BUG: max on axis=1 returns wrong values on type datetime64[ns] when NaT is present in values

[antoinefalck]

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

import pandas as pd
df = pd.DataFrame(["NaT", "2024-04-16 09:20:00.123456789"], dtype="datetime64[ns]")
df.max(axis=1)

# prints:
#
# 0                             NaT
# 1   2024-04-16 09:20:00.123456768
# dtype: datetime64[ns]

Issue Description

The max (axis=1) of a DataFrame of one column should return that column (type pd.Series).
However, when the dtype is datetime64[ns] and the column contains a NaT, there are small differences with the original and returned column.
In the MWE the difference is 21 nanoseconds.

Note: This only happens if there is a NaT in column values.

Expected Behavior

Expected behavior is to print

0                             NaT
1   2024-04-16 09:20:00.123456789
dtype: datetime64[ns]

Installed Versions

INSTALLED VERSIONS
------------------
commit                : 0691c5cf90477d3503834d983f69350f250a6ff7
python                : 3.11.10
python-bits           : 64
OS                    : Linux
OS-release            : 4.14.355-271.569.amzn2.x86_64
Version               : #1 SMP Tue Nov 5 10:11:37 UTC 2024
machine               : x86_64
processor             : x86_64
byteorder             : little
LC_ALL                : None
LANG                  : C.UTF-8
LOCALE                : en_US.UTF-8

pandas                : 2.2.3
numpy                 : 1.26.0
pytz                  : 2024.1
dateutil              : 2.9.0
pip                   : 24.3.1
Cython                : 0.29.30
sphinx                : None
IPython               : 8.21.0
adbc-driver-postgresql: None
adbc-driver-sqlite    : None
bs4                   : 4.12.3
blosc                 : None
bottleneck            : None
dataframe-api-compat  : None
fastparquet           : None
fsspec                : 2024.10.0
html5lib              : 1.1
hypothesis            : None
gcsfs                 : None
jinja2                : 3.1.4
lxml.etree            : None
matplotlib            : 3.8.1
numba                 : None
numexpr               : 2.10.1
odfpy                 : None
openpyxl              : 3.1.5
pandas_gbq            : None
psycopg2              : None
pymysql               : None
pyarrow               : 15.0.2
pyreadstat            : None
pytest                : None
python-calamine       : None
pyxlsb                : None
s3fs                  : 2024.10.0
scipy                 : 1.14.1
sqlalchemy            : None
tables                : 3.8.0
tabulate              : 0.9.0
xarray                : None
xlrd                  : None
xlsxwriter            : None
zstandard             : None
tzdata                : 2024.2
qtpy                  : None
pyqt5                 : None

[rhshadrach]

Thanks for the report! In _nanminmax, the presence of NaT causes pandas to coerce to float here:

pandas/pandas/core/nanops.py

Line 1100 in 1be2637

result = _maybe_null_out(result, axis, mask, values.shape)

It seems like it should be possible to avoid this coercion for datetimes as they are stored as integers here, NaT being the smallest (signed) integer value. Further investigations and PRs to fix are welcome!