Edit

BoxPanel, an example custom panel

  • Article

Learn to write code for a custom Panel class, implementing ArrangeOverride and MeasureOverride methods, and using the Children property.

Important APIs: Panel, ArrangeOverride,MeasureOverride

The example code shows a custom panel implementation, but we don't devote a lot of time explaining the layout concepts that influence how you can customize a panel for different layout scenarios. If you want more info about these layout concepts and how they might apply to your particular layout scenario, see XAML custom panels overview.

A panel is an object that provides a layout behavior for child elements it contains, when the XAML layout system runs and your app UI is rendered. You can define custom panels for XAML layout by deriving a custom class from the Panel class. You provide behavior for your panel by overriding the ArrangeOverride and MeasureOverride methods, supplying logic that measures and arranges the child elements. This example derives from Panel. When you start from Panel, ArrangeOverride and MeasureOverride methods don't have a starting behavior. Your code is providing the gateway by which child elements become known to the XAML layout system and get rendered in the UI. So, it's really important that your code accounts for all child elements and follows the patterns the layout system expects.

Your layout scenario

When you define a custom panel, you're defining a layout scenario.

A layout scenario is expressed through:

  • What the panel will do when it has child elements
  • When the panel has constraints on its own space
  • How the logic of the panel determines all the measurements, placement, positions, and sizings that eventually result in a rendered UI layout of children

With that in mind, the BoxPanel shown here is for a particular scenario. In the interest of keeping the code foremost in this example, we won't explain the scenario in detail yet, and instead concentrate on the steps needed and the coding patterns. If you want to know more about the scenario first, skip ahead to "The scenario for BoxPanel", and then come back to the code.

Start by deriving from Panel

Start by deriving a custom class from Panel. Probably the easiest way to do this is to define a separate code file for this class, using the Add | New Item | Class context menu options for a project from the Solution Explorer in Microsoft Visual Studio. Name the class (and file) BoxPanel.

The template file for a class doesn't start with many using statements because it's not specifically for Windows apps. So first, add using statements. The template file also starts with a few using statements that you probably don't need, and can be deleted. Here's a suggested list of using statements that can resolve types you'll need for typical custom panel code:

using System;
using System.Collections.Generic; // if you need to cast IEnumerable for iteration, or define your own collection properties
using Windows.Foundation; // Point, Size, and Rect
using Windows.UI.Xaml; // DependencyObject, UIElement, and FrameworkElement
using Windows.UI.Xaml.Controls; // Panel
using Windows.UI.Xaml.Media; // if you need Brushes or other utilities

Now that you can resolve Panel, make it the base class of BoxPanel. Also, make BoxPanel public:

public class BoxPanel : Panel
{
}

At the class level, define some int and double values that will be shared by several of your logic functions, but which won't need to be exposed as public API. In the example, these are named: maxrc, rowcount, colcount, cellwidth, cellheight, maxcellheight, aspectratio.

After you've done this, the complete code file looks like this (removing comments on using, now that you know why we have them):

using System;
using System.Collections.Generic;
using Windows.Foundation;
using Windows.UI.Xaml;
using Windows.UI.Xaml.Controls;
using Windows.UI.Xaml.Media;

public class BoxPanel : Panel 
{
    int maxrc, rowcount, colcount;
    double cellwidth, cellheight, maxcellheight, aspectratio;
}

From here on out, we'll be showing you one member definition at a time, be that a method override or something supporting such as a dependency property. You can add these to the skeleton above in any order.

MeasureOverride

protected override Size MeasureOverride(Size availableSize)
{
    // Determine the square that can contain this number of items.
    maxrc = (int)Math.Ceiling(Math.Sqrt(Children.Count));
    // Get an aspect ratio from availableSize, decides whether to trim row or column.
    aspectratio = availableSize.Width / availableSize.Height;

    // Now trim this square down to a rect, many times an entire row or column can be omitted.
    if (aspectratio > 1)
    {
        rowcount = maxrc;
        colcount = (maxrc > 2 && Children.Count <= maxrc * (maxrc - 1)) ? maxrc - 1 : maxrc;
    } 
    else 
    {
        rowcount = (maxrc > 2 && Children.Count <= maxrc * (maxrc - 1)) ? maxrc - 1 : maxrc;
        colcount = maxrc;
    }

    // Now that we have a column count, divide available horizontal, that's our cell width.
    cellwidth = (int)Math.Floor(availableSize.Width / colcount);
    // Next get a cell height, same logic of dividing available vertical by rowcount.
    cellheight = Double.IsInfinity(availableSize.Height) ? Double.PositiveInfinity : availableSize.Height / rowcount;
           
    foreach (UIElement child in Children)
    {
        child.Measure(new Size(cellwidth, cellheight));
        maxcellheight = (child.DesiredSize.Height > maxcellheight) ? child.DesiredSize.Height : maxcellheight;
    }
    return LimitUnboundedSize(availableSize);
}

