BUG: loc against CategoricalIndex may results in normal Index

[sinhrks]

Closes #11586.

After the PR:

import pandas as pd
index = pd.CategoricalIndex(list('aabbca'), categories=list('cabe'))
df = pd.DataFrame({'A' : np.arange(6,dtype='int64')}, index=index)

# OK (not changed)
df.loc[['a', 'b']].index
# CategoricalIndex([u'a', u'a', u'a', u'b', u'b'], categories=[u'c', u'a', u'b', u'e'], ordered=False, dtype='category')

# Fixed to return Categorical Index if value exists in categories
df.loc[['a', 'b', 'e']].index
# CategoricalIndex([u'a', u'a', u'a', u'b', u'b', u'e'], categories=[u'a', u'b', u'e'], ordered=False, dtype='category')

# raise KeyError otherwise (not changed)
df.loc[['a', 'b', 'x']].index
# KeyError: 'a list-indexer must only include values that are in the categories'

There are separate paths when CategoricalIndex values (codes) are unique and not-unique and both fixed.

Also, .reindex intentionally returns normal Index when passed values are not Categorical, I kept the behavior as it is. If .reindex can always return CategoricalIndex, above 2 separate fixes are not required (fixing .reindex to return CategoricalIndex should work both paths).

# return normal Index(not changed)
df.index.reindex(['a', 'b'])
# (Index([u'a', u'a', u'a', u'b', u'b'], dtype='object'), array([0, 1, 5, 2, 3]))

# return CategoricalIndex(not changed)
df.index.reindex(pd.Categorical(['a', 'b']))
# (CategoricalIndex([u'a', u'a', u'a', u'b', u'b'], categories=[u'a', u'b'], ordered=False, dtype='category'), array([0, 1, 5, 2, 3]))

[jreback]

s reindex should give back a like index to the target / meaning if u put in a Categorical we try to give u a Categorical; if u out in an Index u get an Index

note that what u get is not dependent on what the actual index is! this is on purpose

this is essentially the difference between .loc and .reindex

[sinhrks]

Ah yes, thus the behavior is unchanged (as in the example at the bottom of the body). The fix should only applied to .loc.

[jreback]

thanks!