Closed at-wat closed 3 years ago
There seems be an another flaky condition (which is not fixed by this PR) at the same point of the test: http://build.ros.org/job/Npr__geometry2__ubuntu_focal_amd64/55/, http://build.ros.org/job/Npr__geometry2__ubuntu_focal_amd64/39/ and http://build.ros.org/job/Npr__geometry2__ubuntu_focal_amd64/62/.
In this case, time jump back is not detected at:
https://github.com/ros/geometry2/blob/dae26a225e6c14a5aabd25a8b08c3031e6563535/tf2_ros/src/transform_listener.cpp#L106-L111
even when the test publishes old time to /clock
at:
https://github.com/ros/geometry2/blob/dae26a225e6c14a5aabd25a8b08c3031e6563535/tf2_ros/test/time_reset_test.cpp#L84-L85
In this case, time jump back is not detected at:
Yes we're tracking time_backwards in #129 and don't have a root cause of the flakiness. Longer timeouts decrease the frequency but don't guarantee passing.
transform_listener_backwards_reset
fails almost 100% on my AMD EPYC environment. Sleep byspin_for_a_second()
seems to be too small and slept only a millisecond. Fix it to a second to match with the function name. This makes the test possible to pass on EPYC environment.