Potential regression in master re empty Extension Indexes

[max-sixty]

Setting an additional index on a DataFrame with an empty PeriodIndex raises a couple of exceptions:

In [2]: 
df = pd.DataFrame(dict(a=1), index=pd.PeriodIndex(start='2000', freq='B', periods=0, name='date'))
df
Out[2]:
Empty DataFrame
Columns: [a]
Index: []


[ins] In [10]: df.set_index('a', append=True)
---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
~/workspace/pandas/pandas/core/arrays/categorical.py in __init__(self, values, categories, ordered, dtype, fastpath)
    389             try:
--> 390                 codes, categories = factorize(values, sort=True)
    391             except TypeError:

~/workspace/pandas/pandas/util/_decorators.py in wrapper(*args, **kwargs)
    176                     kwargs[new_arg_name] = new_arg_value
--> 177             return func(*args, **kwargs)
    178         return wrapper

~/workspace/pandas/pandas/core/algorithms.py in factorize(values, sort, order, na_sentinel, size_hint)
    631
--> 632     uniques = _reconstruct_data(uniques, dtype, original)
    633

~/workspace/pandas/pandas/core/algorithms.py in _reconstruct_data(values, dtype, original)
    147     if is_extension_array_dtype(dtype):
--> 148         values = dtype.construct_array_type()._from_sequence(values)
    149     elif is_datetime64tz_dtype(dtype) or is_period_dtype(dtype):

~/workspace/pandas/pandas/core/arrays/period.py in _from_sequence(cls, scalars, dtype, copy)
    201
--> 202         freq = freq or libperiod.extract_freq(periods)
    203         ordinals = libperiod.extract_ordinals(periods, freq)

~/workspace/pandas/pandas/_libs/tslibs/period.pyx in pandas._libs.tslibs.period.extract_freq()
   1485
-> 1486     raise ValueError('freq not specified and cannot be inferred')
   1487

ValueError: freq not specified and cannot be inferred

During handling of the above exception, another exception occurred:

NotImplementedError                       Traceback (most recent call last)
<ipython-input-10-7d64bc1a5646> in <module>
----> 1 df.set_index('a', append=True)

~/workspace/pandas/pandas/core/frame.py in set_index(self, keys, drop, append, inplace, verify_integrity)
   4051                     to_remove.append(col)
   4052
-> 4053         index = ensure_index_from_sequences(arrays, names)
   4054
   4055         if verify_integrity and not index.is_unique:

~/workspace/pandas/pandas/core/indexes/base.py in ensure_index_from_sequences(sequences, names)
   5058         return Index(sequences[0], name=names)
   5059     else:
-> 5060         return MultiIndex.from_arrays(sequences, names=names)
   5061
   5062

~/workspace/pandas/pandas/core/indexes/multi.py in from_arrays(cls, arrays, sortorder, names)
   1332         from pandas.core.arrays.categorical import _factorize_from_iterables
   1333
-> 1334         labels, levels = _factorize_from_iterables(arrays)
   1335         if names is None:
   1336             names = [getattr(arr, "name", None) for arr in arrays]

~/workspace/pandas/pandas/core/arrays/categorical.py in _factorize_from_iterables(iterables)
   2601         # For consistency, it should return a list of 2 lists.
   2602         return [[], []]
-> 2603     return map(list, lzip(*[_factorize_from_iterable(it) for it in iterables]))

~/workspace/pandas/pandas/core/arrays/categorical.py in <listcomp>(.0)
   2601         # For consistency, it should return a list of 2 lists.
   2602         return [[], []]
-> 2603     return map(list, lzip(*[_factorize_from_iterable(it) for it in iterables]))

~/workspace/pandas/pandas/core/arrays/categorical.py in _factorize_from_iterable(values)
   2573         # but only the resulting categories, the order of which is independent
   2574         # from ordered. Set ordered to False as default. See GH #15457
-> 2575         cat = Categorical(values, ordered=False)
   2576         categories = cat.categories
   2577         codes = cat.codes

