---
title: "RemotiveTopology Framework changelog"
description: "Release notes for the RemotiveTopology Python framework (remotivelabs-topology), with the new features, fixes and breaking changes in each version."
source: https://developer.remotivelabs.com/docs/reference/release-notes/remotive-topology-framework
last_updated: 2026-11-19
---

# RemotiveTopology Framework changelog

## 0.23.1 - 2026-11-13

### Bug fixes

- **comm:** Let InvalidStateError take a message

  InvalidStateError was a @dataclass with no fields, so its generated
  __init__ took no arguments and every raise InvalidStateError("...")
  raised a TypeError instead. That hid the message and kept callers
  from catching the error.
- **control:** Ignore late replies to timed-out requests

  A reply that arrived in the same event-loop iteration as its request's
  timeout found its future still in _pending_requests, but already
  cancelled. set_result() then raised asyncio.InvalidStateError, which
  ended the transport's receive task, so the ControlClient never received
  another response.

  The client now ignores a reply whose future is already done. The
  receive task also logs and drops a message that fails to decode or
  handle, rather than ending, so one bad message cannot cost every
  message after it. This protects ControlServer too.
- **control:** Report a dead receive task and drop the redundant outer timeout

  If the transport's receive task has ended with an error, send() now
  raises InvalidStateError saying so, chained from the original error,
  instead of "must be started".

  ControlClient.send wrapped TransportClient.send in a second wait_for
  with the same deadline. The outer one could cancel the response future
  before the inner one timed out, which widened the late-reply race.
  TransportClient.send already enforces the timeout.

  Adds broker-free unit tests for TransportClient.

## 0.23.0 - 2026-10-26

> **Danger: Breaking changes**
>
>
> - **gateway-mapping:** schema has changed

### Bug fixes

- **gateway-mapping:** Resolve ethernet PDU channels to socket-pair namespaces

  Topology-generated brokers wire one namespace per SoAd socket
  connection, spelled &lt;ecu>-&lt;source_socket>-&lt;target_socket>-&lt;channel>.

- **gateway-mapping:** Drive mapping targets by their declared bus

  Broker chain type "udp" covers both CAN-over-UDP and ethernet PDU
  tunnels, so MappingModel built a CanNamespace for ethernet PDU
  targets and drove them through a restbus whose writes never reach
  the wire. The mapping's own address types now decide: a new `Bus`
  value type carries the declared protocol, targets get a restbus
  (CAN), a frame cache (LIN), or a plain direct-publish namespace
  (ethernet PDU) — which also makes their sources true passthrough
  subscriptions instead of on_change.

  Message-level routes are now typed values (`MessageRoute` /
  `MessageEndpoint`) so the addresses — with their sockets and bus —
  survive to namespace provisioning; `NamespacePlan` replaces the
  stringly-typed target-message dict, tracks source-only namespaces,
  and rejects a namespace resolved under two different buses.

- **gateway-mapping:** Skip mapping routes missing from the broker

  A platform-generated mapping may cover more channels than a topology
  instance wires up. Entries referencing namespaces the broker does not
  have were silently dropped from namespace provisioning while their
  routes kept subscribing and publishing to nonexistent namespaces —
  invisible at runtime.

  MappingModel now binds the mapping to the broker's actual namespaces
  (`ResolvedMapping`): unavailable entries and fan-out targets are
  removed from every engine's routing table, source and target side,
  and reported in one aggregated warning. The mapping is re-resolved on
  every reconnect attempt, so namespaces that appear later are picked
  up. Config errors (bus conflicts, missing broker client) now raise
  before the retry loop instead of being masked as transient. The
  publisher cache also warns when a managed namespace is missing
  instead of silently falling back to raw publishes; registered custom
  types travel as one `Registrations` value.

- **gateway-mapping:** Route each forwarded message by its cycle time

  Forwarding drove every CAN target through a managed restbus. A
  restbus never transmits a frame without a cycle time, so gatewaying
  to a sporadic frame — the common case for gateway frames — silently
  sent nothing (verified against a topology-generated broker).

  Output is now cadence-aware per target message, carried by a
  `ManagedOutputs` value: cyclic CAN frames keep the restbus and LIN
  frames the cache (time-decoupled, COM-style gatewaying — their
  sources still subscribe on_change), while cycle-less CAN frames and
  ethernet PDUs are relayed one-shot per received message (PduR-style
  IF-gatewaying, which AUTOSAR specifies for event PDUs). Behavior for
  cyclic targets is unchanged from 0.22.

  The engines' `namespaces` argument is replaced by `outputs`
  (a `ManagedOutputs`), since a per-namespace dict cannot carry
  per-message cadence.

