Enable support of access to tables from triggers. Thus fix bug #8406 "Triggers

crash if referencing a table" and several other related bugs.
Fix for bug #11834 "Re-execution of prepared statement with dropped function
crashes server." which was spotted during work on previous bugs.

Also couple of nice cleanups:
- Replaced two separate hashes for stored routines used by statement with one.
- Now instead of doing one pass through all routines used in statement for
  caching them and then doing another pass for adding their tables to table
  list, we do only one pass during which do both things.


mysql-test/r/sp-error.result:
  Added test for bug #11834 "Re-execution of prepared statement with dropped
  function crashes server" also covering handling of prepared statements
  which use stored functions but does not require prelocking.
mysql-test/r/sp.result:
  Updated test for LOCK TABLES with views in table list.
  (Old version of statement used in this test will work ok now, since prelocking
   algorithm was tuned and will lock only one multi-set of tables for each routine
   even if this routine is used in several different views).
mysql-test/r/trigger.result:
  Added several tests for triggers using tables.
mysql-test/t/sp-error.test:
  Added test for bug #11834 "Re-execution of prepared statement with dropped
  function crashes server" also covering handling of prepared statements
  which use stored functions but does not require prelocking.
mysql-test/t/sp.test:
  Updated comment about recursive views to reflect current situation.
  Updated test for LOCK TABLES with views in table list.
  (Old version of statement used in this test will work ok now, since prelocking
   algorithm was tuned and will lock only one multi-set of tables for each routine
   even if this routine is used in several different views).
mysql-test/t/trigger.test:
  Added several tests for triggers using tables.
sql/item_func.cc:
  Item_func_sp::cleanup():
    By next statement execution stored function can be dropped or altered so
    we can't assume that sp_head object for it will be still valid.
sql/sp.cc:
  - Added Sroutine_hash_entry structure that represents element in the set of
    stored routines used by statement or routine. We can't as before use
    LEX_STRING for this purprose because we want link all elements of this set
    in list.
  - Replaced sp_add_to_hash() with sp_add_used_routine() which takes into account
    that now we use one hash for stored routines used by statement instead of two
    and which mantains list linking all elelemnts in this hash.
  - Renamed sp_merge_hash() to sp_update_sp_used_routines().
  - Introduced sp_update_stmt_used_routines() for adding elements to the set of
    routines used by statement from another similar set for statement or routine.
    This function will also mantain list linking elements of destination set.
  - Now instead of one sp_cache_routines() function we have family of 
    sp_cache_routines_and_add_tables() functions which are also responsible for
    adding tables used by routines being cached to statement table list. Nice
    optimization - thanks to list linking all elements in the hash of routines
    used by statement we don't need to perform several iterations over this hash
    (as it was before in cases when we have added new elements to it).
sql/sp.h:
  Added declarations of functions used for manipulations with set (hash) of stored
  routines used by statement.
sql/sp_head.cc:
  sp_name::init_qname():
    Now sp_name also holds key identifying routine in the set (hash) of
    stored routines used by statement. 
  sp_head:
    Instead of two separate hashes sp_funs/m_spprocs representing sets of stored
    routines used by this routine we use one hash - m_sroutines. 
  sp_instr_set_trigger_field:
    Added support for subqueries in assignments to row accessors in triggers.
  Removed definition of sp_add_sp_tables_to_table_list() and auxilary functions 
  since now we don't have separate stage on which we add tables used by routines
  used by statement to table list for prelocking. We do it on the same stage as
  we load those routines in SP cache. So all this functionality moved to
  sp_cache_routines_and_add_tables() family of functions.
sql/sp_head.h:
  sp_name:
    Now this class also holds key identifying routine in the set (hash) of stored
    routines used by statement. 
  sp_head:
    Instead of two separate hashes sp_funs/m_spprocs representing sets of stored
    routines used by this routine we use one hash - m_sroutines. 
  sp_instr_set_trigger_field:
    Added support for subqueries in assignments to row accessors in triggers.
  Removed declaration of sp_add_sp_tables_to_table_list() since now we don't have
  separate stage on which we add tables used by routines used by statement to
  table list for prelocking. We do it on the same stage as we load those routines
  in SP cache.
sql/sql_base.cc:
  open_tables():
  - LEX::spfuns/spprocs hashes were replaced with one LEX::sroutines hash.
  - Now instead of doing one pass through all routines used in statement for
    caching them and then doing another pass for adding their tables to table
    list, we do only one pass during which do both things. It is easy to do
    since all routines in the set of routines used by statement are linked in
    the list. This also allows us to calculate table list for prelocking more
    precisely.
  - Now triggers properly inform prelocking algorithm about tables they use.
sql/sql_lex.cc:
  lex_start():
    Replaced LEX::spfuns/spprocs with with one LEX::sroutines hash.
    Added LEX::sroutines_list list linking all elements in this hash.
  st_lex::st_lex():
    Moved definition of LEX constructor to sql_lex.cc file to be able
    use sp_sroutine_key declaration from sp.h in it.
sql/sql_lex.h:
  LEX:
    Replaced two separate hashes for stored routines used by statement with one.
    Added list linking all elements in this hash to be able to iterate through all
    elements and add new elements to this hash at the same time.
    Moved constructor definition to sql_lex.cc.
sql/sql_parse.cc:
  mysql_execute_command():
    Replaced LEX::spfuns/spprocs with one LEX::sroutines hash.
sql/sql_trigger.cc:
  Added missing GNU GPL notice.
  Table_triggers_list::check_n_load()
    Added initialization of sroutines_key which stores key representing
    triggers of this table in the set (hash) of routines used by this statement.
sql/sql_trigger.h:
  Added missing GNU GPL notice.
  Table_triggers_list:
    Added sroutines_key member to store key representing triggers of this
    table in the set (hash) of routines used by this statement.
    Declared sp_cache_routines_and_add_tables_for_triggers() as friend since
    it needs access to sroutines_key and trigger bodies.
sql/sql_yacc.yy:
  - Now we use sp_add_used_routine() instead of sp_add_to_hash() for adding
    elements to the set of stored routines used in statement.
  - Enabled support of subqueries as right sides in assignments to triggers' row
    accessors.

