Note for future ourselves: disable -march=native when using PCL from Debian sources

[petrapa6]

Using optimization flag -march=native in catkin workspaces may induce random errors arising from different data arrangements in the memory of two packages working with Point Cloud Library (PCL) data. This is due to the PCL available at Debian sources is pre-built without the hardware-specific optimizations, which we use by default in our catkin workspaces.

Currently, there is no need to resolve this in mrs_lib as we use only PCL-ROS bridges in form of pcl-ros and pcl-conversions for simple conversions. So far, no errors have occurred to me in my daily use of PCL and mrs_lib, although I did not test the doTransform() method of mrs_lib/Transformer. However, I created this discussion to keep in mind, that once we start using features/algorithms from PCL (e.g., filters) in our wrappers, this might become an active issue.

For future use, I have prepared a pipeline for compiling and installing the PCL here (private, sorry non-MRS), so feel free to experiment with this. I'm currently deploying it on one of our drones to verify that everything works seamlessly.

Note for current active PCL users: if you are working with PCL and MRS setup, I suggest overriding the flag -march=native in all your PCL-dependent packages e.g., set(CMAKE_CXX_FLAGS "-std=c++17") in your CMakeLists.txt or compile PCL on your own.