- **gateway-mapping:** Plan and resolve Transformer sources despite source registration

  `register_source("can", ...)` only redirects pairs that involve something
  custom — a built-in CAN→CAN pair still runs on the Transformer, which
  subscribes through the broker. But `_namespace_plans` and
  `_resolve_signal_entry` skipped the source whenever its type was registered,
  so such a pair had its source namespace neither planned nor checked for
  availability: an absent source was kept and subscribed to blind.

  One predicate now decides whether a pair reads its source via the broker:
  always on the Transformer path; on the pipe path only when the source type
  is unregistered and the pair actually runs. Both planning and resolution use
  it.

- **gateway-mapping:** Relay custom flows into cycle-less CAN frames one-shot

  Pipes wrote their built-in targets through `TopologyTarget(namespace)`, which
  picks the write path from the namespace type alone: a CAN target always went
  into the restbus. The model deliberately keeps cycle-less frames out of the
  restbus (it never transmits a frame without a cycle time), so a registered
  custom source or transform feeding a sporadic CAN frame parked its value
  there and nothing reached the wire — the very bug the engines were fixed
  for, still live on the pipe path.

  `Pipe.from_mapping` now takes the same per-(namespace, message) publisher
  factory the engines use, and `TopologyTarget` accepts a ready `Publisher`
  alongside a `Namespace`. One code path decides how a message is written.

- **topology:** Close target namespaces when their setup does not complete

  `_open_target_namespaces` recorded a target only after `open()` and
  `restbus.start()` had both succeeded, and its cleanup caught `Exception`.
  So a restbus that failed to start left its already-opened namespace behind,
  and a cancellation mid-setup — not an `Exception` — skipped cleanup
  entirely, while the caller held an empty target set and could close
  nothing. A restbus from an abandoned attempt could keep transmitting while
  the reconnect loop retried.

  Record each target before awaiting its setup, and clean up on any
  `BaseException` before re-raising. `Namespace.close()` is idempotent, so
  closing a target whose `open()` never finished is safe.

- **gateway-mapping:** Name the mapping schema without a version in its title

  The version lives in the schema's $id and nowhere else; the title said 0.1 as well. The
  schema stays at 0.1: that version has never been published, so nothing pins a title, and a bump
  would drop the bundled 0.1 that every mapping file written so far resolves against
  (remotivelabs/basecamp#1760).

- **topology:** Close and reset only the restbus frames a namespace added

  The broker runs one restbus per namespace, shared by every client. Closing
  a namespace removed the whole broker restbus for it, so when several models
  shared a namespace, as the mock ECUs do, stopping one stopped them all.

  `Restbus` now keeps track of the frames it added, from `restbus_configs` or
  `add()`, and `close()` removes only those.

### Features

- **gateway-mapping:** Release the mapping schema as 0.2 at its new URL

  0.1 was published at releases.remotivelabs.com and released frameworks wrote
  that URL into mapping files. A published version is immutable and every change
  is a new one, so moving the host is 0.2: the same schema, at a URL that
  resolves. Files that name 0.1 keep loading, since the loader takes the schema
  key as a string, and 0.1 stays published as it was.

- **topology:** Stop behavioral models cleanly on SIGTERM

  `docker stop` sends SIGTERM, which Python handles by dying on the spot, so
  the `async with` blocks around a BehavioralModel never unwound and its
  teardown never ran. `asyncio.run` only turns Ctrl-C into cancellation.

  `remotivelabs.topology.cli.run` is a drop-in for `asyncio.run` that also
  cancels `main` on SIGTERM, then exits with status 143 once it has unwound.
  A second SIGTERM is ignored so it cannot interrupt that teardown. The mock
  and mapping entry points now start through it.

## 0.22.0 - 2026-08-12

### Features

- **topology:** Add gateway-mapping format

  Add the remotivelabs.topology.mapping submodule: a YAML mapping format
  plus JSON schema, the Forwarder (frame forwarding) and Transformer
  (signal transforms) engines, protocol-aware output (CAN restbus, LIN
  cache, Ethernet PDU publish), and a Pipeline runner that composes those
  with custom source/target pipes. Includes the schema-publishing flow.

## 0.21.0 - 2026-07-03

### Bug fixes

- Updated LICENSE text
- (remotive-topology-lib) Skip e2e signals when initializing LIN cache

  Since LIN namespaces can force e2e signal values, initializing them
  might break the e2e functionality.

### Features

- **release:** Render commit bodies and decouple changelog publishing

  The dev changelog now renders each commit body beneath its subject. A
  dedicated workflow publishes the checked-in topology changelog to the
  docs site on push to main, so the release workflow only publishes the
  versioned API docs. README recommends the basecamp commit-msg hook.

