Topic List - #3764
Draft
jeff-hykin wants to merge 16 commits into
Draft
Conversation
A port's launch-line value can now be an array of topic names instead of a single name, so a module can subscribe a group of same-typed topics to one callback that is told which topic each message arrived on. Rust: `InputGroup<T>` port with `recv() -> Option<(usize, T)>` and a `#[input_group(decode = ..., handler = ...)]` derive attribute. Python: `NativeModuleConfig.topic_groups: dict[str, TopicGroup]`. Existing single-topic ports are unaffected.
jeff-hykin
requested review from
Dreamsorcerer,
aclauer,
leshy,
mustafab0,
paul-nechifor and
spomichter
as code owners
August 29, 2026 19:21
jeff-hykin
changed the title
jeff-hykin
marked this pull request as draft
August 29, 2026 19:23
Codecov Report❌ Patch coverage is
@@ Coverage Diff @@
## main #3764 +/- ##
==========================================
+ Coverage 77.44% 77.63% +0.18%
==========================================
Files 1315 1318 +3
Lines 124238 125011 +773
Branches 10830 11046 +216
==========================================
+ Hits 96220 97056 +836
+ Misses 24895 24821 -74
- Partials 3123 3134 +11
Flags with carried forward coverage won't be shown. Click here to find out more.
... and 42 files with indirect coverage changes 🚀 New features to boost your workflow:
|
Contributor
Greptile SummaryThis change adds grouped native input ports so several same-typed topics can feed one indexed Rust handler. Group handling works in the Rust runtime, but baked module metadata still represents an input group as a scalar topic. A focused host fixture verified that this produces an empty Confidence Score: 4/5Not merge-safe: baked modules using There is one independent, verified P1 finding and it is not security-related, which maps to a score of 4. Files Needing Attention: Update
 T-Rex LogsWhat T-Rex did
Reviews (1): Last reviewed commit: "feat: one handler for N same-typed topic..." | Re-trigger Greptile |
![greptile-apps[bot]](https://avatars.githubusercontent.com/in/867647?s=60&v=4){kind=small}
| .iter() | ||
| .filter(|f| match f.kind { | ||
| FieldKind::Input { .. } => want_input, | ||
| FieldKind::Input { .. } | FieldKind::InputGroup { .. } => want_input, |
Contributor
There was a problem hiding this comment.
InputGroup fields are registered as ordinary scalar inputs here, so baked metadata emits a single topic value instead of the topic_groups array required by grouped ports. The host parses that value into Topics.single, while Builder::input_group reads only Topics.grouped; consequently, a baked module starts with an empty group and its indexed handler receives no input. Preserve grouped-port metadata through bake generation and host configuration rather than representing the group as a scalar input.
Artifacts
Focused Rust host fixture source
- The uploaded host test fixture constructs an InputGroup from the scalar baked metadata shape and asserts that its group length is zero, confirming the metadata-shape loss takeaway.
Focused Rust host fixture observed output
- The captured cargo test output records `scalar baked cams metadata -> InputGroup length: Some(0)` and a passing assertion over the executed host construction path, confirming the empty InputGroup takeaway.
Group entries were written as backend topic strings (`/cam0/color`), which reads as LCM-specific and bypasses the stream-name vocabulary the rest of a blueprint uses. They are now stream names as they read after remapping and namespacing, resolved to wire channels the same way a declared stream name is, with a leading `/` rejected outright.
Group entries used to be resolved to wire channels directly, so `.remappings()`, `.namespace()` and transport pins silently did not apply to them. Each entry now becomes a synthetic `In` stream: the blueprint atom declares it, the coordinator wires it like a declared port, and the module (python or native) reads the wired transport instead of computing a channel by convention.
`StreamGroup`/`stream_groups` becomes `TopicFunnel`/`topic_funnels` in python, and rust's `InputGroup`/`#[input_group]` becomes `TopicFunnel`/`#[topic_funnel]` to match. Naming only, no behavior change.
Rust: fold TopicFunnel<T> into Input<T> — every port takes one topic or an
array from the launch line. Handlers keep their (msg) signature; a handler
that needs the source opts in with #[input(meta)] and receives
(msg, Metadata { index, topic }).
Python: opt in by signature instead — a two-parameter handle_<port> or
handle_<input> receives (msg, Metadata(index, name)), where name is the
declared stream name pre-remapping; a one-parameter handler just gets the
message.
TopicFunnel.names may be a dict mapping each stream name to arbitrary info
(is this camera rectified, what are the depth units) that the handler
receives on Metadata.info. On the launch line such an entry becomes a
{"topic", "info"} object instead of a plain string; rust exposes it as
serde_json::Value, python as a dict. Entries without info stay plain
strings, so existing wiring is untouched.
The handler captured as a closure default was still typed Any | None from the getattr, and baked_host's _argv/_stdin_blob overrides never widened to the funnel-aware topics type.
Fanning several same-typed streams into one handler is stream routing, so it belongs alongside the other routing rewrites rather than in a module kwarg. A list of names (or a dict of name -> info) as a remapping value now fans that port in; a string still renames a single stream. .remappings() expands the fan-in into the blueprint atom right away: the port's StreamRef is replaced by one In per entry, typed from the declaration. remapping_map stays string-valued, so autoconnect, namespacing and transport pins are untouched. The declared port also supplies the message type, so TopicFunnel.msg_type is gone, and naming a port that does not exist is now an error instead of a silent no-op.
jeff-hykin
commented
Aug 31, 2026
The topics map was typed loosely enough that the baked host had to widen its override to Any. Name the info-carrying entry after the native TopicEntry it serializes to, and give the port's value its own alias so the host can state what it passes through. The remaining Any is the info payload itself, which is arbitrary by design.
jeff-hykin
commented
Aug 31, 2026
jeff-hykin
commented
Aug 31, 2026
| "(use a manual self.<input>.subscribe(...) for sync handlers)" | ||
| ) | ||
| handler: Callable[..., Any] = declared_handler | ||
| if _handler_wants_metadata(handler, f"{type(self).__name__}.handle_{input_name}"): |
Member
Author
There was a problem hiding this comment.
# doesn't get metadata
def callback(msg): pass
# gets metadata
def callback(msg, metadata): pass
jeff-hykin
commented
Aug 31, 2026
Metadata is a word this codebase already uses for several unrelated things, so the bare name says nothing about what the handler is being handed. Name it for what it describes: which topic the message arrived on.
Namespacing walks a blueprint's streams uniformly, so prefixing a funnel entry was already covered by the generic namespace tests once the entries became streams. The port's absence from that same list and the leading-slash rejection were each one assertion about an internal structure, not about anything a module does.
Filtering the signature down to positional parameters made a keyword-only or variadic handler quietly count as one-argument. A handler takes the message and optionally the metadata, so count what it declares and reject anything else.
Review feedback: remappings needs no prose, TopicFunnel is better shown than described, and _argv's existing one-liner was fine as it was.
A pydantic model is keyword-only, so building one meant naming both stored fields at every call site even though authoring takes a single list-or-dict. A plain dataclass takes that one argument positionally and splits it itself. Also name the funnel tests' delivery timeout rather than repeating the literal.
A plain port is a funnel of one, so the two bind routines were the same thirty lines twice. Treating a declared port as a single-stream funnel folds them together and drops the second one, along with the extra rx layer the funnel path was going through to reach the same dispatcher. Funnel streams now get their default transport when they are built, which is what makes the merged path safe standalone; a coordinator overwrites it as before. That also leaves channel_for with no callers.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
config a list of topics that can be specified at blueprint build time