BUG: inconsistent None handling for ISO and non-ISO formats

[MarcoGorelli]

Pandas version checks

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

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

• I have confirmed this bug exists on the main branch of pandas.

Reproducible Example

In [15]: to_datetime(['2020-01-01 00:00:00+01:00', '2020-01-01 00:00:00+02:00', None], format='%Y-%m-%d %H:%M:%S%z')
Out[15]: Index([2020-01-01 00:00:00+01:00, 2020-01-01 00:00:00+02:00, None], dtype='object')

In [16]: to_datetime(['2020-01-01 00:00:00+01:00', '2020-01-01 00:00:00+02:00', None], format='%Y-%d-%m %H:%M:%S%z')
Out[16]: Index([2020-01-01 00:00:00+01:00, 2020-01-01 00:00:00+02:00, NaT], dtype='object')

Issue Description

In the first (ISO) case, None stays None. But in the second (non-ISO) case, None becomes NaT.

Expected Behavior

@jbrockmendel any thoughts on what should be correct?

I think I'd prefer the latter (just change to NaT)

Installed Versions

INSTALLED VERSIONS

commit : 8dab54d
python : 3.8.15.final.0
python-bits : 64
OS : Linux
OS-release : 5.10.102.1-microsoft-standard-WSL2
Version : #1 SMP Wed Mar 2 00:30:59 UTC 2022
machine : x86_64
processor : x86_64
byteorder : little
LC_ALL : None
LANG : en_GB.UTF-8
LOCALE : en_GB.UTF-8

pandas : 1.5.2
numpy : 1.23.5
pytz : 2022.6
dateutil : 2.8.2
setuptools : 65.5.1
pip : 22.3.1
Cython : None
pytest : None
hypothesis : None
sphinx : None
blosc : None
feather : None
xlsxwriter : None
lxml.etree : None
html5lib : None
pymysql : None
psycopg2 : None
jinja2 : 3.1.2
IPython : 8.6.0
pandas_datareader: None
bs4 : 4.11.1
bottleneck : None
brotli : None
fastparquet : None
fsspec : None
gcsfs : None
matplotlib : None
numba : None
numexpr : None
odfpy : None
openpyxl : None
pandas_gbq : None
pyarrow : None
pyreadstat : None
pyxlsb : None
s3fs : None
scipy : None
snappy : None
sqlalchemy : None
tables : None
tabulate : None
xarray : None
xlrd : None
xlwt : None
zstandard : None
tzdata : None
None

[MarcoGorelli]

@mroeschke do you have thoughts on which is correct?

[jbrockmendel]

i think id expect both to be NaT and both to be DatetimeIndex

[MarcoGorelli]

I don't think DatetimeIndex is possible with mixed offsets, is it? (note how the strings have different offsets)

[jbrockmendel]

didnt notice that, you're right

[mroeschke]

Should the errors keyword be applicable here?

Since there's an implicit to_datetime(..., format=..., errors="raise") shouldn't this raise since format can't be applied to None? Not sure if we have testing for None being special

[MarcoGorelli]

yeah it's tested here

pandas/pandas/tests/tools/test_to_datetime.py

Lines 1343 to 1363 in 8da8743

	vals = [
	"nan",
	Timestamp("1990-01-01"),
	"2015-03-14T16:15:14.123-08:00",
	"2019-03-04T21:56:32.620-07:00",
	None,
	]
	ser = Series(vals)
	assert all(ser[i] is vals[i] for i in range(len(vals))) # GH#40111
	mixed = to_datetime(ser)
	expected = Series(
	[
	"NaT",
	Timestamp("1990-01-01"),
	Timestamp("2015-03-14T16:15:14.123-08:00").to_pydatetime(),
	Timestamp("2019-03-04T21:56:32.620-07:00").to_pydatetime(),
	None,
	],
	dtype=object,
	)

I just think expected is wrong on two counts:

• the first "NaT" should be NaT
• the None should be NaT

[mroeschke]

I guess in that particular example format is not provided thought.

I haven't looked at mixed-argument to_datetime testing in a while but leads me to 2 questions

1. If format is provided, should all arguments be strings (or datetimes that were recently added?) or can other arguments be parsed with non-formatting paths?
2. Should in theory a None.strftime(format) -> NaT parsing be supported? Since NaT.strp/stftime are designed to raise errors my initial thought is no but could be convinced otherwise

[mroeschke]

I just think expected is wrong on two counts:

Agreed with your conclusions

[MarcoGorelli]

Not sure it should depend on whether or not format was passed - what I'm leaning towards is:

• skip over any NaT-y (None, NaT, nan, 'nan', etc.) -> convert all to NaT
• skip (or rather, preserve) pydatetime objects
• if format is provided, parse according to that (respecting exact). Else, guess/infer

Note that with PDEP4, it'll always behave as if format was passed (unless pandas can't guess the format)

[mroeschke]

Ah okay I like those guidelines then.