## 0.20.0 - 2026-06-15

> **Danger: Breaking changes**
>
>
> - LinNamespace no longer automatically populated the cache
>   with all frames. You must now explicitly configure which frames to cache
>   via cache_config.

### Features

- LinNamespace cache population is configured with cache_config
  - **BREAKING**: LinNamespace no longer automatically populated the cache
    with all frames. You must now explicitly configure which frames to cache
    via cache_config.

  Previously, LinNamespace automatically populated the LIN cache with all
  frames on open(), even frames not belonging to this ECU. This behavior
  is now controlled by the cache_config parameter, which fixes the problem
  and also allows more fine-grained control over the cache seeding.

## 0.19.3 - 2026-06-10

### Bug fixes

- **docs:** Tuck the theme toggle and version nav into the hamburger menu

  On small screens the fixed-position toggle and version nav overlapped pdoc's hamburger button.
  Render them in the nav footer (new module.html.jinja2) so on mobile they sit inside the opened
  hamburger menu; on desktop they still float top-right as before.

  Addresses the same overlap as remotiveplatform-docs ac8c76c.

- Make it easier to import FMUBehavioralModel etc

## 0.19.2 - 2026-05-26

### Features

- New FMUBehavioralModel

## 0.19.0 - 2026-05-22

### Features

- Metrics package is no longer optional, all related packages are always included

  Keep the [metrics] for backward compat

- Removed optional backwards compat, [metrics] not longer supported

## 0.18.0 - 2026-05-20

### Features

- **topology:** Add LinNamespace

  Encapsulates LIN-specific behavior that previously leaked into ECUMock:
  type validation on open, automatic broker-side cache priming with each
  signal's database start_value, header subscription, and set_schedule.

  ECUMock now dispatches on broker namespace type to the protocol-specific
  namespace class (CanNamespace/LinNamespace/SomeIPNamespace/
  ScriptedNamespace), falling back to GenericNamespace for unknown types.
  The LIN init-cache loop is gone since LinNamespace.open() handles it.

  Example LIN ECUs (rl, rlcm, polestar_backlight) updated to use
  LinNamespace and inherit BehavioralModel directly.

  Also fix subtle bugs in someip tests surfacing when bumping broker
  version from 1.19 to 1.24 in integration tests.

## 0.17.0 - 2026-05-13

> **Danger: Breaking changes**
>
>
> - **topology:** Register input handlers directly on namespaces

### Bug fixes

- **topology:** Restore subscription metrics on SubscriptionMonitor

  The frames-received counter and handler-duration histogram lived in
  NamespaceClient until the namespace refactor (61717812) deleted that
  class — the metrics went with it. SubscriptionMonitor is the new shared
  home for subscription plumbing, so move the instrumentation there. Both
  BehavioralModel input handlers and direct namespace subscriptions now
  emit the same metrics.

  Also:
  - plumb `callback_label` through Namespace.\_subscribe so input handlers
    surface their handler name (not the internal `_wrapped` closure) in
    the `callback` tag of handler.duration
  - bind handler.duration to second-scale buckets via a View; the OTel SDK
    ignores `explicit_bucket_boundaries_advisory` and falls back to
    ms-scale defaults, which collapses sub-5ms latencies into one bucket
    and makes histogram_quantile return the bucket midpoint
  - regression test asserts second-scale histogram buckets in the export

- Install [metrics] extras in topology docker image

### Features

- **topology:** Register input handlers directly on namespaces
  - **BREAKING**: register input handlers directly on namespaces

  Users can now attach input handlers to a namespace without routing them
  through a BehavioralModel. Filters + callback go straight onto the
  namespace (via constructor or `subscribe_input_handler`), using the same
  FrameHandler pipeline as BM — so signal filters and named-value decoding
  behave identically on both paths. This makes standalone namespaces
  useful on their own and keeps one mental model for input handling.

  Under the hood, every namespace now owns a SubscriptionMonitor in the
  base class that runs each callback in its own task with centralized
  exception propagation. BehavioralModel wires its input_handlers through
  the same monitor instead of reimplementing subscription management,
  which let us delete NamespaceClient entirely.
  - add `SubscriptionMonitor` (namespaces/monitor.py) and move shared
    subscription plumbing onto the `Namespace` base class
  - add `subscribe_input_handler` / `unsubscribe_input_handler` and an
    `input_handlers` constructor param on `GenericNamespace`
  - define `InputHandlerSpec` alongside `InputHandler` in input_handlers.py
    so BM and namespaces share one definition
  - replace per-subclass `open()` overrides with a `_validate(info)` hook
    on the base, removing duplicate `get_namespace()` round-trips
  - delete `NamespaceClient`; `BehavioralModel` now owns namespace
    lifecycle directly (breaking for any external NamespaceClient users)
  - add unit tests for the new API and an integration test that asserts
    namespace- and BM-level input handlers coexist on the same namespace

## 0.16.0 - 2026-04-29

### Features

- Accept (SignalName, SignalValue) tuples in publish() and TransactionContext

  GenericNamespace.publish() now accepts tuples alongside WriteSignal, consistent
  with the existing pattern in Restbus.update_signals() and TransactionContext.update().
  TransactionContext.updates field type widened to match what .update() already accepted.

