A cleanup for MDEV-9217 Split Item::tmp_table_field_from_field_type() into virtual methods in Type_handler

Fixing that the server tried to create the old decimal for:

  CREATE TABLE t2 AS SELECT old_decimal_field FROM t1
               UNION SELECT bigint_field FROM t1;

  CREATE TABLE t2 AS SELECT old_decimal_field FROM t1
               UNION SELECT mediumint_field FROM t1;

  CREATE TABLE t2 AS SELECT year FROM t1
               UNION SELECT old_decimal_field FROM t1;

  CREATE TABLE t2 AS SELCT COALESCE(old_decimal_field) FROM t1;

Solution:

1. field_types_merge_rules[][] had three MYSQL_TYPE_DECIMAL remainders.
   Fixing to MYSQL_TYPE_NEWDECIMAL, like it is done for all other type pairs.
   This fixes the above queries with UNION.

2. Convert MYSQL_TYPE_DECIMAL to MYSQL_TYPE_NEWDECIMAL in
   Item::tmp_table_field_from_field_type().
   This fixed the above query with COALESCE.

3. Adding "new Field_decimal" into Type_handler_olddecimal::make_table_field().
   In case if something goes wrong it will crash on assert only in debug builds,
   while create the old decimal in release.
   Note, this "new Field_decimal" will be needed later anyway,
   when we reuse Type_handler::make_table_field() in make_field() in field.cc.

mysql-test/r/type_decimal.result

@@ -1019,3 +1019,52 @@ cast('-0.0' as decimal(5,1)) < 0
 #
 # End of 5.5 tests
 #
+#
+# Start of 10.3 tests
+#
+#
+# MDEV-9217 Split Item::tmp_table_field_from_field_type() into virtual methods in Type_handler
+#
+# This creates the old DECIMAL. Will be fixed in MDEV-12574.
+CREATE TABLE t1 AS SELECT MAX(a) FROM t1dec102;
+SHOW CREATE TABLE t1;
+Table	Create Table
+t1	CREATE TABLE `t1` (
+  `MAX(a)` decimal(10,2)/*old*/ DEFAULT NULL
+) ENGINE=MyISAM DEFAULT CHARSET=latin1
+DROP TABLE t1;
+CREATE TABLE t1 AS SELECT COALESCE(a) FROM t1dec102;
+SHOW CREATE TABLE t1;
+Table	Create Table
+t1	CREATE TABLE `t1` (
+  `COALESCE(a)` decimal(12,2) DEFAULT NULL
+) ENGINE=MyISAM DEFAULT CHARSET=latin1
+DROP TABLE t1;
+CREATE TABLE t1 (a BIGINT);
+CREATE TABLE t2 AS SELECT a FROM t1dec102 UNION SELECT a FROM t1;
+SHOW CREATE TABLE t2;
+Table	Create Table
+t2	CREATE TABLE `t2` (
+  `a` decimal(21,2) DEFAULT NULL
+) ENGINE=MyISAM DEFAULT CHARSET=latin1
+DROP TABLE t2;
+DROP TABLE t1;
+CREATE TABLE t1 (a MEDIUMINT);
+CREATE TABLE t2 AS SELECT a FROM t1dec102 UNION SELECT a FROM t1;
+SHOW CREATE TABLE t2;
+Table	Create Table
+t2	CREATE TABLE `t2` (
+  `a` decimal(12,2) DEFAULT NULL
+) ENGINE=MyISAM DEFAULT CHARSET=latin1
+DROP TABLE t2;
+DROP TABLE t1;
+CREATE TABLE t1 (a YEAR);
+CREATE TABLE t2 AS SELECT a FROM t1 UNION SELECT a FROM t1dec102;
+SHOW CREATE TABLE t2;
+Table	Create Table
+t2	CREATE TABLE `t2` (
+  `a` decimal(12,2) DEFAULT NULL
+) ENGINE=MyISAM DEFAULT CHARSET=latin1
+DROP TABLE t2;
+DROP TABLE t1;
+DROP TABLE t1dec102;

