| 1 |
/* Copyright (C) 2000-2006 MySQL AB |
| 2 |
|
| 3 |
This program is free software; you can redistribute it and/or modify |
| 4 |
it under the terms of the GNU General Public License as published by |
| 5 |
the Free Software Foundation; version 2 of the License. |
| 6 |
|
| 7 |
This program is distributed in the hope that it will be useful, |
| 8 |
but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 9 |
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
| 10 |
GNU General Public License for more details. |
| 11 |
|
| 12 |
You should have received a copy of the GNU General Public License |
| 13 |
along with this program; if not, write to the Free Software |
| 14 |
Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA */ |
| 15 |
|
| 16 |
|
| 17 |
/* mysql_select and join optimization */ |
| 18 |
|
| 19 |
#ifdef USE_PRAGMA_IMPLEMENTATION |
| 20 |
#pragma implementation // gcc: Class implementation |
| 21 |
#endif |
| 22 |
|
| 23 |
#include "mysql_priv.h" |
| 24 |
#include "sql_select.h" |
| 25 |
#include "sql_cursor.h" |
| 26 |
|
| 27 |
#include <m_ctype.h> |
| 28 |
#include <hash.h> |
| 29 |
#include <ft_global.h> |
| 30 |
|
| 31 |
const char *join_type_str[]={ "UNKNOWN","system","const","eq_ref","ref", |
| 32 |
"MAYBE_REF","ALL","range","index","fulltext", |
| 33 |
"ref_or_null","unique_subquery","index_subquery", |
| 34 |
"index_merge" |
| 35 |
}; |
| 36 |
|
| 37 |
struct st_sargable_param; |
| 38 |
|
| 39 |
static void optimize_keyuse(JOIN *join, DYNAMIC_ARRAY *keyuse_array); |
| 40 |
static bool make_join_statistics(JOIN *join, TABLE_LIST *leaves, COND *conds, |
| 41 |
DYNAMIC_ARRAY *keyuse); |
| 42 |
static bool update_ref_and_keys(THD *thd, DYNAMIC_ARRAY *keyuse, |
| 43 |
JOIN_TAB *join_tab, |
| 44 |
uint tables, COND *conds, |
| 45 |
COND_EQUAL *cond_equal, |
| 46 |
table_map table_map, SELECT_LEX *select_lex, |
| 47 |
st_sargable_param **sargables); |
| 48 |
static int sort_keyuse(KEYUSE *a,KEYUSE *b); |
| 49 |
static void set_position(JOIN *join,uint index,JOIN_TAB *table,KEYUSE *key); |
| 50 |
static bool create_ref_for_key(JOIN *join, JOIN_TAB *j, KEYUSE *org_keyuse, |
| 51 |
table_map used_tables); |
| 52 |
static bool choose_plan(JOIN *join,table_map join_tables); |
| 53 |
|
| 54 |
static void best_access_path(JOIN *join, JOIN_TAB *s, THD *thd, |
| 55 |
table_map remaining_tables, uint idx, |
| 56 |
double record_count, double read_time); |
| 57 |
static void optimize_straight_join(JOIN *join, table_map join_tables); |
| 58 |
static bool greedy_search(JOIN *join, table_map remaining_tables, |
| 59 |
uint depth, uint prune_level); |
| 60 |
static bool best_extension_by_limited_search(JOIN *join, |
| 61 |
table_map remaining_tables, |
| 62 |
uint idx, double record_count, |
| 63 |
double read_time, uint depth, |
| 64 |
uint prune_level); |
| 65 |
static uint determine_search_depth(JOIN* join); |
| 66 |
static int join_tab_cmp(const void* ptr1, const void* ptr2); |
| 67 |
static int join_tab_cmp_straight(const void* ptr1, const void* ptr2); |
| 68 |
/* |
| 69 |
TODO: 'find_best' is here only temporarily until 'greedy_search' is |
| 70 |
tested and approved. |
| 71 |
*/ |
| 72 |
static bool find_best(JOIN *join,table_map rest_tables,uint index, |
| 73 |
double record_count,double read_time); |
| 74 |
static uint cache_record_length(JOIN *join,uint index); |
| 75 |
static double prev_record_reads(JOIN *join,table_map found_ref); |
| 76 |
static bool get_best_combination(JOIN *join); |
| 77 |
static store_key *get_store_key(THD *thd, |
| 78 |
KEYUSE *keyuse, table_map used_tables, |
| 79 |
KEY_PART_INFO *key_part, char *key_buff, |
| 80 |
uint maybe_null); |
| 81 |
static bool make_simple_join(JOIN *join,TABLE *tmp_table); |
| 82 |
static void make_outerjoin_info(JOIN *join); |
| 83 |
static bool make_join_select(JOIN *join,SQL_SELECT *select,COND *item); |
| 84 |
static void make_join_readinfo(JOIN *join, ulonglong options); |
| 85 |
static bool only_eq_ref_tables(JOIN *join, ORDER *order, table_map tables); |
| 86 |
static void update_depend_map(JOIN *join); |
| 87 |
static void update_depend_map(JOIN *join, ORDER *order); |
| 88 |
static ORDER *remove_const(JOIN *join,ORDER *first_order,COND *cond, |
| 89 |
bool change_list, bool *simple_order); |
| 90 |
static int return_zero_rows(JOIN *join, select_result *res,TABLE_LIST *tables, |
| 91 |
List<Item> &fields, bool send_row, |
| 92 |
ulonglong select_options, const char *info, |
| 93 |
Item *having); |
| 94 |
static COND *build_equal_items(THD *thd, COND *cond, |
| 95 |
COND_EQUAL *inherited, |
| 96 |
List<TABLE_LIST> *join_list, |
| 97 |
COND_EQUAL **cond_equal_ref); |
| 98 |
static COND* substitute_for_best_equal_field(COND *cond, |
| 99 |
COND_EQUAL *cond_equal, |
| 100 |
void *table_join_idx); |
| 101 |
static COND *simplify_joins(JOIN *join, List<TABLE_LIST> *join_list, |
| 102 |
COND *conds, bool top); |
| 103 |
static bool check_interleaving_with_nj(JOIN_TAB *last, JOIN_TAB *next); |
| 104 |
static void restore_prev_nj_state(JOIN_TAB *last); |
| 105 |
static void reset_nj_counters(List<TABLE_LIST> *join_list); |
| 106 |
static uint build_bitmap_for_nested_joins(List<TABLE_LIST> *join_list, |
| 107 |
uint first_unused); |
| 108 |
|
| 109 |
static COND *optimize_cond(JOIN *join, COND *conds, |
| 110 |
List<TABLE_LIST> *join_list, |
| 111 |
Item::cond_result *cond_value); |
| 112 |
static bool const_expression_in_where(COND *conds,Item *item, Item **comp_item); |
| 113 |
static bool open_tmp_table(TABLE *table); |
| 114 |
static bool create_myisam_tmp_table(TABLE *table,TMP_TABLE_PARAM *param, |
| 115 |
ulonglong options); |
| 116 |
static int do_select(JOIN *join,List<Item> *fields,TABLE *tmp_table, |
| 117 |
Procedure *proc); |
| 118 |
|
| 119 |
static enum_nested_loop_state |
| 120 |
evaluate_join_record(JOIN *join, JOIN_TAB *join_tab, |
| 121 |
int error, my_bool *report_error); |
| 122 |
static enum_nested_loop_state |
| 123 |
evaluate_null_complemented_join_record(JOIN *join, JOIN_TAB *join_tab); |
| 124 |
static enum_nested_loop_state |
| 125 |
flush_cached_records(JOIN *join, JOIN_TAB *join_tab, bool skip_last); |
| 126 |
static enum_nested_loop_state |
| 127 |
end_send(JOIN *join, JOIN_TAB *join_tab, bool end_of_records); |
| 128 |
static enum_nested_loop_state |
| 129 |
end_send_group(JOIN *join, JOIN_TAB *join_tab, bool end_of_records); |
| 130 |
static enum_nested_loop_state |
| 131 |
end_write(JOIN *join, JOIN_TAB *join_tab, bool end_of_records); |
| 132 |
static enum_nested_loop_state |
| 133 |
end_update(JOIN *join, JOIN_TAB *join_tab, bool end_of_records); |
| 134 |
static enum_nested_loop_state |
| 135 |
end_unique_update(JOIN *join, JOIN_TAB *join_tab, bool end_of_records); |
| 136 |
static enum_nested_loop_state |
| 137 |
end_write_group(JOIN *join, JOIN_TAB *join_tab, bool end_of_records); |
| 138 |
|
| 139 |
static int test_if_group_changed(List<Cached_item> &list); |
| 140 |
static int join_read_const_table(JOIN_TAB *tab, POSITION *pos); |
| 141 |
static int join_read_system(JOIN_TAB *tab); |
| 142 |
static int join_read_const(JOIN_TAB *tab); |
| 143 |
static int join_read_key(JOIN_TAB *tab); |
| 144 |
static int join_read_always_key(JOIN_TAB *tab); |
| 145 |
static int join_read_last_key(JOIN_TAB *tab); |
| 146 |
static int join_no_more_records(READ_RECORD *info); |
| 147 |
static int join_read_next(READ_RECORD *info); |
| 148 |
static int join_init_quick_read_record(JOIN_TAB *tab); |
| 149 |
static int test_if_quick_select(JOIN_TAB *tab); |
| 150 |
static int join_init_read_record(JOIN_TAB *tab); |
| 151 |
static int join_read_first(JOIN_TAB *tab); |
| 152 |
static int join_read_next(READ_RECORD *info); |
| 153 |
static int join_read_next_same(READ_RECORD *info); |
| 154 |
static int join_read_last(JOIN_TAB *tab); |
| 155 |
static int join_read_prev_same(READ_RECORD *info); |
| 156 |
static int join_read_prev(READ_RECORD *info); |
| 157 |
static int join_ft_read_first(JOIN_TAB *tab); |
| 158 |
static int join_ft_read_next(READ_RECORD *info); |
| 159 |
int join_read_always_key_or_null(JOIN_TAB *tab); |
| 160 |
int join_read_next_same_or_null(READ_RECORD *info); |
| 161 |
static COND *make_cond_for_table(COND *cond,table_map table, |
| 162 |
table_map used_table); |
| 163 |
static Item* part_of_refkey(TABLE *form,Field *field); |
| 164 |
uint find_shortest_key(TABLE *table, const key_map *usable_keys); |
| 165 |
static bool test_if_skip_sort_order(JOIN_TAB *tab,ORDER *order, |
| 166 |
ha_rows select_limit, bool no_changes); |
| 167 |
static bool list_contains_unique_index(TABLE *table, |
| 168 |
bool (*find_func) (Field *, void *), void *data); |
| 169 |
static bool find_field_in_item_list (Field *field, void *data); |
| 170 |
static bool find_field_in_order_list (Field *field, void *data); |
| 171 |
static int create_sort_index(THD *thd, JOIN *join, ORDER *order, |
| 172 |
ha_rows filesort_limit, ha_rows select_limit); |
| 173 |
static int remove_duplicates(JOIN *join,TABLE *entry,List<Item> &fields, |
| 174 |
Item *having); |
| 175 |
static int remove_dup_with_compare(THD *thd, TABLE *entry, Field **field, |
| 176 |
ulong offset,Item *having); |
| 177 |
static int remove_dup_with_hash_index(THD *thd,TABLE *table, |
| 178 |
uint field_count, Field **first_field, |
| 179 |
|
| 180 |
ulong key_length,Item *having); |
| 181 |
static int join_init_cache(THD *thd,JOIN_TAB *tables,uint table_count); |
| 182 |
static ulong used_blob_length(CACHE_FIELD **ptr); |
| 183 |
static bool store_record_in_cache(JOIN_CACHE *cache); |
| 184 |
static void reset_cache_read(JOIN_CACHE *cache); |
| 185 |
static void reset_cache_write(JOIN_CACHE *cache); |
| 186 |
static void read_cached_record(JOIN_TAB *tab); |
| 187 |
static bool cmp_buffer_with_ref(JOIN_TAB *tab); |
| 188 |
static bool setup_new_fields(THD *thd, List<Item> &fields, |
| 189 |
List<Item> &all_fields, ORDER *new_order); |
| 190 |
static ORDER *create_distinct_group(THD *thd, Item **ref_pointer_array, |
| 191 |
ORDER *order, List<Item> &fields, |
| 192 |
List<Item> &all_fields, |
| 193 |
bool *all_order_by_fields_used); |
| 194 |
static bool test_if_subpart(ORDER *a,ORDER *b); |
| 195 |
static TABLE *get_sort_by_table(ORDER *a,ORDER *b,TABLE_LIST *tables); |
| 196 |
static void calc_group_buffer(JOIN *join,ORDER *group); |
| 197 |
static bool make_group_fields(JOIN *main_join, JOIN *curr_join); |
| 198 |
static bool alloc_group_fields(JOIN *join,ORDER *group); |
| 199 |
// Create list for using with tempory table |
| 200 |
static bool change_to_use_tmp_fields(THD *thd, Item **ref_pointer_array, |
| 201 |
List<Item> &new_list1, |
| 202 |
List<Item> &new_list2, |
| 203 |
uint elements, List<Item> &items); |
| 204 |
// Create list for using with tempory table |
| 205 |
static bool change_refs_to_tmp_fields(THD *thd, Item **ref_pointer_array, |
| 206 |
List<Item> &new_list1, |
| 207 |
List<Item> &new_list2, |
| 208 |
uint elements, List<Item> &items); |
| 209 |
static void init_tmptable_sum_functions(Item_sum **func); |
| 210 |
static void update_tmptable_sum_func(Item_sum **func,TABLE *tmp_table); |
| 211 |
static void copy_sum_funcs(Item_sum **func_ptr, Item_sum **end); |
| 212 |
static bool add_ref_to_table_cond(THD *thd, JOIN_TAB *join_tab); |
| 213 |
static bool setup_sum_funcs(THD *thd, Item_sum **func_ptr); |
| 214 |
static bool init_sum_functions(Item_sum **func, Item_sum **end); |
| 215 |
static bool update_sum_func(Item_sum **func); |
| 216 |
static void select_describe(JOIN *join, bool need_tmp_table,bool need_order, |
| 217 |
bool distinct, const char *message=NullS); |
| 218 |
static Item *remove_additional_cond(Item* conds); |
| 219 |
static void add_group_and_distinct_keys(JOIN *join, JOIN_TAB *join_tab); |
| 220 |
|
| 221 |
|
| 222 |
/* |
| 223 |
This handles SELECT with and without UNION |
| 224 |
*/ |
| 225 |
|
| 226 |
bool handle_select(THD *thd, LEX *lex, select_result *result, |
| 227 |
ulong setup_tables_done_option) |
| 228 |
{ |
| 229 |
bool res; |
| 230 |
register SELECT_LEX *select_lex = &lex->select_lex; |
| 231 |
DBUG_ENTER("handle_select"); |
| 232 |
|
| 233 |
if (select_lex->next_select() || select_lex->master_unit()->fake_select_lex) |
| 234 |
res= mysql_union(thd, lex, result, &lex->unit, setup_tables_done_option); |
| 235 |
else |
| 236 |
{ |
| 237 |
SELECT_LEX_UNIT *unit= &lex->unit; |
| 238 |
unit->set_limit(unit->global_parameters); |
| 239 |
/* |
| 240 |
'options' of mysql_select will be set in JOIN, as far as JOIN for |
| 241 |
every PS/SP execution new, we will not need reset this flag if |
| 242 |
setup_tables_done_option changed for next rexecution |
| 243 |
*/ |
| 244 |
res= mysql_select(thd, &select_lex->ref_pointer_array, |
| 245 |
(TABLE_LIST*) select_lex->table_list.first, |
| 246 |
select_lex->with_wild, select_lex->item_list, |
| 247 |
select_lex->where, |
| 248 |
select_lex->order_list.elements + |
| 249 |
select_lex->group_list.elements, |
| 250 |
(ORDER*) select_lex->order_list.first, |
| 251 |
(ORDER*) select_lex->group_list.first, |
| 252 |
select_lex->having, |
| 253 |
(ORDER*) lex->proc_list.first, |
| 254 |
select_lex->options | thd->options | |
| 255 |
setup_tables_done_option, |
| 256 |
result, unit, select_lex); |
| 257 |
} |
| 258 |
DBUG_PRINT("info",("res: %d report_error: %d", res, |
| 259 |
thd->net.report_error)); |
| 260 |
res|= thd->net.report_error; |
| 261 |
if (unlikely(res)) |
| 262 |
result->abort(); |
| 263 |
|
| 264 |
DBUG_RETURN(res); |
| 265 |
} |
| 266 |
|
| 267 |
|
| 268 |
/* |
| 269 |
Fix fields referenced from inner selects. |
| 270 |
|
| 271 |
SYNOPSIS |
| 272 |
fix_inner_refs() |
| 273 |
thd Thread handle |
| 274 |
all_fields List of all fields used in select |
| 275 |
select Current select |
| 276 |
ref_pointer_array Array of references to Items used in current select |
| 277 |
|
| 278 |
DESCRIPTION |
| 279 |
The function serves 3 purposes - adds fields referenced from inner |
| 280 |
selects to the current select list, resolves which class to use |
| 281 |
to access referenced item (Item_ref of Item_direct_ref) and fixes |
| 282 |
references (Item_ref objects) to these fields. |
| 283 |
|
| 284 |
If a field isn't already in the select list and the ref_pointer_array |
| 285 |
is provided then it is added to the all_fields list and the pointer to |
| 286 |
it is saved in the ref_pointer_array. |
| 287 |
|
| 288 |
The class to access the outer field is determined by the following rules: |
| 289 |
1. If the outer field isn't used under an aggregate function |
| 290 |
then the Item_ref class should be used. |
| 291 |
2. If the outer field is used under an aggregate function and this |
| 292 |
function is aggregated in the select where the outer field was |
| 293 |
resolved or in some more inner select then the Item_direct_ref |
| 294 |
class should be used. |
| 295 |
The resolution is done here and not at the fix_fields() stage as |
| 296 |
it can be done only after sum functions are fixed and pulled up to |
| 297 |
selects where they are have to be aggregated. |
| 298 |
When the class is chosen it substitutes the original field in the |
| 299 |
Item_outer_ref object. |
| 300 |
|
| 301 |
After this we proceed with fixing references (Item_outer_ref objects) to |
| 302 |
this field from inner subqueries. |
| 303 |
|
| 304 |
RETURN |
| 305 |
TRUE an error occured |
| 306 |
FALSE ok |
| 307 |
*/ |
| 308 |
|
| 309 |
bool |
| 310 |
fix_inner_refs(THD *thd, List<Item> &all_fields, SELECT_LEX *select, |
| 311 |
Item **ref_pointer_array) |
| 312 |
{ |
| 313 |
Item_outer_ref *ref; |
| 314 |
bool res= FALSE; |
| 315 |
bool direct_ref= FALSE; |
| 316 |
|
| 317 |
List_iterator<Item_outer_ref> ref_it(select->inner_refs_list); |
| 318 |
while ((ref= ref_it++)) |
| 319 |
{ |
| 320 |
Item *item= ref->outer_ref; |
| 321 |
Item **item_ref= ref->ref; |
| 322 |
Item_ref *new_ref; |
| 323 |
/* |
| 324 |