- Intrumenting python libs with open-telemetry and prometheus

  Intial metrics that we can build upon. Grafana dashboard in signal-broker gets copied during generate.

## 0.15.0 - 2026-04-23

> **Danger: Breaking changes**
>
>
> - **topology:** Subscriptions no longer deliver every frame by default.
>   Callbacks that relied on firing on unchanged frames (e.g. for cycle-time
>   heartbeats) must opt back in by passing `on_change=False`.

### Bug fixes

- Subscribe_frames() now respects constructor decode_named_values

### Documentation

- **topology:** Cover on_change default and E2eSignalsFilter usage

- Include input handlers in docs
  - Snippet was never included

### Features

- **topology:** Add input_filters to BehavioralModel

  BehavioralModel accepts a cross-cutting input_filters sequence that is
  appended to every input handler via a new InputHandler.extend_filters
  hook. Useful for cross-cutting concerns like excluding E2E signals:
  `input_filters=[E2eSignalsFilter(exclude=True)]`.

  Re-export E2eSignalsFilter, FrameSignalFilter.

- **topology:** Default BehavioralModel to use on changed
  - **BREAKING**: subscriptions no longer deliver every frame by default.
    Callbacks that relied on firing on unchanged frames (e.g. for cycle-time
    heartbeats) must opt back in by passing `on_change=False`.

  BehavioralModel and NamespaceClient accept an `on_change` option,
  defaulting to true (OnChangeMode.MERGED) so subscriptions only deliver frames
  whose subscribed signals actually changed.

## 0.14.0 - 2026-04-14

### Bug fixes

- Tester images now correctly installs dev version of lib

  Also removes duplicate filters implementation to use filters from
  broker lib instead. The current filters module is kept as a proxy
  module for backwards compability.

### Features

- **testing:** Replace SignalValueAccumulator with Capture-based test helpers

  Introduces a composable testing API for asserting async signal sequences:
  - `Capture[T]` — generic async iterator consumer with stateful cursor and
    retry-based waiting; replaces the old SignalValueAccumulator pattern
  - `capture_frames()` / `wait_for_frame()` / `wait_for_frames()` /
    `wait_for_signal_value()` / `wait_for_signal_values()` in `frames.py`
  - `capture_events()` / `wait_for_event()` / `wait_for_event_parameter_values()`
    in `some_ip.py` for SOME/IP event assertions
  - Refactors `retry` into a standalone generic utility (removes hamcrest
    dependency from the retry path)
  - Deletes `SignalValueAccumulator`, `async_utils`, and the old `await_at_most`
    accumulator pattern

  All integration and example tests updated to use the new API.

## 0.13.1 - 2026-04-01

### Documentation

- Improved docs

## 0.12.0 - 2026-01-27

### Features

- Default initial_empty to True in subscribe calls

  In a flow when the user wants to
  1. Subscribe to a signal
  2. Do something
  3. Verify that the signal was received
     The subscribe needs to be blocking until the broker has done what it
     should do, e.g. in some/ip do a subscribe call

### Refactor

- Filters now works with built-ins

  remotivelabs.topology.namespaces.filters now works with python
  built-in functions like `filter()`.

## 0.10.0 - 2025-10-15

### Bug fixes

- Clean up scripted namespace

  Remove some parts that were copied from generic namespace but are
  not relevant for scripted.

- Update scripted namespace documentation

### Features

- **topology:** Add support for scripted namespace in namespace client

## 0.9.4 - 2025-10-09

### Bug fixes

- Simple NamespaceClient
- BehavioralModel error when missing virt

## 0.9.3 - 2025-08-28

### Bug fixes

- Fix dependencies that could be considered circular

## 0.9.2 - 2025-08-27

### Bug fixes

- **remotivelabs-topology:** Bump broker version
  - fix accidental breaking change

- Handle empty event
- Import statements
  - import from leafe inside own module

### Documentation

- Added examples to documentation

### Features

- Add BCM reboot to lighting and steering
- BehavioralModel args only parses known args

## 0.9.1 - 2025-06-10

### Documentation

- Moved docs from docs.remotivelabs to examples

## 0.9.0 - 2025-06-09

### Documentation

- Added snippets to package and docs to **init**.py
  - So that it is possible to generate docs from the wheel
