diff options
-rw-r--r-- | innobase/data/data0type.c | 3 | ||||
-rw-r--r-- | innobase/include/data0type.h | 38 | ||||
-rw-r--r-- | innobase/include/data0type.ic | 19 | ||||
-rw-r--r-- | innobase/rem/rem0cmp.c | 16 | ||||
-rw-r--r-- | sql/ha_innodb.cc | 25 |
5 files changed, 54 insertions, 47 deletions
diff --git a/innobase/data/data0type.c b/innobase/data/data0type.c index 39c80b773a2..a1cd34acd23 100644 --- a/innobase/data/data0type.c +++ b/innobase/data/data0type.c @@ -18,8 +18,9 @@ column definitions, or records in the insert buffer, we use this charset-collation code for them. */ ulint data_mysql_default_charset_coll = 99999999; +ulint data_mysql_latin1_swedish_charset_coll = 99999999; -dtype_t dtype_binary_val = {DATA_BINARY, 0, 0, 0, 0}; +dtype_t dtype_binary_val = {DATA_BINARY, 0, 0, 0}; dtype_t* dtype_binary = &dtype_binary_val; /************************************************************************* diff --git a/innobase/include/data0type.h b/innobase/include/data0type.h index b01a1be95f2..00a0845ace8 100644 --- a/innobase/include/data0type.h +++ b/innobase/include/data0type.h @@ -12,6 +12,7 @@ Created 1/16/1996 Heikki Tuuri #include "univ.i" extern ulint data_mysql_default_charset_coll; +extern ulint data_mysql_latin1_swedish_charset_coll; /* SQL data type struct */ typedef struct dtype_struct dtype_t; @@ -30,8 +31,8 @@ extern dtype_t* dtype_binary; #define DATA_BINARY 4 /* binary string */ #define DATA_BLOB 5 /* binary large object, or a TEXT type; if prtype & DATA_BINARY_TYPE == 0, then this is - actually a TEXT column; see also below about - the flag DATA_NONLATIN1 */ + actually a TEXT column (or a BLOB created + with < 4.0.14) */ #define DATA_INT 6 /* integer: can be any size 1 - 8 bytes */ #define DATA_SYS_CHILD 7 /* address of the child page in node pointer */ #define DATA_SYS 8 /* system column */ @@ -59,7 +60,7 @@ Tables created by a MySQL user have the following convention: code (not applicable for system columns). - In the second least significant byte we OR flags DATA_NOT_NULL, -DATA_UNSIGNED, DATA_BINARY_TYPE, DATA_NONLATIN1. +DATA_UNSIGNED, DATA_BINARY_TYPE. - In the third least significant byte of the precise type of string types we store the MySQL charset-collation code. In DATA_BLOB columns created with @@ -73,6 +74,23 @@ installation. If the stored charset code is 0 in the system table SYS_COLUMNS of InnoDB, that means that the default charset of this MySQL installation should be used. +When loading a table definition from the system tables to the InnoDB data +dictionary cache in main memory, InnoDB versions >= 4.1.2 and >= 5.0.1 check +if the stored charset-collation is 0, and if that is the case and the type is +a non-binary string, replace that 0 by the default charset-collation code of +this MySQL installation. In short, in old tables, the charset-collation code +in the system tables on disk can be 0, but in in-memory data structures +(dtype_t), the charset-collation code is always != 0 for non-binary string +types. + +In new tables, in binary string types, the charset-collation code is the +MySQL code for the 'binary charset', that is, != 0. + +For binary string types and for DATA_CHAR, DATA_VARCHAR, and for those +DATA_BLOB which are binary or have the charset-collation latin1_swedish_ci, +InnoDB performs all comparisons internally, without resorting to the MySQL +comparison functions. This is to save CPU time. + InnoDB's own internal system tables have different precise types for their columns, and for them the precise type is usually not used at all. */ @@ -112,14 +130,10 @@ be less than 256 */ string, this is ORed to the precise type: this only holds for tables created with >= MySQL-4.0.14 */ -#define DATA_NONLATIN1 2048 /* If the data type is DATA_BLOB with - the prtype & DATA_BINARY_TYPE == 0, that is, - TEXT, then in versions 4.0.14 - 4.0.xx this - flag is set to 1, if the charset is not - latin1. In version 4.1.1 this was set - to 1 for all TEXT columns. In versions >= 4.1.2 - this is set to 1 if the charset-collation of a - TEXT column is not latin1_swedish_ci. */ +/* #define DATA_NONLATIN1 2048 This is a relic from < 4.1.2 and < 5.0.1. + In earlier versions this was set for some + BLOB columns. +*/ /*-------------------------------------------*/ /* This many bytes we need to store the type information affecting the @@ -210,7 +224,7 @@ ulint dtype_form_prtype( /*==============*/ ulint old_prtype, /* in: the MySQL type code and the flags - DATA_NONLATIN1 etc. */ + DATA_BINARY_TYPE etc. */ ulint charset_coll); /* in: MySQL charset-collation code */ /************************************************************************* Gets