bokeh: Can't change type of figure.x_range dynamically

ALL software version info (bokeh, python, notebook, OS, browser, any other relevant packages)

$ bokeh info
Python version      :  3.7.0 (v3.7.0:1bf9cc5093, Jun 26 2018, 23:26:24)
IPython version     :  7.1.1
Tornado version     :  5.1.1
Bokeh version       :  1.0.1
BokehJS static path :  /path/to/my/python3.7/site-packages/bokeh/server/static
node.js version     :  v9.7.1
npm version         :  5.6.0

Description of expected behavior and the observed behavior

I am trying to implement a plot where the user can select a column from a dropdown and see the histogram of values in the column. Some columns may be categorical, and some are continuous, so I’ve using a vbar and setting the width accordingly. However, this means that I sometimes need a DataRange1d, and sometimes a FactorRange. Changing my figure’s x_range to a FactorRange causes my plot to be reset and not show any output. There is no error, the glyphs just disappear.

Complete, minimal, self-contained example code that reproduces the issue

This is about as minimal as I could get it, sorry.

from bokeh.plotting import Figure, ColumnDataSource
from bokeh.models.widgets import Dropdown
from bokeh.models.ranges import DataRange1d, FactorRange
from bokeh.layouts import row
from bokeh.io import curdoc

import pandas as pd
import numpy as np


def _col_stats(col,):
    """Value counts or histogram of column"""
    is_str = isinstance(col.values[0], str)

    if is_str:
        counts = col.value_counts().sort_index()
        hist = counts.values
        edges = np.array(counts.index)

    else:
        hist, edges = np.histogram(col, bins="auto")

    return hist, edges


def _hist_source_dict(hist, edges):
    if isinstance(edges[0], str):
        return {"top": hist, "x": edges, "width": 0.8 * np.ones_like(edges)}

    else:
        w = np.diff(edges)
        x = np.array(edges[:-1]) + w / 2
        return {"top": hist, "x": x, "width": w}


def _update_histogram(figure, source, hist, edges, name=""):
    """Why doesn't this work?"""
    if isinstance(edges[0], str):
        figure.x_range = FactorRange(factors=edges)
    else:
        figure.x_range = DataRange1d()

    source.data = _hist_source_dict(hist, edges)

    if name:
        figure.title.text = name


df = pd.DataFrame(
    {"a": [1, 2, 3, 4, 3, 1, 2, 3, 4, 1], "b": list("abcdcbaeda")}
)
columns = df.columns

# initial values
first_col = columns[0]
first_hist, first_edges = _col_stats(df[first_col])

# set up data sources
hist_source = ColumnDataSource(data=_hist_source_dict(first_hist, first_edges))

# set up dashboard components
plot = Figure(
    title=first_col,
    tools="xpan,xwheel_zoom,box_zoom,reset",
    active_drag="xpan",
    active_scroll="xwheel_zoom",
)

plot.vbar(top="top", x="x", width="width", bottom=0, source=hist_source)

dropdown = Dropdown(
    label="Column", menu=[(col, col) for col in df.columns], value=first_col
)

# configure callbacks
def dropdown_callback(attr, old, new):
    global df, plot, hist_source
    hist, edges = _col_stats(df[new])
    _update_histogram(plot, hist_source, hist, edges, new)


dropdown.on_change("value", dropdown_callback)

# Position elements in document
root = row(plot, dropdown)
curdoc().add_root(root)

If you save this and run it with bokeh serve, selecting ‘b’ from the dropdown will blank out the plot, but there is no error.

Screenshots or screencasts of the bug in action

Continuous column:

Categorical column: