Invoking a Multi-Threaded DLL at Run-Time - winforms

All, I call a .NET DLL containing a WinForm at run-time from a WinForm C# application. To do this I use the following:
DLL = Assembly.LoadFrom(strDllPath);
classType = DLL.GetType(String.Format("{0}.{1}", strNamespaceName, strClassName));
if (classType != null)
{
if (bDllIsWinForm)
{
classInst = Activator.CreateInstance(classType);
Form dllWinForm = (Form)classInst;
dllWinForm.Show();
// Invoke required method.
MethodInfo methodInfo = classType.GetMethod(strMethodName);
if (methodInfo != null)
{
object result = null;
result = methodInfo.Invoke(classInst, new object[] { dllParams });
return result == null ? String.Empty : result.ToString();
}
}
}
This is invoking the WinForm DLL and the required method fine for serial methods within the DLL. However, I am now invoking a multi-threaded DLL, and calling the following method:
public async void ExecuteTest(object[] args)
{
Result result = new Result();
if (!BuildParameterObjects(args[0].ToString(), args[1].ToString()))
return;
IProgress<ProgressInfo> progressIndicator = new Progress<ProgressInfo>(ReportProgress);
List<Enum> enumList = new List<Enum>()
{
Method.TestSqlConnection,
Method.ImportReferenceTables
};
Task task = Task.Factory.StartNew(() =>
{
foreach (Method method in enumList)
{
result = Process.ProcessStrategyFactory.Execute(Parameters.Instance, progressIndicator,
Process.ProcessStrategyFactory.GetProcessType(method));
if (!result.Succeeded)
{
// Display error.
return;
}
}
});
await task;
Utilities.InfoMsg("VCDC run executed successfully.");
}
But this is returning control to the caller right away due to the await (which is expected). However, the return causes the calling method to exit which closes the DLL WinForm.
What is the best method to keep the DLL WinForm active/open?
Thanks for your time.
Edit. Following the suggestion by Stephen below I have decided to turn my DLL intery method type to Task<object> and set up a continuation as follows
if (classType != null)
{
if (bDllIsWinForm)
{
// To pass object array to constructor use the following.
// classInst = Activator.CreateInstance(classType, new object[] {dllParams});
classInst = Activator.CreateInstance(classType);
dllWinForm = (Form)classInst;
dllWinForm.Show();
// Invoke required method.
MethodInfo methodInfo = classType.GetMethod(strMethodName);
if (methodInfo != null)
{
object result = null;
result = methodInfo.Invoke(classInst, new object[] { dllParams });
if (result != null)
{
if (result.GetType() == typeof(Task<object>))
{
Task<object> task = (Task<object>)result;
task.ContinueWith(ant =>
{
object innerResult = task.Result;
return innerResult == null ? String.Empty : innerResult.ToString();
});
}
return result.ToString();
}
return String.Empty;
}
}
}
I decided to set up the continuation instead of the await to avoid the chaining that would occur with the await keyword - that is making the calling method (that calling the DLL of type Task<String> etc. up the call stack.
The DLL entry method now becomes:
public Task<object> ExecuteTest(object[] args)
{
Task<object> task = null;
Result result = new Result();
if (!BuildParameterObjects(args[0].ToString(), args[1].ToString()))
return task;
IProgress<ProgressInfo> progressIndicator = new Progress<ProgressInfo>(ReportProgress);
List<Enum> enumList = new List<Enum>()
{
Method.TestSqlConnection,
Method.ImportReferenceTables
};
task = Task.Factory.StartNew<object>(() =>
{
foreach (Method method in enumList)
{
result = Process.ProcessStrategyFactory.Execute(Parameters.Instance, progressIndicator,
Process.ProcessStrategyFactory.GetProcessType(method));
if (!result.Succeeded)
{
// Display error.
}
task.Wait(5000); // Wait to prevent the method returning too quickly for testing only.
}
return null;
});
return task;
}
But this causes the DLL WinForm to be show for a split second and then disapear. i even attempted to make the Form dllWinForm global to keep the refernce to the object active, but this also has not worked. I want to note that the call to the DLL (N.B. the calling method is already running on a background thread-pool thread).
Any further helps is appreciated.

