A few days ago I discovered a very interesting project. Ryan Corradini has a project named Canvassa that is a HTML5/Canvas based Zelda clone. He does more or less the same thing I do with SilverTile but his approach is quite different. His blog is quite interesting and especially the posts about reverse engineering the game map based on a huge picture was very interesting.

Canvassa

Inspired by Ryans posts I wrote a simple C# program that would parse the entire map for The Legend of Zelda (I found the map here) and spit out tile images and Xaml code that SilverTile could utilize. (More on this in a later post - it was quite fun actually!)

The entire game world - click the image for the full 4096 pixel wide version

With this in place I think SilverTile is ready to be tested by a broader audience than me and a couple of friends. So that is where you dear reader come in.

I need you to test out two versions of SilverTile, one running under Silverlight 4 beta and one running under Silverlight 3. If you only have Silverlight 3 installed then just run that one. :-) What I am after is feedback on wheter I need to optimize the game engine or not. In its current state the engine is full of naive implementations that could probably be easily optimized, but I don't want to spend unnecessary time on it. I am also interested in getting feedback on the game engine itself. Does it work? Is it playable? What should I change?

Monsters! This is what awaits you if you move left from the start area

Once the game is running you can control your character using the arrow keys. Move to the open edges on the screen to visit other areas of the game world. If you move to the left you will face hordes of enemies and if you move up you will face a few enemies. All other areas are empty, and you cannot fight the monsters at the moment. If you die just reload the page to start again.

What I want you to do is the following:

• First move around the screen and see if you can control the player sprite properly.
• Then move up to the next area and see if the monsters run around more or less randomly. If you run into a monster your life (the hearts) should decrease.
• Then move back down and to the left. Here you will meet a huge number of monsters. Watch the monster counter and let me know how many monsters there are when the game starts to feel sluggish. Aslo watch the frame rate counter in the bottom of your browser window. It should be around 38-40. How long can you survive before the monsters eat you?
• Finally explore the map a little and see how this works. Does the game engine crash or behave strangely?

When you are done testing I want you to leave a comment to let me know what kind of computer you are running and what your results were. I am particularily interested in hearing how many monsters the game engine could handle and if you experience anything strange. Which operating system and browser you are using is also interesting.

Click here for the Silverlight 4 version and here for the Silerlight 3 version.

Thanks in advance for helping!

.NET, Silverlight .net, silverlight, silvertile

It's time for another project update.

I just uploaded a new version of the SilverTile sample application. The new version has four areas that you can move between freely and lots of monsters you can hunt. The monsters have different AI, the blue ones will try to hunt you while the red ones will move randomly.

Lots of dangerous monsters!

Since my last update I have added some features to the game engine:

• A simple keyboard input handler
• Improved area transitions
• Improved collision detection
• Layering of sprites (z-index)
• Improvements to the monster spawning process

All the changes I have done are based on the needs of my sample game. A more structured keyboard input was needed as the complexity of the game grew.

Area transitions needed to be directional so that you can transition into a tile with an area transition back without being stuck in an endless loop.

I found some bugs in the collision detection algorithm that I fixed (Yay for TDD!), but my implementation is really naive. I would be grateful if anyone could help with a better algorithm. :-)

Layering of sprites was something I really wanted the engine to have, but I didn't need it until I was to render the players sword. The sword is drawn underneath the player sprite when the attack animation starts, and I needed to make sure the sword wasn't draw on top of the player.

The monster spawning process was modified so that monsters can be spawned immediately when a level is loaded. The classic The Legend of Zelda spawns enemies this way, so I need to support it.

All game levels were drawn using the editor application I wrote. The application is really simple, but it allows me to draw the game levels and export them as Xaml easily, so I am happy with it.

Whats next

With a more complete game engine I think it is time to build a better user interface for the game. I have added hit points to the player, but I need to display them as hearts as in the classic Zelda.

I also need to build more game levels and start adding items to the game. Monsters should drop coins that you can use to buy stuff.

And then I need to add text rendering and an intro screen and a game over screen!

... So there is still lots of work to do.

.NET, Silverlight .net, silverlight, silvertile

A few days ago I blogged about a tiny fun project of mine, where I added dynamic type support to the DataTable and DataRow classes.

This was fun, and the code works as intended, but it has a two shortcomings: There was a bug preventing you from setting values of regular properties on the DynamicDataRow type, and you couldn't add new columns using dynamic properties.

Naturally, I had to fix this, so here it is: The DynamicDataTable v2!

What is new?

