[SPARK-28349][SQL] Add FALSE and SETMINUS to ansiNonReserved

[wangyum]

What changes were proposed in this pull request?

How to reproduce this issue:

spark-sql> set spark.sql.parser.ansi.enabled=true;
spark.sql.parser.ansi.enabled	true
spark-sql> select 1 as false;
Error in query:
no viable alternative at input 'false'(line 1, pos 12)

== SQL ==
select 1 as false
------------^^^

spark-sql> select 1 as minus;
Error in query:
no viable alternative at input 'minus'(line 1, pos 12)

== SQL ==
select 1 as minus
------------^^^

Other databases behaviour:
PostgreSQL:

postgres=# select 1 as false, 1 as minus;
 false | minus
-------+-------
     1 |     1
(1 row)

Vertica:

dbadmin=> select 1 as false, 1 as minus;
 false | minus
-------+-------
     1 |     1
(1 row)

SQL Server:

1> select 1 as false, 1 as minus
2> go
false       minus
----------- -----------
          1           1

DB2:

Oracle:

SQL> select 1 as false, 1 as minus from dual;
select 1 as false, 1 as minus from dual
                        *
ERROR at line 1:
ORA-00923: FROM keyword not found where expected

Teradata:

This pr add FALSE and SETMINUS to ansiNonReserved to fix this issue.

How was this patch tested?

unit tests and manual tests:

spark-sql> select 1 as false, 1 as minus;
1	1
spark-sql>

[SparkQA]

Test build #107530 has finished for PR 25114 at commit 3efbda6.

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

[wangyum]

cc @maropu @dongjoon-hyun

[maropu]

Is this correct? I don't check carefully check yet, but I think the 'reserved' meaning is essentially different between spark and the other systems. For example, these keyworkds below are reserved in postgresql, but they can be used as alias names;

postgres=# select 1 as select;
 select 
--------
      1
(1 row)

postgres=# select 1 as from;
 from 
------
    1
(1 row)

[gatorsmile]

First, FALSE is reserved word.

Second, see https://www.postgresql.org/docs/7.3/sql-keywords-appendix.html

In the PostgreSQL parser life is a bit more complicated. There are several different classes of tokens ranging from those that can never be used as an identifier to those that have absolutely no special status in the parser as compared to an ordinary identifier. (The latter is usually the case for functions specified by SQL.) Even reserved key words are not completely reserved in PostgreSQL, but can be used as column labels (for example, SELECT 55 AS CHECK, even though CHECK is a reserved key word).

We do not need to follow PostgreSQL to support reserved words in column alias.

[wangyum]

Thank you @gatorsmile

[wangyum]

spark-sql> set spark.sql.parser.ansi.enabled=true;
spark.sql.parser.ansi.enabled	true
spark-sql> select extract(year from timestamp '2001-02-16 20:38:40')  ;
Error in query:
no viable alternative at input 'year'(line 1, pos 15)

== SQL ==
select extract(year from timestamp '2001-02-16 20:38:40')
---------------^^^

spark-sql> set spark.sql.parser.ansi.enabled=false;
spark.sql.parser.ansi.enabled	false
spark-sql> select extract(year from timestamp '2001-02-16 20:38:40')  ;
2001

Do we need to add YEAR, MONTH, DAY , HOUR, MINUTE and SECOND to ansiNonReserved? These are all non-reserved for PostgreSQL and Oracle.

https://github.com/antlr/grammars-v4/blob/b7f5b6d8c4da4d45d3dcb8146df310b89cd58114/plsql/PlSqlParser.g4#L4959-L6703

[maropu]

I think we don't need to do so cuz we basically follow SQL-2011 now.

[wangyum]

or add a regular_id? Some UDFs are also not available in ansi mode:

spark-sql> select left('12345', 2);
12
spark-sql> set spark.sql.parser.ansi.enabled=true;
spark.sql.parser.ansi.enabled	true
spark-sql> select left('12345', 2);
Error in query:
no viable alternative at input 'left'(line 1, pos 7)

== SQL ==
select left('12345', 2)
-------^^^

[maropu]

Rather, itd be better to list up these functions and rename them if necessary? IMO making these keywords non-reserved worsens parser error messages...