[HIVE-13942] Correctness of CASE folding in the presence of NULL values

• Type: Sub-task
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 2.1.1, 2.2.0
• Fix Version/s: 2.1.0
• Component/s: CBO
• Assignee: Jesús Camacho Rodríguez Assign to me
• Reporter: Jesús Camacho Rodríguez
• Created: 03/Jun/16 20:54
• Updated: 28/Feb/24 01:24
• Resolved: 07/Jun/16 13:05

Description

Introduced in HIVE-13068.

(case when key<'90' then 2 else 4 end) > 3 should not fold to key >= '90', as these two expressions are not equivalent (consider null values). Instead, it should fold to not NVL((key < '90'),false).

This is caused by 1) some methods still calling original RexUtil.simplify method where the bug was originally present, and 2) further improvements needed in HiveRexUtil.simplify.

Issue Links

• is broken by: HIVE-13068 Disable Hive ConstantPropagate optimizer when CBO has optimized the plan II
• is related to: HIVE-12742 NULL table comparison within CASE does not work as previous hive versions
• links to: RB link