[nativeaot] fix default `SynchronizationContext` #9883

jonathanpeppers · 2025-03-05T19:42:47Z

Context: b6c22f3909

Running Mono.Android-Tests under NativeAOT has the test failure:

03-04 23:06:20.021  6191  6211 I NUnit   : SynchronizationContext_Is_ThreadingSynchronizationContextCurrent
03-04 23:06:20.032  6191  6211 E NUnit   : 	[FAIL]
03-04 23:06:20.032  6191  6211 E NUnit   :  :   Expected: True
03-04 23:06:20.032  6191  6211 E NUnit   :   But was:  False
03-04 23:06:20.032  6191  6211 E NUnit   :    at libMono.Android.NET-Tests!<BaseAddress>+0x1482f63
03-04 23:06:20.032  6191  6211 E NUnit   :    at System.Reflection.DynamicInvokeInfo.Invoke(Object, IntPtr, Object[], BinderBundle, Boolean) + 0xf3

Initially, if we called JNIEnvInit.SetSynchronizationContext() at startup, this can result in the static ctor for
Android.App.Application running too soon before JNIEnvInit.InitializeJniRuntime() has completed.

To fix this:

Add an empty static ctor to Android.App.Application to remove the beforefieldinit flag from the type.
Call JNIEnvInit.SetSynchronizationContext() at startup for NativeAOT.

With this change in place, no crashes occur at startup and the test passes.

Running `Mono.Android-Tests` under NativeAOT has the test failure: 03-04 23:06:20.021 6191 6211 I NUnit : SynchronizationContext_Is_ThreadingSynchronizationContextCurrent 03-04 23:06:20.032 6191 6211 E NUnit : [FAIL] 03-04 23:06:20.032 6191 6211 E NUnit : : Expected: True 03-04 23:06:20.032 6191 6211 E NUnit : But was: False 03-04 23:06:20.032 6191 6211 E NUnit : at libMono.Android.NET-Tests!<BaseAddress>+0x1482f63 03-04 23:06:20.032 6191 6211 E NUnit : at System.Reflection.DynamicInvokeInfo.Invoke(Object, IntPtr, Object[], BinderBundle, Boolean) + 0xf3 Initially, if we called `JNIEnvInit.SetSynchronizationContext()` at startup, this can result in the static ctor for `Android.App.Application` running *too soon* before `JNIEnvInit.InitializeJniRuntime()` has completed. To fix this: * Add an empty `static ctor` to `Android.App.Application` to remove the `beforefieldinit` flag from the type. * Call `JNIEnvInit.SetSynchronizationContext()` at startup for NativeAOT. With this change in place, no crashes occur at startup and the test passes.

jonathanpeppers · 2025-03-05T22:23:58Z

The test passes here:

https://devdiv.visualstudio.com/DevDiv/_build/results?buildId=11134260&view=ms.vss-test-web.build-test-results-tab&runId=123783871&resultId=100002&paneView=debug

dellis1972 · 2025-03-06T13:16:15Z

I've tested this locally and it doesn't seem to fix my test case.

jonathanpeppers · 2025-03-06T14:22:43Z

I've tested this locally and it doesn't seem to fix my test case.

I think @dellis1972's issue might be something different:

03-06 13:32:32.094  6966  7016 E AndroidGameView: Java.Lang.ClassNotFoundException: mono.java.lang.RunnableImplementor
03-06 13:32:32.094  6966  7016 E AndroidGameView:    at Java.Interop.JniEnvironment.Types.TryFindClass(String classname, Boolean throwOnError) + 0x418
03-06 13:32:32.094  6966  7016 E AndroidGameView:    at Android.Runtime.JNIEnv.FindClass(String classname) + 0x30
03-06 13:32:32.094  6966  7016 E AndroidGameView:    at Android.Runtime.JNIEnv.AllocObject(String jniClassName) + 0x10
03-06 13:32:32.094  6966  7016 E AndroidGameView:    at Java.Lang.Thread.RunnableImplementor..ctor(Action handler, Boolean removable) + 0x38
03-06 13:32:32.094  6966  7016 E AndroidGameView:    at Android.OS.Handler.Post(Action action) + 0x34
03-06 13:32:32.094  6966  7016 E AndroidGameView:    at Android.App.SyncContext.Send(SendOrPostCallback d, Object state) + 0xf4
03-06 13:32:32.094  6966  7016 E AndroidGameView:    at Microsoft.Xna.Framework.MonoGameAndroidGameView.WorkerThreadFrameDispatcher(SynchronizationContext uiThreadSyncContext) + 0x180
03-06 13:32:32.094  6966  7016 E AndroidGameView:   --- End of managed Java.Lang.ClassNotFoundException stack trace ---
03-06 13:32:32.094  6966  7016 E AndroidGameView: java.lang.ClassNotFoundException: mono.java.lang.RunnableImplementor

jonpryor · 2025-03-07T19:43:18Z

Context: https://github.com/dotnet/android/pull/9846
Context: b6c22f3909741aa0418b2c17eef7bf3895a797bb

