[BUG] Impossible to change FactorRange to a lower dimension with different number of factors

Initially reported at https://discourse.bokeh.org/t/y-ticks-y-range-error-while-changing-from-2-dimensional-x-range-factors-to-1-dimensional/5106

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

Bokeh 2.0.1

Description of expected behavior and the observed behavior

It should be possible to switch between factors of a FactorRange without any issues.

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

from bokeh.io import show
from bokeh.layouts import column
from bokeh.models import Button, CustomJS, ColumnDataSource, FactorRange
from bokeh.plotting import figure

fig = figure(x_range=FactorRange(factors=[('a', 'b'), ('b', 'c')]))

ds = ColumnDataSource(data=dict(x=[['a', 'b'], ['b', 'c']], y=[1, 2]))
fig.vbar(x='x', top='y', source=ds)

b = Button(label="Switch to single level")

b.js_on_click(CustomJS(args=dict(fig=fig, ds=ds),
                       code="""\
                           fig.x_range.factors = ['a'];
                           ds.data = {x: ['a'], y: [1]};
                           ds.change.emit();
                       """))

show(column(b, fig))

At least one of the causes is that FactorRange.tops is not updated if we switch to 1D factors from 2D or 3D factors: https://github.com/bokeh/bokeh/blob/cd74b12ed245c3d53c8bf21420de0f14979aebd0/bokehjs/src/lib/models/ranges/factor_range.ts#L240 There seems to be at least one other issue because the error that you see may depend on the order of the instructions within the CustomJS.code above. Seems like a race condition.

[mattpap]

In bokeh this:

plot.x_range.update(start=1, end=2)

and in bokehjs:

plot.x_range.setv({start: 1, end: 2})

are equivalent to:

plot.x_range.start = 0
plot.y_range.start = 0

when it comes to client-server updates (signals/events are a separate concern). If you look at the implementation, you will see a simple for loop distributing over changed properties in both cases. Those for loops need to be folded to allow pushing multiple changes in one protocol-level change event.

Additionally it would be nice to allow to synchronize a block of code like this:

plot.x_range.start = 0
plot.y_range.start = 0

so that one isn’t forced to use setv() or update(). I already implemented this locally and it will be part of bokeh 3.0. In bokehjs one will write:

plot.synchronized(() => {
  plot.x_range.start = 0
  plot.y_range.start = 0
})

and in bokeh there will be a context manager (as a substitute for multiline lambdas).

[bryevdv]

Yes, I mean that we should take existing APIs and improve them. Especially since Document is the correct place for these kind of synchronizations, not Plot (there may not be a plot at all, or changes that span multiple Plots). AFAICT there is an obvious place for unhold to generate a single protocol message instead of looping at the end.