I've written an assertion method Ensure.CurrentlyOnUiThread(), below, that checks that the current thread is a UI thread.
- Is this going to be reliable in detecting the Winforms UI thread?
- Our app is mixed WPF and Winforms, how best to detect a valid WPF UI thread?
- Is there a better way to do this? Perhaps code contracts?
Ensure.cs
using System.Diagnostics;
using System.Windows.Forms;
public static class Ensure
{
[Conditional("DEBUG")]
public static void CurrentlyOnUiThread()
{
if (!Application.MessageLoop)
{
throw new ThreadStateException("Assertion failed: not on the UI thread");
}
}开发者_如何学JAVA
}
Don't use
if(Dispatcher.CurrentDispatcher.Thread == Thread.CurrentThread)
{
// Do something
}
Dispatcher.CurrentDispatcher
will, if the current thread do not have a dispatcher, create and return a new Dispatcher
associated with the current thread.
Instead do like this
Dispatcher dispatcher = Dispatcher.FromThread(Thread.CurrentThread);
if (dispatcher != null)
{
// We know the thread have a dispatcher that we can use.
}
To be sure you have the correct dispatcher or are on the correct thread you have the following options
Dispatcher _myDispatcher;
public void UnknownThreadCalling()
{
if (_myDispatcher.CheckAccess())
{
// Calling thread is associated with the Dispatcher
}
try
{
_myDispatcher.VerifyAccess();
// Calling thread is associated with the Dispatcher
}
catch (InvalidOperationException)
{
// Thread can't use dispatcher
}
}
CheckAccess()
and VerifyAccess()
do not show up in intellisense.
Also, if you have to resort to these kinds of things its likely due to bad design. You should know which threads run what code in your program.
For WPF, I use the following:
public static void InvokeIfNecessary (Action action)
{
if (Thread.CurrentThread == Application.Current.Dispatcher.Thread)
action ();
else {
Application.Current.Dispatcher.Invoke(action);
}
}
The key is instead of checking Dispatcher.CurrentDispatcher (which will give you the dispatcher for the current thread), you need to check if the current thread matches the dispatcher of the application or another control.
Within WinForms you would normally use
if(control.InvokeRequired)
{
// Do non UI thread stuff
}
for WPF
if (!control.Dispatcher.CheckAccess())
{
// Do non UI Thread stuff
}
I would probably write a little method that uses a Generic constraint to determine which of these you should be calling. e.g.
public static bool CurrentlyOnUiThread<T>(T control)
{
if(T is System.Windows.Forms.Control)
{
System.Windows.Forms.Control c = control as System.Windows.Forms.Control;
return !c.InvokeRequired;
}
else if(T is System.Windows.Controls.Control)
{
System.Windows.Controls.Control c = control as System.Windows.Control.Control;
return c.Dispatcher.CheckAccess()
}
}
For WPF:
// You are on WPF UI thread!
if (Thread.CurrentThread == System.Windows.Threading.Dispatcher.CurrentDispatcher.Thread)
For WinForms:
// You are NOT on WinForms UI thread for this control!
if (someControlOrWindow.InvokeRequired)
Maybe Control.InvokeRequired
(WinForms) and Dispatcher.CheckAccess
(WPF) are OK for you?
You're pushing knowledge of your UI down into your logic. This is not a good design.
Your UI layer should be handling threading, as ensuring the UI thread isn't abused is within the purview of the UI.
This also allows you to use IsInvokeRequired in winforms and Dispatcher.Invoke in WPF... and allows you to use your code within synchronous and asynchronous asp.net requests as well...
I've found in practice that trying to handle threading at a lower level within your application logic often adds lots of unneeded complexity. In fact, practically the entire framework is written with this point conceded--almost nothing in the framework is thread safe. Its up to callers (at a higher level) to ensure thread safety.
