Fix xvals for datetime axis

[scottshambaugh]

Closes #208

@cphyc do you have setup instructions for the image comparison tests? I cannot get them to fail if I change them.

[scottshambaugh]

Now that CI is updated I'm able to update the image test, so this PR is good to go. Would still appreciate some docs/pointers on how to get the image tests running locally!

[scottshambaugh]

I'll note that while I didn't want to add pandas as a dependency, this does work for pandas timestamps as well. Here was my local test script:

import matplotlib.pyplot as plt
import pandas as pd
import numpy as np
from labellines import labelLines, labelLine

x = np.linspace(0, 10, 100)
t = pd.date_range(start='2020-01-01', periods=100)  # Creates a DatetimeIndex
df = pd.DataFrame({'x': x, 't': t})

fig, ax = plt.subplots()
ax.plot(df['t'], df['x'], label='x')
ax.plot(df['t'], df['x'] + 1, label='x')
labelLines(ax.get_lines(), zorder=2.5, xvals=[df['t'].iloc[0], df['t'].iloc[-1]])  # df['t'].iloc[-1] is a pandas Timestamp
plt.show(block=True)


fig, ax = plt.subplots()
datetimes = pd.to_datetime(df['t']).values  # Creates a numpy array with dtype np.datetime64
ax.plot(datetimes, df['x'], label='x')
labelLine(ax.get_lines()[0], x=datetimes[-1])
plt.show(block=True)

[cphyc]

The easy™ solution is to run on the CI and the "correct" version will appear in the artefacts of the build which you can download and add to your PR in hindsight.

[cphyc]

Shouldn't this place the first label on 2018/11/01 and the second on 2018/11/03?

[scottshambaugh]

That's true for a list input xvals=[dates[0], dates[-1]], but the tuple input adds padding equivalent to a blank label at the extremes automatically