Does JointTrajectoryController assume that joints are rotational - continuous, revolute - instead of linear - prismatic -? I cannot find any such assumption documented on the wiki or in the code, but maybe I missed something.
I see the angle_wraparound_ member, apparently used to differentiate continuous from revolute joints.
The calculation for the state_joint_error_ on position at https://github.com/ros-controls/ros_controllers/blob/melodic-devel/joint_trajectory_controller/include/joint_trajectory_controller/joint_trajectory_controller_impl.h#L390 and that for the state_error on position at https://github.com/ros-controls/ros_controllers/blob/melodic-devel/joint_trajectory_controller/include/joint_trajectory_controller/joint_trajectory_controller_impl.h#L394 use angles::shortest_angular_distance which will always give you a result in [-pi;pi].
It would be confusing to find that the position error on my linear joint mysteriously doesn't want to go beyond [-3.14meters;3.14meters], although a linear joint with such a length is probably not very common...