Scattergl freezes when y values have lots of zeros

This might be more to do with plotly.js or plotly.py, but I only have dash code to generate the plot so I am posting here. I do a lot of time series plotting for my work, so Scattergl comes in handy because I routinely plot 1000's of points on a trace. After updating to dash 1.21.0, I noticed the Scattergl plots hang on hovering when my trace has lots of consecutive zero y-values. In most cases the hover event will freeze the entire page and I need to close and reopen the page.

EDIT: I pinpointed this error to the upgrade from plotly.js v2.0.0-rc.1 to plotly.js v2.0.0-rc.2. If you put rc.1 in the assets folder, scattergl works fast like in the previous version of dash. If you put rc.2 in the assets folder, the issues return.

I am not a js developer, so I don't know how best to move this issue over to plotly.js with equivalent js example. Nevertheless, I looked the difference in rc.1 and rc.2 changelogs and found that in rc.2 there is an additional reference to scattergl: Improve scattergl performance when using typed arrays #5632, maybe that is causing the issue?

Check the code below for an example:

import plotly.graph_objects as go
import numpy as np
import dash
import dash_core_components as dcc
import pandas as pd
import random

# build dataset
N = 200000
index = pd.date_range("7/1/2020", "7/1/2021", N)
vals = np.zeros(N)

# add some random non-zero data around the dataset
for i in range(50):
    start = random.randint(0, N - 50)
    vals[start : start + 50] = np.random.random(50)

# uncomment this line and see how Scattergl is much faster without the consecutive zeros
# vals = np.random.random(N)


# change Scattergl to Scatter to see how Scatter performs better with consecutive zeros
fig = go.Figure()
fig.add_trace(
    go.Scattergl(
        x=index,
        y=vals,
    )
)

app = dash.Dash()
app.layout = dcc.Graph(figure=fig)

if __name__ == "__main__":
    app.run_server(host="0.0.0.0", port=8080, debug=True)

Commencez par exécuter la reproduction fournie en Dash/Python avec 200 000 points et des valeurs y nulles consécutives, puis comparez plotly.js v2.0.0-rc.1 à v2.0.0-rc.2. Lisez les modifications liées à scattergl dans PR #5632 ; le travail est terminé lorsque le survol ne fige plus la page pour le schéma de données signalé, tout en préservant le comportement existant de Scattergl.