Change the return type of Execute to Task and await it.

It's difficult to guess what you have in dll but eventually your dll code + exposed in your question:
dllWinForm.Show();
should eventually have been after juxtaposing:
new Thread
(
() => new Form().ShowDialog()
)
.Start();
Probably, you should change dllWinForm.Show(); by dllWinForm.ShowDialog().Start();
ShowDialog(), in contrast with Show(), starts its own message pumping and returns only when explicitly closed.
Update (in respnse to comment):
It is not absolutely necessary to launch a form from UI.
Since you are on .NET 4.5, it is probably simpler to use WPF (instead of Windows Form) form.
Here is the code for WPF form which, in order to adjust for Windows Form, you should change Dispatcher parts by initializing WindowsFormsSynchronizationContext
Though, IMO, WinForms code will be much more complex.

Related

How do I make sure the UI is updated during long running processes in a WPF application?

In a WPF app that follows the MVVM pattern, I've run across a common issue where a user clicks on a button which fires an event in the ViewModel. This event should enable a "Please Wait" spinner animation, do some processing which may take a few seconds, then hide the spinner. I'm not really sure of a good pattern I can use to make sure the spinner animation always appears.
As an example, I have a login process which does the following:
Displays spinner (set property on VM to true, spinner is bound to it)
Attempt to connect to server (can take a few seconds depending on connection)
On a failure, display a failure message
On success, save off some info about the user so it's available to the rest of the app.
What I'm finding is that the spinner never actually appears. I have tried wrapping the longer-running process in a Task.Run call, but that hasn't seemed to help.
Here's an approximation of what the code looks like:
// When true, spinner should be visible
protected bool _authenticatingIsVisible = false;
public bool AuthenticatingIsVisible
{
get { return _authenticatingIsVisible; }
set
{
_authenticatingIsVisible = value;
NotifyOfPropertyChange(() => AuthenticatingIsVisible);
}
}
public void Login()
{
try
{
AuthenticationIsVisible = true;
AuthCode result = AuthCode.NoAuthenticated;
Task.Run(() => { result = _client.Authenticate() }).Wait();
AuthenticationIsVisible = false;
if (result == AuthCode.Authenticated)
{
// Bit of misc. code to set up the environment
// Another check to see if something has failed
// If it has, displays a dialog.
// ex.
var error = new Error("Something Failed", "Details Here", Answer.Ok);
var vm = new DialogViewModel() { Dialog = error };
_win.ShowDialog(vm);
return;
}
else
{
DisplayAuthMessage(result);
}
}
finally
{
AuthenticationIsVisible = false;
}
}
The proper way would be not to block the UI thread (which is what you are doing right now with .Wait()), and use AsyncAwait instead.
private Task<AuthCode> Authenticate()
{
return Task.Run<AuthCode>(()=>
{
return _client.Authenticate();
});
}
public async void Login()
{
AuthenticationIsVisible = true;
AuthCode result = await Authenticate();
AuthenticationIsVisible = false;
}

