Hyukjin Kwon created SPARK-61316:
------------------------------------
Summary: Encode datetime literals exactly, as PySpark's
TimestampType does
Key: SPARK-61316
URL: https://issues.apache.org/jira/browse/SPARK-61316
Project: Spark
Issue Type: Bug
Components: Connect
Affects Versions: connect-rust-4.3.0
Reporter: Hyukjin Kwon
The drop-in computes the microseconds of a datetime (lit() and createDataFrame)
from a float of seconds, `datetime.timestamp() * 1e6`, rounded. The float loses
the microseconds after the year 2255, so `F.lit(datetime(2300, 1, 1, 0, 0, 0,
123457, tzinfo=UTC))` sends .123456, and `F.lit(datetime(9999, 12, 31, 23, 59,
59, 999999))` raises OverflowError.
Official PySpark 4.3.0 computes the seconds as an integer (calendar.timegm of
the UTC tuple for an aware datetime, time.mktime of the local tuple for a naive
one) and adds the microseconds, so the values are exact from year 1 to year
9999.
--
This message was sent by Atlassian Jira
(v10.3.26#10030026)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
