JanStaschulat
commented
1 year ago Hi @alecGraves thanks for your question. (CC @fujitatomoya @pablogs9 ) Some arguments that support RCLC as a separate package on top of RCL:
- The ROS2 stack is structured in layers: RMW, RCL, language frontends like RCLCPP, RCLPY, and RCLC. We consider RCLC as a language frontend, rather than a core library that is used by other frontends.
- integrating rclc into rcl would lead to the situation having two executors in two layers, one in RCL, one in RCLCPP, which are using the same underlying rcl interface (rcl_wait, rcl_take, rcl_publish).
- "I think a lot of other clients might want the ability to define sequential execution or perform priority-based event handling." Yes, I fully agree. But what is the drawback in installing the rclc package, which provides the rclc-Executor with the mentioned features, in your use-case?
There have been several attempts to provide a C++ interface for RCLC (see issue https://github.com/ros2/rclc/issues/126). Do you also prefer a C++ API or would you prefer a C API for your project? Such a C++ API would have to be community effort.
On the other hand, the real-time working group has been working on a real-time Executor concept. A common goal could be to bundle the efforts in the embedded community (micro-ROS) and ROS 2 community to develop such an Executor with deterministic and real-time behavior.
alecGraves
Would we want to merge the work done on the rclc executor back into RCL at any point? I think a lot of other clients might want the ability to define sequential execution or perform priority-based event handling. This is a very common need in robotic systems.
Is there a point to keeping this separate from the RCL?