Monthly Archives: August 2009

by dkt

Finding all of the bindings on an object in WPF

Disclaimer: Don’t do it. Whatever brought you here, chances are you’re doing something wrong, because you’re most likely about to do something very gnarly that you shouldn’t be doing with your view and a whole pile of code-behind. You could make your viewmodel more intelligent, you could implement IDataErrorInfo or otherwise use an attached behavior and put your errors on the viewmodel where they belong—chances are, there is something that you could be doing that won’t require you to need to identify all of the bindings on an object.

That being said, here’s how it’s done.

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20

public static Dictionary<DependencyProperty, BindingBase> GetAllBindings(DependencyObject d)
{
    if (d == null)
    {
        throw new ArgumentNullException("d");
    }
 
    var bindings = new Dictionary<DependencyProperty, BindingBase>();
    var lve = d.GetLocalValueEnumerator();
    while (lve.MoveNext())
    {
        DependencyProperty dp = lve.Current.Property;
        var expr = BindingOperations.GetBindingBase(d, dp);
        if (expr != null)
        {
            bindings.Add(dp, expr);
        }
    }
    return bindings;
}

If you are doing (or are thinking about doing) anything fancy with reflection to determine all properties and values on an object, you may want to consider using DependencyObjects instead. You get everything that DependencyObjects give you for free, and you can do everything without resorting to slow (and often obfuscating) reflection. —DKT

by dkt

Bindable Validation Errors

Wouldn’t it be great if you could do this in XAML:

1
2

<TextBox Text="{Binding Path=IngredientName}"
         Validation.Error="{Binding Path=IngredientNameError}"/>

Turns out you can:

1
2

<TextBox Text="{Binding Path=IngredientName}"
         vh:ValidationHelper.Error="{Binding Path=IngredientNameError}"/>
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159

using System;
using System.Globalization;
using System.Windows;
using System.Windows.Controls;
using System.Windows.Data;
 
namespace Pelebyte.ValidationUtils
{
  /// <summary>
  /// Provides validation helper methods and utilities.
  /// </summary>
  public static class ValidationHelper
  {
    /// <summary>
    /// The <see cref="DependencyProperty"/> that identifies the
    /// <see cref="P:ValidationHelper.Error"/> attached property.
    /// </summary>
    public static readonly DependencyProperty ErrorProperty =
      DependencyProperty.RegisterAttached(
        "Error", typeof(string), typeof(ValidationHelper),
        new PropertyMetadata(null, OnErrorChanged));
 
    /// <summary>
    /// Gets the custom error applied to the control.
    /// </summary>
    /// <param name="d">
    /// The control to get the current custom error for.
    /// </param>
    /// <returns>
    /// A string that represents the custom error.
    /// </returns>
    public static string GetError(DependencyObject d)
    {
      return (string)d.GetValue(ErrorProperty);
    }
 
    /// <summary>
    /// Sets the custom error applied to the control.
    /// </summary>
    /// <param name="d">
    /// The control to set the current custom error for.
    /// </param>
    /// <param name="value">
    /// A string that represents the custom error. An empty or
    /// <c>null</c> string clears the error on the field.
    /// </param>
    public static void SetError(DependencyObject d, string value)
    {
      d.SetValue(ErrorProperty, value);
    }
 
    /// <summary>
    /// Called when the <see cref="P:ValidationHelper.Error"/>
    /// attached property changes value.
    /// </summary>
    /// <param name="d">
    /// The <see cref="DependencyObject"/> that is having its
    /// <see cref="P:ValidationHelper.Error"/> property changing
    /// values.
    /// </param>
    /// <param name="e">
    /// The <see cref="DependencyPropertyChangedEventArgs"/>
    /// instance containing the event data.
    /// </param>
    private static void OnErrorChanged(
      DependencyObject d, DependencyPropertyChangedEventArgs e)
    {
      BindingExpressionBase expr;
 
        expr = BindingOperations.GetBindingExpressionBase(
          d, BindingErrorTargetProperty);
 