Running `Mono.Android-Tests` under NativeAOT has the test failure:

	I NUnit   : SynchronizationContext_Is_ThreadingSynchronizationContextCurrent
	E NUnit   : 	[FAIL]
	E NUnit   :  :   Expected: True
	E NUnit   :   But was:  False
	E NUnit   :    at libMono.Android.NET-Tests!<BaseAddress>+0x1482f63
	E NUnit   :    at System.Reflection.DynamicInvokeInfo.Invoke(Object, IntPtr, Object[], BinderBundle, Boolean) + 0xf3

`SynchronizationContext_Is_ThreadingSynchronizationContextCurrent()`
failed because we never called `JNIEnvInit.SetSynchronizationContext()`.

Call `JNIEnvInit.SetSynchronizationContext()`, and…

    --------- beginning of crash
    03-07 15:48:39.566 17058 17119 F libc    : Fatal signal 6 (SIGABRT), code -1 (SI_QUEUE) in tid 17119 (Thread-11), pid 17058 (droid.NET_Tests)
    03-07 15:48:39.661 17145 17145 I crash_dump64: obtaining output fd from tombstoned, type: kDebuggerdTombstoneProto
    03-07 15:48:39.662   693   693 I tombstoned: received crash request for pid 17119
    03-07 15:48:39.663 17145 17145 I crash_dump64: performing dump of process 17058 (target tid = 17119)
    03-07 15:48:39.828 17145 17145 F DEBUG   : *** *** *** *** *** *** *** *** *** *** *** *** *** *** *** ***
    03-07 15:48:39.828 17145 17145 F DEBUG   : Build fingerprint: 'google/panther/panther:14/AP2A.240905.003/12231197:user/release-keys'
    03-07 15:48:39.828 17145 17145 F DEBUG   : Revision: 'MP1.0'
    03-07 15:48:39.828 17145 17145 F DEBUG   : ABI: 'arm64'
    03-07 15:48:39.828 17145 17145 F DEBUG   : Timestamp: 2025-03-07 15:48:39.677241696-0600
    03-07 15:48:39.828 17145 17145 F DEBUG   : Process uptime: 2s
    03-07 15:48:39.828 17145 17145 F DEBUG   : Cmdline: Mono.Android.NET_Tests
    03-07 15:48:39.828 17145 17145 F DEBUG   : pid: 17058, tid: 17119, name: Thread-11  >>> Mono.Android.NET_Tests <<<
    03-07 15:48:39.828 17145 17145 F DEBUG   : uid: 10582
    03-07 15:48:39.828 17145 17145 F DEBUG   : tagged_addr_ctrl: 0000000000000001 (PR_TAGGED_ADDR_ENABLE)
    03-07 15:48:39.828 17145 17145 F DEBUG   : signal 6 (SIGABRT), code -1 (SI_QUEUE), fault addr --------
    03-07 15:48:39.828 17145 17145 F DEBUG   :     x0  0000000000000000  x1  00000000000042df  x2  0000000000000006  x3  0000007c6b4838f0
    03-07 15:48:39.828 17145 17145 F DEBUG   :     x4  0000000000000000  x5  0000000000000000  x6  0000000000000000  x7  0000007f36a58004
    03-07 15:48:39.828 17145 17145 F DEBUG   :     x8  00000000000000f0  x9  0000007f18a6d350  x10 0000000000000001  x11 0000007f18abe170
    03-07 15:48:39.828 17145 17145 F DEBUG   :     x12 0000000000000004  x13 51c34e921a529271  x14 0000000000000001  x15 0000000000000001
    03-07 15:48:39.828 17145 17145 F DEBUG   :     x16 0000007f18b24fd0  x17 0000007f18b10560  x18 0000007c694dc000  x19 00000000000042a2
    03-07 15:48:39.828 17145 17145 F DEBUG   :     x20 00000000000042df  x21 00000000ffffffff  x22 0000001a580093f8  x23 0000007c6b483e18
    03-07 15:48:39.828 17145 17145 F DEBUG   :     x24 0000000000000010  x25 0000000000000014  x26 0000001a5bc27ab8  x27 0000001a5bc27b40
    03-07 15:48:39.828 17145 17145 F DEBUG   :     x28 0000001a56005b38  x29 0000007c6b483970
    03-07 15:48:39.828 17145 17145 F DEBUG   :     lr  0000007f18aa78b8  sp  0000007c6b4838d0  pc  0000007f18aa78e4  pst 0000000000001000
    03-07 15:48:39.828 17145 17145 F DEBUG   : 2 total frames
    03-07 15:48:39.828 17145 17145 F DEBUG   : backtrace:
    03-07 15:48:39.828 17145 17145 F DEBUG   :       #00 pc 000000000005d8e4  /apex/com.android.runtime/lib64/bionic/libc.so (abort+164) (BuildId: 1d36f8ae6e0af6158793abea7d4f4f2b)
    03-07 15:48:39.828 17145 17145 F DEBUG   :       #01 pc 0000000000034828  /data/app/~~rqTsxkU5NjJYcDmnf-0haw==/Mono.Android.NET_Tests-fIi1xEY1rqvY2p4EDO81Ww==/split_config.arm64_v8a.apk (offset 0xa6c000)

