Record/play custom messages without having to source workspace

[weidinger-c]

As I understood, MCAP is, like ROS 1 Bags, an encapsulated container that contains all msg definitions.
Right now when recording MCAP bags in ROS 2 Humble, I still have to source a workspace that contains custom messages to be able to record them.

Same goes when I try to play an MCAP bag with "ros2 bag play". Custom messages are skipped with the message:
[WARN] [1689771449.920892226] [rosbag2_player]: Ignoring a topic '/TOPIC', reason: package 'PACKAGE' not found, searching: [/ros_ws/install/ros2bag_tools, /ros_ws/install/rosbag2_tools, /ros_ws/install/rosbag2_rviz_plugins, /opt/ros/humble].
This is because I played the ROS 2 Bag in an environment, where the custom messages were not installed.

Playing custom messages was possible with ROS 1 bags, is there anything planned to support this behaviour with ROS 2 and MCAP?

[emersonknapp]

As of current ROS 2 state, message types must be installed in the local space in order to play them back in the bag. The schemas stored in the MCAP file are for use by other tooling, not the ROS 2 system itself.

To highlight a few details about it, the rclcpp::GenericPublisher that is used by the Rosbag2 Player class still requires the ability to fetch the appropriate typesupport for the type. The benefit from the "generic" publisher is the ability to publish already-serialized messages, saving performance.

The ability to playback without message types installed locally doesn't yet exist in the ros2 core, so it's not an MCAP issue specifically. There is some work begun in rclcpp (ros2/ros2#1374) to support "dynamic" subscription, which would require only a schema description locally and would lead naturally into the "dynamic" publisher. That feature is a part of making this "fully encapsulated playback" possible.

[emersonknapp]

Further note - we had ros2/rosbag2#1373 filed for "dynamic recording", but I've just filed a related ros2/rosbag2#1428 to track the "dynamic playback" complementary feature

[weidinger-c]

Thank you for that clarification. Looking forward to using "dynamic recording/playback“ at some point!