[
https://issues.apache.org/jira/browse/HIVE-13555?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15453842#comment-15453842
]
Zoltan Haindrich commented on HIVE-13555:
-----------------------------------------
i've checked for reference postgresql's nullif...it converts types like:
{code}
nullif(integer,integer) => integer
nullif(text,integer) => integer
nullif(integer,text) => integer
nullif(text,text) => text
{code}
it demotes the the text argument into integer - in case the otherone is an
integer...and produces a syntax exception for nullif(1,'x')
the proposed method will keep the 'first' argument type, but will convert the
second if needed...in case someone puts in a string...this is good or bad?
i think this is somewhat expected...and will align with 'case when' current
implementation.
should I remove the conversion - be strict about types as much as possible -
and stick to the standard?
> Add nullif udf
> --------------
>
> Key: HIVE-13555
> URL: https://issues.apache.org/jira/browse/HIVE-13555
> Project: Hive
> Issue Type: Sub-task
> Components: UDF
> Reporter: Ashutosh Chauhan
> Assignee: Zoltan Haindrich
> Attachments: HIVE-13555.1.patch
>
>
> {{nullif(exp1, exp2)}} is shorthand for: {{case when exp1 = exp2 then null
> else exp1}}
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
