Timestamp returns a different type depending on whether the data is empty or not.

[kitagry]

Environment details

• OS type and version: macOS sonoma 14.6
• Python version: 3.11.9
• pip version: I use rye instead of pip. and rye version is 0.35
• pandas-gbq version: 0.26.1
• pandas version: 2.2.2

Code example

import pandas_gbq

df = pandas_gbq.read_gbq('SELECT timestamp FROM `bigquery-public-data.pypi.file_downloads` LIMIT 1')
print(df.dtypes)

df = pandas_gbq.read_gbq('SELECT timestamp FROM `bigquery-public-data.pypi.file_downloads` LIMIT 0')
print(df.dtypes)

got the following. I expected either us or ns

timestamp    datetime64[us, UTC]
dtype: object
timestamp    datetime64[ns, UTC]
dtype: object

When I downgrade pandas version to 1.5.3, the above both code returns datetime64[ns, UTC].

Thanks!

[tswast]

Thanks @kitagry for the report and the proposed fix!