TODO: this field item already might be present in the select list. |
| 325 |
In this case instead of adding new field item we could use an |
| 326 |
existing one. The change will lead to less operations for copying fields, |
| 327 |
smaller temporary tables and less data passed through filesort. |
| 328 |
*/ |
| 329 |
if (ref_pointer_array && !ref->found_in_select_list) |
| 330 |
{ |
| 331 |
int el= all_fields.elements; |
| 332 |
ref_pointer_array[el]= item; |
| 333 |
/* Add the field item to the select list of the current select. */ |
| 334 |
all_fields.push_front(item); |
| 335 |
/* |
| 336 |
If it's needed reset each Item_ref item that refers this field with |
| 337 |
a new reference taken from ref_pointer_array. |
| 338 |
*/ |
| 339 |
item_ref= ref_pointer_array + el; |
| 340 |
} |
| 341 |
|
| 342 |
if (ref->in_sum_func) |
| 343 |
{ |
| 344 |
Item_sum *sum_func; |
| 345 |
if (ref->in_sum_func->nest_level > select->nest_level) |
| 346 |
direct_ref= TRUE; |
| 347 |
else |
| 348 |
{ |
| 349 |
for (sum_func= ref->in_sum_func; sum_func && |
| 350 |
sum_func->aggr_level >= select->nest_level; |
| 351 |
sum_func= sum_func->in_sum_func) |
| 352 |
{ |
| 353 |
if (sum_func->aggr_level == select->nest_level) |
| 354 |
{ |
| 355 |
direct_ref= TRUE; |
| 356 |
break; |
| 357 |
} |
| 358 |
} |
| 359 |
} |
| 360 |
} |
| 361 |
new_ref= direct_ref ? |
| 362 |
new Item_direct_ref(ref->context, item_ref, ref->field_name, |
| 363 |
ref->table_name, ref->alias_name_used) : |
| 364 |
new Item_ref(ref->context, item_ref, ref->field_name, |
| 365 |
ref->table_name, ref->alias_name_used); |
| 366 |
if (!new_ref) |
| 367 |
return TRUE; |
| 368 |
ref->outer_ref= new_ref; |
| 369 |
ref->ref= &ref->outer_ref; |
| 370 |
|
| 371 |
if (!ref->fixed && ref->fix_fields(thd, 0)) |
| 372 |
return TRUE; |
| 373 |
thd->used_tables|= item->used_tables(); |
| 374 |
} |
| 375 |
return res; |
| 376 |
} |
| 377 |
|
| 378 |
/* |
| 379 |
Function to setup clauses without sum functions |
| 380 |
*/ |
| 381 |
inline int setup_without_group(THD *thd, Item **ref_pointer_array, |
| 382 |
TABLE_LIST *tables, |
| 383 |
TABLE_LIST *leaves, |
| 384 |
List<Item> &fields, |
| 385 |
List<Item> &all_fields, |
| 386 |
COND **conds, |
| 387 |
ORDER *order, |
| 388 |
ORDER *group, bool *hidden_group_fields) |
| 389 |
{ |
| 390 |
int res; |
| 391 |
nesting_map save_allow_sum_func=thd->lex->allow_sum_func ; |
| 392 |
DBUG_ENTER("setup_without_group"); |
| 393 |
|
| 394 |
thd->lex->allow_sum_func&= ~(1 << thd->lex->current_select->nest_level); |
| 395 |
res= setup_conds(thd, tables, leaves, conds); |
| 396 |
|
| 397 |
thd->lex->allow_sum_func|= 1 << thd->lex->current_select->nest_level; |
| 398 |
res= res || setup_order(thd, ref_pointer_array, tables, fields, all_fields, |
| 399 |
order); |
| 400 |
thd->lex->allow_sum_func&= ~(1 << thd->lex->current_select->nest_level); |
| 401 |
res= res || setup_group(thd, ref_pointer_array, tables, fields, all_fields, |
| 402 |
group, hidden_group_fields); |
| 403 |
thd->lex->allow_sum_func= save_allow_sum_func; |
| 404 |
DBUG_RETURN(res); |
| 405 |
} |
| 406 |
|
| 407 |
/***************************************************************************** |
| 408 |
Check fields, find best join, do the select and output fields. |
| 409 |
mysql_select assumes that all tables are already opened |
| 410 |
*****************************************************************************/ |
| 411 |
|
| 412 |
/* |
| 413 |
Prepare of whole select (including sub queries in future). |
| 414 |
return -1 on error |
| 415 |
0 on success |
| 416 |
*/ |
| 417 |
int |
| 418 |
JOIN::prepare(Item ***rref_pointer_array, |
| 419 |
TABLE_LIST *tables_init, |
| 420 |
uint wild_num, COND *conds_init, uint og_num, |
| 421 |
ORDER *order_init, ORDER *group_init, |
| 422 |
Item *having_init, |
| 423 |
ORDER *proc_param_init, SELECT_LEX *select_lex_arg, |
| 424 |
SELECT_LEX_UNIT *unit_arg) |
| 425 |
{ |
| 426 |
DBUG_ENTER("JOIN::prepare"); |
| 427 |
|
| 428 |
// to prevent double initialization on EXPLAIN |
| 429 |
if (optimized) |
| 430 |
DBUG_RETURN(0); |
| 431 |
|
| 432 |
conds= conds_init; |
| 433 |
order= order_init; |
| 434 |
group_list= group_init; |
| 435 |
having= having_init; |
| 436 |
proc_param= proc_param_init; |
| 437 |
tables_list= tables_init; |
| 438 |
select_lex= select_lex_arg; |
| 439 |
select_lex->join= this; |
| 440 |
join_list= &select_lex->top_join_list; |
| 441 |
union_part= (unit_arg->first_select()->next_select() != 0); |
| 442 |
|
| 443 |
thd->lex->current_select->is_item_list_lookup= 1; |
| 444 |
/* |
| 445 |
If we have already executed SELECT, then it have not sense to prevent |
| 446 |
its table from update (see unique_table()) |
| 447 |
*/ |
| 448 |
if (thd->derived_tables_processing) |
| 449 |
select_lex->exclude_from_table_unique_test= TRUE; |
| 450 |
|
| 451 |
/* Check that all tables, fields, conds and order are ok */ |
| 452 |
|
| 453 |
if (!(select_options & OPTION_SETUP_TABLES_DONE) && |
| 454 |
setup_tables_and_check_access(thd, &select_lex->context, join_list, |
| 455 |
tables_list, &conds, |
| 456 |
&select_lex->leaf_tables, FALSE, |
| 457 |
SELECT_ACL, SELECT_ACL)) |
| 458 |
DBUG_RETURN(-1); |
| 459 |
|
| 460 |
TABLE_LIST *table_ptr; |
| 461 |
for (table_ptr= select_lex->leaf_tables; |
| 462 |
table_ptr; |
| 463 |
table_ptr= table_ptr->next_leaf) |
| 464 |
tables++; |
| 465 |
|
| 466 |
if (setup_wild(thd, tables_list, fields_list, &all_fields, wild_num) || |
| 467 |
select_lex->setup_ref_array(thd, og_num) || |
| 468 |
setup_fields(thd, (*rref_pointer_array), fields_list, 1, |
| 469 |
&all_fields, 1) || |
| 470 |
setup_without_group(thd, (*rref_pointer_array), tables_list, |
| 471 |
select_lex->leaf_tables, fields_list, |
| 472 |
all_fields, &conds, order, group_list, |
| 473 |
&hidden_group_fields)) |
| 474 |
DBUG_RETURN(-1); /* purecov: inspected */ |
| 475 |
|
| 476 |
ref_pointer_array= *rref_pointer_array; |
| 477 |
|
| 478 |
if (having) |
| 479 |
{ |
| 480 |
nesting_map save_allow_sum_func= thd->lex->allow_sum_func; |
| 481 |
thd->where="having clause"; |
| 482 |
thd->lex->allow_sum_func|= 1 << select_lex_arg->nest_level; |
| 483 |
select_lex->having_fix_field= 1; |
| 484 |
bool having_fix_rc= (!having->fixed && |
| 485 |
(having->fix_fields(thd, &having) || |
| 486 |
having->check_cols(1))); |
| 487 |
select_lex->having_fix_field= 0; |
| 488 |
if (having_fix_rc || thd->net.report_error) |
| 489 |
DBUG_RETURN(-1); /* purecov: inspected */ |
| 490 |
thd->lex->allow_sum_func= save_allow_sum_func; |
| 491 |
} |
| 492 |
|
| 493 |
if (!thd->lex->view_prepare_mode) |
| 494 |
{ |
| 495 |
Item_subselect *subselect; |
| 496 |
/* Is it subselect? */ |
| 497 |
if ((subselect= select_lex->master_unit()->item)) |
| 498 |
{ |
| 499 |
Item_subselect::trans_res res; |
| 500 |
if ((res= subselect->select_transformer(this)) != |
| 501 |
Item_subselect::RES_OK) |
| 502 |
{ |
| 503 |
select_lex->fix_prepare_information(thd, &conds, &having); |
| 504 |
DBUG_RETURN((res == Item_subselect::RES_ERROR)); |
| 505 |
} |
| 506 |
} |
| 507 |
} |
| 508 |
|
| 509 |
select_lex->fix_prepare_information(thd, &conds, &having); |
| 510 |
|
| 511 |
if (order) |
| 512 |
{ |
| 513 |
ORDER *ord; |
| 514 |
for (ord= order; ord; ord= ord->next) |
| 515 |
{ |
| 516 |
Item *item= *ord->item; |
| 517 |
if (item->with_sum_func && item->type() != Item::SUM_FUNC_ITEM) |
| 518 |
item->split_sum_func(thd, ref_pointer_array, all_fields); |
| 519 |
} |
| 520 |
} |
| 521 |
|
| 522 |
if (having && having->with_sum_func) |
| 523 |
having->split_sum_func2(thd, ref_pointer_array, all_fields, |
| 524 |
&having, TRUE); |
| 525 |
if (select_lex->inner_sum_func_list) |
| 526 |
{ |
| 527 |
Item_sum *end=select_lex->inner_sum_func_list; |
| 528 |
Item_sum *item_sum= end; |
| 529 |
do |
| 530 |
{ |
| 531 |
item_sum= item_sum->next; |
| 532 |
item_sum->split_sum_func2(thd, ref_pointer_array, |
| 533 |
all_fields, item_sum->ref_by, FALSE); |
| 534 |
} while (item_sum != end); |
| 535 |
} |
| 536 |
|
| 537 |
if (select_lex->inner_refs_list.elements && |
| 538 |
fix_inner_refs(thd, all_fields, select_lex, ref_pointer_array)) |
| 539 |
DBUG_RETURN(-1); |
| 540 |
|
| 541 |
if (group_list) |
| 542 |
{ |
| 543 |
/* |
| 544 |
Because HEAP tables can't index BIT fields we need to use an |
| 545 |
additional hidden field for grouping because later it will be |
| 546 |
converted to a LONG field. Original field will remain of the |
| 547 |
BIT type and will be returned to a client. |
| 548 |
*/ |
| 549 |
for (ORDER *ord= group_list; ord; ord= ord->next) |
| 550 |
{ |
| 551 |
if ((*ord->item)->type() == Item::FIELD_ITEM && |
| 552 |
(*ord->item)->field_type() == MYSQL_TYPE_BIT) |
| 553 |
{ |
| 554 |
Item_field *field= new Item_field(thd, *(Item_field**)ord->item); |
| 555 |
int el= all_fields.elements; |
| 556 |
ref_pointer_array[el]= field; |
| 557 |
all_fields.push_front(field); |
| 558 |
ord->item= ref_pointer_array + el; |
| 559 |
} |
| 560 |
} |
| 561 |
} |
| 562 |
|
| 563 |
if (setup_ftfuncs(select_lex)) /* should be after having->fix_fields */ |
| 564 |
DBUG_RETURN(-1); |
| 565 |
|
| 566 |
|
| 567 |
/* |
| 568 |
Check if one one uses a not constant column with group functions |
| 569 |
and no GROUP BY. |
| 570 |
TODO: Add check of calculation of GROUP functions and fields: |
| 571 |
SELECT COUNT(*)+table.col1 from table1; |
| 572 |
*/ |
| 573 |
{ |
| 574 |
if (!group_list) |
| 575 |
{ |
| 576 |
uint flag=0; |
| 577 |
List_iterator_fast<Item> it(fields_list); |
| 578 |
Item *item; |
| 579 |
while ((item= it++)) |
| 580 |
{ |
| 581 |
if (item->with_sum_func) |
| 582 |
flag|=1; |
| 583 |
else if (!(flag & 2) && !item->const_during_execution()) |
| 584 |
flag|=2; |
| 585 |
} |
| 586 |
if (flag == 3) |
| 587 |
{ |
| 588 |
my_message(ER_MIX_OF_GROUP_FUNC_AND_FIELDS, |
| 589 |
ER(ER_MIX_OF_GROUP_FUNC_AND_FIELDS), MYF(0)); |
| 590 |
DBUG_RETURN(-1); |
| 591 |
} |
| 592 |
} |
| 593 |
} |
| 594 |
{ |
| 595 |
/* Caclulate the number of groups */ |
| 596 |
send_group_parts= 0; |
| 597 |
for (ORDER *group_tmp= group_list ; group_tmp ; group_tmp= group_tmp->next) |
| 598 |
send_group_parts++; |
| 599 |
} |
| 600 |
|
| 601 |
procedure= setup_procedure(thd, proc_param, result, fields_list, &error); |
| 602 |
if (error) |
| 603 |
goto err; /* purecov: inspected */ |
| 604 |
if (procedure) |
| 605 |
{ |
| 606 |
if (setup_new_fields(thd, fields_list, all_fields, |
| 607 |
procedure->param_fields)) |
| 608 |
goto err; /* purecov: inspected */ |
| 609 |
if (procedure->group) |
| 610 |
{ |
| 611 |
if (!test_if_subpart(procedure->group,group_list)) |
| 612 |
{ /* purecov: inspected */ |
| 613 |
my_message(ER_DIFF_GROUPS_PROC, ER(ER_DIFF_GROUPS_PROC), |
| 614 |
MYF(0)); /* purecov: inspected */ |
| 615 |
goto err; /* purecov: inspected */ |
| 616 |
} |
| 617 |
} |
| 618 |
#ifdef NOT_NEEDED |
| 619 |
else if (!group_list && procedure->flags & PROC_GROUP) |
| 620 |
{ |
| 621 |
my_message(ER_NO_GROUP_FOR_PROC, MYF(0)); |
| 622 |
goto err; |
| 623 |
} |
| 624 |
#endif |
| 625 |
if (order && (procedure->flags & PROC_NO_SORT)) |
| 626 |
{ /* purecov: inspected */ |
| 627 |
my_message(ER_ORDER_WITH_PROC, ER(ER_ORDER_WITH_PROC), |
| 628 |
MYF(0)); /* purecov: inspected */ |
| 629 |
goto err; /* purecov: inspected */ |
| 630 |
} |
| 631 |
} |
| 632 |
|
| 633 |
if (!procedure && result && result->prepare(fields_list, unit_arg)) |
| 634 |
goto err; /* purecov: inspected */ |
| 635 |
|
| 636 |
/* Init join struct */ |
| 637 |
count_field_types(select_lex, &tmp_table_param, all_fields, 0); |
| 638 |
ref_pointer_array_size= all_fields.elements*sizeof(Item*); |
| 639 |
this->group= group_list != 0; |
| 640 |
unit= unit_arg; |
| 641 |
|
| 642 |
#ifdef RESTRICTED_GROUP |
| 643 |
if (sum_func_count && !group_list && (func_count || field_count)) |
| 644 |
{ |
| 645 |
my_message(ER_WRONG_SUM_SELECT,ER(ER_WRONG_SUM_SELECT),MYF(0)); |
| 646 |
goto err; |
| 647 |
} |
| 648 |
#endif |
| 649 |
if (select_lex->olap == ROLLUP_TYPE && rollup_init()) |
| 650 |
goto err; |
| 651 |
if (alloc_func_list()) |
| 652 |
goto err; |
| 653 |
|
| 654 |
DBUG_RETURN(0); // All OK |
| 655 |
|
| 656 |
err: |
| 657 |
delete procedure; /* purecov: inspected */ |
| 658 |
procedure= 0; |
| 659 |
DBUG_RETURN(-1); /* purecov: inspected */ |
| 660 |
} |
| 661 |
|
| 662 |
|
| 663 |
/* |
| 664 |
Remove the predicates pushed down into the subquery |
| 665 |
|
| 666 |
SYNOPSIS |
| 667 |
JOIN::remove_subq_pushed_predicates() |
| 668 |
where IN Must be NULL |
| 669 |
OUT The remaining WHERE condition, or NULL |
| 670 |
|
| 671 |
DESCRIPTION |
| 672 |
Given that this join will be executed using (unique|index)_subquery, |
| 673 |
without "checking NULL", remove the predicates that were pushed down |
| 674 |
into the subquery. |
| 675 |
|
| 676 |
We can remove the equalities that will be guaranteed to be true by the |
| 677 |
fact that subquery engine will be using index lookup. |
| 678 |
|
| 679 |
If the subquery compares scalar values, we can remove the condition that |
| 680 |
was wrapped into trig_cond (it will be checked when needed by the subquery |
| 681 |
engine) |
| 682 |
|
| 683 |
If the subquery compares row values, we need to keep the wrapped |
| 684 |
equalities in the WHERE clause: when the left (outer) tuple has both NULL |
| 685 |
and non-NULL values, we'll do a full table scan and will rely on the |
| 686 |
equalities corresponding to non-NULL parts of left tuple to filter out |
| 687 |
non-matching records. |
| 688 |
*/ |
| 689 |
|
| 690 |
void JOIN::remove_subq_pushed_predicates(Item **where) |
| 691 |
{ |
| 692 |
if (conds->type() == Item::FUNC_ITEM && |
| 693 |
((Item_func *)this->conds)->functype() == Item_func::EQ_FUNC && |
| 694 |
((Item_func *)conds)->arguments()[0]->type() == Item::REF_ITEM && |
| 695 |
((Item_func *)conds)->arguments()[1]->type() == Item::FIELD_ITEM) |
| 696 |
{ |
| 697 |
*where= 0; |
| 698 |
return; |
| 699 |
} |
| 700 |
if (conds->type() == Item::COND_ITEM && |
| 701 |
((class Item_func *)this->conds)->functype() == |
| 702 |
Item_func::COND_AND_FUNC) |
| 703 |
{ |
| 704 |
*where= remove_additional_cond(conds); |
| 705 |
} |
| 706 |
} |
| 707 |
|
| 708 |
|
| 709 |
/* |
| 710 |
Index lookup-based subquery: save some flags for EXPLAIN output |
| 711 |
|
| 712 |
SYNOPSIS |
| 713 |
save_index_subquery_explain_info() |
| 714 |
join_tab Subquery's join tab (there is only one as index lookup is |
| 715 |
only used for subqueries that are single-table SELECTs) |
| 716 |
where Subquery's WHERE clause |
| 717 |
|
| 718 |
DESCRIPTION |
| 719 |
For index lookup-based subquery (i.e. one executed with |
| 720 |
subselect_uniquesubquery_engine or subselect_indexsubquery_engine), |
| 721 |
check its EXPLAIN output row should contain |
| 722 |
"Using index" (TAB_INFO_FULL_SCAN_ON_NULL) |
| 723 |
"Using Where" (TAB_INFO_USING_WHERE) |
| 724 |
"Full scan on NULL key" (TAB_INFO_FULL_SCAN_ON_NULL) |
| 725 |
and set appropriate flags in join_tab->packed_info. |
| 726 |
*/ |
| 727 |
|
| 728 |
static void save_index_subquery_explain_info(JOIN_TAB *join_tab, Item* where) |
| 729 |
{ |
| 730 |
join_tab->packed_info= TAB_INFO_HAVE_VALUE; |
| 731 |
if (join_tab->table->used_keys.is_set(join_tab->ref.key)) |
| 732 |
join_tab->packed_info |= TAB_INFO_USING_INDEX; |
| 733 |
if (where) |
| 734 |
join_tab->packed_info |= TAB_INFO_USING_WHERE; |
| 735 |
for (uint i = 0; i < join_tab->ref.key_parts; i++) |
| 736 |
{ |
| 737 |
if (join_tab->ref.cond_guards[i]) |
| 738 |
{ |
| 739 |
join_tab->packed_info |= TAB_INFO_FULL_SCAN_ON_NULL; |
| 740 |
break; |
| 741 |
} |
| 742 |
} |
| 743 |
} |
| 744 |
|
| 745 |
|
| 746 |
/* |
| 747 |
global select optimisation. |
| 748 |
return 0 - success |
| 749 |
1 - error |
| 750 |
error code saved in field 'error' |
| 751 |
*/ |
| 752 |
|
| 753 |
int |
| 754 |
JOIN::optimize() |
| 755 |
{ |
| 756 |
DBUG_ENTER("JOIN::optimize"); |
