When a user name (or host name) is empty in the user specification for SET SESSION AUTHORIZATION command, ER_NO_SUCH_USER is produced regardless whether the user with the empty value actually exists.

According to set_authorization.test it is expected, but still, it looks strange, so I want to have it at least filed for those who come later wondering.

MariaDB [test]> create or replace user ''@localhost;

Query OK, 0 rows affected (0.013 sec)

MariaDB [test]> show grants for ''@localhost;

+--------------------------------------+

| Grants for @localhost                |

+--------------------------------------+

| GRANT USAGE ON *.* TO ``@`localhost` |

+--------------------------------------+

1 row in set (0.000 sec)

MariaDB [test]> set session authorization ''@localhost;

ERROR 1449 (HY000): The user ''@'localhost' does not exist

It would be easier to understand if it were some "not allowed" error, but getting ER_NO_SUCH_USER when the user very clearly exists is just weird.

WIth an empty host it's a bit different, because even though CREATE and GRANT accept the empty host name, they convert it into the wildcard '%', so it could be claimed that SET SESSION AUTHORIZATION doesn't make the adjustment and expects the value to be exact – still a weak argument, but more than one can say for the user name.

MariaDB [test]> create or replace user 'x'@'';

Query OK, 0 rows affected (0.013 sec)

MariaDB [test]> show grants for 'x'@'';

+-------------------------------+

| Grants for x@%                |

+-------------------------------+

| GRANT USAGE ON *.* TO `x`@`%` |

+-------------------------------+

1 row in set (0.000 sec)

MariaDB [test]> set session authorization 'x'@'';

ERROR 1449 (HY000): The user 'x'@'' does not exist

A strong argument could be that the SQL standard demands this behavior. I don't have access to the standard to check if it's so.