Details
-
Bug
-
Status: Closed (View Workflow)
-
Major
-
Resolution: Fixed
-
10.2(EOL)
Description
The window_spec clause in sql_yacc.yy has a shift/reduce conflict.
In a query like this:
SELECT .. COUNT(*) OVER (rows ...); |
It's not clear what rows stands for. It can be ident in opt_window_ref:
SELECT ... COUNT(*) OVER (rows ORDER BY ...); |
or the ROWS keyword in window_frame_units:
SELECT ... COUNT(*) OVER (ROWS BETWEEN 1 PRECEDING AND 1 FOLLOWING); |
The conflict happens because ROWS is a non-reserved keyword in MariaDB.
Possible solutions:
- Make ROWS a reserved keyword, as in the SQL standard
- Or change the ident part in opt_window_ref to a new rule ident_window_ref which will not include ROWS.
Attachments
Issue Links
- relates to
-
-
- Closed
-
Activity
Igor,
I don't know why ROWS is not reserved. It has been non-reserved in MySQL/MariaDB for years. Perhaps it did not cause any conflicts earlier.
PRECEDING must be a non-reserved keyword. See MDEV-9979, which is now in review.
ROWS is a reserved keyword in the SQL standard. I don't mind to make it reserved in MariaDB. But usually we try not to introduce new reserved keywords, unless it's absolutely necessary.
This conflict causes the problems of MDEV-9976.
So I'd rather get rid of it.
Alexander,
Could you, please, tell me why ROWS is considered as a non-reserved keyword in our parser (as opposed to PRECEDING)?
To my mind window_ref can be ident, and ident can be keyword. So we still have a conflict