opt_range.cc 263 KB
Newer Older
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
/* Copyright (C) 2000 MySQL AB & MySQL Finland AB & TCX DataKonsult AB

   This program is free software; you can redistribute it and/or modify
   it under the terms of the GNU General Public License as published by
   the Free Software Foundation; either version 2 of the License, or
   (at your option) any later version.

   This program is distributed in the hope that it will be useful,
   but WITHOUT ANY WARRANTY; without even the implied warranty of
   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
   GNU General Public License for more details.

   You should have received a copy of the GNU General Public License
   along with this program; if not, write to the Free Software
   Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA */

17 18 19 20 21
/*
  TODO:
  Fix that MAYBE_KEY are stored in the tree so that we can detect use
  of full hash keys for queries like:

22 23
  select s.id, kws.keyword_id from sites as s,kws where s.id=kws.site_id and kws.keyword_id in (204,205);

24 25
*/

26 27
/*
  Classes in this file are used in the following way:
28 29
  1. For a selection condition a tree of SEL_IMERGE/SEL_TREE/SEL_ARG objects
     is created. #of rows in table and index statistics are ignored at this
30
     step.
31 32 33 34
  2. Created SEL_TREE and index stats data are used to construct a
     TABLE_READ_PLAN-derived object (TRP_*). Several 'candidate' table read
     plans may be created.
  3. The least expensive table read plan is used to create a tree of
35 36 37 38
     QUICK_SELECT_I-derived objects which are later used for row retrieval.
     QUICK_RANGEs are also created in this step.
*/

bk@work.mysql.com's avatar
bk@work.mysql.com committed
39 40 41 42 43 44 45 46 47 48 49 50 51
#ifdef __GNUC__
#pragma implementation				// gcc: Class implementation
#endif

#include "mysql_priv.h"
#include <m_ctype.h>
#include "sql_select.h"

#ifndef EXTRA_DEBUG
#define test_rb_tree(A,B) {}
#define test_use_count(A) {}
#endif

52
/*
53
  Convert double value to #rows. Currently this does floor(), and we
54 55
  might consider using round() instead.
*/
56
#define double2rows(x) ((ha_rows)(x))
57

bk@work.mysql.com's avatar
bk@work.mysql.com committed
58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82
static int sel_cmp(Field *f,char *a,char *b,uint8 a_flag,uint8 b_flag);

static char is_null_string[2]= {1,0};

class SEL_ARG :public Sql_alloc
{
public:
  uint8 min_flag,max_flag,maybe_flag;
  uint8 part;					// Which key part
  uint8 maybe_null;
  uint16 elements;				// Elements in tree
  ulong use_count;				// use of this sub_tree
  Field *field;
  char *min_value,*max_value;			// Pointer to range

  SEL_ARG *left,*right,*next,*prev,*parent,*next_key_part;
  enum leaf_color { BLACK,RED } color;
  enum Type { IMPOSSIBLE, MAYBE, MAYBE_KEY, KEY_RANGE } type;

  SEL_ARG() {}
  SEL_ARG(SEL_ARG &);
  SEL_ARG(Field *,const char *,const char *);
  SEL_ARG(Field *field, uint8 part, char *min_value, char *max_value,
	  uint8 min_flag, uint8 max_flag, uint8 maybe_flag);
  SEL_ARG(enum Type type_arg)
83 84 85
    :elements(1),use_count(1),left(0),next_key_part(0),color(BLACK),
     type(type_arg)
  {}
bk@work.mysql.com's avatar
bk@work.mysql.com committed
86 87
  inline bool is_same(SEL_ARG *arg)
  {
88
    if (type != arg->type || part != arg->part)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186
      return 0;
    if (type != KEY_RANGE)
      return 1;
    return cmp_min_to_min(arg) == 0 && cmp_max_to_max(arg) == 0;
  }
  inline void merge_flags(SEL_ARG *arg) { maybe_flag|=arg->maybe_flag; }
  inline void maybe_smaller() { maybe_flag=1; }
  inline int cmp_min_to_min(SEL_ARG* arg)
  {
    return sel_cmp(field,min_value, arg->min_value, min_flag, arg->min_flag);
  }
  inline int cmp_min_to_max(SEL_ARG* arg)
  {
    return sel_cmp(field,min_value, arg->max_value, min_flag, arg->max_flag);
  }
  inline int cmp_max_to_max(SEL_ARG* arg)
  {
    return sel_cmp(field,max_value, arg->max_value, max_flag, arg->max_flag);
  }
  inline int cmp_max_to_min(SEL_ARG* arg)
  {
    return sel_cmp(field,max_value, arg->min_value, max_flag, arg->min_flag);
  }
  SEL_ARG *clone_and(SEL_ARG* arg)
  {						// Get overlapping range
    char *new_min,*new_max;
    uint8 flag_min,flag_max;
    if (cmp_min_to_min(arg) >= 0)
    {
      new_min=min_value; flag_min=min_flag;
    }
    else
    {
      new_min=arg->min_value; flag_min=arg->min_flag; /* purecov: deadcode */
    }
    if (cmp_max_to_max(arg) <= 0)
    {
      new_max=max_value; flag_max=max_flag;
    }
    else
    {
      new_max=arg->max_value; flag_max=arg->max_flag;
    }
    return new SEL_ARG(field, part, new_min, new_max, flag_min, flag_max,
		       test(maybe_flag && arg->maybe_flag));
  }
  SEL_ARG *clone_first(SEL_ARG *arg)
  {						// min <= X < arg->min
    return new SEL_ARG(field,part, min_value, arg->min_value,
		       min_flag, arg->min_flag & NEAR_MIN ? 0 : NEAR_MAX,
		       maybe_flag | arg->maybe_flag);
  }
  SEL_ARG *clone_last(SEL_ARG *arg)
  {						// min <= X <= key_max
    return new SEL_ARG(field, part, min_value, arg->max_value,
		       min_flag, arg->max_flag, maybe_flag | arg->maybe_flag);
  }
  SEL_ARG *clone(SEL_ARG *new_parent,SEL_ARG **next);

  bool copy_min(SEL_ARG* arg)
  {						// Get overlapping range
    if (cmp_min_to_min(arg) > 0)
    {
      min_value=arg->min_value; min_flag=arg->min_flag;
      if ((max_flag & (NO_MAX_RANGE | NO_MIN_RANGE)) ==
	  (NO_MAX_RANGE | NO_MIN_RANGE))
	return 1;				// Full range
    }
    maybe_flag|=arg->maybe_flag;
    return 0;
  }
  bool copy_max(SEL_ARG* arg)
  {						// Get overlapping range
    if (cmp_max_to_max(arg) <= 0)
    {
      max_value=arg->max_value; max_flag=arg->max_flag;
      if ((max_flag & (NO_MAX_RANGE | NO_MIN_RANGE)) ==
	  (NO_MAX_RANGE | NO_MIN_RANGE))
	return 1;				// Full range
    }
    maybe_flag|=arg->maybe_flag;
    return 0;
  }

  void copy_min_to_min(SEL_ARG *arg)
  {
    min_value=arg->min_value; min_flag=arg->min_flag;
  }
  void copy_min_to_max(SEL_ARG *arg)
  {
    max_value=arg->min_value;
    max_flag=arg->min_flag & NEAR_MIN ? 0 : NEAR_MAX;
  }
  void copy_max_to_min(SEL_ARG *arg)
  {
    min_value=arg->max_value;
    min_flag=arg->max_flag & NEAR_MAX ? 0 : NEAR_MIN;
  }
187
  void store_min(uint length,char **min_key,uint min_key_flag)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
188
  {
189 190 191
    if ((min_flag & GEOM_FLAG) ||
        (!(min_flag & NO_MIN_RANGE) &&
	!(min_key_flag & (NO_MIN_RANGE | NEAR_MIN))))
bk@work.mysql.com's avatar
bk@work.mysql.com committed
192 193 194 195
    {
      if (maybe_null && *min_value)
      {
	**min_key=1;
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
196
	bzero(*min_key+1,length-1);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
197 198
      }
      else
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
199 200
	memcpy(*min_key,min_value,length);
      (*min_key)+= length;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
201
    }
202
  }
bk@work.mysql.com's avatar
bk@work.mysql.com committed
203 204 205
  void store(uint length,char **min_key,uint min_key_flag,
	     char **max_key, uint max_key_flag)
  {
206
    store_min(length, min_key, min_key_flag);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
207 208 209 210 211 212
    if (!(max_flag & NO_MAX_RANGE) &&
	!(max_key_flag & (NO_MAX_RANGE | NEAR_MAX)))
    {
      if (maybe_null && *max_value)
      {
	**max_key=1;
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
213
	bzero(*max_key+1,length-1);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
214 215
      }
      else
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
216 217
	memcpy(*max_key,max_value,length);
      (*max_key)+= length;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
218 219 220 221 222 223
    }
  }

  void store_min_key(KEY_PART *key,char **range_key, uint *range_key_flag)
  {
    SEL_ARG *key_tree= first();
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
224
    key_tree->store(key[key_tree->part].store_length,
bk@work.mysql.com's avatar
bk@work.mysql.com committed
225 226 227 228 229 230 231 232 233 234 235 236
		    range_key,*range_key_flag,range_key,NO_MAX_RANGE);
    *range_key_flag|= key_tree->min_flag;
    if (key_tree->next_key_part &&
	key_tree->next_key_part->part == key_tree->part+1 &&
	!(*range_key_flag & (NO_MIN_RANGE | NEAR_MIN)) &&
	key_tree->next_key_part->type == SEL_ARG::KEY_RANGE)
      key_tree->next_key_part->store_min_key(key,range_key, range_key_flag);
  }

  void store_max_key(KEY_PART *key,char **range_key, uint *range_key_flag)
  {
    SEL_ARG *key_tree= last();
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
237
    key_tree->store(key[key_tree->part].store_length,
bk@work.mysql.com's avatar
bk@work.mysql.com committed
238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290
		    range_key, NO_MIN_RANGE, range_key,*range_key_flag);
    (*range_key_flag)|= key_tree->max_flag;
    if (key_tree->next_key_part &&
	key_tree->next_key_part->part == key_tree->part+1 &&
	!(*range_key_flag & (NO_MAX_RANGE | NEAR_MAX)) &&
	key_tree->next_key_part->type == SEL_ARG::KEY_RANGE)
      key_tree->next_key_part->store_max_key(key,range_key, range_key_flag);
  }

  SEL_ARG *insert(SEL_ARG *key);
  SEL_ARG *tree_delete(SEL_ARG *key);
  SEL_ARG *find_range(SEL_ARG *key);
  SEL_ARG *rb_insert(SEL_ARG *leaf);
  friend SEL_ARG *rb_delete_fixup(SEL_ARG *root,SEL_ARG *key, SEL_ARG *par);
#ifdef EXTRA_DEBUG
  friend int test_rb_tree(SEL_ARG *element,SEL_ARG *parent);
  void test_use_count(SEL_ARG *root);
#endif
  SEL_ARG *first();
  SEL_ARG *last();
  void make_root();
  inline bool simple_key()
  {
    return !next_key_part && elements == 1;
  }
  void increment_use_count(long count)
  {
    if (next_key_part)
    {
      next_key_part->use_count+=count;
      count*= (next_key_part->use_count-count);
      for (SEL_ARG *pos=next_key_part->first(); pos ; pos=pos->next)
	if (pos->next_key_part)
	  pos->increment_use_count(count);
    }
  }
  void free_tree()
  {
    for (SEL_ARG *pos=first(); pos ; pos=pos->next)
      if (pos->next_key_part)
      {
	pos->next_key_part->use_count--;
	pos->next_key_part->free_tree();
      }
  }

  inline SEL_ARG **parent_ptr()
  {
    return parent->left == this ? &parent->left : &parent->right;
  }
  SEL_ARG *clone_tree();
};

291
class SEL_IMERGE;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
292

293

bk@work.mysql.com's avatar
bk@work.mysql.com committed
294 295 296 297 298
class SEL_TREE :public Sql_alloc
{
public:
  enum Type { IMPOSSIBLE, ALWAYS, MAYBE, KEY, KEY_SMALLER } type;
  SEL_TREE(enum Type type_arg) :type(type_arg) {}
299
  SEL_TREE() :type(KEY)
300
  {
301
    keys_map.clear_all();
302 303
    bzero((char*) keys,sizeof(keys));
  }
bk@work.mysql.com's avatar
bk@work.mysql.com committed
304
  SEL_ARG *keys[MAX_KEY];
305 306
  key_map keys_map;        /* bitmask of non-NULL elements in keys */

307 308
  /*
    Possible ways to read rows using index_merge. The list is non-empty only
309 310 311
    if type==KEY. Currently can be non empty only if keys_map.is_clear_all().
  */
  List<SEL_IMERGE> merges;
312

313 314
  /* The members below are filled/used only after get_mm_tree is done */
  key_map ror_scans_map;   /* bitmask of ROR scan-able elements in keys */
315
  uint    n_ror_scans;     /* number of set bits in ror_scans_map */
316 317 318 319

  struct st_ror_scan_info **ror_scans;     /* list of ROR key scans */
  struct st_ror_scan_info **ror_scans_end; /* last ROR scan */
  /* Note that #records for each key scan is stored in table->quick_rows */
bk@work.mysql.com's avatar
bk@work.mysql.com committed
320 321 322 323
};


typedef struct st_qsel_param {
324
  THD	*thd;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
325
  TABLE *table;
326 327
  KEY_PART *key_parts,*key_parts_end;
  KEY_PART *key[MAX_KEY]; /* First key parts of keys used in the query */
328 329
  MEM_ROOT *mem_root;
  table_map prev_tables,read_tables,current_table;
330
  uint baseflag, max_key_part, range_count;
331

332 333 334
  uint keys; /* number of keys used in the query */

  /* used_key_no -> table_key_no translation table */
335
  uint real_keynr[MAX_KEY];
336

bk@work.mysql.com's avatar
bk@work.mysql.com committed
337 338
  char min_key[MAX_KEY_LENGTH+MAX_FIELD_WIDTH],
    max_key[MAX_KEY_LENGTH+MAX_FIELD_WIDTH];
339
  bool quick;				// Don't calulate possible keys
340
  COND *cond;
341

342
  uint fields_bitmap_size;
343 344 345 346
  MY_BITMAP needed_fields;    /* bitmask of fields needed by the query */

  key_map *needed_reg;        /* ptr to SQL_SELECT::needed_reg */

347 348
  uint *imerge_cost_buff;     /* buffer for index_merge cost estimates */
  uint imerge_cost_buff_size; /* size of the buffer */
349 350 351

 /* TRUE if last checked tree->key can be used for ROR-scan */
  bool is_ror_scan;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
352 353
} PARAM;

354 355 356 357 358
class TABLE_READ_PLAN;
  class TRP_RANGE;
  class TRP_ROR_INTERSECT;
  class TRP_ROR_UNION;
  class TRP_ROR_INDEX_MERGE;
359
  class TRP_GROUP_MIN_MAX;
360 361 362

struct st_ror_scan_info;

