Cleaning up some task failure detection problems

* fixed a bug in which failures from a with_* loop were not being
  caught correctly, leading to tasks continuing when they should stop
* when ignore_errors is enabled, the failure will no longer count
  towards the number of failed tasks

lib/ansible/executor/task_result.py

@@ -43,7 +43,7 @@ class TaskResult:
         return self._check_key('skipped')
 
     def is_failed(self):
-        if 'failed_when_result' in self._result:
+        if 'results' in self._result and True in [True for x in self._result['results'] if 'failed_when_result' in x]:
             return self._check_key('failed_when_result')
         else:
             return self._check_key('failed') or self._result.get('rc', 0) != 0
@@ -57,5 +57,6 @@ class TaskResult:
             for res in self._result.get('results', []):
                 if isinstance(res, dict):
                     flag |= res.get(key, False)
+            return flag
         else:
             return self._result.get(key, False)

lib/ansible/plugins/strategies/__init__.py

@@ -154,7 +154,9 @@ class StrategyBase:
                             debug("marking %s as failed" % host.name)
                             iterator.mark_host_failed(host)
                             self._tqm._failed_hosts[host.name] = True
-                        self._tqm._stats.increment('failures', host.name)
+                            self._tqm._stats.increment('failures', host.name)
+                        else:
+                            self._tqm._stats.increment('ok', host.name)
                         self._tqm.send_callback('v2_runner_on_failed', task_result)
                     elif result[0] == 'host_unreachable':
                         self._tqm._unreachable_hosts[host.name] = True