[ARROW-371] Python: Table with null timestamp becomes float in pandas

• Type: Bug
• Status: Resolved
• Priority: Minor
• Resolution: Fixed
• Affects Version/s: 0.1.0
• Fix Version/s: 0.2.0
• Component/s: Python
• External issue URL: https://github.com/apache/arrow/issues/15934
• Assignee: Uwe Korn
• Reporter: Jochen Ott
• Created: 08/Nov/16 10:20
• Updated: 11/Jan/23 07:08
• Resolved: 16/Nov/16 21:19

Description

I'm testing round-trip of pandas DataFrame to arrow Table and back.

Converting a pandas DataFrame with pd.Timestamp columns and NaT entries to a Table seems to work fine (null count and data type of the column match expectations). However, converting the Table back to a DataFrame represents the column as float64, not as pandas.Timestamp/numpy.datetime64[ns] as expected.