363
static SEL_TREE * get_mm_parts(PARAM *param,COND *cond_func,Field *field,
bk@work.mysql.com's avatar
bk@work.mysql.com committed
364 365
			       Item_func::Functype type,Item *value,
			       Item_result cmp_type);
366 367
static SEL_ARG *get_mm_leaf(PARAM *param,COND *cond_func,Field *field,
			    KEY_PART *key_part,
bk@work.mysql.com's avatar
bk@work.mysql.com committed
368 369
			    Item_func::Functype type,Item *value);
static SEL_TREE *get_mm_tree(PARAM *param,COND *cond);
370 371

static bool is_key_scan_ror(PARAM *param, uint keynr, uint8 nparts);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
372 373 374 375 376
static ha_rows check_quick_select(PARAM *param,uint index,SEL_ARG *key_tree);
static ha_rows check_quick_keys(PARAM *param,uint index,SEL_ARG *key_tree,
				char *min_key,uint min_key_flag,
				char *max_key, uint max_key_flag);

377
QUICK_RANGE_SELECT *get_quick_select(PARAM *param,uint index,
378
                                     SEL_ARG *key_tree,
379
                                     MEM_ROOT *alloc = NULL);
380
static TRP_RANGE *get_key_scans_params(PARAM *param, SEL_TREE *tree,
381
                                       bool index_read_must_be_used,
382 383 384 385 386 387
                                       double read_time);
static
TRP_ROR_INTERSECT *get_best_ror_intersect(const PARAM *param, SEL_TREE *tree,
                                          double read_time,
                                          bool *are_all_covering);
static
388 389
TRP_ROR_INTERSECT *get_best_covering_ror_intersect(PARAM *param,
                                                   SEL_TREE *tree,
390 391 392 393
                                                   double read_time);
static
TABLE_READ_PLAN *get_best_disjunct_quick(PARAM *param, SEL_IMERGE *imerge,
                                         double read_time);
394 395
static
TRP_GROUP_MIN_MAX *get_best_group_min_max(PARAM *param, SEL_TREE *tree);
396
static int get_index_merge_params(PARAM *param, key_map& needed_reg,
397
                           SEL_IMERGE *imerge, double *read_time,
398
                           ha_rows* imerge_rows);
399
inline double get_index_only_read_time(const PARAM* param, ha_rows records,
400 401
                                       int keynr);

bk@work.mysql.com's avatar
bk@work.mysql.com committed
402
#ifndef DBUG_OFF
403 404
static void print_sel_tree(PARAM *param, SEL_TREE *tree, key_map *tree_map,
                           const char *msg);
405 406
static void print_ror_scans_arr(TABLE *table, const char *msg,
                                struct st_ror_scan_info **start,
407 408 409
                                struct st_ror_scan_info **end);
static void print_rowid(byte* val, int len);
static void print_quick(QUICK_SELECT_I *quick, const key_map *needed_reg);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
410
#endif
411

bk@work.mysql.com's avatar
bk@work.mysql.com committed
412 413 414 415 416 417
static SEL_TREE *tree_and(PARAM *param,SEL_TREE *tree1,SEL_TREE *tree2);
static SEL_TREE *tree_or(PARAM *param,SEL_TREE *tree1,SEL_TREE *tree2);
static SEL_ARG *sel_add(SEL_ARG *key1,SEL_ARG *key2);
static SEL_ARG *key_or(SEL_ARG *key1,SEL_ARG *key2);
static SEL_ARG *key_and(SEL_ARG *key1,SEL_ARG *key2,uint clone_flag);
static bool get_range(SEL_ARG **e1,SEL_ARG **e2,SEL_ARG *root1);
418
bool get_quick_keys(PARAM *param,QUICK_RANGE_SELECT *quick,KEY_PART *key,
bk@work.mysql.com's avatar
bk@work.mysql.com committed
419 420 421 422 423
			   SEL_ARG *key_tree,char *min_key,uint min_key_flag,
			   char *max_key,uint max_key_flag);
static bool eq_tree(SEL_ARG* a,SEL_ARG *b);

static SEL_ARG null_element(SEL_ARG::IMPOSSIBLE);
424
static bool null_part_in_key(KEY_PART *key_part, const char *key,
425
                             uint length);
426 427 428 429
bool sel_trees_can_be_ored(SEL_TREE *tree1, SEL_TREE *tree2, PARAM* param);


/*
430
  SEL_IMERGE is a list of possible ways to do index merge, i.e. it is
431
  a condition in the following form:
432
   (t_1||t_2||...||t_N) && (next)
433

434
  where all t_i are SEL_TREEs, next is another SEL_IMERGE and no pair
435 436 437 438 439 440 441 442 443 444 445
  (t_i,t_j) contains SEL_ARGS for the same index.

  SEL_TREE contained in SEL_IMERGE always has merges=NULL.

  This class relies on memory manager to do the cleanup.
*/

class SEL_IMERGE : public Sql_alloc
{
  enum { PREALLOCED_TREES= 10};
public:
446
  SEL_TREE *trees_prealloced[PREALLOCED_TREES];
447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463
  SEL_TREE **trees;             /* trees used to do index_merge   */
  SEL_TREE **trees_next;        /* last of these trees            */
  SEL_TREE **trees_end;         /* end of allocated space         */

  SEL_ARG  ***best_keys;        /* best keys to read in SEL_TREEs */

  SEL_IMERGE() :
    trees(&trees_prealloced[0]),
    trees_next(trees),
    trees_end(trees + PREALLOCED_TREES)
  {}
  int or_sel_tree(PARAM *param, SEL_TREE *tree);
  int or_sel_tree_with_checks(PARAM *param, SEL_TREE *new_tree);
  int or_sel_imerge_with_checks(PARAM *param, SEL_IMERGE* imerge);
};


464
/*
465 466
  Add SEL_TREE to this index_merge without any checks,

467 468
  NOTES
    This function implements the following:
469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500
      (x_1||...||x_N) || t = (x_1||...||x_N||t), where x_i, t are SEL_TREEs

  RETURN
     0 - OK
    -1 - Out of memory.
*/

int SEL_IMERGE::or_sel_tree(PARAM *param, SEL_TREE *tree)
{
  if (trees_next == trees_end)
  {
    const int realloc_ratio= 2;		/* Double size for next round */
    uint old_elements= (trees_end - trees);
    uint old_size= sizeof(SEL_TREE**) * old_elements;
    uint new_size= old_size * realloc_ratio;
    SEL_TREE **new_trees;
    if (!(new_trees= (SEL_TREE**)alloc_root(param->mem_root, new_size)))
      return -1;
    memcpy(new_trees, trees, old_size);
    trees=      new_trees;
    trees_next= trees + old_elements;
    trees_end=  trees + old_elements * realloc_ratio;
  }
  *(trees_next++)= tree;
  return 0;
}


/*
  Perform OR operation on this SEL_IMERGE and supplied SEL_TREE new_tree,
  combining new_tree with one of the trees in this SEL_IMERGE if they both
  have SEL_ARGs for the same key.
501

502 503 504 505 506
  SYNOPSIS
    or_sel_tree_with_checks()
      param    PARAM from SQL_SELECT::test_quick_select
      new_tree SEL_TREE with type KEY or KEY_SMALLER.

507
  NOTES
508
    This does the following:
509 510
    (t_1||...||t_k)||new_tree =
     either
511 512 513
       = (t_1||...||t_k||new_tree)
     or
       = (t_1||....||(t_j|| new_tree)||...||t_k),
514

515
     where t_i, y are SEL_TREEs.
516 517
    new_tree is combined with the first t_j it has a SEL_ARG on common
    key with. As a consequence of this, choice of keys to do index_merge
518 519
    read may depend on the order of conditions in WHERE part of the query.

520
  RETURN
521
    0  OK
522
    1  One of the trees was combined with new_tree to SEL_TREE::ALWAYS,
523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545
       and (*this) should be discarded.
   -1  An error occurred.
*/

int SEL_IMERGE::or_sel_tree_with_checks(PARAM *param, SEL_TREE *new_tree)
{
  for (SEL_TREE** tree = trees;
       tree != trees_next;
       tree++)
  {
    if (sel_trees_can_be_ored(*tree, new_tree, param))
    {
      *tree = tree_or(param, *tree, new_tree);
      if (!*tree)
        return 1;
      if (((*tree)->type == SEL_TREE::MAYBE) ||
          ((*tree)->type == SEL_TREE::ALWAYS))
        return 1;
      /* SEL_TREE::IMPOSSIBLE is impossible here */
      return 0;
    }
  }

546
  /* New tree cannot be combined with any of existing trees. */
547 548 549 550 551 552 553 554 555
  return or_sel_tree(param, new_tree);
}


/*
  Perform OR operation on this index_merge and supplied index_merge list.

  RETURN
    0 - OK
556
    1 - One of conditions in result is always TRUE and this SEL_IMERGE
557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573
        should be discarded.
   -1 - An error occurred
*/

int SEL_IMERGE::or_sel_imerge_with_checks(PARAM *param, SEL_IMERGE* imerge)
{
  for (SEL_TREE** tree= imerge->trees;
       tree != imerge->trees_next;
       tree++)
  {
    if (or_sel_tree_with_checks(param, *tree))
      return 1;
  }
  return 0;
}


574
/*
575
  Perform AND operation on two index_merge lists and store result in *im1.
576 577 578 579 580 581 582 583 584 585 586
*/

inline void imerge_list_and_list(List<SEL_IMERGE> *im1, List<SEL_IMERGE> *im2)
{
  im1->concat(im2);
}


/*
  Perform OR operation on 2 index_merge lists, storing result in first list.

587
  NOTES
588 589 590
    The following conversion is implemented:
     (a_1 &&...&& a_N)||(b_1 &&...&& b_K) = AND_i,j(a_i || b_j) =>
      => (a_1||b_1).
591 592

    i.e. all conjuncts except the first one are currently dropped.
593 594
    This is done to avoid producing N*K ways to do index_merge.

monty@mysql.com's avatar
monty@mysql.com committed
595
    If (a_1||b_1) produce a condition that is always TRUE, NULL is returned
596
    and index_merge is discarded (while it is actually possible to try
597
    harder).
598

599 600
    As a consequence of this, choice of keys to do index_merge read may depend
    on the order of conditions in WHERE part of the query.
601 602

  RETURN
603
    0     OK, result is stored in *im1
604 605 606
    other Error, both passed lists are unusable
*/

607
int imerge_list_or_list(PARAM *param,
608 609 610 611 612 613
                        List<SEL_IMERGE> *im1,
                        List<SEL_IMERGE> *im2)
{
  SEL_IMERGE *imerge= im1->head();
  im1->empty();
  im1->push_back(imerge);
614

615 616 617 618 619 620 621 622
  return imerge->or_sel_imerge_with_checks(param, im2->head());
}


/*
  Perform OR operation on index_merge list and key tree.

  RETURN
623
    0     OK, result is stored in *im1.
624 625 626
    other Error
*/

627
int imerge_list_or_tree(PARAM *param,
628 629 630 631 632 633 634 635 636 637 638 639
                        List<SEL_IMERGE> *im1,
                        SEL_TREE *tree)
{
  SEL_IMERGE *imerge;
  List_iterator<SEL_IMERGE> it(*im1);
  while((imerge= it++))
  {
    if (imerge->or_sel_tree_with_checks(param, tree))
      it.remove();
  }
  return im1->is_empty();
}
bk@work.mysql.com's avatar
bk@work.mysql.com committed
640 641

/***************************************************************************
642
** Basic functions for SQL_SELECT and QUICK_RANGE_SELECT
bk@work.mysql.com's avatar
bk@work.mysql.com committed
643 644 645 646 647 648 649 650 651
***************************************************************************/

	/* make a select from mysql info
	   Error is set as following:
	   0 = ok
	   1 = Got some error (out of memory?)
	   */

SQL_SELECT *make_select(TABLE *head, table_map const_tables,
monty@mysql.com's avatar
monty@mysql.com committed
652 653 654 655
			table_map read_tables, COND *conds,
                        bool allow_null_cond,
                        int *error)
                        
bk@work.mysql.com's avatar
bk@work.mysql.com committed
656 657 658 659 660
{
  SQL_SELECT *select;
  DBUG_ENTER("make_select");

  *error=0;
661 662

  if (!conds && !allow_null_cond)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
663 664 665
    DBUG_RETURN(0);
  if (!(select= new SQL_SELECT))
  {
666 667
    *error= 1;			// out of memory
    DBUG_RETURN(0);		/* purecov: inspected */
bk@work.mysql.com's avatar
bk@work.mysql.com committed
668 669 670 671 672 673
  }
  select->read_tables=read_tables;
  select->const_tables=const_tables;
  select->head=head;
  select->cond=conds;

igor@hundin.mysql.fi's avatar
igor@hundin.mysql.fi committed
674
  if (head->sort.io_cache)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
675
  {
igor@hundin.mysql.fi's avatar
igor@hundin.mysql.fi committed
676
    select->file= *head->sort.io_cache;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
677 678
    select->records=(ha_rows) (select->file.end_of_file/
			       head->file->ref_length);
igor@hundin.mysql.fi's avatar
igor@hundin.mysql.fi committed
679 680
    my_free((gptr) (head->sort.io_cache),MYF(0));
    head->sort.io_cache=0;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
681 682 683 684 685 686 687
  }
  DBUG_RETURN(select);
}


SQL_SELECT::SQL_SELECT() :quick(0),cond(0),free_cond(0)
{
serg@serg.mylan's avatar
serg@serg.mylan committed
688
  quick_keys.clear_all(); needed_reg.clear_all();
bk@work.mysql.com's avatar
bk@work.mysql.com committed
689 690 691 692
  my_b_clear(&file);
}


693
void SQL_SELECT::cleanup()
bk@work.mysql.com's avatar
bk@work.mysql.com committed
694 695
{
  delete quick;
696
  quick= 0;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
697
  if (free_cond)
698 699
  {
    free_cond=0;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
700
    delete cond;
701
    cond= 0;
702
  }
bk@work.mysql.com's avatar
bk@work.mysql.com committed
703 704 705
  close_cached_file(&file);
}

706 707 708 709 710 711

SQL_SELECT::~SQL_SELECT()
{
  cleanup();
}

712
#undef index					// Fix for Unixware 7
bk@work.mysql.com's avatar
bk@work.mysql.com committed
713

sergefp@mysql.com's avatar
sergefp@mysql.com committed
714 715 716 717 718
QUICK_SELECT_I::QUICK_SELECT_I()
  :max_used_key_length(0),
   used_key_parts(0)
{}

