Iterate over IntraProcessBuffer instead of creating a vector

[thomasmoore-torc]

rclcpp: Reduce unnecessary copies/allocations in intra-process publish

Branch: intra-process-reduce-copies (this repo: ros2/rclcpp)
Related PR: #3271 — "Populate message info for intra-process messages" (intra-process-message-info). See Landing order below — no functional dependency, but the two touch overlapping code.

Summary

Two avoidable allocation/copy hot spots in the intra-process publish path:

1. IntraProcessManager recomputes and concatenates the take-ownership subscription ID lists (inter-process-need + intra-process-need) into a fresh std::vector on every single publish call, even though the set of registered subscriptions only changes on add_subscription/remove_subscription.
2. Code that needs to visit every message currently buffered for a subscription (e.g. transient-local replay to a newly-added subscription) builds a full std::vector copy of the ring buffer's contents first, then iterates it, rather than visiting the buffer's storage directly.

Changes

• Pre-compute concatenated take ownership subscriptions — caches the merged take-ownership subscription id list, invalidating/recomputing it only when subscriptions are added or removed, instead of rebuilding it on every publish.
• Iterate over IntraProcessBuffer instead of creating a vector — adds for_each_shared/for_each_unique visitor-style iteration over IntraProcessBuffer's ring-buffer storage (covering all shared/unique storage × shared/unique request combinations, including the move-only unique-ptr case), and switches vector-materializing call sites to use it instead.

Compatibility

Internal to rclcpp's intra-process implementation (rclcpp::experimental::*); no public API or observable behavior changes — this is a pure performance/allocation-reduction change.

Testing

Built and tested against rolling in a ros:rolling-ros-base container via colcon build/colcon test for rclcpp (3091 tests, 0 failures), plus uncrustify/cpplint lint targets. All green.

Landing order

intra-process-message-info (opened separately) touches the same files (intra_process_buffer.hpp, intra_process_manager.hpp, and both intra-process test files) as part of an unrelated feature adding message-info propagation. The two branches are independently buildable and testable, but not free to merge in either order without a rebase — real textual conflicts exist between them (confirmed via a trial merge). No functional dependency either way; whichever of the two merges second will need to reconcile against the other.

🤖 Generated with Claude Code

[jmachowinski]

This AI generated PR description is pretty muss useless.
What is the goal here?
Did you somehow verify that there is a real performance enhancement ?

[thomasmoore-torc]

The goal is to eliminate unnecessary memory allocations. This is achieved by iterating over the IntraPreocessBuffer instead of having it create and return a std::vector.

[jmachowinski]

Did you somehow verify that there is a real performance enhancement ?