element.Arrange call that is performed for each child for a call to the AnimatedPanelHelper.ArrangeChild() static method.

That is all you need to do!

Acknowledgements

Thanks to Ed Foh and Kevin Moore for their great work.

Download the source code here.

Recently I needed to replace a UniformGrid control where the number of columns was known at design time, with something equivalent that would be bound to a Prism region (update: if you are unfamiliar with Prism, the RegionName attached property will basically pull the items into the ItemsSource property). The goal was for each cell to hold a view, and to grow horizontally as more views were added. For some reason, at the time was under the impression that Grid and UniformGrid did not support ItemsControl, so I started with a StackPanel with horizontal Orientation, but of course, that doesn’t work well when you have a narrow display area. I then thought of a couple of approaches: either making my own RegionAdapter for ItemsControl, or deriving from ItemsControl. Those seemed more complex than required, so after looking long and hard on the web for a way to automatically set the Grid.Column attached property based on the number of views assigned, I realized that UniformGrid does support ItemsControl. All I had to do was bind the UniformGrid.Columns property to the ItemsControl.ItemsSource.Count property, and voilá! I got myself a dynamically sized UniformGrid:

<ItemsControl cal:RegionManager.RegionName="{x:Static
    Extensibility:RegionNames.SprintOverviewChartsRegion}">
    <ItemsControl.ItemsPanel>
        <ItemsPanelTemplate>
            <UniformGrid Columns="{Binding Path=ItemsSource.Count,
                RelativeSource={RelativeSource FindAncestor,
                AncestorType={x:Type ItemsControl}}}"
                IsItemsHost="True" />
        </ItemsPanelTemplate>
    </ItemsControl.ItemsPanel>
</ItemsControl>

For the last couple of days, I have been fighting with the Composite Application Guidance for WPF (aka Prism V2) after deciding to port an experimental version of the Scrum Sprint Monitor from Prism V1. It turns out that the ninject-contrib project where I got the Ninject bootstrapper from is still using an old version of the guidance. Long story short, it turned out that most of the problems I experienced stemmed from the fact that the bootstrapper has been significantly changed in V2, and I had not been able to find an article detailing the new layout of the bootstrapper, such as this one did for V1. The process of moving to the current version turned out somewhat painful, and I am documenting it here so that hopefully others can have an easier time going through the same process.

Updated NinjectContrib.CompositeWpf

I am attaching an updated version of the NinjectContrib.CompositeWpf project, now renamed to NinjectContrib.CompositePresentation so as to follow Prism’s naming scheme. You can drop that in as a replacement to the current version when migrating to V2.

Back to the problem at hand

I started seeing this error a lot:

"This RegionManager does not contain a Region with the name …"

Apparently, Prism V1 demanded regions to be registered before a module could register a view in it.

I set out to debug the problem. I banged my head around the DelayedRegionCreationBehavior class, wondering why the region was not appearing in the RegionManager – even though I could see it being created in response to the attached property declaration. I scoured the web for help, but no one seemed to be having the same issue (or they weren’t posting about it). Finally, on the second day I came across the UnityBootstrapper using .NET Reflector, and started seeing light at the end of the tunnel.

It turns out that some new methods and services were added in V2, missing from ninject-contrib’s implementation:

One of them adds some default behaviors:

/// <summary>
/// Configures the <see cref="IRegionBehaviorFactory"/>. This will be the list
/// of default behaviors that will be added to a region.
/// </summary>
protected virtual IRegionBehaviorFactory ConfigureDefaultRegionBehaviors()
{
    try
    {
        var defaultRegionBehaviorTypesDictionary =
            Kernel.Get<IRegionBehaviorFactory>();
        if (defaultRegionBehaviorTypesDictionary != null)
        {
            defaultRegionBehaviorTypesDictionary.AddIfMissing(
                "AutoPopulate", typeof(AutoPopulateRegionBehavior));
            defaultRegionBehaviorTypesDictionary.AddIfMissing(
                "ContextToDependencyObject",
                typeof(BindRegionContextToDependencyObjectBehavior));
            defaultRegionBehaviorTypesDictionary.AddIfMissing(
                "ActiveAware", typeof(RegionActiveAwareBehavior));
            defaultRegionBehaviorTypesDictionary.AddIfMissing(
                SyncRegionContextWithHostBehavior.BehaviorKey,
                typeof(SyncRegionContextWithHostBehavior));
            defaultRegionBehaviorTypesDictionary.AddIfMissing(
                RegionManagerRegistrationBehavior.BehaviorKey,
                typeof(RegionManagerRegistrationBehavior));
        }
        return defaultRegionBehaviorTypesDictionary;
    }
    catch (ActivationException)
    {
        return null;
    }
}

The other is responsible for registering any Ninject exception types considered to be root exceptions:

/// <summary>
/// Registers in the <see cref="IKernel"/> the <see cref="Type"/>
/// of the Exceptions that are not considered root exceptions
/// by the <see cref="ExceptionExtensions"/>.
/// </summary>
protected virtual void RegisterFrameworkExceptionTypes()
{
    ExceptionExtensions.RegisterFrameworkExceptionType(
        typeof(
            Microsoft.Practices.ServiceLocation.ActivationException
            ));
    ExceptionExtensions.RegisterFrameworkExceptionType(
        typeof(ActivationException));
}

There have also been changes in namespaces and interface names, all of which are detailed in this article. Hopefully by using the attached zip file, you will be able to sidestep these issues, and have a more painless transition to V2. I will try to get this code into the ninject-contrib project, once I find out how to do that.

