[
https://issues.apache.org/jira/browse/HIVE-28901?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17944419#comment-17944419
]
Raghav Aggarwal commented on HIVE-28901:
----------------------------------------
After further debugging, the issue can be seen in a simple query like (check
above q files for reference)
{code:java}
SELECT * from t1 where id = cast(474874000125288438 as string); {code}
in query plan, _{*}predicate:(UDFToDouble(id) = 4.7487400012528845E17D){*},_
both the bigint and string are converted to double compared before the value of
*E* which is rounded off and giving wrong result for large number.
*The join query is wrong.q is giving 2 rows (expected output) in Spark (3.4.1)
and in mysql.*
Here is the code which is converting string to DOUBLE:
[https://github.com/apache/hive/blob/master/serde/src/java/org/apache/hadoop/hive/serde2/typeinfo/TypeInfoUtils.java#L957-L959]
If I change it to LONG then it is working fine. _I understand for decimals,
converting to LONG will result in loss of scale._
[~zabetak] , [~kkasa] , can you please provide your take on this?
> Wrong query result for join between string and bigint
> -----------------------------------------------------
>
> Key: HIVE-28901
> URL: https://issues.apache.org/jira/browse/HIVE-28901
> Project: Hive
> Issue Type: Bug
> Reporter: Raghav Aggarwal
> Assignee: Raghav Aggarwal
> Priority: Major
> Attachments: working.q, working.q.out, wrong.q, wrong.q.out
>
>
> Attaching the wrong and expected output q and q.out file to reproduce the
> issue.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
