SystemDefaultsQoS - force to have depth > 0 for Publishers

[SteveMacenski]

Feature request

Feature description

For SystemDefaultsQoS to work out of the box for IPC, it needs to always strictly have a depth setting > 0. When used in a subscription out of the box, this works fine. For example:

  footprint_sub_ = create_subscription<geometry_msgs::msg::Polygon>(
    "footprint",
    nav2_util::DefaultQoS(),
    std::bind(&Costmap2DROS::setRobotFootprintPolygon, this, std::placeholders::_1));

However, when used in a publisher, it fails. For example:

  footprint_pub_ = create_publisher<geometry_msgs::msg::PolygonStamped>(
    "published_footprint", rclcpp::SystemDefaultsQoS());

or

  state_pub_ = this->create_publisher<nav2_msgs::msg::CollisionDetectorState>(
    "collision_detector_state", rclcpp::SystemDefaultsQoS());

with the error

[component_container_isolated-7] [ERROR] [1734048098.404790594] [local_costmap.local_costmap]: Original error: intraprocess communication is not allowed with a zero qos history depth value

---

Thus, I feel that we should update the default for SystemDefaultsQoS for publishers (or in general) to be 1 if not already > 1 so that folks migrating or implementing IPC don't run into a cacophony of issues, one for each publisher. I understand this comes from the RMW/DDS, but it seems to me this needs to be updated since SystemDefaultsQoS is a key "default" QoS profile. Having that simply not work for a core ROS 2 feature seems intrinsically problematic :-) It was a bit subtle for me to identify this, so fixing this for everyone else in the future would be ideal.

For now, I'm creating a new nav2_util::DefaultPublisherQoS() to resolve this on my end, but I believe it should be fixed in rclcpp / below for the general user case

[fujitatomoya]

@SteveMacenski thanks for detailed description, that really helped.

i believe this is a bug for the intra-process publisher, #2707 should fix this.

[SteveMacenski]

I'll openly admit the use of qos vs get_actual_qos() is a subtlety I cannot contribute much to the discussion on :-)

I will say that in your test, I see it with use_intra_process_comms(true) without messing with the publisher-specific IPC settings like you do in your current test. I think these are functionally the same, but if you wanted to verify it fixed what I was seeing, a test might be simply:

TEST(TestPublisher, test_publisher_with_system_default_qos) {
  auto node_options = rclcpp::NodeOptions().use_intra_process_comms(true);
  auto test_node = std::make_shared<rclcpp::Node>(node_options);
  using test_msgs::msg::Empty;
  ASSERT_NO_THROW(
  {
    auto publisher = test_node->create_publisher<Empty>("topic", rclcpp::SystemDefaultsQoS());
  });
}

If you see that fail without your change and pass with it, then it fixes the issue case I see 😄