Proper validation with MVVM

Warning: Very long and detailed post.

Okay, validation in WPF when using MVVM. I’ve read many things now, looked at many SO questions, and tried many approaches, but everything feels somewhat hacky at some point and I’m really not sure how to do it the right way™.

Ideally, I want to have all validation happen in the view model using IDataErrorInfo; so that’s what I did. There are however different aspects that make this solution be not a complete solution for the whole validation topic.

The situation

Let’s take the following simple form. As you can see, it’s nothing fancy. We just have a two textboxes which bind to a string and int property in the view model each. Furthermore we have a button that is bound to an ICommand.

So for the validation we now have a two choices:

1. We can run the validation automatically whenever the value of a text box changes. As such the user gets an instant response when he entered something invalid.

• We can take this one step further to disable the button when there are any errors.

2. Or we can run the validation only explicitly when the button is pressed, then showing all errors if applicable. Obviously we can’t disable the button on errors here.

Ideally, I want to implement choice 1. For normal data bindings with activated ValidatesOnDataErrors this is default behavior. So when the text changes, the binding updates the source and triggers the IDataErrorInfo validation for that property; errors are reported back the view. So far so good.

Validation status in the view model

The interesting bit is to let the view model, or the button in this case, know if there are any errors. The way IDataErrorInfo works, it is mainly there to report errors back to the view. So the view can easily see if there are any errors, display them and even show annotations using Validation.Errors. Furthermore, validation always happens looking at a single property.

So having the view model know when there are any errors, or if the validation succeeded, is tricky. A common solution is to simply trigger the IDataErrorInfo validation for all properties in the view model itself. This is often done using a separate IsValid property. The benefit is that this can also be easily used for disabling the command. The drawback is that this might run the validation on all properties a bit too often, but most validations should be simply enough to not hurt the performance. Another solution would be to remember which properties produced errors using the validation and only check those, but that seems a bit overcomplicated and unnecessary for most times.

The bottom line is that this could work fine. IDataErrorInfo provides the validation for all properties, and we can simply use that interface in the view model itself to run the validation there too for the whole object. Introducing the problem:

Binding exceptions

The view model uses actual types for its properties. So in our example, the integer property is an actual int. The text box used in the view however natively only supports text. So when binding to the int in the view model, the data binding engine will automatically perform type conversions—or at least it will try. If you can enter text in a text box meant for numbers, the chances are high that there won’t always be valid numbers inside: So the data binding engine will fail to convert and throw a FormatException.

On the view side, we can easily see that. Exceptions from the binding engine are automatically caught by WPF and are displayed as errors—there isn’t even a need to enable Binding.ValidatesOnExceptions which would be required for exceptions thrown in the setter. The error messages do have a generic text though, so that could be a problem. I have solved this for myself by using a Binding.UpdateSourceExceptionFilter handler, inspecting the exception being thrown and looking at the source property and then generating a less generic error message instead. All that capsulated away into my own Binding markup extension, so I can have all the defaults I need.

So the view is fine. The user makes an error, sees some error feedback and can correct it. The view model however is lost. As the binding engine threw the exception, the source was never updated. So the view model is still on the old value, which isn’t what’s being displayed to the user, and the IDataErrorInfo validation obviously doesn’t apply.

What’s worse, there is no good way for the view model to know this. At least, I haven’t found a good solution for this yet. What would be possible is to have the view report back to the view model that there was an error. This could be done by data binding the Validation.HasError property back to the view model (which isn’t possible directly), so the view model could check the view’s state first.

Another option would be to relay the exception handled in Binding.UpdateSourceExceptionFilter to the view model, so it would be notified of it as well. The view model could even provide some interface for the binding to report these things, allowing for custom error messages instead of generic per-type ones. But that would create a stronger coupling from the view to the view model, which I generally want to avoid.

Another “solution” would be to get rid of all typed properties, use plain string properties and do the conversion in the view model instead. This obviously would move all validation to the view model, but also mean an incredible amount of duplication of things the data binding engine usually takes care of. Furthermore it would change the semantics of the view model. For me, a view is built for the view model and not the reverse—of course the design of the view model depends on what we imagine the view to do, but there’s still general freedom how the view does that. So the view model defines an int property because there is a number; the view can now use a text box (allowing all these problems), or use something that natively works with numbers. So no, changing the types of the properties to string is not an option for me.

In the end, this is a problem of the view. The view (and its data binding engine) is responsible for giving the view model proper values to work with. But in this case, there seems to be no good way to tell the view model that it should invalidate the old property value.

