I'm trying to implement my first application using the MVVM pattern. I've manged to get most things working, but now I'm facing a problem with the following (IMHO pretty common) scenario:
Pressing a Button (View) shall invoke a Method (Model). Using a ICommand (ViewModel) this is pretty easy. But what to do if a time consuming operation has to be executed?
My current solution required me to implement a WorkQueue class containing WorkQueueItems. The WorkQueue has a Thread associated with it which executes the WorkQueueItems. Each WorkQueueItem has a Name, a Status and a Progress which is updated during execution.
Each Window has its own WorkQueue - visualized as StatusBar.
My problem: How can a ViewModel find the appropriate WorkQueue? Do I have to pass the WorkQueue to each ViewModel I create (this would be really be annoying)? Or are there other mechanism I could use?
I'm not really familiar with RoutedCommands - tough the basic concept seems to go into this direction. What'd love to see is a solution where I can bind a WorkQueueItem to a Command/Event which then bubbles up to the containing Window where it is added to the Window's WorkQueue.
I also considered making WorkQueue a Singleton - but this only works if I only have one Window at a time.
With the later .Net Frameworks (4.0+) and WPF you can utilize the System.Threading.Tasks library to provide a lot of this work under the hood.
If say your Command on your needs to update a property on your View Model, but it has to wait for the information, you simply start a task to perform the IO:
this.FindDataCommand = new RelayCommand<string>(
/* ICommand.Execute */
value =>
{
Task.Factory
.StartNew<IEnumerable<Foo>>(() => FindData(value))
.ContinueWith(
task =>
{
this.foundData.Clear();
this.foundData.AddRange(task.Result);
},
TaskScheduler.FromCurrentSynchronizationContext());
},
/* ICommand.CanExecute */
value => !String.IsNullOrWhitespace(value));
Breaking this down into manageable parts, we're starting a new task which calls some method IEnumerable<Foo> FindData(string). This is the plain old boring synchronous code you've always written. Likely it already exists on your view model!
Next we tell the framework to start a new task when that one finishes using ContinueWith, but to do it on the WPF Dispatcher instead. This allows you to avoid the hassles of cross-thread problems with UI elements.
You can extend this for monitoring with a helper class:
public class TaskManager
{
private static ConcurrentDictionary<Dispatcher, TaskManager> _map
= new ConcurrentDictionary<Dispatcher, TaskManager>();
public ObservableCollection<WorkItem> Running
{
get;
private set;
}
public TaskManager()
{
this.Running = new ObservableCollection<WorkItem>();
}
public static TaskManager Get(Dispatcher dispatcher)
{
return _map.GetOrAdd(dispatcher, new TaskManager());
}
// ...
Using this class in XAML would be along the lines of adding its instance to your Window's ViewModel:
public TaskManager CurrentTaskManager
{
get { return TaskManager.Get(Dispatcher.CurrentDispatcher); }
}
// <StatusBarItem Content="{Binding CurrentTaskManager.Running.Count}" />
You would then add a method to your TaskManager to handle the adding of tasks to and from the Running collection:
public Task<TResult> StartNew<TResult>(Func<TResult> work)
{
var task = Task.Factory
.StartNew<TResult>(work);
// build our view model
var workItem = new WorkItem(task);
this.Running.Add(workItem);
// Pass the result back using ContinueWith
return task.ContinueWith(
t => { this.Running.Remove(workItem); return t.Result; },
TaskScheduler.FromCurrentSynchronizationContext());
}
Now we simply change our FindDataCommand implementation:
TaskManager.Get(Dispatcher.CurrentDispatcher)
.StartNew<IEnumerable<Foo>>(() => FindData(value))
.ContinueWith(
task =>
{
this.foundData.Clear();
this.foundData.AddRange(task.Result);
},
TaskScheduler.FromCurrentSynchronizationContext());
The WorkItem class could expose the properties on the Task class to the UI, or it could be extended to encapsulate a CancellationToken to support cancellation in the future.
I'm not sure I got the question right, but I feel that using buil in Dispatcher would solve your problem and you do not need implementing WorkQueue manually since Dispatcher implements such a queue for you and able dispatching "worker items" to the UI/any thred using predefined set of priorities. You can execute an operation either synchronously or asynchronously using Dispatcher.Invoke() or Dispatcher.BeginInvoke()
Useful links:
MSDN Magazine: WPF Threads, Build More Responsive Apps With The Dispatcher
Related
I'm trying to follow along with Jonathan Worthington's airport announcement example in An Event-driven and Reactive Future
It compiles.