Most of the code is the same as last time, but this time the useage is even simpler. You just create a new DynamicDataTable object, add a new row using the NewRow method and assign values to the properties you need. Look at this:

// Create a DataTable
var tb = new DynamicDataTable();

// Row using dynamic syntax
dynamic row = tb.NewRow();
row.aa = "Dynamic here!";
row.bb = 44;
row.cc = DateTime.Now; 
tb.Rows.Add(row);

// Access column values using property accessors instead of using the indexer
Console.WriteLine("AA: " + row.aa);
Console.WriteLine("BB: " + row.bb);
Console.WriteLine("CC: " + row.cc);

If you compare this with the previous version you see that I have removed the column definitions. Now when the DynamicDataRow intercepts a call to a property set for a property that doesn't exist it will check its DataTable to see if there is a column with that name. If it can't find a column with that name it will add a new column and set its datatype to the type of the assigned value. In the above example our table ends up with three columns: "aa" of type String, "bb" of type Int32 and "cc" of type DateTime. Later you can access the DataTable as a regular table if you want.

The updated source code

I have uploaded an updated version of the source code if you are interested: DynamicDataTable.zip (4.57 kb)

.NET .net

This is just for fun, so don't use this at work! :-)

Some day ago we had a discussion on Twitter about using the DataTable and DataSet types in .Net. We talked about the differences between typed and untyped DataSets and the merits of both. 

Personally I don't like typed DataSets, not only do I see them as unnecessary, but the DataSet Designer in Visual Studio is terribly buggy and unstable. The one place where typed DataSets have an advantage is in code readability - using property names is easier to read than accessing the DataRow using the indexer property.

Having just read about the dynamic type support in C#4 I got the idea that we could use it to get the pretty syntax of the typed DataSets without all their extra annoyances. Not that it would be very useful, but it would be a good chance to get to know dynamic typing better while writing some fun code. After all, this combines the "old" techonology of the DataTables with the new fancy technology of dynamic types. What could be better?

What I would like to do

Since this is just for fun, I don't care about writing a very robust implementation. What I want is a way to do the following:

// Create a DataTable object
var tb = new DynamicDataTable();

// Add some columns using old-style syntax
tb.Columns.Add("aa", typeof(string));
tb.Columns.Add("bb", typeof(Int32));

// Add a row using dynamic syntax
dynamic row = tb.NewRow();
row.aa = "Dynamic here!";
row.bb = 42;
tb.Rows.Add(row)

// Access column values using property accessors instead of using the indexer
Console.WriteLine("AA: " + row.aa);
Console.WriteLine("BB: " + row.bb);

Note that this syntax is somewhere between the regular DataTable syntax and the syntax for typed DataTables.

Implementing it

I implemented the DynamicDataTable using three classes: The DynamicDataTable itself, a new DynamicRowsCollection that replaces the DataRowCollection of regular DataTables and the DynamicDataRow that extends the DataRow type with dynamic capabilities.

The DynamicDataTable simply wraps a DataTable and overrides or overwrites a few methods. By overriding the NewRowFromBuilder, GetRowType methods and then overwriting the NewRow method I can force the DynamicDataTable to return my DynamicDataRow objects. In addition I overwrite the Rows collection with my own DynamicRowsCollection type to ensure that we only store DynamicDataRow objects.

public class DynamicDataTable : DataTable
{

  private DynamicRowsCollection _rows;

  protected override DataRow NewRowFromBuilder(DataRowBuilder builder)
  {
    return new DynamicDataRow(builder);
  }

  protected override Type GetRowType()
  {
    return typeof(DynamicDataRow);
  }

  public new DynamicRowsCollection Rows
  { 
    get
    {
      if (_rows == null)
      {
        _rows = new DynamicRowsCollection(base.Rows);
      }
      return _rows;
    }
  }

  public new DynamicDataRow NewRow()
  {
    return (DynamicDataRow)base.NewRow();
  }

}

I then wrote a DynamicRowsCollection type as a replacement for the old RowCollection. It implements the IList<DynamicDataRow> interface but all it does is to reroute all method calls to the contained RowCollection:

public class DynamicRowsCollection : IList<DynamicDataRow>
{

  private DataRowCollection _rows;

  public DynamicRowsCollection(DataRowCollection rows
  {
    _rows = rows;
  }

  #region IList implementation

  // Snipped lots of dumb code...

Finally I implemented the DynamicDataRow. This is my specialized DataRow that reroutes all attemps at getting or setting properties to access the indexer instead. I hit upon a challenge here where I first rerouted all property calls, also those that were to existing properties. This meant that trying to access the DataRow.Table property ended up in an attemt to access a column named Table. To fix this I added a check of whether there already exists a property with the given name. I should also add the same check when setting property values, but I haven't done so yet.

