[
https://issues.apache.org/jira/browse/HIVE-15758?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16088261#comment-16088261
]
Vineet Garg commented on HIVE-15758:
------------------------------------
Your example looks semantically correct to me. Current rewrite in hive for
queries such as {code:sql} select * from part where p_size <> (select
count(p_size) from part pp where part.p_type = pp.p_type) {code} only require
one join (left outer join) but with the example you provided It will need two
joins.
Current rewrite is similar to your example but instead of row_id we use
correlated columns to group by and to join.
> Allow correlated scalar subqueries with aggregates which has non-equi join
> predicates
> -------------------------------------------------------------------------------------
>
> Key: HIVE-15758
> URL: https://issues.apache.org/jira/browse/HIVE-15758
> Project: Hive
> Issue Type: Sub-task
> Components: Logical Optimizer
> Reporter: Vineet Garg
> Assignee: Vineet Garg
> Labels: sub-query
> Attachments: HIVE-15758.1.patch, HIVE-15758.2.patch
>
>
> Queries such as
> {code} select * from part where p_size <> (select count(p_size) from part pp
> where part.p_type <> pp.p_type); {code} are currently not allowed since HIVE
> doesn't know how to rewrite such queries to preserve the correctness for
> cases when there is zero row
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
