Force log axes to remain positive

I have noticed when linking the axes of a log plot to a linear plot of the same data, the resulting figure fails gloriously when dragging the linear axis negative. This is related to issue #5389, #5549, and PR #5477.

I see two ways to prevent problems when users inadvertently designate a situation that does not really make sense for a log plot,

1. impose an axis-limit-override for log scale so the lower limit is ALWAYS greater than zero, i.e., ignore commands to set the limit to zero or negative
2. somehow determine minimum value to show, below the the lowest data value, then mirror the axis to negative, then show that space empty (log(n) > 0 so it would never have data), i.e., the smallest data value is 1e-9 so draw the axis at 1e-10 and mirror across the axis

I am not sure which would be preferred/easier, and there are likely other options worth considering, ideas?

Here is an simple example to demonstrate the strange behavior (note that this was run before PR #5477):

import numpy as np
from bokeh.models import Range1d
from bokeh.layouts import gridplot
from bokeh.plotting import figure, show, output_notebook
output_notebook()

x = np.linspace(0.001, 0.9, 300)
y = (np.sin(15 * x) / (15 * x)) ** 2 # fake small-angle scattering data

p1 = figure(title='Linear Scale', x_axis_label='q (1/A)', y_axis_label='I(q)',
            width=400, height=400)
p2 = figure(title='Log Scale', x_axis_label='q (1/A)', y_axis_label='I(q)', 
            width=400, height=400, x_axis_type='log', y_axis_type='log')

p1.circle(x, y)
p2.circle(x, y)

# pre PR #5477 bit to force the log range to be correct
x_range = np.min(x[x > 0]) * 0.9, np.max(x) * 1.1
y_range = np.min(y[y > 0]) * 0.9, np.max(y) * 1.1
p2.x_range = Range1d(*x_range)
p2.y_range = Range1d(*y_range)

# set the ranges to match
p1.x_range = p2.x_range
p1.y_range = p2.y_range

fig = gridplot([[p1, p2]])
show(fig)

When dragging the plots, the axis can reverse negative-to-positive, the zoom changes to where you can lose track of where the data is, and possibly other strange behaviors. Here is the original plot

and here is an example of the y-axes inverting

[StevenCHowell]

Thank you @clairetang6 for addressing this issue!

[StevenCHowell]

I just tested panning the plots generated by the code in my original comment. In the current dev build, 0.12.5dev11, panning works great. When the dragging the linear plot beyond the limits of the log plot, the log plot pauses until and the linear plot returns to possible limits of the log plot.

I also tested the current release build, 0.12.4 (0.12.4-py35_0), and the problems persist there.

As long as nothing gets reverted, I think this safe to close as the 0.12.5 update should resolve the issue.