The necessary pattern of a MeasureOverride implementation is the loop through each element in Panel.Children. Always call the Measure method on each of these elements. Measure has a parameter of type Size. What you're passing here is the size that your panel is committing to have available for that particular child element. So, before you can do the loop and start calling Measure, you need to know how much space each cell can devote. From the MeasureOverride method itself, you have the availableSize value. That is the size that the panel's parent used when it called Measure, which was the trigger for this MeasureOverride being called in the first place. So a typical logic is to devise a scheme whereby each child element divides the space of the panel's overall availableSize. You then pass each division of size to Measure of each child element.

How BoxPanel divides size is fairly simple: it divides its space into a number of boxes that's largely controlled by the number of items. Boxes are sized based on row and column count and the available size. Sometimes one row or column from a square isn't needed, so it's dropped and the panel becomes a rectangle rather than square in terms of its row : column ratio. For more info about how this logic was arrived at, skip ahead to "The scenario for BoxPanel".

So what does the measure pass do? It sets a value for the read-only DesiredSize property on each element where Measure was called. Having a DesiredSize value is possibly important once you get to the arrange pass, because the DesiredSize communicates what the size can or should be when arranging and in the final rendering. Even if you don't use DesiredSize in your own logic, the system still needs it.

It's possible for this panel to be used when the height component of availableSize is unbounded. If that's true, the panel doesn't have a known height to divide. In this case, the logic for the measure pass informs each child that it doesn't have a bounded height, yet. It does so by passing a Size to the Measure call for children where Size.Height is infinite. That's legal. When Measure is called, the logic is that the DesiredSize is set as the minimum of these: what was passed to Measure, or that element's natural size from factors such as explicitly-set Height and Width.

Note

The internal logic of StackPanel also has this behavior: StackPanel passes an infinite dimension value to Measure on children, indicating that there is no constraint on children in the orientation dimension. StackPanel typically sizes itself dynamically, to accommodate all children in a stack that grows in that dimension.

However, the panel itself can't return a Size with an infinite value from MeasureOverride; that throws an exception during layout. So, part of the logic is to find out the maximum height that any child requests, and use that height as the cell height in case that isn't coming from the panel's own size constraints already. Here's the helper function LimitUnboundedSize that was referenced in previous code, which then takes that maximum cell height and uses it to give the panel a finite height to return, as well as assuring that cellheight is a finite number before the arrange pass is initiated:

// This method limits the panel height when no limit is imposed by the panel's parent.
// That can happen to height if the panel is close to the root of main app window.
// In this case, base the height of a cell on the max height from desired size
// and base the height of the panel on that number times the #rows.
Size LimitUnboundedSize(Size input)
{
    if (Double.IsInfinity(input.Height))
    {
        input.Height = maxcellheight * colcount;
        cellheight = maxcellheight;
    }
    return input;
}

ArrangeOverride

protected override Size ArrangeOverride(Size finalSize)
{
     int count = 1;
     double x, y;
     foreach (UIElement child in Children)
     {
          x = (count - 1) % colcount * cellwidth;
          y = ((int)(count - 1) / colcount) * cellheight;
          Point anchorPoint = new Point(x, y);
          child.Arrange(new Rect(anchorPoint, child.DesiredSize));
          count++;
     }
     return finalSize;
}

The necessary pattern of an ArrangeOverride implementation is the loop through each element in Panel.Children. Always call the Arrange method on each of these elements.

Note how there aren't as many calculations as in MeasureOverride; that's typical. The size of children is already known from the panel's own MeasureOverride logic, or from the DesiredSize value of each child set during the measure pass. However, we still need to decide the location within the panel where each child will appear. In a typical panel, each child should render at a different position. A panel that creates overlapping elements isn't desirable for typical scenarios (although it's not out of the question to create panels that have purposeful overlaps, if that's really your intended scenario).

This panel arranges by the concept of rows and columns. The number of rows and columns was already calculated (it was necessary for measurement). So now the shape of the rows and columns plus the known sizes of each cell contribute to the logic of defining a rendering position (the anchorPoint) for each element that this panel contains. That Point, along with the Size already known from measure, are used as the two components that construct a Rect. Rect is the input type for Arrange.

Panels sometimes need to clip their content. If they do, the clipped size is the size that's present in DesiredSize, because the Measure logic sets it as the minimum of what was passed to Measure, or other natural size factors. So you don't typically need to specifically check for clipping during Arrange; the clipping just happens based on passing the DesiredSize through to each Arrange call.

