title author date CreateTime categories
WPF Main thread gets a deadlock when stylus input thread is waiting for the window to close
lindexi
2018-11-01 09:32:42 +0800
2018-10-30 16:57:37 +0800
WPF

We found two way that can make the main thread locked. And we can not write any code to solve it and it can only be circumvented.
The easiest way to reproduce this issue is to wait for the window in the main thread to close in the stylus input thread.

We have found two ways, the first way always happens, and the second way is probabilistic.

Before we tell you about it, we need to tell you something about the touch thread and why it can make the main thread wait forever.

Theory

The stylus input thread gets the input event when the user touches the screen.

There is a ThreadProc method running in the stylus input thread and this method has a loop inside which will never end until the application exists.

void ThreadProc()
{
// The loop that never ends.
while (!__disposed)
{ }
}

There are nested two loops in the ThreadProc method. In the outside one, it adds and removes PenContext and in the inside one, it will be blocked by the PENIMC and can be continued when the user touches the screen.

void ThreadProc()
{
while (!__disposed)
{
// The outside loop
// To remove or add the PenContext while (true)
{
// The inside loop
// Tt will be blocked by the PENIMC
if(!Penimc.UnsafeNativeMethods.GetPenEvent(/*the thread locker*/))
{
// If the `_pimcResetHandle` is released, this if branch will enter so the inside loop will end with the `break` and the code runs back to the outside loop.
break;
} FireEvent(/*fire the touch events*/);
}
}
}

When a window is closed, it calls HwndSource.DisposeStylusInputProvider and this causes the PenContext.Disable be calling with the calling stack trace showing below.

System.Windows.Input.PenThreadWorker.WorkerRemovePenContext(System.Windows.Input.PenContext penContext)
System.Windows.Input.PenContext.Disable(bool shutdownWorkerThread)
System.Windows.Input.PenContexts.Disable(bool shutdownWorkerThread)
System.Windows.Input.StylusWisp.WispLogic.UnRegisterHwndForInput(System.Windows.Interop.HwndSource hwndSource)
System.Windows.Interop.HwndStylusInputProvider.Dispose()

Let us see the PenThreadWorker.WorkerRemovePenContext that run in the main thread.

internal bool WorkerRemovePenContext(PenContext penContext)
{
var operationRemoveContext = new PenThreadWorker.WorkerOperationRemoveContext(penContext, this); _workerOperation.Add((PenThreadWorker.WorkerOperation) operationRemoveContext);
// Release the _pimcResetHandle lock
UnsafeNativeMethods.RaiseResetEvent(this._pimcResetHandle.Value);
// Wait for the operationRemoveContext to finish
operationRemoveContext.DoneEvent.WaitOne();
operationRemoveContext.DoneEvent.Close();
return operationRemoveContext.Result;
}

From the code above we can learn that the main thread releases the _pimcResetHandle and it makes the ThreadProc breaking the inside loop and go back to the outside one to remove the PenContext.

Normally we should run the code in the stylus input thread to remove the PenContext and keep the main thread waiting for the operationRemoveContext to finish. But if the stylus input thread never remove the PenContext and the main thread waits for it, the main thread will never continue.

The first way

The first way is to write a custom class implementing StylusPlugIn and wait for a window to close in the OnStylusUp method.

Let's create a new empty window named FooWindow.

public class FooWindow : Window
{ }

Then we create a FooStylusPlugIn class to implement the StylusPlugIn with overriding the OnStylusUp method. We add some code to wait for the window to close by calling Invoke which will wait by pumping a new message loop.

public class FooStylusPlugIn : StylusPlugIn
{
public FooStylusPlugIn(FooWindow fooWindow)
{
FooWindow = fooWindow;
} public FooWindow FooWindow { get; } /// <inheritdoc />
protected override void OnStylusUp(RawStylusInput rawStylusInput)
{
FooWindow.Dispatcher.Invoke(() => FooWindow.Close());
base.OnStylusUp(rawStylusInput);
}
}