public class DynamicDataRow : System.Data.DataRow, IDynamicMetaObjectProvider
{

  public DynamicDataRow(DataRowBuilder builder) : base(builder)
  { }

  public DynamicMetaObject GetMetaObject(Linq.Expressions.Expression parameter)
  {
    return new DataRowDynamicMetaObject(parameter, this);
  }

  private class DataRowDynamicMetaObject : DynamicMetaObject
  {

    private DataRow _row;

    private DataTable _table;

    public DataRowDynamicMetaObject(Expression parameter, DataRow row)
      : base(parameter, BindingRestrictions.Empty, row)
    {
      _row = row;
      _table = row.Table;
    }

    public override DynamicMetaObject BindGetMember(GetMemberBinder binder)
    {

      // Look for an existing property with this name
      var existingProperty = typeof(DataRow).GetProperty(binder.Name);

      if (existingProperty != null)
      {
        // Call the property directly
        var callProperty = Expression.Property(Expression.Constant(_row),  existingProperty);
        var callPropertyConverted = Expression.Convert(callProperty, binder.ReturnType);
        return new DynamicMetaObject(callPropertyConverted, BindingRestrictions.GetInstanceRestriction(Expression, _row), _row);
      }
      else
      {
        // Try to find a column with the name of the property
        var indexer = typeof(DataRow).GetProperty("Item", new Type[] { typeof(string) });
        var getRow = Expression.Constant(_row);
        var colName = Expression.Constant(binder.Name);
        var getIndexerValue = Expression.Property(getRow, indexer, colName);
        return new DynamicMetaObject(getIndexerValue, BindingRestrictions.GetInstanceRestriction(Expression, _row), _row);
      }

      public override DynamicMetaObject BindSetMember(SetMemberBinder binder, DynamicMetaObject value)
      {

        var indexer = typeof(DataRow).GetProperty("Item", new Type[] { typeof(string) });
        var getRow = Expression.Constant(_row);
        var colName = Expression.Constant(binder.Name);
        var setIndexerValue = Expression.Property(getRow, indexer, colName);
        var newValue = Expression.Constant(value.Value, typeof(object));
        var assignColValue = Expression.Assign(setIndexerValue, newValue);
        return new DynamicMetaObject(assignColValue, BindingRestrictions.GetInstanceRestriction(Expression, _row), _row);
      }
    }
  }

The first thing to note here is that the DynamicDataRow itself is really simple. All it does is to inherit from the DataRow and add an implementation of the IDynamicMetaObjectProvider interface. This interface has only one method named GetMetaObject. My implementation of GetMetaObject is also simple - all it does is to return an instance of a DataRowDynamicMetaObject.

The DataRowDynamicMetaObject is more complex. It implements the BindGetMember and BindSetMember methods that handles all property calls for the dynamic object.

The implementation of BindGetMember will first check if there exists a property with the requested name. This is done by checking the return value of the Type.GetProperty method. If a property exists we just call itand return the value. If the property doesn't exist, we first get a reference to the indexer property of the DataRow. We then call the indexer passing it the name of the property/column we ask for and return the value.

The implementation of BindSetMember is somewhat easier since we don't care about existing properties here. The code is similar to that for BindGetMember but we end up with an Assign expression where we assign the new value to the column.

Possible improvements

My code is not at all meant to be used in a real-life scenario, so I haven't cared about performance of error checking at all. The most obvious improvement would be to cache the generated DynamicMetaObject objects for each property. This could be done using a simple Dictionary shared between all instances of the DataRowDynamicMetaObject type. In addition we should fix the BindSetMember method to check if the property exists on the DataRow type. On top of that we should add some error checking code to make sure the types we assign to the columns are compatible with the columns data type.

Another nice improvement would be to add new columns to the DataTable the first time we assigned a value to a property. This would save us the calls to DataTable.Columns.Add that defines the table.

Conclusion

The dynamic type support in .Net 4 is really nice and opens up some really nice possibilities. It also enables some terrible hacks like the DynamicDataTable.

Feel free to download the source code if you want: DynamicDataTable.zip (4.46 kb)

.NET .net

This is something I had to look up several times myself, so I thought I should write it down for later reference.

I was building a simple level editor for SilverTile (more about that in another post later) and wanted to place the map tiles in a grid-like fashion. This could be done using my dynamic Grid control, but I wanted pixel perfect control of where the tiles were placed on screen. Note the gaps between the grid rows in the following screen shot.

Fail! Note the gaps between the tiles!

To do this I would need to use a Canvas control and place each tile using the Canvas.Left and Canvas.Top properties to set the position.

Win! This is what we want. Pixel perfect placement of our images!

The challenge is that the Canvas control doesn’t support list binding directly, so you must use some other control to accomplish this. Luckily there is a solution available by using the ItemsControl class. The ItemsControl is a list control that lets you specify which control should be the root element. In our case we will use a Canvas control as the root and then use data binding to specify the values for the Canvas.Left and Canvas.Top properties.

Using the ItemsControl

To exemplify the usage of the ItemsControl I first created a simple view model for the application. This was simply a list of TileData objects where each describes one tile. Each TileData object has an image source and Left and Top properties. The Left and Top properties tell where the tile should be placed on screen. In a real application your view model would probably be more complex and possibly calculate the positions based on some other data.

// The TileData class describes a tile
public class TileData
{
    public Uri Image { get; set; }
    public double Left { get; set; }
    public double Top { get; set; }
}

Now, with the view model in place we can setup our application.

