• malff/marcsql@weblab.(none)'s avatar
    Bug#23703 (DROP TRIGGER needs an IF EXISTS) · 429b0e6c
    malff/marcsql@weblab.(none) authored
    This change set implements the DROP TRIGGER IF EXISTS functionality.
    
    This fix is considered a bug and not a feature, because without it,
    there is no known method to write a database creation script that can create
    a trigger without failing, when executed on a database that may or may not
    contain already a trigger of the same name.
    
    Implementing this functionality closes an orthogonality gap between triggers
    and stored procedures / stored functions (which do support the DROP IF
    EXISTS syntax).
    
    In sql_trigger.cc, in mysql_create_or_drop_trigger,
    the code has been reordered to:
    - perform the tests that do not depend on the file system (access()),
    - get the locks (wait_if_global_read_lock, LOCK_open)
    - call access()
    - perform the operation
    - write to the binlog
    - unlock (LOCK_open, start_waiting_global_read_lock)
    
    This is to ensure that all the code that depends on the presence of the
    trigger file is executed in the same critical section,
    and prevents race conditions similar to the case fixed by Bug 14262 :
    
    - thread 1 executes DROP TRIGGER IF EXISTS, access() returns a failure
    - thread 2 executes CREATE TRIGGER
    - thread 2 logs CREATE TRIGGER
    - thread 1 logs DROP TRIGGER IF EXISTS
    
    The patch itself is based on code contributed by the MySQL community,
    under the terms of the Contributor License Agreement (See Bug 18161).
    429b0e6c
trigger.test 47.6 KB