Michel Hidalgo
Michel Hidalgo
FYI I added pipeline diagrams in https://github.com/ros2/design/pull/310/commits/78cd763d9c1be51179ed8b30ebad04b91c95d570. Would a data-flow diagram that shows how the current architecture can be cast into the proposed one help?
> Also matching the proposed tools to the stage in the pipeline would make it more clear I think. +1 > Am I somewhat close to what you are imagining?...
@sloretz @jacobperron I've updated the proposal, added a draft CBS schema, added examples. PTAL!
> I guess in my example I wouldn't supply an exact build command, since it would be dependent on what system is chosen (CMake or Gradle), but we could supply...
I can still reproduce on `master`.
> Reader and writer discovery do run independently in RTPS and so it is technically possible that the server has discovered the client's request writer and handled the request without...
> What I would find more elegant is for the client to first wait for the server to have discovered the response reader, and only then send the request. IMHO...
IIRC https://github.com/ros2/rmw_cyclonedds/pull/187 worked around this problem. @eboasson is there an ETA for a complete solution (e.g. a `DDS-RPC` like implementation for robust discovery or similar)?
Some re-design discussion took place in https://github.com/ros2/rmw_fastrtps/pull/418.
Even if compliance with DDS-RPC cannot be achieved, the discovery algorithm it suggests would provide for the behavior we expect. Some of your suggestions in #187 do not fall far...