BindingGroups

Binding groups are one way I tried to tackle this. Binding groups have the ability to group all validations, including IDataErrorInfo and thrown exceptions. If available to the view model, they even have a mean to check the validation status for all of those validation sources, for example using CommitEdit.

By default, binding groups implement choice 2 from above. They make the bindings update explicitly, essentially adding an additional uncommitted state. So when clicking the button, the command can commit those changes, trigger the source updates and all validations and get a single result if it succeeded. So the command’s action could be this:

 if (bindingGroup.CommitEdit())
     SaveEverything();

CommitEdit will only return true if all validations succeeded. It will take IDataErrorInfo into account and also check binding exceptions. This seems to be a perfect solution for choice 2. The only thing that is a bit of a hassle is managing the binding group with the bindings, but I’ve built myself something that mostly takes care of this (related).

If a binding group is present for a binding, the binding will default to an explicit UpdateSourceTrigger. To implement choice 1 from above using binding groups, we basically have to change the trigger. As I have a custom binding extension anyway, this is rather simple, I just set it to LostFocus for all.

So now, the bindings will still update whenever a text field changes. If the source could be updated (binding engine throws no exception) then IDataErrorInfo will run as usual. If it couldn’t be updated the view is still able to see it. And if we click our button, the underlying command can call CommitEdit (although nothing needs to be committed) and get the total validation result to see if it can continue.

We might not be able to disable the button easily this way. At least not from the view model. Checking the validation over and over is not really a good idea just to update the command status, and the view model isn’t notified when a binding engine exception is thrown anyway (which should disable the button then)—or when it goes away to enable the button again. We could still add a trigger to disable the button in the view using the Validation.HasError so it’s not impossible.

Solution?

So overall, this seems to be the perfect solution. What is my problem with it though? To be honest, I’m not entirely sure. Binding groups are a complex thing that seem to be usually used in smaller groups, possibly having multiple binding groups in a single view. By using one big binding group for the whole view just to ensure my validation, it feels as if I’m abusing it. And I just keep thinking, that there must be a better way to solve this whole situation, because surely I can’t be the only one having these problems. And so far I haven’t really seen many people use binding groups for validation with MVVM at all, so it just feels odd.

So, what exactly is the proper way to do validation in WPF with MVVM while being able to check for binding engine exceptions?

---

My solution (/hack)

First of all, thanks for your input! As I have written above, I’m using IDataErrorInfo already to do my data validation and I personally believe it’s the most comfortable utility to do the validation job. I’m using utilities similar to what Sheridan suggested in his answer below, so maintaining works fine too.

In the end, my problem boiled down to the binding exception issue, where the view model just wouldn’t know about when it happened. While I could handle this with binding groups as detailed above, I still decided against it, as I just didn’t feel all that comfortable with it. So what did I do instead?

As I mentioned above, I detect binding exceptions on the view-side by listening to a binding’s UpdateSourceExceptionFilter. In there, I can get a reference to the view model from the binding expression’s DataItem. I then have an interface IReceivesBindingErrorInformation which registers the view model as a possible receiver for information about binding errors. I then use that to pass the binding path and the exception to the view model:

object OnUpdateSourceExceptionFilter(object bindExpression, Exception exception)
{
    BindingExpression expr = (bindExpression as BindingExpression);
    if (expr.DataItem is IReceivesBindingErrorInformation)
    {
        ((IReceivesBindingErrorInformation)expr.DataItem).ReceiveBindingErrorInformation(expr.ParentBinding.Path.Path, exception);
    }

    // check for FormatException and produce a nicer error
    // ...
 }

In the view model I then remember whenever I am notified about a path’s binding expression:

HashSet<string> bindingErrors = new HashSet<string>();
void IReceivesBindingErrorInformation.ReceiveBindingErrorInformation(string path, Exception exception)
{
    bindingErrors.Add(path);
}

And whenever the IDataErrorInfo revalidates a property, I know that the binding worked, and I can clear the property from the hash set.

In the view model I then can check if the hash set contains any items and abort any action that requires the data to be validated completely. It might not be the nicest solution due to the coupling from the view to the view model, but using that interface it’s at least somewhat less a problem.

>Warning: Long answer also

I use the IDataErrorInfo interface for validation, but I have customised it to my needs. I think that you'll find that it solves some of your problems too. One difference to your question is that I implement it in my base data type class.

