[SPARK-20915][SQL] Make lpad/rpad with empty pad string same as MySQL.

[wangyum]

What changes were proposed in this pull request?

Spark SQL rpad/lpad with empty pad string:

spark-sql> select rpad('hello', -2, ''), rpad('hello', -1, '') , rpad('hello', 0, ''), rpad('hello', 1, ''), rpad('hello', 3, ''), rpad('hello', 5, ''), rpad('hello', 6, '');
			h	hel	hello	hello
spark-sql> select lpad('hello', -2, ''), lpad('hello', -1, '') , lpad('hello', 0, ''), lpad('hello', 1, ''), lpad('hello', 3, ''), lpad('hello', 5, ''), lpad('hello', 6, '');
			h	hel	hello	hello
spark-sql>

but the MySQL result:

mysql> select rpad('hello', -2, ''), rpad('hello', -1, '') , rpad('hello', 0, ''), rpad('hello', 1, ''), rpad('hello', 3, ''), rpad('hello', 5, ''), rpad('hello', 6, '') from dual;
+-----------------------+-----------------------+----------------------+----------------------+----------------------+----------------------+----------------------+
| rpad('hello', -2, '') | rpad('hello', -1, '') | rpad('hello', 0, '') | rpad('hello', 1, '') | rpad('hello', 3, '') | rpad('hello', 5, '') | rpad('hello', 6, '') |
+-----------------------+-----------------------+----------------------+----------------------+----------------------+----------------------+----------------------+
| NULL                  | NULL                  |                      | h                    | hel                  | hello                | NULL                 |
+-----------------------+-----------------------+----------------------+----------------------+----------------------+----------------------+----------------------+
1 row in set (0.00 sec)

mysql> select lpad('hello', -2, ''), lpad('hello', -1, '') , lpad('hello', 0, ''), lpad('hello', 1, ''), lpad('hello', 3, ''), lpad('hello', 5, ''), lpad('hello', 6, '')  from dual;
+-----------------------+-----------------------+----------------------+----------------------+----------------------+----------------------+----------------------+
| lpad('hello', -2, '') | lpad('hello', -1, '') | lpad('hello', 0, '') | lpad('hello', 1, '') | lpad('hello', 3, '') | lpad('hello', 5, '') | lpad('hello', 6, '') |
+-----------------------+-----------------------+----------------------+----------------------+----------------------+----------------------+----------------------+
| NULL                  | NULL                  |                      | h                    | hel                  | hello                | NULL                 |
+-----------------------+-----------------------+----------------------+----------------------+----------------------+----------------------+----------------------+
1 row in set (0.01 sec)

This PR fix this issue, after this PR:

spark-sql> select rpad('hello', -2, ''), rpad('hello', -1, '') , rpad('hello', 0, ''), rpad('hello', 1, ''), rpad('hello', 3, ''), rpad('hello', 5, ''), rpad('hello', 6, '');
NULL	NULL		h	hel	hello	NULL
spark-sql> select lpad('hello', -2, ''), lpad('hello', -1, '') , lpad('hello', 0, ''), lpad('hello', 1, ''), lpad('hello', 3, ''), lpad('hello', 5, ''), lpad('hello', 6, '');
NULL	NULL		h	hel	hello	NULL
spark-sql> 

How was this patch tested?

unit tests

[SparkQA]

Test build #77505 has finished for PR 18138 at commit 3ac9fb0.

• This patch fails Spark unit tests.
• This patch merges cleanly.
• This patch adds no public classes.

[SparkQA]

Test build #77512 has finished for PR 18138 at commit 895f414.

• This patch passes all tests.
• This patch merges cleanly.
• This patch adds no public classes.

[jiangxb1987]

Can you also examine the HIVE behavior under such cases?

[wangyum]

Hive will throw ArrayIndexOutOfBoundsException at runtime: https://issues.apache.org/jira/browse/HIVE-17077

[jiangxb1987]

Seems our current behavior doesn't follow each way :(