Async-await seems to use the UI thread

In a view-model I use a factory:
private async Task<BaseData> InitializeAsync()
{
await InstancesAsync();
await ProjectsAsync();
await AdminAsync();
return this;
}
public static async Task<BaseData> CreateAsync()
{
var ret = new BaseData();
return await ret.InitializeAsync();
}
The awaited methods are rather staightforward, with e.g.
var instances = await TaskEx.Run(new Func<List<string>>(() => Agent.GetInstances()));
In the wpf view I want to set the DataContext in the constructor:
Loaded += delegate
{
Dispatcher.Invoke(new Action(async () => { DataContext = await BasisGegevens.CreateAsync(); }));
};
Although it works, I feel rather uncomfortable because the UI thread is used everywhere, also after the callbacks when the awaits complete. What am I missing?
Also I don't understand how to use the factory pattern for the DataContext because without the Invoke above I get the error that a different thread owns the object.
EDIT: using the ideas of Mr. Cleary I get:
Loaded += async (object sender, RoutedEventArgs e) =>
{ DataContext = await BaseData.CreateAsync(); };
public static Task<BaseData> CreateAsync()
{
var ret = new BaseData();
return ret.InitializeAsync();
}
private async Task<BaseData> InitializeAsync()
{
// UI thread here
await InstancesAsync().ConfigureAwait(false);
// thread 'a' here
await ProjectsAsync().ConfigureAwait(false);
// thread 'a' sometimes 'b' here
await AdminAsync().ConfigureAwait(false);
// thread 'a' or 'b' here
return this;
}
This works fine, except I cannot understand how ConfigureAwait(false) works.
Inside the method InstancesAsync() I have the awaited task:
var instances = await TaskEx.Run(new Func<List<string>>(() => Agent.GetInstances()));
After awaiting the repsonse, I return in the UI thread - I never expected that to happen!
Note that ProjectsAsync() and AdminAsync() behave the same, although they start on a worker (or background) thread!
I thougth that ConfigureAwait(true) has the effect of returning in the calling thread (in my case UI thread). I tested that and it is so.
Why do I see this with ConfigureAwait(false) too: because of a nested await, see comments.
I find it most useful to treat the ViewModel as having UI thread affinity. Think of it as the logical UI, even if it's not the actual UI. So all property and observable collection updates on ViewModel classes should be done on the UI thread.
In your async methods, if you don't need to return to the UI thread, then you can use ConfigureAwait(false) to avoid resuming on the UI thread. For example, if your various initialization methods are independent, you could do something like this:
private async Task<BaseData> InitializeAsync()
{
// Start all methods on the UI thread.
var instancesTask = InstancesAsync();
var projectsTask = ProjectsAsync();
var adminTask = AdminAsync();
// Await for them all to complete, and resume this method on a background thread.
await Task.WhenAll(instancesTask, projectsTask, adminTask).ConfigureAwait(false);
return this;
}
Also, any time you have return await, take another look to see if you can just avoid async/await entirely:
public static Task<BaseData> CreateAsync()
{
var ret = new BaseData();
return ret.InitializeAsync();
}
Finally, you should strongly avoid Dispatcher. Your Loaded event could be simplified to this:
Loaded += async ()
{
DataContext = await BasisGegevens.CreateAsync();
};