The problem is that the `Android.App.Application` static constructor
was running *too early*, which caused an
`Android.Runtime.XAPeerMembers` instance to be constructed
*before the runtime had finished initialization*.

Which leads us to a digression: when is static constructor run?
The answer to this depends upon the [`beforefieldinit` flag][0]:

  * When `beforefieldinit` is set, the runtime has lots of
    flexibility over when the static constructor is executed.
    The static constructor could be executed far earlier than one may
    normally expect.

  * When `beforefieldinit` *is not* set, the static constructor
    is executed "immediately" before the type is used.

In C#, if a type does not contain a static constructor, then the
type has `beforefieldinit` set:

	// has beforefieldinit
	partial class Application {
	  static readonly JniPeerMembers _members = new XAPeerMembers ("android/app/Application", typeof (Application));
	}

If a type has a static constructor, then it does not have
`beforefieldinit` set:

	// has beforefieldinit
	partial class Application {
	  static readonly JniPeerMembers _members = new XAPeerMembers ("android/app/Application", typeof (Application));

	  static Application()
	  {
	  }
	}

We hit this same scenario in b6c22f39, and fixed that by using
`[MethodImpl(MethodImplOptions.NoInlining)]` on
`SetSynchronizationContext()`, on the hope that doing so would delay
execution of the `Application` static constructor until the invocation
of `SetSynchronizationContext()`.  This worked for b6c22f39, but not
within NativeAOT.

To fix this:

  * Add an empty static constructor to `Android.App.Application` to
    remove the `beforefieldinit` flag from the type.

  * Call `JNIEnvInit.SetSynchronizationContext()` at startup for
    NativeAOT.

With this change in place, no crashes occur at startup and the test
passes.

[0]: https://csharpindepth.com/Articles/BeforeFieldInit

jonpryor · 2025-03-07T19:43:41Z

@jonathanpeppers: would you be able to provide the TODO in my draft commit message?

jonathanpeppers · 2025-03-07T21:51:36Z

It is a native crash, not very descriptive...

