[BUG]: Sunburst marker.colors reassignment changes segment colors

[ym-xie]

Description

A Plotly Express sunburst chart changes color assignment when marker.colors is
reassigned to its current value.

The chart structure and labels stay the same, but the segment colors differ. This
looks like marker.colors is not idempotent for sunburst traces: explicitly assigning
the existing color tuple changes the color resolution path.

Screenshots/Video

Steps to reproduce

import io
import json

import plotly
import plotly.express as px
import plotly.io as pio
from PIL import Image, ImageChops
from playwright.sync_api import sync_playwright


print(f"Plotly version: {plotly.__version__}")


def make_fig():
    df = px.data.tips()
    return px.sunburst(
        df,
        path=["sex", "day", "time"],
        values="total_bill",
        color="time",
    )


colors = tuple(make_fig().data[0].marker.colors)

# Case 1: reassign marker.colors before the initial render.
fig1 = make_fig()
fig1.data[0].marker.colors = colors
assert tuple(fig1.data[0].marker.colors) == colors
html1 = pio.to_html(fig1, full_html=True, include_plotlyjs=True)

# Case 2: render first, then reassign the same value with Plotly.restyle.
fig2 = make_fig()
html2 = pio.to_html(fig2, full_html=True, include_plotlyjs=True)
html2 = html2.replace(
    "</body>",
    f"""
<script>
window.addEventListener("DOMContentLoaded", () => {{
    const plotEl = document.getElementsByClassName("plotly-graph-div")[0];
    setTimeout(() => {{
        Plotly.restyle(plotEl, {{"marker.colors": {json.dumps(colors)}}}, [0]);
    }}, 1000);
}});
</script>
</body>""",
)

with sync_playwright() as playwright:
    browser = playwright.chromium.launch(
        headless=True,
        args=["--no-sandbox", "--disable-dev-shm-usage", "--disable-gpu"],
    )
    page = browser.new_page(viewport={"width": 1200, "height": 800})

    page.set_content(html1, wait_until="domcontentloaded")
    page.wait_for_timeout(1500)
    img1 = page.screenshot(path="plot1.png")

    page.set_content(html2, wait_until="domcontentloaded")
    page.wait_for_timeout(2000)
    img2 = page.screenshot(path="plot2.png")

    browser.close()

diff = ImageChops.difference(
    Image.open(io.BytesIO(img1)).convert("RGB"),
    Image.open(io.BytesIO(img2)).convert("RGB"),
)

print("Saved plot1.png (Python-side reassignment before render)")
print("Saved plot2.png (JS-side restyle after render)")
assert diff.getbbox() is not None, "Reproducer did not reproduce: images are identical."
print("Bug reproduced: the two renderings differ.")

Notes

Add info here that doesn't fit in the other sections.

[beanscg]

I checked the Plotly.py side of this, and I think a useful first regression is JSON idempotency before deciding whether the bug lives in Plotly.py or Plotly.js.

On the Python side:

• assigning through figure paths delegates down to the trace/property setter path in basedatatypes.py;
• sunburst.marker.colors is a normal generated array property;
• the generated marker docs say colors sets the color of each sector, and if it is not specified then the default trace color set is used.

Given that, reassigning fig.data[0].marker.colors = fig.data[0].marker.colors should be a no-op in serialized JSON. If fig.to_plotly_json() changes after that assignment, the Python setter/validator path is implicated. If the JSON is identical but the rendered colors change, the more likely bug is the downstream plotly.js/defaults/render path interpreting an explicitly present marker.colors differently from an omitted/defaulted one.

The regression I would add first is:

• capture before = fig.to_plotly_json(),
• assign the existing marker.colors back to itself,
• capture after = fig.to_plotly_json(),
• assert whether before == after.

That would make the ownership boundary clear and avoid debugging the renderer before confirming whether Python changed the payload.