hive-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Zheng Shao (JIRA)" <>
Subject [jira] Commented: (HIVE-521) Move size, if, isnull, isnotnull to GenericUDF
Date Sun, 31 May 2009 23:42:07 GMT


Zheng Shao commented on HIVE-521:

Nice work.

Several issues:
1. It seems the code does not work if the boolean expression evaluates to null. Can you add
a test case (like ql/test/query/clientpositive/udf*.q) and run "ant test -Dtestcase=TestCliDriver
-Dqfile=udf_if.q -Doverwrite=true) to generate the standard output of the query? Also we need
to do "svn add" for both the query file and the result file.
2. GenericUDFIsNull and GenericUDFIsNotNull can be very simple - just test if the Object is
null or not in evaluate. We are now doing it in UDFIsNull etc but we did add special handling
code to make sure we can pass a complex object into UDF. That solution is half-baked (can
only test whether it's null/not null but there is no way to inspect the internal structure
because no ObjectInspector is passed) and very ugly. That's why we want to move it to GenericUDF.
3. The above argument is also true for size(). We have special handling code in UDF to make
sure we can convert a Hive map/list data structure to java map/list, which is also half-baked
and hacky.

Let me know if you have any questions on this.

> Move size, if, isnull, isnotnull to GenericUDF
> ----------------------------------------------
>                 Key: HIVE-521
>                 URL:
>             Project: Hadoop Hive
>          Issue Type: Improvement
>    Affects Versions: 0.4.0
>            Reporter: Zheng Shao
>            Assignee: Min Zhou
>             Fix For: 0.4.0
>         Attachments: HIVE-521-IF-2.patch, HIVE-521-IF.patch
> See HIVE-511 for an example of the move.
> size, if, isnull, isnotnull are all implemented with UDF but they are actually working
on variable types of objects. We should move them to GenericUDF for better type handling.
> This also helps to clean up the hack in doing type matching/type conversion in UDF.

This message is automatically generated by JIRA.
You can reply to this email to add a comment to the issue online.

View raw message