| 757 |
// to prevent double initialization on EXPLAIN |
| 758 |
if (optimized) |
| 759 |
DBUG_RETURN(0); |
| 760 |
optimized= 1; |
| 761 |
|
| 762 |
if (thd->lex->orig_sql_command != SQLCOM_SHOW_STATUS) |
| 763 |
thd->status_var.last_query_cost= 0.0; |
| 764 |
|
| 765 |
thd_proc_info(thd, "optimizing"); |
| 766 |
row_limit= ((select_distinct || order || group_list) ? HA_POS_ERROR : |
| 767 |
unit->select_limit_cnt); |
| 768 |
/* select_limit is used to decide if we are likely to scan the whole table */ |
| 769 |
select_limit= unit->select_limit_cnt; |
| 770 |
if (having || (select_options & OPTION_FOUND_ROWS)) |
| 771 |
select_limit= HA_POS_ERROR; |
| 772 |
do_send_rows = (unit->select_limit_cnt) ? 1 : 0; |
| 773 |
// Ignore errors of execution if option IGNORE present |
| 774 |
if (thd->lex->ignore) |
| 775 |
thd->lex->current_select->no_error= 1; |
| 776 |
#ifdef HAVE_REF_TO_FIELDS // Not done yet |
| 777 |
/* Add HAVING to WHERE if possible */ |
| 778 |
if (having && !group_list && !sum_func_count) |
| 779 |
{ |
| 780 |
if (!conds) |
| 781 |
{ |
| 782 |
conds= having; |
| 783 |
having= 0; |
| 784 |
} |
| 785 |
else if ((conds=new Item_cond_and(conds,having))) |
| 786 |
{ |
| 787 |
/* |
| 788 |
Item_cond_and can't be fixed after creation, so we do not check |
| 789 |
conds->fixed |
| 790 |
*/ |
| 791 |
conds->fix_fields(thd, &conds); |
| 792 |
conds->change_ref_to_fields(thd, tables_list); |
| 793 |
conds->top_level_item(); |
| 794 |
having= 0; |
| 795 |
} |
| 796 |
} |
| 797 |
#endif |
| 798 |
SELECT_LEX *sel= thd->lex->current_select; |
| 799 |
if (sel->first_cond_optimization) |
| 800 |
{ |
| 801 |
/* |
| 802 |
The following code will allocate the new items in a permanent |
| 803 |
MEMROOT for prepared statements and stored procedures. |
| 804 |
*/ |
| 805 |
|
| 806 |
Query_arena *arena= thd->stmt_arena, backup; |
| 807 |
if (arena->is_conventional()) |
| 808 |
arena= 0; // For easier test |
| 809 |
else |
| 810 |
thd->set_n_backup_active_arena(arena, &backup); |
| 811 |
|
| 812 |
sel->first_cond_optimization= 0; |
| 813 |
|
| 814 |
/* Convert all outer joins to inner joins if possible */ |
| 815 |
conds= simplify_joins(this, join_list, conds, TRUE); |
| 816 |
build_bitmap_for_nested_joins(join_list, 0); |
| 817 |
|
| 818 |
sel->prep_where= conds ? conds->copy_andor_structure(thd) : 0; |
| 819 |
|
| 820 |
if (arena) |
| 821 |
thd->restore_active_arena(arena, &backup); |
| 822 |
} |
| 823 |
|
| 824 |
conds= optimize_cond(this, conds, join_list, &cond_value); |
| 825 |
if (thd->net.report_error) |
| 826 |
{ |
| 827 |
error= 1; |
| 828 |
DBUG_PRINT("error",("Error from optimize_cond")); |
| 829 |
DBUG_RETURN(1); |
| 830 |
} |
| 831 |
|
| 832 |
{ |
| 833 |
having= optimize_cond(this, having, join_list, &having_value); |
| 834 |
if (thd->net.report_error) |
| 835 |
{ |
| 836 |
error= 1; |
| 837 |
DBUG_PRINT("error",("Error from optimize_cond")); |
| 838 |
DBUG_RETURN(1); |
| 839 |
} |
| 840 |
if (select_lex->where) |
| 841 |
select_lex->cond_value= cond_value; |
| 842 |
if (select_lex->having) |
| 843 |
select_lex->having_value= having_value; |
| 844 |
|
| 845 |
if (cond_value == Item::COND_FALSE || having_value == Item::COND_FALSE || |
| 846 |
(!unit->select_limit_cnt && !(select_options & OPTION_FOUND_ROWS))) |
| 847 |
{ /* Impossible cond */ |
| 848 |
DBUG_PRINT("info", (having_value == Item::COND_FALSE ? |
| 849 |
"Impossible HAVING" : "Impossible WHERE")); |
| 850 |
zero_result_cause= having_value == Item::COND_FALSE ? |
| 851 |
"Impossible HAVING" : "Impossible WHERE"; |
| 852 |
error= 0; |
| 853 |
DBUG_RETURN(0); |
| 854 |
} |
| 855 |
} |
| 856 |
|
| 857 |
/* Optimize count(*), min() and max() */ |
| 858 |
if (tables_list && tmp_table_param.sum_func_count && ! group_list) |
| 859 |
{ |
| 860 |
int res; |
| 861 |
/* |
| 862 |
opt_sum_query() returns HA_ERR_KEY_NOT_FOUND if no rows match |
| 863 |
to the WHERE conditions, |
| 864 |
or 1 if all items were resolved, |
| 865 |
or 0, or an error number HA_ERR_... |
| 866 |
*/ |
| 867 |
if ((res=opt_sum_query(select_lex->leaf_tables, all_fields, conds))) |
| 868 |
{ |
| 869 |
if (res == HA_ERR_KEY_NOT_FOUND) |
| 870 |
{ |
| 871 |
DBUG_PRINT("info",("No matching min/max row")); |
| 872 |
zero_result_cause= "No matching min/max row"; |
| 873 |
error=0; |
| 874 |
DBUG_RETURN(0); |
| 875 |
} |
| 876 |
if (res > 1) |
| 877 |
{ |
| 878 |
thd->fatal_error(); |
| 879 |
error= res; |
| 880 |
DBUG_PRINT("error",("Error from opt_sum_query")); |
| 881 |
DBUG_RETURN(1); |
| 882 |
} |
| 883 |
if (res < 0) |
| 884 |
{ |
| 885 |
DBUG_PRINT("info",("No matching min/max row")); |
| 886 |
zero_result_cause= "No matching min/max row"; |
| 887 |
error=0; |
| 888 |
DBUG_RETURN(0); |
| 889 |
} |
| 890 |
DBUG_PRINT("info",("Select tables optimized away")); |
| 891 |
zero_result_cause= "Select tables optimized away"; |
| 892 |
tables_list= 0; // All tables resolved |
| 893 |
/* |
| 894 |
Extract all table-independent conditions and replace the WHERE |
| 895 |
clause with them. All other conditions were computed by opt_sum_query |
| 896 |
and the MIN/MAX/COUNT function(s) have been replaced by constants, |
| 897 |
so there is no need to compute the whole WHERE clause again. |
| 898 |
Notice that make_cond_for_table() will always succeed to remove all |
| 899 |
computed conditions, because opt_sum_query() is applicable only to |
| 900 |
conjunctions. |
| 901 |
Preserve conditions for EXPLAIN. |
| 902 |
*/ |
| 903 |
if (conds && !(thd->lex->describe & DESCRIBE_EXTENDED)) |
| 904 |
{ |
| 905 |
COND *table_independent_conds= |
| 906 |
make_cond_for_table(conds, PSEUDO_TABLE_BITS, 0); |
| 907 |
DBUG_EXECUTE("where", |
| 908 |
print_where(table_independent_conds, |
| 909 |
"where after opt_sum_query()");); |
| 910 |
conds= table_independent_conds; |
| 911 |
} |
| 912 |
} |
| 913 |
} |
| 914 |
if (!tables_list) |
| 915 |
{ |
| 916 |
DBUG_PRINT("info",("No tables")); |
| 917 |
error= 0; |
| 918 |
DBUG_RETURN(0); |
| 919 |
} |
| 920 |
error= -1; // Error is sent to client |
| 921 |
sort_by_table= get_sort_by_table(order, group_list, select_lex->leaf_tables); |
| 922 |
|
| 923 |
/* Calculate how to do the join */ |
| 924 |
thd_proc_info(thd, "statistics"); |
| 925 |
if (make_join_statistics(this, select_lex->leaf_tables, conds, &keyuse) || |
| 926 |
thd->is_fatal_error) |
| 927 |
{ |
| 928 |
DBUG_PRINT("error",("Error: make_join_statistics() failed")); |
| 929 |
DBUG_RETURN(1); |
| 930 |
} |
| 931 |
|
| 932 |
/* Remove distinct if only const tables */ |
| 933 |
select_distinct= select_distinct && (const_tables != tables); |
| 934 |
thd_proc_info(thd, "preparing"); |
| 935 |
if (result->initialize_tables(this)) |
| 936 |
{ |
| 937 |
DBUG_PRINT("error",("Error: initialize_tables() failed")); |
| 938 |
DBUG_RETURN(1); // error == -1 |
| 939 |
} |
| 940 |
if (const_table_map != found_const_table_map && |
| 941 |
!(select_options & SELECT_DESCRIBE) && |
| 942 |
(!conds || |
| 943 |
!(conds->used_tables() & RAND_TABLE_BIT) || |
| 944 |
select_lex->master_unit() == &thd->lex->unit)) // upper level SELECT |
| 945 |
{ |
| 946 |
zero_result_cause= "no matching row in const table"; |
| 947 |
DBUG_PRINT("error",("Error: %s", zero_result_cause)); |
| 948 |
error= 0; |
| 949 |
DBUG_RETURN(0); |
| 950 |
} |
| 951 |
if (!(thd->options & OPTION_BIG_SELECTS) && |
| 952 |
best_read > (double) thd->variables.max_join_size && |
| 953 |
!(select_options & SELECT_DESCRIBE)) |
| 954 |
{ /* purecov: inspected */ |
| 955 |
my_message(ER_TOO_BIG_SELECT, ER(ER_TOO_BIG_SELECT), MYF(0)); |
| 956 |
error= -1; |
| 957 |
DBUG_RETURN(1); |
| 958 |
} |
| 959 |
if (const_tables && !thd->locked_tables && |
| 960 |
!(select_options & SELECT_NO_UNLOCK)) |
| 961 |
mysql_unlock_some_tables(thd, table, const_tables); |
| 962 |
if (!conds && outer_join) |
| 963 |
{ |
| 964 |
/* Handle the case where we have an OUTER JOIN without a WHERE */ |
| 965 |
conds=new Item_int((longlong) 1,1); // Always true |
| 966 |
} |
| 967 |
select= make_select(*table, const_table_map, |
| 968 |
const_table_map, conds, 1, &error); |
| 969 |
if (error) |
| 970 |
{ /* purecov: inspected */ |
| 971 |
error= -1; /* purecov: inspected */ |
| 972 |
DBUG_PRINT("error",("Error: make_select() failed")); |
| 973 |
DBUG_RETURN(1); |
| 974 |
} |
| 975 |
|
| 976 |
reset_nj_counters(join_list); |
| 977 |
make_outerjoin_info(this); |
| 978 |
|
| 979 |
/* |
| 980 |
Among the equal fields belonging to the same multiple equality |
| 981 |
choose the one that is to be retrieved first and substitute |
| 982 |
all references to these in where condition for a reference for |
| 983 |
the selected field. |
| 984 |
*/ |
| 985 |
if (conds) |
| 986 |
{ |
| 987 |
conds= substitute_for_best_equal_field(conds, cond_equal, map2table); |
| 988 |
conds->update_used_tables(); |
| 989 |
DBUG_EXECUTE("where", print_where(conds, "after substitute_best_equal");); |
| 990 |
} |
| 991 |
|
| 992 |
/* |
| 993 |
Permorm the the optimization on fields evaluation mentioned above |
| 994 |
for all on expressions. |
| 995 |
*/ |
| 996 |
for (JOIN_TAB *tab= join_tab + const_tables; tab < join_tab + tables ; tab++) |
| 997 |
{ |
| 998 |
if (*tab->on_expr_ref) |
| 999 |
{ |
| 1000 |
*tab->on_expr_ref= substitute_for_best_equal_field(*tab->on_expr_ref, |
| 1001 |
tab->cond_equal, |
| 1002 |
map2table); |
| 1003 |
(*tab->on_expr_ref)->update_used_tables(); |
| 1004 |
} |
| 1005 |
} |
| 1006 |
|
| 1007 |
if (conds &&!outer_join && const_table_map != found_const_table_map && |
| 1008 |
(select_options & SELECT_DESCRIBE) && |
| 1009 |
select_lex->master_unit() == &thd->lex->unit) // upper level SELECT |
| 1010 |
{ |
| 1011 |
conds=new Item_int((longlong) 0,1); // Always false |
| 1012 |
} |
| 1013 |
if (make_join_select(this, select, conds)) |
| 1014 |
{ |
| 1015 |
zero_result_cause= |
| 1016 |
"Impossible WHERE noticed after reading const tables"; |
| 1017 |
DBUG_RETURN(0); // error == 0 |
| 1018 |
} |
| 1019 |
|
| 1020 |
error= -1; /* if goto err */ |
| 1021 |
|
| 1022 |
/* Optimize distinct away if possible */ |
| 1023 |
{ |
| 1024 |
ORDER *org_order= order; |
| 1025 |
order=remove_const(this, order,conds,1, &simple_order); |
| 1026 |
if (thd->net.report_error) |
| 1027 |
{ |
| 1028 |
error= 1; |
| 1029 |
DBUG_PRINT("error",("Error from remove_const")); |
| 1030 |
DBUG_RETURN(1); |
| 1031 |
} |
| 1032 |
|
| 1033 |
/* |
| 1034 |
If we are using ORDER BY NULL or ORDER BY const_expression, |
| 1035 |
return result in any order (even if we are using a GROUP BY) |
| 1036 |
*/ |
| 1037 |
if (!order && org_order) |
| 1038 |
skip_sort_order= 1; |
| 1039 |
} |
| 1040 |
/* |
| 1041 |
Check if we can optimize away GROUP BY/DISTINCT. |
| 1042 |
We can do that if there are no aggregate functions, the |
| 1043 |
fields in DISTINCT clause (if present) and/or columns in GROUP BY |
| 1044 |
(if present) contain direct references to all key parts of |
| 1045 |
an unique index (in whatever order) and if the key parts of the |
| 1046 |
unique index cannot contain NULLs. |
| 1047 |
Note that the unique keys for DISTINCT and GROUP BY should not |
| 1048 |
be the same (as long as they are unique). |
| 1049 |
|
| 1050 |
The FROM clause must contain a single non-constant table. |
| 1051 |
*/ |
| 1052 |
if (tables - const_tables == 1 && (group_list || select_distinct) && |
| 1053 |
!tmp_table_param.sum_func_count && |
| 1054 |
(!join_tab[const_tables].select || |
| 1055 |
!join_tab[const_tables].select->quick || |
| 1056 |
join_tab[const_tables].select->quick->get_type() != |
| 1057 |
QUICK_SELECT_I::QS_TYPE_GROUP_MIN_MAX)) |
| 1058 |
{ |
| 1059 |
if (group_list && |
| 1060 |
list_contains_unique_index(join_tab[const_tables].table, |
| 1061 |
find_field_in_order_list, |
| 1062 |
(void *) group_list)) |
| 1063 |
{ |
| 1064 |
/* |
| 1065 |
We have found that grouping can be removed since groups correspond to |
| 1066 |
only one row anyway, but we still have to guarantee correct result |
| 1067 |
order. The line below effectively rewrites the query from GROUP BY |
| 1068 |
<fields> to ORDER BY <fields>. One exception is if skip_sort_order is |
| 1069 |
set (see above), then we can simply skip GROUP BY. |
| 1070 |
*/ |
| 1071 |
order= skip_sort_order ? 0 : group_list; |
| 1072 |
group_list= 0; |
| 1073 |
group= 0; |
| 1074 |
} |
| 1075 |
if (select_distinct && |
| 1076 |
list_contains_unique_index(join_tab[const_tables].table, |
| 1077 |
find_field_in_item_list, |
| 1078 |
(void *) &fields_list)) |
| 1079 |
{ |
| 1080 |
select_distinct= 0; |
| 1081 |
} |
| 1082 |
} |
| 1083 |
if (group_list || tmp_table_param.sum_func_count) |
| 1084 |
{ |
| 1085 |
if (! hidden_group_fields && rollup.state == ROLLUP::STATE_NONE) |
| 1086 |
select_distinct=0; |
| 1087 |
} |
| 1088 |
else if (select_distinct && tables - const_tables == 1) |
| 1089 |
{ |
| 1090 |
/* |
| 1091 |
We are only using one table. In this case we change DISTINCT to a |
| 1092 |
GROUP BY query if: |
| 1093 |
- The GROUP BY can be done through indexes (no sort) and the ORDER |
| 1094 |
BY only uses selected fields. |
| 1095 |
(In this case we can later optimize away GROUP BY and ORDER BY) |
| 1096 |
- We are scanning the whole table without LIMIT |
| 1097 |
This can happen if: |
| 1098 |
- We are using CALC_FOUND_ROWS |
| 1099 |
- We are using an ORDER BY that can't be optimized away. |
| 1100 |
|
| 1101 |
We don't want to use this optimization when we are using LIMIT |
| 1102 |
because in this case we can just create a temporary table that |
| 1103 |
holds LIMIT rows and stop when this table is full. |
| 1104 |
*/ |
| 1105 |
JOIN_TAB *tab= &join_tab[const_tables]; |
| 1106 |
bool all_order_fields_used; |
| 1107 |
if (order) |
| 1108 |
skip_sort_order= test_if_skip_sort_order(tab, order, select_limit, 1); |
| 1109 |
if ((group_list=create_distinct_group(thd, select_lex->ref_pointer_array, |
| 1110 |
order, fields_list, all_fields, |
| 1111 |
&all_order_fields_used))) |
| 1112 |
{ |
| 1113 |
bool skip_group= (skip_sort_order && |
| 1114 |
test_if_skip_sort_order(tab, group_list, select_limit, |
| 1115 |
1) != 0); |
| 1116 |
count_field_types(select_lex, &tmp_table_param, all_fields, 0); |
| 1117 |
if ((skip_group && all_order_fields_used) || |
| 1118 |
select_limit == HA_POS_ERROR || |
| 1119 |
(order && !skip_sort_order)) |
| 1120 |
{ |
| 1121 |
/* Change DISTINCT to GROUP BY */ |
| 1122 |
select_distinct= 0; |
| 1123 |
no_order= !order; |
| 1124 |
if (all_order_fields_used) |
| 1125 |
{ |
| 1126 |
if (order && skip_sort_order) |
| 1127 |
{ |
| 1128 |
/* |
| 1129 |
Force MySQL to read the table in sorted order to get result in |
| 1130 |
ORDER BY order. |
| 1131 |
*/ |
| 1132 |
tmp_table_param.quick_group=0; |
| 1133 |
} |
| 1134 |
order=0; |
| 1135 |
} |
| 1136 |
group=1; // For end_write_group |
| 1137 |
} |
| 1138 |
else |
| 1139 |
group_list= 0; |
| 1140 |
} |
| 1141 |
else if (thd->is_fatal_error) // End of memory |
| 1142 |
DBUG_RETURN(1); |
| 1143 |
} |
| 1144 |
simple_group= 0; |
| 1145 |
{ |
| 1146 |
ORDER *old_group_list; |
| 1147 |
group_list= remove_const(this, (old_group_list= group_list), conds, |
| 1148 |
rollup.state == ROLLUP::STATE_NONE, |
| 1149 |
&simple_group); |
| 1150 |
if (thd->net.report_error) |
| 1151 |
{ |
| 1152 |
error= 1; |
| 1153 |
DBUG_PRINT("error",("Error from remove_const")); |
| 1154 |
DBUG_RETURN(1); |
| 1155 |
} |
| 1156 |
if (old_group_list && !group_list) |
| 1157 |
select_distinct= 0; |
| 1158 |
} |
| 1159 |
if (!group_list && group) |
| 1160 |
{ |
| 1161 |
order=0; // The output has only one row |
| 1162 |
simple_order=1; |
| 1163 |
select_distinct= 0; // No need in distinct for 1 row |
| 1164 |
group_optimized_away= 1; |
| 1165 |
} |
| 1166 |
|
| 1167 |
calc_group_buffer(this, group_list); |
| 1168 |
send_group_parts= tmp_table_param.group_parts; /* Save org parts */ |
| 1169 |
if (procedure && procedure->group) |
| 1170 |
{ |
