.NET 10 Native AOT causes WinUI 3 desktop applications to not respond

[Gaoyifei1011]

Description

.NET 10 Native AOT 导致 WinUI 3 桌面应用程序未响应

--------------------------------------------------------------------

.NET 10 Native AOT causes WinUI 3 desktop applications to not respond

Reproduction Steps

我的应用程序之前使用的是 .NET 9 版本，今天发布了 .NET 10 后，升级 WinUI 3 桌面应用到 .NET 10 版本，在 Debug 模式下开启了 Native AOT 后，桌面应用程序运行一段时间后（通常在 10 秒到 20 秒左右）就会导致未响应，而且也没有触发任何异常，在 .NET 9 版本（无论是否开了 AOT）和 .NET 10 without Native AOT 正常运行，所以我猜测是 .NET 10 Native AOT 出现了问题，希望 .NET 团队能定位一下 Native AOT 的这个 bug 并尽快修复。

--------------------------------------------------------------

My application was previously using.NET 9 version. After.NET 10 was released today, I upgraded the WinUI 3 desktop application to.NET 10 version and enabled Native AOT in Debug mode. After the desktop application runs for a period of time (usually around 10 to 20 seconds), it will cause unresponse and no exceptions will be triggered. It runs normally in.NET 9 version (regardless of whether AOT is enabled or not) and.NET 10 without Native AOT. So I guess there is a problem the.NET 10 Native AOT. I hope the.NET team can locate this bug of Native AOT and fix it as soon as possible.

Expected behavior

.NET 10 和 .NET 9 在 AOT 模式下正常运行

--------------------------------------------------------------------

.NET 10 and.NET 9 operate normally in AOT mode

Actual behavior

.NET 10 WinUI 3 桌面应用程序在开启 Native AOT 后会发生未响应，且不触发异常，难以找到具体问题

--------------------------------------------------------------------

.NET 10 WinUI 3 desktop applications will not respond after enabling Native AOT, and no exception will be triggered. It is difficult to find the specific problem

Regression?

继续使用 .NET 9

--------------------------------------------------------------------

Continue using .NET 9

Known Workarounds

继续使用 .NET 9

--------------------------------------------------------------------

Continue using .NET 9

Configuration

.NET 10.0
Windows 11 26100.7070
X64
None
None

Other information

截图和我的项目以及主线程调用堆栈

--------------------------------------------------------------------

Screenshots, my project and the main thread call stack

https://github.com/Gaoyifei1011/GetStoreApp

[dotnet-policy-service]

Tagging subscribers to this area: @agocke, @MichalStrehovsky, @jkotas
See info in area-owners.md if you want to be subscribed.

[jkotas]

enabled Native AOT in Debug mode

Do you see the hang with Native AOT in Release mode as well, or is it Debug mode specific?

[Gaoyifei1011]

enabled Native AOT in Debug mode

Do you see the hang with Native AOT in Release mode as well, or is it Debug mode specific?

I copied the exe generated in the release mode to the debug directory for debugging.

[jkotas]

Is it intentional that you are disabling optimizations unconditionally for your app https://github.com/Gaoyifei1011/GetStoreApp/blob/main/GetStoreApp/GetStoreApp.csproj#L25? The bug is triggered by disabled optimizations. Could you please check whether the problem goes away when you delete <Optimize>false</Optimize> from your project file?

[jkotas]

The problem is that https://github.com/dotnet/runtime/blob/a12ecf93aedc8f111ec8cc5a8f6f51c899413d9e/src/coreclr/nativeaot/System.Private.CoreLib/src/System/Runtime/InteropServices/TrackerObjectManager.NativeAot.cs#L211C55-L211C83 assumes that the static constructor is interpreted by the AOT compiler. It does not happen when the optimizations are turned off. We may need to explicitly trigger the static constructor earlier to make this work with optimizations off.

Likely introduced by #116081 . cc @AaronRobinsonMSFT

[Gaoyifei1011]

The problem is that https://github.com/dotnet/runtime/blob/a12ecf93aedc8f111ec8cc5a8f6f51c899413d9e/src/coreclr/nativeaot/System.Private.CoreLib/src/System/Runtime/InteropServices/TrackerObjectManager.NativeAot.cs#L211C55-L211C83 assumes that the static constructor is interpreted by the AOT compiler. It does not happen when the optimizations are turned off. We may need to explicitly trigger the static constructor earlier to make this work with optimizations off.

Likely introduced by #116081 . cc @AaronRobinsonMSFT

However, that was not the case. The application still did not respond.

[jkotas]

Are you running debug build or release build? The bug should not repro on release build with optimizations on.

If you still see the hang, could you please check the stack trace? If you see EnsureClassConstructorRun and FindReferenceTargetsCallback in the stacktrace of the hang, it is still the same problem - the app is compiled with optimizations off.

[AaronRobinsonMSFT]

@Gaoyifei1011 The property ForceAotCompile is setting a value of ture, I assume it should be true.

[Gaoyifei1011]

@Gaoyifei1011 The property ForceAotCompile is setting a value of ture, I assume it should be true.

When conducting the local test, I changed it to "true".

[Gaoyifei1011]

@Gaoyifei1011 The property ForceAotCompile is setting a value of ture, I assume it should be true.

I did it this way in Net 9 without any problem. I wonder what improvements were made in Net 10's native AOT, but now this strange bug has occurred.

[Sergio0694]

"The problem is that https://github.com/dotnet/runtime/blob/a12ecf93aedc8f111ec8cc5a8f6f51c899413d9e/src/coreclr/nativeaot/System.Private.CoreLib/src/System/Runtime/InteropServices/TrackerObjectManager.NativeAot.cs#L211C55-L211C83 assumes that the static constructor is interpreted by the AOT compiler. It does not happen when the optimizations are turned off. We may need to explicitly trigger the static constructor earlier to make this work with optimizations off."

@jkotas can you clarify what's causing the issue here exactly? Looking at the code I'd expect the .cctor to just be triggered normally before the field is accessed (or like when the method is called, since the type is not beforefieldinit), and then that vtable field has [FixedAddressValueType], so taking the raw address should still be fine even if it wasn't pre-initialized. Of course I'm missing something since this is causing problems, but could you help me understand what's happening? Thank you 😄

[AaronRobinsonMSFT]

@Sergio0694 It is being trigger as expected. But that trigger is causing an allocation during a GC. The issue is a GC is being triggered during a GC and causing the deadlock.