As you pointed out, this interface just deals with one property at a time, but clearly in this day and age, that's no good. So I just added a collection property to use instead:

protected ObservableCollection<string> errors = new ObservableCollection<string>();

public virtual ObservableCollection<string> Errors
{
    get { return errors; }
}

To address your problem of not being able to display external errors (in your case from the view, but in mine from the view model), I simply added another collection property:

protected ObservableCollection<string> externalErrors = new ObservableCollection<string>();

public ObservableCollection<string> ExternalErrors
{
    get { return externalErrors; }
}

I have an HasError property which looks at my collection:

public virtual bool HasError
{
    get { return Errors != null && Errors.Count > 0; }
}

This enables me to bind this to Grid.Visibility using a custom BoolToVisibilityConverter, eg. to show a Grid with a collection control inside that shows the errors when there are any. It also lets me change a Brush to Red to highlight an error (using another Converter), but I guess you get the idea.

Then in each data type, or model class, I override the Errors property and implement the Item indexer (simplified in this example):

public override ObservableCollection<string> Errors
{
    get
    {
        errors = new ObservableCollection<string>();
        errors.AddUniqueIfNotEmpty(this["Name"]);
        errors.AddUniqueIfNotEmpty(this["EmailAddresses"]);
        errors.AddUniqueIfNotEmpty(this["SomeOtherProperty"]);
        errors.AddRange(ExternalErrors);
        return errors;
    }
}

public override string this[string propertyName]
{
    get
    {
        string error = string.Empty;
        if (propertyName == "Name" && Name.IsNullOrEmpty()) error = "You must enter the Name field.";
        else if (propertyName == "EmailAddresses" && EmailAddresses.Count == 0) error = "You must enter at least one e-mail address into the Email address(es) field.";
        else if (propertyName == "SomeOtherProperty" && SomeOtherProperty.IsNullOrEmpty()) error = "You must enter the SomeOtherProperty field.";
        return error;
    }
}

The AddUniqueIfNotEmpty method is a custom extension method and 'does what is says on the tin'. Note how it will call each property that I want to validate in turn and compile a collection from them, ignoring duplicate errors.

Using the ExternalErrors collection, I can validate things that I can't validate in the data class:

private void ValidateUniqueName(Genre genre)
{
    string errorMessage = "The genre name must be unique";
    if (!IsGenreNameUnique(genre))
    {
        if (!genre.ExternalErrors.Contains(errorMessage)) genre.ExternalErrors.Add(errorMessage);
    }
    else genre.ExternalErrors.Remove(errorMessage);
}

To address your point regarding the situation where a user enters an alphabetical character into a int field, I tend to use a custom IsNumeric AttachedProperty for the TextBox, eg. I don't let them make these kinds of errors. I always feel that it's better to stop it, than to let it happen and then fix it.

Overall I'm really happy with my validation ability in WPF and am not left wanting at all.

To end with and for completeness, I felt that I should alert you to the fact that there is now an INotifyDataErrorInfo interface which includes some of this added functionality. You can find out more from the INotifyDataErrorInfo Interface page on MSDN.

---

UPDATE >>>

Yes, the ExternalErrors property just let's me add errors that relate to a data object from outside that object... sorry, my example wasn't complete... if I'd have shown you the IsGenreNameUnique method, you would have seen that it uses LinQ on all of the Genre data items in the collection to determine whether the object's name is unique or not:

private bool IsGenreNameUnique(Genre genre)
{
    return Genres.Where(d => d.Name != string.Empty && d.Name == genre.Name).Count() == 1;
}

As for your int/string problem, the only way I can see you getting those errors in your data class is if you declare all your properties as object, but then you'd have an awful lot of casting to do. Perhaps you could double your properties like this:

public object FooObject { get; set; } // Implement INotifyPropertyChanged

public int Foo
{
    get { return FooObject.GetType() == typeof(int) ? int.Parse(FooObject) : -1; }
}

Then if Foo was used in code and FooObject was used in the Binding, you could do this:

public override string this[string propertyName]
{
    get
    {
        string error = string.Empty;
        if (propertyName == "FooObject" && FooObject.GetType() != typeof(int)) 
            error = "Please enter a whole number for the Foo field.";
        ...
        return error;
    }
}

That way you could fulfil your requirements, but you'll have a lot of extra code to add.

In my opinion, the problem lies in validation happening at too many places. I also wished to write all my validation login in ViewModel but all those number binding were making my ViewModel crazy.

I solved this problem by creating a binding that never fails. Obviously, if a binding is always successful then the type itself has to handle the error conditions gracefully.

