fix(rmw-zenoh-rs): publish on the DDS-mangled type name, as the wire requires

[YuanYuYuan]

Summary

rmw_zenoh_rs derives the topic key expression from a slash-separated type name. The hiroz library, and upstream rmw_zenoh_cpp, use the DDS-mangled form. They differ in one segment, so a hiroz library node and a ROS 2 node running rmw_zenoh_rs match no topic and exchange nothing. Neither side reports an error.

side	key expression
hiroz library	0/chatter/std_msgs::msg::dds_::String_/RIHS01_df668c74…
rmw_zenoh_rs	0/chatter/std_msgs/msg/String/RIHS01_df668c74…

The domain matches. The type hash matches byte for byte. Only the type name differs.

get_type_info() feeds both the key expression and the liveliness token. It read get_ros_type_name(), while get_type_prefix() sat beside it in the same file and produced the mangled form.

Which side is wrong was measured

A demo_nodes_cpp talker ran under RMW_IMPLEMENTATION=rmw_zenoh_cpp. A raw zenoh subscriber on ** recorded the key it declared:

0/chatter/std_msgs::msg::dds_::String_/RIHS01_df668c740482bbd48fb39d76a70dfd4bd59db1288021743503259e948f6b1a18

The library matches the wire. rmw_zenoh_rs is the outlier, so rmw_zenoh_rs is what changes here. Patching the library instead would have broken the test.yml leg that passes today.

The key expression is also the only difference. Raw CDR bytes published on the rmw_zenoh_rs key, produced by no ROS code and carrying no attachment, reached a demo_nodes_cpp listener. The two stacks already agree on domain, type hash, encapsulation and envelope.

Why CI is green

A missing edge, not a weak test. test.yml pairs the library with rmw_zenoh_cpp. rmw-zenoh-rs.yml pairs rmw_zenoh_rs with ROS packages. No leg pairs the library with rmw_zenoh_rs.

What this changes

	today	with this change
get_type_info()	reads get_ros_type_name()	reads get_type_prefix()
get_ros_type_name()	exists, called twice	deleted
type_name_dyn() vs type_info_dyn()	disagree	agree
the key on the wire	…/std_msgs/msg/String/…	…/std_msgs::msg::dds_::String_/…
a hiroz node and an rmw_zenoh_rs node	exchange nothing, silently	exchange messages
ros2 topic list -t, local topic	right, from the wrong name	right, demangled at the boundary
ros2 topic list -t, remote topic	std_msgs::msg::dds_::String_	std_msgs/msg/String

The graph leak on the last row is older than this change. hiroz builds the graph from liveliness tokens. Those tokens already carry the mangled form for every hiroz peer and every rmw_zenoh_cpp peer. So rmw_get_topic_names_and_types already reports the mangled name for a remote publisher on main today. On its own this change would extend that to local topics as well.

ros_from_dds_strict closes both halves. It is a port of _demangle_if_ros_type in rmw_zenoh_cpp's graph_cache.cpp, applied at all eight sites that copy a graph type name into an RMW result.

One rule, ten call sites

Review asked why several derivations existed at all. The rule now lives once, in hiroz_schema::type_name, and ten production sites call it: hiroz-protocol, both hiroz-codegen generators, hiroz-derive, two hiroz::dynamic paths, two action blocks in hiroz::node, hiroz-py, and hiroz-union. Four of those were independent inverses. dds_:: now appears in no production format! or replace outside that module.

hiroz-schema is the home because it already owns the canonical form, and because it is a leaf crate. Nothing pulls zenoh into a build script or a proc macro.

Two inverses remain, on purpose. ros_from_dds_strict matches the reference byte for byte, which an RMW boundary requires. ros_from_dds falls back for a name that has no dds_:: segment, because such a name must still resolve against the schema registry. Both differences were accidental before. A test now pins each.

