What's odd is that I think we have a test for this? Maybe?

From the abort message in #10314:

7-21 15:25:22.624  1610  1610 F DEBUG   : Abort message: 'art/runtime/thread.cc:1223] Native thread exited without calling DetachCurrentThread: Thread[15,tid=4355,Native,Thread*=0x7f3c09434600,peer=0x12c89580,"Thread-234"]'

which suggests that if we:

1. Create a new Thread
2. Do some JNI work on that thread, and
3. Exit the thread

then the assertion should be triggered, no?

We have tests that do that, e.g.:

so why didn't we hit this before?

Are thread pool threads different from "normal" new System.Threading.Thread()?

Do we need to do "different" kinds of JNI calls on the created thread to trigger the assertion? It likely wouldn't be a bad idea to have a test that does:

var t = new Thread(() => {
    var list = new Java.Util.ArrayList();
    list.Add(new Java.Lang.String("a");
    list.Add(new Java.Lang.Integer(42);
});
t.Start();
t.Join();

and do a bit more than straight JNIEnv calls on the created thread.

The assert message, as-is, does not imply to me that we need to worry about app exit. (And that's ignoring the fact that "app exit" is a very nebulous concept on Android in the first place!)