[IMPALA-850] Incorrect results with codegen on multi-column group by with NULLs.

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Affects Version/s: Impala 1.2.3
• Fix Version/s: Impala 1.3
• Assignee: Nong Li
• Reporter: Alexander Behm
• Created: 04/Mar/14 07:40
• Updated: 10/Mar/14 22:27
• Resolved: 10/Mar/14 22:27

Description

Looks like a bug in subexpression elimination. I manually disabled it and the query results were correct even with codegen enabled.

Data

create table foo (x int, y int);
insert into foo values(1, 1);
insert into foo values(1, 2);
insert into foo values(2, 2);
insert into foo values(1, NULL);
insert into foo values(2, NULL);

Query with codegen enabled

Query: select x > 1, y > 1, x from foo group by 1, 2, 3
+-------+-------+------+
| x > 1 | y > 1 | x    |
+-------+-------+------+
| false | NULL  | NULL |
| false | false | 1    |
| true  | NULL  | NULL |
| false | true  | 1    |
| true  | true  | 2    |
+-------+-------+------+

Query with codegen disabled (or with codegen enabled but subexpr elim commented out)

Query: select x > 1, y > 1, x from foo group by 1, 2, 3
+-------+-------+---+
| x > 1 | y > 1 | x |
+-------+-------+---+
| true  | NULL  | 2 |
| false | false | 1 |
| false | NULL  | 1 |
| false | true  | 1 |
| true  | true  | 2 |
+-------+-------+---+