BUG: pd.to_datetime() raises InvalidIndexError with Null-like arguments

[blinkseb]

• I have checked that this issue has not already been reported.

• I have confirmed this bug exists on the latest version of pandas.

• (optional) I have confirmed this bug exists on the master branch of pandas.

---

Note: Please read this guide detailing how to provide the necessary information for us to reproduce your bug.

Code Sample, a copy-pastable example

import pandas as pd

s = pd.Series([pd.NaT] * 2000 + [None] * 2000, dtype='object')
pd.to_datetime(s)

Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/home/sbrochet/venvs/tmp-fa372ee62ee9bef/lib64/python3.8/site-packages/pandas/core/tools/datetimes.py", line 801, in to_datetime
    result = arg.map(cache_array)
  File "/home/sbrochet/venvs/tmp-fa372ee62ee9bef/lib64/python3.8/site-packages/pandas/core/series.py", line 3970, in map
    new_values = super()._map_values(arg, na_action=na_action)
  File "/home/sbrochet/venvs/tmp-fa372ee62ee9bef/lib64/python3.8/site-packages/pandas/core/base.py", line 1131, in _map_values
    indexer = mapper.index.get_indexer(values)
  File "/home/sbrochet/venvs/tmp-fa372ee62ee9bef/lib64/python3.8/site-packages/pandas/core/indexes/base.py", line 2980, in get_indexer
    raise InvalidIndexError(
pandas.errors.InvalidIndexError: Reindexing only valid with uniquely valued Index objects

Problem description

pd.to_datetime() crash if the input contains a mix of NaT and None with a pandas.errors.InvalidIndexError: Reindexing only valid with uniquely valued Index objects exception.
This issue is similar to #22305 which was fixed a while ago. It only occurs if the input is large enough to force the caching mechanism.

Expected Output

No crash

Output of pd.show_versions()

INSTALLED VERSIONS

commit : f2ca0a2
python : 3.8.5.final.0
python-bits : 64
OS : Linux
OS-release : 5.7.15-200.fc32.x86_64
Version : #1 SMP Tue Aug 11 16:36:14 UTC 2020
machine : x86_64
processor : x86_64
byteorder : little
LC_ALL : None
LANG : en_GB.UTF-8
LOCALE : en_GB.UTF-8

pandas : 1.1.1
numpy : 1.19.1
pytz : 2020.1
dateutil : 2.8.1
pip : 20.0.2
setuptools : 46.1.3
Cython : None
pytest : None
hypothesis : None
sphinx : None
blosc : None
feather : None
xlsxwriter : None
lxml.etree : None
html5lib : None
pymysql : 0.10.0
psycopg2 : None
jinja2 : None
IPython : None
pandas_datareader: None
bs4 : None
bottleneck : None
fsspec : None
fastparquet : None
gcsfs : None
matplotlib : None
numexpr : None
odfpy : None
openpyxl : None
pandas_gbq : None
pyarrow : None
pytables : None
pyxlsb : None
s3fs : None
scipy : None
sqlalchemy : 1.3.19
tables : None
tabulate : 0.8.7
xarray : None
xlrd : None
xlwt : None
numba : None

[pbjones]

I don't know if it's helpful but I also found the same / similar error today with None and to_datetime. If you are unable to replicate it, I can probably reduce the complexity of mine down (c. 58,000 rows, mixed None with dates).

From my workaround attempts, I found that I can use '' instead of None to get my desired outcome NaT. If memory serves me right, I couldn't do this previously. So amending the example above, this provides 4000 NaTs

    s = pd.Series([pd.NaT] * 2000 + [''] * 2000, dtype='object')
    pd.to_datetime(s)

[TomAugspurger]

I think the root problem is something like this: We try to get the unique dates and then pass it to an index.

In [10]: pd.unique(arg)
Out[10]: array([NaT, None], dtype=object)

In [11]: pd.Index(arg)
Out[11]: DatetimeIndex(['NaT', 'NaT'], dtype='datetime64[ns]', freq=None)

But Index infers datetime dtype and converts None to NaT. So we need to somewhere standardize the NA values. I'm not sure exactly where that should be done though. We'd welcome more investigation!

[mroeschke]

It appears this works on master now. Could use a test

In [4]: import pandas as pd
   ...:
   ...: s = pd.Series([pd.NaT] * 2000 + [None] * 2000, dtype='object')
   ...: pd.to_datetime(s)
Out[4]:
0      NaT
1      NaT
2      NaT
3      NaT
4      NaT
        ..
3995   NaT
3996   NaT
3997   NaT
3998   NaT
3999   NaT
Length: 4000, dtype: datetime64[ns]

[ryangilmour]

My first time contributing - but the PR above should hopefully be a sufficient test. Not sure these tests require an addition to whatsnew?