BUG: Fix bug in to_datetime that occasionally throws FloatingPointErr…

[snitish]

…or when called on a float array with missing values

• closes BUG: not reproducible error FloatingPointError: overflow encountered in multiply in the following sequence: read_csv followed by to_datetime with pandas version 2.2.2 #58419
• Tests added and passed if fixing a bug or adding a new feature
• All code checks passed.
• Added type annotations to new arguments/methods/functions.
• Added an entry in the latest doc/source/whatsnew/v3.0.0.rst file if fixing a bug or adding a new feature.

The cause of the error is a numpy array created via np.empty(). Sometimes it contains garbage values which subsequently get passed to np.round causing errors.

[mroeschke]

Thanks @snitish

[vgolskiy]

Pandas v.2.2.3, the bug is still there. I am processing a pd.Series filled with float timestamps. Overflows while to_datetime conversion with following dt.round to ms:
time = pd.to_datetime(time, unit='ms', errors='coerce').dt.round(freq='ms')
The error is related to pd.to_datetime, you can see below.

More details:
File "../pandas/core/tools/datetimes.py", line 1067, in to_datetime
values = convert_listlike(arg._values, format)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "../pandas/core/tools/datetimes.py", line 407, in _convert_listlike_datetimes
return _to_datetime_with_unit(arg, unit, name, utc, errors)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "../pandas/core/tools/datetimes.py", line 512, in _to_datetime_with_unit
arr = cast_from_unit_vectorized(arg, unit=unit)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "conversion.pyx", line 149, in pandas._libs.tslibs.conversion.cast_from_unit_vectorized
File "../numpy/_core/fromnumeric.py", line 3710, in round
return _wrapfunc(a, 'round', decimals=decimals, out=out)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "../numpy/_core/fromnumeric.py", line 57, in _wrapfunc
return bound(*args, **kwds)
^^^^^^^^^^^^^^^^^^^^
FloatingPointError: overflow encountered in multiply

[asishm]

@vgolskiy yes, the fix will be released with 3.0 (unless this gets backported to the 2.3.x branch)

[RyuuOujiXS]

What workaround is recommended until pandas 3 is released?

[snitish]

@RyuuOujiXS you can call to_datetime() after dropping NaNs, then add back the NaNs in the right places.

[vgolskiy]

@snitish , so it was the NaN value that was causing overflow? Also, after removing NaNs, it is better to return NaTs as the data will have a datetime type by that moment.
@RyuuOujiXS I described the workaround in the last comment here #58419 (comment)