Failable Value Type

I started by creating a generic type which would gracefully support the failed conversions:

public struct Failable<T>
{
    public T Value { get; private set; }
    public string Text { get; private set; }
    public bool IsValid { get; private set; }

    public Failable(T value)
    {
        Value = value;

        try
        {
            var converter = TypeDescriptor.GetConverter(typeof(T));
            Text = converter.ConvertToString(value);
            IsValid = true;
        }
        catch
        {
            Text = String.Empty;
            IsValid = false;
        }
    }

    public Failable(string text)
    {
        Text = text;

        try
        {
            var converter = TypeDescriptor.GetConverter(typeof(T));
            Value = (T)converter.ConvertFromString(text);
            IsValid = true;
        }
        catch
        {
            Value = default(T);
            IsValid = false;
        }
    }
}

Note that even if the type fails to initialise because of invalid input string (second constructor), it quietly stores the invalid state along with invalid text also. This is required in order to support the round-trip of binding even in case of wrong input.

Generic Value Converter

A generic value converter could be written using above type:

public class StringToFailableConverter<T> : IValueConverter
{
    public object Convert(object value, Type targetType, object parameter, CultureInfo culture)
    {
        if (value.GetType() != typeof(Failable<T>))
            throw new InvalidOperationException("Invalid value type.");

        if (targetType != typeof(string))
            throw new InvalidOperationException("Invalid target type.");

        var rawValue = (Failable<T>)value;
        return rawValue.Text;
    }

    public object ConvertBack(object value, Type targetType, object parameter, CultureInfo culture)
    {
        if (value.GetType() != typeof(string))
            throw new InvalidOperationException("Invalid value type.");

        if (targetType != typeof(Failable<T>))
            throw new InvalidOperationException("Invalid target type.");

        return new Failable<T>(value as string);
    }
}

XAML Handy Converters

Since creating and using the instances of generics is pain in XAML, lets make static instances of common converters:

public static class Failable
{
    public static StringToFailableConverter<Int32> Int32Converter { get; private set; }
    public static StringToFailableConverter<double> DoubleConverter { get; private set; }

    static Failable()
    {
        Int32Converter = new StringToFailableConverter<Int32>();
        DoubleConverter = new StringToFailableConverter<Double>();
    }
}

Other value types can be extended easily.

Usage

Usage is pretty simple, just need to change the type from int to Failable<int>:

ViewModel

public Failable<int> NumberValue
{
    //Custom logic along with validation
    //using IsValid property
}

XAML

<TextBox Text="{Binding NumberValue,Converter={x:Static local:Failable.Int32Converter}}"/>

This way, you can use the same validation mechanism (IDataErrorInfo or INotifyDataErrorInfo or anything else) in ViewModel by checking the IsValid property. If IsValid is true, you can directly use the Value.

Ok, I believe I have found the answer you were looking for...
It will not be easy to explain - but..
Very easy to understand once explained...
I do think It is most accurate/"certified" to MVVM viewed as "standard" or at the least attempted standard.

But before we begin.. you need to change a concept which you got used to regarding MVVM:

"Furthermore it would change the semantics of the view model. For me, a view is built for the view model and not the reverse—of course the design of the view model depends on what we imagine the view to do, but there’s still general freedom how the view does that"

That paragraph is the source of your problem.. - why?

Because you are stating the View-Model has no role to adjust itself to the View..
That is wrong in many ways - as I'll prove to you very simply..

If you have a property such as:

public Visibility MyPresenter { get...

What is Visibility if not something that serves the View?
The type itself and the name that will be given to the property is definitely made up for the view.

There two distinguishable View-Models categories in MVVM according to my experience:

• Presenter View Model - which is to be hooked to buttons, menus, Tab Items etc....
• Entity View Model - which is to be hocked to controls that brings the entity data to screen.

These are two different - completely different concerns.

And now to the solution:

public abstract class ViewModelBase : INotifyPropertyChanged
{
   public event PropertyChangedEventHandler PropertyChanged;

   public void RaisePropertyChanged([CallerMemberName] string propertyName = null)
   {
	  if (PropertyChanged != null)
		 PropertyChanged(this, new PropertyChangedEventArgs(propertyName));
   }
}

public class VmSomeEntity : ViewModelBase, INotifyDataErrorInfo
{
	//This one is part of INotifyDataErrorInfo interface which I will not use,
	//perhaps in more complicated scenarios it could be used to let some other VM know validation changed.
	public event EventHandler<DataErrorsChangedEventArgs> ErrorsChanged; 

	//will hold the errors found in validation.
	public Dictionary<string, string> ValidationErrors = new Dictionary<string, string>();

