[jira] [Updated] (DRILL-1180) Case messes up the datatype returned by function surrounding it

["Mehant Baid (JIRA)" <ji...@apache.org>]

     [ https://issues.apache.org/jira/browse/DRILL-1180?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Mehant Baid updated DRILL-1180:
-------------------------------

    Attachment: DRILL-1180.patch

This patch can be independently merged into drill without breaking any tests, but part of the bug fix is also in Optiq, pull request: https://github.com/jacques-n/optiq/pull/10

> Case messes up the datatype returned by function surrounding it
> ---------------------------------------------------------------
>
>                 Key: DRILL-1180
>                 URL: https://issues.apache.org/jira/browse/DRILL-1180
>             Project: Apache Drill
>          Issue Type: Bug
>            Reporter: Ramana Inukonda Nagaraj
>            Assignee: Mehant Baid
>            Priority: Critical
>             Fix For: 0.4.0
>
>         Attachments: DRILL-1180.patch
>
>
> Hit this while investigating tpch data variation between postgres and drill
> Simplified tpch14 to the following query:
> select 
>  sum(case
>     when l.L_RETURNFLAG like 'R%'
>       then l.l_extendedprice * (1 - l.l_discount)
>     else 0
>   end)
> from lineitem l;
> returns bigint in the case of drill and double in the case of postgres. Extendedprice and discount are double though.
> Drill:507996494
> Postgres:507996454.406699
> However when the case is removed and we use an equivalent filter instead drill and postgres return the same results:
> select 
> sum(l.l_extendedprice * (1 - l.l_discount)
> from lineitem l where l.L_RETURNFLAG like 'R%';
> Postgres: 507996454.406699
> Drill: 5.0799645440669966E8
> This would explain the data mismatch for both TPCH14 and 8
> git.commit.id.abbrev=e5c2da0



--
This message was sent by Atlassian JIRA
(v6.2#6252)