The problem: SayGateChange is never called. I'm new to Rx. I must be leaving something out. What I have here is his code as exactly as I could transcribe it. Sadly, there is no source available online.
AddGateChange is supposed to push a new item onto EventStreams.GateChanged, which in turn is supposed to be watched by Announcer.Announcements, which is supposed to be watched by SayGateChange.
I'm in Windows forms, not WPF, if that makes a difference.
I will gladly put it into a console app or LinqPad if that will make it work.
using Microsoft.VisualBasic;
using System;
using System.Collections;
using System.Collections.Generic;
using System.Data;
using System.Diagnostics;
using System.Collections.ObjectModel;
using System.Collections.Specialized;
using System.Reactive.Concurrency;
using System.Reactive.Linq;
using System.Reactive.Subjects;
using System.Threading;
public class frmAnnouncements
{
Announcer _Announcer = new Announcer();
ObservableCollection<string> Announcements = new ObservableCollection<string>();
private void frmRx_Load(System.Object sender, System.EventArgs e)
{
PopulateAnnouncements();
AddGateChange();
}
private void AddGateChange()
{
EventStreams.GateChanged.OnNext(new GateChanged {
Destination = "DAL",
FlightCode = 1503
});
}
private void PopulateAnnouncements()
{
_Announcer.Announcements.ObserveOnDispatcher().Subscribe(SayGateChange);
}
private void SayGateChange(string Message)
{
Interaction.MsgBox(Message);
}
public class GateChanged
{
public string FlightCode;
public string Destination;
}
public class EventStreams
{
public static Subject<GateChanged> GateChanged = new Subject<GateChanged>();
}
public class Announcer
{
public Announcer()
{
this.Announcements = EventStreams.GateChanged.Select(e => string.Format("gate change {0} to {1} ", e.FlightCode, e.Destination));
}
public IObservable<string> Announcements;
}
public frmAnnouncements()
{
Load += frmRx_Load;
}
}
As #Enigmativity stated, using ObserveOnDispatcher() is a problem - although without looking at Interaction.MsgBox its hard to be 100% certain it's the whole story - I guess it may be in the video, but it's rather long and I didn't watch it all.
The use of ObservableOnDispatcher() suggests you have pulled in the wrong nuget package for Rx:
For WPF applications, use rx-xaml (deprecated synonym rx-wpf), which provides the extension method ObserveOnDispatcher()
For Winforms applications, use rx-winforms, which provides the extension method overload ObserveOn(Control)
Both Winforms and WPF have a similar design where the user interface runs on a dedicated thread. In Winforms this is known as the "UI Thread" and in WPF as the "Dispatcher". Although the concept is very similar, the implementation is quite different.
ObserveOnDispatcher in WPF will cause the observer notifications OnXXX to be invoked on the dispatcher thread.
In WinForms, where you use ObserveOn(this), the this will generally be the form itself. For any WinForms control, this will locate the control's SynchronizationContext and Post OnXXX notifications to that.
Both overloads are smart in that invocations are direct if you happen to be on the correct Dispatcher thread or UI thread already.
I do seem to remember that WinForms is a lot more tolerant of updating UI off the UI thread - although this problem occurs in WPF too. This isn't a good thing, since it can lead to unpredictable results that are hard to debug. I note that the WinForms MessageBox.Show method, for example, doesn't care which thread it is invoked on since it creates it's own window. In general, use of some form of ObserveOn/ObserveOnDispatcher is always recommended in UI scenarios.
For this reason, it's a good idea to understand how these work in detail. For this, and to learn about the related SubscribeOn, have a look at this question.
I am surprised that you didn't get an informative InvalidOperationException stating that "The current thread has no Dispatcher associated with it." I can only think some other part of your code is swallowing exceptions, or you are using WPF code in your app as well and a Dispatcher had been created associated with the Winforms UI thread. That code behind Interaction.MsgBox is probably to blame for swallowing an error. Either way, I suggest removing rx-xaml to avoid confusion.
I want to unit test a view model which contains a registration like:
public SampleViewModel(IUnityContainer container)
{
...
Observable.FromEventPattern<PropertyChangedEventArgs>(gridViewModel, "PropertyChanged")
.**ObserveOnDispatcher()**
.Subscribe(_ => this.Update());
...
}
When I run the unit test it tells me that "The current thread has no Dispatcher associated with it." when reaching this code.
One solution would be to use a Scheduler but I don't want to modify the Viewmodel.
Is there a solution to make the unit test pass this statement without getting an error?