      if ((expr == null) && (e.NewValue != null))
      {
          // create a new binding between two properties that
          // we're only going to use so that we have an avenue
          // of our own to attach binding errors
          Binding b = new Binding();
          b.Source = d;
          b.Path = new PropertyPath(BindingErrorSourceProperty);
          b.Mode = BindingMode.OneWayToSource;
 
          b.ValidationRules.Add(new InternalRule(d));
 
          expr = BindingOperations.SetBinding(
              d, BindingErrorTargetProperty, b);
      }
 
        if (expr != null)
      {
        expr.UpdateSource();
      }
      }
    }
 
    /// <summary>
    /// The internal implementation of <see cref="ValidationRule"/>
    /// that returns our real "error" whenever we want.
    /// </summary>
    private sealed class InternalRule : ValidationRule
    {
      private readonly DependencyObject _d;
 
      /// <summary>
      /// Initializes a new instance of the
      /// <see cref="InternalRule"/> class specific to a
      /// particular object. The
      /// <see cref="P:ValidationHelper.Error"/> property of the
      /// given object will be used to determine the error on the
      /// object.
      /// </summary>
      /// <param name="d">
      /// The <see cref="DependencyObject"/> to return errors
      /// for.
      /// </param>
      public InternalRule(DependencyObject d)
      {
        _d = d;
      }
 
      public override ValidationResult Validate(
          object value, CultureInfo cultureInfo)
      {
        // completely ignore /value/ and look for the error
        // on the DependencyObject that was given to us in
        // our constructor
        string error = GetError(_d);
 
        if (string.IsNullOrEmpty(error))
        {
          // an empty or null string means no error
          return ValidationResult.ValidResult;
        }
        else
        {
          // anything else means an error
          return new ValidationResult(false, error);
        }
      }
    }
 
    // two private dependency properties that we use internally to
    // set up our useless binding
 
    private static readonly DependencyProperty
      BindingErrorSourceProperty =
        DependencyProperty.RegisterAttached(
          "BindingErrorSource", typeof(object),
          typeof(ValidationHelper),
          new PropertyMetadata(null));
 
    private static readonly DependencyProperty
      BindingErrorTargetProperty =
        DependencyProperty.RegisterAttached(
          "BindingErrorTarget", typeof(object),
          typeof(ValidationHelper),
          new PropertyMetadata(null));
  }
}

Why it works

The System.Windows.Controls.Validation.Errors property is a collection for a reason—it’s a collection of all of the binding errors on the object.

For most controls, it’s not readily apparent that more than one binding on the same control could actually fail:

1

<TextBox Text="{Binding Path=IngredientName, ValidatesOnDataErrors=True}"/>

But if you had a complex control where more than one property was controlled directly by the user, it’s more obvious why you’d possibly need a collection instead of a single object:

1
2
3
4
5

<!-- this slider has two thumbs; the user drags both of
     them around to specify a range -->
<my:DoubleSlider
    MinValue="{Binding Path=Minimum, ValidatesOnDataErrors=True}"
    MaxValue="{Binding Path=Maximum, ValidatesOnDataErrors=True}"/>

If both of these properties had errors, WPF would collect both of them in Validation.Errors.

So the ValidationHelper code above is essentially emulating the following XAML snippet in C# (note that the code in green isn’t actually possible—that’s why we’re writing this code in C#):

<TextBox x:Name="MyTextBox"
Text="{Binding Path=IngredientName, ValidatesOnDataErrors=True}"
<vh:ValidationHelper.BindingErrorTarget>
<Binding Source="MyTextBox"
Path="(vh:ValidationHelper.BindingErrorSource)">
Mode="OneWayToSource"
<vh:ValidationHelper+InternalRule (MyTextBox)>
</Binding>
</vh:ValidationHelper.BindingErrorTarget>
</TextBox>

(It may help at this point to open another a window with my little sketch of how WPF data binding data flows around.)