the type length. */ diff --git a/innobase/include/data0type.ic b/innobase/include/data0type.ic index bba9fc07eca..41127a65183 100644 --- a/innobase/include/data0type.ic +++ b/innobase/include/data0type.ic @@ -143,7 +143,7 @@ dtype_get_charset_coll( /*===================*/ ulint prtype) /* in: precise data type */ { - return((prtype >> 16) & 0xFF); + return((prtype >> 16) & 0xFFUL); } /************************************************************************* @@ -153,7 +153,7 @@ ulint dtype_form_prtype( /*==============*/ ulint old_prtype, /* in: the MySQL type code and the flags - DATA_NONLATIN1 etc. */ + DATA_BINARY_TYPE etc. */ ulint charset_coll) /* in: MySQL charset-collation code */ { ut_a(old_prtype < 256 * 256); @@ -237,9 +237,10 @@ dtype_new_store_for_order_and_null_size( buf[0] = buf[0] | 128; } - if (type->prtype & DATA_NONLATIN1) { - buf[0] = buf[0] | 64; - } + /* In versions < 4.1.2 we had: if (type->prtype & DATA_NONLATIN1) { + buf[0] = buf[0] | 64; + } + */ buf[1] = (byte)(type->prtype & 0xFFUL); @@ -271,10 +272,6 @@ dtype_read_for_order_and_null_size( type->prtype = type->prtype | DATA_BINARY_TYPE; } - if (buf[0] & 64) { - type->prtype = type->prtype | DATA_NONLATIN1; - } - type->len = mach_read_from_2(buf + 2); type->prtype = dtype_form_prtype(type->prtype, @@ -303,10 +300,6 @@ dtype_new_read_for_order_and_null_size( type->prtype = type->prtype | DATA_BINARY_TYPE; } - if (buf[0] & 64) { - type->prtype = type->prtype | DATA_NONLATIN1; - } - type->len = mach_read_from_2(buf + 2); mach_read_from_2(buf + 4); diff --git a/innobase/rem/rem0cmp.c b/innobase/rem/rem0cmp.c index b6f32c83b35..86818168db9 100644 --- a/innobase/rem/rem0cmp.c +++ b/innobase/rem/rem0cmp.c @@ -331,7 +331,9 @@ cmp_data_data_slow( if (cur_type->mtype >= DATA_FLOAT || (cur_type->mtype == DATA_BLOB - && (cur_type->prtype & DATA_NONLATIN1))) { + && 0 == (cur_type->prtype & DATA_BINARY_TYPE) + && dtype_get_charset_coll(cur_type->prtype) != + data_mysql_latin1_swedish_charset_coll)) { return(cmp_whole_field(cur_type, data1, len1, data2, len2)); } @@ -532,8 +534,10 @@ cmp_dtuple_rec_with_match( } if (cur_type->mtype >= DATA_FLOAT - || (cur_type->mtype == DATA_BLOB - && (cur_type->prtype & DATA_NONLATIN1))) { + || (cur_type->mtype == DATA_BLOB + && 0 == (cur_type->prtype & DATA_BINARY_TYPE) + && dtype_get_charset_coll(cur_type->prtype) != + data_mysql_latin1_swedish_charset_coll)) { ret = cmp_whole_field(cur_type, dfield_get_data(dtuple_field), dtuple_f_len, @@ -854,8 +858,10 @@ cmp_rec_rec_with_match( } if (cur_type->mtype >= DATA_FLOAT - || (cur_type->mtype == DATA_BLOB - && (cur_type->prtype & DATA_NONLATIN1))) { + || (cur_type->mtype == DATA_BLOB + && 0 == (cur_type->prtype & DATA_BINARY_TYPE) + && dtype_get_charset_coll(cur_type->prtype) != + data_mysql_latin1_swedish_charset_coll)) { ret = cmp_whole_field(cur_type, rec1_b_ptr, rec1_f_len, diff --git a/sql/ha_innodb.cc b/sql/ha_innodb.cc index 53e0d5aca61..40ba7ab1cef 100644 --- a/sql/ha_innodb.cc +++ b/sql/ha_innodb.cc @@ -755,11 +755,6 @@ innobase_init(void) srv_set_thread_priorities = TRUE; srv_query_thread_priority = QUERY_PRIOR; } - - /* Store the default charset-collation number of this MySQL - installation */ - - data_mysql_default_charset_coll = (ulint)default_charset_info->number; /* Set InnoDB initialization parameters according to the values read from MySQL .cnf file */ @@ -875,6 +870,14 @@ innobase_init(void) srv_print_verbose_log = mysql_embedded ? 0 : 1; + /* Store the default charset-collation number of this MySQL + installation */ + + data_mysql_default_charset_coll = (ulint)default_charset_info->number; + + data_mysql_latin1_swedish_charset_coll = + (ulint)my_charset_latin1.number; + /* Store the latin1_swedish_ci character ordering table to InnoDB. For non-latin1_swedish_ci charsets we use the MySQL comparison functions, and consequently we do not need to know the ordering internally in @@ -3260,7 +3263,6 @@ create_table_def( ulint nulls_allowed; ulint unsigned_type; ulint binary_type; - ulint nonlatin1_type; ulint charset_no; ulint i; @@ -3290,17 +3292,8 @@ create_table_def( unsigned_type = 0; } - if (col_type == DATA_BLOB - && strcmp(field->charset()->name, - "latin1_swedish_ci") != 0) { - nonlatin1_type = DATA_NONLATIN1; - } else { - nonlatin1_type = 0; - } - if (field->binary()) { binary_type = DATA_BINARY_TYPE; - nonlatin1_type = 0; } else { binary_type = 0; } @@ -3319,7 +3312,7 @@ create_table_def( col_type, dtype_form_prtype( (ulint)field->type() | nulls_allowed | unsigned_type - | nonlatin1_type | binary_type, + | binary_type, + charset_no), field->pack_length(), 0); } |