matplotlib.ticker.LinearLocator view_limits algorithm improvement?

[maqifrnswa]

Inspecting the code of matplotlib.ticker.LinearLocator

matplotlib/lib/matplotlib/ticker.py

Line 1161 in 7d1a7c2

class LinearLocator(Locator):

you can see that the view limits are chosen such that the difference between vmin and vmax is an interger multiple of scale, which is itself a power of 10. Therefore, the range will be nicely divisible when divided by 10 (11 tickmarks).

matplotlib/lib/matplotlib/ticker.py

Line 1213 in 7d1a7c2

def view_limits(self, vmin, vmax):

Therefore, the view_limits function determines the limits assuming there are 11 ticks. This assumption is implicit in two lines:

matplotlib/lib/matplotlib/ticker.py

Line 1224 in 7d1a7c2

exponent, remainder = divmod(math.log10(vmax - vmin), 1)

matplotlib/lib/matplotlib/ticker.py

Line 1227 in 7d1a7c2

scale = 10 ** (-exponent)

Code is repeated here:

exponent, remainder = divmod(math.log10(vmax - vmin), 1)
if remainder < 0.5:
    exponent -= 1
scale = 10 ** (-exponent)
vmin = math.floor(scale * vmin) / scale
vmax = math.ceil(scale * vmax) / scale

Since we know the number of ticks, from self.num_ticks, we can generalize the current algorithm to be better suited for any number of ticks. Suggested generalized algorithm:

exponent, remainder = divmod(math.log10(vmax - vmin), math.log10(self.num_ticks-1))
if remainder < 0.5:
    exponent -= 1
scale = (self.num_ticks-1) ** (-exponent)
vmin = math.floor(scale * vmin) / scale
vmax = math.ceil(scale * vmax) / scale

This generalized expression reduces to the current form when self.num_ticks==11 (which is the current default). For other cases, here is an example:
when num_ticks = 10, vmin = 20, vmax=90
Current algorithm returns vmin = 20, vmax=90, corresponding to ticks spaced by 7.77778.

The proposed algorithm returns vmin = 18, vmax = 90, corresponding to ticks spaced by 8.

Is this something worth doing? The patch is trivial -- just changing two lines of code. I can turn this in to a pull request to illustrate if it is helpful.

[tacaswell]

A pull request would help to clairify (and will allow the tests to run).

[tacaswell]

provisionally tagging an 2.0 in case we like this change and it breaks tests, can be pushed out if needed.

[maqifrnswa]

No problem, here's a pull request:
#6146