I would suggest that you provide you own IScheduler implementation to ObserveOn(IScheduler) instead of using the ObserveOnDispatcher() operator. I have used techniques for loading a DispatcherFrame or a Dispatcher but the problem is that you are still using a Dispatcher. Eventually I found that you just "fall off the cliff" especially once you have long running background threads involved. Following the guidelines of "No threading in Unit tests" just dont let the dispatcher get near your ViewModels! Your Unit tests will run much, much faster.
A far superior way to deal with this is to inject an interface that gives access to your Dispatcher Scheduler (via the IScheduler interface). This allows you to substitute in an implementation that exposes the TestScheduler. You now can control time in your unit test. You can control and validate which actions are marshalled to each scheduler.
This is a really old (pre-Rx) post on 'Unit' testing WPF with Dispatcher calls from early 2009. It seemed like a good idea at the time.
https://leecampbell.com/2009/02/17/responsive-wpf-user-interfaces-part-5/
More information on Testing with Rx and the TestScheduler is found in my other site on Rx
http://introtorx.com/Content/v1.0.10621.0/16_TestingRx.html
This works for me.
When setting up the unit test I create an application to simulate the environment for my VM:
static Application App;
static void BeforeTestRun()
{
var waitForApplicationRun = new ManualResetEventSlim();
Task.Run(() =>
{
App = new Application();
App.Startup += (s, e) => { waitForApplicationRun.Set(); };
App.Run();
});
waitForApplicationRun.Wait();
}
and this is how I use it to instanciate the view model.
App.Dispatcher.Invoke(() => { this.viewModel = new ViewModel(); });
To properly unit test your viewmodel, you really need to be able to supply all of its dependencies. In this case, your viewmodel has a dependency upon the dispatcher. Making your viewmodel take a IScheduler dependency is the ideal way. But if you really don't want to do that, then try looking at this duplicate question: Unit test IObservable<T> with ObserveOnDispatcher
I found a solution for avoiding the error, simply from Unit Test code instantiate the ViewModel by using a dispatcher like:
SampleViewModel sampleViewModel;
var dispatcher = Application.Current != null ? Application.Current.Dispatcher : Dispatcher.CurrentDispatcher;
dispatcher.Invoke((Action)(() => sampleViewModel = new SampleViewModel(this.container);
That's all and seems to work without modifying current code, maybe there are also better solutions.
I've got an ObservableCollection in a ViewModel that I need to modify in code on a background thread (using a Task object). I've also got an ICollectionView attached to that collection. When I modify the collection at run-time, the code runs fine. In unit testing I get the error, "This type of CollectionView does not support changes to its SourceCollection from a thread different from the Dispatcher thread."
I create the collection thusly (on the UI thread at run-time):
CashDeliveryDepots = new ObservableCollection<CashDeliveryDepot>();
FilteredCashDeliveryDepots = CollectionViewSource.GetDefaultView(CashDeliveryDepots);
and modify it using this code (which will be running on a non-UI thread at run-time):
CurrentDispatcher.Invoke(() =>
{
foreach (var depot in depots)
{
CashDeliveryDepots.Add(depot);
}
});
The CurrentDispatcher class executes code according to whether or not there is an application dispatcher.
internal static class CurrentDispatcher
{
internal static void Invoke(Action action)
{
if (Application.Current != null)
Application.Current.Dispatcher.Invoke(action);
else
action();
}
}
So at run-time the dispatcher is picked up correctly, the observable collection is updated and the filtered view updates too. When running the unit tests I get the exception at the point of trying to add a new item to the collection.
Can anyone suggest a way around this so I can get my tests working?
My questions are many. Since I saw. NET 4.5, I was very impressed. Unfortunately all my projects are .NET 4.0 and I am not thinking about migrating. So I would like to simplify my code.
Currently, most of my code that usually take enough time to freeze the screen, I do the following:
BackgroundWorker bd = new BackgroundWorker();
bd.DoWork += (a, r) =>
{
r.Result = ProcessMethod(r.Argument);
};
bd.RunWorkerCompleted += (a, r) =>
{
UpdateView(r.Result);
};
bd.RunWorkerAsync(args);
Honestly, I'm tired of it. And that becomes a big problem when there is a logic complex user interaction.
I wonder, how to simplify this logic? (Remember that I'm with. Net 4.0) I noticed a few things by google, but not found anything easy to implement and suitable for my needs.
I thought this solution below:
var foo = args as Foo;
var result = AsyncHelper.CustomInvoke<Foo>(ProcessMethod, foo);
UpdateView(result);
public static class AsyncHelper
{
public static T CustomInvoke<T>(Func<T, T> func, T param) where T : class
{
T result = null;
DispatcherFrame frame = new DispatcherFrame();
Task.Factory.StartNew(() =>
{
result = func(param);
frame.Continue = false;
});
Dispatcher.PushFrame(frame);
return result;
}
}
I am not sure about the impact is on manipulating the dispatcher frame.
