Implement UnhandledException hook for finalizer scenario in Mono.

[VSadov]

A follow-up for #109806

Most of the implementation is in the shared corelib code and works for Mono, but finalization scenario needs special handling as that part is not shared.

[Copilot]

Pull Request Overview

This PR implements the UnhandledException hook for finalizer scenarios in Mono runtime, building on previous work in shared corelib code. The implementation ensures that unhandled exceptions in finalizers are properly handled through the global exception handler mechanism.

Key changes:

• Introduces a new GuardedFinalize method that wraps finalizer calls with exception handling
• Updates Mono's finalizer execution path to use the guarded version instead of direct finalizer calls
• Enables previously excluded UnhandledException tests for Mono (with some exceptions)

Reviewed Changes

Copilot reviewed 5 out of 5 changed files in this pull request and generated 2 comments.

Show a summary per file

File	Description
src/tests/issues.targets	Enables UnhandledException tests for Mono, excluding only specific problematic test cases
src/mono/mono/mini/aot-compiler.c	Updates AOT compilation to use GuardedFinalize instead of Finalize for runtime-invoke
src/mono/mono/metadata/gc.c	Modifies finalizer execution to call GuardedFinalize and updates method resolution
src/mono/System.Private.CoreLib/src/System/Object.Mono.cs	Implements GuardedFinalize method with exception handling using UnsafeAccessor
src/mono/System.Private.CoreLib/src/ILLink/ILLink.Descriptors.xml	Preserves GuardedFinalize method from IL linking removal

[VSadov]

I think this is ready for a review.

There are some failures on ios/tvos, but I cannot tell if these are caused by the change.
In theory the approach should work on all platforms. - it seems to work even for WASM

[VSadov]

This keeps failing on ios/tvos. I am out of ideas why. @lambdageek - could you take a look?

[lambdageek]

@VSadov I can't really see anything in the CI logs, but probably something needs to tell Mono's AOT compiler to compile GuardedFinalize

[kotlarmilos]

Let me run it locally to try to get better diagnostics/ stack trace.

[kotlarmilos]

I am getting the following error on iOS arm64 device:

((null) error) * Assertion at /Users/miloskotlar/dotnet/runtime-unsafe-accessor/src/mono/mono/metadata/gc.c:304, condition `is_ok (error)' not met, function:mono_gc_run_finalize, Attempting to JIT compile method '(wrapper runtime-invoke) object object:runtime_invoke_direct_void__this__ (object,intptr,intptr,intptr)' while running in aot-only mode. See https://learn.microsoft.com/xamarin/ios/internals/limitations for more information.

[VSadov]

I am getting the following error on iOS arm64 device:

((null) error) * Assertion at /Users/miloskotlar/dotnet/runtime-unsafe-accessor/src/mono/mono/metadata/gc.c:304, condition `is_ok (error)' not met, function:mono_gc_run_finalize, Attempting to JIT compile method '(wrapper runtime-invoke) object object:runtime_invoke_direct_void__this__ (object,intptr,intptr,intptr)' while running in aot-only mode. See https://learn.microsoft.com/xamarin/ios/internals/limitations for more information.

I think I am precompiling the wrapper. We had the same pattern before, just the wrapper was for Finalize and now it is for GuardedFinalize.
I wonder what is the difference?

The wrapper for Finalize was a virtual wrapper, since Finalize is a virtual method. The GuardedFinalize is not virtual, since it does not need to be.
I can make GuardedFinalize virtual and see what happens.

[VSadov]

Making GuardedFinalize virtual actually helps with ios/tvos. So it is not an issue with UnsafeAccessor. It is an issue with precompiling direct nonvirtual invoke wrapper.

Virtual GuardedFinalize could be ok, but less than ideal.
I will now try a nonvirtual again, but with non-direct wrapper.

[VSadov]

LLVMFULLAOT fails in Regression_2 JIT testcase. I do not think it is related.

The rest is passing now, including the newly enabled tests for UnhandledExceptionHandler - that is on all tested configs: JIT, interpreted, WASM or AOT/ios variants.

[VSadov]

/ba-g LLVMFULLAOT failure in Regression_2 is unrelated. Happens without this change as well (see: NOOP change in #118642)

[VSadov]

Thanks for reviewing!

[VSadov]

/backport to release/10.0-rc1

[github-actions]

Started backporting to release/10.0-rc1: https://github.com/dotnet/runtime/actions/runs/17064173533