~/workspace/pandas/pandas/core/arrays/categorical.py in __init__(self, values, categories, ordered, dtype, fastpath)
    400
    401                 # FIXME
--> 402                 raise NotImplementedError("> 1 ndim Categorical are not "
    403                                           "supported at this time")
    404

NotImplementedError: > 1 ndim Categorical are not supported at this time

Expected Output

Problem description

This works fine on 0.23.4:

[ins] In [1]: import pandas as pd

[ins] In [2]: df = pd.DataFrame(dict(a=1), index=pd.PeriodIndex(start='2000', freq='B', periods=0, name='date'))
         ...:          ...: df
Out[2]:
Empty DataFrame
Columns: [a]
Index: []

[ins] In [3]: df.set_index('a', append=True)
Out[3]:
Empty DataFrame
Columns: []
Index: []

Output of pd.show_versions()

INSTALLED VERSIONS

commit: b7294dd
python: 3.7.1.final.0
python-bits: 64
OS: Darwin
OS-release: 18.0.0
machine: x86_64
processor: i386
byteorder: little
LC_ALL: None
LANG: en_US.UTF-8
LOCALE: en_US.UTF-8

pandas: 0.19.0+4049.gb7294dd3e
pytest: 3.9.2
pip: 18.1
setuptools: 40.6.2
Cython: 0.28.5
numpy: 1.15.2
scipy: None
pyarrow: None
xarray: 0.10.9
IPython: 7.1.1
sphinx: None
patsy: 0.5.0
dateutil: 2.7.5
pytz: 2018.5
blosc: None
bottleneck: 1.2.1
tables: None
numexpr: 2.6.8
feather: None
matplotlib: 3.0.0
openpyxl: None
xlrd: 1.1.0
xlwt: 1.3.0
xlsxwriter: None
lxml: None
bs4: 4.6.0
html5lib: 1.0.1
sqlalchemy: 1.2.12
pymysql: None
psycopg2: None
jinja2: 2.8.1
s3fs: None
fastparquet: None
pandas_gbq: 0.6.1+2.gd98c621
pandas_datareader: None
gcsfs: None

[TomAugspurger]

Root cause is that we don't factorize empty period arrays correctly

In [2]: arr = pd.PeriodIndex([], freq='D')

In [3]: pd.factorize(arr)
---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
<ipython-input-3-65d36072b155> in <module>()
----> 1 pd.factorize(arr)

~/sandbox/pandas/pandas/util/_decorators.py in wrapper(*args, **kwargs)
    175                 else:
    176                     kwargs[new_arg_name] = new_arg_value
--> 177             return func(*args, **kwargs)
    178         return wrapper
    179     return _deprecate_kwarg

~/sandbox/pandas/pandas/core/algorithms.py in factorize(values, sort, order, na_sentinel, size_hint)
    630                                         assume_unique=True)
    631
--> 632     uniques = _reconstruct_data(uniques, dtype, original)
    633
    634     # return original tenor

~/sandbox/pandas/pandas/core/algorithms.py in _reconstruct_data(values, dtype, original)
    146     from pandas import Index
    147     if is_extension_array_dtype(dtype):
--> 148         values = dtype.construct_array_type()._from_sequence(values)
    149     elif is_datetime64tz_dtype(dtype) or is_period_dtype(dtype):
    150         values = Index(original)._shallow_copy(values, name=None)

~/sandbox/pandas/pandas/core/arrays/period.py in _from_sequence(cls, scalars, dtype, copy)
    200             periods = periods.copy()
    201
--> 202         freq = freq or libperiod.extract_freq(periods)
    203         ordinals = libperiod.extract_ordinals(periods, freq)
    204         return cls(ordinals, freq=freq)

~/sandbox/pandas/pandas/_libs/tslibs/period.pyx in pandas._libs.tslibs.period.extract_freq()
   1484             pass
   1485
-> 1486     raise ValueError('freq not specified and cannot be inferred')
   1487
   1488

ValueError: freq not specified and cannot be inferred

[TomAugspurger]

PeriodArray._from_sequence should probably check if isinstance(scalars, cls): return cls. Will want to check that dtype is None / matches scalars.dtype and handle copying correctly.

It'd be good to add a base extension test for factorizing an empty array.