<Grid Name="Root" Loaded="Root_Loaded">
    <Grid.Resources>
        <Style x:Key="ContainerStyle">
            <Setter Property="Canvas.Left" Value="{Binding Left}" />
            <Setter Property="Canvas.Top" Value="{Binding Top}" />
        </Style>
    </Grid.Resources>
    <ItemsControl Name="map" ItemsSource="{Binding}" ItemContainerStyle="{StaticResource ContainerStyle}">
        <ItemsControl.ItemsPanel>
            <ItemsPanelTemplate>
                <Canvas Width="{Binding Path=Root.ActualWidth}" Height="{Binding Path=Root.ActualHeight}" />
            </ItemsPanelTemplate>
        </ItemsControl.ItemsPanel>
        <ItemsControl.ItemTemplate>
            <DataTemplate>
                <Image Source="{Binding Image}" />
            </DataTemplate>
        </ItemsControl.ItemTemplate>
    </ItemsControl>
</Grid>

Here we have a Grid as the root element of the application. I’ve used a grid since it will be sized dynamically when the window is resized, something the Canvas control doesn’t. We then have the ItemsControl that has its ItemsSource bound to our data context. The ItemsSource is the list of TileData objects that is used to describe the map.

And now we get to the interesting part: The ItemsControl.ItemsPanel property. This property defines which control should be the container into which all our list items are placed. By setting this to a Canvas control we end up with a layout where you can set your child controls position as you want.

Further we have the ItemsControl.ItemsTemplate that tells how you want each object in your ItemsSource to be rendered. In our case we render it as an Image control with its source bound to the Image property of your TileData object.

The final missing part is how we place our controls. The obvious way to do this would be to set the Canvas.Left and Canvas.Top properties of our Image control by binding them to the Left and Top properties of our TileData object. One would expect this to work but if you try it you will see that all our controls are placed at the top left corner of the canvas, with Canvas.Left = 0 and Canvas.Top = 0. The solution is to create a style that sets the Canvas.Left and Canvas.Top properties and to assign that to the ItemContainerStyle property of the ItemsContainer. By doing this you will see that our Image controls are placed correctly where we want them.

Using Snoop to find out what really happened here

To understand what really happened here we need to examine the control hierarchy our Xaml code generates. A good tool to do this is Snoop. It lets you attach to a running Wpf application to examine all properties of all generated controls.

Using Snoop to look at our applications control hierarchy

Look at the above screen shot. Here you can see the control hierarchy of the GridWin application. If you look down the tree you will find an ItemsControl control with a child Border control. The Border contains an ItemsPresenter Control that contains a Canvas. This Canvas control is the one we assigned to our ItemsControl.ItemsPanel property.

The Canvas contains a bunch of ContentPresenter controls that each contains one Image control. And this is the reason why we cannot set the Canvas.Left and Canvas.Top properties of our Image controls directly. Since the images are contained in the ContentPresenter controls we need to set the properties here instead. And this is what we do when we set the ItemContainerStyle property. The style is applied to the ContentPresenter which is placed at the correct position.

Using the code

I am using this approach in a level editor application for SilverTile that I will be writing about soon. The application stores the map as a list of Tile objects where each Tile knows which row and column they should be placed in. The Canvas.Left and Top properties are then bound to the X and Y properties that are caluclated by multiplying the row or column number with the tiles width or height in pixels. The tile list is used as the ItemsSource for an ItemsControl that renders the tiles at the wanted position.

The ItemsControl is available in Silverlight as well as in WPF, so this approach works just as well there. The only challenge is that the binding of the Canvas' size to the size of the Root grid cannot be done in Silverlight 3 since it doesn't support element to element binding. This is fixed in Silverlight 4 though, so there you should be able to use the exact same approach.

.NET, Silverlight, WPF .net, silverlight, wpf

It's time I posted another progress report.

Since my last progress report I have implemented most of the features from my feature list. I'm not entirely happy about the Xaml support for level design, and I am still missing the support for per-pixel collision detection, but the game engine works well enough that I have implemented a VERY simple game using it. 

The game, available here, lets you control your character on a field throwing swords at some monsters. You get points for how many monsters you kill and if you are touched by a monster you die. 

Hunting monsters

As simple as it is the game demonstrates the features needed for a proper game:

• We have collision detection - both between moving sprites (character, monster, sword) and between sprites and background (you cannot walk across mountains). 
• We have animation support and an animated sprite can have many animations that we can choose.
• The game area and all game assets are defined in Xaml. 
• Graphics are reused between sprites. This means that two identical monsters share the same bitmaps in memory.

The engine is still missing some features. The most obvious one is the pixel perfect collision detection. This is a simple feature to add, but I fear that it will be a real performance killer so I postpone it until I really need it. Apart from that I feel that the framework surrounding game setup needs some work.

The current implementation lets you define one or many GameAreaGrid objects. The GameAreaGrid lets you place one rectangular sprite in each cell. You then tell the Viewport to load the grid and it will be setup as the background area of your game. To change maps you tell the viewport to load another GameAreaGrid object.

While this works well enough I would like it to be even simpler to use.

