Register our signal handler in rclpy_init vs each wait

[dhood]

This is my current thinking towards addressing ros2/launch#88

Opensplice python nodes would crash on sigint when calling the original signal handler. It is suspected that the signal handler by that point is changed to one that is opensplice specific, which is not what we are intending to call (we are intending to call the python signal module's handler).

I suspect that the point at which opensplice registers their signal handler is when a node is created, because if the creation of launch_ros' rclpy node is skipped (but we still call spin), then the crash on shutdown doesn't occur.

Registering the signal handler in rclpy_init (and only in init) matches the behaviour in rclcpp: https://github.com/ros2/rclcpp/blob/07e5be76218bd830c787b1d5c4af12e9eb6d5f6e/rclcpp/src/rclcpp/utilities.cpp#L187

This is in progress as I do more testing but feedback on the approach is appreciated

[dhood]

I missed @wjwwood 's comment in #192 (comment) that also suggested that this is a reasonable approach.

I've testing this on windows, osx and xenial and it's behaving as expected.

One thing I noticed was that with opensplice on windows the return code is set to 3221225786, which according to this page just indicates the program was stopped with ctrl-c. That seems reasonable to me (note: the following invocation was including a workaround for ros2/launch#91)

C:\J\workspace\ci_packaging_windows\ws>ros2 launch demo_nodes_cpp talker_listener.launch.py
[INFO] [launch]: process[talker.EXE-1]: started with pid [10812]
[INFO] [launch]: process[listener.EXE-2]: started with pid [5184]
[INFO] [talker]: Publishing: 'Hello World: 1'
[INFO] [listener]: I heard: [Hello World: 1]
[INFO] [talker]: Publishing: 'Hello World: 2'
[INFO] [listener]: I heard: [Hello World: 2]
[INFO] [talker]: Publishing: 'Hello World: 3'
[INFO] [listener]: I heard: [Hello World: 3]
[INFO] [talker]: Publishing: 'Hello World: 4'
[INFO] [listener]: I heard: [Hello World: 4]
[WARNING] [launch.LaunchService]: user interrupted with ctrl-c (SIGINT)
ssignal_handler(2)
[WARNING] [launch.LaunchService]: user interrupted with ctrl-c (SIGINT) again, ignoring...
ignal_handler(2)
[ERROR] [launch]: process[talker.EXE-1] process has died [pid 10812, exit code 3221225786, cmd 'C:\J\workspace\ci_packag
ing_windows\ws\install\lib\demo_nodes_cpp\talker.EXE'].
[ERROR] [launch]: process[listener.EXE-2] process has died [pid 5184, exit code 3221225786, cmd 'C:\J\workspace\ci_packa
ging_windows\ws\install\lib\demo_nodes_cpp\listener.EXE'].

C:\J\workspace\ci_packaging_windows\ws>ros2 launch demo_nodes_cpp talker_listener.launch.py
[INFO] [launch]: process[talker.EXE-1]: started with pid [9456]
[INFO] [launch]: process[listener.EXE-2]: started with pid [12252]
[INFO] [talker]: Publishing: 'Hello World: 1'
[INFO] [listener]: I heard: [Hello World: 1]
signal_handler(2)
signal_handler(2)
[WARNING] [launch.LaunchService]: user interrupted with ctrl-c (SIGINT)
[WARNING] [launch.LaunchService]: user interrupted with ctrl-c (SIGINT)
[ERROR] [launch]: process[listener.EXE-2] process has died [pid 12252, exit code 3221225786, cmd 'C:\J\workspace\ci_pack
aging_windows\ws\install\lib\demo_nodes_cpp\listener.EXE'].
[ERROR] [launch]: process[talker.EXE-1] process has died [pid 9456, exit code 3221225786, cmd 'C:\J\workspace\ci_packagi
ng_windows\ws\install\lib\demo_nodes_cpp\talker.EXE'].

So I will run CI on this and put it in review

[dhood]

Full CI, including opensplice, before 8573fe7

• Linux
• Linux-aarch64
• macOS
• Windows

CI after 8573fe7 including packages that call shutdown:

• Linux
• Linux-aarch64
• macOS
• Windows

[dhood]

windows looks alright but I'm not expecting linux to come back green because I'm seeing a lot of failing launch tests.

[dhood]

The issue was 69c19df: I had been testing with processes that were always calling wait and so had guard conditions set up, but that's not the case for processes that just have publishers, for example.

New CI (full, including opensplice):

• Linux
• Linux-aarch64
• macOS
• Windows