719
QUICK_RANGE_SELECT::QUICK_RANGE_SELECT(THD *thd, TABLE *table, uint key_nr,
sergefp@mysql.com's avatar
sergefp@mysql.com committed
720
                                       bool no_alloc, MEM_ROOT *parent_alloc)
721
  :dont_free(0),error(0),free_file(0),in_range(0),cur_range(NULL),range(0)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
722
{
monty@mysql.com's avatar
monty@mysql.com committed
723
  sorted= 0;
sergefp@mysql.com's avatar
sergefp@mysql.com committed
724 725
  index= key_nr;
  head=  table;
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
726
  key_part_info= head->key_info[index].key_part;
727
  my_init_dynamic_array(&ranges, sizeof(QUICK_RANGE*), 16, 16);
sergefp@mysql.com's avatar
sergefp@mysql.com committed
728

sergefp@mysql.com's avatar
sergefp@mysql.com committed
729
  /* 'thd' is not accessible in QUICK_RANGE_SELECT::reset(). */
ingo@mysql.com's avatar
ingo@mysql.com committed
730 731 732 733 734 735
  multi_range_bufsiz= thd->variables.read_rnd_buff_size;
  multi_range_count= thd->variables.multi_range_count;
  multi_range_length= 0;
  multi_range= NULL;
  multi_range_buff= NULL;

sergefp@mysql.com's avatar
sergefp@mysql.com committed
736
  if (!no_alloc && !parent_alloc)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
737
  {
738 739
    // Allocates everything through the internal memroot
    init_sql_alloc(&alloc, thd->variables.range_alloc_block_size, 0);
740
    thd->mem_root= &alloc;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
741 742 743
  }
  else
    bzero((char*) &alloc,sizeof(alloc));
744 745
  file= head->file;
  record= head->record[0];
bk@work.mysql.com's avatar
bk@work.mysql.com committed
746 747
}

monty@mysql.com's avatar
monty@mysql.com committed
748

749 750
int QUICK_RANGE_SELECT::init()
{
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
751
  DBUG_ENTER("QUICK_RANGE_SELECT::init");
ingo@mysql.com's avatar
ingo@mysql.com committed
752

monty@mysql.com's avatar
monty@mysql.com committed
753 754 755 756 757 758 759 760 761 762
  if (file->inited == handler::NONE)
    DBUG_RETURN(error= file->ha_index_init(index));
  error= 0;
  DBUG_RETURN(0);
}


void QUICK_RANGE_SELECT::range_end()
{
  if (file->inited != handler::NONE)
763
    file->ha_index_or_rnd_end();
bk@work.mysql.com's avatar
bk@work.mysql.com committed
764 765
}

monty@mysql.com's avatar
monty@mysql.com committed
766

767
QUICK_RANGE_SELECT::~QUICK_RANGE_SELECT()
768
{
769
  DBUG_ENTER("QUICK_RANGE_SELECT::~QUICK_RANGE_SELECT");
770 771
  if (!dont_free)
  {
772 773
    /* file is NULL for CPK scan on covering ROR-intersection */
    if (file) 
774
    {
775 776 777 778 779 780 781
      range_end();
      file->extra(HA_EXTRA_NO_KEYREAD);
      if (free_file)
      {
        DBUG_PRINT("info", ("Freeing separate handler %p (free=%d)", file,
                            free_file));
        file->reset();
782
        file->external_lock(current_thd, F_UNLCK);
783 784
        file->close();
      }
785
    }
786
    delete_dynamic(&ranges); /* ranges are allocated in alloc */
787 788
    free_root(&alloc,MYF(0));
  }
ingo@mysql.com's avatar
ingo@mysql.com committed
789 790 791 792
  if (multi_range)
    my_free((char*) multi_range, MYF(0));
  if (multi_range_buff)
    my_free((char*) multi_range_buff, MYF(0));
793
  DBUG_VOID_RETURN;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
794 795
}

796

797
QUICK_INDEX_MERGE_SELECT::QUICK_INDEX_MERGE_SELECT(THD *thd_param,
798
                                                   TABLE *table)
sergefp@mysql.com's avatar
sergefp@mysql.com committed
799
  :pk_quick_select(NULL), thd(thd_param)
800
{
801
  DBUG_ENTER("QUICK_INDEX_MERGE_SELECT::QUICK_INDEX_MERGE_SELECT");
802 803
  index= MAX_KEY;
  head= table;
804
  bzero(&read_record, sizeof(read_record));
805
  init_sql_alloc(&alloc, thd->variables.range_alloc_block_size, 0);
806
  DBUG_VOID_RETURN;
807 808 809 810
}

int QUICK_INDEX_MERGE_SELECT::init()
{
sergefp@mysql.com's avatar
sergefp@mysql.com committed
811 812
  DBUG_ENTER("QUICK_INDEX_MERGE_SELECT::init");
  DBUG_RETURN(0);
813 814
}

815
int QUICK_INDEX_MERGE_SELECT::reset()
816
{
817
  DBUG_ENTER("QUICK_INDEX_MERGE_SELECT::reset");
sergefp@mysql.com's avatar
sergefp@mysql.com committed
818
  DBUG_RETURN(read_keys_and_merge());
819 820
}

821
bool
822 823
QUICK_INDEX_MERGE_SELECT::push_quick_back(QUICK_RANGE_SELECT *quick_sel_range)
{
824 825
  /*
    Save quick_select that does scan on clustered primary key as it will be
826
    processed separately.
827
  */
828
  if (head->file->primary_key_is_clustered() &&
829
      quick_sel_range->index == head->s->primary_key)
830 831 832 833
    pk_quick_select= quick_sel_range;
  else
    return quick_selects.push_back(quick_sel_range);
  return 0;
834 835 836 837
}

QUICK_INDEX_MERGE_SELECT::~QUICK_INDEX_MERGE_SELECT()
{
sergefp@mysql.com's avatar
sergefp@mysql.com committed
838 839
  List_iterator_fast<QUICK_RANGE_SELECT> quick_it(quick_selects);
  QUICK_RANGE_SELECT* quick;
840
  DBUG_ENTER("QUICK_INDEX_MERGE_SELECT::~QUICK_INDEX_MERGE_SELECT");
sergefp@mysql.com's avatar
sergefp@mysql.com committed
841 842 843
  quick_it.rewind();
  while ((quick= quick_it++))
    quick->file= NULL;
844
  quick_selects.delete_elements();
845
  delete pk_quick_select;
846
  free_root(&alloc,MYF(0));
847
  DBUG_VOID_RETURN;
848 849
}

850 851 852 853 854

QUICK_ROR_INTERSECT_SELECT::QUICK_ROR_INTERSECT_SELECT(THD *thd_param,
                                                       TABLE *table,
                                                       bool retrieve_full_rows,
                                                       MEM_ROOT *parent_alloc)
sergefp@mysql.com's avatar
sergefp@mysql.com committed
855 856
  : cpk_quick(NULL), thd(thd_param), need_to_fetch_row(retrieve_full_rows),
    scans_inited(false)
857 858
{
  index= MAX_KEY;
859
  head= table;
860 861
  record= head->record[0];
  if (!parent_alloc)
862
    init_sql_alloc(&alloc, thd->variables.range_alloc_block_size, 0);
863 864
  else
    bzero(&alloc, sizeof(MEM_ROOT));
865
  last_rowid= (byte*)alloc_root(parent_alloc? parent_alloc : &alloc,
866 867 868
                                head->file->ref_length);
}

869

870
/*
871 872 873
  Do post-constructor initialization.
  SYNOPSIS
    QUICK_ROR_INTERSECT_SELECT::init()
874

875 876 877 878 879
  RETURN
    0      OK
    other  Error code
*/

880 881
int QUICK_ROR_INTERSECT_SELECT::init()
{
sergefp@mysql.com's avatar
sergefp@mysql.com committed
882 883 884
  DBUG_ENTER("QUICK_ROR_INTERSECT_SELECT::init");
 /* Check if last_rowid was successfully allocated in ctor */
  DBUG_RETURN(!last_rowid);
885 886 887 888
}


/*
889 890 891 892
  Initialize this quick select to be a ROR-merged scan.

  SYNOPSIS
    QUICK_RANGE_SELECT::init_ror_merged_scan()
monty@mysql.com's avatar
monty@mysql.com committed
893
      reuse_handler If TRUE, use head->file, otherwise create a separate
894 895 896 897
                    handler object

  NOTES
    This function creates and prepares for subsequent use a separate handler
898
    object if it can't reuse head->file. The reason for this is that during
899 900 901
    ROR-merge several key scans are performed simultaneously, and a single
    handler is only capable of preserving context of a single key scan.

902
    In ROR-merge the quick select doing merge does full records retrieval,
903
    merged quick selects read only keys.
904 905

  RETURN
906 907 908 909
    0  ROR child scan initialized, ok to use.
    1  error
*/

910
int QUICK_RANGE_SELECT::init_ror_merged_scan(bool reuse_handler)
911 912
{
  handler *save_file= file;
913
  DBUG_ENTER("QUICK_RANGE_SELECT::init_ror_merged_scan");
914

915 916 917 918
  if (reuse_handler)
  {
    DBUG_PRINT("info", ("Reusing handler %p", file));
    if (file->extra(HA_EXTRA_KEYREAD) ||
919
        file->extra(HA_EXTRA_RETRIEVE_PRIMARY_KEY) ||
920 921 922 923
        init() || reset())
    {
      DBUG_RETURN(1);
    }
monty@mysql.com's avatar
monty@mysql.com committed
924
    DBUG_RETURN(0);
925 926 927 928 929 930 931 932
  }

  /* Create a separate handler object for this quick select */
  if (free_file)
  {
    /* already have own 'handler' object. */
    DBUG_RETURN(0);
  }
933

934
  THD *thd= current_thd;
935
  if (!(file= get_new_handler(head, head->s->db_type)))
936 937
    goto failure;
  DBUG_PRINT("info", ("Allocated new handler %p", file));
938
  if (file->ha_open(head->s->path, head->db_stat, HA_OPEN_IGNORE_IF_LOCKED))
939
  {
940
    /* Caller will free the memory */
941 942
    goto failure;
  }
943 944
  if (file->external_lock(thd, F_RDLCK))
    goto failure;
945 946

  if (file->extra(HA_EXTRA_KEYREAD) ||
947
      file->extra(HA_EXTRA_RETRIEVE_PRIMARY_KEY) ||
948 949
      init() || reset())
  {
950
    file->external_lock(thd, F_UNLCK);
951 952 953
    file->close();
    goto failure;
  }
monty@mysql.com's avatar
monty@mysql.com committed
954
  free_file= TRUE;
955 956 957 958 959 960 961 962
  last_rowid= file->ref;
  DBUG_RETURN(0);

failure:
  file= save_file;
  DBUG_RETURN(1);
}

963 964 965 966 967

/*
  Initialize this quick select to be a part of a ROR-merged scan.
  SYNOPSIS
    QUICK_ROR_INTERSECT_SELECT::init_ror_merged_scan()
monty@mysql.com's avatar
monty@mysql.com committed
968
      reuse_handler If TRUE, use head->file, otherwise create separate
969
                    handler object.
970
  RETURN
971 972 973 974
    0     OK
    other error code
*/
int QUICK_ROR_INTERSECT_SELECT::init_ror_merged_scan(bool reuse_handler)
975 976 977
{
  List_iterator_fast<QUICK_RANGE_SELECT> quick_it(quick_selects);
  QUICK_RANGE_SELECT* quick;
978
  DBUG_ENTER("QUICK_ROR_INTERSECT_SELECT::init_ror_merged_scan");
979 980

  /* Initialize all merged "children" quick selects */
sergefp@mysql.com's avatar
sergefp@mysql.com committed
981
  DBUG_ASSERT(!need_to_fetch_row || reuse_handler);
982 983 984
  if (!need_to_fetch_row && reuse_handler)
  {
    quick= quick_it++;
985
    /*
986
      There is no use of this->file. Use it for the first of merged range
987 988
      selects.
    */
monty@mysql.com's avatar
monty@mysql.com committed
989
    if (quick->init_ror_merged_scan(TRUE))
990 991 992 993 994
      DBUG_RETURN(1);
    quick->file->extra(HA_EXTRA_KEYREAD_PRESERVE_FIELDS);
  }
  while((quick= quick_it++))
  {
monty@mysql.com's avatar
monty@mysql.com committed
995
    if (quick->init_ror_merged_scan(FALSE))
996 997
      DBUG_RETURN(1);
    quick->file->extra(HA_EXTRA_KEYREAD_PRESERVE_FIELDS);
998
    /* All merged scans share the same record buffer in intersection. */
999 1000 1001
    quick->record= head->record[0];
  }

monty@mysql.com's avatar
monty@mysql.com committed
1002
  if (need_to_fetch_row && head->file->ha_rnd_init(1))
1003 1004 1005 1006 1007 1008 1009
  {
    DBUG_PRINT("error", ("ROR index_merge rnd_init call failed"));
    DBUG_RETURN(1);
  }
  DBUG_RETURN(0);
}

1010

1011
/*
1012 1013 1014 1015 1016 1017 1018 1019
  Initialize quick select for row retrieval.
  SYNOPSIS
    reset()
  RETURN
    0      OK
    other  Error code
*/

1020 1021 1022
int QUICK_ROR_INTERSECT_SELECT::reset()
{
  DBUG_ENTER("QUICK_ROR_INTERSECT_SELECT::reset");
sergefp@mysql.com's avatar
sergefp@mysql.com committed
1023 1024 1025 1026 1027 1028 1029 1030
  if (!scans_inited && init_ror_merged_scan(TRUE))
    DBUG_RETURN(1);
  scans_inited= true;
  List_iterator_fast<QUICK_RANGE_SELECT> it(quick_selects);
  QUICK_RANGE_SELECT *quick;
  while ((quick= it++))
    quick->reset();
  DBUG_RETURN(0);
1031 1032
}

1033 1034 1035

/*
  Add a merged quick select to this ROR-intersection quick select.
1036

1037 1038 1039 1040 1041 1042
  SYNOPSIS
    QUICK_ROR_INTERSECT_SELECT::push_quick_back()
      quick Quick select to be added. The quick select must return
            rows in rowid order.
  NOTES
    This call can only be made before init() is called.
1043

1044
  RETURN
1045
    FALSE OK
monty@mysql.com's avatar
monty@mysql.com committed
1046
    TRUE  Out of memory.
1047 1048
*/

1049
bool
1050 1051
QUICK_ROR_INTERSECT_SELECT::push_quick_back(QUICK_RANGE_SELECT *quick)
{
1052
  return quick_selects.push_back(quick);
1053 1054 1055
}

QUICK_ROR_INTERSECT_SELECT::~QUICK_ROR_INTERSECT_SELECT()
1056
{
1057
  DBUG_ENTER("QUICK_ROR_INTERSECT_SELECT::~QUICK_ROR_INTERSECT_SELECT");
1058
  quick_selects.delete_elements();
1059 1060
  delete cpk_quick;
  free_root(&alloc,MYF(0));
monty@mysql.com's avatar
monty@mysql.com committed
1061 1062
  if (need_to_fetch_row && head->file->inited != handler::NONE)
    head->file->ha_rnd_end();
1063 1064 1065
  DBUG_VOID_RETURN;
}

monty@mysql.com's avatar
monty@mysql.com committed
1066

1067 1068
QUICK_ROR_UNION_SELECT::QUICK_ROR_UNION_SELECT(THD *thd_param,
                                               TABLE *table)
sergefp@mysql.com's avatar
sergefp@mysql.com committed
1069
  : thd(thd_param), scans_inited(false)
1070 1071 1072 1073 1074 1075
{
  index= MAX_KEY;
  head= table;
  rowid_length= table->file->ref_length;
  record= head->record[0];
  init_sql_alloc(&alloc, thd->variables.range_alloc_block_size, 0);
monty@mysql.com's avatar
monty@mysql.com committed
1076
  thd_param->mem_root= &alloc;
1077 1078
}

1079 1080 1081 1082 1083

/*
  Do post-constructor initialization.
  SYNOPSIS
    QUICK_ROR_UNION_SELECT::init()
1084

1085 1086 1087 1088 1089
  RETURN
    0      OK
    other  Error code
*/

1090 1091
int QUICK_ROR_UNION_SELECT::init()
{
sergefp@mysql.com's avatar
sergefp@mysql.com committed
1092
  DBUG_ENTER("QUICK_ROR_UNION_SELECT::init");
1093
  if (init_queue(&queue, quick_selects.elements, 0,
monty@mysql.com's avatar
monty@mysql.com committed
1094
                 FALSE , QUICK_ROR_UNION_SELECT::queue_cmp,
1095 1096 1097
                 (void*) this))
  {
    bzero(&queue, sizeof(QUEUE));
sergefp@mysql.com's avatar
sergefp@mysql.com committed
1098
    DBUG_RETURN(1);
1099
  }
1100

1101
  if (!(cur_rowid= (byte*)alloc_root(&alloc, 2*head->file->ref_length)))
sergefp@mysql.com's avatar
sergefp@mysql.com committed
1102
    DBUG_RETURN(1);
1103
  prev_rowid= cur_rowid + head->file->ref_length;
sergefp@mysql.com's avatar
sergefp@mysql.com committed
1104
  DBUG_RETURN(0);
1105 1106
}

1107

1108
/*
1109
  Comparison function to be used QUICK_ROR_UNION_SELECT::queue priority
1110 1111
  queue.