The idea is to add support for area transitions to cells in the GameAreaGrid so that when the player enters one of these we automatically loads the new grid and places the player at the new location. This should allow us to setup an entire game world using only Xaml. I'm not entirely sure what would be the best way to define an area transition but I think it would be nice to add an AreaTransitions collection property to the GameAreaGrid where each object in the collection described a transition that would be performed when the player enters a given cell. A transition would be from a cell (grid row and column) to a new cell in another GameAreaGrid. The reference to the GameAreaGrid would be a regular object reference that could be modeled as a binding to a StaticResource in Xaml. This still gives us the option of designing the GameAreaGrid in Xaml or in code if we prefer. 

I would also like to allow random placement of enemies in a GameAreaGrid. Either completely random or by identifying cells where mosnters can spawn. The types of monsters, the number of monsters and how often they spawn would also have to be specified.

So that will be my plan for the week until christmas. Implementing more support for game mechanics, paving the way for a complete game engine :-)

.NET, Silverlight .net, silverlight, silvertile

It's time for a progress report.

The last week I have been adding basic features to SilverTile. The first feature was to add basic rendering support. My first attempt used Array-copy functions to copy pixel data from the sprite to the viewport. This worked well, but it did not support alpha blending (transparency) of pixel data. To improve this I added some code from the Silverlight Games 101 blog that allowed rendering of bitmap data from one WriteableBitmap to another. I modified the code somewhat to support rendering of data from any IRenderable and then I had the rendering support I needed.

The intial architecture of SilverTile defined each sprite object as having its own pixel data. This is fine as long as you have only a few sprites or if each sprite have an unique appearance, but in most cases you will have many sprites using the same image. One example is the walls of a game area. I model each wall segment as an unique sprite, but each is built up of the same image. So instead of having each sprite have its own pixel data I refactored out the pixel data into a Frame object that could be reused between sprites. This also lay the ground for animation support later.

Link walking!

To set up a game area I want to support two modes. One mode should be building game areas in code by using the AddBackgroundSprite() method of the Viewport class. This lets you add one sprite at a time at the desired location. While building the game area in code works well, I wanted to support using Xaml to describe the game area and the graphical resources used by a game. Since Silverlight only allows classes that inherit from UIElement to be named (given the x:Name and x:key attributes) in Xaml I had a choice of either letting the Sprite classes derive from UIElement or using a separate set of classes to describe the game area. I cholse the latter option and built a set of classes that can be instansiated in Xaml and that describes a game area, a game tile and a frame. I then added a SetGameArea helper method to the viewport that reads the GameArea data and performs the setup.

Finally I added support for animated sprites. An animation is a list of IFrame objects in combination with a frame rate. Each time the Update method of the animated sprite is called, we check if the elapsed time is greater than the time a frame should last and if it is we switch frames to the next frame. When we reach the last frame in the animation we go back to the first frame again. Although I will need for sprites to support multiple animations (like walk left and walk right), this is a good starting point.

On top of all this I have written tests for everything. This really helped me when refactoring the code to use IFrame and the bitmap blitting.

If you have the Silverlight 4 beta installed you can watch a live demo here: http://rag.no/st/silvertile.html. (Try clicking on the game area!) Or you can download the source code here: http://silvertile.codeplex.com/

.NET, Silverlight .net, silverlight, silvertile

I will try to keep SilverTile simple with just the features needed to get my game running.

At the very least SilverTile will support the following:

• Rendering to a surface (WriteableBitmap)
  • I think this is the fastest way to render pixel graphics in Silverlight (done)
• Support both tiles for background / terrain, and moveable sprites
  • Sprites should support animation (done)
  • Sprites should support transparent pixels (done)
• XAML for game level setup
  • Support using XAML to setup the game levels/maps (done)
  • SUpport using XAML to define sprites and animated sprites (done)
• Collision detection
  • Tiles should support bounding-box collision detection (done)
  • Sprites should support bounding-box collision detection (done) and possibly per-pixel collision detection (not yet)
• Input control
  • Keyboard input should be handled in a structured way
• Game infrastructure support
  • Support for transitions between levels/maps defined in XAML (done)

In addition I think SilverTile needs a way to handle the game and a game session as an entity, but that can wait until later. This is implemented in the SilverTileGameWorld class.

Any comments? Did I forget anything essential? I will update this page with any new features and with some progress information as soon as I have anything to report.

.NET, Silverlight .net, silverlight, silvertile

The last days I have been playing with WPF trying to get an understanding of how it works ”under the hood”. So I have written a bunch of tiny applications where I prototype a function. When building these applications I don’t care much about its user interface, but I still like to have my controls lined up neatly. To do this I usually use a grid control with some rows and columns. This usually works well, but the grids syntax gets tiresome to write and rewrite all the time.