mysql-test/t/type_decimal.test

@@ -612,3 +612,46 @@ select cast('-0.0' as decimal(5,1)) < 0;
 --echo #
 --echo # End of 5.5 tests
 --echo #
+
+--echo #
+--echo # Start of 10.3 tests
+--echo #
+
+--echo #
+--echo # MDEV-9217 Split Item::tmp_table_field_from_field_type() into virtual methods in Type_handler
+--echo #
+
+let $MYSQLD_DATADIR= `select @@datadir`;
+
+--copy_file std_data/old_decimal/t1dec102.frm $MYSQLD_DATADIR/test/t1dec102.frm
+--copy_file std_data/old_decimal/t1dec102.MYD $MYSQLD_DATADIR/test/t1dec102.MYD
+--copy_file std_data/old_decimal/t1dec102.MYI $MYSQLD_DATADIR/test/t1dec102.MYI
+
+--echo # This creates the old DECIMAL. Will be fixed in MDEV-12574.
+CREATE TABLE t1 AS SELECT MAX(a) FROM t1dec102;
+SHOW CREATE TABLE t1;
+DROP TABLE t1;
+
+CREATE TABLE t1 AS SELECT COALESCE(a) FROM t1dec102;
+SHOW CREATE TABLE t1;
+DROP TABLE t1;
+
+CREATE TABLE t1 (a BIGINT);
+CREATE TABLE t2 AS SELECT a FROM t1dec102 UNION SELECT a FROM t1;
+SHOW CREATE TABLE t2;
+DROP TABLE t2;
+DROP TABLE t1;
+
+CREATE TABLE t1 (a MEDIUMINT);
+CREATE TABLE t2 AS SELECT a FROM t1dec102 UNION SELECT a FROM t1;
+SHOW CREATE TABLE t2;
+DROP TABLE t2;
+DROP TABLE t1;
+
+CREATE TABLE t1 (a YEAR);
+CREATE TABLE t2 AS SELECT a FROM t1 UNION SELECT a FROM t1dec102;
+SHOW CREATE TABLE t2;
+DROP TABLE t2;
+DROP TABLE t1;
+
+DROP TABLE t1dec102;

sql/field.cc

@@ -125,7 +125,7 @@ static enum_field_types field_types_merge_rules [FIELDTYPE_NUM][FIELDTYPE_NUM]=
   //MYSQL_TYPE_NULL         MYSQL_TYPE_TIMESTAMP
     MYSQL_TYPE_NEWDECIMAL,  MYSQL_TYPE_VARCHAR,
   //MYSQL_TYPE_LONGLONG     MYSQL_TYPE_INT24
-    MYSQL_TYPE_DECIMAL,     MYSQL_TYPE_DECIMAL,
+    MYSQL_TYPE_NEWDECIMAL,  MYSQL_TYPE_NEWDECIMAL,
   //MYSQL_TYPE_DATE         MYSQL_TYPE_TIME
     MYSQL_TYPE_VARCHAR,     MYSQL_TYPE_VARCHAR,
   //MYSQL_TYPE_DATETIME     MYSQL_TYPE_YEAR
@@ -520,7 +520,7 @@ static enum_field_types field_types_merge_rules [FIELDTYPE_NUM][FIELDTYPE_NUM]=
   /* MYSQL_TYPE_YEAR -> */
   {
   //MYSQL_TYPE_DECIMAL      MYSQL_TYPE_TINY
-    MYSQL_TYPE_DECIMAL,     MYSQL_TYPE_TINY,
+    MYSQL_TYPE_NEWDECIMAL,  MYSQL_TYPE_TINY,
   //MYSQL_TYPE_SHORT        MYSQL_TYPE_LONG
     MYSQL_TYPE_SHORT,       MYSQL_TYPE_LONG,
   //MYSQL_TYPE_FLOAT        MYSQL_TYPE_DOUBLE

sql/item.cc

@@ -6284,6 +6284,9 @@ Field *Item::tmp_table_field_from_field_type(TABLE *table)
   Record_addr addr(maybe_null);
 
   switch (handler->field_type()) {
+  case MYSQL_TYPE_DECIMAL:
+    handler= &type_handler_newdecimal;
+    break;
   case MYSQL_TYPE_NULL:
   case MYSQL_TYPE_STRING:
   case MYSQL_TYPE_ENUM:

sql/sql_type.cc

@@ -1281,11 +1281,15 @@ Type_handler_olddecimal::make_table_field(const LEX_CSTRING *name,
   /*
     Currently make_table_field() is used for Item purpose only.
     On Item level we have type_handler_newdecimal only.
-    Will be implemented when we reuse Type_handler::make_table_field()
+    For now we have DBUG_ASSERT(0).
+    It will be removed when we reuse Type_handler::make_table_field()
     in make_field() in field.cc, to open old tables with old decimal.
   */
   DBUG_ASSERT(0);
-  return NULL;
+  return new (table->in_use->mem_root)
+         Field_decimal(addr.ptr, attr.max_length, addr.null_ptr, addr.null_bit,
+                       Field::NONE, name, attr.decimals,
+                       0/*zerofill*/,attr.unsigned_flag);
 }