1112 1113 1114 1115 1116 1117 1118 1119
  SYNPOSIS
    QUICK_ROR_UNION_SELECT::queue_cmp()
      arg   Pointer to QUICK_ROR_UNION_SELECT
      val1  First merged select
      val2  Second merged select
*/
int QUICK_ROR_UNION_SELECT::queue_cmp(void *arg, byte *val1, byte *val2)
{
1120
  QUICK_ROR_UNION_SELECT *self= (QUICK_ROR_UNION_SELECT*)arg;
1121 1122 1123 1124
  return self->head->file->cmp_ref(((QUICK_SELECT_I*)val1)->last_rowid,
                                   ((QUICK_SELECT_I*)val2)->last_rowid);
}

1125

1126
/*
1127 1128 1129
  Initialize quick select for row retrieval.
  SYNOPSIS
    reset()
1130

1131 1132 1133 1134 1135
  RETURN
    0      OK
    other  Error code
*/

1136 1137 1138 1139 1140
int QUICK_ROR_UNION_SELECT::reset()
{
  QUICK_SELECT_I* quick;
  int error;
  DBUG_ENTER("QUICK_ROR_UNION_SELECT::reset");
monty@mysql.com's avatar
monty@mysql.com committed
1141
  have_prev_rowid= FALSE;
sergefp@mysql.com's avatar
sergefp@mysql.com committed
1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153
  if (!scans_inited)
  {
    QUICK_SELECT_I *quick;
    List_iterator_fast<QUICK_SELECT_I> it(quick_selects);
    while ((quick= it++))
    {
      if (quick->init_ror_merged_scan(FALSE))
        DBUG_RETURN(1);
    }
    scans_inited= true;
  }
  queue_remove_all(&queue);
1154 1155
  /*
    Initialize scans for merged quick selects and put all merged quick
1156 1157 1158 1159 1160
    selects into the queue.
  */
  List_iterator_fast<QUICK_SELECT_I> it(quick_selects);
  while ((quick= it++))
  {
sergefp@mysql.com's avatar
sergefp@mysql.com committed
1161
    if (quick->reset())
1162
      DBUG_RETURN(1);
1163 1164 1165 1166
    if ((error= quick->get_next()))
    {
      if (error == HA_ERR_END_OF_FILE)
        continue;
monty@mysql.com's avatar
monty@mysql.com committed
1167
      DBUG_RETURN(error);
1168 1169 1170 1171 1172
    }
    quick->save_last_pos();
    queue_insert(&queue, (byte*)quick);
  }

monty@mysql.com's avatar
monty@mysql.com committed
1173
  if (head->file->ha_rnd_init(1))
1174 1175 1176 1177 1178 1179 1180 1181 1182
  {
    DBUG_PRINT("error", ("ROR index_merge rnd_init call failed"));
    DBUG_RETURN(1);
  }

  DBUG_RETURN(0);
}


1183
bool
1184 1185 1186 1187 1188 1189 1190 1191 1192
QUICK_ROR_UNION_SELECT::push_quick_back(QUICK_SELECT_I *quick_sel_range)
{
  return quick_selects.push_back(quick_sel_range);
}

QUICK_ROR_UNION_SELECT::~QUICK_ROR_UNION_SELECT()
{
  DBUG_ENTER("QUICK_ROR_UNION_SELECT::~QUICK_ROR_UNION_SELECT");
  delete_queue(&queue);
1193
  quick_selects.delete_elements();
1194 1195
  if (head->file->inited != handler::NONE)
    head->file->ha_rnd_end();
1196 1197
  free_root(&alloc,MYF(0));
  DBUG_VOID_RETURN;
1198 1199
}

1200

bk@work.mysql.com's avatar
bk@work.mysql.com committed
1201 1202 1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230 1231 1232 1233 1234 1235 1236 1237 1238 1239 1240 1241 1242 1243 1244 1245 1246 1247 1248 1249 1250 1251 1252 1253
QUICK_RANGE::QUICK_RANGE()
  :min_key(0),max_key(0),min_length(0),max_length(0),
   flag(NO_MIN_RANGE | NO_MAX_RANGE)
{}

SEL_ARG::SEL_ARG(SEL_ARG &arg) :Sql_alloc()
{
  type=arg.type;
  min_flag=arg.min_flag;
  max_flag=arg.max_flag;
  maybe_flag=arg.maybe_flag;
  maybe_null=arg.maybe_null;
  part=arg.part;
  field=arg.field;
  min_value=arg.min_value;
  max_value=arg.max_value;
  next_key_part=arg.next_key_part;
  use_count=1; elements=1;
}


inline void SEL_ARG::make_root()
{
  left=right= &null_element;
  color=BLACK;
  next=prev=0;
  use_count=0; elements=1;
}

SEL_ARG::SEL_ARG(Field *f,const char *min_value_arg,const char *max_value_arg)
  :min_flag(0), max_flag(0), maybe_flag(0), maybe_null(f->real_maybe_null()),
   elements(1), use_count(1), field(f), min_value((char*) min_value_arg),
   max_value((char*) max_value_arg), next(0),prev(0),
   next_key_part(0),color(BLACK),type(KEY_RANGE)
{
  left=right= &null_element;
}

SEL_ARG::SEL_ARG(Field *field_,uint8 part_,char *min_value_,char *max_value_,
		 uint8 min_flag_,uint8 max_flag_,uint8 maybe_flag_)
  :min_flag(min_flag_),max_flag(max_flag_),maybe_flag(maybe_flag_),
   part(part_),maybe_null(field_->real_maybe_null()), elements(1),use_count(1),
   field(field_), min_value(min_value_), max_value(max_value_),
   next(0),prev(0),next_key_part(0),color(BLACK),type(KEY_RANGE)
{
  left=right= &null_element;
}

SEL_ARG *SEL_ARG::clone(SEL_ARG *new_parent,SEL_ARG **next_arg)
{
  SEL_ARG *tmp;
  if (type != KEY_RANGE)
  {
1254 1255
    if (!(tmp= new SEL_ARG(type)))
      return 0;					// out of memory
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1256 1257 1258 1259 1260 1261
    tmp->prev= *next_arg;			// Link into next/prev chain
    (*next_arg)->next=tmp;
    (*next_arg)= tmp;
  }
  else
  {
1262 1263 1264
    if (!(tmp= new SEL_ARG(field,part, min_value,max_value,
			   min_flag, max_flag, maybe_flag)))
      return 0;					// OOM
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1265 1266 1267 1268 1269 1270 1271 1272 1273 1274
    tmp->parent=new_parent;
    tmp->next_key_part=next_key_part;
    if (left != &null_element)
      tmp->left=left->clone(tmp,next_arg);

    tmp->prev= *next_arg;			// Link into next/prev chain
    (*next_arg)->next=tmp;
    (*next_arg)= tmp;

    if (right != &null_element)
1275 1276
      if (!(tmp->right= right->clone(tmp,next_arg)))
	return 0;				// OOM
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1277 1278
  }
  increment_use_count(1);
1279
  tmp->color= color;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1280 1281 1282 1283 1284 1285 1286 1287 1288 1289 1290 1291 1292 1293 1294 1295 1296 1297 1298 1299 1300 1301 1302
  return tmp;
}

SEL_ARG *SEL_ARG::first()
{
  SEL_ARG *next_arg=this;
  if (!next_arg->left)
    return 0;					// MAYBE_KEY
  while (next_arg->left != &null_element)
    next_arg=next_arg->left;
  return next_arg;
}

SEL_ARG *SEL_ARG::last()
{
  SEL_ARG *next_arg=this;
  if (!next_arg->right)
    return 0;					// MAYBE_KEY
  while (next_arg->right != &null_element)
    next_arg=next_arg->right;
  return next_arg;
}

1303

bk@work.mysql.com's avatar
bk@work.mysql.com committed
1304 1305 1306
/*
  Check if a compare is ok, when one takes ranges in account
  Returns -2 or 2 if the ranges where 'joined' like  < 2 and >= 2
1307
*/
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1308 1309 1310 1311 1312 1313 1314 1315 1316 1317 1318 1319 1320 1321 1322 1323 1324 1325 1326 1327 1328 1329 1330

static int sel_cmp(Field *field, char *a,char *b,uint8 a_flag,uint8 b_flag)
{
  int cmp;
  /* First check if there was a compare to a min or max element */
  if (a_flag & (NO_MIN_RANGE | NO_MAX_RANGE))
  {
    if ((a_flag & (NO_MIN_RANGE | NO_MAX_RANGE)) ==
	(b_flag & (NO_MIN_RANGE | NO_MAX_RANGE)))
      return 0;
    return (a_flag & NO_MIN_RANGE) ? -1 : 1;
  }
  if (b_flag & (NO_MIN_RANGE | NO_MAX_RANGE))
    return (b_flag & NO_MIN_RANGE) ? 1 : -1;

  if (field->real_maybe_null())			// If null is part of key
  {
    if (*a != *b)
    {
      return *a ? -1 : 1;
    }
    if (*a)
      goto end;					// NULL where equal
monty@hundin.mysql.fi's avatar
monty@hundin.mysql.fi committed
1331
    a++; b++;					// Skip NULL marker
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1332 1333 1334 1335 1336 1337 1338 1339 1340 1341 1342 1343 1344 1345 1346 1347 1348 1349 1350 1351 1352 1353 1354 1355
  }
  cmp=field->key_cmp((byte*) a,(byte*) b);
  if (cmp) return cmp < 0 ? -1 : 1;		// The values differed

  // Check if the compared equal arguments was defined with open/closed range
 end:
  if (a_flag & (NEAR_MIN | NEAR_MAX))
  {
    if ((a_flag & (NEAR_MIN | NEAR_MAX)) == (b_flag & (NEAR_MIN | NEAR_MAX)))
      return 0;
    if (!(b_flag & (NEAR_MIN | NEAR_MAX)))
      return (a_flag & NEAR_MIN) ? 2 : -2;
    return (a_flag & NEAR_MIN) ? 1 : -1;
  }
  if (b_flag & (NEAR_MIN | NEAR_MAX))
    return (b_flag & NEAR_MIN) ? -2 : 2;
  return 0;					// The elements where equal
}


SEL_ARG *SEL_ARG::clone_tree()
{
  SEL_ARG tmp_link,*next_arg,*root;
  next_arg= &tmp_link;
1356
  root= clone((SEL_ARG *) 0, &next_arg);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1357 1358
  next_arg->next=0;				// Fix last link
  tmp_link.next->prev=0;			// Fix first link
1359 1360
  if (root)					// If not OOM
    root->use_count= 0;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1361 1362 1363
  return root;
}

1364

serg@serg.mylan's avatar
serg@serg.mylan committed
1365
/*
1366
  Table rows retrieval plan. Range optimizer creates QUICK_SELECT_I-derived
1367 1368 1369 1370 1371
  objects from table read plans.
*/
class TABLE_READ_PLAN
{
public:
1372 1373
  /*
    Plan read cost, with or without cost of full row retrieval, depending
1374 1375
    on plan creation parameters.
  */
1376
  double read_cost;
1377
  ha_rows records; /* estimate of #rows to be examined */
serg@serg.mylan's avatar
serg@serg.mylan committed
1378

1379 1380
  /*
    If TRUE, the scan returns rows in rowid order. This is used only for
1381 1382
    scans that can be both ROR and non-ROR.
  */
1383
  bool is_ror;
1384

1385 1386 1387 1388 1389
  /*
    Create quick select for this plan.
    SYNOPSIS
     make_quick()
       param               Parameter from test_quick_select
monty@mysql.com's avatar
monty@mysql.com committed
1390
       retrieve_full_rows  If TRUE, created quick select will do full record
1391 1392
                           retrieval.
       parent_alloc        Memory pool to use, if any.
1393

1394 1395
    NOTES
      retrieve_full_rows is ignored by some implementations.
1396 1397

    RETURN
1398 1399 1400
      created quick select
      NULL on any error.
  */
1401 1402 1403 1404
  virtual QUICK_SELECT_I *make_quick(PARAM *param,
                                     bool retrieve_full_rows,
                                     MEM_ROOT *parent_alloc=NULL) = 0;

1405
  /* Table read plans are allocated on MEM_ROOT and are never deleted */
1406 1407
  static void *operator new(size_t size, MEM_ROOT *mem_root)
  { return (void*) alloc_root(mem_root, (uint) size); }
1408
  static void operator delete(void *ptr,size_t size) { TRASH(ptr, size); }
1409
  static void operator delete(void *ptr, MEM_ROOT *mem_root) { /* Never called */ }
1410 1411 1412 1413 1414 1415 1416
};

class TRP_ROR_INTERSECT;
class TRP_ROR_UNION;
class TRP_INDEX_MERGE;


1417
/*
1418
  Plan for a QUICK_RANGE_SELECT scan.
1419 1420 1421
  TRP_RANGE::make_quick ignores retrieve_full_rows parameter because
  QUICK_RANGE_SELECT doesn't distinguish between 'index only' scans and full
  record retrieval scans.
serg@serg.mylan's avatar
serg@serg.mylan committed
1422
*/
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1423

1424
class TRP_RANGE : public TABLE_READ_PLAN
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1425
{
1426
public:
1427 1428
  SEL_ARG *key; /* set of intervals to be used in "range" method retrieval */
  uint     key_idx; /* key number in PARAM::key */
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1429