| 1171 |
group_list= procedure->group= remove_const(this, procedure->group, conds, |
| 1172 |
1, &simple_group); |
| 1173 |
if (thd->net.report_error) |
| 1174 |
{ |
| 1175 |
error= 1; |
| 1176 |
DBUG_PRINT("error",("Error from remove_const")); |
| 1177 |
DBUG_RETURN(1); |
| 1178 |
} |
| 1179 |
calc_group_buffer(this, group_list); |
| 1180 |
} |
| 1181 |
|
| 1182 |
if (test_if_subpart(group_list, order) || |
| 1183 |
(!group_list && tmp_table_param.sum_func_count)) |
| 1184 |
order=0; |
| 1185 |
|
| 1186 |
// Can't use sort on head table if using row cache |
| 1187 |
if (full_join) |
| 1188 |
{ |
| 1189 |
if (group_list) |
| 1190 |
simple_group=0; |
| 1191 |
if (order) |
| 1192 |
simple_order=0; |
| 1193 |
} |
| 1194 |
|
| 1195 |
/* |
| 1196 |
Check if we need to create a temporary table. |
| 1197 |
This has to be done if all tables are not already read (const tables) |
| 1198 |
and one of the following conditions holds: |
| 1199 |
- We are using DISTINCT (simple distinct's are already optimized away) |
| 1200 |
- We are using an ORDER BY or GROUP BY on fields not in the first table |
| 1201 |
- We are using different ORDER BY and GROUP BY orders |
| 1202 |
- The user wants us to buffer the result. |
| 1203 |
*/ |
| 1204 |
need_tmp= (const_tables != tables && |
| 1205 |
((select_distinct || !simple_order || !simple_group) || |
| 1206 |
(group_list && order) || |
| 1207 |
test(select_options & OPTION_BUFFER_RESULT))); |
| 1208 |
|
| 1209 |
// No cache for MATCH |
| 1210 |
make_join_readinfo(this, |
| 1211 |
(select_options & (SELECT_DESCRIBE | |
| 1212 |
SELECT_NO_JOIN_CACHE)) | |
| 1213 |
(select_lex->ftfunc_list->elements ? |
| 1214 |
SELECT_NO_JOIN_CACHE : 0)); |
| 1215 |
|
| 1216 |
/* Perform FULLTEXT search before all regular searches */ |
| 1217 |
if (!(select_options & SELECT_DESCRIBE)) |
| 1218 |
init_ftfuncs(thd, select_lex, test(order)); |
| 1219 |
|
| 1220 |
/* |
| 1221 |
is this simple IN subquery? |
| 1222 |
*/ |
| 1223 |
if (!group_list && !order && |
| 1224 |
unit->item && unit->item->substype() == Item_subselect::IN_SUBS && |
| 1225 |
tables == 1 && conds && |
| 1226 |
!unit->first_select()->next_select()) |
| 1227 |
{ |
| 1228 |
if (!having) |
| 1229 |
{ |
| 1230 |
Item *where= 0; |
| 1231 |
if (join_tab[0].type == JT_EQ_REF && |
| 1232 |
join_tab[0].ref.items[0]->name == in_left_expr_name) |
| 1233 |
{ |
| 1234 |
remove_subq_pushed_predicates(&where); |
| 1235 |
save_index_subquery_explain_info(join_tab, where); |
| 1236 |
join_tab[0].type= JT_UNIQUE_SUBQUERY; |
| 1237 |
error= 0; |
| 1238 |
DBUG_RETURN(unit->item-> |
| 1239 |
change_engine(new |
| 1240 |
subselect_uniquesubquery_engine(thd, |
| 1241 |
join_tab, |
| 1242 |
unit->item, |
| 1243 |
where))); |
| 1244 |
} |
| 1245 |
else if (join_tab[0].type == JT_REF && |
| 1246 |
join_tab[0].ref.items[0]->name == in_left_expr_name) |
| 1247 |
{ |
| 1248 |
remove_subq_pushed_predicates(&where); |
| 1249 |
save_index_subquery_explain_info(join_tab, where); |
| 1250 |
join_tab[0].type= JT_INDEX_SUBQUERY; |
| 1251 |
error= 0; |
| 1252 |
DBUG_RETURN(unit->item-> |
| 1253 |
change_engine(new |
| 1254 |
subselect_indexsubquery_engine(thd, |
| 1255 |
join_tab, |
| 1256 |
unit->item, |
| 1257 |
where, |
| 1258 |
NULL, |
| 1259 |
0))); |
| 1260 |
} |
| 1261 |
} else if (join_tab[0].type == JT_REF_OR_NULL && |
| 1262 |
join_tab[0].ref.items[0]->name == in_left_expr_name && |
| 1263 |
having->name == in_having_cond) |
| 1264 |
{ |
| 1265 |
join_tab[0].type= JT_INDEX_SUBQUERY; |
| 1266 |
error= 0; |
| 1267 |
conds= remove_additional_cond(conds); |
| 1268 |
save_index_subquery_explain_info(join_tab, conds); |
| 1269 |
DBUG_RETURN(unit->item-> |
| 1270 |
change_engine(new subselect_indexsubquery_engine(thd, |
| 1271 |
join_tab, |
| 1272 |
unit->item, |
| 1273 |
conds, |
| 1274 |
having, |
| 1275 |
1))); |
| 1276 |
} |
| 1277 |
|
| 1278 |
} |
| 1279 |
/* |
| 1280 |
Need to tell Innobase that to play it safe, it should fetch all |
| 1281 |
columns of the tables: this is because MySQL may build row |
| 1282 |
pointers for the rows, and for all columns of the primary key the |
| 1283 |
field->query_id has not necessarily been set to thd->query_id by |
| 1284 |
MySQL. |
| 1285 |
*/ |
| 1286 |
|
| 1287 |
#ifdef HAVE_INNOBASE_DB |
| 1288 |
if (need_tmp || select_distinct || group_list || order) |
| 1289 |
{ |
| 1290 |
for (uint i_h = const_tables; i_h < tables; i_h++) |
| 1291 |
{ |
| 1292 |
TABLE* table_h = join_tab[i_h].table; |
| 1293 |
table_h->file->extra(HA_EXTRA_RETRIEVE_PRIMARY_KEY); |
| 1294 |
} |
| 1295 |
} |
| 1296 |
#endif |
| 1297 |
|
| 1298 |
DBUG_EXECUTE("info",TEST_join(this);); |
| 1299 |
|
| 1300 |
if (const_tables != tables) |
| 1301 |
{ |
| 1302 |
/* |
| 1303 |
Because filesort always does a full table scan or a quick range scan |
| 1304 |
we must add the removed reference to the select for the table. |
| 1305 |
We only need to do this when we have a simple_order or simple_group |
| 1306 |
as in other cases the join is done before the sort. |
| 1307 |
*/ |
| 1308 |
if ((order || group_list) && |
| 1309 |
join_tab[const_tables].type != JT_ALL && |
| 1310 |
join_tab[const_tables].type != JT_FT && |
| 1311 |
join_tab[const_tables].type != JT_REF_OR_NULL && |
| 1312 |
(order && simple_order || group_list && simple_group)) |
| 1313 |
{ |
| 1314 |
if (add_ref_to_table_cond(thd,&join_tab[const_tables])) { |
| 1315 |
DBUG_RETURN(1); |
| 1316 |
} |
| 1317 |
} |
| 1318 |
|
| 1319 |
if (!(select_options & SELECT_BIG_RESULT) && |
| 1320 |
((group_list && |
| 1321 |
(!simple_group || |
| 1322 |
!test_if_skip_sort_order(&join_tab[const_tables], group_list, |
| 1323 |
unit->select_limit_cnt, 0))) || |
| 1324 |
select_distinct) && |
| 1325 |
tmp_table_param.quick_group && !procedure) |
| 1326 |
{ |
| 1327 |
need_tmp=1; simple_order=simple_group=0; // Force tmp table without sort |
| 1328 |
} |
| 1329 |
if (order) |
| 1330 |
{ |
| 1331 |
/* |
| 1332 |
Force using of tmp table if sorting by a SP or UDF function due to |
| 1333 |
their expensive and probably non-deterministic nature. |
| 1334 |
*/ |
| 1335 |
for (ORDER *tmp_order= order; tmp_order ; tmp_order=tmp_order->next) |
| 1336 |
{ |
| 1337 |
Item *item= *tmp_order->item; |
| 1338 |
if (item->walk(&Item::is_expensive_processor,(byte*)0)) |
| 1339 |
{ |
| 1340 |
/* Force tmp table without sort */ |
| 1341 |
need_tmp=1; simple_order=simple_group=0; |
| 1342 |
break; |
| 1343 |
} |
| 1344 |
} |
| 1345 |
} |
| 1346 |
} |
| 1347 |
|
| 1348 |
tmp_having= having; |
| 1349 |
if (select_options & SELECT_DESCRIBE) |
| 1350 |
{ |
| 1351 |
error= 0; |
| 1352 |
DBUG_RETURN(0); |
| 1353 |
} |
| 1354 |
having= 0; |
| 1355 |
|
| 1356 |
/* |
| 1357 |
The loose index scan access method guarantees that all grouping or |
| 1358 |
duplicate row elimination (for distinct) is already performed |
| 1359 |
during data retrieval, and that all MIN/MAX functions are already |
| 1360 |
computed for each group. Thus all MIN/MAX functions should be |
| 1361 |
treated as regular functions, and there is no need to perform |
| 1362 |
grouping in the main execution loop. |
| 1363 |
Notice that currently loose index scan is applicable only for |
| 1364 |
single table queries, thus it is sufficient to test only the first |
| 1365 |
join_tab element of the plan for its access method. |
| 1366 |
*/ |
| 1367 |
if (join_tab->is_using_loose_index_scan()) |
| 1368 |
tmp_table_param.precomputed_group_by= TRUE; |
| 1369 |
|
| 1370 |
/* Create a tmp table if distinct or if the sort is too complicated */ |
| 1371 |
if (need_tmp) |
| 1372 |
{ |
| 1373 |
DBUG_PRINT("info",("Creating tmp table")); |
| 1374 |
thd_proc_info(thd, "Creating tmp table"); |
| 1375 |
|
| 1376 |
init_items_ref_array(); |
| 1377 |
|
| 1378 |
tmp_table_param.hidden_field_count= (all_fields.elements - |
| 1379 |
fields_list.elements); |
| 1380 |
ORDER *tmp_group= ((!simple_group && !procedure && |
| 1381 |
!(test_flags & TEST_NO_KEY_GROUP)) ? group_list : |
| 1382 |
(ORDER*) 0); |
| 1383 |
/* |
| 1384 |
Pushing LIMIT to the temporary table creation is not applicable |
| 1385 |
when there is ORDER BY or GROUP BY or there is no GROUP BY, but |
| 1386 |
there are aggregate functions, because in all these cases we need |
| 1387 |
all result rows. |
| 1388 |
*/ |
| 1389 |
ha_rows tmp_rows_limit= ((order == 0 || skip_sort_order) && |
| 1390 |
!tmp_group && |
| 1391 |
!thd->lex->current_select->with_sum_func) ? |
| 1392 |
select_limit : HA_POS_ERROR; |
| 1393 |
|
| 1394 |
if (!(exec_tmp_table1 = |
| 1395 |
create_tmp_table(thd, &tmp_table_param, all_fields, |
| 1396 |
tmp_group, |
| 1397 |
group_list ? 0 : select_distinct, |
| 1398 |
group_list && simple_group, |
| 1399 |
select_options, |
| 1400 |
tmp_rows_limit, |
| 1401 |
(char *) ""))) |
| 1402 |
{ |
| 1403 |
DBUG_RETURN(1); |
| 1404 |
} |
| 1405 |
|
| 1406 |
/* |
| 1407 |
We don't have to store rows in temp table that doesn't match HAVING if: |
| 1408 |
- we are sorting the table and writing complete group rows to the |
| 1409 |
temp table. |
| 1410 |
- We are using DISTINCT without resolving the distinct as a GROUP BY |
| 1411 |
on all columns. |
| 1412 |
|
| 1413 |
If having is not handled here, it will be checked before the row |
| 1414 |
is sent to the client. |
| 1415 |
*/ |
| 1416 |
if (tmp_having && |
| 1417 |
(sort_and_group || (exec_tmp_table1->distinct && !group_list))) |
| 1418 |
having= tmp_having; |
| 1419 |
|
| 1420 |
/* if group or order on first table, sort first */ |
| 1421 |
if (group_list && simple_group) |
| 1422 |
{ |
| 1423 |
DBUG_PRINT("info",("Sorting for group")); |
| 1424 |
thd_proc_info(thd, "Sorting for group"); |
| 1425 |
if (create_sort_index(thd, this, group_list, |
| 1426 |
HA_POS_ERROR, HA_POS_ERROR) || |
| 1427 |
alloc_group_fields(this, group_list) || |
| 1428 |
make_sum_func_list(all_fields, fields_list, 1) || |
| 1429 |
setup_sum_funcs(thd, sum_funcs)) |
| 1430 |
{ |
| 1431 |
DBUG_RETURN(1); |
| 1432 |
} |
| 1433 |
group_list=0; |
| 1434 |
} |
| 1435 |
else |
| 1436 |
{ |
| 1437 |
if (make_sum_func_list(all_fields, fields_list, 0) || |
| 1438 |
setup_sum_funcs(thd, sum_funcs)) |
| 1439 |
{ |
| 1440 |
DBUG_RETURN(1); |
| 1441 |
} |
| 1442 |
|
| 1443 |
if (!group_list && ! exec_tmp_table1->distinct && order && simple_order) |
| 1444 |
{ |
| 1445 |
thd_proc_info(thd, "Sorting for order"); |
| 1446 |
if (create_sort_index(thd, this, order, |
| 1447 |
HA_POS_ERROR, HA_POS_ERROR)) |
| 1448 |
{ |
| 1449 |
DBUG_RETURN(1); |
| 1450 |
} |
| 1451 |
order=0; |
| 1452 |
} |
| 1453 |
} |
| 1454 |
|
| 1455 |
/* |
| 1456 |
Optimize distinct when used on some of the tables |
| 1457 |
SELECT DISTINCT t1.a FROM t1,t2 WHERE t1.b=t2.b |
| 1458 |
In this case we can stop scanning t2 when we have found one t1.a |
| 1459 |
*/ |
| 1460 |
|
| 1461 |
if (exec_tmp_table1->distinct) |
| 1462 |
{ |
| 1463 |
table_map used_tables= thd->used_tables; |
| 1464 |
JOIN_TAB *last_join_tab= join_tab+tables-1; |
| 1465 |
do |
| 1466 |
{ |
| 1467 |
if (used_tables & last_join_tab->table->map) |
| 1468 |
break; |
| 1469 |
last_join_tab->not_used_in_distinct=1; |
| 1470 |
} while (last_join_tab-- != join_tab); |
| 1471 |
/* Optimize "select distinct b from t1 order by key_part_1 limit #" */ |
| 1472 |
if (order && skip_sort_order) |
| 1473 |
{ |
| 1474 |
/* Should always succeed */ |
| 1475 |
if (test_if_skip_sort_order(&join_tab[const_tables], |
| 1476 |
order, unit->select_limit_cnt, 0)) |
| 1477 |
order=0; |
| 1478 |
} |
| 1479 |
} |
| 1480 |
|
| 1481 |
/* |
| 1482 |
If this join belongs to an uncacheable subquery save |
| 1483 |
the original join |
| 1484 |
*/ |
| 1485 |
if (select_lex->uncacheable && !is_top_level_join() && |
| 1486 |
init_save_join_tab()) |
| 1487 |
DBUG_RETURN(-1); /* purecov: inspected */ |
| 1488 |
} |
| 1489 |
|
| 1490 |
#ifdef ENABLE_SENNA |
| 1491 |
if (my_thread_var->sen_flags & SENNA_USE_2IND) |
| 1492 |
DEBUG_2IND(my_thread_var->sen_flags |= SENNA_FIRST_CALL); |
| 1493 |
#endif /* ENABLE_SENNA */ |
| 1494 |
|
| 1495 |
error= 0; |
| 1496 |
DBUG_RETURN(0); |
| 1497 |
} |
| 1498 |
|
| 1499 |
|
| 1500 |
/* |
| 1501 |
Restore values in temporary join |
| 1502 |
*/ |
| 1503 |
void JOIN::restore_tmp() |
| 1504 |
{ |
| 1505 |
memcpy(tmp_join, this, (size_t) sizeof(JOIN)); |
| 1506 |
} |
| 1507 |
|
| 1508 |
|
| 1509 |
int |
| 1510 |
JOIN::reinit() |
| 1511 |
{ |
| 1512 |
DBUG_ENTER("JOIN::reinit"); |
| 1513 |
|
| 1514 |
unit->offset_limit_cnt= (ha_rows)(select_lex->offset_limit ? |
| 1515 |
select_lex->offset_limit->val_uint() : |
| 1516 |
ULL(0)); |
| 1517 |
|
| 1518 |
first_record= 0; |
| 1519 |
|
| 1520 |
if (exec_tmp_table1) |
| 1521 |
{ |
| 1522 |
exec_tmp_table1->file->extra(HA_EXTRA_RESET_STATE); |
| 1523 |
exec_tmp_table1->file->delete_all_rows(); |
| 1524 |
free_io_cache(exec_tmp_table1); |
| 1525 |
filesort_free_buffers(exec_tmp_table1,0); |
| 1526 |
} |
| 1527 |
if (exec_tmp_table2) |
| 1528 |
{ |
| 1529 |
exec_tmp_table2->file->extra(HA_EXTRA_RESET_STATE); |
| 1530 |
exec_tmp_table2->file->delete_all_rows(); |
| 1531 |
free_io_cache(exec_tmp_table2); |
| 1532 |
filesort_free_buffers(exec_tmp_table2,0); |
| 1533 |
} |
| 1534 |
if (items0) |
| 1535 |
set_items_ref_array(items0); |
| 1536 |
|
| 1537 |
if (join_tab_save) |
| 1538 |
memcpy(join_tab, join_tab_save, sizeof(JOIN_TAB) * tables); |
| 1539 |
|
| 1540 |
if (tmp_join) |
| 1541 |
restore_tmp(); |
| 1542 |
|
| 1543 |
/* Reset of sum functions */ |
| 1544 |
if (sum_funcs) |
| 1545 |
{ |
| 1546 |
Item_sum *func, **func_ptr= sum_funcs; |
| 1547 |
while ((func= *(func_ptr++))) |
| 1548 |
func->clear(); |
| 1549 |
} |
| 1550 |
|
| 1551 |
DBUG_RETURN(0); |
| 1552 |
} |
| 1553 |
|
| 1554 |
/** |
| 1555 |
@brief Save the original join layout |
| 1556 |
|
| 1557 |
@details Saves the original join layout so it can be reused in |
| 1558 |
re-execution and for EXPLAIN. |
| 1559 |
|
| 1560 |
@return Operation status |
| 1561 |
@retval 0 success. |
| 1562 |
@retval 1 error occurred. |
| 1563 |
*/ |
| 1564 |
|
| 1565 |
bool |
| 1566 |
JOIN::init_save_join_tab() |
| 1567 |
{ |
| 1568 |
if (!(tmp_join= (JOIN*)thd->alloc(sizeof(JOIN)))) |
| 1569 |
return 1; /* purecov: inspected */ |
| 1570 |
error= 0; // Ensure that tmp_join.error= 0 |
| 1571 |
restore_tmp(); |
| 1572 |
return 0; |
| 1573 |
} |
| 1574 |
|
| 1575 |
|
| 1576 |
bool |
| 1577 |
JOIN::save_join_tab() |
| 1578 |
{ |
| 1579 |
if (!join_tab_save && select_lex->master_unit()->uncacheable) |
| 1580 |
{ |
| 1581 |
if (!(join_tab_save= (JOIN_TAB*)thd->memdup((gptr) join_tab, |
| 1582 |
sizeof(JOIN_TAB) * tables))) |
| 1583 |
return 1; |
| 1584 |
} |
| 1585 |
return 0; |
| 1586 |
} |
| 1587 |
|
| 1588 |
|
| 1589 |
/* |
| 1590 |
Exec select |
| 1591 |
*/ |
| 1592 |
void |
| 1593 |
JOIN::exec() |
| 1594 |
{ |
| 1595 |
List<Item> *columns_list= &fields_list; |
| 1596 |
int tmp_error; |
| 1597 |
DBUG_ENTER("JOIN::exec"); |
| 1598 |
|
| 1599 |
thd_proc_info(thd, "executing"); |
| 1600 |
error= 0; |
| 1601 |
if (procedure) |
| 1602 |
{ |
| 1603 |
procedure_fields_list= fields_list; |
| 1604 |
if (procedure->change_columns(procedure_fields_list) || |
| 1605 |
result->prepare(procedure_fields_list, unit)) |
| 1606 |
{ |
| 1607 |
thd->limit_found_rows= thd->examined_row_count= 0; |
| 1608 |
DBUG_VOID_RETURN; |
| 1609 |
} |
| 1610 |
columns_list= &procedure_fields_list; |
| 1611 |
} |
| 1612 |
(void) result->prepare2(); // Currently, this cannot fail. |
| 1613 |
|
| 1614 |
if (!tables_list && (tables || !select_lex->with_sum_func)) |
| 1615 |
{ // Only test of functions |
| 1616 |
if (select_options & SELECT_DESCRIBE) |
| 1617 |
select_describe(this, FALSE, FALSE, FALSE, |
| 1618 |
(zero_result_cause?zero_result_cause:"No tables used")); |
| 1619 |
else |
| 1620 |
{ |
| 1621 |
result->send_fields(*columns_list, |
| 1622 |
Protocol::SEND_NUM_ROWS | Protocol::SEND_EOF); |
| 1623 |
/* |
| 1624 |
