test(qos): add QoS conformance matrix + rmw_zenoh_cpp interop tests

[YuanYuYuan]

Background

PR #183 fixed TRANSIENT_LOCAL interop with rmw_zenoh_cpp by switching to AdvancedPublisher/AdvancedSubscriber. The root cause — express(true) being mistaken for a caching mechanism — went undetected because:

1. No existing test exercised the late-joiner case (subscriber created after publisher emits samples). All QoS tests subscribed first, then published, which works fine even without a cache.
2. No automated test verified ros-z behaviour against a running rmw_zenoh_cpp node, only ros-z ↔ ros-z.

PR #183 adds a handful of late-joiner tests, but they only cover a small slice of the QoS space. A more thorough conformance effort is warranted to catch the next class of silent QoS bug.

Goals

1. Internal QoS conformance matrix (ros-z ↔ ros-z)

Add a parameterised test matrix in ros-z-tests/ (or crates/ros-z/tests/) that, for every meaningful combination of:

• QosDurability: Volatile, TransientLocal
• QosReliability: Reliable, BestEffort
• QosHistory: KeepLast(1), KeepLast(10), KeepAll
• Join ordering: subscriber-first, publisher-first (late-join)
• Subscribers: single, two (early + late)

…verifies that the subscriber receives the expected sample set. The expectation table should be derived from ROS 2 QoS semantics, not ros-z behaviour, so it serves as a spec.

Suggested implementation:

• Use rstest or a build.rs-generated table to keep the test file maintainable.
• Use the wait_for_count polling helper introduced in PR feat(pubsub-use-advanced): use AdvancedPublisher/AdvancedSubscriber to interoperate with rmw_zenoh_cpp transient_local topics #183 — no sleep-based assertions.

2. Cross-implementation interop tests (ros-z ↔ rmw_zenoh_cpp)

The harder part: automated tests that spawn a real rmw_zenoh_cpp publisher (or subscriber) and verify ros-z agrees on delivery semantics across the wire.

Open design questions:

• How is the rmw_zenoh_cpp node spawned from a Rust test? Likely via the existing ros-z-bridge test infrastructure or ./scripts/test-ros-packages.nu. (See .claude/ros-z/rules/testing.md — nix develop subprocesses are forbidden; everything must come from the flake.)
• Which subset of the conformance matrix should run in cross-impl mode? CI cost will dominate; pick the QoS combos that are common in practice (/tf_static, /parameter_events, defaults).
• Should this live in ros-z-tests/ or in a new crate?

Not blocking on these answers — file a design sub-issue once we start.

Why both, not just one

The internal matrix is cheap to add and catches drift inside ros-z. But it cannot catch wire-format / advanced-builder configuration mismatches with the C++ rmw — that's exactly the class of bug PR #183 fixed. Both layers are needed.

Related

• PR feat(pubsub-use-advanced): use AdvancedPublisher/AdvancedSubscriber to interoperate with rmw_zenoh_cpp transient_local topics #183 (transient_local interop fix)
• Support the transient_local QoS rmw-zenoh-mcap-writer#3 (same underlying issue, fixed independently)

[YuanYuYuan]

Additional pre-existing divergences from rmw_zenoh_cpp found while reviewing PR #183 (not introduced by it, but worth covering in the conformance matrix):

1. Publisher congestion_control — rmw_zenoh maps: BestEffort → Drop; Reliable + KeepLast → Drop; Reliable + KeepAll → Block. ros-z typed path: Reliable → Block (regardless of history); BestEffort → Drop. ros-z FFI path: always Block.

2. Publisher reliability — rmw_zenoh sets the zenoh reliability field explicitly (Z_RELIABILITY_RELIABLE/BEST_EFFORT). ros-z only sets congestion_control, leaving zenoh's reliability at the default.

Both could cause subtle interop differences in saturating-publisher or burst-loss scenarios and should be part of the conformance matrix.