[BUG] Clearing lines from streamed data using ColumnDataSource fails if data is arrays

[Lif3line]

Issue

When attempting to clear drawn lines from streamed data via source.stream(data, rollover=1) the clear doesn't work if the line's x and y are numpy arrays in the associated ColumnDataSource. Instead a new line is drawn using the last of the old data and the new data.

This is different to the behaviour if either x or y is a list.

Minimal Example

Tested under Windows 10 + Mozilla Firefox in Jupyter Lab with the following packages

python=3.7.7
bokeh=2.0.2
numpy=1.18.1
jupyterlab=1.2.6
@bokeh/jupyter_bokeh v1.2.0

Actual Example:

from bokeh.io import output_notebook, show, push_notebook
from bokeh.models import ColumnDataSource
from bokeh.plotting import figure

import numpy as np

output_notebook()

notebook_url="http://localhost:8888"

p = figure(match_aspect=True)

source1 = ColumnDataSource(data={"x": [-1, 0], "y1": [-1, -1] })
p.line(x='x', y='y1', source=source1, line_width=6, legend_label="List Source", line_color="blue")

source2 = ColumnDataSource(data={"x": np.array([0, 1]), "y1": np.array([0, 0]) })
p.line(x='x', y='y1', source=source2, line_width=6, legend_label="Array Source", line_color="red")

source3 = ColumnDataSource(data={"x": np.array([1, 2]), "y1": [1, 1] })
p.line(x='x', y='y1', source=source3, line_width=6, legend_label="Array X, List Y", line_color="yellow")

source4 = ColumnDataSource(data={"x": [2, 3], "y1": np.array([2, 2]) })
p.line(x='x', y='y1', source=source4, line_width=6, legend_label="List X, Array Y", line_color="orange")

plot_handle = show(p, notebook_url=notebook_url, notebook_handle=True)

### Attempt to clear data
# Update list only source
source1.stream({"x": [4], "y1": [4]}, rollover=1)

# Update array only source
source2.stream({"x": np.array([4]), "y1": np.array([4])}, rollover=1)

# Update array x, list y source
source3.stream({"x": np.array([4]), "y1": [4]}, rollover=1)

# Update array x, list y source
source4.stream({"x": [4], "y1": np.array([4])}, rollover=1)

push_notebook(handle=plot_handle)

This raises a Javascript console warning: [bokeh] data source has columns of inconsistent lengths caused by the sources with mixed data types. Same behaviour exists when these are removed.

[bryevdv]

Noting first that I cannot reproduce any issue in classic notebook. Will have to try later with jupyterlab

[mattpap]

Can't reproduce on Linux in jupyterlab in FF and Chrome. Maybe this is a Windows specific issue.

[Lif3line]

Interesting, I found I had the same issue in a classic notebook. I tested in a fresh anaconda environment with the following:

python=3.8.2
bokeh=2.0.2
numpy=1.18.4
notebook=6.0.3

[Lif3line]

Can confirm that I don't get the same issue under Ubuntu 18.04, using the same clean environemnt. Agreed - it appears to be Windows specific.