Indhumathi Muthumurugesh created HIVE-30228:
-----------------------------------------------
Summary: Vectorized LEAD/LAG returns wrong default value when the
default literal's type differs from the window column's type (e.g. DATE column,
string-literal default)
Key: HIVE-30228
URL: https://issues.apache.org/jira/browse/HIVE-30228
Project: Hive
Issue Type: Bug
Reporter: Indhumathi Muthumurugesh
When LEAD/LAG is used over a window with a literal default value whose SQL type
differs from the value column's type — e.g. LEAD(date_col, 1, '9999-01-01')
where the default is a STRING literal implicitly convertible to the column's
DATE type — vectorized execution silently returns the wrong default (epoch
1970-01-01) instead of the specified value, while non-vectorized (row-mode)
execution returns the correct value.
Repro:
CREATE TABLE t1 (id INT, cat STRING, app_date DATE, active STRING);
INSERT INTO t1 VALUES (1,'A',DATE'2024-03-01','Y');
SELECT id, cat, app_date,
LEAD(date_add(app_date,-1), 1, ('9999-01-01'))
OVER (PARTITION BY id, cat ORDER BY id, cat, app_date) AS app_end_date
FROM t1;
-- vectorized: returns 1970-01-01 for the last row in each partition (WRONG)
SET hive.vectorized.execution.reduce.enabled=false;
-- same query now correctly returns 9999-01-01
--
This message was sent by Atlassian Jira
(v10.3.26#10030026)