1430
  TRP_RANGE(SEL_ARG *key_arg, uint idx_arg)
1431 1432
   : key(key_arg), key_idx(idx_arg)
  {}
1433

1434 1435 1436 1437 1438 1439 1440 1441 1442 1443 1444 1445 1446
  QUICK_SELECT_I *make_quick(PARAM *param, bool retrieve_full_rows,
                             MEM_ROOT *parent_alloc)
  {
    DBUG_ENTER("TRP_RANGE::make_quick");
    QUICK_RANGE_SELECT *quick;
    if ((quick= get_quick_select(param, key_idx, key, parent_alloc)))
    {
      quick->records= records;
      quick->read_time= read_cost;
    }
    DBUG_RETURN(quick);
  }
};
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1447 1448


1449 1450
/* Plan for QUICK_ROR_INTERSECT_SELECT scan. */

1451 1452 1453
class TRP_ROR_INTERSECT : public TABLE_READ_PLAN
{
public:
1454
  QUICK_SELECT_I *make_quick(PARAM *param, bool retrieve_full_rows,
1455
                             MEM_ROOT *parent_alloc);
1456

1457
  /* Array of pointers to ROR range scans used in this intersection */
1458
  struct st_ror_scan_info **first_scan;
1459 1460
  struct st_ror_scan_info **last_scan; /* End of the above array */
  struct st_ror_scan_info *cpk_scan;  /* Clustered PK scan, if there is one */
monty@mysql.com's avatar
monty@mysql.com committed
1461
  bool is_covering; /* TRUE if no row retrieval phase is necessary */
1462
  double index_scan_costs; /* SUM(cost(index_scan)) */
1463 1464
};

1465

1466
/*
1467 1468
  Plan for QUICK_ROR_UNION_SELECT scan.
  QUICK_ROR_UNION_SELECT always retrieves full rows, so retrieve_full_rows
1469
  is ignored by make_quick.
1470
*/
1471

1472 1473 1474
class TRP_ROR_UNION : public TABLE_READ_PLAN
{
public:
1475
  QUICK_SELECT_I *make_quick(PARAM *param, bool retrieve_full_rows,
1476
                             MEM_ROOT *parent_alloc);
1477 1478
  TABLE_READ_PLAN **first_ror; /* array of ptrs to plans for merged scans */
  TABLE_READ_PLAN **last_ror;  /* end of the above array */
1479 1480
};

1481 1482 1483 1484

/*
  Plan for QUICK_INDEX_MERGE_SELECT scan.
  QUICK_ROR_INTERSECT_SELECT always retrieves full rows, so retrieve_full_rows
1485
  is ignored by make_quick.
1486 1487
*/

1488 1489 1490
class TRP_INDEX_MERGE : public TABLE_READ_PLAN
{
public:
1491
  QUICK_SELECT_I *make_quick(PARAM *param, bool retrieve_full_rows,
1492
                             MEM_ROOT *parent_alloc);
1493 1494
  TRP_RANGE **range_scans; /* array of ptrs to plans of merged scans */
  TRP_RANGE **range_scans_end; /* end of the array */
1495 1496 1497
};


1498 1499 1500 1501 1502 1503 1504 1505 1506 1507 1508 1509 1510 1511 1512 1513 1514 1515 1516 1517 1518 1519 1520
/*
  Plan for a QUICK_GROUP_MIN_MAX_SELECT scan. 
*/

class TRP_GROUP_MIN_MAX : public TABLE_READ_PLAN
{
private:
  bool have_min, have_max;
  KEY_PART_INFO *min_max_arg_part;
  uint group_prefix_len;
  uint used_key_parts;
  uint group_key_parts;
  KEY *index_info;
  uint index;
  uint key_infix_len;
  byte key_infix[MAX_KEY_LENGTH];
  SEL_TREE *range_tree; /* Represents all range predicates in the query. */
  SEL_ARG  *index_tree; /* The SEL_ARG sub-tree corresponding to index_info. */
  uint param_idx; /* Index of used key in param->key. */
  /* Number of records selected by the ranges in index_tree. */
public:
  ha_rows quick_prefix_records;
public:
1521 1522 1523 1524
  TRP_GROUP_MIN_MAX(bool have_min_arg, bool have_max_arg,
                    KEY_PART_INFO *min_max_arg_part_arg,
                    uint group_prefix_len_arg, uint used_key_parts_arg,
                    uint group_key_parts_arg, KEY *index_info_arg,
1525 1526
                    uint index_arg, uint key_infix_len_arg,
                    byte *key_infix_arg,
1527 1528 1529 1530 1531 1532 1533 1534 1535
                    SEL_TREE *tree_arg, SEL_ARG *index_tree_arg,
                    uint param_idx_arg, ha_rows quick_prefix_records_arg)
  : have_min(have_min_arg), have_max(have_max_arg),
    min_max_arg_part(min_max_arg_part_arg),
    group_prefix_len(group_prefix_len_arg), used_key_parts(used_key_parts_arg),
    group_key_parts(group_key_parts_arg), index_info(index_info_arg),
    index(index_arg), key_infix_len(key_infix_len_arg), range_tree(tree_arg),
    index_tree(index_tree_arg), param_idx(param_idx_arg),
    quick_prefix_records(quick_prefix_records_arg)
1536 1537 1538 1539
    {
      if (key_infix_len)
        memcpy(this->key_infix, key_infix_arg, key_infix_len);
    }
1540 1541 1542 1543 1544 1545

  QUICK_SELECT_I *make_quick(PARAM *param, bool retrieve_full_rows,
                             MEM_ROOT *parent_alloc);
};


1546
/*
1547
  Fill param->needed_fields with bitmap of fields used in the query.
1548
  SYNOPSIS
1549 1550
    fill_used_fields_bitmap()
      param Parameter from test_quick_select function.
1551

1552 1553 1554
  NOTES
    Clustered PK members are not put into the bitmap as they are implicitly
    present in all keys (and it is impossible to avoid reading them).
1555 1556 1557
  RETURN
    0  Ok
    1  Out of memory.
1558 1559 1560 1561 1562
*/

static int fill_used_fields_bitmap(PARAM *param)
{
  TABLE *table= param->table;
1563
  param->fields_bitmap_size= (table->s->fields/8 + 1);
1564 1565 1566
  uchar *tmp;
  uint pk;
  if (!(tmp= (uchar*)alloc_root(param->mem_root,param->fields_bitmap_size)) ||
1567
      bitmap_init(&param->needed_fields, tmp, param->fields_bitmap_size*8,
monty@mysql.com's avatar
monty@mysql.com committed
1568
                  FALSE))
1569
    return 1;
1570

1571
  bitmap_clear_all(&param->needed_fields);
1572
  for (uint i= 0; i < table->s->fields; i++)
1573 1574 1575 1576 1577
  {
    if (param->thd->query_id == table->field[i]->query_id)
      bitmap_set_bit(&param->needed_fields, i+1);
  }

1578
  pk= param->table->s->primary_key;
1579 1580
  if (param->table->file->primary_key_is_clustered() && pk != MAX_KEY)
  {
1581
    /* The table uses clustered PK and it is not internally generated */
1582
    KEY_PART_INFO *key_part= param->table->key_info[pk].key_part;
1583
    KEY_PART_INFO *key_part_end= key_part +
1584 1585 1586 1587 1588 1589 1590 1591 1592 1593
                                 param->table->key_info[pk].key_parts;
    for(;key_part != key_part_end; ++key_part)
    {
      bitmap_clear_bit(&param->needed_fields, key_part->fieldnr);
    }
  }
  return 0;
}


serg@serg.mylan's avatar
serg@serg.mylan committed
1594
/*
1595
  Test if a key can be used in different ranges
serg@serg.mylan's avatar
serg@serg.mylan committed
1596 1597

  SYNOPSIS
1598 1599 1600 1601 1602
    SQL_SELECT::test_quick_select()
      thd               Current thread
      keys_to_use       Keys to use for range retrieval
      prev_tables       Tables assumed to be already read when the scan is
                        performed (but not read at the moment of this call)
1603 1604 1605
      limit             Query limit
      force_quick_range Prefer to use range (instead of full table scan) even
                        if it is more expensive.
1606 1607 1608 1609 1610

  NOTES
    Updates the following in the select parameter:
      needed_reg - Bits for keys with may be used if all prev regs are read
      quick      - Parameter to use when reading records.
1611

1612 1613 1614
    In the table struct the following information is updated:
      quick_keys - Which keys can be used
      quick_rows - How many rows the key matches
serg@serg.mylan's avatar
serg@serg.mylan committed
1615

1616 1617 1618 1619
  TODO
   Check if this function really needs to modify keys_to_use, and change the
   code to pass it by reference if it doesn't.

1620
   In addition to force_quick_range other means can be (an usually are) used
1621 1622
   to make this function prefer range over full table scan. Figure out if
   force_quick_range is really needed.
1623

1624 1625 1626 1627
  RETURN
   -1 if impossible select (i.e. certainly no rows will be selected)
    0 if can't use quick_select
    1 if found usable ranges and quick select has been successfully created.
serg@serg.mylan's avatar
serg@serg.mylan committed
1628
*/
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1629

1630 1631
int SQL_SELECT::test_quick_select(THD *thd, key_map keys_to_use,
				  table_map prev_tables,
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1632 1633 1634 1635
				  ha_rows limit, bool force_quick_range)
{
  uint idx;
  double scan_time;
1636
  DBUG_ENTER("SQL_SELECT::test_quick_select");
serg@serg.mylan's avatar
serg@serg.mylan committed
1637 1638 1639
  DBUG_PRINT("enter",("keys_to_use: %lu  prev_tables: %lu  const_tables: %lu",
		      keys_to_use.to_ulonglong(), (ulong) prev_tables,
		      (ulong) const_tables));
1640
  DBUG_PRINT("info", ("records=%lu", (ulong)head->file->records));
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1641 1642
  delete quick;
  quick=0;
1643 1644 1645
  needed_reg.clear_all();
  quick_keys.clear_all();
  if ((specialflag & SPECIAL_SAFE_MODE) && ! force_quick_range ||
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1646 1647
      !limit)
    DBUG_RETURN(0); /* purecov: inspected */
1648 1649
  if (keys_to_use.is_clear_all())
    DBUG_RETURN(0);
1650
  records= head->file->records;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1651 1652
  if (!records)
    records++;					/* purecov: inspected */
1653 1654
  scan_time= (double) records / TIME_FOR_COMPARE + 1;
  read_time= (double) head->file->scan_time() + scan_time + 1.1;
1655 1656
  if (head->force_index)
    scan_time= read_time= DBL_MAX;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1657
  if (limit < records)
1658
    read_time= (double) records + scan_time + 1; // Force to use index
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1659
  else if (read_time <= 2.0 && !force_quick_range)
1660
    DBUG_RETURN(0);				/* No need for quick select */
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1661

1662
  DBUG_PRINT("info",("Time to scan table: %g", read_time));
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1663

1664 1665
  keys_to_use.intersect(head->keys_in_use_for_query);
  if (!keys_to_use.is_clear_all())
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1666 1667
  {
    MEM_ROOT *old_root,alloc;
1668
    SEL_TREE *tree= NULL;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1669
    KEY_PART *key_parts;
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
1670
    KEY *key_info;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1671
    PARAM param;
serg@serg.mylan's avatar
serg@serg.mylan committed
1672

bk@work.mysql.com's avatar
bk@work.mysql.com committed
1673
    /* set up parameter that is passed to all functions */
1674
    param.thd= thd;
monty@mysql.com's avatar
monty@mysql.com committed
1675
    param.baseflag=head->file->table_flags();
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1676 1677 1678 1679 1680
    param.prev_tables=prev_tables | const_tables;
    param.read_tables=read_tables;
    param.current_table= head->map;
    param.table=head;
    param.keys=0;
1681
    param.mem_root= &alloc;
1682
    param.needed_reg= &needed_reg;
1683
    param.imerge_cost_buff_size= 0;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1684

monty@mashka.mysql.fi's avatar
monty@mashka.mysql.fi committed
1685
    thd->no_errors=1;				// Don't warn about NULL
1686
    init_sql_alloc(&alloc, thd->variables.range_alloc_block_size, 0);
1687 1688 1689 1690
    if (!(param.key_parts= (KEY_PART*) alloc_root(&alloc,
                                                  sizeof(KEY_PART)*
                                                  head->s->key_parts)) ||
        fill_used_fields_bitmap(&param))
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1691
    {
monty@mashka.mysql.fi's avatar
monty@mashka.mysql.fi committed
1692
      thd->no_errors=0;
1693
      free_root(&alloc,MYF(0));			// Return memory & allocator
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1694 1695 1696
      DBUG_RETURN(0);				// Can't use range
    }
    key_parts= param.key_parts;
1697 1698
    old_root= thd->mem_root;
    thd->mem_root= &alloc;
1699 1700 1701 1702

    /*
      Make an array with description of all key parts of all table keys.
      This is used in get_mm_parts function.
1703
    */
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
1704
    key_info= head->key_info;
1705
    for (idx=0 ; idx < head->s->keys ; idx++, key_info++)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1706
    {
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
1707
      KEY_PART_INFO *key_part_info;
1708
      if (!keys_to_use.is_set(idx))
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1709 1710 1711 1712 1713
	continue;
      if (key_info->flags & HA_FULLTEXT)
	continue;    // ToDo: ft-keys in non-ft ranges, if possible   SerG

      param.key[param.keys]=key_parts;
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
1714 1715 1716
      key_part_info= key_info->key_part;
      for (uint part=0 ; part < key_info->key_parts ;
	   part++, key_parts++, key_part_info++)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1717
      {
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
1718 1719 1720 1721 1722 1723
	key_parts->key=		 param.keys;
	key_parts->part=	 part;
	key_parts->length=       key_part_info->length;
	key_parts->store_length= key_part_info->store_length;
	key_parts->field=	 key_part_info->field;
	key_parts->null_bit=	 key_part_info->null_bit;
1724
        key_parts->image_type =
1725
          (key_info->flags & HA_SPATIAL) ? Field::itMBR : Field::itRAW;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1726 1727 1728 1729 1730
      }
      param.real_keynr[param.keys++]=idx;
    }
    param.key_parts_end=key_parts;

sergefp@mysql.com's avatar
sergefp@mysql.com committed
1731 1732 1733 1734
    /* Calculate cost of full index read for the shortest covering index */
    if (!head->used_keys.is_clear_all())
    {
      int key_for_use= find_shortest_key(head, &head->used_keys);
1735 1736 1737
      double key_read_time= (get_index_only_read_time(&param, records,
                                                     key_for_use) +
                             (double) records / TIME_FOR_COMPARE);
sergefp@mysql.com's avatar
sergefp@mysql.com committed
1738 1739 1740 1741 1742
      DBUG_PRINT("info",  ("'all'+'using index' scan will be using key %d, "
                           "read time %g", key_for_use, key_read_time));
      if (key_read_time < read_time)
        read_time= key_read_time;
    }
1743

