abort() in python on windows debug nightly

[nuclearsandwich]

I spent a chunk of today investigating https://ci.ros2.org/view/nightly/job/nightly_win_deb/1243/

I started by reproducing the failure in https://ci.ros2.org/job/ci_windows/6824/ and then investigated the CI workspace. When running the ros2action tests manually I could repro and got a popup that python_d had abort()ed.

I eventually got as far as triggering the abort() just by running rclpy.init() from the python debug interpreter. I tried reverting a few recent rclpy packages, eventually reverting back to ros2/rclpy@d340737 but the issue persisted. I needed to change to a more local host for further debugging but hit different issues and got bogged down. I've assigned myself to make sure that this gets attention but I'll be unavailable until next week so if someone else wants to take it feel free to kick me off.

[mjcarroll]

I believe that this is related to a larger set of test failures on Debug Windows builds.

Over the last week, I discovered that the current log4cxx choco package had incorrectly-built .lib/.dll pairs for debug, which were pointing to the release dll, regardless of the build type.

I corrected the package and included it on the current release in the choco-packages repo: https://github.com/ros2/choco-packages/releases/tag/2019-02-15-1, as well as deployed it on the ROS2 CI machines.

After doing that, I surfaced some CMake issues, which need to be addressed via this PR: ros2/rcl_logging#13

[clalancette]

This is now "fixed" by ros2/rcl#436 (https://ci.ros2.org/view/nightly/job/nightly_win_deb/1251/ is the first non-red build in a while), so I'll close this out.