fleetctl: determine job state to be changed more accurately

So far assertUnitState() has determined whether a unit's job state is to
be changed, simply by comparing current unit state with the target unit
state. That's actually not always correct, as it's unnecessary to do
that also for downgrading states. (e.g. launched -> loaded)

To run it only when upgrading states (e.g. loaded -> launched),
introduce a helper jsToBeChanged(), which returns true only the target
state is a more activated one than the current state. Use
jsToBeChanged() instead of simple comparison in assertUnitState().

fleetctl/fleetctl.go

@@ -1045,7 +1045,7 @@ func assertUnitState(name string, js job.JobState, out io.Writer) (ret bool) {
 		state = u.CurrentState
 	}
 
-	if job.JobState(state) != js {
+	if jsToBeChanged(job.JobState(state), js) {
 		log.Debugf("Waiting for Unit(%s) state(%s) to be %s", name, job.JobState(state), js)
 		return
 	}
@@ -1158,6 +1158,30 @@ func machineState(machID string) (*machine.MachineState, error) {
 	return nil, nil
 }
 
+// jsToBeChanged returns true if the target state is a more activated state
+// than the current state. For example, it returns true if it's a transition
+// from inactive to launched, but false for other way around.
+func jsToBeChanged(cur job.JobState, tgt job.JobState) bool {
+	switch tgt {
+	case job.JobStateInactive:
+		return false
+		break
+	case job.JobStateLoaded:
+		if cur == job.JobStateInactive {
+			return true
+		}
+		break
+	case job.JobStateLaunched:
+		if cur == job.JobStateInactive || cur == job.JobStateLoaded {
+			return true
+		}
+		break
+	default:
+		break
+	}
+	return false
+}
+
 // cachedMachineState makes a best-effort to retrieve the MachineState of the given machine ID.
 // It memoizes MachineState information for the life of a fleetctl invocation.
 // Any error encountered retrieving the list of machines is ignored.