all rclpy executor scripts fail on exit

[dirk-thomas]

with something like:

Failed to fini 'rcl_publisher_t'

custom_executor also has the following error:

TypeError: 'tuple' object is not an iterator

[sloretz]

Question: Should executor.shutdown() call node.destroy_node() on each node that has been added to it? A benefit is convenience for the user. A negative is it eliminates the possibility of replacing an executor while keeping the nodes alive, though I'm not imaginative enough to think of a scenario where that would be useful.

The Failed to fini 'rcl_publisher_t' errors happen because the examples don't call destroy_node on the nodes they create. At the end of the program the garbage collector calls __del__ on the nodes, but rclpy.shutdown() has already been called so the rcl instance id doesn't match.

Exception ignored in: <bound method Node.__del__ of <examples_rclpy_executors.custom_executor.Estopper object at 0x7f2d7ab94128>>
Traceback (most recent call last):
  File "/workspace/ros2_ws/install_isolated/rclpy/lib/python3.5/site-packages/rclpy/node.py", line 300, in __del__
  File "/workspace/ros2_ws/install_isolated/rclpy/lib/python3.5/site-packages/rclpy/node.py", line 272, in destroy_node
RuntimeError: Failed to fini 'rcl_subscription_t': rcl node is invalid, rcl instance id does not match, at /workspace/ros2_ws/src/ros2/rcl/rcl/src/rcl/node.c:382

Edit: shutdown() -> destroy_node()

[dirk-thomas]

I would lean toward "no".

The demo talker / listener e.g. call just:

node.destroy_node()
rclpy.shutdown()

The examples could do the same.

[sloretz]

Fixed by #196