API: Make searchsorted raise TypeError on mismatched types

[mroeschke]

In [1]: import pandas as pd

In [2]: s = pd.Series([1, 2, "1"])
   ...: s.sort_values()
TypeError: '<' not supported between instances of 'str' and 'int'

In [3]: s = pd.Series([1, 2])
   ...: s.searchsorted("1")
Out[3]: 0

In [4]: import numpy as np

In [5]: arr = np.array([1, 2])

In [6]: arr.searchsorted("1")
Out[6]: 0

Looks like we are mirroring this behavior from numpy, but since Series does not support sorting incomparable objects then it seems like searchsorted with incomparable objects should raise too

[jbrockmendel]

Im curious how often searchsorted is used by users. We use it in indexing a bunch.

This came up for me the other day in the context of trying to de-duplicate a test. Some but not all of our EAs' searchsorted methods have a check for no-NAs, since the presence of NAs would invalidate the sortedness assumption.

[jorisvandenbossche]

And for example for DatetimeArray, we actually already do raise a TypeError:

In [2]: pd.Series(pd.date_range("2012-01-01", freq="D", periods=3)).searchsorted("1")
...
TypeError: value should be a 'Timestamp', 'NaT', or array of those. Got 'str' instead.

So some (all?) of our EAs already seem to be more strict

[divya1974]

take