BUG: Fix KeyError in merge on CategoricalIndex

[fjetter]

For categorical type indices a KeyError is raised when the index level is used during a merge on an index level

Example:

import pandas as pd
left = pd.DataFrame(
    {"left_data": [1, 2]},
   index=pd.CategoricalIndex(data=["A", "B"], categories=["A", "B"], name='index_col')
)
right = pd.DataFrame(
      {"right_data": [1.0, 2.0]},
   index=pd.CategoricalIndex(data=["A", "B"], categories=["A", "B"], name='index_col')
)
result = left.merge(right, on=['index_col'])

With this fix, the behavior of the test test_merge_datetime_index(self, klass) changed, though. IMHO, the behavior in this PR is more consistent since it is the same for all input types but I'm not sure what the actual behavior should be and I couldn't find a section in the documentation explaining this path.

• closes #xxxx
• tests added / passed
• passes git diff upstream/master -u -- "*.py" | flake8 --diff
• whatsnew entry

[jreback]

can u show. short example of what you are trying to do

[fjetter]

The following code raises a KeyError if the index is of categorical type but is ok for all other types

import pandas as pd
left = pd.DataFrame(
    {"left_data": [1, 2]},
   index=pd.CategoricalIndex(data=["A", "B"], categories=["A", "B"], name='index_col')
)
right = pd.DataFrame(
      {"right_data": [1.0, 2.0]},
   index=pd.CategoricalIndex(data=["A", "B"], categories=["A", "B"], name='index_col')
)
result = left.merge(right, on=['index_col'])

I updated the title and the description

[jreback]

if you can fix up the formatting will have another look. pls add a whatsnew note as well. (reshaping bug fixes)

[jreback]

write like

@pytest.mark.parametrize(
    'index',
     [

so it is closer to the left margin, then each of the indexes can be writen more simply, also don't use the data kwarg
e.g.
```Index(['A', 'B'], name='index_col')``

[jreback]

should be a 1-liner

[codecov]

Codecov Report

Merging #20777 into master will increase coverage by 0.03%.
The diff coverage is 100%.

@@            Coverage Diff             @@
##           master   #20777      +/-   ##
==========================================
+ Coverage   91.81%   91.85%   +0.03%     
==========================================
  Files         153      153              
  Lines       49471    49310     -161     
==========================================
- Hits        45422    45292     -130     
+ Misses       4049     4018      -31

Flag	Coverage Δ
#multiple	90.24% <100%> (+0.03%)	⬆️
#single	41.89% <50%> (+0.04%)	⬆️

Impacted Files	Coverage Δ
pandas/core/reshape/merge.py	94.25% <ø> (ø)	⬆️
pandas/core/algorithms.py	94.38% <100%> (-0.11%)	⬇️
pandas/core/indexing.py	93.08% <0%> (-0.48%)	⬇️
pandas/core/strings.py	98.32% <0%> (-0.31%)	⬇️
pandas/core/dtypes/base.py	91.89% <0%> (-0.22%)	⬇️
pandas/core/dtypes/cast.py	87.85% <0%> (-0.21%)	⬇️
pandas/core/indexes/api.py	98.78% <0%> (-0.15%)	⬇️
pandas/core/series.py	93.9% <0%> (-0.13%)	⬇️
pandas/core/dtypes/missing.py	92.85% <0%> (-0.09%)	⬇️
pandas/core/indexes/datetimelike.py	96.72% <0%> (-0.08%)	⬇️
... and 31 more

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update d3d3352...0664858. Read the comment docs.

[fjetter]

@jreback I fixed the formatting and added a changelog entry for the bug I fixed. Since I didn't know for sure if the other change in behavior is intended I left it out for now. Can add another entry once this is settled

[jreback]

thanks @fjetter nice patch!