No link to shared axes for Axis.set_units

[duncanmmacleod]

Bug report

Bug summary

The Axis.set_units method does not emit to linked axes in the same way that Axes.set_xlim does (by default).

Code for reproduction

from matplotlib import pyplot
from astropy import units
from astropy.visualization import quantity_support

data = units.Quantity([1, 2, 3, 4, 5], 'm')

with quantity_support():
    fig = pyplot.figure()
    ax1 = fig.add_subplot(2, 1, 1)
    ax2 = fig.add_subplot(2, 1, 2, sharey=ax1)
    ax1.plot(data)
    ax2.plot(data)
    print('After first plot:')
    print(ax1.yaxis.get_units())
    print(ax2.yaxis.get_units())
    ax1.yaxis.set_units(units.cm)
    print('After manual set_units():')
    print(ax1.yaxis.get_units())
    print(ax2.yaxis.get_units())

Actual outcome

After first plot:
m
m
After manual set_units():
m
cm

Expected outcome

If set_units was linked to all shared axes, I would expect the call out to ax1.yaxis.set_units to also operate on ax2.

This is probably best exposed via new set_xunits and set_yunits methods for the Axes object, that include an emit=True keyword to use self._shared_x_axes.get_siblings in the same way as Axes.set_{x,y}lim.

Matplotlib version

• Operating system: macOS 10.13.2
• Matplotlib version: 2.1.1
• Matplotlib backend (print(matplotlib.get_backend())): 'agg'
• Python version: 2.7.14
• Jupyter version (if applicable):
• Other libraries: astropy 2.0.2

all installed with macports.

[jklymak]

Units support are a bit inconsistent. Does astropy regularly use them for plotting? It'd be good if a dev who uses them regularly could help w/ them in Matplotlib. We have some old examples JPL wrote, which if I recall correctly don't quite work if you extend them at all.

See #9713

[duncanmmacleod]

@jklymak, astropy provides a ConversionInterface that integrates its own units package with the matplotlib infrastructure, but I have no idea if its widely used (I have only just started playing with it). I am happy to try and extend the unit infrastructure as I go along, which might not amount to much.

[jklymak]

It seems the Matplotlib should consider using a minimal units system for testing. We have one in examples/units/basic_units.py, but it doesn't do sensible things if you start mixing units from what I can tell. What happens if you do

ax.xaxis.set_units(cm)
ax.plot(x * cm, x)
ax.xaxis.set_units(m)
plt.show()

I'd expect to see things plotted in meters, but the conversion happens when plot is called, so it never knows about the m units. What happens if we subsequently do

ax.plot(datetime(2018,1,1))

on that axis?

This'll all take a bit of thought to do correctly, and maybe when JPL originally did this, they put the thought in, but subsequently its no longer as cohesive.