1744 1745 1746 1747 1748 1749 1750 1751
    TABLE_READ_PLAN *best_trp= NULL;
    TRP_GROUP_MIN_MAX *group_trp;
    double best_read_time= read_time;

    if (cond)
      tree= get_mm_tree(&param,cond);

    if (tree && tree->type == SEL_TREE::IMPOSSIBLE)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1752
    {
1753 1754 1755 1756 1757 1758 1759 1760 1761 1762 1763 1764 1765 1766 1767 1768 1769 1770 1771 1772 1773
      records=0L;                      /* Return -1 from this function. */
      read_time= (double) HA_POS_ERROR;
      goto free_mem;
    }
    else if (tree && tree->type != SEL_TREE::KEY &&
                     tree->type != SEL_TREE::KEY_SMALLER)
      goto free_mem;


    /*
      Try to construct a QUICK_GROUP_MIN_MAX_SELECT.
      Notice that it can be constructed no matter if there is a range tree.
    */
    group_trp= get_best_group_min_max(&param, tree);
    if (group_trp && group_trp->read_cost < best_read_time)
    {
      best_trp= group_trp;
      best_read_time= best_trp->read_cost;
    }

    if (tree)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1774
    {
monty@mysql.com's avatar
monty@mysql.com committed
1775 1776 1777
      /*
        It is possible to use a range-based quick select (but it might be
        slower than 'all' table scan).
1778 1779
      */
      if (tree->merges.is_empty())
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1780
      {
1781 1782 1783 1784 1785 1786 1787 1788 1789 1790 1791 1792
        TRP_RANGE         *range_trp;
        TRP_ROR_INTERSECT *rori_trp;
        bool can_build_covering= FALSE;

        /* Get best 'range' plan and prepare data for making other plans */
        if ((range_trp= get_key_scans_params(&param, tree, FALSE,
                                             best_read_time)))
        {
          best_trp= range_trp;
          best_read_time= best_trp->read_cost;
        }

1793
        /*
1794 1795 1796
          Simultaneous key scans and row deletes on several handler
          objects are not allowed so don't use ROR-intersection for
          table deletes.
1797
        */
1798 1799 1800 1801
        if ((thd->lex->sql_command != SQLCOM_DELETE))
#ifdef NOT_USED
          if ((thd->lex->sql_command != SQLCOM_UPDATE))
#endif
1802
        {
1803
          /*
1804 1805
            Get best non-covering ROR-intersection plan and prepare data for
            building covering ROR-intersection.
1806
          */
1807 1808
          if ((rori_trp= get_best_ror_intersect(&param, tree, best_read_time,
                                                &can_build_covering)))
1809
          {
1810 1811
            best_trp= rori_trp;
            best_read_time= best_trp->read_cost;
1812 1813
            /*
              Try constructing covering ROR-intersect only if it looks possible
1814 1815
              and worth doing.
            */
1816 1817 1818 1819
            if (!rori_trp->is_covering && can_build_covering &&
                (rori_trp= get_best_covering_ror_intersect(&param, tree,
                                                           best_read_time)))
              best_trp= rori_trp;
1820 1821
          }
        }
1822 1823 1824 1825 1826 1827 1828 1829 1830 1831 1832 1833
      }
      else
      {
        /* Try creating index_merge/ROR-union scan. */
        SEL_IMERGE *imerge;
        TABLE_READ_PLAN *best_conj_trp= NULL, *new_conj_trp;
        LINT_INIT(new_conj_trp); /* no empty index_merge lists possible */

        DBUG_PRINT("info",("No range reads possible,"
                           " trying to construct index_merge"));
        List_iterator_fast<SEL_IMERGE> it(tree->merges);
        while ((imerge= it++))
1834
        {
1835 1836 1837 1838
          new_conj_trp= get_best_disjunct_quick(&param, imerge, best_read_time);
          if (!best_conj_trp || (new_conj_trp && new_conj_trp->read_cost <
                                 best_conj_trp->read_cost))
            best_conj_trp= new_conj_trp;
1839
        }
1840 1841 1842 1843
        if (best_conj_trp)
          best_trp= best_conj_trp;
      }
    }
1844

1845
    thd->mem_root= old_root;
1846 1847 1848 1849 1850 1851 1852 1853 1854

    /* If we got a read plan, create a quick select from it. */
    if (best_trp)
    {
      records= best_trp->records;
      if (!(quick= best_trp->make_quick(&param, TRUE)) || quick->init())
      {
        delete quick;
        quick= NULL;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1855 1856
      }
    }
1857 1858

  free_mem:
1859
    free_root(&alloc,MYF(0));			// Return memory & allocator
1860
    thd->mem_root= old_root;
monty@mashka.mysql.fi's avatar
monty@mashka.mysql.fi committed
1861
    thd->no_errors=0;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1862
  }
1863

1864
  DBUG_EXECUTE("info", print_quick(quick, &needed_reg););
1865

bk@work.mysql.com's avatar
bk@work.mysql.com committed
1866 1867 1868 1869 1870 1871 1872
  /*
    Assume that if the user is using 'limit' we will only need to scan
    limit rows if we are using a key
  */
  DBUG_RETURN(records ? test(quick) : -1);
}

1873

1874
/*
1875 1876 1877 1878
  Get cost of 'sweep' full records retrieval.
  SYNOPSIS
    get_sweep_read_cost()
      param            Parameter from test_quick_select
1879
      records          # of records to be retrieved
1880
  RETURN
1881
    cost of sweep
1882
*/
1883

1884
double get_sweep_read_cost(const PARAM *param, ha_rows records)
1885
{
1886
  double result;
1887
  DBUG_ENTER("get_sweep_read_cost");
1888 1889
  if (param->table->file->primary_key_is_clustered())
  {
1890
    result= param->table->file->read_time(param->table->s->primary_key,
1891
                                          records, records);
1892 1893
  }
  else
1894
  {
1895
    double n_blocks=
1896
      ceil(ulonglong2double(param->table->file->data_file_length) / IO_SIZE);
1897 1898 1899 1900
    double busy_blocks=
      n_blocks * (1.0 - pow(1.0 - 1.0/n_blocks, rows2double(records)));
    if (busy_blocks < 1.0)
      busy_blocks= 1.0;
1901
    DBUG_PRINT("info",("sweep: nblocks=%g, busy_blocks=%g", n_blocks,
1902
                       busy_blocks));
1903
    /*
1904
      Disabled: Bail out if # of blocks to read is bigger than # of blocks in
1905 1906 1907 1908 1909 1910 1911 1912
      table data file.
    if (max_cost != DBL_MAX  && (busy_blocks+index_reads_cost) >= n_blocks)
      return 1;
    */
    JOIN *join= param->thd->lex->select_lex.join;
    if (!join || join->tables == 1)
    {
      /* No join, assume reading is done in one 'sweep' */
1913
      result= busy_blocks*(DISK_SEEK_BASE_COST +
1914 1915 1916 1917
                          DISK_SEEK_PROP_COST*n_blocks/busy_blocks);
    }
    else
    {
1918
      /*
1919 1920 1921
        Possibly this is a join with source table being non-last table, so
        assume that disk seeks are random here.
      */
1922
      result= busy_blocks;
1923 1924
    }
  }
1925
  DBUG_PRINT("info",("returning cost=%g", result));
1926
  DBUG_RETURN(result);
1927
}
1928 1929


1930 1931 1932 1933
/*
  Get best plan for a SEL_IMERGE disjunctive expression.
  SYNOPSIS
    get_best_disjunct_quick()
1934 1935
      param     Parameter from check_quick_select function
      imerge    Expression to use
1936
      read_time Don't create scans with cost > read_time
1937

1938
  NOTES
1939
    index_merge cost is calculated as follows:
1940
    index_merge_cost =
1941 1942 1943 1944 1945
      cost(index_reads) +         (see #1)
      cost(rowid_to_row_scan) +   (see #2)
      cost(unique_use)            (see #3)

    1. cost(index_reads) =SUM_i(cost(index_read_i))
1946 1947
       For non-CPK scans,
         cost(index_read_i) = {cost of ordinary 'index only' scan}
1948 1949 1950 1951 1952
       For CPK scan,
         cost(index_read_i) = {cost of non-'index only' scan}

    2. cost(rowid_to_row_scan)
      If table PK is clustered then
1953
        cost(rowid_to_row_scan) =
1954
          {cost of ordinary clustered PK scan with n_ranges=n_rows}
1955 1956

      Otherwise, we use the following model to calculate costs:
1957
      We need to retrieve n_rows rows from file that occupies n_blocks blocks.
1958
      We assume that offsets of rows we need are independent variates with
1959
      uniform distribution in [0..max_file_offset] range.
1960

1961 1962
      We'll denote block as "busy" if it contains row(s) we need to retrieve
      and "empty" if doesn't contain rows we need.
1963

1964
      Probability that a block is empty is (1 - 1/n_blocks)^n_rows (this
1965
      applies to any block in file). Let x_i be a variate taking value 1 if
1966
      block #i is empty and 0 otherwise.
1967

1968 1969
      Then E(x_i) = (1 - 1/n_blocks)^n_rows;

1970 1971
      E(n_empty_blocks) = E(sum(x_i)) = sum(E(x_i)) =
        = n_blocks * ((1 - 1/n_blocks)^n_rows) =
1972 1973 1974 1975
       ~= n_blocks * exp(-n_rows/n_blocks).

      E(n_busy_blocks) = n_blocks*(1 - (1 - 1/n_blocks)^n_rows) =
       ~= n_blocks * (1 - exp(-n_rows/n_blocks)).
1976

1977 1978
      Average size of "hole" between neighbor non-empty blocks is
           E(hole_size) = n_blocks/E(n_busy_blocks).
1979

1980 1981 1982 1983 1984 1985
      The total cost of reading all needed blocks in one "sweep" is:

      E(n_busy_blocks)*
       (DISK_SEEK_BASE_COST + DISK_SEEK_PROP_COST*n_blocks/E(n_busy_blocks)).

    3. Cost of Unique use is calculated in Unique::get_use_cost function.
1986 1987 1988 1989 1990

  ROR-union cost is calculated in the same way index_merge, but instead of
  Unique a priority queue is used.

  RETURN
1991 1992
    Created read plan
    NULL - Out of memory or no read scan could be built.
1993
*/
1994

1995 1996
static
TABLE_READ_PLAN *get_best_disjunct_quick(PARAM *param, SEL_IMERGE *imerge,
1997
                                         double read_time)
1998 1999 2000 2001 2002 2003 2004
{
  SEL_TREE **ptree;
  TRP_INDEX_MERGE *imerge_trp= NULL;
  uint n_child_scans= imerge->trees_next - imerge->trees;
  TRP_RANGE **range_scans;
  TRP_RANGE **cur_child;
  TRP_RANGE **cpk_scan= NULL;
monty@mysql.com's avatar
monty@mysql.com committed
2005
  bool imerge_too_expensive= FALSE;
2006 2007 2008 2009
  double imerge_cost= 0.0;
  ha_rows cpk_scan_records= 0;
  ha_rows non_cpk_scan_records= 0;
  bool pk_is_clustered= param->table->file->primary_key_is_clustered();
monty@mysql.com's avatar
monty@mysql.com committed
2010 2011
  bool all_scans_ror_able= TRUE;
  bool all_scans_rors= TRUE;
2012 2013 2014 2015 2016 2017 2018 2019 2020
  uint unique_calc_buff_size;
  TABLE_READ_PLAN **roru_read_plans;
  TABLE_READ_PLAN **cur_roru_plan;
  double roru_index_costs;
  ha_rows roru_total_records;
  double roru_intersect_part= 1.0;
  DBUG_ENTER("get_best_disjunct_quick");
  DBUG_PRINT("info", ("Full table scan cost =%g", read_time));

2021
  if (!(range_scans= (TRP_RANGE**)alloc_root(param->mem_root,
2022 2023 2024
                                             sizeof(TRP_RANGE*)*
                                             n_child_scans)))
    DBUG_RETURN(NULL);
2025
  /*
2026 2027 2028
    Collect best 'range' scan for each of disjuncts, and, while doing so,
    analyze possibility of ROR scans. Also calculate some values needed by
    other parts of the code.
2029
  */
2030
  for (ptree= imerge->trees, cur_child= range_scans;
2031
       ptree != imerge->trees_next;
2032
       ptree++, cur_child++)
2033
  {
2034 2035
    DBUG_EXECUTE("info", print_sel_tree(param, *ptree, &(*ptree)->keys_map,
                                        "tree in SEL_IMERGE"););
monty@mysql.com's avatar
monty@mysql.com committed
2036
    if (!(*cur_child= get_key_scans_params(param, *ptree, TRUE, read_time)))
2037 2038
    {
      /*
2039
        One of index scans in this index_merge is more expensive than entire
2040 2041 2042
        table read for another available option. The entire index_merge (and
        any possible ROR-union) will be more expensive then, too. We continue
        here only to update SQL_SELECT members.
2043
      */
monty@mysql.com's avatar
monty@mysql.com committed
2044
      imerge_too_expensive= TRUE;
2045 2046 2047
    }
    if (imerge_too_expensive)
      continue;
2048

2049 2050 2051
    imerge_cost += (*cur_child)->read_cost;
    all_scans_ror_able &= ((*ptree)->n_ror_scans > 0);
    all_scans_rors &= (*cur_child)->is_ror;
2052
    if (pk_is_clustered &&
2053 2054
        param->real_keynr[(*cur_child)->key_idx] ==
        param->table->s->primary_key)
2055
    {
2056 2057
      cpk_scan= cur_child;
      cpk_scan_records= (*cur_child)->records;
2058 2059
    }
    else
2060
      non_cpk_scan_records += (*cur_child)->records;
2061
  }
2062

2063
  DBUG_PRINT("info", ("index_merge scans cost=%g", imerge_cost));
2064
  if (imerge_too_expensive || (imerge_cost > read_time) ||
2065 2066
      (non_cpk_scan_records+cpk_scan_records >= param->table->file->records) &&
      read_time != DBL_MAX)
2067
  {
2068 2069
    /*
      Bail out if it is obvious that both index_merge and ROR-union will be
2070
      more expensive
2071
    */
2072 2073
    DBUG_PRINT("info", ("Sum of index_merge scans is more expensive than "
                        "full table scan, bailing out"));
2074
    DBUG_RETURN(NULL);
2075
  }
2076
  if (all_scans_rors)
2077
  {
2078 2079
    roru_read_plans= (TABLE_READ_PLAN**)range_scans;
    goto skip_to_ror_scan;
2080
  }
2081 2082
  if (cpk_scan)
  {
2083 2084
    /*
      Add one ROWID comparison for each row retrieved on non-CPK scan.  (it
2085 2086 2087
      is done in QUICK_RANGE_SELECT::row_in_ranges)
     */
    imerge_cost += non_cpk_scan_records / TIME_FOR_COMPARE_ROWID;
2088 2089 2090
  }

  /* Calculate cost(rowid_to_row_scan) */
2091
  imerge_cost += get_sweep_read_cost(param, non_cpk_scan_records);
2092
  DBUG_PRINT("info",("index_merge cost with rowid-to-row scan: %g",
2093
                     imerge_cost));
2094 2095
  if (imerge_cost > read_time)
    goto build_ror_index_merge;
2096 2097

