xaxis limits change when plotting time series of different resolution as secondary_y

[bmu]

In [1]: N = 10000

In [2]: rng = pd.date_range('1/1/2000', periods=N, freq='min')

In [3]: ts = pd.Series(np.random.randn(N), index=rng)

In [4]: ax = ts.plot()

In [5]: ax.get_xlim()
Out[5]: (15778080.0, 15788079.0)

In [6]: ts.resample('D').plot(style='ro', ax=ax)
Out[6]: <matplotlib.axes.AxesSubplot at 0x4482c50>

In [7]: ax.get_xlim()
Out[7]: (15778080.0, 15788079.0)

In [8]: ts.resample('D').plot(style='ro', secondary_y=True, ax=ax)
Out[8]: <matplotlib.axes.AxesSubplot at 0x497e450>

In [9]: ax.get_xlim()
Out[9]: (10957.0, 10963.0)

As the xlim changes, the original data is not visible any more.

Could be related to #2185

[rosnfeld]

Possibly related to #2960 ?

[rosnfeld]

Replying to my own comment - given a bit of insight into what's going on in #2960 , I think this is a separate issue.

[rosnfeld]

I think the different resolution is a red herring, it seems basically time-series plotting (in tseries/plotting.py) generally doesn't play well with secondary axes.

This is an example with no resampling that has the same behavior:

index_1 = pd.date_range(start='2000-01-01', periods=4, freq='D')
index_2 = pd.date_range(start='2000-01-05', periods=4, freq='D')

s1 = pd.Series(5, index=index_1)
s2 = pd.Series(7, index=index_2)

ax = s1.plot()
s2.plot(secondary_y=True, ax=ax)

The second plot call overwrites the x_lim, probably because (internal details of tseries/plotting) _get_xlim(ax.get_lines()) only references the current axis, and secondary_y creates a second axis object (via twinx()), and the two axes aren't considered together. This doesn't happen with a non-timeseries plot, e.g. if you change the indices to [1, 2, 3, 4] and [5, 6, 7, 8], though I'm not sure why.