Enjoy!

Now that we have our mock data showing up in Expression Blend, let’s get it working under Cider (the VS 2008 WPF designer). If you are really lucky, you might have hit all the right notes necessary to get it working the first time. If you fall under the other 99%, then follow the tips below:

Cider does not like the Window control for design-time data binding…

That’s right. This one actually took me the longest to figure out. It only hit me because I had one UserControl working correctly, so I started with a stripped down Window to determine the root cause. Turns out, if I change the Window to a UserControl, Cider is happy to do the data binding. So there are two ways you could get around this limitation:

• Create an intermediate UserControl, and host that inside your Window; or
• Add the DataContext to the root container element in your Window (e.g. a Grid). That seems to sidestep the Window data binding problem.

Put UserControl.DataContext after UserControl.Resources! … And,  declare the serviceProvider on the UserControl.Resources too!

I was getting erratic behavior hosting different controls from within the same project in my main Window. It ended up originating from a missing serviceProvider declaration. The hosted control is not able to find my App.xaml declarations, apparently. Therefore, if you host your UserControl Foo inside Control Bar, the designer will cough out the error "Could not create an instance of type ‘Foo" when you open Bar in the designer. Fortunately, you are able to define it in both App.xaml and on the UserControl, and it will not create problems in both Cider or Blend.

Sometimes, Cider calls IValueConverters with strange data

I have had Cider call a ValueConverter with a string, when I was expecting a business object instance. This caused the designer to catch an exception. The solution was to ignore unexpected types when in design mode (and this is actually suggested in a Microsoft Cide blog). Easy enough.

Now, for the final step…

Even though the previous steps should get the mock data showing up in Cider as well, you still have a problem: the design data will show up at run-time as well. There is an easy way to fix this:

public TeamMemberDataUserControl()
{
    InitializeComponent();

    if (!DesignerProperties.GetIsInDesignMode(this))
    {
        // Clear the design time data context
        ClearValue(DataContextProperty);
    }
}

You just clear the DataContext in the constructor of your control. That will execute at run-time, and reset the DataContext as if the XAML declaration wasn’t there.

As Eduardo pointed out to me, this has the side benefit of fixing another issue, as you will see in a bit. There are two different situations where you can see a control in design-mode. The simplest one is when you open the control file (let’s call it A.xaml) directly. The other one is when you open a control (B.xaml) which hosts A.xaml. In the latter situation, the designer will pass in the DataContext to the hosted control, but only if it does not already have a DataContext assigned already. Clearing the DataContext property in the constructor seems to do the trick. Eduardo also worked in a cleaner solution using attached properties, so he might blog a bit about that approach.

Wrap up

By following the steps in these two posts, you should be able to get your application showing design time data both in Expression Blend and Cider. You should not only be able to get that data in simple UserControls, but also Windows that nest other UserControls. If you would like to see the source code for an application that puts this into practice, take a look at the Scrum Sprint Monitor source code at CodePlex. Hope this is useful to you!

Recently, I have been trying to populate my WPF design-time controls with mock data in order to have real-time feedback of the UI, as I am tweaking the XAML. I first saw this in action at a session with Richard Griffin, Felix Corke, and Ian Griffiths at Tech Ed Barcelona 2008, and I was blown away by the prospects of it. This has been the Holy Grail of form design for me, ever since the first releases of Visual Studio came out (think Win32, MFC, Windows Forms), and now it seemed Microsoft nailed it! The gang even got this working with an IoC container! This was a Silverlight project though, and not WPF, so I wasn’t sure how easy it would be to transition the same experience.

After a few tries, and looking at articles like this one and this one, it seemed that the road was going to be a bumpy one. Even though I got it working with Expression Blend initially, it seemed much harder to get it to work consistently under the Visual Studio 2008 designer (Cider). In this blog post, I will outline the gotchas I have found along the way to a solution that works both in Expression Blend 2 SP1 and Visual Studio 2008 SP1, without modification:

The solution I am using involves creating a ServiceLocator object, which exposes my ViewModels as properties:

public class ServiceLocator
{
    static ServiceLocator()
    {
        SingletonContainer.Setup(new NinjectContainerAbstraction());
    }

    public IMainViewViewModel MainWindowViewModel
    {
        get { return SingletonContainer.Resolve<IMainViewViewModel>(); }
    }
}

After adding it to my App.xaml file:

<Application.Resources>
  <local:ServiceLocator x:Key="serviceLocator" />
</Application.Resources>

… I can then bind to a ViewModel like this:

<UserControl.DataContext>
  <Binding Path="MainWindowViewModel" Source="{StaticResource serviceLocator}"/></UserControl.DataContext>

I am then able to start using the ViewModel directly on the XAML:

<Label Content="{Binding Path=CurrentSprintName}">

The serviceLocator will be configured with the design-time mock data, or the run-time data, depending on whether we’re hosted inside a designer. This check is accomplished through DesignerProperties.GetIsInDesignMode. However, as noted in Delay’s Blog, there is a bug in that implementation, so I’m using his recommended workaround.

This should be enough to get the data to show up in Blend. And this is what a lot of articles floating on the Internet will tell you. However, as you will find in part II of this installment, VS2008 is much pickier in what it decides to chew up.