[HIVE-9168] Vectorized Coalesce for strings is broken

• Type: Bug
• Status: Closed
• Priority: Critical
• Resolution: Fixed
• Affects Version/s: 0.13.0, 0.14.0
• Fix Version/s: 1.0.0
• Component/s: Vectorization
• Assignee: Matt McCline
• Reporter: Matt McCline
• Created: 18/Dec/14 23:58
• Updated: 19/Feb/15 18:22
• Resolved: 24/Dec/14 01:26

Description

Vectorized Coalesce uses BytesColumnVector.setElement which does not set the output string length correctly.

create table str_str_orc (str1 string, str2 string) stored as orc;

insert into table str_str_orc values (null, "X"), ("0", "X"), ("1", "X"), (null, "y");

EXPLAIN
SELECT
   str2, ROUND(sum(cast(COALESCE(str1, 0) as int))/60, 2) as result
from str_str_orc
GROUP BY str2;

SELECT
   str2, ROUND(sum(cast(COALESCE(str1, 0) as int))/60, 2) as result
from str_str_orc
GROUP BY str2;

EXPLAIN
SELECT COALESCE(str1, 0) as result
from str_str_orc;

SELECT COALESCE(str1, 0) as result
from str_str_orc;

Produces different results when vectorized and not vectorized.