MarkerCluster not working properly in Streamlit 1.34 onwards

[benjamin-awd]

It seems that Streamlit version 1.34 introduced some kind of regression for Folium -- specifically it disables the Spiderfy functionality in MarkerCluster (and I think any related interactivity)

Expected
(on click)

Current

To reproduce

pip install streamlit==1.34

import streamlit as st

st.set_page_config(
    page_title="streamlit-folium documentation",
    page_icon=":world_map:️",
    layout="wide",
)

left, right = st.columns(2)

with left:
    with st.echo():
        import folium
        import streamlit as st

        from streamlit_folium import st_folium
        from folium.plugins import MarkerCluster

        # center on Liberty Bell, add marker
        m = folium.Map(location=[39.949610, -75.150282], zoom_start=16)
        
        marker_cluster = MarkerCluster().add_to(m)
        folium.Marker(
            [39.949610, -75.150282], popup="Liberty Bell", tooltip="Liberty Bell"
        ).add_to(marker_cluster)
        folium.Marker(
            [39.949610, -75.150282], popup="Liberty Bell", tooltip="Liberty Bell"
        ).add_to(marker_cluster)
        folium.Marker(
            [39.949610, -75.150282], popup="Liberty Bell", tooltip="Liberty Bell"
        ).add_to(marker_cluster)

        # call to render Folium map in Streamlit
        st_data = st_folium(m, width=725)

with right:
    st_data

[benjamin-awd]

Spent a bit of time figuring out the bad commit upstream with git bisect, and it seems to be this one: streamlit/streamlit@ea17620

It seems to be a fairly innocuous change, but somehow breaks the spiderfy functionality.

[blackary]

Spent a bit of time figuring out the bad commit upstream with git bisect, and it seems to be this one: streamlit/streamlit@ea17620

It seems to be a fairly innocuous change, but somehow breaks the spiderfy functionality.

Wow, thanks for doing that digging! I'll reach out to the streamlit team and see if they have any insights.

[benjamin-awd]

thanks @blackary @randyzwitch!

I ended up creating a .tar.gz of 1.38 without that particular commit, and it seems to be working well for my app: https://hdb-kaki.streamlit.app/Town_Analysis

To reproduce, you can try running the script above with Streamlit 1.38 (the spiderfy functionality should be missing). Then drop that specific commit from the Streamlit commit history, recompile using make package, and add the tar.gz file to your dependencies and re-run.

Otherwise if it's useful, I created a demo branch with the original bug: https://github.com/benjamin-awd/hdb-kaki/tree/folium-regression-demo

Btw, had no issues running on Safari even without that commit

[kmcgrady]

Hey @randyzwitch @benjamin-awd ! Thanks for letting us know. I'm having trouble figuring out what is going on from Streamlit's side that could be causing the issue. We did notice that v0.20.0 works but v0.20.1 does not work. Looking at the Network logs I see the Marker Cluster is inserted after the map information (that all being said, I see scripts duplicated multiple times). This leads me to think this might be tied to this change surrounding the dynamic load of the JS instances.

[randyzwitch]

Interesting that this might both be a streamlit and a streamlit-folium issue

[sean-doody]

Edit: Something weird is going on. It initially worked fine with v0.20.0 after creating a fresh virtualenv, but if I refresh the app, it breaks again. If I kill the app and rerun streamlit run app.py, on first load, the data won't render to the map. On refresh, it will (but then the clusters don't respond to clicks).

For what it's worth, I have been experiencing the same issue running Streamlit 1.40 and streamlit-folium v0.23.1. Following @kmcgrady above, I installed v0.20.0 and now MarkerCluster is performing as expected (I am still using Streamlit 1.40).