But I know That it would work very well, for example, I could use it in all the events of controls without bothering to freeze the screen.
My knowledge about generic types, covariance, contravariance is limited, maybe this code can be improved.
I thought of other things using Task.Factory.StartNew and Dispatcher.Invoke, but nothing that seems interesting and simple to use. Can anyone give me some light?
You should just use the Task Parallel Library (TPL). The key is specifying the TaskScheduler for the current SynchronizationContext for any continuations in which you update the UI. For example:
Task.Factory.StartNew(() =>
{
return ProcessMethod(yourArgument);
})
.ContinueWith(antecedent =>
{
UpdateView(antecedent.Result);
},
TaskScheduler.FromCurrentSynchronizationContext());
Aside from some exception handling when accessing the antecedent's Result property, that's all there is too it. By using FromCurrentSynchronizationContext() the ambient SynchronizationContext that comes from WPF (i.e. the DispatcherSynchronizationContext) will be used to execute the continuation. This is the same as calling Dispatcher.[Begin]Invoke, but you are completely abstracted from it.
If you wanted to get even "cleaner", if you control ProcessMethod I would actually rewrite that to return a Task and let it own how that gets spun up (can still use StartNew internally). That way you abstract the caller from the async execution decisions that ProcessMethod might want to make on its own and instead they only have to worry about chaining on a continuation to wait for the result.
UPDATE 5/22/2013
It should be noted that with the advent of .NET 4.5 and the async language support in C# this prescribed technique is outdated and you can simply rely on those features to execute a specific task using await Task.Run and then execution after that will take place on the Dispatcher thread again automagically. So something like this:
MyResultType processingResult = await Task.Run(() =>
{
return ProcessMethod(yourArgument);
});
UpdateView(processingResult);
How about encapsulating the code that is always the same in a reusable component? You could create a Freezable which implements ICommand, exposes a property of Type DoWorkEventHandler and a Result property. On ICommand.Executed, it would create a BackgroundWorker and wire up the delegates for DoWork and Completed, using the value of the DoWorkEventHandler as event handler, and handling Completed in a way that it sets its own Result property to the result returned in the event.
You'd configure the component in XAML, using a converter to bind the DoWorkEventHandler property to a method on the ViewModel (I assume you've got one), and bind your View to the component's Result property, so it gets updated automatically when Result does a change notification.
The advantages of this solution are: it is reusable, and it works with XAML only, so no more glue code in your ViewModel just for handling BackgroundWorkers. If you don't need your background process to report progress, it could even be unaware that it runs on a background thread, so you can decide in the XAML whether you want to call a method synchronously or asynchronously.
A few months have passed, but could this help you?
Using async/await without .NET Framework 4.5
I've start using prism with silverlight 3, but, we are trying to implement it to work with ADO.NET DataServices. The "DataServiceQuery" query type required to use with Silverlight, requires a Asyncronous call to be fired after the query. This will break ous Prism Pattern by what I can see.
Any ideas to get only the data of the query to use in Prism Pattern? Correct-me anyone if i'm wrong!
Making an Asynchronous call to your server doesn't break "Prism Pattern". When your view needs to query the server, its viewmodel fires an asynchronous request and provides a callback. Once callback is called, it handles the result and updates whatever properties it exposes to a view. This will result in view updating according to bindings you set up in your xaml.
PL is exactly right. There's really no patterns that Prism encourages that are incompatible with ADO.NET Data Services. There are just a few things you should know.
Here's a small sample. It's a little tricky... the complete event will sometimes fire outside of the UI thread, so you have to handle it with the dispatcher (at least in SL2 you did):
public class MyViewModel : BaseViewModel
{
public Customer CustomerResult
{
...
}
NorthwindEntities svcContext = null;
public MyViewModel()
{
svcContext =
new NorthwindEntities(new Uri("Northwind.svc", UriKind.Relative));
DataServiceQuery<Customers> query =
svcContext.Customers.Expand("Orders");
// Begin the query execution.
query.BeginExecute(WorkComplete, query);
}
private void WorkComplete(IAsyncResult result)
{
DataServiceQuery<Customers> query =
result.AsyncState as DataServiceQuery<Customers>;
Customers returnedCustomer =
query.EndExecute(result).FirstOrDefault();
//Execute with the dispatcher
Dispatcher.CurrentDispatcher.BeginInvoke( () =>
{
CustomerResult = returnedCustomer;
});
}
}
Of course there is no exception handling in here, but you get the picture hopefully.