Yesterday this annoyed me so much that I decided to fix it! The reason was not only the syntax, but also that I needed a grid that let me set the number of rows and columns dynamically using data binding. The result is the DynamicGrid control. It allows you to replace this code:

<Grid>
    <Grid.RowDefinitions>
        <RowDefinition />
        <RowDefinition />
        <RowDefinition />
        <RowDefinition />
        <RowDefinition />
        <RowDefinition />
        <RowDefinition />
        <RowDefinition />
    </Grid.RowDefinitions>
    <Grid.ColumnDefinitions>
        <ColumnDefinition/>
        <ColumnDefinition/>
        <ColumnDefinition/>
        <ColumnDefinition/>
        <ColumnDefinition/>
        <ColumnDefinition/>
        <ColumnDefinition/>
        <ColumnDefinition/>
    </Grid.ColumnDefinitions>
</Grid>

With this code:

<grid:DynamicGrid NumColumns="8" NumRows="8">       
</grid:DynamicGrid>

Nice! And it also allows you to use data binding to setup the number of rows or columns. Like this:

<grid:DynamicGrid NumColumns="{Binding ColumnCount}" NumRows="{Binding RowCount}">       
</grid:DynamicGrid>

So your grid can adapt to your data source. On top of all that, it even plays nice with the old grid-markup. So if you need a header row with a fixed size then you can insert it using a RowDefinition and let the DynamicGrid add the rest of the rows and columns for you automatically.

How does it work?

It turned out that implementing the grid was really easy. All I needed to do was to subclass the Grid class, add two new dependency properties and add the code that added the columns and rows automatically.

I created two new dependency properties named NumColumns and NumRows. They are both identical except for the naming:

public static readonly DependencyProperty NumColumnsProperty =
    DependencyProperty.Register("NumColumns", typeof(Int32), typeof(DynamicGrid));

public Int32 NumColumns
{
    get { return (Int32)GetValue(NumColumnsProperty); }
    set { SetValue(NumColumnsProperty, value); }
}

There is not much to note about them – we default to one column and one row.

I then wrote an override for the OnInitialized method:

protected override void OnInitialized(EventArgs e)
{
    base.OnInitialized(e);

    RecreateGridCells();
}

This will call the RecreateGridCells method that does the actual setup of the dynamic grid. Since it is called after the base grid is initialized we support its functionality.

The RecreateGridCells method does the custom setup:

private void RecreateGridCells()
{
    int numRows = NumRows;
    int currentNumRows = RowDefinitions.Count;

    while (numRows > currentNumRows)
    {
        RowDefinitions.Add(new RowDefinition { Height = new GridLength(1, GridUnitType.Star) });
        currentNumRows++;
    }

    while (numRows < currentNumRows)
    {
        currentNumRows--;
        RowDefinitions.RemoveAt(currentNumRows);
    }

    int numCols = NumColumns;
    int currentNumCols = ColumnDefinitions.Count;

    while (numCols > currentNumCols)
    {
        ColumnDefinitions.Add(new ColumnDefinition{ Width = new GridLength(1, GridUnitType.Star) });
        currentNumCols++;
    }

    while (numCols < currentNumCols)
    {
        currentNumCols--;
        ColumnDefinitions.RemoveAt(currentNumCols);
    }

}

Here we read out the number of rows we want and the number of rows we have. If we need more rows we add them to the RowDefinitions collection and set their heights to *. The * means that they should scale automatically. If we have too many rows we remove them starting with the last one.

After setting up the rows we setup the columns the same way.

That is all there is to it!

So this code:

<Window x:Class="DynamicGridDemo.MainWindow"
        xmlns="http://schemas.microsoft.com/winfx/2006/xaml/presentation"
        xmlns:x="http://schemas.microsoft.com/winfx/2006/xaml"
        xmlns:grid="clr-namespace:DynamicGridDemo"
        Title="MainWindow" Height="350" Width="525">
    <grid:DynamicGrid NumColumns="2" NumRows="2">
        <Button Grid.Column="0" Grid.Row="0" Content="At 0,0" />
        <Button Grid.Column="0" Grid.Row="1" Content="At 0,1" />
        <Button Grid.Column="1" Grid.Row="0" Content="At 1,0" />
        <Button Grid.Column="1" Grid.Row="1" Content="At 1,1" />
    </grid:DynamicGrid>
</Window>

Gives you this result:

Let’s add a header row:

<Window x:Class="DynamicGridDemo.MainWindow"
        xmlns="http://schemas.microsoft.com/winfx/2006/xaml/presentation"
        xmlns:x="http://schemas.microsoft.com/winfx/2006/xaml"
        xmlns:grid="clr-namespace:DynamicGridDemo"
        Title="MainWindow" Height="350" Width="525">
    <grid:DynamicGrid NumColumns="2" NumRows="3">
        <Label Grid.Row="0" Grid.Column="0" Grid.ColumnSpan="2" Content="Header label here at 0,0 and 1,0" FontSize="20" Background="Yellow" />
        <Button Grid.Column="0" Grid.Row="1" Content="At 0,1" />
        <Button Grid.Column="0" Grid.Row="2" Content="At 0,2" />
        <Button Grid.Column="1" Grid.Row="1" Content="At 1,1" />
        <Button Grid.Column="1" Grid.Row="2" Content="At 1,2" />
    </grid:DynamicGrid>
</Window>

