feat(optimizer)!: annotate regexpreplace for mysql

[PiyaDaswadkar]

This PR annotate the function RegexpReplace for mysql

https://dev.mysql.com/doc/refman/9.7/en/regexp.html#function_regexp-replace

[treysp]

It looks like REGEXP_REPLACE has signatures for binary inputs that all return LONGBLOB, so we'll need to use an annotator to conditionally set the return type based on the input arg types.

All three of this, expression, and replacement must have the same general type family (either text or binary).

There are 4 possible text and 6 possible binary input types:

mysql> CREATE TABLE regex_test_inputs (
    ->   c_text TEXT, 
    ->   c_tinytext TINYTEXT,
    ->   c_mediumtext MEDIUMTEXT,
    ->   c_longtext LONGTEXT,
    ->   c_binary BINARY(3), 
    ->   c_varbinary VARBINARY(3),
    ->   c_blob BLOB,
    ->   c_tinyblob TINYBLOB,
    ->   c_mediumblob MEDIUMBLOB,
    ->   c_longblob LONGBLOB 
    ->  );
Query OK, 0 rows affected (0.010 sec)

mysql>  
    ->  INSERT INTO regex_test_inputs VALUES (
    ->   'abc',
    ->   'abc',
    ->   'abc',
    ->   'abc',
    ->   X'616263', -- abc 
    ->   X'616263',
    ->   X'616263',
    ->   X'616263',
    ->   X'616263',
    ->   X'616263' 
    ->  );
Query OK, 1 row affected (0.001 sec)

mysql>  
    -> CREATE TABLE regex_test_results AS SELECT
    ->   REGEXP_REPLACE(c_text, 'a', 'z') AS from_text,
    ->   REGEXP_REPLACE(c_tinytext, 'a', 'z') AS from_tinytext,
    ->   REGEXP_REPLACE(c_mediumtext, 'a', 'z') AS from_mediumtext,
    ->   REGEXP_REPLACE(c_longtext, 'a', 'z') AS from_longtext,
    ->   REGEXP_REPLACE(c_binary, X'62', X'78') AS from_binary,
    ->   REGEXP_REPLACE(c_varbinary,X'62', X'78') AS from_varbinary, 
    ->   REGEXP_REPLACE(c_tinyblob, X'62', X'78') AS from_tinyblob,
    ->   REGEXP_REPLACE(c_blob, X'62', X'78') AS from_blob,
    ->   REGEXP_REPLACE(c_mediumblob, X'62', X'78') AS from_mediumblob,
    ->   REGEXP_REPLACE(c_longblob, X'62', X'78') AS from_longblob 
    -> FROM regex_test_inputs;
Query OK, 1 row affected (0.005 sec)
Records: 1  Duplicates: 0  Warnings: 0

mysql> DESCRIBE regex_test_results;
+-----------------+----------+------+-----+---------+-------+
| Field           | Type     | Null | Key | Default | Extra |
+-----------------+----------+------+-----+---------+-------+
| from_text       | longtext | YES  |     | NULL    |       |
| from_tinytext   | longtext | YES  |     | NULL    |       |
| from_mediumtext | longtext | YES  |     | NULL    |       |
| from_longtext   | longtext | YES  |     | NULL    |       |
| from_binary     | longblob | YES  |     | NULL    |       |
| from_varbinary  | longblob | YES  |     | NULL    |       |
| from_tinyblob   | longblob | YES  |     | NULL    |       |
| from_blob       | longblob | YES  |     | NULL    |       |
| from_mediumblob | longblob | YES  |     | NULL    |       |
| from_longblob   | longblob | YES  |     | NULL    |       |
+-----------------+----------+------+-----+---------+-------+
10 rows in set (0.003 sec)

[PiyaDaswadkar]

Thanks for the suggestion. I’ve added the binary-family fixtures for REGEXP_REPLACE, covering the same optional argument combinations as the text-family tests. The tests now cover both LONGTEXT and LONGBLOB return types based on the argument family.