Month: May 2023

Delimiters in MySQL and MariaDB

In the SQL standard delimiters are for separating tokens; in MySQL/MariaDB they’re for ending statements, particularly compound statements as in

  VALUES (5);

The problem is that a dumb client could stop at the first semicolon “;” in the statement (the one after VALUES (5)) instead of the one that truly ends (the one after END). The solution is to say that semicolon is no longer the default delimiter, and the way to specify a replacement is with a non-standard DELIMITER statement, such as


and after that the true statement end is //.

The idea isn’t unusual — in PostgreSQL’s psql client you can enclose compound statements with $$ which they call dollar quoting, with Microsoft you get used to saying “go” when you want something shipped to the server, with Oracle you might sometimes see SQLTERMINATOR. But it shouldn’t always be necessary — despite the fact that words like BEGIN are not reserved, a client with even a slight ability to recognize syntax can see where a compound statement ends without needing special tags.


Some DELIMITER quirks are documented in either the MySQL manual or MariaDB manual, some are common to all mysql-client commands, but all are weird. They are:
() DELIMITER itself is not delimited
() After DELIMITER A B the delimiter is A
() After DELIMITER A; the delimiter is A;
() After DELIMITER 12345678901234567890 the delimiter is 123456789012345
() After DELIMITER ‘x’ or DELIMITER “x” or DELIMITER `x` the delimiter is x
() After DELIMITER ;; the delimiter is ;;
() Delimiters inside comments or character string literals or delimited identifiers are ignored
() After \dA the delimiter is A (\d is a substitute for DELIMITER and no space is necessary)
() After DELIMITER \ the effect can be odd, the MySQL manual recommends not to use \
() SELECT 5; DELIMITER ; will cause an error message
() DELIMITER ;; followed by SELECT 5;;; has a different effect from DELIMITER // followed by SELECT 5;//
() \G and \g will work whether or not there is a delimiter

It is not a gotcha that, when statements are in a .sql file and read by the SOURCE file-name statement, the rules are different: Compound statements will be recognized without special delimiters, and semicolons will always end non-compound statements even after a different delimiter has been specified.

It is not a quirk or a gotcha, it’s merely bad design, that DELIMITER forces the fundamental requirement of a parser — the tokenizer — to change what the tokenizing rules are while the tokenizing is going on.

Best practice

Both the MySQL and the MariaDB manual have example statements where the delimiter is // and that’s unlikely to cause confusion, since the only other uses of / are as a beginning of a bracketed comment or as a division operator, or in arguments to PROMPT or SOURCE or TEE.
Less frequently there is a recommendation for $$ but that is bad because $$ can occur inside regular identifiers.
DELIMITER !! is equally bad because !, unlike /, is a repeatable operator, for example SELECT !!5; returns 1.
DELIMITER ;; is worse and has caused bugs in the past, but the mysqldump utility will generate it before CREATE TRIGGER statements.

Avoid non-printable or non-ASCII characters, some clients will support them but others won’t.

You could set a delimiter at the start of a session and let it persist. But I believe the more common practice is to put DELIMITER [something appropriate] immediately before any CREATE PROCEDURE or CREATE TRIGGER or CREATE FUNCTION or BEGIN NOT ATOMIC statement, and DELIMITER ; immediately after. This might help readability by humans since otherwise they’d look at statements that end with mere semicolons and assume they’d go to the server immediately, not realizing that long ago there was a DELIMITER statement.

See also the Descriptive SQL style guide section “Semicolons”.


The ocelotgui client doesn’t need DELIMITER statements because obviously a client that includes a stored-procedure debugger can recognize compound statements, but it will accept them because it tries to be compatible with the mysql client. It does not duplicate mysql client behaviour for the SOURCE statement but that should be fixed in the next version, and the fix is now in the source code which can be downloaded from github.