We have to test for 'conds' here as the WHERE may not be constant |
| 1625 |
even if we don't have any tables for prepared statements or if |
| 1626 |
conds uses something like 'rand()'. |
| 1627 |
*/ |
| 1628 |
if (cond_value != Item::COND_FALSE && |
| 1629 |
(!conds || conds->val_int()) && |
| 1630 |
(!having || having->val_int())) |
| 1631 |
{ |
| 1632 |
if (do_send_rows && |
| 1633 |
(procedure ? (procedure->send_row(procedure_fields_list) || |
| 1634 |
procedure->end_of_records()) : result->send_data(fields_list))) |
| 1635 |
error= 1; |
| 1636 |
else |
| 1637 |
{ |
| 1638 |
error= (int) result->send_eof(); |
| 1639 |
send_records= ((select_options & OPTION_FOUND_ROWS) ? 1 : |
| 1640 |
thd->sent_row_count); |
| 1641 |
} |
| 1642 |
} |
| 1643 |
else |
| 1644 |
{ |
| 1645 |
error=(int) result->send_eof(); |
| 1646 |
send_records= 0; |
| 1647 |
} |
| 1648 |
} |
| 1649 |
/* Single select (without union) always returns 0 or 1 row */ |
| 1650 |
thd->limit_found_rows= send_records; |
| 1651 |
thd->examined_row_count= 0; |
| 1652 |
DBUG_VOID_RETURN; |
| 1653 |
} |
| 1654 |
/* |
| 1655 |
Don't reset the found rows count if there're no tables as |
| 1656 |
FOUND_ROWS() may be called. Never reset the examined row count here. |
| 1657 |
It must be accumulated from all join iterations of all join parts. |
| 1658 |
*/ |
| 1659 |
if (tables) |
| 1660 |
thd->limit_found_rows= 0; |
| 1661 |
|
| 1662 |
if (zero_result_cause) |
| 1663 |
{ |
| 1664 |
(void) return_zero_rows(this, result, select_lex->leaf_tables, |
| 1665 |
*columns_list, |
| 1666 |
send_row_on_empty_set(), |
| 1667 |
select_options, |
| 1668 |
zero_result_cause, |
| 1669 |
having); |
| 1670 |
DBUG_VOID_RETURN; |
| 1671 |
} |
| 1672 |
|
| 1673 |
if (select_options & SELECT_DESCRIBE) |
| 1674 |
{ |
| 1675 |
/* |
| 1676 |
Check if we managed to optimize ORDER BY away and don't use temporary |
| 1677 |
table to resolve ORDER BY: in that case, we only may need to do |
| 1678 |
filesort for GROUP BY. |
| 1679 |
*/ |
| 1680 |
if (!order && !no_order && (!skip_sort_order || !need_tmp)) |
| 1681 |
{ |
| 1682 |
/* |
| 1683 |
Reset 'order' to 'group_list' and reinit variables describing |
| 1684 |
'order' |
| 1685 |
*/ |
| 1686 |
order= group_list; |
| 1687 |
simple_order= simple_group; |
| 1688 |
skip_sort_order= 0; |
| 1689 |
} |
| 1690 |
if (order && |
| 1691 |
(order != group_list || !(select_options & SELECT_BIG_RESULT)) && |
| 1692 |
(const_tables == tables || |
| 1693 |
((simple_order || skip_sort_order) && |
| 1694 |
test_if_skip_sort_order(&join_tab[const_tables], order, |
| 1695 |
select_limit, 0)))) |
| 1696 |
order=0; |
| 1697 |
having= tmp_having; |
| 1698 |
select_describe(this, need_tmp, |
| 1699 |
order != 0 && !skip_sort_order, |
| 1700 |
select_distinct, |
| 1701 |
!tables ? "No tables used" : NullS); |
| 1702 |
DBUG_VOID_RETURN; |
| 1703 |
} |
| 1704 |
|
| 1705 |
JOIN *curr_join= this; |
| 1706 |
List<Item> *curr_all_fields= &all_fields; |
| 1707 |
List<Item> *curr_fields_list= &fields_list; |
| 1708 |
TABLE *curr_tmp_table= 0; |
| 1709 |
/* |
| 1710 |
Initialize examined rows here because the values from all join parts |
| 1711 |
must be accumulated in examined_row_count. Hence every join |
| 1712 |
iteration must count from zero. |
| 1713 |
*/ |
| 1714 |
curr_join->examined_rows= 0; |
| 1715 |
|
| 1716 |
if ((curr_join->select_lex->options & OPTION_SCHEMA_TABLE) && |
| 1717 |
!thd->lex->describe && |
| 1718 |
get_schema_tables_result(curr_join, PROCESSED_BY_JOIN_EXEC)) |
| 1719 |
{ |
| 1720 |
DBUG_VOID_RETURN; |
| 1721 |
} |
| 1722 |
|
| 1723 |
/* Create a tmp table if distinct or if the sort is too complicated */ |
| 1724 |
if (need_tmp) |
| 1725 |
{ |
| 1726 |
if (tmp_join) |
| 1727 |
{ |
| 1728 |
/* |
| 1729 |
We are in a non cacheable sub query. Get the saved join structure |
| 1730 |
after optimization. |
| 1731 |
(curr_join may have been modified during last exection and we need |
| 1732 |
to reset it) |
| 1733 |
*/ |
| 1734 |
curr_join= tmp_join; |
| 1735 |
} |
| 1736 |
curr_tmp_table= exec_tmp_table1; |
| 1737 |
|
| 1738 |
/* Copy data to the temporary table */ |
| 1739 |
thd_proc_info(thd, "Copying to tmp table"); |
| 1740 |
DBUG_PRINT("info", ("%s", thd->proc_info)); |
| 1741 |
if ((tmp_error= do_select(curr_join, (List<Item> *) 0, curr_tmp_table, 0))) |
| 1742 |
{ |
| 1743 |
error= tmp_error; |
| 1744 |
DBUG_VOID_RETURN; |
| 1745 |
} |
| 1746 |
curr_tmp_table->file->info(HA_STATUS_VARIABLE); |
| 1747 |
|
| 1748 |
if (curr_join->having) |
| 1749 |
curr_join->having= curr_join->tmp_having= 0; // Allready done |
| 1750 |
|
| 1751 |
/* Change sum_fields reference to calculated fields in tmp_table */ |
| 1752 |
curr_join->all_fields= *curr_all_fields; |
| 1753 |
if (!items1) |
| 1754 |
{ |
| 1755 |
items1= items0 + all_fields.elements; |
| 1756 |
if (sort_and_group || curr_tmp_table->group) |
| 1757 |
{ |
| 1758 |
if (change_to_use_tmp_fields(thd, items1, |
| 1759 |
tmp_fields_list1, tmp_all_fields1, |
| 1760 |
fields_list.elements, all_fields)) |
| 1761 |
DBUG_VOID_RETURN; |
| 1762 |
} |
| 1763 |
else |
| 1764 |
{ |
| 1765 |
if (change_refs_to_tmp_fields(thd, items1, |
| 1766 |
tmp_fields_list1, tmp_all_fields1, |
| 1767 |
fields_list.elements, all_fields)) |
| 1768 |
DBUG_VOID_RETURN; |
| 1769 |
} |
| 1770 |
curr_join->tmp_all_fields1= tmp_all_fields1; |
| 1771 |
curr_join->tmp_fields_list1= tmp_fields_list1; |
| 1772 |
curr_join->items1= items1; |
| 1773 |
} |
| 1774 |
curr_all_fields= &tmp_all_fields1; |
| 1775 |
curr_fields_list= &tmp_fields_list1; |
| 1776 |
curr_join->set_items_ref_array(items1); |
| 1777 |
|
| 1778 |
if (sort_and_group || curr_tmp_table->group) |
| 1779 |
{ |
| 1780 |
curr_join->tmp_table_param.field_count+= |
| 1781 |
curr_join->tmp_table_param.sum_func_count+ |
| 1782 |
curr_join->tmp_table_param.func_count; |
| 1783 |
curr_join->tmp_table_param.sum_func_count= |
| 1784 |
curr_join->tmp_table_param.func_count= 0; |
| 1785 |
} |
| 1786 |
else |
| 1787 |
{ |
| 1788 |
curr_join->tmp_table_param.field_count+= |
| 1789 |
curr_join->tmp_table_param.func_count; |
| 1790 |
curr_join->tmp_table_param.func_count= 0; |
| 1791 |
} |
| 1792 |
|
| 1793 |
// procedure can't be used inside subselect => we do nothing special for it |
| 1794 |
if (procedure) |
| 1795 |
procedure->update_refs(); |
| 1796 |
|
| 1797 |
if (curr_tmp_table->group) |
| 1798 |
{ // Already grouped |
| 1799 |
if (!curr_join->order && !curr_join->no_order && !skip_sort_order) |
| 1800 |
curr_join->order= curr_join->group_list; /* order by group */ |
| 1801 |
curr_join->group_list= 0; |
| 1802 |
} |
| 1803 |
|
| 1804 |
/* |
| 1805 |
If we have different sort & group then we must sort the data by group |
| 1806 |
and copy it to another tmp table |
| 1807 |
This code is also used if we are using distinct something |
| 1808 |
we haven't been able to store in the temporary table yet |
| 1809 |
like SEC_TO_TIME(SUM(...)). |
| 1810 |
*/ |
| 1811 |
|
| 1812 |
if (curr_join->group_list && (!test_if_subpart(curr_join->group_list, |
| 1813 |
curr_join->order) || |
| 1814 |
curr_join->select_distinct) || |
| 1815 |
(curr_join->select_distinct && |
| 1816 |
curr_join->tmp_table_param.using_indirect_summary_function)) |
| 1817 |
{ /* Must copy to another table */ |
| 1818 |
DBUG_PRINT("info",("Creating group table")); |
| 1819 |
|
| 1820 |
/* Free first data from old join */ |
| 1821 |
curr_join->join_free(); |
| 1822 |
if (make_simple_join(curr_join, curr_tmp_table)) |
| 1823 |
DBUG_VOID_RETURN; |
| 1824 |
calc_group_buffer(curr_join, group_list); |
| 1825 |
count_field_types(select_lex, &curr_join->tmp_table_param, |
| 1826 |
curr_join->tmp_all_fields1, |
| 1827 |
curr_join->select_distinct && !curr_join->group_list); |
| 1828 |
curr_join->tmp_table_param.hidden_field_count= |
| 1829 |
(curr_join->tmp_all_fields1.elements- |
| 1830 |
curr_join->tmp_fields_list1.elements); |
| 1831 |
|
| 1832 |
|
| 1833 |
if (exec_tmp_table2) |
| 1834 |
curr_tmp_table= exec_tmp_table2; |
| 1835 |
else |
| 1836 |
{ |
| 1837 |
/* group data to new table */ |
| 1838 |
|
| 1839 |
/* |
| 1840 |
If the access method is loose index scan then all MIN/MAX |
| 1841 |
functions are precomputed, and should be treated as regular |
| 1842 |
functions. See extended comment in JOIN::exec. |
| 1843 |
*/ |
| 1844 |
if (curr_join->join_tab->is_using_loose_index_scan()) |
| 1845 |
curr_join->tmp_table_param.precomputed_group_by= TRUE; |
| 1846 |
|
| 1847 |
if (!(curr_tmp_table= |
| 1848 |
exec_tmp_table2= create_tmp_table(thd, |
| 1849 |
&curr_join->tmp_table_param, |
| 1850 |
*curr_all_fields, |
| 1851 |
(ORDER*) 0, |
| 1852 |
curr_join->select_distinct && |
| 1853 |
!curr_join->group_list, |
| 1854 |
1, curr_join->select_options, |
| 1855 |
HA_POS_ERROR, |
| 1856 |
(char *) ""))) |
| 1857 |
DBUG_VOID_RETURN; |
| 1858 |
curr_join->exec_tmp_table2= exec_tmp_table2; |
| 1859 |
} |
| 1860 |
if (curr_join->group_list) |
| 1861 |
{ |
| 1862 |
thd_proc_info(thd, "Creating sort index"); |
| 1863 |
if (curr_join->join_tab == join_tab && save_join_tab()) |
| 1864 |
{ |
| 1865 |
DBUG_VOID_RETURN; |
| 1866 |
} |
| 1867 |
if (create_sort_index(thd, curr_join, curr_join->group_list, |
| 1868 |
HA_POS_ERROR, HA_POS_ERROR) || |
| 1869 |
make_group_fields(this, curr_join)) |
| 1870 |
{ |
| 1871 |
DBUG_VOID_RETURN; |
| 1872 |
} |
| 1873 |
sortorder= curr_join->sortorder; |
| 1874 |
} |
| 1875 |
|
| 1876 |
thd_proc_info(thd, "Copying to group table"); |
| 1877 |
DBUG_PRINT("info", ("%s", thd->proc_info)); |
| 1878 |
tmp_error= -1; |
| 1879 |
if (curr_join != this) |
| 1880 |
{ |
| 1881 |
if (sum_funcs2) |
| 1882 |
{ |
| 1883 |
curr_join->sum_funcs= sum_funcs2; |
| 1884 |
curr_join->sum_funcs_end= sum_funcs_end2; |
| 1885 |
} |
| 1886 |
else |
| 1887 |
{ |
| 1888 |
curr_join->alloc_func_list(); |
| 1889 |
sum_funcs2= curr_join->sum_funcs; |
| 1890 |
sum_funcs_end2= curr_join->sum_funcs_end; |
| 1891 |
} |
| 1892 |
} |
| 1893 |
if (curr_join->make_sum_func_list(*curr_all_fields, *curr_fields_list, |
| 1894 |
1, TRUE)) |
| 1895 |
DBUG_VOID_RETURN; |
| 1896 |
curr_join->group_list= 0; |
| 1897 |
if (setup_sum_funcs(curr_join->thd, curr_join->sum_funcs) || |
| 1898 |
(tmp_error= do_select(curr_join, (List<Item> *) 0, curr_tmp_table, |
| 1899 |
0))) |
| 1900 |
{ |
| 1901 |
error= tmp_error; |
| 1902 |
DBUG_VOID_RETURN; |
| 1903 |
} |
| 1904 |
end_read_record(&curr_join->join_tab->read_record); |
| 1905 |
curr_join->const_tables= curr_join->tables; // Mark free for cleanup() |
| 1906 |
curr_join->join_tab[0].table= 0; // Table is freed |
| 1907 |
|
| 1908 |
// No sum funcs anymore |
| 1909 |
if (!items2) |
| 1910 |
{ |
| 1911 |
items2= items1 + all_fields.elements; |
| 1912 |
if (change_to_use_tmp_fields(thd, items2, |
| 1913 |
tmp_fields_list2, tmp_all_fields2, |
| 1914 |
fields_list.elements, tmp_all_fields1)) |
| 1915 |
DBUG_VOID_RETURN; |
| 1916 |
curr_join->tmp_fields_list2= tmp_fields_list2; |
| 1917 |
curr_join->tmp_all_fields2= tmp_all_fields2; |
| 1918 |
} |
| 1919 |
curr_fields_list= &curr_join->tmp_fields_list2; |
| 1920 |
curr_all_fields= &curr_join->tmp_all_fields2; |
| 1921 |
curr_join->set_items_ref_array(items2); |
| 1922 |
curr_join->tmp_table_param.field_count+= |
| 1923 |
curr_join->tmp_table_param.sum_func_count; |
| 1924 |
curr_join->tmp_table_param.sum_func_count= 0; |
| 1925 |
} |
| 1926 |
if (curr_tmp_table->distinct) |
| 1927 |
curr_join->select_distinct=0; /* Each row is unique */ |
| 1928 |
|
| 1929 |
curr_join->join_free(); /* Free quick selects */ |
| 1930 |
if (curr_join->select_distinct && ! curr_join->group_list) |
| 1931 |
{ |
| 1932 |
thd_proc_info(thd, "Removing duplicates"); |
| 1933 |
if (curr_join->tmp_having) |
| 1934 |
curr_join->tmp_having->update_used_tables(); |
| 1935 |
if (remove_duplicates(curr_join, curr_tmp_table, |
| 1936 |
*curr_fields_list, curr_join->tmp_having)) |
| 1937 |
DBUG_VOID_RETURN; |
| 1938 |
curr_join->tmp_having=0; |
| 1939 |
curr_join->select_distinct=0; |
| 1940 |
} |
| 1941 |
curr_tmp_table->reginfo.lock_type= TL_UNLOCK; |
| 1942 |
if (make_simple_join(curr_join, curr_tmp_table)) |
| 1943 |
DBUG_VOID_RETURN; |
| 1944 |
calc_group_buffer(curr_join, curr_join->group_list); |
| 1945 |
count_field_types(select_lex, &curr_join->tmp_table_param, |
| 1946 |
*curr_all_fields, 0); |
| 1947 |
|
| 1948 |
} |
| 1949 |
if (procedure) |
| 1950 |
count_field_types(select_lex, &curr_join->tmp_table_param, |
| 1951 |
*curr_all_fields, 0); |
| 1952 |
|
| 1953 |
if (curr_join->group || curr_join->tmp_table_param.sum_func_count || |
| 1954 |
(procedure && (procedure->flags & PROC_GROUP))) |
| 1955 |
{ |
| 1956 |
if (make_group_fields(this, curr_join)) |
| 1957 |
{ |
| 1958 |
DBUG_VOID_RETURN; |
| 1959 |
} |
| 1960 |
if (!items3) |
| 1961 |
{ |
| 1962 |
if (!items0) |
| 1963 |
init_items_ref_array(); |
| 1964 |
items3= ref_pointer_array + (all_fields.elements*4); |
| 1965 |
setup_copy_fields(thd, &curr_join->tmp_table_param, |
| 1966 |
items3, tmp_fields_list3, tmp_all_fields3, |
| 1967 |
curr_fields_list->elements, *curr_all_fields); |
| 1968 |
tmp_table_param.save_copy_funcs= curr_join->tmp_table_param.copy_funcs; |
| 1969 |
tmp_table_param.save_copy_field= curr_join->tmp_table_param.copy_field; |
| 1970 |
tmp_table_param.save_copy_field_end= |
| 1971 |
curr_join->tmp_table_param.copy_field_end; |
| 1972 |
curr_join->tmp_all_fields3= tmp_all_fields3; |
| 1973 |
curr_join->tmp_fields_list3= tmp_fields_list3; |
| 1974 |
} |
| 1975 |
else |
| 1976 |
{ |
| 1977 |
curr_join->tmp_table_param.copy_funcs= tmp_table_param.save_copy_funcs; |
| 1978 |
curr_join->tmp_table_param.copy_field= tmp_table_param.save_copy_field; |
| 1979 |
curr_join->tmp_table_param.copy_field_end= |
| 1980 |
tmp_table_param.save_copy_field_end; |
| 1981 |
} |
| 1982 |
curr_fields_list= &tmp_fields_list3; |
| 1983 |
curr_all_fields= &tmp_all_fields3; |
| 1984 |
curr_join->set_items_ref_array(items3); |
| 1985 |
|
| 1986 |
if (curr_join->make_sum_func_list(*curr_all_fields, *curr_fields_list, |
| 1987 |
1, TRUE) || |
| 1988 |
setup_sum_funcs(curr_join->thd, curr_join->sum_funcs) || |
| 1989 |
thd->is_fatal_error) |
| 1990 |
DBUG_VOID_RETURN; |
| 1991 |
} |
| 1992 |
if (curr_join->group_list || curr_join->order) |
| 1993 |
{ |
| 1994 |
DBUG_PRINT("info",("Sorting for send_fields")); |
| 1995 |
thd_proc_info(thd, "Sorting result"); |
| 1996 |
/* If we have already done the group, add HAVING to sorted table */ |
| 1997 |
if (curr_join->tmp_having && ! curr_join->group_list && |
| 1998 |
! curr_join->sort_and_group) |
| 1999 |
{ |
| 2000 |
// Some tables may have been const |
| 2001 |
curr_join->tmp_having->update_used_tables(); |
| 2002 |
JOIN_TAB *curr_table= &curr_join->join_tab[curr_join->const_tables]; |
| 2003 |
table_map used_tables= (curr_join->const_table_map | |
| 2004 |
curr_table->table->map); |
| 2005 |
|
| 2006 |
Item* sort_table_cond= make_cond_for_table(curr_join->tmp_having, |
| 2007 |
used_tables, |
| 2008 |
used_tables); |
| 2009 |
if (sort_table_cond) |
| 2010 |
{ |
| 2011 |
if (!curr_table->select) |
| 2012 |
if (!(curr_table->select= new SQL_SELECT)) |
| 2013 |
DBUG_VOID_RETURN; |
| 2014 |
if (!curr_table->select->cond) |
| 2015 |
curr_table->select->cond= sort_table_cond; |
| 2016 |
else // This should never happen |
| 2017 |
{ |
| 2018 |
if (!(curr_table->select->cond= |
| 2019 |
new Item_cond_and(curr_table->select->cond, |
| 2020 |
sort_table_cond))) |
| 2021 |
DBUG_VOID_RETURN; |
| 2022 |
/* |
| 2023 |
Item_cond_and do not need fix_fields for execution, its parameters |
| 2024 |
are fixed or do not need fix_fields, too |
| 2025 |
*/ |
| 2026 |
curr_table->select->cond->quick_fix_field(); |
| 2027 |
} |
| 2028 |
curr_table->select_cond= curr_table->select->cond; |
| 2029 |
curr_table->select_cond->top_level_item(); |
| 2030 |
DBUG_EXECUTE("where",print_where(curr_table->select->cond, |
| 2031 |
"select and having");); |
