| Description | I got an error, which had a stack trace like this:
at MyClass.UpdateThingA()
at MyClass.UpdateStuffInUI()
at System.Threading.Tasks.Task.<>c.<ThrowAsync>b__140_1(Object state)
at System.Threading.QueueUserWorkItemCallback.<>c.<.cctor>b__6_0(QueueUserWorkItemCallback quwi)
at System.Threading.ExecutionContext.RunForThreadPoolUnsafe[TState](ExecutionContext executionContext, Action`1 callback, TState& state)
at System.Threading.QueueUserWorkItemCallback.Execute()
at System.Threading.ThreadPoolWorkQueue.Dispatch()
The problem is, that stack trace should be impossible to get.
Constructor:
System.Threading.Timer timer = new System.Threading.Timer(_ => UpdateStuffInUI());
private async void UpdateStuffInUI()
{
// Compute some stuff while on the background thread that the Timer uses.
string a = timeConsumingMethod();
string b = timeConsumingMethod();
await ThreadSwitcher.ResumeForegroundAsync(Application.Current.Dispatcher)
UpdateThingA(a);
UpdateThingB(b);
}
That stack trace should be impossible. I should see the dispatcher in the stack, like so:
at UIShell.App.DemoAsyncMethod()
at System.Threading.ExecutionContext.RunInternal(ExecutionContext executionContext, ContextCallback callback, Object state)
at System.Runtime.CompilerServices.AsyncTaskMethodBuilder`1.AsyncStateMachineBox`1.MoveNext(Thread threadPoolThread)
at Noesis.Dispatcher.DispatcherOperation.Invoke(Delegate callback, Object args, SynchronizationContext context)
at Noesis.Dispatcher.ProcessQueue()
at Noesis.View.Update(Double timeInSeconds)
at NoesisApp.Window.Render(Double time)
at NoesisApp.Application.<Run>b__55_0(Display d)
at NoesisApp.Win32Display.EnterMessageLoop(Boolean runInBackground)
at NoesisApp.Application.Run()
at UIShell.App.Main(String[] args)
Looking at DispatcherThreadSwitcher (https://github.com/Noesis/Managed/blob/master/Src/NoesisApp/Core/Src/ThreadSwitcher/DispatcherThreadSwitcher.cs), it looks at the Dispatcher's CheckAccess method to determine if it needs to switch threads. The Dispatcher.CheckAccess method (https://github.com/Noesis/Managed/blob/master/Src/Noesis/Core/Src/Core/Dispatcher.cs) looks at the current thread's Thread ID vs. the Dispatcher's Thread ID. Both thread IDs are retrieved from unmanaged method Noesis_GetCurrentThreadId(), which I obviously don't have access to, but looks like it returns the unmanaged thread ID (Win32 method GetThreadId on Windows, some Posix function on Linux). In a quick test on my Windows system, the debugger showed that Dispatcher.ThreadID was 32756, which matched the unmanaged thread ID shown in the debugger. The Managed Thread ID was 1.
In my reading, it looks like the relationship between managed and unmanaged threads is not 1:1. It looks like a managed thread might be run on different unmanaged threads. That would explain why the thread ID that the dispatcher knows and the ThreadPoolWorkQueue thread that my method ran on ended up having the same thread ID.
I cannot think of any reason why the call to ResumeForegroundAsync wouldn't switch threads, other than this changing relationship between unmanaged and managed.
I suggest that CheckAccess, VerifyAccess, and CurrentDispatcher all switch to using managed thread ID, not unmanaged. In the Dispatcher class, this should be a matter of just replacing calls to the Dispatcher's helper property CurrentThreadID with a call to Thread.CurrentThread.ManagedThreadId. I didn't trace through any other objects that have a Thread ID that would need to change. |
|---|