Our hidden BindingErrorTarget property participates in validation just like a regular property. So when the value of ValidationHelper.Error is changed:

  1. Force the binding on ValidationHelper.BindingErrorTarget to be re-evaluated from the target back to the source (call BindingExpression.UpdateSource()).
  2. Our validation rule InternalRule gets called as part of the normal validation process; our rule will return ValidationHelper.Error instead of validating the incoming value.

It doesn’t actually matter what the values of the BindingErrorTarget and BindingErrorSource properties are; they only exist so that we can key into the binding system.

Why?

IDataErrorInfo and ValidatesOnDataErrors would seemingly make this technique redundant: why go through all this trouble to expose a binding site for errors on the viewmodel when you could just implement IDataErrorInfo?

  • IDataErrorInfo is only consulted when the source property changes value—if you have a data source whose errors can dynamically change independently of the source property, there isn’t a clean way from the viewmodel to force the view to pick up your changes in the error. (If your source implements INotifyPropertyChanged, you can raise PropertyChanged for the relevant property, but if you use DependencyObjects, there is no way to force the binding system to re-evaluate the property from the viewmodel—you’d need the BindingExpression, which then requires your viewmodel to have knowledge of the view).
  • It may be inconvenient or impossible from a design standpoint to have the object containing your error property to also implement IDataErrorInfo.
    If your error comes from a different object than your viewmodel, then your implementation of IDataErrorInfo would need to know where to fetch it.
  • If you ever had a situation where you had a ValidationRule that you wanted to bind to, you’ve probably discovered that it’s never going to happen—ValidationRule, not being a DependencyObject, doesn’t support binding—not in code, and certainly not in XAML. An attached behavior or subclass (you never need to subclass in WPF) is really your only recourse for situations like this.
  • You don’t want to rely on .NET 3.0 SP1 or .NET 3.5 SP1. Thankfully, Windows 7 comes out of the box with it, but Vista, and certainly XP, do not. This technique works with every version of WPF.

Remember that any DependencyProperty that is the target of a binding can hold validation errors; also remember that you can add attached properties to any object. That means you can add arbitrary validation errors to any DependencyObject through this trick. You can also drive this error generation off of whatever you want—I chose the simplest example and created an attached property specifically to hold an error that will be reported, unchanged, right back through the binding system. You could create a new attached behavior, have the TextBox.Text property and TextBox.TextChanged event drive the error; then set up an attached behavior that provides validation on the text of the TextBox without having to provide an instance of ValidationRule. —DKT

In the interests of not cluttering the picture, I left out a little magic trick:

  • We could attach a converter to the binding where IValueConverter.ConvertBack() returns Binding.DoNothing; this would stop data from ever flowing to the source.
  • Then we could actually drop the BindingErrorSource property and reuse an existing one (like Tag or something), knowing that our binding will never actually change the value or otherwise interfere with it.

And why use two properties when you could get away with just one?

Sep 24: Updated the OnErrorChanged to fix a bug that would cause the error state to never actually clear…whoops! –DKT

by dkt

Data Flow in a WPF BindingExpression

I was writing up some posts on how ValidationRules work, and of course, I ended up confusing myself, because data binding can get confusing once you start throwing in IValueConverters, ValidationRules, target objects, source objects…yikes.

So I decided to sketch out a simple diagram for my own benefit. And after 15 minutes, I realized it wasn’t so simple:





Click the image for a larger version

Hope it helps make WPF binding less confusing… —DKT

by dkt

Dealing with Large Data Sets

Large data sets are increasingly becoming a reality of front-ends, GUIs and web apps alike. Users are getting more comfortable with tons of data, and are generating a lot of it too. Large data sets are as a much a reality as multi-threading—they’re both incredibly annoying to deal with, and they’re both not ignorable with today’s computers (and users).

Strategies

There are four basic strategies to dealing with large data sets: Ignoring the problem, Filtering, Paging, and Data Virtualization.

Ignoring the problem

Don’t do this unless your users never deal with more than ten things at a time. (Incidentally, I have never met anyone with an e-mail account with less than ten e-mails, or anyone with an iPod with ten songs—have you?)

Instant F. Why’d you bother writing the app in the first place? Your users are furious because they’re either spending large quantities of time staring at progress bars, or worse, a frozen screen.