Gives this result:

And it even plays nice together with the old syntax. This is useful if you need a header row with a fixed size for example:

<Window x:Class="DynamicGridDemo.MainWindow"
        xmlns="http://schemas.microsoft.com/winfx/2006/xaml/presentation"
        xmlns:x="http://schemas.microsoft.com/winfx/2006/xaml"
        xmlns:grid="clr-namespace:DynamicGridDemo"
        Title="MainWindow" Height="350" Width="525">
    <grid:DynamicGrid NumColumns="2" NumRows="3">
        <Grid.RowDefinitions>
            <RowDefinition Height="40" />
        </Grid.RowDefinitions>
        <Label Grid.Row="0" Grid.Column="0" Grid.ColumnSpan="3" Content="Header label here at 0,0 and 1,0" FontSize="20" Background="Yellow" />
        <Button Grid.Column="0" Grid.Row="1" Content="At 0,1" />
        <Button Grid.Column="0" Grid.Row="2" Content="At 0,2" />
        <Button Grid.Column="1" Grid.Row="1" Content="At 1,1" />
        <Button Grid.Column="1" Grid.Row="2" Content="At 1,2" />
    </grid:DynamicGrid>
</Window>

This gives the following output (note the size of the first row):

I haven't done much testing of the code, so it may include some nasty bugs, but I think it may be useful. Either if you are prototyping an application or if you need a grid that sizes to your content.

Download the source code here: DynamicGridDemo.zip (77.44 kb) (Visual Studio 2010 b2, WPF4 required)

.NET, WPF .net, wpf

In this post I will look at the new Relational Database Service (RDS) from Amazon Web Services (AWS).

This was supposed to be a short post, but ended up being huge! So good luck to anyone who wants to read it...

AWS offers a bunch of services ranging from simple data storage to full virtual servers. The common factor for all of them is that they are controlled using standard web services. This means that you can easily manage them programmatically if you want to. In this article I won’t focus on the management bit, but rather on how and why to use the service.

In all simplicity RDS is MySQL running on a virtual server somewhere in Amazons server farm. The service is a port of MySQL so if you can connect to a regular MySQL database you can connect to RDS as well. You also manage RDS the same way you would manage a regular MySQL server, you can connect to it using the standard tools and create databases and tables as you want.

In my opinion there are two scenarios where you would use RDS: Either you have an application using MySQL already running on a virtual server on Amazon EC2 or you have an application that needs to reach a MySQL database from anywhere on the internet. In the first case you can now host the database on a dedicated system instead of a custom virtual server. This allows you to easily scale the performance of your application by scaling up your database. In the second case you have a desktop application that needs to connect to a database while being connected to the internet. In that case you could host the database on your own server, but hosting it at AWS is guaranteed to offer better availability and much better bandwidth.

There are at least four obvious advantages to using RDS as compared to running your own server with MySQL:

• High availability – AWS guarantees an uptime of at least 99.95%. Since it is hosted in Amazons server farm it will probably be much more available than most services.
• Easy to upgrade – Using the management web services you can easily scale your RDS server using a single api call. This is much easier than even upgrading a virtual server.
• Automatic backups – RDS will automatically backup your database and in addition you can set up your own backups if you need to.
• Available anywhere – It is easy to configure the RDS firewall so you can make the database available in any way you need to.

But there are also some disadvantages to RDS:

• You trust AWS with your data – Since AWS are hosting your database you also trust them with your data. In some scenarios this may be a problem.
• You pay per hour – When using RDS you have to pay per hour your database is running. While the prices are quite reasonable they still accumulate over time. It will be cheaper to run your database on an existing server if you have one.
• Limited bandwidth – Depending on your use this may not be a problem, but running your database on a remote server means that you connection will be limited to the capacity of your internet line.

Getting started with RDS

The first thing you need to get started with RDS is an account with AWS. It is free to sign up for AWS, but you must pay to use their services. Signing up is simple, especially if you have shopped at Amazon.com before. Sign up here: https://aws-portal.amazon.com/gp/aws/developer/registration/index.html

After you have signed up for AWS you need to create a digital certificate that identifies you. This is done in the Your Account -> Security Credentials page. To access AWS you need two keys – an Access Key ID and a Secret Access Key.

With your keys in place you are ready to connect to AWS and the RDS service. I’ll show you an example .NET application that manages your database instances later, but first we will use the management tools supplied by AWS. These tools are a bunch of Java command line applications, so you need the Java runtime installed to use them. The tools can be downloaded here: http://developer.amazonwebservices.com/connect/entry.jspa?externalID=2928&categoryID=294

Setting up the command line tools is relatively easy – just unzip the file and follow the instructions in README.txt. You should setup a text file containing your Access Key ID and Secret Access Key based on the template found in the zip file.