Silverlight async unit testing

I'm having a weird issue with Silverlight Unit Test Framework. The very first method executed fails, every time. I have a second test with the exact same code, and it passes. The strange thing about the first time it's called is that it actually waits for the timeout and then executes the repository call (underneath it's an HTTP PUT if you care). Here's the code - the first one fails every time, second one passes every time:
[TestMethod]
public void AuthShouldSucceed()
{
var autoResetEvent = new AutoResetEvent(false);
_authRepository.Authenticate(_username, _password, response =>
{
Assert.IsTrue(response);
autoResetEvent.Set();
});
if (!autoResetEvent.WaitOne(Constants.Timeout))
{
Assert.Fail("Test timed out.");
}
}
[TestMethod]
public void AuthShouldSucceed2()
{
var autoResetEvent = new AutoResetEvent(false);
_authRepository.Authenticate(_username, _password, response =>
{
Assert.IsTrue(response);
autoResetEvent.Set();
});
if (!autoResetEvent.WaitOne(Constants.Timeout))
{
Assert.Fail("Test timed out.");
}
}
Edit:
My final solution is a modification of Vladmir's solution:
[TestMethod]
[Asynchronous]
public void AuthShouldSucceed()
{
var complete = false;
var result = false;
_authRepository.Authenticate(_username, _password, response =>
{
complete = true;
result = response;
});
EnqueueConditional(() => complete);
EnqueueCallback(() => Assert.IsTrue(result));
EnqueueTestComplete();
}
If you're using Silverlight Unit Tests Framework try to rewrite your test next way:
[TestMethod]
[Asynchronous]
public void AuthShouldSucceed()
{
var done = false;
var authResult = false;
_authRepository.Authenticate(_username, _password, response =>
{
var done = true;
authResult = response;
});
EnqueueConditional(() => done);
EnqueueCallback(() => Assert.IsTrue(authResult));
EnqueueTestComplete();
}
Your test class should be derived from SilverlightTest class:
[TestClass]
public class MyTests: SilverlightTest
Summarizing all written you should be aware of couple important points in writing Unit Tests for Silverlight.
[Asynchronous]
Attribute of test method shows that you are testing part which includes async operations and want to syncronize them for testing purposes. Asynchrony itself in Silverlight Unit Tests Framework is a bit distorted. All "Enqueue" directives are like "postpone this until something happens" which is not a sense which the traditional programming asynchrony meaning has.
EnqueueConditional
Otherwords means wait until condition will be met. And this is one of the most important parts of async tests. After condition becomes legal all EnqueueCallbacks until the next EnqueueConditional will be executed.
Important note:
EnqueueConditional repeatedly calls the predicate passed to it on a
timer/background thread, checking each time to see if it returns true.
That's why you should avoid using heavy, complex logic within conditions.
EnqueueCallback
Defers code execution until EnqueueConditional that precedes it will met condition. It enqueues an Actoin or an array of Actions.
EnqueueDelay
Enqueues minimum number of milliseconds/time delay before continuing.
EnqueueTestComplete
Enqueues an action to call TestComplete which signals that a test is complete when using Async testing.
Important Note(from documentation):
If you use your own methods for completing, such as an HtmlTimer or
other threading method, it is possible that this call will occur
AFTER the test has timed out when using Timeouts. As such, be very careful as you could complete the call to the next test.
You can use the Asynchronous keyword in the TestMethod
[TestMethod]
[Asynchronous]
[Description("This test checks when NULL is passed")]
public void Testing()
{
bool done = false;
EnvViewModel test = new EnvViewModel ();
.
.
test.AsyncCallBackCompleted += (() => done = true);
EnqueueCallback(() => test.DataCommand.Execute(null));
EnqueueConditional(() => done);
EnqueueCallback(() => Assert.IsTrue(test.AADTDecisionModelList.Count == 0,
"The result does not have any data."));
EnqueueTestComplete();
}
And make a delegate in the ViewModel to be used to every method that is being tested...
public delegate void AsynCallComplete();
public event AsynCallComplete AsyncCallBackCompleted;
public void InformCallbackCompleted()
{
if (AsyncCallBackCompleted != null)
{
AsyncCallBackCompleted();
}
}

Exception thrown when using TaskSourceCompletion in Silverlight 5

I'm trying to wrap the Event Async Programming model used in RIA Services in a Task.
I have followed the standard way of using a TaskCompletionSource and implemented the following extension method:
public static Task<IEnumerable<T>> LoadAsync<T>(this DomainContext source, EntityQuery<T> query) where T : Entity
{
TaskCompletionSource<IEnumerable<T>> taskCompletionSource = new TaskCompletionSource<IEnumerable<T>>();
source.Load(
query,
loadOperation =>
{
if (loadOperation.HasError && !loadOperation.IsErrorHandled)
{
taskCompletionSource.TrySetException(loadOperation.Error);
loadOperation.MarkErrorAsHandled();
}
else if (loadOperation.IsCanceled)
{
taskCompletionSource.TrySetCanceled();
}
else
{
taskCompletionSource.TrySetResult(loadOperation.Entities);
}
},
null);
return taskCompletionSource.Task;
}
I then use this in the following way:
var task = _context.LoadAsync(_context.GetPlayersQuery());
task.Start();
task.Result;
The problem though is that I get an InvalidOperationException stating that "Start may not be called on a promise-style task". I have tried not starting the task, but then the loadOperation callback never fires.
Can anyone see what I am doing wrong here?
Thanks in advance
Problem is solved. Under the hood the DomainContext.Load() method is already operating in an asynchronous manner. There must have been some conflict with trying to wrap an already asynchronous method in a task.
However, even if I still follow the EAP correctly with the code below, I still get the InvalidOperationException of 'start cannot be called on a promise-style task'
public static Task<IEnumerable<T>> LoadAsync<T>(this DomainContext source, EntityQuery<T> query) where T : Entity
{
TaskCompletionSource<IEnumerable<T>> taskCompletionSource = new TaskCompletionSource<IEnumerable<T>>();
var loadOperation = source.Load(query);
loadOperation.Completed += (obj, args) =>
{
if (loadOperation.HasError && !loadOperation.IsErrorHandled)
{
taskCompletionSource.TrySetException(loadOperation.Error);
loadOperation.MarkErrorAsHandled();
}
else if (loadOperation.IsCanceled)
{
taskCompletionSource.TrySetCanceled();
}
else
{
taskCompletionSource.TrySetResult(loadOperation.Entities);
}
};
return taskCompletionSource.Task;
}
Try this instead
var result = await _context.LoadAsync(_context.GetPlayersQuery());
Try using
task.ContinuewWith(Action<Task<T>> continuation)
That worked for me, as I too got that exception when using task.Start