Filtering

Filtering sometimes avoids the need for data virtualization, but this is usually nothing more than a band-aid for a broken arm—watch as your first user types in a query that somehow manages to return 90% of your data set and bring your app to its knees. In order for filtering to be effective, your user must be able and willing to provide enough criteria to narrow down the data set into something that’s not a “large data set”. From the user’s point of view, there is a very high up-front cost in using the UI—more often than not, “filtering” means twenty text boxes/dropdowns/checkboxes, all using names for fields that no one really understands. (“Hey, Bob—what do you suppose the “Type” dropdown is for? It’s got ‘PCX’, ‘AVW’, and ‘BOO’ in it.”) Those cryptic fields are not self-describing because the only thing that could help—the data itself—is locked behind a dizzying array of options.

This’ll get you anything from a C+ to a F, depending on the data set. Everything will be fine until That User types in “T” for a search query. (By the way, did you know “T” is the ticker symbol for AT&T?)

Paging

Implementing paging shows that you care about your users’ time. The GUI doesn’t feel overloaded or clunky and it’s a familiar concept to the user: everyone knows what “Page 5 of 752 (81–100, 15,035 items total)” means. It’s not a coincidence that most (all?) webmail apps show things in pages of 20 or 50 or so e-mails; the browser would choke over trying to render a <table> with thousands of rows and would provide for a very nasty user experience.

It’s not perfect, though:

  • Changing data. If you’re dealing with a data set where rows are frequently added/removed while the user is looking at it, paging is becomes a mediocre solution. You can then either keep the page boundaries stable (which requires at least the server to remember the full data set at the time of initial query) or ignore the problem (which would cause rows to randomly disappear or appear twice as the user Next Pages through your data set).
  • Slower user experience. Most operations are a request-reply back to the server—sorting, grouping, filtering, etc. Most clicks make the user wait.
  • No table scanning. There is something to be said for quickly scrolling through thousands of items, scanning for something when a search isn’t giving me back what I want. (Was it “Bob’s Restaurant”? Oh! I see it there on Row 452; I was actually looking for “Bill’s Burgers and Fries”.)
  • n-dimeinsional Data. Paging is very linear. If you have more than one dimension of data, paging is not very useful. I’m referring to every data set that looks better as a pretty graph.

This can get you as high as an A– (it definitely works for all the search engines) down to a C– (can you imagine what working with iTunes would be like if they made you “Click here for the next 20 results”?).

Data Virtualization

Make the user think you loaded everything. When they click the sort headers, they don’t know that you’re actually sucking out objects back from disk. When they’re whipping through the scrollbar, they have no idea that you’re grabbing loading the next 100 rows into memory.

The only downside? It’s obnoxiously difficult on most development platforms to get right.

What is virtualization?

Any developer who has worked with a grid control knows how important virtualization is for performance. The basic idea behind virtualization is very simple: if the user can’t see it, then the computer doesn’t need it. You can present the user with a giant scrollable expanse of cells; the grid will handle creating/dropping/recycling grid cells as necessary. Any grid control worth anything supports virtualization; even the basic, built-in ListBox and ListView in WPF (and Silverlight 3.0) support row-based virtualization. But in most simple binding situations, you generally need a collection of all of the data that you want to render for your grid loaded in memory.

For applications that don’t delve into large collections of data, there isn’t much point. The added complexity isn’t worth it. But that being said, a lot of commonly-used types of applications benefit from data virtualization:

  • Music library applications—think iTunes with tens of thousands of files of music, movies, etc.
  • Desktop mail applications—the inbox with thousands of e-mails
  • Google Maps—possibly the only web app that I can think of that shows off what data virtualization can truly look like

It would be really neat if Google could take the same approach with Mail as they do in Maps—a virtual table that doesn’t actually download all the mail, but instead lets you pan through the inbox using a fake scrollbar as if everything was already downloaded, but they don’t because it’s a pain in the neck and paging usually works just as well. (Or maybe they just haven’t thought of it yet. If you see that feature within the next few months or years, you can thank me for giving them the idea.)

