Leave the execution_graph as none until compiled

Instead of forcing a compile when accessing the
execution_graph property just leave it as none until
compiling occurs.

It seems awkward to have the side-effect of automatically
compiling occur when a engine property is accessed.

Change-Id: I2760c369bc5926c6293199cb7e500e98d6c3eeb9

taskflow/engines/action_engine/engine.py

@@ -76,8 +76,14 @@ class ActionEngine(base.EngineBase):
 
     @property
     def execution_graph(self):
-        self.compile()
-        return self._analyzer.execution_graph
+        """The graph of nodes to be executed.
+
+        NOTE(harlowja): Only accessible after compilation has completed.
+        """
+        g = None
+        if self._compiled and self._analyzer:
+            g = self._analyzer.execution_graph
+        return g
 
     @lock_utils.locked
     def run(self):

taskflow/tests/unit/test_action_engine.py

@@ -464,6 +464,7 @@ class EngineGraphFlowTest(utils.EngineTestBase):
             utils.TaskNoRequiresNoReturns(name='task2'))
 
         engine = self._make_engine(flow)
+        engine.compile()
         graph = engine.execution_graph
         self.assertIsInstance(graph, networkx.DiGraph)
 
@@ -471,6 +472,7 @@ class EngineGraphFlowTest(utils.EngineTestBase):
         flow = utils.TaskNoRequiresNoReturns(name='task1')
 
         engine = self._make_engine(flow)
+        engine.compile()
         graph = engine.execution_graph
         self.assertIsInstance(graph, networkx.DiGraph)