[MDEV-14415] Add Oracle-style FOR loop to sql_mode=DEFAULT Created: 2017-11-16  Updated: 2023-10-04  Resolved: 2017-11-16

Status: Closed
Project: MariaDB Server
Component/s: Parser
Fix Version/s: 10.3.3

Type: Task Priority: Major
Reporter: Alexander Barkov Assignee: Alexander Barkov
Resolution: Fixed Votes: 0
Labels: Compatibility

Issue Links:
Blocks
blocks MDEV-12518 Unify sql_yacc.yy and sql_yacc_ora.yy Closed
blocks MDEV-14564 support FOR loop in stored aggregate ... Stalled
Problem/Incident
causes MDEV-32275 getting error 'Illegal parameter data... Closed
Relates
relates to MDEV-10581 sql_mode=ORACLE: Explicit cursor FOR ... Closed
relates to MDEV-16674 Document FOR loop Open

 Description   

This MDEV is a part of MDEV-12518.
Under terms of this task we'll add Oracle-style FOR loop (both for integer ranges and cursors) into sql_mode=DEFAULT.
FOR loops will be a non-standard MariaDB extension.

We won't implement the SQL-standard cursor FOR loop yet, because it would need to open the cursor at parse time to know column names of the cursor.

Proposed syntax:

  • Integer range FOR loop:

    [begin_label:]
    FOR var_name IN [ REVERSE ] lower_bound .. upper_bound
    DO statement_list
    END FOR [ end_label ]
    

  • Explicit cursor FOR loop

    [begin_label:]
    FOR record_name IN cursor_name [ ( cursor_actual_parameter_list)]
    DO statement_list
    END FOR [ end_label ]
    

  • Implicit cursor FOR loop

    [begin_label:]
    FOR record_name IN ( select_statement )
    DO statement_list
    END FOR [ end_label ]
    

The proposed syntax is a compilation of the SQL standard syntax and Oracle's syntax for FOR loop:

  • It generally uses Oracle style syntax
  • however, like in the SQL standard, it uses FOR..DO..END FOR (instead or Oracle's FOR..LOOP..END LOOP)

Note, for cursor FOR loops, there is no a need to do OPEN, FETCH and CLOSE. These commands are done automatically. However, it will still be possible to fetch extra records inside the explicit cursor FOR loop body using explicit FETCH commands.

In sql_mode=DEFAULT, the automatic implicit FETCH which happens on FOR iterations and an explicit FETCH inside the loop body will work differently:

  • The automatic FETCH will never generate errors. On NOT FOUND, it will automatically close the cursor and leave the loop.
  • An explicit FETCH command inside the FOR loop will work like normal FETCH commands: i.e. it will generate an error on NOT FOUND, even when used with the FOR cursor.

This script uses an extra FETCH command inside the loop body:

DELIMITER $$
BEGIN NOT ATOMIC
  DECLARE x INT;
  DECLARE cur CURSOR FOR SELECT 1 AS x;
  FOR rec IN cur
  DO
    FETCH cur INTO x; -- Notice it uses "cur" (the loop cursor)
  END FOR;
END;
$$
DELIMITER ;

It will return:

ERROR 02000: No data - zero rows fetched, selected, or processed

because the explicit FETCH command returns no rows: the first record is fetched automatically in the beginning of the FOR iteration, and there are no more records.

FETCH commands inside the FOR body will be normally handled by "NOT FOUND" handlers.
The following script uses a FETCH command with the FOR loop cursor, in combination with a CONTINUE HANDLER FOR NOT FOUND.

