`--hangdump-timeout` is for the whole test run instead of per-test on macOS

[elinor-fung]

Describe the bug

When using the --hangdump-timeout option, the value is considered for the entire test run on macOS. I would expect --hangdump-timeout to be the equivalent of --blame-hang-timeout, which is per-test. It is per-test on Windows.

Steps To Reproduce

1. Create an xunit v3 test project

   dotnet new install xunit.v3.templates
   dotnet new xunit3 -o xunit-test
   

2. Add a reference to Microsoft.Testing.Extensions.HangDump
3. Add the following test methods:

   [Fact]
   public void Test1() => Thread.Sleep(2000);
   
   [Fact]
   public void Test2() => Thread.Sleep(2000);

4. Build and run the tests, specifying 3s as the timeout:
   xunit3-test --hangdump --hangdump-timeout 3s --no-progress --output Detailed

Expected behavior

The --hangdump-timeout of 3s is per-test and the test run does not hit the timeout. This is the case when running on Windows.

passed xunit3_test.UnitTest1.Test1 (2s 010ms)
passed xunit3_test.UnitTest1.Test2 (2s 001ms)

Actual behavior

The --hangdump-timeout of 3s appears to be for the entire test run, so the run hits the timeout and fails. This happens on macOS, but not Windows.

passed xunit3_test.UnitTest1.Test1 (2s 011ms)
Hang dump timeout of '00:00:03' expired
Dumping process 52649 - xunit3-test
The following tests were still running when dump was taken (format: [<time-elapsed-since-start>] <name>):
[00:00:01] xunit3_test.UnitTest1.Test2

Additional context

<PackageReference Include="Microsoft.NET.Test.Sdk" Version="18.0.1" />
<PackageReference Include="xunit.runner.visualstudio" Version="3.1.5" />

I have tried this with both MTP v2:

<PackageReference Include="xunit.v3.mtp-v2" Version="3.2.1" />
<PackageReference Include="Microsoft.Testing.Extensions.HangDump" Version="2.0.2" />

And v1:

<PackageReference Include="xunit.v3" Version="3.2.1" />
<PackageReference Include="Microsoft.Testing.Extensions.HangDump" Version="1.9.1" />

[nohwnd]

Hello,
this was recently fixed, and we will release the fix in mstest 4.1 / mtp 2.1 #7005
On MacOS the Monitor that is used for indication of activity (tests are starting and finishing) does not reset.

This causes the test run controller to never receive a heartbeat from the test runner, and the run exits after the hang dump timeout, even though test are running.

I would expect --hangdump-timeout to be the equivalent of --blame-hang-timeout, which is per-test. It is per-test on Windows.

This is not how hangdump is designed to work. It has a single timer, and it will dump when there is no activity for the given period of time. In other words with timeout of 20s, you can still run a test that takes 3 minutes, as long as there are other tests running in parallel starting at least every 20s. This will give you a hang dump of tests that would otherwise hang forever, and most likely there won't be any tests that are just "randomly" running at the same time.