--------- beginning of crash
03-07 15:48:39.566 17058 17119 F libc    : Fatal signal 6 (SIGABRT), code -1 (SI_QUEUE) in tid 17119 (Thread-11), pid 17058 (droid.NET_Tests)
03-07 15:48:39.661 17145 17145 I crash_dump64: obtaining output fd from tombstoned, type: kDebuggerdTombstoneProto
03-07 15:48:39.662   693   693 I tombstoned: received crash request for pid 17119
03-07 15:48:39.663 17145 17145 I crash_dump64: performing dump of process 17058 (target tid = 17119)
03-07 15:48:39.828 17145 17145 F DEBUG   : *** *** *** *** *** *** *** *** *** *** *** *** *** *** *** ***
03-07 15:48:39.828 17145 17145 F DEBUG   : Build fingerprint: 'google/panther/panther:14/AP2A.240905.003/12231197:user/release-keys'
03-07 15:48:39.828 17145 17145 F DEBUG   : Revision: 'MP1.0'
03-07 15:48:39.828 17145 17145 F DEBUG   : ABI: 'arm64'
03-07 15:48:39.828 17145 17145 F DEBUG   : Timestamp: 2025-03-07 15:48:39.677241696-0600
03-07 15:48:39.828 17145 17145 F DEBUG   : Process uptime: 2s
03-07 15:48:39.828 17145 17145 F DEBUG   : Cmdline: Mono.Android.NET_Tests
03-07 15:48:39.828 17145 17145 F DEBUG   : pid: 17058, tid: 17119, name: Thread-11  >>> Mono.Android.NET_Tests <<<
03-07 15:48:39.828 17145 17145 F DEBUG   : uid: 10582
03-07 15:48:39.828 17145 17145 F DEBUG   : tagged_addr_ctrl: 0000000000000001 (PR_TAGGED_ADDR_ENABLE)
03-07 15:48:39.828 17145 17145 F DEBUG   : signal 6 (SIGABRT), code -1 (SI_QUEUE), fault addr --------
03-07 15:48:39.828 17145 17145 F DEBUG   :     x0  0000000000000000  x1  00000000000042df  x2  0000000000000006  x3  0000007c6b4838f0
03-07 15:48:39.828 17145 17145 F DEBUG   :     x4  0000000000000000  x5  0000000000000000  x6  0000000000000000  x7  0000007f36a58004
03-07 15:48:39.828 17145 17145 F DEBUG   :     x8  00000000000000f0  x9  0000007f18a6d350  x10 0000000000000001  x11 0000007f18abe170
03-07 15:48:39.828 17145 17145 F DEBUG   :     x12 0000000000000004  x13 51c34e921a529271  x14 0000000000000001  x15 0000000000000001
03-07 15:48:39.828 17145 17145 F DEBUG   :     x16 0000007f18b24fd0  x17 0000007f18b10560  x18 0000007c694dc000  x19 00000000000042a2
03-07 15:48:39.828 17145 17145 F DEBUG   :     x20 00000000000042df  x21 00000000ffffffff  x22 0000001a580093f8  x23 0000007c6b483e18
03-07 15:48:39.828 17145 17145 F DEBUG   :     x24 0000000000000010  x25 0000000000000014  x26 0000001a5bc27ab8  x27 0000001a5bc27b40
03-07 15:48:39.828 17145 17145 F DEBUG   :     x28 0000001a56005b38  x29 0000007c6b483970
03-07 15:48:39.828 17145 17145 F DEBUG   :     lr  0000007f18aa78b8  sp  0000007c6b4838d0  pc  0000007f18aa78e4  pst 0000000000001000
03-07 15:48:39.828 17145 17145 F DEBUG   : 2 total frames
03-07 15:48:39.828 17145 17145 F DEBUG   : backtrace:
03-07 15:48:39.828 17145 17145 F DEBUG   :       #00 pc 000000000005d8e4  /apex/com.android.runtime/lib64/bionic/libc.so (abort+164) (BuildId: 1d36f8ae6e0af6158793abea7d4f4f2b)
03-07 15:48:39.828 17145 17145 F DEBUG   :       #01 pc 0000000000034828  /data/app/~~rqTsxkU5NjJYcDmnf-0haw==/Mono.Android.NET_Tests-fIi1xEY1rqvY2p4EDO81Ww==/split_config.arm64_v8a.apk (offset 0xa6c000)

jonathanpeppers marked this pull request as ready for review March 5, 2025 21:41

jonathanpeppers requested a review from jonpryor as a code owner March 5, 2025 21:41

jonathanpeppers requested a review from dellis1972 March 5, 2025 21:41

jonpryor merged commit ccdc24c into main Mar 7, 2025
58 checks passed

jonpryor deleted the dev/peppers/nativeaotsynchronizationcontext branch March 7, 2025 23:32

jonathanpeppers mentioned this pull request Mar 26, 2025

NativeAOT TODO list #9784

Open

15 tasks

github-actions bot locked and limited conversation to collaborators Apr 7, 2025

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

[nativeaot] fix default `SynchronizationContext` #9883

[nativeaot] fix default `SynchronizationContext` #9883

Uh oh!

jonathanpeppers commented Mar 5, 2025 •

edited

Loading

Uh oh!

jonathanpeppers commented Mar 5, 2025

Uh oh!

dellis1972 commented Mar 6, 2025

Uh oh!

jonathanpeppers commented Mar 6, 2025

Uh oh!

jonpryor commented Mar 7, 2025 •

edited by jonathanpeppers

Loading

Uh oh!

jonpryor commented Mar 7, 2025

Uh oh!

jonathanpeppers commented Mar 7, 2025

Uh oh!

Uh oh!

Uh oh!

[nativeaot] fix default SynchronizationContext #9883

[nativeaot] fix default SynchronizationContext #9883

Uh oh!

Conversation

jonathanpeppers commented Mar 5, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

jonathanpeppers commented Mar 5, 2025

Uh oh!

dellis1972 commented Mar 6, 2025

Uh oh!

jonathanpeppers commented Mar 6, 2025

Uh oh!

jonpryor commented Mar 7, 2025 • edited by jonathanpeppers Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

jonpryor commented Mar 7, 2025

Uh oh!

jonathanpeppers commented Mar 7, 2025

Uh oh!

Uh oh!

Uh oh!

[nativeaot] fix default `SynchronizationContext` #9883

[nativeaot] fix default `SynchronizationContext` #9883

jonathanpeppers commented Mar 5, 2025 •

edited

Loading

jonpryor commented Mar 7, 2025 •

edited by jonathanpeppers

Loading