BUG: Constructing series with Timedelta object results in datetime series

[Casper-Guo]

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

test = pd.Series([pd.Timedelta("NaT")])
print(test)

Issue Description

test is initialized to a series of datetime64 type. This gotcha is not documented anywhere and the result is counter-intuitive. Opening the issue in case this is unintended.

Expected Behavior

test is initialized to a series of timedelta64 type

Installed Versions

INSTALLED VERSIONS

commit : 0691c5c
python : 3.12.3
python-bits : 64
OS : Linux
OS-release : 5.15.167.4-microsoft-standard-WSL2
Version : #1 SMP Tue Nov 5 00:21:55 UTC 2024
machine : x86_64
processor : x86_64
byteorder : little
LC_ALL : None
LANG : C.UTF-8
LOCALE : C.UTF-8

pandas : 2.2.3
numpy : 2.2.3
pytz : 2025.1
dateutil : 2.9.0.post0
pip : 25.0.1
Cython : None
sphinx : 8.2.3
IPython : None
adbc-driver-postgresql: None
adbc-driver-sqlite : None
bs4 : 4.13.3
blosc : None
bottleneck : None
dataframe-api-compat : None
fastparquet : None
fsspec : None
html5lib : None
hypothesis : None
gcsfs : None
jinja2 : 3.1.5
lxml.etree : None
matplotlib : 3.10.1
numba : None
numexpr : None
odfpy : None
openpyxl : None
pandas_gbq : None
psycopg2 : None
pymysql : None
pyarrow : None
pyreadstat : None
pytest : 8.3.5
python-calamine : None
pyxlsb : None
s3fs : None
scipy : 1.15.2
sqlalchemy : None
tables : None
tabulate : None
xarray : None
xlrd : None
xlsxwriter : None
zstandard : None
tzdata : 2025.1
qtpy : None
pyqt5 : None

[arthurlw]

Thanks for opening this! Confirmed on main. I agree that the behavior feels a bit unintuitive.

[arthurlw]

I did some investigating and found that for datetime-related types (datetime64, timedelta64, etc) with the value pd.NaT, pandas stores them all as <class 'pandas._libs.tslibs.nattype.NaTType'>. This makes it impossible to differentiate between a Timestamp("NaT") and a Timedelta("NaT") during Series construction.

[chilin0525]

I also traced the source code and it looks like the dtype_if_all_nat is assigned by dtype_if_all_nat in this case. As @arthurlw mentioned, there's no reliable way to determine whether a NaTType originates from a Timestamp or a Timedelta. 😢

pandas/pandas/core/dtypes/cast.py

Lines 1195 to 1206 in 337d40e

	# error: Incompatible return value type (got "Union[ExtensionArray,
	# ndarray[Any, Any]]", expected "Union[ndarray[Any, Any], DatetimeArray,
	# TimedeltaArray, PeriodArray, IntervalArray]")
	return lib.maybe_convert_objects( # type: ignore[return-value]
	value,
	# Here we do not convert numeric dtypes, as if we wanted that,
	# numpy would have done it for us.
	convert_numeric=False,
	convert_non_numeric=True,
	convert_to_nullable_dtype=convert_to_nullable_dtype,
	dtype_if_all_nat=np.dtype("M8[s]"),
	)

pandas/pandas/_libs/lib.pyx

Lines 2808 to 2831 in 337d40e

	elif seen.nat_:
	if not seen.object_ and not seen.numeric_ and not seen.bool_:
	# all NaT, None, or nan (at least one NaT)
	# see GH#49340 for discussion of desired behavior
	dtype = dtype_if_all_nat
	if cnp.PyArray_DescrCheck(dtype):
	# i.e. isinstance(dtype, np.dtype)
	if dtype.kind not in "mM":
	raise ValueError(dtype)
	else:
	res = np.empty((<object>objects).shape, dtype=dtype)
	res[:] = NPY_NAT
	return res
	elif dtype is not None:
	# i.e. PeriodDtype, DatetimeTZDtype
	cls = dtype.construct_array_type()
	obj = cls._from_sequence([], dtype=dtype)
	taker = -np.ones((<object>objects).shape, dtype=np.intp)
	return obj.take(taker, allow_fill=True)
	else:
	# we don't guess
	seen.object_ = True
	else:
	seen.object_ = True