RerunBridge module and CLI

[leshy]

Sketch level rerun-bridge implementation on top of * subs

• This is a sketch, ready for someone to make mature/prod ready.
• Add RerunBridgeModule that subscribes to all pubsub topics and logs messages with to_rerun() methods to Rerun viewer
• Add rerun-bridge CLI command for quick visualization

Features

• Automatic discovery of messages with to_rerun() methods
• Configurable viewer mode (native, web, none)
• Works with any pubsub that supports subscribe_all() so we are transport agnostic

Testing

totally external bridge

uv sync
dimos --viewer-backend none --replay run unitree-go2
rerun-bridge

or

bridge as a module

dimos --viewer-backend none --replay run unitree-go2-bridge

#1171

WIP

needs global config sorted (discussing with paul) then good to go

[greptile-apps]

Greptile Overview

Greptile Summary

This PR introduces a RerunBridgeModule that enables visualization of pubsub messages by automatically logging any message with a to_rerun() method to a Rerun viewer. The implementation adds support for subscribing to all topics on pubsubs that implement the new SubscribeAllCapable protocol.

Key changes:

• Added SubscribeAllCapable protocol to dimos/protocol/pubsub/spec.py with covariant type variables for proper type safety
• Implemented RerunBridgeModule in dimos/visualization/rerun/bridge.py with configurable viewer modes (native, web, none)
• Added CLI command rerun-bridge for quick visualization
• Extended ViewerBackend type to include "none" option for headless operation
• Integrated bridge into unitree_go2_bridge blueprint

Issues found:

• Critical: Line 115 in bridge.py incorrectly passes pubsub.stop to Disposable() without wrapping in lambda, preventing proper cleanup
• Critical: Line 139 uses signal.pause() which is Unix-only and will crash on Windows

The architectural approach is sound - the protocol-based abstraction allows any pubsub implementing subscribe_all() to work with the bridge, and the automatic discovery of to_rerun() methods provides a clean interface. However, the two critical bugs must be fixed before merging.

Confidence Score: 3/5

• This PR has critical bugs that will cause runtime failures and should be fixed before merging.
• Two critical bugs in the bridge module: incorrect disposable wrapping that won't properly call stop methods, and Unix-only signal.pause() that will crash on Windows. The rest of the changes are solid.
• Pay close attention to dimos/visualization/rerun/bridge.py - it has two critical bugs on lines 115 and 139 that must be fixed.

Important Files Changed

Filename	Overview
dimos/visualization/rerun/bridge.py	New RerunBridgeModule added for visualizing pubsub messages. Has critical bugs in disposable handling and cross-platform signal handling.
dimos/protocol/pubsub/spec.py	Added SubscribeAllCapable protocol with proper covariant type variables for type safety.
dimos/core/global_config.py	Added "none" option to ViewerBackend type alias for headless operation.

Sequence Diagram

sequenceDiagram
    participant CLI as rerun-bridge CLI
    participant Bridge as RerunBridgeModule
    participant RR as Rerun Viewer
    participant PS as PubSub (LCM)
    participant Msg as Message with to_rerun()

    CLI->>Bridge: start()
    Bridge->>RR: rr.init("dimos-bridge")
    Bridge->>RR: rr.spawn() / rr.serve_web_viewer()
    Bridge->>PS: start()
    Bridge->>PS: subscribe_all(callback=_on_message)
    PS-->>Bridge: unsub function
    
    loop On each message
        PS->>Bridge: _on_message(msg, topic)
        alt msg has to_rerun()
            Bridge->>Msg: to_rerun()
            Msg-->>Bridge: rerun_data
            alt rerun_data is list
                loop for each (entity_path, archetype)
                    Bridge->>RR: rr.log(entity_path, archetype)
                end
            else single archetype
                Bridge->>Bridge: _get_entity_path(topic)
                Bridge->>RR: rr.log(entity_path, rerun_data)
            end
        end
    end
    
    CLI->>Bridge: stop() (on SIGINT)
    Bridge->>PS: unsub()
    Bridge->>PS: stop()

Loading

[greptile-apps]

_{3 files reviewed, 2 comments}

_{Edit Code Review Agent Settings | Greptile}

[paul-nechifor]

Since it's no longer frozen, it's probably best to remove @cached_property and replace it with @property. People will forget to update this list of keys.

[paul-nechifor]

This wasn't needed before. Why is it needed now?

[leshy]

Drake .pyi stubs only define the 1-arg overload of AddRigidBody, for some reason not cought in CI due to 3.10 I assume, we'll resolve soon

[leshy]

was always needed for 3.12

[paul-nechifor]

I'm sure toolz is not a large package. But it doesn't feel like it's hard to do the same in a more Pythonic way.

[paul-nechifor]

D'oh. I see that it's cached now. Also, the way _visual_override_for_entity_path is used is a bit odd. It's always called right away. I would assume you use a lambda because you want to save it and reuse it, but it's always called as self._visual_override_for_entity_path(entity_path)(msg). Why not eliminate the lambda and use: self._visual_override_for_entity_path(entity_path, msg)

[leshy]

result = msg
for fn in [*matches, final_convert]:
   result = fn(result)
return result

just feels much worse, I added toolz which is similar to lodash/underscore in ts for functional-ish stuff. I understand this is in direct conflict with "pythonic" stuff but seems like there are cases in which it's much cleaner? idk

[paul-nechifor]

Maybe rename this to navigation_costmap since it's not for debugging anymore.

[leshy]

I did, but kept DEBUG_NAVIGATION toggle, I don't mind this being optional

[paul-nechifor]

This does not depend on go2. Should we use this for g1 as well? Maybe move it and reuse it?

[leshy]

yes we should, just it's a large PR already so didn't want to touch G1, Jeff also did a big rewrite for G1 files that's about to go in, can coalesce in the follow ups

[paul-nechifor]

Did you delete the file?

[leshy]

kept for now, this is used by foxglove