| 2032 |
curr_join->tmp_having= make_cond_for_table(curr_join->tmp_having, |
| 2033 |
~ (table_map) 0, |
| 2034 |
~used_tables); |
| 2035 |
DBUG_EXECUTE("where",print_where(curr_join->tmp_having, |
| 2036 |
"having after sort");); |
| 2037 |
} |
| 2038 |
} |
| 2039 |
{ |
| 2040 |
if (group) |
| 2041 |
curr_join->select_limit= HA_POS_ERROR; |
| 2042 |
else |
| 2043 |
{ |
| 2044 |
/* |
| 2045 |
We can abort sorting after thd->select_limit rows if we there is no |
| 2046 |
WHERE clause for any tables after the sorted one. |
| 2047 |
*/ |
| 2048 |
JOIN_TAB *curr_table= &curr_join->join_tab[curr_join->const_tables+1]; |
| 2049 |
JOIN_TAB *end_table= &curr_join->join_tab[curr_join->tables]; |
| 2050 |
for (; curr_table < end_table ; curr_table++) |
| 2051 |
{ |
| 2052 |
/* |
| 2053 |
table->keyuse is set in the case there was an original WHERE clause |
| 2054 |
on the table that was optimized away. |
| 2055 |
*/ |
| 2056 |
if (curr_table->select_cond || |
| 2057 |
(curr_table->keyuse && !curr_table->first_inner)) |
| 2058 |
{ |
| 2059 |
/* We have to sort all rows */ |
| 2060 |
curr_join->select_limit= HA_POS_ERROR; |
| 2061 |
break; |
| 2062 |
} |
| 2063 |
} |
| 2064 |
} |
| 2065 |
if (curr_join->join_tab == join_tab && save_join_tab()) |
| 2066 |
{ |
| 2067 |
DBUG_VOID_RETURN; |
| 2068 |
} |
| 2069 |
/* |
| 2070 |
Here we sort rows for ORDER BY/GROUP BY clause, if the optimiser |
| 2071 |
chose FILESORT to be faster than INDEX SCAN or there is no |
| 2072 |
suitable index present. |
| 2073 |
Note, that create_sort_index calls test_if_skip_sort_order and may |
| 2074 |
finally replace sorting with index scan if there is a LIMIT clause in |
| 2075 |
the query. XXX: it's never shown in EXPLAIN! |
| 2076 |
OPTION_FOUND_ROWS supersedes LIMIT and is taken into account. |
| 2077 |
*/ |
| 2078 |
if (create_sort_index(thd, curr_join, |
| 2079 |
curr_join->group_list ? |
| 2080 |
curr_join->group_list : curr_join->order, |
| 2081 |
curr_join->select_limit, |
| 2082 |
(select_options & OPTION_FOUND_ROWS ? |
| 2083 |
HA_POS_ERROR : unit->select_limit_cnt))) |
| 2084 |
DBUG_VOID_RETURN; |
| 2085 |
sortorder= curr_join->sortorder; |
| 2086 |
} |
| 2087 |
} |
| 2088 |
/* XXX: When can we have here thd->net.report_error not zero? */ |
| 2089 |
if (thd->net.report_error) |
| 2090 |
{ |
| 2091 |
error= thd->net.report_error; |
| 2092 |
DBUG_VOID_RETURN; |
| 2093 |
} |
| 2094 |
curr_join->having= curr_join->tmp_having; |
| 2095 |
curr_join->fields= curr_fields_list; |
| 2096 |
curr_join->procedure= procedure; |
| 2097 |
|
| 2098 |
#ifdef ENABLE_SENNA |
| 2099 |
if (my_thread_var->sen_flags & SENNA_USE_2IND) |
| 2100 |
DEBUG_2IND(my_thread_var->sen_flags &= ~SENNA_FIRST_CALL); |
| 2101 |
#endif /* ENABLE_SENNA */ |
| 2102 |
|
| 2103 |
if (is_top_level_join() && thd->cursor && tables != const_tables) |
| 2104 |
{ |
| 2105 |
/* |
| 2106 |
We are here if this is JOIN::exec for the last select of the main unit |
| 2107 |
and the client requested to open a cursor. |
| 2108 |
We check that not all tables are constant because this case is not |
| 2109 |
handled by do_select() separately, and this case is not implemented |
| 2110 |
for cursors yet. |
| 2111 |
*/ |
| 2112 |
DBUG_ASSERT(error == 0); |
| 2113 |
/* |
| 2114 |
curr_join is used only for reusable joins - that is, |
| 2115 |
to perform SELECT for each outer row (like in subselects). |
| 2116 |
This join is main, so we know for sure that curr_join == join. |
| 2117 |
*/ |
| 2118 |
DBUG_ASSERT(curr_join == this); |
| 2119 |
/* Open cursor for the last join sweep */ |
| 2120 |
error= thd->cursor->open(this); |
| 2121 |
} |
| 2122 |
else |
| 2123 |
{ |
| 2124 |
thd_proc_info(thd, "Sending data"); |
| 2125 |
DBUG_PRINT("info", ("%s", thd->proc_info)); |
| 2126 |
result->send_fields((procedure ? curr_join->procedure_fields_list : |
| 2127 |
*curr_fields_list), |
| 2128 |
Protocol::SEND_NUM_ROWS | Protocol::SEND_EOF); |
| 2129 |
error= do_select(curr_join, curr_fields_list, NULL, procedure); |
| 2130 |
thd->limit_found_rows= curr_join->send_records; |
| 2131 |
} |
| 2132 |
|
| 2133 |
/* Accumulate the counts from all join iterations of all join parts. */ |
| 2134 |
thd->examined_row_count+= curr_join->examined_rows; |
| 2135 |
DBUG_PRINT("counts", ("thd->examined_row_count: %lu", |
| 2136 |
(ulong) thd->examined_row_count)); |
| 2137 |
|
| 2138 |
/* |
| 2139 |
With EXPLAIN EXTENDED we have to restore original ref_array |
| 2140 |
for a derived table which is always materialized. |
| 2141 |
Otherwise we would not be able to print the query correctly. |
| 2142 |
*/ |
| 2143 |
if (items0 && |
| 2144 |
(thd->lex->describe & DESCRIBE_EXTENDED) && |
| 2145 |
select_lex->linkage == DERIVED_TABLE_TYPE) |
| 2146 |
set_items_ref_array(items0); |
| 2147 |
|
| 2148 |
DBUG_VOID_RETURN; |
| 2149 |
} |
| 2150 |
|
| 2151 |
|
| 2152 |
/* |
| 2153 |
Clean up join. Return error that hold JOIN. |
| 2154 |
*/ |
| 2155 |
|
| 2156 |
int |
| 2157 |
JOIN::destroy() |
| 2158 |
{ |
| 2159 |
DBUG_ENTER("JOIN::destroy"); |
| 2160 |
select_lex->join= 0; |
| 2161 |
|
| 2162 |
if (tmp_join) |
| 2163 |
{ |
| 2164 |
if (join_tab != tmp_join->join_tab) |
| 2165 |
{ |
| 2166 |
JOIN_TAB *tab, *end; |
| 2167 |
for (tab= join_tab, end= tab+tables ; tab != end ; tab++) |
| 2168 |
{ |
| 2169 |
tab->cleanup(); |
| 2170 |
} |
| 2171 |
} |
| 2172 |
tmp_join->tmp_join= 0; |
| 2173 |
tmp_table_param.copy_field=0; |
| 2174 |
DBUG_RETURN(tmp_join->destroy()); |
| 2175 |
} |
| 2176 |
cond_equal= 0; |
| 2177 |
|
| 2178 |
cleanup(1); |
| 2179 |
if (exec_tmp_table1) |
| 2180 |
free_tmp_table(thd, exec_tmp_table1); |
| 2181 |
if (exec_tmp_table2) |
| 2182 |
free_tmp_table(thd, exec_tmp_table2); |
| 2183 |
delete select; |
| 2184 |
delete_dynamic(&keyuse); |
| 2185 |
delete procedure; |
| 2186 |
DBUG_RETURN(error); |
| 2187 |
} |
| 2188 |
|
| 2189 |
/* |
| 2190 |
An entry point to single-unit select (a select without UNION). |
| 2191 |
|
| 2192 |
SYNOPSIS |
| 2193 |
mysql_select() |
| 2194 |
|
| 2195 |
thd thread handler |
| 2196 |
rref_pointer_array a reference to ref_pointer_array of |
| 2197 |
the top-level select_lex for this query |
| 2198 |
tables list of all tables used in this query. |
| 2199 |
The tables have been pre-opened. |
| 2200 |
wild_num number of wildcards used in the top level |
| 2201 |
select of this query. |
| 2202 |
For example statement |
| 2203 |
SELECT *, t1.*, catalog.t2.* FROM t0, t1, t2; |
| 2204 |
has 3 wildcards. |
| 2205 |
fields list of items in SELECT list of the top-level |
| 2206 |
select |
| 2207 |
e.g. SELECT a, b, c FROM t1 will have Item_field |
| 2208 |
for a, b and c in this list. |
| 2209 |
conds top level item of an expression representing |
| 2210 |
WHERE clause of the top level select |
| 2211 |
og_num total number of ORDER BY and GROUP BY clauses |
| 2212 |
arguments |
| 2213 |
order linked list of ORDER BY agruments |
| 2214 |
group linked list of GROUP BY arguments |
| 2215 |
having top level item of HAVING expression |
| 2216 |
proc_param list of PROCEDUREs |
| 2217 |
select_options select options (BIG_RESULT, etc) |
| 2218 |
result an instance of result set handling class. |
| 2219 |
This object is responsible for send result |
| 2220 |
set rows to the client or inserting them |
| 2221 |
into a table. |
| 2222 |
select_lex the only SELECT_LEX of this query |
| 2223 |
unit top-level UNIT of this query |
| 2224 |
UNIT is an artificial object created by the parser |
| 2225 |
for every SELECT clause. |
| 2226 |
e.g. SELECT * FROM t1 WHERE a1 IN (SELECT * FROM t2) |
| 2227 |
has 2 unions. |
| 2228 |
|
| 2229 |
RETURN VALUE |
| 2230 |
FALSE success |
| 2231 |
TRUE an error |
| 2232 |
*/ |
| 2233 |
|
| 2234 |
bool |
| 2235 |
mysql_select(THD *thd, Item ***rref_pointer_array, |
| 2236 |
TABLE_LIST *tables, uint wild_num, List<Item> &fields, |
| 2237 |
COND *conds, uint og_num, ORDER *order, ORDER *group, |
| 2238 |
Item *having, ORDER *proc_param, ulonglong select_options, |
| 2239 |
select_result *result, SELECT_LEX_UNIT *unit, |
| 2240 |
SELECT_LEX *select_lex) |
| 2241 |
{ |
| 2242 |
bool err; |
| 2243 |
bool free_join= 1; |
| 2244 |
DBUG_ENTER("mysql_select"); |
| 2245 |
|
| 2246 |
select_lex->context.resolve_in_select_list= TRUE; |
| 2247 |
JOIN *join; |
| 2248 |
if (select_lex->join != 0) |
| 2249 |
{ |
| 2250 |
join= select_lex->join; |
| 2251 |
/* |
| 2252 |
is it single SELECT in derived table, called in derived table |
| 2253 |
creation |
| 2254 |
*/ |
| 2255 |
if (select_lex->linkage != DERIVED_TABLE_TYPE || |
| 2256 |
(select_options & SELECT_DESCRIBE)) |
| 2257 |
{ |
| 2258 |
if (select_lex->linkage != GLOBAL_OPTIONS_TYPE) |
| 2259 |
{ |
| 2260 |
//here is EXPLAIN of subselect or derived table |
| 2261 |
if (join->change_result(result)) |
| 2262 |
{ |
| 2263 |
DBUG_RETURN(TRUE); |
| 2264 |
} |
| 2265 |
} |
| 2266 |
else |
| 2267 |
{ |
| 2268 |
if (err= join->prepare(rref_pointer_array, tables, wild_num, |
| 2269 |
conds, og_num, order, group, having, proc_param, |
| 2270 |
select_lex, unit)) |
| 2271 |
{ |
| 2272 |
goto err; |
| 2273 |
} |
| 2274 |
} |
| 2275 |
} |
| 2276 |
free_join= 0; |
| 2277 |
join->select_options= select_options; |
| 2278 |
} |
| 2279 |
else |
| 2280 |
{ |
| 2281 |
if (!(join= new JOIN(thd, fields, select_options, result))) |
| 2282 |
DBUG_RETURN(TRUE); |
| 2283 |
thd_proc_info(thd, "init"); |
| 2284 |
thd->used_tables=0; // Updated by setup_fields |
| 2285 |
if (err= join->prepare(rref_pointer_array, tables, wild_num, |
| 2286 |
conds, og_num, order, group, having, proc_param, |
| 2287 |
select_lex, unit)) |
| 2288 |
{ |
| 2289 |
goto err; |
| 2290 |
} |
| 2291 |
} |
| 2292 |
#ifdef ENABLE_SENNA |
| 2293 |
if (my_thread_var->sen_flags & SENNA_USE_2IND) { |
| 2294 |
if (select_lex->ftfunc_list->elements) { |
| 2295 |
DEBUG_2IND(my_thread_var->sen_flags |= SENNA_MATCH); |
| 2296 |
} |
| 2297 |
if (join->select_distinct) { |
| 2298 |
DEBUG_2IND(my_thread_var->sen_flags |= SENNA_DISTINCT); |
| 2299 |
} |
| 2300 |
} |
| 2301 |
#endif /* ENABLE_SENNA */ |
| 2302 |
if ((err= join->optimize())) |
| 2303 |
{ |
| 2304 |
goto err; // 1 |
| 2305 |
} |
| 2306 |
|
| 2307 |
if (thd->lex->describe & DESCRIBE_EXTENDED) |
| 2308 |
{ |
| 2309 |
join->conds_history= join->conds; |
| 2310 |
join->having_history= (join->having?join->having:join->tmp_having); |
| 2311 |
} |
| 2312 |
|
| 2313 |
if (thd->net.report_error) |
| 2314 |
goto err; |
| 2315 |
|
| 2316 |
join->exec(); |
| 2317 |
|
| 2318 |
if (thd->cursor && thd->cursor->is_open()) |
| 2319 |
{ |
| 2320 |
/* |
| 2321 |
A cursor was opened for the last sweep in exec(). |
| 2322 |
We are here only if this is mysql_select for top-level SELECT_LEX_UNIT |
| 2323 |
and there were no error. |
| 2324 |
*/ |
| 2325 |
free_join= 0; |
| 2326 |
} |
| 2327 |
|
| 2328 |
if (thd->lex->describe & DESCRIBE_EXTENDED) |
| 2329 |
{ |
| 2330 |
select_lex->where= join->conds_history; |
| 2331 |
select_lex->having= join->having_history; |
| 2332 |
} |
| 2333 |
|
| 2334 |
err: |
| 2335 |
#ifdef ENABLE_SENNA |
| 2336 |
DEBUG_2IND(my_thread_var->sen_flags &= SENNA_USE_2IND); |
| 2337 |
#endif |
| 2338 |
if (free_join) |
| 2339 |
{ |
| 2340 |
thd_proc_info(thd, "end"); |
| 2341 |
err|= select_lex->cleanup(); |
| 2342 |
thd_proc_info(thd, "end"); |
| 2343 |
DBUG_RETURN(err || thd->net.report_error); |
| 2344 |
} |
| 2345 |
DBUG_RETURN(join->error); |
| 2346 |
} |
| 2347 |
|
| 2348 |
/***************************************************************************** |
| 2349 |
Create JOIN_TABS, make a guess about the table types, |
| 2350 |
Approximate how many records will be used in each table |
| 2351 |
*****************************************************************************/ |
| 2352 |
|
| 2353 |
static ha_rows get_quick_record_count(THD *thd, SQL_SELECT *select, |
| 2354 |
TABLE *table, |
| 2355 |
const key_map *keys,ha_rows limit) |
| 2356 |
{ |
| 2357 |
int error; |
| 2358 |
DBUG_ENTER("get_quick_record_count"); |
| 2359 |
if (select) |
| 2360 |
{ |
| 2361 |
select->head=table; |
| 2362 |
table->reginfo.impossible_range=0; |
| 2363 |
if ((error= select->test_quick_select(thd, *(key_map *)keys,(table_map) 0, |
| 2364 |
limit, 0)) == 1) |
| 2365 |
DBUG_RETURN(select->quick->records); |
| 2366 |
if (error == -1) |
| 2367 |
{ |
| 2368 |
table->reginfo.impossible_range=1; |
| 2369 |
DBUG_RETURN(0); |
| 2370 |
} |
| 2371 |
DBUG_PRINT("warning",("Couldn't use record count on const keypart")); |
| 2372 |
} |
| 2373 |
DBUG_RETURN(HA_POS_ERROR); /* This shouldn't happend */ |
| 2374 |
} |
| 2375 |
|
| 2376 |
/* |
| 2377 |
This structure is used to collect info on potentially sargable |
| 2378 |
predicates in order to check whether they become sargable after |
| 2379 |
reading const tables. |
| 2380 |
We form a bitmap of indexes that can be used for sargable predicates. |
| 2381 |
Only such indexes are involved in range analysis. |
| 2382 |
*/ |
| 2383 |
typedef struct st_sargable_param |
| 2384 |
{ |
| 2385 |
Field *field; /* field against which to check sargability */ |
| 2386 |
Item **arg_value; /* values of potential keys for lookups */ |
| 2387 |
uint num_values; /* number of values in the above array */ |
| 2388 |
} SARGABLE_PARAM; |
| 2389 |
|
| 2390 |
/* |
| 2391 |
Calculate the best possible join and initialize the join structure |
| 2392 |
|
| 2393 |
RETURN VALUES |
| 2394 |
0 ok |
| 2395 |
1 Fatal error |
| 2396 |
*/ |
| 2397 |
|
| 2398 |
static bool |
| 2399 |
make_join_statistics(JOIN *join, TABLE_LIST *tables, COND *conds, |
| 2400 |
DYNAMIC_ARRAY *keyuse_array) |
| 2401 |
{ |
| 2402 |
int error; |
| 2403 |
TABLE *table; |
| 2404 |
uint i,table_count,const_count,key; |
| 2405 |
table_map found_const_table_map, all_table_map, found_ref, refs; |
| 2406 |
key_map const_ref, eq_part; |
| 2407 |
TABLE **table_vector; |
| 2408 |
JOIN_TAB *stat,*stat_end,*s,**stat_ref; |
| 2409 |
KEYUSE *keyuse,*start_keyuse; |
| 2410 |
table_map outer_join=0; |
| 2411 |
SARGABLE_PARAM *sargables= 0; |
| 2412 |
JOIN_TAB *stat_vector[MAX_TABLES+1]; |
| 2413 |
DBUG_ENTER("make_join_statistics"); |
| 2414 |
|
| 2415 |
table_count=join->tables; |
| 2416 |
stat=(JOIN_TAB*) join->thd->calloc(sizeof(JOIN_TAB)*table_count); |
| 2417 |
stat_ref=(JOIN_TAB**) join->thd->alloc(sizeof(JOIN_TAB*)*MAX_TABLES); |
| 2418 |
table_vector=(TABLE**) join->thd->alloc(sizeof(TABLE*)*(table_count*2)); |
| 2419 |
if (!stat || !stat_ref || !table_vector) |
| 2420 |
DBUG_RETURN(1); // Eom /* purecov: inspected */ |
| 2421 |
|
| 2422 |
join->best_ref=stat_vector; |
| 2423 |
|
| 2424 |
stat_end=stat+table_count; |
| 2425 |
found_const_table_map= all_table_map=0; |
| 2426 |
const_count=0; |
| 2427 |
|
| 2428 |
for (s= stat, i= 0; |
| 2429 |
tables; |
| 2430 |
s++, tables= tables->next_leaf, i++) |
| 2431 |
{ |
| 2432 |
TABLE_LIST *embedding= tables->embedding; |
| 2433 |
stat_vector[i]=s; |
| 2434 |
s->keys.init(); |
| 2435 |
s->const_keys.init(); |
| 2436 |
s->checked_keys.init(); |
| 2437 |
s->needed_reg.init(); |
| 2438 |
table_vector[i]=s->table=table=tables->table; |
| 2439 |
table->pos_in_table_list= tables; |
| 2440 |
error= table->file->info(HA_STATUS_VARIABLE | HA_STATUS_NO_LOCK); |
| 2441 |
if(error) |
| 2442 |
{ |
| 2443 |
table->file->print_error(error, MYF(0)); |
| 2444 |
DBUG_RETURN(1); |
| 2445 |
} |
| 2446 |
table->quick_keys.clear_all(); |
| 2447 |
table->reginfo.join_tab=s; |
| 2448 |
table->reginfo.not_exists_optimize=0; |
| 2449 |
bzero((char*) table->const_key_parts, sizeof(key_part_map)*table->s->keys); |
| 2450 |
all_table_map|= table->map; |
| 2451 |
s->join=join; |
| 2452 |
s->info=0; // For describe |
| 2453 |
|
| 2454 |
s->dependent= tables->dep_tables; |
| 2455 |
s->key_dependent= 0; |
| 2456 |
if (tables->schema_table) |
| 2457 |
table->file->records= 2; |
| 2458 |
|
| 2459 |
s->on_expr_ref= &tables->on_expr; |
| 2460 |
if (*s->on_expr_ref) |
| 2461 |
{ |
