fork 29.5.6 and apply ipc topic stats fix

[jayyoung]

Forked from: https://github.com/ros2/rclcpp/tree/29.5.6
Applies the fix: ros2#2913

[Copilot]

Pull request overview

This PR enables subscription topic statistics collection for intra-process (IPC) message delivery by passing a type-erased statistics handler into the intra-process subscription implementation, avoiding a circular include with subscription_topic_statistics.hpp.

Changes:

• Add a type-erased StatsHandlerFn to SubscriptionIntraProcess and invoke it during intra-process message execution.
• Wire up Subscription to provide the stats handler when topic statistics are enabled.
• Add logging/warning indicating message age stats are not accurate for intra-process delivery.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 3 comments.

File	Description
rclcpp/include/rclcpp/subscription.hpp	Creates and passes a type-erased stats handler into the intra-process subscription constructor.
rclcpp/include/rclcpp/experimental/subscription_intra_process.hpp	Adds StatsHandlerFn, stores it, and calls it during intra-process execution with a synthesized timestamp + warning.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

You can also share your feedback on Copilot code review. Take the survey.

[tonynajjar]

@jayyoung I added copilot review FYI, just because why not. I am not familiar with this part of rclcpp so if you want a deeper look from me let me know and I'll put aside some time.

I would recommending opening the PR upstream though so we can get feedback from maintainers

[jayyoung]

@jayyoung I added copilot review FYI, just because why not. I am not familiar with this part of rclcpp so if you want a deeper look from me let me know and I'll put aside some time.

Same. But the fix seems small enough, and the copilot review doesn't seem to flag anything major. The only thing I am concerned about is this, because it does have an implication:

  if (stats_handler_) {
    RCLCPP_WARN_ONCE(
      rclcpp::get_logger("rclcpp"),
      "Intra-process communication does not support accurate message age statistics");
    now = std::chrono::system_clock::now();
    const auto nanos = std::chrono::time_point_cast<std::chrono::nanoseconds>(now);
    msg_info.source_timestamp = nanos.time_since_epoch().count();
  }

If I'm following correctly, that means this:

• Intra-process messages bypass the RMW layer, so there's no real source_timestamp from the publisher to grab onto, so we have to do the above.
• The "fix" works around this by stamping source_timestamp = now() at receive time, making message_age always == 0
• message_period stats are accurate for intra-process
• message_age stats are NOT accurate for intra-process — always near zero, not a real latency measurement
• I think this is unlikely to be a practical problem since intra-process delivery should inherently be near-zero latency anyway, right? (shared pointer, no serialization, no latency?)

So I think this is a fix, but there is a better fix possible, but I think it shouldn't be an issue for now, and we can push for a better fix upstream that doesn't have this caveat.

Does that make sense @tonynajjar ?

[tonynajjar]

Makes sense, LGTM