Now we are ready to create a server! First you should make sure everything works properly. Try to run the rds-describe-db-instances command. If the tools are properly set up it should return an empty list. The next command you should try is rds-create-db-instance. This is the command that creates your server. I used the following command:

rds-create-db-instance --db-instance-identifier MyTestServer123 --allocated-storage 5 
--db-instance-class db.m1.small --engine MySQL5.1 --master-username root 
--master-user-password Secret123

And I got the following response:

DBINSTANCE  mytestserver123  db.m1.small  mysql5.1  5  root  creating  1
      SECGROUP  default  active
      PARAMGRP  default.mysql5.1  in-sync

This gave me a small instance of a database server with 5 Gb of disk storage. The instance is named MyTestServer123 and is running MySQL 5.1.

Now if I run the rds-describe-db-instances command again I get the following result:

DBINSTANCE  mytestserver123  db.m1.small  mysql5.1  5  root  creating  1
      SECGROUP  default  active
      PARAMGRP  default.mysql5.1  in-sync

Note that it says creating. It can take a few minutes to setup a new server and you cannot access it before it is ready. Try running the rds-describe-db-instances command after a minute until you get something like this:

DBINSTANCE  mytestserver123  2009-11-04T11:50:21.554Z  db.m1.small  mysql5.1  5  root  
available  mytestserver123.cqpn2yka5vjx.us-east-1.rds.amazonaws.com  3306  us-east-1b  1
      SECGROUP  default  active
      PARAMGRP  default.mysql5.1  in-sync

Now the status should be available. You should also note the text mytestserver123.cqpn2yka5vjx.us-east-1.rds.amazonaws.com. This is the name of your server and you need it when you want to connect later.
Before you can connect to your server you need to open its firewall to let you through. This is done using the command rds-authorize-db-security-group-ingress. You must run it two times. First you must set the security group for your server and then you must allow your ip-address through the firewall to servers in that security group. Use the following commands:

rds-authorize-db-security-group-ingress MyTestServer123 --ec2-security-group-name default

This will place your server in the “default” security group. The default group always exists so it is ok to use here. Then you use the following command to give your ip address access to the server:

rds-authorize-db-security-group-ingress default --cidr-ip 123.124.125.126/24

Here my public ip address is 123.124.125.126. The /24 after the ip address means that all addresses starting with 123.124.125 should be allowed to connect. Read more about the cidr format here: http://en.wikipedia.org/wiki/CIDR.

Now, finally, your server is ready to be used! Try connecting using your favorite tool. I prefer MySQL Query Browser which is part of the MySQL GUI Tools. Download it here: http://dev.mysql.com/downloads/gui-tools/5.0.html.

To connect you must specify the server name you noted earlier. Run rds-describe-db-instances if you forgot it. The user name and password are the ones you specified when you created the server.

Great! Now we have a database server that we can do anything with! Try creating a database and some users and some tables.

After you are done playing with your server you will probably want to remove it. If you don’t you will be billed per hour until you do, so don’t forget this! Use the following command to delete it:

rds-delete-db-instance mytestserver123 --skip-final-snapshot

The --skip-final-snapshot flag tells the program that you don’t need a backup of the server. This is ok as long as you are just experimenting. When deleting a server you get one final chance to stop the process. Answer Y at the following prompt to start the deletion:

    Once you begin deleting this database, it will no longer be able to accept connections.
    Are you sure you want to delete this database? [Ny]y
DBINSTANCE  mytestserver123  2009-11-04T11:50:21.554Z  db.m1.small mysql5.1  5
 root  deleting  us-east-1b  1
      SECGROUP  default  active
      PARAMGRP  default.mysql5.1  in-sync

Now you server will be deleted after a few minutes.

Using the web services from a .Net application

Well, all this has been nice and useful, but not very fun. So I decided to write an application that manages these servers for us. Since I’m a .Net developer that is the tool I will be using.

MySQL has a nice set of ADO.Net compatible classes for download here: http://dev.mysql.com/downloads/connector/net/6.1.html.
And AWS has a nice set of .Net classes that wrap the RDS web service for download here: http://developer.amazonwebservices.com/connect/entry.jspa?externalID=2946&categoryID=293.

I found a bug in the AmazonRDSClient class where the SkipFinalSnapshot argument was incorrectly encoded. This is easy to fix by changing line 1458 in AmazonRDSClient.cs:

// From
parameters.Add("SkipFinalSnapshot", request.SkipFinalSnapshot + "");
// To
parameters.Add("SkipFinalSnapshot", request.SkipFinalSnapshot.ToString().ToLower() + "");

Using these classes I wrote a simple WPF application that connects to RDS and downloads a list of server instances. It also lets you create new instances, delete instances and test that you can connect to an instance.

Download the source code here. The application is a WPF 4 application that needs Visual Studio 2010 to compile.

RDSApp.zip (781.82 kb)

.NET, Databases .net, mysql, cloudcomputing, aws, rds