| 2462 |
/* s is the only inner table of an outer join */ |
| 2463 |
if (!table->file->records && !embedding) |
| 2464 |
{ // Empty table |
| 2465 |
s->dependent= 0; // Ignore LEFT JOIN depend. |
| 2466 |
set_position(join,const_count++,s,(KEYUSE*) 0); |
| 2467 |
continue; |
| 2468 |
} |
| 2469 |
outer_join|= table->map; |
| 2470 |
s->embedding_map= 0; |
| 2471 |
for (;embedding; embedding= embedding->embedding) |
| 2472 |
s->embedding_map|= embedding->nested_join->nj_map; |
| 2473 |
continue; |
| 2474 |
} |
| 2475 |
if (embedding) |
| 2476 |
{ |
| 2477 |
/* s belongs to a nested join, maybe to several embedded joins */ |
| 2478 |
s->embedding_map= 0; |
| 2479 |
do |
| 2480 |
{ |
| 2481 |
NESTED_JOIN *nested_join= embedding->nested_join; |
| 2482 |
s->embedding_map|=nested_join->nj_map; |
| 2483 |
s->dependent|= embedding->dep_tables; |
| 2484 |
embedding= embedding->embedding; |
| 2485 |
outer_join|= nested_join->used_tables; |
| 2486 |
} |
| 2487 |
while (embedding); |
| 2488 |
continue; |
| 2489 |
} |
| 2490 |
|
| 2491 |
if ((table->s->system || table->file->records <= 1) && ! s->dependent && |
| 2492 |
!(table->file->table_flags() & HA_NOT_EXACT_COUNT) && |
| 2493 |
!table->fulltext_searched && !join->no_const_tables) |
| 2494 |
{ |
| 2495 |
set_position(join,const_count++,s,(KEYUSE*) 0); |
| 2496 |
} |
| 2497 |
} |
| 2498 |
stat_vector[i]=0; |
| 2499 |
join->outer_join=outer_join; |
| 2500 |
|
| 2501 |
if (join->outer_join) |
| 2502 |
{ |
| 2503 |
/* |
| 2504 |
Build transitive closure for relation 'to be dependent on'. |
| 2505 |
This will speed up the plan search for many cases with outer joins, |
| 2506 |
as well as allow us to catch illegal cross references/ |
| 2507 |
Warshall's algorithm is used to build the transitive closure. |
| 2508 |
As we use bitmaps to represent the relation the complexity |
| 2509 |
of the algorithm is O((number of tables)^2). |
| 2510 |
*/ |
| 2511 |
for (i= 0, s= stat ; i < table_count ; i++, s++) |
| 2512 |
{ |
| 2513 |
for (uint j= 0 ; j < table_count ; j++) |
| 2514 |
{ |
| 2515 |
table= stat[j].table; |
| 2516 |
if (s->dependent & table->map) |
| 2517 |
s->dependent |= table->reginfo.join_tab->dependent; |
| 2518 |
} |
| 2519 |
if (s->dependent) |
| 2520 |
s->table->maybe_null= 1; |
| 2521 |
} |
| 2522 |
/* Catch illegal cross references for outer joins */ |
| 2523 |
for (i= 0, s= stat ; i < table_count ; i++, s++) |
| 2524 |
{ |
| 2525 |
if (s->dependent & s->table->map) |
| 2526 |
{ |
| 2527 |
join->tables=0; // Don't use join->table |
| 2528 |
my_message(ER_WRONG_OUTER_JOIN, ER(ER_WRONG_OUTER_JOIN), MYF(0)); |
| 2529 |
DBUG_RETURN(1); |
| 2530 |
} |
| 2531 |
s->key_dependent= s->dependent; |
| 2532 |
} |
| 2533 |
} |
| 2534 |
|
| 2535 |
if (conds || outer_join) |
| 2536 |
if (update_ref_and_keys(join->thd, keyuse_array, stat, join->tables, |
| 2537 |
conds, join->cond_equal, |
| 2538 |
~outer_join, join->select_lex, &sargables)) |
| 2539 |
DBUG_RETURN(1); |
| 2540 |
|
| 2541 |
/* Read tables with 0 or 1 rows (system tables) */ |
| 2542 |
join->const_table_map= 0; |
| 2543 |
|
| 2544 |
for (POSITION *p_pos=join->positions, *p_end=p_pos+const_count; |
| 2545 |
p_pos < p_end ; |
| 2546 |
p_pos++) |
| 2547 |
{ |
| 2548 |
int tmp; |
| 2549 |
s= p_pos->table; |
| 2550 |
s->type=JT_SYSTEM; |
| 2551 |
join->const_table_map|=s->table->map; |
| 2552 |
if ((tmp=join_read_const_table(s, p_pos))) |
| 2553 |
{ |
| 2554 |
if (tmp > 0) |
| 2555 |
DBUG_RETURN(1); // Fatal error |
| 2556 |
} |
| 2557 |
else |
| 2558 |
found_const_table_map|= s->table->map; |
| 2559 |
} |
| 2560 |
|
| 2561 |
/* loop until no more const tables are found */ |
| 2562 |
int ref_changed; |
| 2563 |
do |
| 2564 |
{ |
| 2565 |
more_const_tables_found: |
| 2566 |
ref_changed = 0; |
| 2567 |
found_ref=0; |
| 2568 |
|
| 2569 |
/* |
| 2570 |
We only have to loop from stat_vector + const_count as |
| 2571 |
set_position() will move all const_tables first in stat_vector |
| 2572 |
*/ |
| 2573 |
|
| 2574 |
for (JOIN_TAB **pos=stat_vector+const_count ; (s= *pos) ; pos++) |
| 2575 |
{ |
| 2576 |
table=s->table; |
| 2577 |
|
| 2578 |
/* |
| 2579 |
If equi-join condition by a key is null rejecting and after a |
| 2580 |
substitution of a const table the key value happens to be null |
| 2581 |
then we can state that there are no matches for this equi-join. |
| 2582 |
*/ |
| 2583 |
if ((keyuse= s->keyuse) && *s->on_expr_ref && !s->embedding_map) |
| 2584 |
{ |
| 2585 |
/* |
| 2586 |
When performing an outer join operation if there are no matching rows |
| 2587 |
for the single row of the outer table all the inner tables are to be |
| 2588 |
null complemented and thus considered as constant tables. |
| 2589 |
Here we apply this consideration to the case of outer join operations |
| 2590 |
with a single inner table only because the case with nested tables |
| 2591 |
would require a more thorough analysis. |
| 2592 |
TODO. Apply single row substitution to null complemented inner tables |
| 2593 |
for nested outer join operations. |
| 2594 |
*/ |
| 2595 |
while (keyuse->table == table) |
| 2596 |
{ |
| 2597 |
if (!(keyuse->val->used_tables() & ~join->const_table_map) && |
| 2598 |
keyuse->val->is_null() && keyuse->null_rejecting) |
| 2599 |
{ |
| 2600 |
s->type= JT_CONST; |
| 2601 |
mark_as_null_row(table); |
| 2602 |
found_const_table_map|= table->map; |
| 2603 |
join->const_table_map|= table->map; |
| 2604 |
set_position(join,const_count++,s,(KEYUSE*) 0); |
| 2605 |
goto more_const_tables_found; |
| 2606 |
} |
| 2607 |
keyuse++; |
| 2608 |
} |
| 2609 |
} |
| 2610 |
|
| 2611 |
if (s->dependent) // If dependent on some table |
| 2612 |
{ |
| 2613 |
// All dep. must be constants |
| 2614 |
if (s->dependent & ~(found_const_table_map)) |
| 2615 |
continue; |
| 2616 |
if (table->file->records <= 1L && |
| 2617 |
!(table->file->table_flags() & HA_NOT_EXACT_COUNT) && |
| 2618 |
!table->pos_in_table_list->embedding) |
| 2619 |
{ // system table |
| 2620 |
int tmp= 0; |
| 2621 |
s->type=JT_SYSTEM; |
| 2622 |
join->const_table_map|=table->map; |
| 2623 |
set_position(join,const_count++,s,(KEYUSE*) 0); |
| 2624 |
if ((tmp= join_read_const_table(s, join->positions+const_count-1))) |
| 2625 |
{ |
| 2626 |
if (tmp > 0) |
| 2627 |
DBUG_RETURN(1); // Fatal error |
| 2628 |
} |
| 2629 |
else |
| 2630 |
found_const_table_map|= table->map; |
| 2631 |
continue; |
| 2632 |
} |
| 2633 |
} |
| 2634 |
/* check if table can be read by key or table only uses const refs */ |
| 2635 |
if ((keyuse=s->keyuse)) |
| 2636 |
{ |
| 2637 |
s->type= JT_REF; |
| 2638 |
while (keyuse->table == table) |
| 2639 |
{ |
| 2640 |
start_keyuse=keyuse; |
| 2641 |
key=keyuse->key; |
| 2642 |
s->keys.set_bit(key); // QQ: remove this ? |
| 2643 |
|
| 2644 |
refs=0; |
| 2645 |
const_ref.clear_all(); |
| 2646 |
eq_part.clear_all(); |
| 2647 |
do |
| 2648 |
{ |
| 2649 |
if (keyuse->val->type() != Item::NULL_ITEM && !keyuse->optimize) |
| 2650 |
{ |
| 2651 |
if (!((~found_const_table_map) & keyuse->used_tables)) |
| 2652 |
const_ref.set_bit(keyuse->keypart); |
| 2653 |
else |
| 2654 |
refs|=keyuse->used_tables; |
| 2655 |
eq_part.set_bit(keyuse->keypart); |
| 2656 |
} |
| 2657 |
keyuse++; |
| 2658 |
} while (keyuse->table == table && keyuse->key == key); |
| 2659 |
|
| 2660 |
if (eq_part.is_prefix(table->key_info[key].key_parts) && |
| 2661 |
!table->fulltext_searched && |
| 2662 |
!table->pos_in_table_list->embedding) |
| 2663 |
{ |
| 2664 |
if ((table->key_info[key].flags & (HA_NOSAME | HA_END_SPACE_KEY)) |
| 2665 |
== HA_NOSAME) |
| 2666 |
{ |
| 2667 |
if (const_ref == eq_part) |
| 2668 |
{ // Found everything for ref. |
| 2669 |
int tmp; |
| 2670 |
ref_changed = 1; |
| 2671 |
s->type= JT_CONST; |
| 2672 |
join->const_table_map|=table->map; |
| 2673 |
set_position(join,const_count++,s,start_keyuse); |
| 2674 |
if (create_ref_for_key(join, s, start_keyuse, |
| 2675 |
found_const_table_map)) |
| 2676 |
DBUG_RETURN(1); |
| 2677 |
if ((tmp=join_read_const_table(s, |
| 2678 |
join->positions+const_count-1))) |
| 2679 |
{ |
| 2680 |
if (tmp > 0) |
| 2681 |
DBUG_RETURN(1); // Fatal error |
| 2682 |
} |
| 2683 |
else |
| 2684 |
found_const_table_map|= table->map; |
| 2685 |
break; |
| 2686 |
} |
| 2687 |
else |
| 2688 |
found_ref|= refs; // Table is const if all refs are const |
| 2689 |
} |
| 2690 |
else if (const_ref == eq_part) |
| 2691 |
s->const_keys.set_bit(key); |
| 2692 |
} |
| 2693 |
} |
| 2694 |
} |
| 2695 |
} |
| 2696 |
} while (join->const_table_map & found_ref && ref_changed); |
| 2697 |
|
| 2698 |
/* |
| 2699 |
Update info on indexes that can be used for search lookups as |
| 2700 |
reading const tables may has added new sargable predicates. |
| 2701 |
*/ |
| 2702 |
if (const_count && sargables) |
| 2703 |
{ |
| 2704 |
for( ; sargables->field ; sargables++) |
| 2705 |
{ |
| 2706 |
Field *field= sargables->field; |
| 2707 |
JOIN_TAB *join_tab= field->table->reginfo.join_tab; |
| 2708 |
key_map possible_keys= field->key_start; |
| 2709 |
possible_keys.intersect(field->table->keys_in_use_for_query); |
| 2710 |
bool is_const= 1; |
| 2711 |
for (uint j=0; j < sargables->num_values; j++) |
| 2712 |
is_const&= sargables->arg_value[j]->const_item(); |
| 2713 |
if (is_const) |
| 2714 |
join_tab[0].const_keys.merge(possible_keys); |
| 2715 |
} |
| 2716 |
} |
| 2717 |
|
| 2718 |
/* Calc how many (possible) matched records in each table */ |
| 2719 |
|
| 2720 |
for (s=stat ; s < stat_end ; s++) |
| 2721 |
{ |
| 2722 |
if (s->type == JT_SYSTEM || s->type == JT_CONST) |
| 2723 |
{ |
| 2724 |
/* Only one matching row */ |
| 2725 |
s->found_records=s->records=s->read_time=1; s->worst_seeks=1.0; |
| 2726 |
continue; |
| 2727 |
} |
| 2728 |
/* Approximate found rows and time to read them */ |
| 2729 |
s->found_records=s->records=s->table->file->records; |
| 2730 |
s->read_time=(ha_rows) s->table->file->scan_time(); |
| 2731 |
|
| 2732 |
/* |
| 2733 |
Set a max range of how many seeks we can expect when using keys |
| 2734 |
This is can't be to high as otherwise we are likely to use |
| 2735 |
table scan. |
| 2736 |
*/ |
| 2737 |
s->worst_seeks= min((double) s->found_records / 10, |
| 2738 |
(double) s->read_time*3); |
| 2739 |
if (s->worst_seeks < 2.0) // Fix for small tables |
| 2740 |
s->worst_seeks=2.0; |
| 2741 |
|
| 2742 |
/* |
| 2743 |
Add to stat->const_keys those indexes for which all group fields or |
| 2744 |
all select distinct fields participate in one index. |
| 2745 |
*/ |
| 2746 |
add_group_and_distinct_keys(join, s); |
| 2747 |
|
| 2748 |
if (!s->const_keys.is_clear_all() && |
| 2749 |
!s->table->pos_in_table_list->embedding) |
| 2750 |
{ |
| 2751 |
ha_rows records; |
| 2752 |
SQL_SELECT *select; |
| 2753 |
select= make_select(s->table, found_const_table_map, |
| 2754 |
found_const_table_map, |
| 2755 |
*s->on_expr_ref ? *s->on_expr_ref : conds, |
| 2756 |
1, &error); |
| 2757 |
if (!select) |
| 2758 |
DBUG_RETURN(1); |
| 2759 |
records= get_quick_record_count(join->thd, select, s->table, |
| 2760 |
&s->const_keys, join->row_limit); |
| 2761 |
s->quick=select->quick; |
| 2762 |
s->needed_reg=select->needed_reg; |
| 2763 |
select->quick=0; |
| 2764 |
if (records == 0 && s->table->reginfo.impossible_range) |
| 2765 |
{ |
| 2766 |
/* |
| 2767 |
Impossible WHERE or ON expression |
| 2768 |
In case of ON, we mark that the we match one empty NULL row. |
| 2769 |
In case of WHERE, don't set found_const_table_map to get the |
| 2770 |
caller to abort with a zero row result. |
| 2771 |
*/ |
| 2772 |
join->const_table_map|= s->table->map; |
| 2773 |
set_position(join,const_count++,s,(KEYUSE*) 0); |
| 2774 |
s->type= JT_CONST; |
| 2775 |
if (*s->on_expr_ref) |
| 2776 |
{ |
| 2777 |
/* Generate empty row */ |
| 2778 |
s->info= "Impossible ON condition"; |
| 2779 |
found_const_table_map|= s->table->map; |
| 2780 |
s->type= JT_CONST; |
| 2781 |
mark_as_null_row(s->table); // All fields are NULL |
| 2782 |
} |
| 2783 |
} |
| 2784 |
if (records != HA_POS_ERROR) |
| 2785 |
{ |
| 2786 |
s->found_records=records; |
| 2787 |
s->read_time= (ha_rows) (s->quick ? s->quick->read_time : 0.0); |
| 2788 |
} |
| 2789 |
delete select; |
| 2790 |
} |
| 2791 |
} |
| 2792 |
|
| 2793 |
join->join_tab=stat; |
| 2794 |
join->map2table=stat_ref; |
| 2795 |
join->table= join->all_tables=table_vector; |
| 2796 |
join->const_tables=const_count; |
| 2797 |
join->found_const_table_map=found_const_table_map; |
| 2798 |
|
| 2799 |
/* Find an optimal join order of the non-constant tables. */ |
| 2800 |
if (join->const_tables != join->tables) |
| 2801 |
{ |
| 2802 |
optimize_keyuse(join, keyuse_array); |
| 2803 |
if (choose_plan(join, all_table_map & ~join->const_table_map)) |
| 2804 |
DBUG_RETURN(TRUE); |
| 2805 |
} |
| 2806 |
else |
| 2807 |
{ |
| 2808 |
memcpy((gptr) join->best_positions,(gptr) join->positions, |
| 2809 |
sizeof(POSITION)*join->const_tables); |
| 2810 |
join->best_read=1.0; |
| 2811 |
} |
| 2812 |
/* Generate an execution plan from the found optimal join order. */ |
| 2813 |
DBUG_RETURN(join->thd->killed || get_best_combination(join)); |
| 2814 |
} |
| 2815 |
|
| 2816 |
|
| 2817 |
/***************************************************************************** |
| 2818 |
Check with keys are used and with tables references with tables |
| 2819 |
Updates in stat: |
| 2820 |
keys Bitmap of all used keys |
| 2821 |
const_keys Bitmap of all keys with may be used with quick_select |
| 2822 |
keyuse Pointer to possible keys |
| 2823 |
*****************************************************************************/ |
| 2824 |
|
| 2825 |
typedef struct key_field_t { // Used when finding key fields |
| 2826 |
Field *field; |
| 2827 |
Item *val; // May be empty if diff constant |
| 2828 |
uint level; |
| 2829 |
uint optimize; |
| 2830 |
bool eq_func; |
| 2831 |
/* |
| 2832 |
If true, the condition this struct represents will not be satisfied |
| 2833 |
when val IS NULL. |
| 2834 |
*/ |
| 2835 |
bool null_rejecting; |
| 2836 |
bool *cond_guard; /* See KEYUSE::cond_guard */ |
| 2837 |
} KEY_FIELD; |
| 2838 |
|
| 2839 |
/* Values in optimize */ |
| 2840 |
#define KEY_OPTIMIZE_EXISTS 1 |
| 2841 |
#define KEY_OPTIMIZE_REF_OR_NULL 2 |
| 2842 |
|
| 2843 |
/* |
| 2844 |
Merge new key definitions to old ones, remove those not used in both |
| 2845 |
|
| 2846 |
This is called for OR between different levels |
| 2847 |
|
| 2848 |
To be able to do 'ref_or_null' we merge a comparison of a column |
| 2849 |
and 'column IS NULL' to one test. This is useful for sub select queries |
| 2850 |
that are internally transformed to something like: |
| 2851 |
|
| 2852 |
SELECT * FROM t1 WHERE t1.key=outer_ref_field or t1.key IS NULL |
| 2853 |
|
| 2854 |
KEY_FIELD::null_rejecting is processed as follows: |
| 2855 |
result has null_rejecting=true if it is set for both ORed references. |
| 2856 |
for example: |
| 2857 |
(t2.key = t1.field OR t2.key = t1.field) -> null_rejecting=true |
| 2858 |
(t2.key = t1.field OR t2.key <=> t1.field) -> null_rejecting=false |
| 2859 |
*/ |
| 2860 |
|
| 2861 |
static KEY_FIELD * |
| 2862 |
merge_key_fields(KEY_FIELD *start,KEY_FIELD *new_fields,KEY_FIELD *end, |
| 2863 |
uint and_level) |
| 2864 |
{ |
| 2865 |
if (start == new_fields) |
| 2866 |
return start; // Impossible or |
| 2867 |
if (new_fields == end) |
| 2868 |
return start; // No new fields, skip all |
| 2869 |
|
| 2870 |
KEY_FIELD *first_free=new_fields; |
| 2871 |
|
| 2872 |
/* Mark all found fields in old array */ |
| 2873 |
for (; new_fields != end ; new_fields++) |
| 2874 |
{ |
| 2875 |
for (KEY_FIELD *old=start ; old != first_free ; old++) |
| 2876 |
{ |
| 2877 |
if (old->field == new_fields->field) |
| 2878 |
{ |
| 2879 |
/* |
| 2880 |
NOTE: below const_item() call really works as "!used_tables()", i.e. |
| 2881 |
it can return FALSE where it is feasible to make it return TRUE. |
| 2882 |
|
| 2883 |
