fix(hu): resolve the message schema before subscribe returns

[YuanYuYuan]

Last slice of 5, splitting the original #311. #317, #318, #319 and #320 are merged. This one now targets main.

Fixes #297. Fixes #298. Fixes #299.

What this does

The plugin host resolves a message schema before subscribe() returns. The subscribe path also gains a fallback to a local .msg file.

What fails without this

The host resolved the schema on a spawned task. subscribe() had already returned Ok by then. The plugin's failure branch could therefore never run. The dropped channel sender looked like an idle topic.

check	before	after
hu meter echo /chatter	no output, exit 0	prints the decoded message
a publisher without the type description service	no output, exit 0	exit 1, and names the type it could not resolve
hu meter delay	runs forever	reports the failure
a message that fails to decode	skipped in silence	reported once loudly, then at debug

Resolution now happens before the host mints the resource. The host allocates the rep only after the subscription exists.

Discovery fails against any node that lacks the type description service. The host then falls back to a .msg file on HIROZ_MSG_PATH. The type name still comes from the graph, because subscribe carries only a topic.

The fallback refuses a mismatched schema, and accepts an unverifiable one

CDR is positional. A skewed schema therefore yields plausible but wrong field values, not a decode error. The fallback compares the local hash against the advertised hash.

advertised hash	action
equal to the local hash	decode
different	refuse, and name both
not advertised at all	warn, then decode

The third row matters most. A Humble node advertises nothing. Any peer built without type hashing also advertises nothing. An earlier version treated that as a mismatch, which broke every cross-distro case. Those cases are the main reason to have a disk fallback.

Compare hash values, never rendered strings. hiroz_protocol's renderer is gated on no-type-hash. Under that feature it returns one constant for every value, so a string comparison passes on anything. The diagnostic renders through an ungated formatter for the same reason.

Why the guest did not report the failure before

wasmtime gives each guest an epoch budget in wall clock. A blocking host call spends that budget while the guest does not run. A 5 s discovery therefore trapped the guest on return, before it could run the error branch that the host had just given it. A reported failure became a hang.

#319 fixed that. It is merged, so HostBlockGuard suspends the epoch ticker across this wait.

Breaking changes

what changes	who is affected	before to after	action
hu meter echo <topic> on a topic with no resolvable schema	anything that scripts its exit status	exit 0 after printing nothing, to exit non-zero with the reason	drop || true guards. A silent success was indistinguishable from an idle topic.

hz and bw still work on a topic whose type does not resolve. Their numbers come from a raw counting subscriber that never decodes. The typed subscription they used to hold is now best-effort, and it moves out of the mode. The host keeps it only so that hz and bw stay visible in the ROS graph.

hu meter echo now writes subscribed to <topic>; waiting for messages to stderr. A subscribe that succeeds and then prints nothing looks the same as a broken one.

What this does not reach

#300 records the one case. A publisher may advertise no type at all. The fallback needs a type name, and the graph is its only source. That case needs an addition to the WIT interface.

[Copilot]

Pull request overview

Resolves schemas synchronously for WASM subscriptions, adds local .msg fallback, and improves runtime-typed example support.

Changes:

• Propagates schema failures before subscribe() returns and pauses epoch accounting during host waits.
• Adds hash-checked local schema fallback and enables type-description services in publishing examples.
• Improves diagnostics, regression tests, and Axum route compatibility.

Reviewed changes

Copilot reviewed 19 out of 19 changed files in this pull request and generated 3 comments.

Show a summary per file

File	Description
crates/hiroz/src/node.rs	Documents type-description opt-in.
crates/hiroz/src/lifecycle/node.rs	Adds lifecycle builder pass-through.
crates/hiroz/src/dynamic/type_info.rs	Publicizes DDS type normalization.
crates/hiroz/src/dynamic/mod.rs	Re-exports type normalization.
crates/hiroz/src/context.rs	Initializes lifecycle option.
crates/hiroz/examples/z_pubsub.rs	Enables schema service and renames nodes.
crates/hiroz/examples/z_custom_message.rs	Enables schema service.
crates/hiroz/examples/z_cache/talker.rs	Enables schema service.
crates/hiroz/examples/twist_pub.rs	Enables schema service.
crates/hiroz/examples/shm_pointcloud2.rs	Enables schema service.
crates/hiroz/examples/lifecycle/talker.rs	Uses lifecycle pass-through.
crates/hiroz/examples/laser_scan.rs	Enables schema service.
crates/hiroz/examples/demo_nodes/talker.rs	Enables schema service.
crates/hiroz-union/src/plugin/wasm/mod.rs	Adds epoch suspension guard.
crates/hiroz-union/src/plugin/wasm/host/ros.rs	Resolves schemas synchronously with fallback.
crates/hiroz-union/src/modes/web.rs	Fixes Axum wildcard routing.
crates/hiroz-union/src/core/logger.rs	Includes hu in default filters.
crates/hiroz-union/plugins/hu-meter/src/lib.rs	Updates subscription behavior and status output.
crates/hiroz-tests/tests/hu_meter.rs	Tests unresolved-schema failure reporting.

Suppressed comments (3)

crates/hiroz-union/src/plugin/wasm/host/ros.rs:38

• Only inspecting .first() can miss a usable type. Graph ordering is not guaranteed, so when the first publisher has type_info: None and a later publisher advertises the type, the fallback incorrectly reports that no endpoint advertises one. Scan all returned entities for the first endpoint with Some(type_info).

                .get_entities_by_topic(kind, topic)
                .first()
                .and_then(|ent| match ent.as_ref() {
                    Entity::Endpoint(ep) => ep.type_info.clone(),
                    _ => None,

crates/hiroz-union/src/plugin/wasm/host/ros.rs:91

• The graph stores qualified topic names, but this fallback queries it with the caller's raw topic. create_dyn_sub_auto("chatter", ...) correctly qualifies to /chatter; after discovery fails, this lookup searches for chatter, misses the live publisher, and prevents the local .msg fallback. Qualify the topic using the node namespace/name before querying the graph while still passing the original topic to the subscriber builder.

    let Some(ti) = live_topic_type_info(graph, topic) else {

crates/hiroz-union/src/plugin/wasm/host/ros.rs:566

• call_raw has the same process-wide suspension issue: the guest controls timeout_ms up to about 49 days, and this guard disables epoch interruption for all stores for that entire wait. Cap the wait with a host-defined maximum or account for blocking time only on the calling store rather than pausing the shared ticker.

            let _epoch = super::super::HostBlockGuard::enter();
            replies.recv()

---

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