id,node_id,number,title,user,state,locked,assignee,milestone,comments,created_at,updated_at,closed_at,author_association,active_lock_reason,draft,pull_request,body,reactions,performed_via_github_app,state_reason,repo,type
1087126812,I_kwDOAMm_X85AzD0c,6102,Regression in datetime handling in plots,1562854,closed,0,,,10,2021-12-22T19:31:53Z,2022-01-09T20:33:29Z,2022-01-09T20:33:29Z,CONTRIBUTOR,,,,"#5794 (ea2886136dec7047186d) introduced a regression in whether or not pandas datetime converters are loaded or Matplotlib's.  This leads to basic Matplotlib-native plotting failing https://github.com/matplotlib/matplotlib/issues/22023  Previously matplotlib's converters were loaded, now pandas are being loaded, despite the downstream user not ever using xarray's plotting utilities.  

## test code

```python
import matplotlib.pyplot as plt
import numpy as np
import xarray as xr

import matplotlib.units as munits
print(munits.registry)
ds = xr.Dataset({""time"": [np.datetime64('2000-01-01'), np.datetime64('2000-01-02')],
                     ""sir"": [0, 1]})

fig, ax = plt.subplots()
# crashes:
ax.scatter(ds['time'], ds['sir'])
plt.show()
```

#### Previously:
```
{... 
<class 'numpy.datetime64'>: <matplotlib.dates._SwitchableDateConverter object at 0x106434ac0>, 
<class 'datetime.date'>: <matplotlib.dates._SwitchableDateConverter object at 0x106434ac0>, 
<class 'datetime.datetime'>: <matplotlib.dates._SwitchableDateConverter object at 0x106434ac0>}
```

#### Now:

```
{... <class 'numpy.datetime64'>: <pandas.plotting._matplotlib.converter.DatetimeConverter object at 0x17f288160>, 
<class 'datetime.date'>: <pandas.plotting._matplotlib.converter.DatetimeConverter object at 0x17f182250>, 
<class 'datetime.datetime'>: <pandas.plotting._matplotlib.converter.DatetimeConverter object at 0x17f1821c0>, 
<class 'pandas._libs.tslibs.timestamps.Timestamp'>: <pandas.plotting._matplotlib.converter.DatetimeConverter object at 0x17f16ff40>, 
<class 'pandas._libs.tslibs.period.Period'>: <pandas.plotting._matplotlib.converter.PeriodConverter object at 0x17f16ffa0>, 
<class 'datetime.time'>: <pandas.plotting._matplotlib.converter.TimeConverter object at 0x17f288130>}
```

As you can see, the pandas converters have been loaded without any use of pandas nor xarray plotting utilities.  

### Suggestion

Of course if xarray plotting is loaded, you should use and register what date converters you would like (I'd suggest `matplotlib.dates.ConciseConverter`, but your mileage may vary).  But I think if the user is just trying to use xarray to load a data set, they should not have decisions made for them about the converter (or any other plotting functions), and to prevent confusion they should get the default matplotlib converter since it handles `datetime64` just fine.  

I think it could also be argued that this is a pandas issue, in that just importing pandas should not automatically register their converters unless their plotting is used.  ping @TomAugspurger because I thought that was the plan, but apparently things changed.  And it indeed appears their converter has a bug in it for matplotlib scatter.  

Thanks!   
","{""url"": ""https://api.github.com/repos/pydata/xarray/issues/6102/reactions"", ""total_count"": 0, ""+1"": 0, ""-1"": 0, ""laugh"": 0, ""hooray"": 0, ""confused"": 0, ""heart"": 0, ""rocket"": 0, ""eyes"": 0}",,completed,13221727,issue