To combine both the critical codes above, we write some codes in the MainWindow. The FooWindow is instanced in the constructor and the StylusPlugIn is plugged in it. We also make a button in the XAML that can let us know whether the main thread is still running or not.

public partial class MainWindow : Window
{
public MainWindow()
{
InitializeComponent();
_fooWindow = new FooWindow();
StylusPlugIns.Add(new FooStylusPlugIn(_fooWindow));
_fooWindow.Show();
} private void Button_OnClick(object sender, RoutedEventArgs e)
{
} private FooWindow _fooWindow;
}

Run the project, touch the main window, and you'll find that the main window never responds to your interaction. Try to click the button to view the responding and you'll soon verify what I'm talking.

The reason is that the OnStylusUp in FooStylusPlugIn is running in the stylus input thread which is also running the inside loop of the ThreadProc method. It needs to go back to the outside loop to remove the PenContext when a window is closed. The stylus input thread is waiting for the main thread to close a window and the main thread is also waiting for the stylus input thread remove PenContext. Thus, the deadlock occurred.

The demo in github

The second way

If a touch happens exactly during a window closing, the main thread will enter a lock.

The difference between the first method and the second method is that the first one will lock both the main thread and the stylus input thread but the second one will only lock the main thread.

From the theory, we know that the PenContext should be removed correctly in the outside loop. But in the second way, the stylus input thread is firing the touch event exactly when we run the code to remove the PenContext in the stylus input thread. As you can see we need to run the code to remove PenContext in the outside loop but at this moment the code is firing the touch event in the second loop.

The firing of the touch event means the _pimcResetHandle is released. Although the main thread has also released the lock the code cannot run to the outside loop to remove the PenContext and the main thread can no longer wait for the moment when the PenContext removal is finished.

void ThreadProc()
{
while (!__disposed)
{
// The outside loop
// To remove or add the PenContext
// The main thread is waiting for its finishing.
RemovePenContext(); while (true)
{
// The inside loop
// Tt will be blocked by the PENIMC
if(!Penimc.UnsafeNativeMethods.GetPenEvent(/*wait the lock*/))
{
// If the `_pimcResetHandle` is released, this if branch will enter so the inside loop will end with the `break` and the code runs back to the outside loop.
break;
} FireEvent(/*fire the touch events*/); // the code is running in this line
// and the `_pimcResetHandle` is released.
// the main thread release the `_pimcResetHandle` but the code can not go to RemovePenContext for it will no longer break.
}
}
}

The main thread has released the lock but the stylus input thread doesn't need to wait for the lock. The stylus input thread cannot go back to the outside loop to remove the PenContext and main thread can no longer wait for the moment when the PenContext removal is finished.

Thanks to walterlv for proofreading the English translation of this post.

感谢 吕毅 对本文的英文翻译进行校对。

2018-11-1-WPF-Main-thread-gets-a-deadlock-when-stylus-input-thread-is-waiting-for-the-window-to-clos...的更多相关文章

  1. OI生涯回忆录 2018.11.12~2019.4.15

    上一篇:OI生涯回忆录 2017.9.10~2018.11.11 一次逆风而行的成功,是什么都无法代替的 ………… 历经艰难 我还在走着 一 NOIP之后,全机房开始了省选知识的自学. 动态DP,LC ...

  2. China Intelligent Office Summit(2018.11.21)

    时间:2018.11.21地点:中关村软件园国际会议中心

  3. International Programming Retreat Day(2018.11.17)

    时间:2018.11.17地点:北京国华投资大厦

  4. Intel Artificial Intelligence Conference(2018.11.14)

    时间:2018.11.14地点:北京国贸大酒店

  5. Debian系统中当安装deb软件时出现:deb cdrom:[Debian GNU/Linux 9.3.0 _Stretch_ - Official amd64 DVD Binary-1 20171209-12:11]/ stretch contrib main

    vi /etc/apt/sources.list // 注释掉下面这句话# deb cdrom:[Debian GNU/Linux 9.3.0 _Stretch_ - Official amd64 D ...

