Implement a collation for identifiers

In the code we use two things to compare identifiers:

• my_strcasecmp(system_charset_info, ...), or just system_charset_info->strcasecmp(..) when used directly without the macro. This comparison way is used when the two strings are known to be 0-terminated.
• system_charset_info->strnncoll(...). This way is used when we know the lengths of the operands and/or the strings are not necessarily 0-terminated.

There is a problem with this choice:

• strcasecmp() performs accent sensitive case insensitive comparison
• strnncoll() performs accent insensitive case insensitive comparison

This mismatch causes troubles (see examples below).

To avoid this mismatch, we need a dedicated collation for identifiers, whose strnncoll() compares in the accent sensitive case insensitive way, to have both strcasecmp() and strnncoll() compare in the same style.

A tentative name is utf8_general_as_ci, or even utf8_identifier_ci.
Under terms of this task, it may not even be necessary to expose this collation to the SQL level. But later, we should revise INFORMATION_SCHEMA to use this collation for certain columns.

The following scripts demonstrate the difference:

• Column names are accent sensitive (correct)

  CREATE OR REPLACE TABLE t1 (a INT, ä INT);

  SHOW CREATE TABLE t1;

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

  | Table | Create Table |

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

  | t1 | CREATE TABLE `t1` (

  `a` int(11) DEFAULT NULL,

  `ä` int(11) DEFAULT NULL

  ) ENGINE=InnoDB DEFAULT CHARSET=latin1 |

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

  1 row in set (0.002 sec)

  Notice, the two columns co-exist.

• Index names demonstrate both accent sensitive (correct) and accent insensitive (wrong) behavior

  CREATE OR REPLACE TABLE t1 (a INT, KEY a(a), KEY ä(a));

  SHOW WARNINGS;

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

  | Level | Code | Message |

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

  | Note | 1831 | Duplicate index `ä`. This is deprecated and will be disallowed in a future release |

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

  Notice, index names were compared in accent insensitive way to display the warning.
  However, on ALTER, they are compared in accent sensitive way as expected:

  ALTER TABLE t1 DROP INDEX ä;

  SHOW CREATE TABLE t1;

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

  | Table | Create Table |

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

  | t1 | CREATE TABLE `t1` (

  `a` int(11) DEFAULT NULL,

  KEY `a` (`a`)

  ) ENGINE=InnoDB DEFAULT CHARSET=latin1 |

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

  The second ALTER does not drop anything:

  ALTER TABLE t1 DROP INDEX ä;

  ERROR 1091 (42000): Can't DROP INDEX `ä`; check that it exists

• SP variables are accent insensitive (wrong)

  DELIMITER $$

  BEGIN NOT ATOMIC

  DECLARE a INT;

  DECLARE ä INT;

  END;

  $$

  DELIMITER ;

  ERROR 1331 (42000): Duplicate variable: ä

• ROW fields are accent sensitive (correct) on explicit declaration but accent insensitive (wrong) on anchored declaration

  DELIMITER $$

  BEGIN NOT ATOMIC

  DECLARE rec ROW(a INT, ä INT) DEFAULT (1,2);

  SELECT rec.a, rec.ä;

  END;

  $$

  DELIMITER ;

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

  | rec.a | rec.ä |

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

  | 1 | 2 |

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

  The above is correct.

  CREATE OR REPLACE TABLE t1 (a INT, ä INT);

  INSERT INTO t1 VALUES (1,2);

  DELIMITER $$

  BEGIN NOT ATOMIC

  DECLARE rec ROW TYPE OF t1 DEFAULT (1,2);

  SELECT rec.a, rec.ä;

  END;

  $$

  DELIMITER ;

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

  | rec.a | rec.ä |

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

  | 1 | 1 |

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

  The above is wrong. Notice, the expected value for rec.ä should be 2, like in the first example.