Feedback on the experimental StateChange Archetype

[ntjohnson1]

This issue is a place to discuss requests or feedback on the recently added StateChange archetype

[rbuckley5]

hi, this feature is just what I've been looking for! However Ive experienced issues with the timeline scaling when adding it to the blueprint via the python sdk. If I dont add it to the blueprint here, and let the user add it in the viewer via the Blueprint panel, it works fine. But if I add it to the blueprint via the SDK, and include it in the RRD file, the resulting panel only shows the first few 10s or 100s of milliseconds of the start of the recording, which is 20s long.

[ntjohnson1]

hi, this feature is just what I've been looking for! However Ive experienced issues with the timeline scaling when adding it to the blueprint via the python sdk. If I dont add it to the blueprint here, and let the user add it in the viewer via the Blueprint panel, it works fine. But if I add it to the blueprint via the SDK, and include it in the RRD file, the resulting panel only shows the first few 10s or 100s of milliseconds of the start of the recording, which is 20s long.

Thanks for trying it out so quickly! Do you have a video or minimal repro? Often times that helps us diagnose more quickly.

[rbuckley5]

I can't share my code, but I will try to come up with a MRE to share

[rbuckley5]

copilot suggests the issue is due to early initialisation of the blueprint, when its added via the sdk

[rbuckley5]

here's a MRE. When i run this on my system (macos arm, python 3.12.11), and load resulting rrd file using

rerun reset && rerun mre_state_timeline.rrd

then close the viewer and repeat 5-10 times, sometime it loads with the timeline of the StateTimelineView wrong. Sometimes its 0..300ms, sometimes 0..10s, where the correct state is 0..23s ish. If it comes up in the 'wrong' state, adding a second statetimeline from the blueprint window gives a correct version. below is a screenshot in that state

#!/usr/bin/env -S uv run --script
# /// script
# requires-python = ">=3.12"
# dependencies = [
#     "rerun-sdk==0.32.0",
# ]
# ///
from __future__ import annotations

import time
import uuid
from pathlib import Path

import rerun as rr
import rerun.blueprint as rrb

RECORDING_START_S = 0.0
RECORDING_DURATION_S = 23.24
RECORDING_END_S = RECORDING_START_S + RECORDING_DURATION_S

TIMELINE = "time"

STATE_ENTITIES: dict[str, list[str]] = {
    "state1": ["value1", "value2"],
    "state2": ["value3", "value4"],
    "state3": ["true", "false"],
    "state4": ["true"],
    "state5": ["true"],
}

STATE_TRANSITIONS: dict[str, list[tuple[float, str]]] = {
    "state1": [
        (0.000, "value1"),
        (0.270, "value2"),
    ],
    "state3": [
        (0.050, "true"),
        (10.110, "false"),
        (10.860, "true"),
    ],
    "state4": [
        (0.050, "true"),
    ],
    "state5": [
        (0.000, "true"),
    ],
}


def log_static_state_configuration() -> None:
    stream = rr.get_global_data_recording()
    for entity, values in STATE_ENTITIES.items():
        rr.log(
            entity,
            rr.StateConfiguration(
                values=list(values),
                labels=list(values),
            ),
            static=True,
        )
        if stream is not None:
            stream.flush()


def log_interleaved_stream() -> None:
    Event = tuple[float, str, str, str]

    num_behavior_updates = int(RECORDING_DURATION_S * 100)
    events: list[Event] = [
        (
            RECORDING_START_S + RECORDING_DURATION_S * i / num_behavior_updates,
            "behavior",
            "",
            "",
        )
        for i in range(num_behavior_updates)
    ]
    for entity, transitions in STATE_TRANSITIONS.items():
        for t, value in transitions:
            events.append((RECORDING_START_S + t, "state", entity, value))
        if transitions:
            events.append((RECORDING_END_S, "sentinel", entity, transitions[-1][1]))

    events.sort(key=lambda ev: ev[0])

    sleep_per_event = 8.0 / len(events)
    stream = rr.get_global_data_recording()
    for t, kind, entity, value in events:
        rr.set_time(TIMELINE, duration=t)
        if kind == "behavior":
            rr.log(
                "/tree",
                rr.GraphNodes(
                    node_ids=["root", "child_a", "child_b"],
                    labels=["Root", "A", "B"],
                    positions=[[0.0, 0.0], [-1.0, -1.0], [1.0, -1.0]],
                    colors=[
                        [0x4C, 0xAF, 0x50, 0xFF],
                        [0xEF, 0x53, 0x50, 0xFF],
                        [0x42, 0xA5, 0xF5, 0xFF],
                    ],
                ),
            )
            rr.log(
                "/tree",
                rr.GraphEdges(
                    edges=[("root", "child_a"), ("root", "child_b")],
                    graph_type="directed",
                ),
            )
        else:
            rr.log(entity, rr.StateChange(state=value))
            if stream is not None:
                stream.flush()
        time.sleep(sleep_per_event)


def build_blueprint() -> rrb.Blueprint:
    spatial_view = rrb.Spatial3DView(name="3D Scene", origin="/")

    main_row = rrb.Horizontal(
        spatial_view,
    )

    state_view = rrb.StateTimelineView(name="States", origin="/")

    return rrb.Blueprint(
        rrb.Vertical(main_row, state_view, row_shares=[4, 1]),
    )


def main() -> None:
    out_path = Path(__file__).with_name("mre_state_timeline.rrd")

    stream = rr.RecordingStream(
        "a",
        recording_id=str(uuid.uuid4()),
        make_default=True,
    )

    blueprint = build_blueprint()
    stream.save(str(out_path), default_blueprint=blueprint)
    rr.send_blueprint(blueprint, make_active=True, make_default=True)

    log_static_state_configuration()
    log_interleaved_stream()

    stream.flush()

    print(f"Wrote {out_path}")


if __name__ == "__main__":
    main()