The cause is as follows: Some of the tables are already known to be |
| 2884 |
const tables (the detection code is in make_join_statistics(), |
| 2885 |
above the update_ref_and_keys() call), but we didn't propagate |
| 2886 |
information about this: TABLE::const_table is not set to TRUE, and |
| 2887 |
Item::update_used_tables() hasn't been called for each item. |
| 2888 |
The result of this is that we're missing some 'ref' accesses. |
| 2889 |
TODO: OptimizerTeam: Fix this |
| 2890 |
*/ |
| 2891 |
if (!new_fields->val->const_item()) |
| 2892 |
{ |
| 2893 |
/* |
| 2894 |
If the value matches, we can use the key reference. |
| 2895 |
If not, we keep it until we have examined all new values |
| 2896 |
*/ |
| 2897 |
if (old->val->eq(new_fields->val, old->field->binary())) |
| 2898 |
{ |
| 2899 |
old->level= and_level; |
| 2900 |
old->optimize= ((old->optimize & new_fields->optimize & |
| 2901 |
KEY_OPTIMIZE_EXISTS) | |
| 2902 |
((old->optimize | new_fields->optimize) & |
| 2903 |
KEY_OPTIMIZE_REF_OR_NULL)); |
| 2904 |
old->null_rejecting= (old->null_rejecting && |
| 2905 |
new_fields->null_rejecting); |
| 2906 |
} |
| 2907 |
} |
| 2908 |
else if (old->eq_func && new_fields->eq_func && |
| 2909 |
old->val->eq(new_fields->val, old->field->binary())) |
| 2910 |
|
| 2911 |
{ |
| 2912 |
old->level= and_level; |
| 2913 |
old->optimize= ((old->optimize & new_fields->optimize & |
| 2914 |
KEY_OPTIMIZE_EXISTS) | |
| 2915 |
((old->optimize | new_fields->optimize) & |
| 2916 |
KEY_OPTIMIZE_REF_OR_NULL)); |
| 2917 |
old->null_rejecting= (old->null_rejecting && |
| 2918 |
new_fields->null_rejecting); |
| 2919 |
} |
| 2920 |
else if (old->eq_func && new_fields->eq_func && |
| 2921 |
((old->val->const_item() && old->val->is_null()) || |
| 2922 |
new_fields->val->is_null())) |
| 2923 |
{ |
| 2924 |
/* field = expression OR field IS NULL */ |
| 2925 |
old->level= and_level; |
| 2926 |
old->optimize= KEY_OPTIMIZE_REF_OR_NULL; |
| 2927 |
/* |
| 2928 |
Remember the NOT NULL value unless the value does not depend |
| 2929 |
on other tables. |
| 2930 |
*/ |
| 2931 |
if (!old->val->used_tables() && old->val->is_null()) |
| 2932 |
old->val= new_fields->val; |
| 2933 |
/* The referred expression can be NULL: */ |
| 2934 |
old->null_rejecting= 0; |
| 2935 |
} |
| 2936 |
else |
| 2937 |
{ |
| 2938 |
/* |
| 2939 |
We are comparing two different const. In this case we can't |
| 2940 |
use a key-lookup on this so it's better to remove the value |
| 2941 |
and let the range optimzier handle it |
| 2942 |
*/ |
| 2943 |
if (old == --first_free) // If last item |
| 2944 |
break; |
| 2945 |
*old= *first_free; // Remove old value |
| 2946 |
old--; // Retry this value |
| 2947 |
} |
| 2948 |
} |
| 2949 |
} |
| 2950 |
} |
| 2951 |
/* Remove all not used items */ |
| 2952 |
for (KEY_FIELD *old=start ; old != first_free ;) |
| 2953 |
{ |
| 2954 |
if (old->level != and_level) |
| 2955 |
{ // Not used in all levels |
| 2956 |
if (old == --first_free) |
| 2957 |
break; |
| 2958 |
*old= *first_free; // Remove old value |
| 2959 |
continue; |
| 2960 |
} |
| 2961 |
old++; |
| 2962 |
} |
| 2963 |
return first_free; |
| 2964 |
} |
| 2965 |
|
| 2966 |
|
| 2967 |
/* |
| 2968 |
Add a possible key to array of possible keys if it's usable as a key |
| 2969 |
|
| 2970 |
SYNPOSIS |
| 2971 |
add_key_field() |
| 2972 |
key_fields Pointer to add key, if usable |
| 2973 |
and_level And level, to be stored in KEY_FIELD |
| 2974 |
cond Condition predicate |
| 2975 |
field Field used in comparision |
| 2976 |
eq_func True if we used =, <=> or IS NULL |
| 2977 |
value Value used for comparison with field |
| 2978 |
usable_tables Tables which can be used for key optimization |
| 2979 |
sargables IN/OUT Array of found sargable candidates |
| 2980 |
|
| 2981 |
NOTES |
| 2982 |
If we are doing a NOT NULL comparison on a NOT NULL field in a outer join |
| 2983 |
table, we store this to be able to do not exists optimization later. |
| 2984 |
|
| 2985 |
RETURN |
| 2986 |
*key_fields is incremented if we stored a key in the array |
| 2987 |
*/ |
| 2988 |
|
| 2989 |
static void |
| 2990 |
add_key_field(KEY_FIELD **key_fields,uint and_level, Item_func *cond, |
| 2991 |
Field *field, bool eq_func, Item **value, uint num_values, |
| 2992 |
table_map usable_tables, SARGABLE_PARAM **sargables) |
| 2993 |
{ |
| 2994 |
uint exists_optimize= 0; |
| 2995 |
if (!(field->flags & PART_KEY_FLAG)) |
| 2996 |
{ |
| 2997 |
// Don't remove column IS NULL on a LEFT JOIN table |
| 2998 |
if (!eq_func || (*value)->type() != Item::NULL_ITEM || |
| 2999 |
!field->table->maybe_null || field->null_ptr) |
| 3000 |
return; // Not a key. Skip it |
| 3001 |
exists_optimize= KEY_OPTIMIZE_EXISTS; |
| 3002 |
DBUG_ASSERT(num_values == 1); |
| 3003 |
} |
| 3004 |
else |
| 3005 |
{ |
| 3006 |
table_map used_tables=0; |
| 3007 |
bool optimizable=0; |
| 3008 |
for (uint i=0; i<num_values; i++) |
| 3009 |
{ |
| 3010 |
used_tables|=(value[i])->used_tables(); |
| 3011 |
if (!((value[i])->used_tables() & (field->table->map | RAND_TABLE_BIT))) |
| 3012 |
optimizable=1; |
| 3013 |
} |
| 3014 |
if (!optimizable) |
| 3015 |
return; |
| 3016 |
if (!(usable_tables & field->table->map)) |
| 3017 |
{ |
| 3018 |
if (!eq_func || (*value)->type() != Item::NULL_ITEM || |
| 3019 |
!field->table->maybe_null || field->null_ptr) |
| 3020 |
return; // Can't use left join optimize |
| 3021 |
exists_optimize= KEY_OPTIMIZE_EXISTS; |
| 3022 |
} |
| 3023 |
else |
| 3024 |
{ |
| 3025 |
JOIN_TAB *stat=field->table->reginfo.join_tab; |
| 3026 |
key_map possible_keys=field->key_start; |
| 3027 |
possible_keys.intersect(field->table->keys_in_use_for_query); |
| 3028 |
stat[0].keys.merge(possible_keys); // Add possible keys |
| 3029 |
|
| 3030 |
/* |
| 3031 |
Save the following cases: |
| 3032 |
Field op constant |
| 3033 |
Field LIKE constant where constant doesn't start with a wildcard |
| 3034 |
Field = field2 where field2 is in a different table |
| 3035 |
Field op formula |
| 3036 |
Field IS NULL |
| 3037 |
Field IS NOT NULL |
| 3038 |
Field BETWEEN ... |
| 3039 |
Field IN ... |
| 3040 |
*/ |
| 3041 |
stat[0].key_dependent|=used_tables; |
| 3042 |
|
| 3043 |
bool is_const=1; |
| 3044 |
for (uint i=0; i<num_values; i++) |
| 3045 |
is_const&= value[i]->const_item(); |
| 3046 |
if (is_const) |
| 3047 |
stat[0].const_keys.merge(possible_keys); |
| 3048 |
else if (!eq_func) |
| 3049 |
{ |
| 3050 |
/* |
| 3051 |
Save info to be able check whether this predicate can be |
| 3052 |
considered as sargable for range analisis after reading const tables. |
| 3053 |
We do not save info about equalities as update_const_equal_items |
| 3054 |
will take care of updating info on keys from sargable equalities. |
| 3055 |
*/ |
| 3056 |
(*sargables)--; |
| 3057 |
(*sargables)->field= field; |
| 3058 |
(*sargables)->arg_value= value; |
| 3059 |
(*sargables)->num_values= num_values; |
| 3060 |
} |
| 3061 |
/* |
| 3062 |
We can't always use indexes when comparing a string index to a |
| 3063 |
number. cmp_type() is checked to allow compare of dates to numbers. |
| 3064 |
eq_func is NEVER true when num_values > 1 |
| 3065 |
*/ |
| 3066 |
if (!eq_func) |
| 3067 |
{ |
| 3068 |
/* |
| 3069 |
Additional optimization: if we're processing |
| 3070 |
"t.key BETWEEN c1 AND c1" then proceed as if we were processing |
| 3071 |
"t.key = c1". |
| 3072 |
TODO: This is a very limited fix. A more generic fix is possible. |
| 3073 |
There are 2 options: |
| 3074 |
A) Make equality propagation code be able to handle BETWEEN |
| 3075 |
(including cases like t1.key BETWEEN t2.key AND t3.key) |
| 3076 |
B) Make range optimizer to infer additional "t.key = c" equalities |
| 3077 |
and use them in equality propagation process (see details in |
| 3078 |
OptimizerKBAndTodo) |
| 3079 |
*/ |
| 3080 |
if ((cond->functype() != Item_func::BETWEEN) || |
| 3081 |
((Item_func_between*) cond)->negated || |
| 3082 |
!value[0]->eq(value[1], field->binary())) |
| 3083 |
return; |
| 3084 |
eq_func= TRUE; |
| 3085 |
} |
| 3086 |
|
| 3087 |
if (field->result_type() == STRING_RESULT) |
| 3088 |
{ |
| 3089 |
if ((*value)->result_type() != STRING_RESULT) |
| 3090 |
{ |
| 3091 |
if (field->cmp_type() != (*value)->result_type()) |
| 3092 |
return; |
| 3093 |
} |
| 3094 |
else |
| 3095 |
{ |
| 3096 |
/* |
| 3097 |
We can't use indexes if the effective collation |
| 3098 |
of the operation differ from the field collation. |
| 3099 |
*/ |
| 3100 |
if (field->cmp_type() == STRING_RESULT && |
| 3101 |
((Field_str*)field)->charset() != cond->compare_collation()) |
| 3102 |
return; |
| 3103 |
} |
| 3104 |
} |
| 3105 |
} |
| 3106 |
} |
| 3107 |
/* |
| 3108 |
For the moment eq_func is always true. This slot is reserved for future |
| 3109 |
extensions where we want to remembers other things than just eq comparisons |
| 3110 |
*/ |
| 3111 |
DBUG_ASSERT(eq_func); |
| 3112 |
/* Store possible eq field */ |
| 3113 |
(*key_fields)->field= field; |
| 3114 |
(*key_fields)->eq_func= eq_func; |
| 3115 |
(*key_fields)->val= *value; |
| 3116 |
(*key_fields)->level= and_level; |
| 3117 |
(*key_fields)->optimize= exists_optimize; |
| 3118 |
/* |
| 3119 |
If the condition has form "tbl.keypart = othertbl.field" and |
| 3120 |
othertbl.field can be NULL, there will be no matches if othertbl.field |
| 3121 |
has NULL value. |
| 3122 |
We use null_rejecting in add_not_null_conds() to add |
| 3123 |
'othertbl.field IS NOT NULL' to tab->select_cond. |
| 3124 |
*/ |
| 3125 |
(*key_fields)->null_rejecting= ((cond->functype() == Item_func::EQ_FUNC || |
| 3126 |
cond->functype() == Item_func::MULT_EQUAL_FUNC) && |
| 3127 |
((*value)->type() == Item::FIELD_ITEM) && |
| 3128 |
((Item_field*)*value)->field->maybe_null()); |
| 3129 |
(*key_fields)->cond_guard= NULL; |
| 3130 |
(*key_fields)++; |
| 3131 |
} |
| 3132 |
|
| 3133 |
/* |
| 3134 |
Add possible keys to array of possible keys originated from a simple predicate |
| 3135 |
|
| 3136 |
SYNPOSIS |
| 3137 |
add_key_equal_fields() |
| 3138 |
key_fields Pointer to add key, if usable |
| 3139 |
and_level And level, to be stored in KEY_FIELD |
| 3140 |
cond Condition predicate |
| 3141 |
field Field used in comparision |
| 3142 |
eq_func True if we used =, <=> or IS NULL |
| 3143 |
value Value used for comparison with field |
| 3144 |
Is NULL for BETWEEN and IN |
| 3145 |
usable_tables Tables which can be used for key optimization |
| 3146 |
sargables IN/OUT Array of found sargable candidates |
| 3147 |
|
| 3148 |
NOTES |
| 3149 |
If field items f1 and f2 belong to the same multiple equality and |
| 3150 |
a key is added for f1, the the same key is added for f2. |
| 3151 |
|
| 3152 |
RETURN |
| 3153 |
*key_fields is incremented if we stored a key in the array |
| 3154 |
*/ |
| 3155 |
|
| 3156 |
static void |
| 3157 |
add_key_equal_fields(KEY_FIELD **key_fields, uint and_level, |
| 3158 |
Item_func *cond, Item_field *field_item, |
| 3159 |
bool eq_func, Item **val, |
| 3160 |
uint num_values, table_map usable_tables, |
| 3161 |
SARGABLE_PARAM **sargables) |
| 3162 |
{ |
| 3163 |
Field *field= field_item->field; |
| 3164 |
add_key_field(key_fields, and_level, cond, field, |
| 3165 |
eq_func, val, num_values, usable_tables, sargables); |
| 3166 |
Item_equal *item_equal= field_item->item_equal; |
| 3167 |
if (item_equal) |
| 3168 |
{ |
| 3169 |
/* |
| 3170 |
Add to the set of possible key values every substitution of |
| 3171 |
the field for an equal field included into item_equal |
| 3172 |
*/ |
| 3173 |
Item_equal_iterator it(*item_equal); |
| 3174 |
Item_field *item; |
| 3175 |
while ((item= it++)) |
| 3176 |
{ |
| 3177 |
if (!field->eq(item->field)) |
| 3178 |
{ |
| 3179 |
add_key_field(key_fields, and_level, cond, item->field, |
| 3180 |
eq_func, val, num_values, usable_tables, |
| 3181 |
sargables); |
| 3182 |
} |
| 3183 |
} |
| 3184 |
} |
| 3185 |
} |
| 3186 |
|
| 3187 |
static void |
| 3188 |
add_key_fields(JOIN *join, KEY_FIELD **key_fields, uint *and_level, |
| 3189 |
COND *cond, table_map usable_tables, |
| 3190 |
SARGABLE_PARAM **sargables) |
| 3191 |
{ |
| 3192 |
if (cond->type() == Item_func::COND_ITEM) |
| 3193 |
{ |
| 3194 |
List_iterator_fast<Item> li(*((Item_cond*) cond)->argument_list()); |
| 3195 |
KEY_FIELD *org_key_fields= *key_fields; |
| 3196 |
|
| 3197 |
if (((Item_cond*) cond)->functype() == Item_func::COND_AND_FUNC) |
| 3198 |
{ |
| 3199 |
Item *item; |
| 3200 |
while ((item=li++)) |
| 3201 |
add_key_fields(join, key_fields, and_level, item, usable_tables, |
| 3202 |
sargables); |
| 3203 |
for (; org_key_fields != *key_fields ; org_key_fields++) |
| 3204 |
org_key_fields->level= *and_level; |
| 3205 |
} |
| 3206 |
else |
| 3207 |
{ |
| 3208 |
(*and_level)++; |
| 3209 |
add_key_fields(join, key_fields, and_level, li++, usable_tables, |
| 3210 |
sargables); |
| 3211 |
Item *item; |
| 3212 |
while ((item=li++)) |
| 3213 |
{ |
| 3214 |
KEY_FIELD *start_key_fields= *key_fields; |
| 3215 |
(*and_level)++; |
| 3216 |
add_key_fields(join, key_fields, and_level, item, usable_tables, |
| 3217 |
sargables); |
| 3218 |
*key_fields=merge_key_fields(org_key_fields,start_key_fields, |
| 3219 |
*key_fields,++(*and_level)); |
| 3220 |
} |
| 3221 |
} |
| 3222 |
return; |
| 3223 |
} |
| 3224 |
|
| 3225 |
/* |
| 3226 |
Subquery optimization: Conditions that are pushed down into subqueries |
| 3227 |
are wrapped into Item_func_trig_cond. We process the wrapped condition |
| 3228 |
but need to set cond_guard for KEYUSE elements generated from it. |
| 3229 |
*/ |
| 3230 |
{ |
| 3231 |
if (cond->type() == Item::FUNC_ITEM && |
| 3232 |
((Item_func*)cond)->functype() == Item_func::TRIG_COND_FUNC) |
| 3233 |
{ |
| 3234 |
Item *cond_arg= ((Item_func*)cond)->arguments()[0]; |
| 3235 |
if (!join->group_list && !join->order && |
| 3236 |
join->unit->item && |
| 3237 |
join->unit->item->substype() == Item_subselect::IN_SUBS && |
| 3238 |
!join->unit->first_select()->next_select()) |
| 3239 |
{ |
| 3240 |
KEY_FIELD *save= *key_fields; |
| 3241 |
add_key_fields(join, key_fields, and_level, cond_arg, usable_tables, |
| 3242 |
sargables); |
| 3243 |
// Indicate that this ref access candidate is for subquery lookup: |
| 3244 |
for (; save != *key_fields; save++) |
| 3245 |
save->cond_guard= ((Item_func_trig_cond*)cond)->get_trig_var(); |
| 3246 |
} |
| 3247 |
return; |
| 3248 |
} |
| 3249 |
} |
| 3250 |
|
| 3251 |
/* If item is of type 'field op field/constant' add it to key_fields */ |
| 3252 |
if (cond->type() != Item::FUNC_ITEM) |
| 3253 |
return; |
| 3254 |
Item_func *cond_func= (Item_func*) cond; |
| 3255 |
switch (cond_func->select_optimize()) { |
| 3256 |
case Item_func::OPTIMIZE_NONE: |
| 3257 |
break; |
| 3258 |
case Item_func::OPTIMIZE_KEY: |
| 3259 |
{ |
| 3260 |
Item **values; |
| 3261 |
// BETWEEN, IN, NE |
| 3262 |
if (cond_func->key_item()->real_item()->type() == Item::FIELD_ITEM && |
| 3263 |
!(cond_func->used_tables() & OUTER_REF_TABLE_BIT)) |
| 3264 |
{ |
| 3265 |
values= cond_func->arguments()+1; |
| 3266 |
if (cond_func->functype() == Item_func::NE_FUNC && |
| 3267 |
cond_func->arguments()[1]->real_item()->type() == Item::FIELD_ITEM && |
| 3268 |
!(cond_func->arguments()[0]->used_tables() & OUTER_REF_TABLE_BIT)) |
| 3269 |
values--; |
| 3270 |
DBUG_ASSERT(cond_func->functype() != Item_func::IN_FUNC || |
| 3271 |
cond_func->argument_count() != 2); |
| 3272 |
add_key_equal_fields(key_fields, *and_level, cond_func, |
| 3273 |
(Item_field*) (cond_func->key_item()->real_item()), |
| 3274 |
0, values, |
| 3275 |
cond_func->argument_count()-1, |
| 3276 |
usable_tables, sargables); |
| 3277 |
} |
| 3278 |
if (cond_func->functype() == Item_func::BETWEEN) |
| 3279 |
{ |
| 3280 |
values= cond_func->arguments(); |
| 3281 |
for (uint i= 1 ; i < cond_func->argument_count() ; i++) |
| 3282 |
{ |
| 3283 |
Item_field *field_item; |
| 3284 |
if (cond_func->arguments()[i]->real_item()->type() == Item::FIELD_ITEM |
| 3285 |
| |