[BUG] data duplication on save (maybe render?)

[knaaptime]

Context

if i create several independent Maps and write each out to file, each successive file contains all of the data from prior maps until the kernel is restarted

Resulting behaviour, error message or logs

every map after the first contains data from all prior maps

Environment

macos 15
geopandas : 1.0.1
geodatasets: 2024.8.0
lonboard : 0.10.3

Steps to reproduce the bug

reproducible example. This will write 5 test files, the first of which is 3.3mb, the last of which is 16.7

import geodatasets
import geopandas as gpd
from lonboard import Map, PolygonLayer, viz

gdf = gpd.read_file(geodatasets.get_path("geoda.milwaukee1"))
gdf = gdf[["HH_INC", "geometry"]]

for i in range(5):
    Map([PolygonLayer.from_geopandas(gdf)]).to_html(f'test{i}.html')

maybe this is anywidget related? I'm pretty sure this happens each time a map is rendered, not necessarily each time its written to file. If you recreate a bunch of maps in the same notebook (as in re-run the cells, not generate a billion maps) it will blow up the RAM and eventually crash the kernel

[mgax]

Even creating PolygonLayer instances has an effect. Running this code as-is generates a 3.3M file, and with the loop uncommented, the file is 5M:

import geodatasets
import geopandas as gpd

from lonboard import Map, PolygonLayer

gdf = gpd.read_file(geodatasets.get_path("geoda.milwaukee1"))
gdf = gdf[["HH_INC", "geometry"]]

## uncomment to get larger file:
# for i in range(5):
#     PolygonLayer.from_geopandas(gdf)

Map([PolygonLayer.from_geopandas(gdf)]).to_html("test.html")

[mgax]

So what I think is causing the issue is that Map.to_html() calls embed_minimal_html from ipywidgets, which calls embed_data. But embed_data is called with no state, so it calls Widget.get_manager_state, which returns the state of all widgets ever created.

[kylebarron]

Thanks for tracking that down! Would you be able to test out #726?

[mgax]

Would you be able to test out #726?

@kylebarron I've run the same code on your branch and it the file size stays consistent at 3MB 👍

If a program generates a lot of these files in a loop, it's still going to leak memory, but that seems like a known ipywidgets issue (jupyter-widgets/ipywidgets#1345). The ticket suggests a workaround – explicitly closing down widgets when we're done with them – but that's tedious and error-prone.