Here is a snippet of code I use in WPF to catch attempts to modify UI Properties (that implement INotifyPropertyChanged) from a non-UI thread:
public event PropertyChangedEventHandler PropertyChanged;
private void NotifyPropertyChanged(String info)
{
// Uncomment this to catch attempts to modify UI properties from a non-UI thread
//bool oopsie = false;
//if (Thread.CurrentThread != Application.Current.Dispatcher.Thread)
//{
// oopsie = true; // place to set a breakpt
//}
if (PropertyChanged != null)
{
PropertyChanged(this, new PropertyChangedEventArgs(info));
}
}
For WPF:
I've needed to know is Dispatcher on my thread is actually started, or not. Because if you create any WPF class on the thread, the accepted answer will state that the dispatcher is there, even if you never do the Dispatcher.Run()
. I've ended up with some reflection:
public static class WpfDispatcherUtils
{
private static readonly Type dispatcherType = typeof(Dispatcher);
private static readonly FieldInfo frameDepthField = dispatcherType.GetField("_frameDepth", BindingFlags.Instance | BindingFlags.NonPublic);
public static bool IsInsideDispatcher()
{
// get dispatcher for current thread
Dispatcher currentThreadDispatcher = Dispatcher.FromThread(Thread.CurrentThread);
if (currentThreadDispatcher == null)
{
// no dispatcher for current thread, we're definitely outside
return false;
}
// get current dispatcher frame depth
int currentFrameDepth = (int) frameDepthField.GetValue(currentThreadDispatcher);
return currentFrameDepth != 0;
}
}
You can compare thread ids like this :
var managedThreadId = System.Windows.Threading.Dispatcher.FromThread(System.Threading.Thread.CurrentThread)?.Thread.ManagedThreadId;
var dispatcherManagedThreadId = System.Windows.Application.Current.Dispatcher.Thread.ManagedThreadId;
if (managedThreadId == dispatcherManagedThreadId)
{
//works in ui dispatcher thread
}
Using MVVM it is actually fairly easy. What I do is put something like the following in, say, ViewModelBase...
protected readonly SynchronizationContext SyncContext = SynchronizationContext.Current;
or...
protected readonly TaskScheduler Scheduler = TaskScheduler.Current;
Then when a particular ViewModel needs to touch anything "observable", you can check the context and react accordingly...
public void RefreshData(object state = null /* for direct calls */)
{
if (SyncContext != SynchronizationContext.Current)
{
SyncContext.Post(RefreshData, null); // SendOrPostCallback
return;
}
// ...
}
or do something else in the background before returning to context ...
public void RefreshData()
{
Task<MyData>.Factory.StartNew(() => GetData())
.ContinueWith(t => {/* Do something with t.Result */}, Scheduler);
}
Normally, if you follow MVVM (or any other architecture) in an orderly fashion, it is easy to tell where the responsibility for UI synchronization will be situated. But you can basically do this anywhere to return to the context where your objects are created. I'm sure it would be easy to create a "Guard" to handle this cleanly and consistently in a large and complex system.
I think it makes sense to say that your only responsibility is to get back to your own original context. It is a client's responsibility to do the same.
FOR WPF:
Here's a snippet based on the top answer, using a delegate meaning it is very generic.
/// <summary>
/// Invokes the Delegate directly on the main UI thread, based on the calling threads' <see cref="Dispatcher"/>.
/// NOTE this is a blocking call.
/// </summary>
/// <param name="method">Method to invoke on the Main ui thread</param>
/// <param name="args">Argumens to pass to the method</param>
/// <returns>The return object of the called object, which can be null.</returns>
private object InvokeForUiIfNeeded(Delegate method, params object[] args)
{
if (method == null) throw new ArgumentNullException(nameof(method));
var dispatcher = Application.Current.Dispatcher;
if (dispatcher.Thread != Thread.CurrentThread)
{
// We're on some other thread, Invoke it directly on the main ui thread.
return dispatcher.Invoke(method, args);
}
else
{
// We're on the dispatchers' thread, which (in wpf) is the main UI thread.
// We can safely update ui here, and not going through the dispatcher which safes some (minor) overhead.
return method.DynamicInvoke(args);
}
}
/// <inheritdoc cref="InvokeForUiIfNeeded(Delegate, object[])"/>
public TReturn InvokeForUiIfNeeded<TReturn>(Delegate method, params object[] args)
=> (TReturn) InvokeForUiIfNeeded(method, args);
The second method allows for a more type safe return type.
I've also added some overloads that automatically take the Func
and Action
parameters in my code, e.g:
/// <inheritdoc cref="InvokeForUiIfNeeded(System.Delegate, object[])"/>
private void InvokeForUiIfNeeded(Action action)
=> InvokeForUiIfNeeded((Delegate) action);
Note; the Func
and Action
inherit from Delegate
so we can just cast it.
You could also add your own generic overloads that take actions, i did not bother creating a bunch of overloads but you definitely could e.g;
/// <inheritdoc cref="InvokeForUiIfNeeded(System.Delegate, object[])"/>
private void InvokeForUiIfNeeded<T1>(Action<T1> action, T1 p1)
=> InvokeForUiIfNeeded((Delegate)action, p1);
/// <inheritdoc cref="InvokeForUiIfNeeded(System.Delegate, object[])"/>
private TReturn InvokeForUiIfNeeded<T1, TReturn>(Func<T1, TReturn> action, T1 p1)
=> (TReturn)InvokeForUiIfNeeded((Delegate)action, p1);
Thread.CurrentThread.ManagedThreadId == Dispatcher.Thread.ManagedThreadId
Is a better way to check this
精彩评论