Terminated Agent Process leads to lost Test Results

[ghost]

In our observation, if the test agent process crashes (e.g. due to an unmanaged exception in a test case, or simulated by an Environment.Exit()), the test results of the whole test run are lost. The TestResult.xml file contains only the exception that the connection to the agent process was interrupted.

Sample Testcase:

[Test]
public void MyTestmethod()
{
    Environment.Exit(1);
}

Outcome:

System.Net.Sockets.SocketException: An existing connection was forcibly closed by the remote host

Test Run Summary
  Overall result: Failed
  Test Count: 0, Passed: 0, Failed: 0, Warnings: 0, Inconclusive: 0, Skipped: 0

Is this behavior intended?

We are working in a quite big project, where a test run may consist of several test assemblies with several 10K test cases. Even if the test case which leads to the crash is the last test case in the last test assembly, all previous test results are lost.
Therefore, we are currently searching for a possibility to handle terminations of the nunit agent process without loosing the information of the already executed test cases.
Is there any such possibility?
Can NUnit be extended somehow such that it is able to continuously log the test progress?

The use cases which justify this requirement would be in our case the following:

UseCase 1:
As a developer, I want to be able to analyze the root cause why a test agent has crashed. To do this, I need to know
a) at what point within the test case the crash happened.
b) what are the outcomes of previously run test cases.

UseCase 2:
As an Integrator, I want to be able to kill a running test agent in certain situations (process hangs, …) without loosing the results of the already executed tests in the current test run.

Additional Information:
• We run our tests with process model “Separate” (--process=Separate)
• NUnit Framework Version: 3.10.1
• NUnit Console Version: 3.8.0

[CharliePoole]

The info you see in the console output as well as th XML result is by design because, at the point where it is produced, it's the best we can do. The failing agents are gone with no trace left behind. So, what you are is in lieu of exiting without any output.

Both use cases would probably require changes to both the framework and the engine.

[CharliePoole]

Moving this to 4.0 for discussion. Not sure if breaking changes are needed to do it.