You don't always need a count while going through the loop if all the info you need for defining the rendering position is known by other means. For example, in Canvas layout logic, the position in the Children collection doesn't matter. All the info needed to position each element in a Canvas is known by reading Canvas.Left and Canvas.Top values of children as part of the arrange logic. The BoxPanel logic happens to need a count to compare to the colcount so it's known when to begin a new row and offset the y value.

It's typical that the input finalSize and the Size you return from a ArrangeOverride implementation are the same. For more info about why, see "ArrangeOverride" section of XAML custom panels overview.

A refinement: controlling the row vs. column count

You could compile and use this panel just as it is now. However, we'll add one more refinement. In the code just shown, the logic puts the extra row or column on the side that's longest in aspect ratio. But for greater control over the shapes of cells, it might be desirable to choose a 4x3 set of cells instead of 3x4 even if the panel's own aspect ratio is "portrait." So we'll add an optional dependency property that the panel consumer can set to control that behavior. Here's the dependency property definition, which is very basic:

// Property
public Orientation Orientation
{
    get { return (Orientation)GetValue(OrientationProperty); }
    set { SetValue(OrientationProperty, value); }
}

// Dependency Property Registration
public static readonly DependencyProperty OrientationProperty =
        DependencyProperty.Register(nameof(Orientation), typeof(Orientation), typeof(BoxPanel), new PropertyMetadata(null, OnOrientationChanged));

// Changed callback so we invalidate our layout when the property changes.
private static void OnOrientationChanged(DependencyObject dependencyObject, DependencyPropertyChangedEventArgs args)
{
    if (dependencyObject is BoxPanel panel)
    {
        panel.InvalidateMeasure();
    }
}

And below is how using Orientation impacts the measure logic in MeasureOverride. Really all it's doing is changing how rowcount and colcount are derived from maxrc and the true aspect ratio, and there are corresponding size differences for each cell because of that. When Orientation is Vertical (default), it inverts the value of the true aspect ratio before using it for row and column counts for our "portrait" rectangle layout.

// Get an aspect ratio from availableSize, decides whether to trim row or column.
aspectratio = availableSize.Width / availableSize.Height;

// Transpose aspect ratio based on Orientation property.
if (Orientation == Orientation.Vertical) { aspectratio = 1 / aspectratio; }

The scenario for BoxPanel

The particular scenario for BoxPanel is that it's a panel where one of the main determinants of how to divide space is by knowing the number of child items, and dividing the known available space for the panel. Panels are innately rectangle shapes. Many panels operate by dividing that rectangle space into further rectangles; that's what Grid does for its cells. In Grid's case, the size of the cells is set by ColumnDefinition and RowDefinition values, and elements declare the exact cell they go into with Grid.Row and Grid.Column attached properties. Getting good layout from a Grid usually requires knowing the number of child elements beforehand, so that there are enough cells and each child element sets its attached properties to fit into its own cell.

But what if the number of children is dynamic? That's certainly possible; your app code can add items to collections, in response to any dynamic run-time condition you consider to be important enough to be worth updating your UI. If you're using data binding to backing collections/business objects, getting such updates and updating the UI is handled automatically, so that's often the preferred technique (see Data binding in depth).

But not all app scenarios lend themselves to data binding. Sometimes, you need to create new UI elements at runtime and make them visible. BoxPanel is for this scenario. A changing number of child items is no problem for BoxPanel because it's using the child count in calculations, and adjusts both the existing and new child elements into a new layout so they all fit.

An advanced scenario for extending BoxPanel further (not shown here) could both accommodate dynamic children and use a child's DesiredSize as a stronger factor for the sizing of individual cells. This scenario might use varying row or column sizes or non-grid shapes so that there's less "wasted" space. This requires a strategy for how multiple rectangles of various sizes and aspect ratios can all fit into a containing rectangle both for aesthetics and smallest size. BoxPanel doesn't do that; it's using a simpler technique for dividing space. BoxPanel's technique is to determine the least square number that's greater than the child count. For example, 9 items would fit in a 3x3 square. 10 items require a 4x4 square. However, you can often fit items while still removing one row or column of the starting square, to save space. In the count=10 example, that fits in a 4x3 or 3x4 rectangle.

You might wonder why the panel wouldn't instead choose 5x2 for 10 items, because that fits the item number neatly. However, in practice, panels are sized as rectangles that seldom have a strongly oriented aspect ratio. The least-squares technique is a way to bias the sizing logic to work well with typical layout shapes and not encourage sizing where the cell shapes get odd aspect ratios.

Reference

Concepts