  /* Add Unique operations cost */
2098 2099
  unique_calc_buff_size=
    Unique::get_cost_calc_buff_size(non_cpk_scan_records,
2100 2101 2102 2103 2104 2105
                                    param->table->file->ref_length,
                                    param->thd->variables.sortbuff_size);
  if (param->imerge_cost_buff_size < unique_calc_buff_size)
  {
    if (!(param->imerge_cost_buff= (uint*)alloc_root(param->mem_root,
                                                     unique_calc_buff_size)))
2106
      DBUG_RETURN(NULL);
2107 2108 2109
    param->imerge_cost_buff_size= unique_calc_buff_size;
  }

2110
  imerge_cost +=
2111
    Unique::get_use_cost(param->imerge_cost_buff, non_cpk_scan_records,
2112 2113
                         param->table->file->ref_length,
                         param->thd->variables.sortbuff_size);
2114
  DBUG_PRINT("info",("index_merge total cost: %g (wanted: less then %g)",
2115 2116 2117 2118 2119 2120 2121
                     imerge_cost, read_time));
  if (imerge_cost < read_time)
  {
    if ((imerge_trp= new (param->mem_root)TRP_INDEX_MERGE))
    {
      imerge_trp->read_cost= imerge_cost;
      imerge_trp->records= non_cpk_scan_records + cpk_scan_records;
2122
      imerge_trp->records= min(imerge_trp->records,
2123 2124 2125 2126 2127 2128
                               param->table->file->records);
      imerge_trp->range_scans= range_scans;
      imerge_trp->range_scans_end= range_scans + n_child_scans;
      read_time= imerge_cost;
    }
  }
2129

2130
build_ror_index_merge:
2131 2132
  if (!all_scans_ror_able || param->thd->lex->sql_command == SQLCOM_DELETE)
    DBUG_RETURN(imerge_trp);
2133

2134 2135
  /* Ok, it is possible to build a ROR-union, try it. */
  bool dummy;
2136
  if (!(roru_read_plans=
2137 2138 2139 2140 2141 2142 2143 2144 2145 2146 2147 2148 2149
          (TABLE_READ_PLAN**)alloc_root(param->mem_root,
                                        sizeof(TABLE_READ_PLAN*)*
                                        n_child_scans)))
    DBUG_RETURN(imerge_trp);
skip_to_ror_scan:
  roru_index_costs= 0.0;
  roru_total_records= 0;
  cur_roru_plan= roru_read_plans;

  /* Find 'best' ROR scan for each of trees in disjunction */
  for (ptree= imerge->trees, cur_child= range_scans;
       ptree != imerge->trees_next;
       ptree++, cur_child++, cur_roru_plan++)
2150
  {
2151 2152
    /*
      Assume the best ROR scan is the one that has cheapest full-row-retrieval
2153 2154
      scan cost.
      Also accumulate index_only scan costs as we'll need them to calculate
2155 2156 2157 2158 2159 2160 2161
      overall index_intersection cost.
    */
    double cost;
    if ((*cur_child)->is_ror)
    {
      /* Ok, we have index_only cost, now get full rows scan cost */
      cost= param->table->file->
2162
              read_time(param->real_keynr[(*cur_child)->key_idx], 1,
2163 2164 2165 2166 2167 2168 2169
                        (*cur_child)->records) +
              rows2double((*cur_child)->records) / TIME_FOR_COMPARE;
    }
    else
      cost= read_time;

    TABLE_READ_PLAN *prev_plan= *cur_child;
2170
    if (!(*cur_roru_plan= get_best_ror_intersect(param, *ptree, cost,
2171 2172 2173 2174 2175 2176 2177 2178 2179
                                                 &dummy)))
    {
      if (prev_plan->is_ror)
        *cur_roru_plan= prev_plan;
      else
        DBUG_RETURN(imerge_trp);
      roru_index_costs += (*cur_roru_plan)->read_cost;
    }
    else
2180 2181
      roru_index_costs +=
        ((TRP_ROR_INTERSECT*)(*cur_roru_plan))->index_scan_costs;
2182
    roru_total_records += (*cur_roru_plan)->records;
2183
    roru_intersect_part *= (*cur_roru_plan)->records /
2184
                           param->table->file->records;
2185
  }
2186

2187 2188
  /*
    rows to retrieve=
2189
      SUM(rows_in_scan_i) - table_rows * PROD(rows_in_scan_i / table_rows).
2190
    This is valid because index_merge construction guarantees that conditions
2191 2192 2193
    in disjunction do not share key parts.
  */
  roru_total_records -= (ha_rows)(roru_intersect_part*
2194 2195 2196
                                  param->table->file->records);
  /* ok, got a ROR read plan for each of the disjuncts
    Calculate cost:
2197 2198 2199 2200 2201 2202
    cost(index_union_scan(scan_1, ... scan_n)) =
      SUM_i(cost_of_index_only_scan(scan_i)) +
      queue_use_cost(rowid_len, n) +
      cost_of_row_retrieval
    See get_merge_buffers_cost function for queue_use_cost formula derivation.
  */
2203

2204
  double roru_total_cost;
2205 2206 2207
  roru_total_cost= roru_index_costs +
                   rows2double(roru_total_records)*log((double)n_child_scans) /
                   (TIME_FOR_COMPARE_ROWID * M_LN2) +
2208 2209
                   get_sweep_read_cost(param, roru_total_records);

2210
  DBUG_PRINT("info", ("ROR-union: cost %g, %d members", roru_total_cost,
2211 2212 2213 2214 2215 2216 2217 2218 2219 2220 2221 2222 2223 2224
                      n_child_scans));
  TRP_ROR_UNION* roru;
  if (roru_total_cost < read_time)
  {
    if ((roru= new (param->mem_root) TRP_ROR_UNION))
    {
      roru->first_ror= roru_read_plans;
      roru->last_ror= roru_read_plans + n_child_scans;
      roru->read_cost= roru_total_cost;
      roru->records= roru_total_records;
      DBUG_RETURN(roru);
    }
  }
  DBUG_RETURN(imerge_trp);
2225 2226 2227 2228 2229 2230 2231
}


/*
  Calculate cost of 'index only' scan for given index and number of records.

  SYNOPSIS
2232
    get_index_only_read_time()
2233 2234 2235 2236 2237
      param    parameters structure
      records  #of records to read
      keynr    key to read

  NOTES
2238
    It is assumed that we will read trough the whole key range and that all
2239 2240 2241 2242
    key blocks are half full (normally things are much better). It is also
    assumed that each time we read the next key from the index, the handler
    performs a random seek, thus the cost is proportional to the number of
    blocks read.
2243 2244 2245 2246 2247 2248

  TODO:
    Move this to handler->read_time() by adding a flag 'index-only-read' to
    this call. The reason for doing this is that the current function doesn't
    handle the case when the row is stored in the b-tree (like in innodb
    clustered index)
2249 2250
*/

2251
inline double get_index_only_read_time(const PARAM* param, ha_rows records,
2252
                                       int keynr)
2253 2254 2255 2256 2257 2258 2259
{
  double read_time;
  uint keys_per_block= (param->table->file->block_size/2/
			(param->table->key_info[keynr].key_length+
			 param->table->file->ref_length) + 1);
  read_time=((double) (records+keys_per_block-1)/
             (double) keys_per_block);
2260
  return read_time;
2261 2262
}

2263

2264 2265
typedef struct st_ror_scan_info
{
2266 2267 2268 2269 2270
  uint      idx;      /* # of used key in param->keys */
  uint      keynr;    /* # of used key in table */
  ha_rows   records;  /* estimate of # records this scan will return */

  /* Set of intervals over key fields that will be used for row retrieval. */
2271
  SEL_ARG   *sel_arg;
2272 2273

  /* Fields used in the query and covered by this ROR scan. */
2274 2275
  MY_BITMAP covered_fields;
  uint      used_fields_covered; /* # of set bits in covered_fields */
2276
  int       key_rec_length; /* length of key record (including rowid) */
2277 2278

  /*
2279 2280
    Cost of reading all index records with values in sel_arg intervals set
    (assuming there is no need to access full table records)
2281 2282
  */
  double    index_read_cost;
2283 2284 2285
  uint      first_uncovered_field; /* first unused bit in covered_fields */
  uint      key_components; /* # of parts in the key */
} ROR_SCAN_INFO;
2286 2287 2288


/*
2289
  Create ROR_SCAN_INFO* structure with a single ROR scan on index idx using
2290
  sel_arg set of intervals.
2291

2292 2293
  SYNOPSIS
    make_ror_scan()
2294 2295 2296
      param    Parameter from test_quick_select function
      idx      Index of key in param->keys
      sel_arg  Set of intervals for a given key
2297
  RETURN
2298
    NULL - out of memory
2299
    ROR scan structure containing a scan for {idx, sel_arg}
2300 2301 2302 2303 2304 2305 2306 2307 2308 2309 2310 2311 2312 2313 2314
*/

static
ROR_SCAN_INFO *make_ror_scan(const PARAM *param, int idx, SEL_ARG *sel_arg)
{
  ROR_SCAN_INFO *ror_scan;
  uchar *bitmap_buf;
  uint keynr;
  DBUG_ENTER("make_ror_scan");
  if (!(ror_scan= (ROR_SCAN_INFO*)alloc_root(param->mem_root,
                                             sizeof(ROR_SCAN_INFO))))
    DBUG_RETURN(NULL);

  ror_scan->idx= idx;
  ror_scan->keynr= keynr= param->real_keynr[idx];
2315
  ror_scan->key_rec_length= param->table->key_info[keynr].key_length +
2316 2317 2318
                            param->table->file->ref_length;
  ror_scan->sel_arg= sel_arg;
  ror_scan->records= param->table->quick_rows[keynr];
2319 2320

  if (!(bitmap_buf= (uchar*)alloc_root(param->mem_root,
2321 2322
                                      param->fields_bitmap_size)))
    DBUG_RETURN(NULL);
2323

2324
  if (bitmap_init(&ror_scan->covered_fields, bitmap_buf,
monty@mysql.com's avatar
monty@mysql.com committed
2325
                  param->fields_bitmap_size*8, FALSE))
2326 2327
    DBUG_RETURN(NULL);
  bitmap_clear_all(&ror_scan->covered_fields);
2328

2329
  KEY_PART_INFO *key_part= param->table->key_info[keynr].key_part;
2330
  KEY_PART_INFO *key_part_end= key_part +
2331 2332 2333 2334 2335 2336 2337 2338 2339 2340
                               param->table->key_info[keynr].key_parts;
  uint n_used_covered= 0;
  for (;key_part != key_part_end; ++key_part)
  {
    if (bitmap_is_set(&param->needed_fields, key_part->fieldnr))
    {
      n_used_covered++;
      bitmap_set_bit(&ror_scan->covered_fields, key_part->fieldnr);
    }
  }
2341
  ror_scan->index_read_cost=
2342 2343 2344 2345 2346 2347
    get_index_only_read_time(param, param->table->quick_rows[ror_scan->keynr],
                             ror_scan->keynr);
  DBUG_RETURN(ror_scan);
}


2348
/*
2349 2350 2351 2352 2353 2354 2355
  Compare two ROR_SCAN_INFO** by  E(#records_matched) * key_record_length.
  SYNOPSIS
    cmp_ror_scan_info()
      a ptr to first compared value
      b ptr to second compared value

  RETURN
2356
   -1 a < b
2357 2358
    0 a = b
    1 a > b
2359
*/
2360
static int cmp_ror_scan_info(ROR_SCAN_INFO** a, ROR_SCAN_INFO** b)
2361 2362 2363 2364 2365 2366 2367
{
  double val1= rows2double((*a)->records) * (*a)->key_rec_length;
  double val2= rows2double((*b)->records) * (*b)->key_rec_length;
  return (val1 < val2)? -1: (val1 == val2)? 0 : 1;
}

/*
2368 2369 2370
  Compare two ROR_SCAN_INFO** by
   (#covered fields in F desc,
    #components asc,
2371
    number of first not covered component asc)
2372 2373 2374 2375 2376 2377 2378

  SYNOPSIS
    cmp_ror_scan_info_covering()
      a ptr to first compared value
      b ptr to second compared value

  RETURN
2379
   -1 a < b
2380 2381
    0 a = b
    1 a > b
2382
*/
2383
static int cmp_ror_scan_info_covering(ROR_SCAN_INFO** a, ROR_SCAN_INFO** b)
2384 2385 2386 2387 2388 2389 2390 2391 2392 2393 2394 2395 2396 2397 2398 2399 2400
{
  if ((*a)->used_fields_covered > (*b)->used_fields_covered)
    return -1;
  if ((*a)->used_fields_covered < (*b)->used_fields_covered)
    return 1;
  if ((*a)->key_components < (*b)->key_components)
    return -1;
  if ((*a)->key_components > (*b)->key_components)
    return 1;
  if ((*a)->first_uncovered_field < (*b)->first_uncovered_field)
    return -1;
  if ((*a)->first_uncovered_field > (*b)->first_uncovered_field)
    return 1;
  return 0;
}

/* Auxiliary structure for incremental ROR-intersection creation */
2401
typedef struct
2402 2403 2404
{
  const PARAM *param;
  MY_BITMAP covered_fields; /* union of fields covered by all scans */
2405
  /*
2406
    Fraction of table records that satisfies conditions of all scans.
2407
    This is the number of full records that will be retrieved if a
2408 2409
    non-index_only index intersection will be employed.
  */
2410 2411 2412 2413
  double out_rows;
  /* TRUE if covered_fields is a superset of needed_fields */
  bool is_covering;

2414
  ha_rows index_records; /* sum(#records to look in indexes) */
2415 2416
  double index_scan_costs; /* SUM(cost of 'index-only' scans) */
  double total_cost;
2417
} ROR_INTERSECT_INFO;
2418 2419


2420 2421 2422 2423
/*
  Allocate a ROR_INTERSECT_INFO and initialize it to contain zero scans.

  SYNOPSIS
2424 2425 2426
    ror_intersect_init()
      param         Parameter from test_quick_select

2427 2428 2429 2430 2431 2432
  RETURN
    allocated structure
    NULL on error
*/

static
2433
ROR_INTERSECT_INFO* ror_intersect_init(const PARAM *param)
2434 2435 2436
{
  ROR_INTERSECT_INFO *info;
  uchar* buf;
2437
  if (!(info= (ROR_INTERSECT_INFO*)alloc_root(param->mem_root,
2438 2439 2440 2441 2442 2443
                                              sizeof(ROR_INTERSECT_INFO))))
    return NULL;
  info->param= param;
  if (!(buf= (uchar*)alloc_root(param->mem_root, param->fields_bitmap_size)))
    return NULL;
  if (bitmap_init(&info->covered_fields, buf, param->fields_bitmap_size*8,
monty@mysql.com's avatar
monty@mysql.com committed
2444
                  FALSE))
2445
    return NULL;
2446
  info->is_covering= FALSE;
2447
  info->index_scan_costs= 0.0;
2448 2449 2450
  info->index_records= 0;
  info->out_rows= param->table->file->records;
  bitmap_clear_all(&info->covered_fields);
2451 2452 2453
  return info;
}

2454 2455 2456 2457 2458 2459 2460 2461 2462 2463 2464
void ror_intersect_cpy(ROR_INTERSECT_INFO *dst, const ROR_INTERSECT_INFO *src)
{
  dst->param= src->param;
  memcpy(dst->covered_fields.bitmap, src->covered_fields.bitmap, 
         src->covered_fields.bitmap_size);
  dst->out_rows= src->out_rows;
  dst->is_covering= src->is_covering;
  dst->index_records= src->index_records;
  dst->index_scan_costs= src->index_scan_costs;
  dst->total_cost= src->total_cost;
}
2465
/*
2466
  Get selectivity of a ROR scan wrt ROR-intersection.
2467

2468
  SYNOPSIS
2469 2470 2471 2472
    ror_scan_selectivity()
      info  ROR-interection 
      scan  ROR scan
      
2473
  NOTES
2474
    Suppose we have a condition on several keys
2475 2476
    cond=k_11=c_11 AND k_12=c_12 AND ...  // parts of first key
         k_21=c_21 AND k_22=c_22 AND ...  // parts of second key
2477
          ...
2478
         k_n1=c_n1 AND k_n3=c_n3 AND ...  (1) //parts of the key used by *scan
2479

2480 2481 2482 2483 2484
    where k_ij may be the same as any k_pq (i.e. keys may have common parts).

