Throwing results in test never completing

[Vijay-Braidable]

Describe the bug

When I throw a cyclical exception from a test (like is thrown when a resilience pipeline fails) the test from which it is thrown never exits.

Steps To Reproduce

1. Run the test TestMethod1 in the attached project.
2. See that the test runs forever (well I left it 5 mins)

Expected behavior

The test should finish as a failure.

Actual behavior

The test never exits.

Additional context

MSTest.XYZ Version: 3.6.1
Microsoft.NET.Test.Sdk Version: 17.11.0
Operating System: Windows 10 Pro
Visual Studio Version: 2022 (64-bit) - Preview (17.12.0 Preview 4.0)
.NET Version: net8.0

Toy test code for a repro if you don't want to download the zip:

    [TestMethod]
    public void TestMethod1()
    {
        // Arrange
        var exceptionA = new Exception("Exception A");
        var exceptionB = new Exception("Exception B", exceptionA);

        // Use reflection to set exceptionA's InnerException to exceptionB, creating a cycle
        FieldInfo innerExceptionField = typeof(Exception).GetField("_innerException", BindingFlags.Instance | BindingFlags.NonPublic);
        innerExceptionField.SetValue(exceptionA, exceptionB);

        // Act: Throw exceptionA, which now has a cyclic InnerException reference
        throw exceptionA;
    }

MSTest.Bug.Repro.zip

[thomhurst]

Looking at this, I don't understand why you'd set the inner exception later on?

A has an inner B exception
Then B has an inner A exception

What's the use case?

From a purely (maybe basic) logical perspective they both can't have caused each other.

One must be the initial cause, which makes sense to be an inner exception if wrapped, but creating a cyclical exception doesn't make sense to me?

[Vijay-Braidable]

Wrapped Exceptions: The pipeline might be wrapping an exception multiple times as it moves through different stages, leading to nested or cyclical exceptions. This can happen if each policy wraps the caught exception before rethrowing it, especially in complex configurations.

Inner Exception Referencing: Some policies might capture and throw a new exception with the original one as the InnerException. When this happens repeatedly, it creates a cyclical reference if an exception somewhere in the chain references one from an earlier stage, creating a loop of inner exceptions.

My view is that why it's happening doesn't matter, it can happen, and the test runner shouldn't run forever when it does.

[Vijay-Braidable]

I should add, this is what I believe is happening in my codebase, I can't be sure. I took a stab at the reasoning and made a repro, I don't personally throw nested exceptions. My guess was that some nested data object (thrown my the pipeline) in the exception was cyclical and that was causing the issue. It's why System.Text.Json has ReferenceHandler.Preserve, though as I say, it was a guess on my part that this was the reason.

[thomhurst]

I think this is actually a runtime problem. The hang occurs at the throw call.

I've reproduced it in a console app with 0 dependencies -> dotnet/runtime#109262

[thomhurst]

They've closed that issue as by design, so nothing we can do. We can't bypass the runtime rules.

[Evangelink]

I would have been tempted to close as won't fix but as we are already doing some special casing, it seems that the hashset solution isn't going to add too much complexity.

[Evangelink]

Problem is actually when the exception is thrown so nothing linked to MSTest. Closing as by-design on runtime.