Clarify asynchronous test behaviour

The tests was occasionally failing due to
a race between the goroutine waiting for a timeout and the goroutine
setting the test state.

This commit changes the test so that it does not close the channel in
time, so that the chance of flakiness is significantely reduced.

Fixes #465

Co-authored-by: William Martin <wmartin@pivotal.io>

internal/leafnodes/runner.go

@@ -87,6 +87,9 @@ func (r *runner) runAsync() (outcome types.SpecState, failure types.SpecFailure)
 		finished = true
 	}()
 
+	// If this goroutine gets no CPU time before the select block,
+	// the <-done case may complete even if the test took longer than the timeoutThreshold.
+	// This can cause flaky behaviour, but we haven't seen it in the wild.
 	select {
 	case <-done:
 	case <-time.After(r.timeoutThreshold):

internal/leafnodes/shared_runner_test.go

@@ -199,21 +199,18 @@ func AsynchronousSharedRunnerBehaviors(build func(body interface{}, timeout time
 			})
 		})
 
-		Context("when the function times out", func() {
+		Context("when the function doesn't close the done channel in time", func() {
 			var guard chan struct{}
 
 			BeforeEach(func() {
 				guard = make(chan struct{})
 				outcome, failure = build(func(done Done) {
 					didRun = true
-					time.Sleep(20 * time.Millisecond)
 					close(guard)
-					defer close(done)
-					panic("doesn't matter")
 				}, 10*time.Millisecond, failer, componentCodeLocation).Run()
 			})
 
-			It("should return the timeout", func() {
+			It("should return a timeout", func() {
 				<-guard
 				Ω(didRun).Should(BeTrue())