What fails without this

stage	result
RED	library_and_rmw_derive_the_same_topic_key fails on the assertion: left: "std_msgs/msg/String", right: "std_msgs::msg::dds_::String_"
GREEN	the crate's tests pass, cargo fmt --check clean
end to end	library talker to demo_nodes_cpp listener under rmw_zenoh_rs: 35 received. Reverse: 29 received. The sniffed key matches the rmw_zenoh_cpp measurement
re-falsified	reverting the mangling turns the tests red and drops the listener to 0 received of 22 published

Tests

All 15 tests this change adds live in dedicated files, not an inline mod tests. Both crates already carried a tests/ directory.

file	tests	covers
crates/hiroz-schema/tests/type_name.rs	10	the forward rule, both inverses, every kind, the empty namespace, action sub-types, and the two cases where the inverses must disagree
crates/hiroz-protocol/tests/type_name_agreement.rs	5	the topic key, the service key, the liveliness token, the three action interfaces, and the service-name rule

The pre-existing inline module in rmw_zenoh.rs keeps its own 33 tests. Moving those too would add churn to a defect fix; that belongs in its own pull request.

Breaking changes

tag	what changes	who is affected	before → after	action
BC1	MessageTypeSupport::get_ros_type_name() is removed	any external caller	std_msgs/msg/String → call get_type_prefix()	call get_type_prefix()
BC2	the key rmw_zenoh_rs publishes on changes	a peer matching the old slash-form key	…/std_msgs/msg/String/… → …/std_msgs::msg::dds_::String_/…	none — the new key is the one rmw_zenoh_cpp and the library already use
BC3	RMW graph queries demangle	a caller expecting the mangled form from a remote peer	std_msgs::msg::dds_::String_ → std_msgs/msg/String	none — this is the form ROS tooling expects

No caller of get_ros_type_name() exists in this repository beyond the two this change rewrites.

Coverage

what	where it runs
the derivation, both inverses, service and action names, the liveliness token	15 unit tests in two dedicated files
the library against rmw_zenoh_cpp, four distros	Interop Tests, which runs hiroz-tests with ros-interop
the Humble no-type-hash configuration	the same job's humble leg, --no-default-features --features ros-interop,humble
rmw_zenoh_rs against ROS packages	rmw_zenoh_rs with ROS 2 Jazzy
the RMW graph reporting the ROS type name	a new test-ros.nu check, run by that same workflow

The check lives in scripts/test-ros.nu, which already registers this project's ROS-dependent tests. It starts a talker under RMW_IMPLEMENTATION=rmw_zenoh_rs, then asserts ros2 topic list -t prints std_msgs/msg/String and no dds_::. It also asserts /chatter appeared at all, so a probe that saw nothing fails instead of passing.

It stays out of the default pipeline, because it needs rmw_zenoh_rs built and on the RMW search path. Only the rmw_zenoh_rs workflow sets that up, and it runs the check by name.

That required one fix to the script. main validated a requested test against the pipeline, so any test registered in the map but left out of the default order was unrunnable. It now validates against the map, and --list marks which tests the default run skips.

The detector was proven to detect, against a stub ros2:

the listing says	result
std_msgs/msg/String	passes
std_msgs::msg::dds_::String_	fails: "the graph did not report std_msgs/msg/String"
nothing, talker dead	fails: "/chatter never appeared, so this run proves nothing"

The check starts its own rmw_zenohd, because rmw_zenoh peers need a router and nothing else in that step starts one. It guards both job kill calls: a talker that already exited makes an unguarded kill throw "Job N not found", which hides the real diagnosis behind a Nushell error.

Not verified

The end-to-end runs used std_msgs/String only. Tests cover the service path, the action interfaces and the liveliness token at the formatter. No live ROS service exercises them.

This touches no file that #325 touches, so the two can land in either order.

[Copilot]

Pull request overview

Aligns rmw-zenoh-rs topic and service keys with DDS-mangled type names used by hiroz and rmw_zenoh_cpp.

Changes:

• Adds a shared DDS type-name formatter.
• Uses mangled names for message and service wire metadata.
• Adds topic/service formatting tests.

Reviewed changes

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

File	Description
crates/rmw-zenoh-rs/src/type_support.rs	Applies DDS names to runtime type information.
crates/hiroz-protocol/src/format/rmw_zenoh.rs	Adds the formatter and regression tests.

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[Copilot]

Pull request overview

Copilot reviewed 17 out of 17 changed files in this pull request and generated no new comments.