1 files changed, 82 insertions, 1 deletions

diff --git a/mysql-test/t/trigger.test b/mysql-test/t/trigger.test
index 0c5ef077159..75b2f6de57a 100644
--- a/mysql-test/t/trigger.test
+++ b/mysql-test/t/trigger.test
@@ -3,9 +3,10 @@
 #
 
 --disable_warnings
-drop table if exists t1, t2;
+drop table if exists t1, t2, t3;
 drop view if exists v1;
 drop database if exists mysqltest;
+drop function if exists f1;
 --enable_warnings
 
 create table t1 (i int);
@@ -200,6 +201,86 @@ drop table t1;
 
 
 #
+# Let us test triggers which access other tables.
+#
+# Trivial trigger which inserts data into another table
+create table t1 (id int primary key, data varchar(10), fk int);
+create table t2 (event varchar(100));
+create table t3 (id int primary key);
+create trigger t1_ai after insert on t1 for each row 
+  insert into t2 values (concat("INSERT INTO t1 id=", new.id, " data='", new.data, "'"));
+insert into t1 (id, data) values (1, "one"), (2, "two");
+select * from t1;
+select * from t2;
+drop trigger t1.t1_ai;
+# Trigger which uses couple of tables (and partially emulates FK constraint)
+delimiter |;
+create trigger t1_bi before insert on t1 for each row
+begin
+  if exists (select id from t3 where id=new.fk) then
+    insert into t2 values (concat("INSERT INTO t1 id=", new.id, " data='", new.data, "' fk=", new.fk));
+  else
+    insert into t2 values (concat("INSERT INTO t1 FAILED id=", new.id, " data='", new.data, "' fk=", new.fk));
+    set new.id= NULL;
+  end if;
+end|
+delimiter ;|
+insert into t3 values (1);
+--error 1048
+insert into t1 values (4, "four", 1), (5, "five", 2);
+select * from t1;
+select * from t2;
+drop table t1, t2, t3;
+# Trigger which invokes function
+create table t1 (id int primary key, data varchar(10));
+create table t2 (seq int);
+insert into t2 values (10);
+create function f1 () returns int return (select max(seq) from t2);
+delimiter |;
+create trigger t1_bi before insert on t1 for each row
+begin
+  if new.id > f1() then
+    set new.id:= f1();
+  end if;
+end|
+delimiter ;|
+# Remove this once bug #11554 will be fixed.
+select f1();
+insert into t1 values (1, "first");
+insert into t1 values (f1(), "max");
+select * from t1;
+drop table t1, t2;
+drop function f1;
+# Trigger which forces invocation of another trigger
+# (emulation of FK on delete cascade policy)
+create table t1 (id int primary key, fk_t2 int);
+create table t2 (id int primary key, fk_t3 int);
+create table t3 (id int primary key);
+insert into t1 values (1,1), (2,1), (3,2);
+insert into t2 values (1,1), (2,2);
+insert into t3 values (1), (2);
+create trigger t3_ad after delete on t3 for each row
+  delete from t2 where fk_t3=old.id;
+create trigger t2_ad after delete on t2 for each row
+  delete from t1 where fk_t2=old.id;
+delete from t3 where id = 1;
+select * from t1 left join (t2 left join t3 on t2.fk_t3 = t3.id) on t1.fk_t2 = t2.id;
+drop table t1, t2, t3;
+# Trigger which assigns value selected from table to field of row
+# being inserted/updated.
+create table t1 (id int primary key, copy int);
+create table t2 (id int primary key, data int);
+insert into t2 values (1,1), (2,2);
+create trigger t1_bi before insert on t1 for each row
+  set new.copy= (select data from t2 where id = new.id);
+create trigger t1_bu before update on t1 for each row
+  set new.copy= (select data from t2 where id = new.id);
+insert into t1 values (1,3), (2,4), (3,3);
+update t1 set copy= 1 where id = 2;
+select * from t1;
+drop table t1, t2;
+
+#
 # Test of wrong column specifiers in triggers
 #
 create table t1 (i int);