BUG: Series([big_timestamp], dtype=pd.DatetimeTZDtype("us", "US/Pacific")) raises

[jbrockmendel]

ts = pd.Timestamp(2999, 1, 1)
ts2 = ts.tz_localize("US/Pacific")

dtype = pd.DatetimeTZDtype("us", "US/Pacific")

>>> ser = pd.Series([ts2], dtype=dtype)
[...]
pandas._libs.tslibs.np_datetime.OutOfBoundsDatetime: Cannot cast 2999-01-01 00:00:00-08:00 to unit='ns' without overflow.

[AdrianDAlessandro]

take

[AdrianDAlessandro]

Further investigation reproduces this bug in multiple ways. It appears the underlying Cython code (in pandas._libs.tslib.pyx) still requires ns resolution. This is not limited to timezone-aware datetimes.

>>> pd.Series([pd.Timestamp(2999,1,1)], dtype="datetime64[us]")
[...]
pandas._libs.tslibs.np_datetime.OutOfBoundsDatetime: Cannot cast 2999-01-01 00:00:00 to unit='ns' without overflow., at position 0

Also if you don't specify the dtype for an out-of-bounds datetime then you just get an array with object dtype, and it doesn't infer it.

>>> pd.Series([pd.Timestamp(2000,1,1)])
0   2000-01-01
dtype: datetime64[ns]
>>> pd.Series([pd.Timestamp(2999,1,1)])
0    2999-01-01 00:00:00
dtype: object

[jbrockmendel]

closed by #55768