Surviving TPL, Delegates, Threads, and Invokes

I'm facing a serious problem of deadlock in a multithreaded desktop/windows application. I fear I'm not using the correct approach to delegates in a very async environment. Also, even though I "sink" my events into the calling UI thread, if possible, I still have to Invoke on the UI thread to see some action. Follows, is the details.
The application is basically a client for users of an online file storage service. That service exposes functionality through REST calls. I first created a managed code wrapper DLL for such calls that allows for a .NET consumer to create a static instance of this DLL and call the functions. I'll take the file upload operation as an example.
Now, in the wrapper, here is the public interface for a file upload:
public Int32 UploadFile(FileSystemObject FolderToUploadTo, FileInfo LocalFileAndPath, OperationProgressReportEventHandler onOperationProgressReport, FileSystemObjectUploadCompletedEventHandler onOperationCompleted) {
Int32 ReplyNumber = 0;
try {
var TheOperation = new UploadFileObjectOperation(FolderToUploadTo, LocalFileAndPath, _User.APIKey) {
onProgressReport = onOperationProgressReport,
onUploadCompleted = onOperationCompleted
};
//Add it to the pool of operations
OperationPool.Add(TheOperation);
//Start the operation through the factory
OperationFactory.StartNew(() => {
TheOperation.Start();
});
//Chain the *actual* TPL Task to flush after usage
TheOperation.InnerTask.ContinueWith(t => {
t.Dispose(); //Dispose the inner task
OperationPool.Remove(TheOperation); //Remove the operation from the pool
TheOperation = null; //Nullify the Operation
});
ReplyNumber = TheOperation.TaskId;
}
catch {
ReplyNumber = 0;
}
return ReplyNumber;
}
As you can see, the actual UI application, that will refer this DLL, will be sending delegates for progress and completed to the operation(s). Now, the body of the operation itself:
public class UploadFileObjectOperation : BaseOperation, IDisposable {
//Store
public FileSystemObjectUploadCompletedEventHandler onUploadCompleted;
//Constructors
//Disposing stuff
protected override void PerformWork() {
try {
//Init the WebClient
UploadClient.UploadProgressChanged += (UploadProgressChanged_s, UploadProgressChanged_e) => {
//This is my event in base class being raised
ReportProgress(UploadProgressChanged_e.ProgressPercentage, UploadProgressChanged_e);
};
UploadClient.UploadFileCompleted += (UploadFileCompleted_s, UploadFileCompleted_e) => {
if (UploadFileCompleted_e.Error != null) {
throw new ApplicationException("Upload failed. " + UploadFileCompleted_e.Error.Message);
}
JObject JSONLiveObject = JObject.Parse(Encoding.UTF8.GetString(UploadFileCompleted_e.Result));
if (String.Compare((String)JSONLiveObject["status"], Constants._CONST_RESTRESPONSE_STATUS_VALUE_FAIL, false) == 0) {
throw new ApplicationException("Upload response failed. " + (String)JSONLiveObject["result"]["message"]);
}
//Eureka! Success! We have an upload!
//This is my event being raised
UploadTaskCompleted(new UploadFileObjectOperationEventArg {
Error = null,
ResultSource = OperationResultSource.Fresh,
Status = OperationExitStatus.Success,
TaskId = TaskId,
UploadedFileSystemObject = _UploadedFile
});
};
//Start the async upload
UploadClient.UploadFileAsync(AddressOfRESTURI, UploadingMethod, _FileToUpload.FullName);
}
catch (OperationCanceledException exp_Canceled) {
UploadTaskCompleted(new UploadFileObjectOperationEventArg {
Error = exp_Canceled,
ResultSource = OperationResultSource.Fresh,
Status = OperationExitStatus.Canceled,
TaskId = TaskId,
UploadedFileSystemObject = _UploadedFile
});
// To ensure that the calling code knows the task was canceled
//throw;
}
catch (Exception exp) {
