gazebo_ros: Use relative namespace for performance metrics

[mitschron]

No description provided.

[chapulina]

I don't think this will work. Gazebo is publishing at an exact topic without a namespace:

https://github.com/osrf/gazebo/blob/10863ffa8ec28e8e0e3d3a3a588d21cdc78055ab/gazebo/sensors/SensorManager.cc#L186

Also, the ~ will resolve to /gazebo/worldname, not just /gazebo.

Are you trying to solve some issue?

[mitschron]

I don't think this will work. Gazebo is publishing at an exact topic without a namespace:

https://github.com/osrf/gazebo/blob/10863ffa8ec28e8e0e3d3a3a588d21cdc78055ab/gazebo/sensors/SensorManager.cc#L186

Also, the ~ will resolve to /gazebo/worldname, not just /gazebo.

Are you trying to solve some issue?

Please have a look at issue #1175. Adding the performance metrics seems to be breaking the plugins.

[scpeters]

I don't think this will work. Gazebo is publishing at an exact topic without a namespace:
https://github.com/osrf/gazebo/blob/10863ffa8ec28e8e0e3d3a3a588d21cdc78055ab/gazebo/sensors/SensorManager.cc#L186
Also, the ~ will resolve to /gazebo/worldname, not just /gazebo.
Are you trying to solve some issue?

Please have a look at issue #1175. Adding the performance metrics seems to be breaking the plugins.

I believe the underlying issue in #1175 is gazebosim/gazebo-classic#2902, and it can occur whenever there is a subscriber to the gazebo topic /gazebo/performance_metrics and gazebo's SensorManager has sensors in 2 or more SensorContainers (ie a camera and a lidar, or a lidar and an IMU). This pull request works around the issue by breaking the gazebo subscriber so that it subscribes to an incorrect topic. I've submitted a different workaround in #1202 to noetic-devel (and #1203 for melodic and #1205 for ros2) that only subscribes to the gazebo topic if there is anyone subscribing to the corresponding ROS topic. So it will still be possible to encounter the deadlock, but it requires an extra step of subscribing to the /performance_metrics topic. I've opened #1209 to backport this workaround to kinetic-devel, so I think this can be closed once that is merged.

[scpeters]

I'm going to close this since I've just merged #1209 to kinetic-devel, and I believe it's a suitable workaround to #1175