  6. 2018.11.23 浪在ACM 集训队第六次测试赛

    2018.11.23 浪在ACM 集训队第六次测试赛 整理人:刘文胜 div 2: A: Jam的计数法 参考博客:[1] 万众 B:数列 参考博客: [1] C:摆花 参考博客: [1] D:文化之 ...

  7. 2018.11.16 浪在ACM 集训队第五次测试赛

    2018.11.16 浪在ACM 集训队第五次测试赛 整理人:李继朋 Problem A : 参考博客:[1]朱远迪 Problem B : 参考博客: Problem C : 参考博客:[1]马鸿儒 ...

  8. 2018.11.9浪在ACM集训队第四次测试赛

    2018.11.9浪在ACM集训队第四次测试赛 整理人:朱远迪 A 生活大爆炸版 石头剪刀布           参考博客:[1] 刘凯 B 联合权值            参考博客: [1]田玉康 ...

  9. 2018.11.2浪在ACM集训队第三次测试赛

    2018.11.2 浪在ACM 集训队第三次测试赛 整理人:孔晓霞 A 珠心算测试 参考博客:[1]李继朋  B 比例简化 参考博客: [1]李继朋 C 螺旋矩阵 参考博客:[1]朱远迪 D 子矩阵 ...

  10. Thread.Sleep(0) vs Sleep(1) vs Thread.Yeild()

    注:本文为个人学习摘录,原文地址:http://www.cnblogs.com/stg609/p/3857242.html 本文将要提到的线程及其相关内容,均是指 Windows 操作系统中的线程,不 ...

随机推荐

  1. 一个windows 两个jar

    设置两个子JAVA_HOME,一个总设置两个子JAVA_HOME:JAVA_HOME6 = C:\Program Files\Java\jdk1.6.0_43JAVA_HOME8 = C:\Progr ...

  2. String方法之fromCharCode()和charCodeAt()

    1.fromCharCode fromCharCode() 可接受一个指定的 Unicode 值,然后返回一个字符串. 语法   我们可以根据 Unicode 来输出 "HELLO" ...

  3. 安装hadoop伪分布式

    修改hosts cat /etc/hosts 127.0.0.1 mo.don.com 创建用户 useradd hadoop passwd hadoop sudo授权 visudo hadoop A ...

  4. Java—重写与重载的区别

    1.重写(Override) 子类继承了父类原有的方法,但有时子类并不想原封不动的继承父类中的某个方法,所以在方法名,参数列表,返回类型(除了子类中方法的返回值是父类中方法返回值的子类时)都相同的情况 ...

  5. java扫描某个包下的所有java类并加载

    最近在学习java的反射和注解,实际情景中需要扫描某个包下的所有java类,然后使用类加载器加载类. 基本思路,获得程序的路径扫描src下某个包内的子包和java类,实现也比较简单. 运行环境:win ...

  6. 1、Redis的使用

    非关系型数据库分类: 分类 典型代表 典型应用场景 数据类型 优点 缺点 键值 (key-value) Tokyo Cabinet/Tyrant, Redis, Voldemort, Oracle B ...

  7. Ubuntu 16.04下apt安装ssh及卸载(含FileZilla 使用过程)

    前言:文件发布测试过程中需要使用Windows和Ubuntu实现文件共享. 以下内容转载自:http://blog.csdn.net/jiang13479/article/details/535014 ...

  8. Odoo Documentation : Fields

    Fields Basic fields class openerp.fields.Field(string=None, **kwargs) The field descriptor contains ...

  9. [转]模块化——Common规范及Node模块实现

    Node在实现中并非完全按照CommonJS规范实现,而是对模块规范进行了一定的取舍,同时也增加了少许自身需要的特性.本文将详细介绍NodeJS的模块实现 引入 nodejs是区别于javascrip ...

  10. 升级gitk后,Error in startup script: unknown color name "lime"

    $ gitkError in startup script: unknown color name "greeen" (processing "-fore" o ...