UploadTaskCompleted(new UploadFileObjectOperationEventArg {
Error = exp,
ResultSource = OperationResultSource.Fresh,
Status = OperationExitStatus.Error,
TaskId = TaskId,
UploadedFileSystemObject = _UploadedFile
});
// If the calling code also needs to know.
//throw;
}
}
protected void UploadTaskCompleted(UploadFileObjectOperationEventArg arg) {
if (onUploadCompleted == null)
return;
//Sinking into calling UI thread, if possible
if (onUploadCompleted.Target is Control) {
Control targetForm = onUploadCompleted.Target as Control;
targetForm.Invoke(onUploadCompleted, new object[] { arg });
}
else {
onUploadCompleted(arg);
}
Status = OperationRunningStatus.Completed;
}
}
The PerformWork() raises the two events: Progress reporting and completion. Note that while raising an event, I check if can get a route to the calling thread and push the event directly so to avoid invokes at UI.
Now, lets see how I'm using all of the above in a desktop client:
private void UploadFile(FileInfo DraggedFileInfo, FileSystemObject ParentDefination) {
SessionLifetimeStuff.APICore.UploadFile(ParentDefination, DraggedFileInfo,
(PercentageCompleted) => {
#region Progress
this.InvokeEx(f => {
UpdateTaskProgress(newTaskQueue.OID, PercentageCompleted.Progress, PercentageCompleted);
});
#endregion
}, (Result) => {
#region Completion
this.InvokeEx(f => {
switch (Result.Status) {
case OperationExitStatus.Success:
Console.WriteLine(String.Format("File: {0} uploaded to {1}", Result.UploadedFileSystemObject.DocumentFullname, Result.UploadedFileSystemObject.FolderId));
break;
case OperationExitStatus.Canceled:
DialogManager.ShowDialog(DialogTypeEnum.Warning, "Dropbox", "Upload canceled.", null, this);
break;
case OperationExitStatus.Error:
DialogManager.ShowDialog(DialogTypeEnum.Error, "Dropbox", "Upload failed.", Result.Error, this);
break;
}
});
#endregion
});
}
I'm using an extension method I found on Stackoverflow for adding the Invoking functionality:
public static class InvokeExtensions {
public static void InvokeEx<T>(this T #this, Action<T> action) where T : Control {
if (#this.InvokeRequired) {
#this.Invoke(action, new object[] { #this });
}
else {
if (!#this.IsHandleCreated)
return;
if (#this.IsDisposed)
throw new ObjectDisposedException("#this is disposed.");
action(#this);
}
}
public static IAsyncResult BeginInvokeEx<T>(this T #this, Action<T> action)
where T : Control {
return #this.BeginInvoke((Action)(() => #this.InvokeEx(action)));
}
public static void EndInvokeEx<T>(this T #this, IAsyncResult result)
where T : Control {
#this.EndInvoke(result);
}
}
In my code, i have commented out the invokes as I though i don't need then as the events being raised are coming in sinked. However, i realized that my UI was not doing anything at all. So, I added the InvokeEx({ code; }) and my UI started to shoe activity.
Now, why do I need to invoke?
If I attempt different operations from UI, eventually, my UI freezes although the application still behaves as normally functioning.
I found an old article at http://msdn.microsoft.com/en-us/library/ff649143.aspx#scag-ch06_topic4 that described the usage of delegates and I see that there is a IAsyncResult involved.
Could someone point me as to where I'm going wrong here?
Update:
Ok, with the invoking code commented on the UI, I get no activity at all. But upon using the this.InvokeEx or wrapping a work in this.BeginInvokeEx, I get UI updates but after a while, here are the two exceptions occurring (in this order):
Invoke or BeginInvoke cannot be called on a control until the window handle has been created.
A Task's exception(s) were not observed either by Waiting on the Task or accessing its Exception property. As a result, the unobserved exception was rethrown by the finalizer thread.

Resources