Killing thread cancels test run

[CharliePoole]

I've only seen this once, but it's pretty scary. In the following CI build, an entire test run is cancelled while we are testing ThreadUtility.Kill. I've never see this before.

https://ci.appveyor.com/project/CharliePoole/nunit/build/3.8.0-ci-04129-pr-2287

[rprouse]

Interesting, we had a timeout in those same tests and also in .NET 3.5 only in #2283 a few days ago.

[jnm2]

This is a test I wrote, so I'll take this one. I'll work on it tonight EST.

[jnm2]

Well, I have extra work so I'll have to put this off a day.

[jnm2]

Yikes. Worked 14 hours, slept 4.5, worked 11.5. Brain is cooked for today. But work deadline is over so I'll try to do this tomorrow at some point.

[jnm2]

The worst thing that's happening is a call to thread.Abort() on a thread I created, not the test thread, so I'm not sure how that got propagated back up to NUnit.

[jnm2]

I'm running stress tests via build -t test35 and not getting a repro on my computer.

[CharliePoole]

It shouldn't... for exactly this reason. There is code that NUnit uses to cancel an Abort unless a particular flag is set in the TestExecutionContext that tells us "Yes, we really want to abort the entire run."

[jnm2]

Do we have a second example of this failure on AppVeyor? I haven't found one so far.

AppVeyor needs a log search option.

[CharliePoole]

I saw it once before. Same stack trace as far as I remember. I wrote the bug on the second appearance of the problem.

[jnm2]

I can't think of any lines of attack. Can't think of any ways that a ThreadAbortException could be raised on a test thread via this test.

[CharliePoole]

I took a look at it. You have a TimeoutAttribute set, which will kill the test thread if it takes more than 1000ms. Maybe that interacts with the kill in your test in some unforseen way. You could try setting that timeout very short to see what happens.

[jnm2]

Now I feel silly. :D