	//the actual value - notice it is 'int' and not 'string'..
	private int storageCapacityInBytes;

	//this is just to keep things sane - otherwise the view will not be able to send whatever the user throw at it.
	//we want to consume what the user throw at us and validate it - right? :)
	private string storageCapacityInBytesWrapper;

	//This is a property to be served by the View.. important to understand the tactic used inside!
	public string StorageCapacityInBytes
	{
	   get { return storageCapacityInBytesWrapper ?? storageCapacityInBytes.ToString(); }
	   set
	   {
		  int result;
		  var isValid = int.TryParse(value, out result);
		  if (isValid)
		  {
			 storageCapacityInBytes = result;
			 storageCapacityInBytesWrapper = null;
			 RaisePropertyChanged();
		  }
		  else
			 storageCapacityInBytesWrapper = value;         

		  HandleValidationError(isValid, "StorageCapacityInBytes", "Not a number.");
	   }
	}

	//Manager for the dictionary
	private void HandleValidationError(bool isValid, string propertyName, string validationErrorDescription)
	{
		if (!string.IsNullOrEmpty(propertyName))
		{
			if (isValid)
			{
				if (ValidationErrors.ContainsKey(propertyName))
					ValidationErrors.Remove(propertyName);
			}
			else
			{
				if (!ValidationErrors.ContainsKey(propertyName))
					ValidationErrors.Add(propertyName, validationErrorDescription);
				else
					ValidationErrors[propertyName] = validationErrorDescription;
			}
		}
	}
    
	// this is another part of the interface - will be called automatically
	public IEnumerable GetErrors(string propertyName)
	{
		return ValidationErrors.ContainsKey(propertyName)
			? ValidationErrors[propertyName]
			: null;
	}

	// same here, another part of the interface - will be called automatically
	public bool HasErrors
	{
		get
		{
			return ValidationErrors.Count > 0;
		}
	}
}

And now somewhere in your code - your button command 'CanExecute' method can add to its implementation a call to VmEntity.HasErrors.

And may peace be upon your code regarding validation from now on :)

> The drawback is that this might run the validation on all properties a > bit too often, but most validations should be simply enough to not > hurt the performance. Another solution would be to remember which > properties produced errors using the validation and only check those, > but that seems a bit overcomplicated and unecessary for most times.

You don't need to track which properties have errors; you only need to know that errors exist. The view model can maintain a list of errors (also useful for displaying an error summary), and the IsValid property can simply be a reflection of whether the list has anything. You don't need to check everything each time IsValid is called, as long as you ensure that the error summary is current and that IsValid is refreshed each time it changes.

---

> In the end, this is a problem of the view. The view (and its data > binding engine) is responsible for giving the view model proper values > to work with. But in this case, there seems to be no good way to tell > the view model that it should invalidate the old property value.

You can listen to errors within the container that is bound to the view model:

container.AddHandler(Validation.ErrorEvent, Container_Error);

...

void Container_Error(object sender, ValidationErrorEventArgs e) {
    ...
}

This notifies you when errors are added or removed, and you can identify binding exceptions by whether e.Error.Exception exists, so your view can maintain a list of binding exceptions and inform the view model of it.

But any solution to this problem will always be a hack, because the view is not filling its role properly, which is giving the user a means of reading and updating the view model structure. This should be seen as a temporary solution until you correctly present the user with some kind of "integer box" instead of a text box.

Here's an effort to simplify things if you don't want to implement tons of additional code...

The scenario is that you have an int property in your viewmodel (could be decimal or another non-string type) and you bind a textbox to it in your view.

You have validation in your viewmodel that fires in the setter of the property.

In the view a user enters 123abc and the view logic highlights the error in the view, but can't set the property because the value is the wrong type. The setter never gets called.

The simplest solution is to change your int property in the viewmodel to be a string property, and cast the values into and out of it from the model. This allows the bad text to hit the setter of your property, and your validation code can then check the data and reject it as appropriate.

IMHO validation in WPF is broken, as can be seen from the elaborate (and ingenious) ways people have tried to work around the problem given previously. For me I don't want to add a huge amount of extra code or implement my own type classes to enable a textbox to validate so basing these properties on strings is something I can live with, even if it does feel like a bit of a kludge.

Microsoft should look at fixing this so that the scenario of invalid user input in a textbox bound to an int or decimal property can somehow communicate this fact elegantly to the viewmodel. It should be possible, for instance for them to create a new bound property for a XAML control to communicate view logic validation errors to a property in the viewmodel.

Thanks and respect to the other guys that have provided detailed answers to this topic.