ros_comm 1.4.1 breaks socketcan_bridge tests

[mathias-luedtke]

I just noticed that one test started to fail for melodic shadow-fixed (1.4.1), but works for main (1.3.6).
https://travis-ci.org/ipa-mdl/ros_canopen/builds/381469509

The failing test, runs ros::spinOnce() to fetch one message.

Digging through the commits #1365 raised my attention.
I will run further tests, but this seems to change the behaviour at least for this test.

[mathias-luedtke]

I was not able to fix this test with an additional wait. It might be that the message got lost.
So it looks more like #1393 is the cause.

The only external depedencies are

executing command [apt-get install -y ros-melodic-roscpp]
executing command [apt-get install -y ros-melodic-class-loader]
executing command [apt-get install -y ros-melodic-message-runtime]
executing command [apt-get install -y ros-melodic-std-msgs]
executing command [apt-get install -y ros-melodic-roslint]
executing command [apt-get install -y ros-melodic-rostest]
executing command [apt-get install -y ros-melodic-rosunit]
executing command [apt-get install -y ros-melodic-message-generation]
executing command [apt-get install -y libboost-all-dev]

[dirk-thomas]

@guillaumeautran Can you please take a closer look at this.

[mathias-luedtke]

I was not able to dermine the cause of this problem, perhaps it's the combination of different commits.
However, I was able to come up with a test case.
For some reason the /rosout handling seems to be involved.

[mikepurvis]

Assuming you're on Linux, I'd be doubtful that #1393 could directly be the issue, since that was a fix for non-epoll systems such as macOS.

Does the test still fail if you build ros_comm from source but override HAVE_EPOLL=0 in this spot?

https://github.com/ros/ros_comm/pull/1217/files#diff-af96143090deaa643970030f962c964dR70

That would kick it back to the legacy poll implementation for handling IO.

[guillaumeautran]

I am taking a look right now but the specific tests passes for me at this time. Still doing some more digging.

[mathias-luedtke]

Assuming you're on Linux

The test case fails on travis, ROS buildfarm (#1409) and in a local docker environment (run_ci).

Does the test still fail if you build ros_comm from source but override HAVE_EPOLL=0 in this spot?

I will push a commit to the test case PR

I did not have a deeper look at the other PRs.

[guillaumeautran]

So, looks like a setup/teardown problem. I was able to reproduce the issue (only happens on the ros_comm melodic branch) but a soon as I comment out the first test case, the second test (which was failing originally) starts to pass.
Something must not have been cleaned up properly between the tests.

[mathias-luedtke]

The only other difference of package versions is rosconsole: 1.13.6 -> 1.13.7

[mathias-luedtke]

Something must not have been cleaned up properly between the tests.

If I add a ros::NodeHandle to main(), both tests pass.

[mathias-luedtke]

Finally, I ran git bisect. The test case started to fail with d21a33c (#1241)

[dirk-thomas]

Finally, I ran git bisect. The test case started to fail with d21a33c (#1241)

@trobertson-cpr FYI

[mathias-luedtke]

I have resolved this by adding the NodeHandle.
Nevertheless, #1241 changed the behavior and affects at least the logging performance in rospy.