[
https://issues.apache.org/jira/browse/HIVE-10716?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14560403#comment-14560403
]
Gopal V commented on HIVE-10716:
--------------------------------
The easiest fix to the problem seems to be an additional filter expr to produce
an AND()
{code}
hive> explain select avg(ss_sold_date_sk) from store_sales where (case
ss_sold_date when '1998-01-02' then 1 else null end)=1;
Map Operator Tree:
TableScan
alias: store_sales
filterExpr: CASE (ss_sold_date) WHEN ('1998-01-02') THEN
(true) ELSE (null) END (type: int)
Statistics: Num rows: 2474913 Data size: 9899654 Basic stats:
COMPLETE Column stats: COMPLETE
{code}
vs
{code}
hive> explain select avg(ss_sold_date_sk) from store_sales where (case
ss_sold_date when '1998-01-02' then 1 else null end)=1 and ss_sold_time_Sk > 0;
Map Operator Tree:
TableScan
alias: store_sales
filterExpr: ((ss_sold_date = '1998-01-02') and
(ss_sold_time_sk > 0)) (type: boolean)
Statistics: Num rows: 1237456 Data size: 9899654 Basic stats:
COMPLETE Column stats: COMPLETE
Filter Operator
predicate: (ss_sold_time_sk > 0) (type: boolean)
{code}
[~ashutoshc]: any idea why the extra filter helps in fixing the PPD case?
> Fold case/when udf for expression involving nulls in filter operator.
> ---------------------------------------------------------------------
>
> Key: HIVE-10716
> URL: https://issues.apache.org/jira/browse/HIVE-10716
> Project: Hive
> Issue Type: New Feature
> Components: Logical Optimizer
> Affects Versions: 1.3.0
> Reporter: Ashutosh Chauhan
> Assignee: Ashutosh Chauhan
> Attachments: HIVE-10716.patch
>
>
> From HIVE-10636 comments, more folding is possible.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
