Memory leak when trying to repeatedly display plots using the same output widget

[spott]

When I try to use bokeh in Jupyterlab to plot multiple times in a single Output widget, the old bokeh plots appear to stick around in memory even though they aren't actually accessible.

The following code will reproduce the issue (open up chrome's memory profiler and watch the memory increase as you click next).

from bokeh.plotting import figure, output_notebook, show

import ipywidgets as widgets
from ipywidgets import Layout, VBox
output_notebook()
import numpy as np

def plot_sample_bokeh(i):
    f = figure(title="test")
    f.line(np.arange(100), np.random.randn(100))

    show(f)

to_annotate = np.arange(100)
cur_index = 0

def display_img_widget(i):
    with out_widget:
        plot_sample_bokeh(i)

def on_button_click(b):
    global to_annotate, cur_index
    cur_id = to_annotate[cur_index]

    cur_index += 1

    if cur_index == len(to_annotate):
        print('End of collection')
        return
    out_widget.clear_output(wait=True)
    display_img_widget(to_annotate[cur_index])

a_button = widgets.Button(description='next', button_style='primary')
a_button.on_click(on_button_click)

out_widget = widgets.Output(width=1100, height=600)

display_img_widget(to_annotate[0])

form_item_layout = Layout(
    display='flex',
    flex_flow='row',
    justify_content='flex-start', width='100%')

form = VBox([a_button, out_widget], layout=form_item_layout)

form

I'm fairly sure that this is a bokeh specific thing (replacing plot_sample_bokeh with an ipywidget.Label doesn't show the same behavior, but that is a pretty simple widget).

jupyterlab: 0.35.4 jupyterlab_bokeh: v0.6.3

[bryevdv]

This needs to be validated with jlab >= 3.0 and ext 3.x, or closed.

[bryevdv]

Closed for lack of information.