Stored Procedures implemented 2004-01-29: Summary of what's implemented: - SQL PROCEDUREs/FUNCTIONs (CREATE/DROP) - CALL - DECLARE of local variables - BEGIN/END, SET, CASE, IF, LOOP, WHILE, REPEAT, ITERATE, LEAVE - SELECT INTO local variables - "Non-query" FUNCTIONs only - Prepared SP caching - CONDITIONs and HANDLERs - Simple read-only CURSORs. - SHOW CREATE PROCEDURE/FUNCTION and SHOW PROCEDURE/FUNCTION STATUS Summary of Not Yet Implemented: - SQL statements using tables (like SELECT, INSERT, UPDATE etc) in FUNCTIONs - External languages - Access control - SQL-99 COMMIT (related to BEGIN/END) - FOR-loops - CASCADE/RESTRICT for ALTER and DROP - ALTER/DROP METHOD (as it implies User Defined Types) - SIGNAL and RESIGNAL, and UNDO handlers List of what's implemented: - CREATE PROCEDURE|FUNCTION name ( args ) characteristics body where characteristics is: LANGUAGE SQL | [NOT] DETERMINISTIC | SQL SECURITY [DEFINER|INVOKER] | COMMENT string However the DETERMINISTIC setting is not currently used. - ALTER PROCEDURE|FUNCTION name characteristics CASCADE/RESTRICT is not implemented. characteristics is: COMMENT string | SQL SECURITY [DEFINER|INVOKER] | NAME newname - DROP PROCEDURE|FUNCTION [IF EXISTS] name CASCADE/RESTRICT is not implemented. - CALL name (args) OUT and INOUT parameters are also works for user variables ("global" variables) - i.e., if a procedure is defined as: CREATE PROCEDURE foo(OUT p INT) ...; a call like: CALL foo(@x); will set @x to the output value. - Function/Procedure body: - BEGIN/END Is parsed, but not the real thing with (optional) transaction control, it only serves as block syntax for multiple statements (and local variable binding). Note: Multiple statements requires a client that can send bodies containing ";". This is handled in the CLI clients mysql and mysqltest with the "delimiter" command. Changing the end-of-query delimiter ";" to for instance "|" allows ";" to be used in the routine body. - SET of local variables Implemented as part of the pre-existing SET syntax. This allows an extended syntax of "SET a=x, b=y, ..." where different variable types (SP local and global) can be mixed. This also allows combinations of local variables and some options that only make sense for global/system variables; in that case the options are accepted but ignored. - The flow control constructs: CASE, IF, LOOP, WHILE, ITERATE and LEAVE are fully implemented. - SELECT ... INTO local variables (as well as global session variables) is implemented. (Note: This is not SQL-99 feature, but common in other databases.) - A FUNCTION can have flow control contructs, but must not contain an SQL query/statement, like SELECT, INSERT, UPDATE, etc. The reason is that it's hard to allow this is that a FUNCTION is executed as part of another query (unlike a PROCEDURE, which is called as a statement). The table locking scheme used makes it difficult to allow "subqueries" during FUNCTION invokation. - SPs are cached, but with a separate cache for each thread (THD). There are still quite a few non-reentrant constructs in the lexical context which makes sharing prepared SPs impossible. And, even when this is resolved, it's not necessarily the case that it will be faster than a cache per thread. A global cache requires locks, which might become a bottleneck. (It would save memory though.) - CONDITIONs and HANDLERs are implemented, but not the SIGNAL and RESIGNAL statements. (It's unclear if these can be implemented.) The semantics of CONDITIONs is expanded to allow catching MySQL error codes as well. UNDO handlers are not implemented (since we don't have SQL-99 style transaction control yet). - Simple read-only CURSORs are implemented, but not yet any of the optional arguments to DECLARE (SCROLL, SENSITIVE, etc) or FETCH (NEXT, PRIOR, etc). Cursors are ASENSITIVE, READ-ONLY, non-SCROLLing. (The additional syntax will be added for completeness, but for the most part unsupported with the current underlying cursor mechanism.) N.B. The current implementation is temporary and only works within a stored procedure, and may not perform well for very large result sets. A "real" cursor implementation is under development; this will replace the current one when it's finished. - SHOW procedures and functions SHOW CREATE PROCEDURE|FUNCTION returns the definition of a routine. SHOW PROCEDURE|FUNCTION STATUS [LIKE ] returns characteristics of routines, like the name, type, creator, creation and modification dates, etc.