A+, if you pull off the illusion flawlessly.

How do I implement it?

Most grid controls allow you to hook into their virtualization so that you could conceivably provide your own data virtualization to go along with the control virtualization native to the control, but there isn’t much out of the box in .NET to help you with the data side. There is no built-in “virtual” ICollection<T>, although there is nothing to stop someone from implementing one. Indeed, Beatriz Costa wrote about some data virtualization techniques that can be used in WPF and Silverlight, but they aren’t perfect.

Although .NET is a great platform for quickly building apps, Cocoa is a great platform for quickly building apps that can handle lots and lots of data. Cocoa on the Mac (and, as of iPhone OS 3.0, Cocoa Touch) supports data virtualization out-of-the-box using Core Data. Core Data provides a mechanism for defining a data model, and then it takes care of persistence of objects for you. On Cocoa Touch, NSFetchedResultsController acts as an intermediary between Core Data and your controls, essentially handling loading and unloading of data as the UI requires it. Given the memory constraints on the iPhone, Apple really had to provide a solid solution for data virtualization—keeping a list of even a few hundred moderately complicated objects could cause your app to run out of memory and crash.

Core Data uses SQLite under the hood*, and it is a very good solution for rolling your own data virtualization scheme. SQLite is an embedded database engine. There are no servers to install, no processes to run—it’s just a library that allows SQL access to a file on the file system. It has the semantics of a database and the semantics of a file, depending on which one is more convenient. System.Data.SQLite is an excellent ADO.NET provider that you can use in .NET (unfortunately, because it uses native code, it’s off-limits in Silverlight):

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36

using System.Data.SQLite;
 
public static class SQLiteTest
{
    public static void Main()
    {
        // colors.db is just a file that will be created in the
        // current directory if it doesn't exist
        using (var conn = new SQLiteConnection("Data Source=colors.db"))
        {
            conn.Open();
            using (IDbCommand cmd = conn.CreateCommand())
            {
                cmd.CommandText = "CREATE TABLE Colors(id INTEGER PRIMARY KEY, name TEXT)";
                cmd.ExecuteNonQuery();
 
                cmd.CommandText = "INSERT INTO Colors(name) VALUES(\"Red\")";
                cmd.ExecuteNonQuery();
 
                cmd.CommandText = "INSERT INTO Colors(name) VALUES(\"Green\")";
                cmd.ExecuteNonQuery();
 
                cmd.CommandText = "INSERT INTO Colors(name) VALUES(\"Blue\")";
                cmd.ExecuteNonQuery();
            }
            using (IDbCommand cmd = conn.CreateCommand())
            {
                cmd.CommandText = "SELECT name FROM Colors";
                using (IDataReader reader = cmd.ExecuteReader())
                {
                    System.Console.WriteLine(reader.Read(0));
                }
            }
        }
    }
}

Because the database is local, queries are fast from start to finish—no network latency issues here. And because it’s just a file, you can throw it away when you’re done. SQLite makes an excellent backbone to any data virtualization scheme because of its hybrid semantics (simplified random access like a database, convenience like a flat file).

*I’m purposely leaving out Core Data’s binary and XML serialization because they require the entire object graph to be loaded into memory, and if you’re going to do that, then what’s the point?

But large data sets aren’t that important…

Every mature application that I have ever been a part of has had to tackle the issue of larger data sets at some point. In my personal experience, thousands of entities (hundreds if you’re talking about images) is enough to be considered a “large data set” in that your application begins to visibly suffer.

I’d love to recommend data virtualization as the route to go for handling large data sets in .NET, but there really isn’t enough in the way of frameworks that provides an out-of-the-box solution. Implement paging unless you really need data virtualization. If you’re working on iPhone apps, drop everything and learn Core Data if you haven’t already. You’d be surprised at how much you don’t need to worry about. But regardless of what platform you’re developing a front-end on, you should always at least ask yourself how your GUI might respond to thousands of x. because one day it’s going to have to. —DKT