DELIMITER $$
BEGIN NOT ATOMIC
  DECLARE done INT DEFAULT 0;
  DECLARE cur CURSOR FOR SELECT 1 AS x, 'y1' AS y UNION
                         SELECT 2,'y2' UNION
                         SELECT 3,'y3';
  DECLARE CONTINUE HANDLER FOR NOT FOUND SET done = TRUE;
  forrec:
  FOR rec IN cur
  DO
    SELECT CONCAT(rec.x, ' ', rec.y) AS 'Implicit FETCH';
    FETCH cur INTO rec;
    IF done THEN
      SELECT 'NO DATA' AS `Explicit FETCH`;
      LEAVE forrec;
    ELSE
      SELECT CONCAT(rec.x, ' ', rec.y) AS 'Explicit FETCH';
    END IF;
  END FOR;
END;
$$
DELIMITER ;

It will return without errors, with the following output:

Implicit FETCH
1 y1
Explicit FETCH
2 y2
Implicit FETCH
3 y3
Explicit FETCH
NO DATA

Notice:

  • some records are fetched using the automatic FETCH on FOR iterations
  • some records are fetched using the explicit FETCH command inside the loop body
  • the last explicit FETCH fails on NOT FOUND, which is caught by the CONTINUE handler, and the error is suppressed.


 Comments   
Comment by Robert Dyas [ 2018-07-03 ]

I can't find this in the MariaDB documentation... is it there and I just missed it?

Especially useful to note in the docs that 10.3.3+ has ANSI SQL cursor FOR loop with simple cursor example. So much easier than the exiting way for 99% of my uses cases.

Comment by Sergei Golubchik [ 2018-07-03 ]

MariaDB 10.3 does not support ANSI SQL cursor FOR. It's kind-of-like-in-Oracle cursor FOR. Functionality is pretty much the same, syntax is a bit different. Still, strictly speaking, not standard.

Comment by Alexander Barkov [ 2018-07-04 ]

The documentation is currently in progress:
MDEV-16674 Document FOR loop

In the meanwhile please have a look into this MDEV-14415, as well as these test files:

https://github.com/MariaDB/server/blob/10.3/mysql-test/main/sp-for-loop.test
https://github.com/MariaDB/server/blob/10.3/mysql-test/main/sp-cursor.test

Comment by Robert Dyas [ 2018-07-04 ]

In the documentation if you could clarify what happens with rec.col references if col is ambiguous from a join... is it simply rec.table.col as you would guess? Example:

CREATE TABLE t1 (id INT, name VARCHAR(10));
INSERT INTO t1 VALUES (1,'b1'), (2,'b2'), (3,'b3');
CREATE TABLE t2 (id INT, name VARCHAR(10));
INSERT INTO t1 VALUES (1,'bob'), (2,'sam'), (3,'frank');
 
BEGIN NOT ATOMIC
  DECLARE cur CURSOR FOR 
     SELECT * FROM t1 JOIN t2 ON (t2.id = t1.id);
 
  FOR rec IN cur DO
     SELECT rec.id, rec.name; -- both id and name are ambiguous ... what is correct syntax? rec.t2.name?
  END FOR;
 
END;

Comment by Alexander Barkov [ 2018-07-05 ]

I tested this script:

DROP TABLE IF EXISTS t1, t2;
CREATE TABLE t1 (id INT, name VARCHAR(10));
INSERT INTO t1 VALUES (1,'b1'), (2,'b2'), (3,'b3');
CREATE TABLE t2 (id INT, name VARCHAR(10));
INSERT INTO t2 VALUES (1,'bob'), (2,'sam'), (3,'frank');
 
DELIMITER $$
BEGIN NOT ATOMIC
  DECLARE cur CURSOR FOR 
     SELECT * FROM t1 JOIN t2 ON (t2.id = t1.id);
  FOR rec IN cur DO
     SELECT rec.id, rec.name; -- both id and name are ambiguous ... what is correct syntax? rec.t2.name?
  END FOR;
END;
$$
DELIMITER ;

and it correctly returns this error:

ERROR 1060 (42S21): Duplicate column name 'id'

The correct syntax is to give unique names to columns using AS, e.g. like this:

SELECT id, t1.name AS name1, t2.name AS name2 FROM t1 JOIN t2 USING (id);

Generated at Thu Feb 08 08:13:22 UTC 2024 using Jira 8.20.16#820016-sha1:9d11dbea5f4be3d4cc21f03a88dd11d8c8687422.