[jira] [Commented] (DRILL-7195) Query returns incorrect result or does not fail when cast with is null is used in filter condition

["Aman Sinha (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/DRILL-7195?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16824243#comment-16824243 ] 

Aman Sinha commented on DRILL-7195:
-----------------------------------

[~vvysotskyi] regarding this issue, it is worth debating whether or not it really is a blocker.  The first scenario where no error is shown does not seem like a blocker. The second case with wrong result occurs only when the cast_empty_string_to_null is set to True, which is not the default setting.  Looking at the history of IS NULL simplification in Calcite, it is somewhat complex issue and I feel that any fix we do could potentially introduce other side effects which will prolong the release.  Can we document this for 1.16 and fix it in master ? 

> Query returns incorrect result or does not fail when cast with is null is used in filter condition
> --------------------------------------------------------------------------------------------------
>
>                 Key: DRILL-7195
>                 URL: https://issues.apache.org/jira/browse/DRILL-7195
>             Project: Apache Drill
>          Issue Type: Bug
>    Affects Versions: 1.16.0
>            Reporter: Volodymyr Vysotskyi
>            Assignee: Volodymyr Vysotskyi
>            Priority: Blocker
>             Fix For: 1.16.0
>
>
> 1. For the case when a query contains filter with a {{cast}} which cannot be done with {{is null}}, the query does not fail:
> {code:sql}
> select * from dfs.tmp.`a.json` as t where cast(t.a as integer) is null;
> +---+
> | a |
> +---+
> +---+
> No rows selected (0.142 seconds)
> {code}
> where
> {noformat}
> cat /tmp/a.json
> {"a":"aaa"}
> {noformat}
> But for the case when this condition is specified in project, query, as it is expected, fails:
> {code:sql}
> select cast(t.a as integer) is null from dfs.tmp.`a.json` t;
> Error: SYSTEM ERROR: NumberFormatException: aaa
> Fragment 0:0
> Please, refer to logs for more information.
> [Error Id: ed3982ce-a12f-4d63-bc6e-cafddf28cc24 on user515050-pc:31010] (state=,code=0)
> {code}
> This is a regression, for Drill 1.15 the first and the second queries are failed:
> {code:sql}
> select * from dfs.tmp.`a.json` as t where cast(t.a as integer) is null;
> Error: SYSTEM ERROR: NumberFormatException: aaa
> Fragment 0:0
> Please, refer to logs for more information.
> [Error Id: 2f878f15-ddaa-48cd-9dfb-45c04db39048 on user515050-pc:31010] (state=,code=0)
> {code}
> 2. For the case when {{drill.exec.functions.cast_empty_string_to_null}} is enabled, this issue will cause wrong results:
> {code:sql}
> alter system set `drill.exec.functions.cast_empty_string_to_null`=true;
> select * from dfs.tmp.`a1.json` t where cast(t.a as integer) is null;
> +---+
> | a |
> +---+
> +---+
> No rows selected (1.759 seconds)
> {code}
> where
> {noformat}
> cat /tmp/a1.json 
> {"a":"1"}
> {"a":""}
> {noformat}
> Result for Drill 1.15.0:
> {code:sql}
> select * from dfs.tmp.`a1.json` t where cast(t.a as integer) is null;
> +----+
> | a  |
> +----+
> |    |
> +----+
> 1 row selected (1.724 seconds)
> {code}



--
This message was sent by Atlassian JIRA
(v7.6.3#76005)