    A full row is retrieved iff entire cond holds.

    The recursive procedure for finding P(cond) is as follows:
2485

2486
    First step:
2487
    Pick 1st part of 1st key and break conjunction (1) into two parts:
2488 2489
      cond= (k_11=c_11 AND R)

2490
    Here R may still contain condition(s) equivalent to k_11=c_11.
2491 2492
    Nevertheless, the following holds:

2493
      P(k_11=c_11 AND R) = P(k_11=c_11) * P(R|k_11=c_11).
2494 2495 2496 2497 2498

    Mark k_11 as fixed field (and satisfied condition) F, save P(F),
    save R to be cond and proceed to recursion step.

    Recursion step:
2499
    We have a set of fixed fields/satisfied conditions) F, probability P(F),
2500 2501 2502
    and remaining conjunction R
    Pick next key part on current key and its condition "k_ij=c_ij".
    We will add "k_ij=c_ij" into F and update P(F).
2503
    Lets denote k_ij as t,  R = t AND R1, where R1 may still contain t. Then
2504

2505
     P((t AND R1)|F) = P(t|F) * P(R1|t|F) = P(t|F) * P(R1|(t AND F)) (2)
2506 2507 2508 2509 2510 2511 2512

    (where '|' mean conditional probability, not "or")

    Consider the first multiplier in (2). One of the following holds:
    a) F contains condition on field used in t (i.e. t AND F = F).
      Then P(t|F) = 1

2513 2514
    b) F doesn't contain condition on field used in t. Then F and t are
     considered independent.
2515

2516
     P(t|F) = P(t|(fields_before_t_in_key AND other_fields)) =
2517 2518
          = P(t|fields_before_t_in_key).

2519 2520
     P(t|fields_before_t_in_key) = #records(fields_before_t_in_key) /
                                   #records(fields_before_t_in_key, t)
2521 2522

    The second multiplier is calculated by applying this step recursively.
2523

2524 2525 2526 2527 2528
  IMPLEMENTATION
    This function calculates the result of application of the "recursion step"
    described above for all fixed key members of a single key, accumulating set
    of covered fields, selectivity, etc.

2529
    The calculation is conducted as follows:
2530
    Lets denote #records(keypart1, ... keypartK) as n_k. We need to calculate
2531

2532 2533
     n_{k1}      n_{k_2}
    --------- * ---------  * .... (3)
2534
     n_{k1-1}    n_{k2_1}
2535

2536 2537 2538 2539 2540
    where k1,k2,... are key parts which fields were not yet marked as fixed
    ( this is result of application of option b) of the recursion step for
      parts of a single key).
    Since it is reasonable to expect that most of the fields are not marked
    as fixed, we calcualate (3) as
2541 2542 2543

                                  n_{i1}      n_{i_2}
    (3) = n_{max_key_part}  / (   --------- * ---------  * ....  )
2544 2545 2546 2547
                                  n_{i1-1}    n_{i2_1}

    where i1,i2, .. are key parts that were already marked as fixed.

2548 2549
    In order to minimize number of expensive records_in_range calls we group
    and reduce adjacent fractions.
2550

2551
  RETURN
2552 2553
    Selectivity of given ROR scan.
    
2554 2555
*/

2556 2557
static double ror_scan_selectivity(const ROR_INTERSECT_INFO *info, 
                                   const ROR_SCAN_INFO *scan)
2558 2559
{
  double selectivity_mult= 1.0;
2560
  KEY_PART_INFO *key_part= info->param->table->key_info[scan->keynr].key_part;
sergefp@mysql.com's avatar
sergefp@mysql.com committed
2561
  byte key_val[MAX_KEY_LENGTH+MAX_FIELD_WIDTH]; /* key values tuple */
2562
  char *key_ptr= (char*) key_val;
2563 2564
  SEL_ARG *sel_arg, *tuple_arg= NULL;
  bool cur_covered;
2565 2566
  bool prev_covered= test(bitmap_is_set(&info->covered_fields,
                                        key_part->fieldnr));
sergefp@mysql.com's avatar
sergefp@mysql.com committed
2567 2568 2569 2570 2571 2572
  key_range min_range;
  key_range max_range;
  min_range.key= (byte*) key_val;
  min_range.flag= HA_READ_KEY_EXACT;
  max_range.key= (byte*) key_val;
  max_range.flag= HA_READ_AFTER_KEY;
2573 2574 2575 2576
  ha_rows prev_records= info->param->table->file->records;
  int i;
  DBUG_ENTER("ror_intersect_selectivity");
  for(i= 0, sel_arg= scan->sel_arg; sel_arg;
2577 2578
      i++, sel_arg= sel_arg->next_key_part)
  {
2579
    DBUG_PRINT("info",("sel_arg step"));
2580 2581
    cur_covered= test(bitmap_is_set(&info->covered_fields,
                                    (key_part + i)->fieldnr));
2582
    if (cur_covered != prev_covered)
2583
    {
2584 2585 2586 2587
      /* create (part1val, ..., part{n-1}val) tuple. */
      {
        if (!tuple_arg)
        {
2588
          tuple_arg= scan->sel_arg;
2589
          tuple_arg->store_min(key_part->length, &key_ptr, 0);
2590 2591 2592
        }
        while (tuple_arg->next_key_part != sel_arg)
        {
2593 2594
          tuple_arg= tuple_arg->next_key_part;
          tuple_arg->store_min(key_part->length, &key_ptr, 0);
2595
        }
2596
      }
2597
      ha_rows records;
2598
      min_range.length= max_range.length= ((char*) key_ptr - (char*) key_val);
2599 2600
      records= info->param->table->file->
                 records_in_range(scan->keynr, &min_range, &max_range);
2601 2602 2603 2604 2605 2606 2607 2608 2609 2610 2611
      if (cur_covered)
      {
        /* uncovered -> covered */
        double tmp= rows2double(records)/rows2double(prev_records);
        DBUG_PRINT("info", ("Selectivity multiplier: %g", tmp));
        selectivity_mult *= tmp;
        prev_records= HA_POS_ERROR;
      }
      else
      {
        /* covered -> uncovered */
2612
        prev_records= records;
2613
      }
2614
    }
2615 2616 2617 2618
    prev_covered= cur_covered;
  }
  if (!prev_covered)
  {
2619
    double tmp= rows2double(info->param->table->quick_rows[scan->keynr]) /
2620 2621
                rows2double(prev_records);
    DBUG_PRINT("info", ("Selectivity multiplier: %g", tmp));
2622
    selectivity_mult *= tmp;
2623
  }
2624 2625 2626
  DBUG_PRINT("info", ("Returning multiplier: %g", selectivity_mult));
  DBUG_RETURN(selectivity_mult);
}
2627

2628 2629 2630 2631 2632 2633 2634 2635 2636 2637 2638 2639 2640 2641 2642 2643 2644 2645 2646 2647 2648 2649 2650 2651 2652 2653 2654 2655 2656 2657 2658 2659 2660 2661 2662 2663 2664 2665 2666 2667 2668 2669 2670 2671 2672 2673 2674 2675
/*
  Check if adding a ROR scan to a ROR-intersection reduces its cost of
  ROR-intersection and if yes, update parameters of ROR-intersection,
  including its cost.

  SYNOPSIS
    ror_intersect_add()
      param        Parameter from test_quick_select
      info         ROR-intersection structure to add the scan to.
      ror_scan     ROR scan info to add.
      is_cpk_scan  If TRUE, add the scan as CPK scan (this can be inferred
                   from other parameters and is passed separately only to
                   avoid duplicating the inference code)

  NOTES
    Adding a ROR scan to ROR-intersect "makes sense" iff the cost of ROR-
    intersection decreases. The cost of ROR-intersection is calculated as
    follows:

    cost= SUM_i(key_scan_cost_i) + cost_of_full_rows_retrieval

    When we add a scan the first increases and the second decreases.

    cost_of_full_rows_retrieval=
      (union of indexes used covers all needed fields) ?
        cost_of_sweep_read(E(rows_to_retrieve), rows_in_table) :
        0

    E(rows_to_retrieve) = #rows_in_table * ror_scan_selectivity(null, scan1) *
                           ror_scan_selectivity({scan1}, scan2) * ... *
                           ror_scan_selectivity({scan1,...}, scanN). 
  RETURN
    TRUE   ROR scan added to ROR-intersection, cost updated.
    FALSE  It doesn't make sense to add this ROR scan to this ROR-intersection.
*/

static bool ror_intersect_add(ROR_INTERSECT_INFO *info,
                       ROR_SCAN_INFO* ror_scan, bool is_cpk_scan)
{
  double selectivity_mult= 1.0;

  DBUG_ENTER("ror_intersect_add");
  DBUG_PRINT("info", ("Current out_rows= %g", info->out_rows));
  DBUG_PRINT("info", ("Adding scan on %s",
                      info->param->table->key_info[ror_scan->keynr].name));
  DBUG_PRINT("info", ("is_cpk_scan=%d",is_cpk_scan));

  selectivity_mult = ror_scan_selectivity(info, ror_scan);
2676 2677 2678
  if (selectivity_mult == 1.0)
  {
    /* Don't add this scan if it doesn't improve selectivity. */
2679
    DBUG_PRINT("info", ("The scan doesn't improve selectivity."));
2680
    DBUG_RETURN(false);
2681
  }
2682 2683 2684 2685
  
  info->out_rows *= selectivity_mult;
  DBUG_PRINT("info", ("info->total_cost= %g", info->total_cost));
  
2686
  if (is_cpk_scan)
2687
  {
2688 2689 2690 2691 2692 2693
    /*
      CPK scan is used to filter out rows. We apply filtering for 
      each record of every scan. Assuming 1/TIME_FOR_COMPARE_ROWID
      per check this gives us:
    */
    info->index_scan_costs += rows2double(info->index_records) / 
2694 2695 2696 2697
                              TIME_FOR_COMPARE_ROWID;
  }
  else
  {
2698
    info->index_records += info->param->table->quick_rows[ror_scan->keynr];
2699 2700
    info->index_scan_costs += ror_scan->index_read_cost;
    bitmap_union(&info->covered_fields, &ror_scan->covered_fields);
2701 2702 2703 2704 2705 2706
    if (!info->is_covering && bitmap_is_subset(&info->param->needed_fields,
                                               &info->covered_fields))
    {
      DBUG_PRINT("info", ("ROR-intersect is covering now"));
      info->is_covering= TRUE;
    }
2707
  }
2708

2709
  info->total_cost= info->index_scan_costs;
2710
  DBUG_PRINT("info", ("info->total_cost= %g", info->total_cost));
2711 2712
  if (!info->is_covering)
  {
2713 2714 2715
    info->total_cost += 
      get_sweep_read_cost(info->param, double2rows(info->out_rows));
    DBUG_PRINT("info", ("info->total_cost= %g", info->total_cost));
2716
  }
2717
  DBUG_PRINT("info", ("New out_rows= %g", info->out_rows));
2718
  DBUG_PRINT("info", ("New cost= %g, %scovering", info->total_cost,
2719
                      info->is_covering?"" : "non-"));
2720
  DBUG_RETURN(TRUE);
2721 2722
}

2723

2724 2725
/*
  Get best ROR-intersection plan using non-covering ROR-intersection search
2726 2727 2728 2729
  algorithm. The returned plan may be covering.

  SYNOPSIS
    get_best_ror_intersect()
2730 2731 2732
      param            Parameter from test_quick_select function.
      tree             Transformed restriction condition to be used to look
                       for ROR scans.
2733
      read_time        Do not return read plans with cost > read_time.
2734
      are_all_covering [out] set to TRUE if union of all scans covers all
2735 2736
                       fields needed by the query (and it is possible to build
                       a covering ROR-intersection)
2737

2738
  NOTES
2739 2740 2741 2742 2743
    get_key_scans_params must be called before this function can be called.
    
    When this function is called by ROR-union construction algorithm it
    assumes it is building an uncovered ROR-intersection (and thus # of full
    records to be retrieved is wrong here). This is a hack.
2744

2745
  IMPLEMENTATION
2746
    The approximate best non-covering plan search algorithm is as follows:
2747

2748 2749 2750 2751
    find_min_ror_intersection_scan()
    {
      R= select all ROR scans;
      order R by (E(#records_matched) * key_record_length).
2752

2753 2754 2755 2756 2757 2758
      S= first(R); -- set of scans that will be used for ROR-intersection
      R= R-first(S);
      min_cost= cost(S);
      min_scan= make_scan(S);
      while (R is not empty)
      {
2759 2760
        firstR= R - first(R);
        if (!selectivity(S + firstR < selectivity(S)))
2761
          continue;
2762
          
2763 2764 2765 2766 2767 2768 2769 2770 2771
        S= S + first(R);
        if (cost(S) < min_cost)
        {
          min_cost= cost(S);
          min_scan= make_scan(S);
        }
      }
      return min_scan;
    }
2772

2773
    See ror_intersect_add function for ROR intersection costs.
2774

2775
    Special handling for Clustered PK scans
2776 2777
    Clustered PK contains all table fields, so using it as a regular scan in
    index intersection doesn't make sense: a range scan on CPK will be less
2778 2779
    expensive in this case.
    Clustered PK scan has special handling in ROR-intersection: it is not used
2780
    to retrieve rows, instead its condition is used to filter row references
2781
    we get from scans on other keys.
2782 2783

  RETURN
2784
    ROR-intersection table read plan
2785
    NULL if out of memory or no suitable plan found.
2786 2787
*/

2788 2789 2790 2791 2792 2793
static
TRP_ROR_INTERSECT *get_best_ror_intersect(const PARAM *param, SEL_TREE *tree,
                                          double read_time,
                                          bool *are_all_covering)
{
  uint idx;
2794
  double min_cost= DBL_MAX;
2795
  DBUG_ENTER("get_best_ror_intersect");
2796

2797
  if ((tree->n_ror_scans < 2) || !param->table->file->records)
2798
    DBUG_RETURN(NULL);
2799 2800

  /*
2801 2802
    Step1: Collect ROR-able SEL_ARGs and create ROR_SCAN_INFO for each of 
    them. Also find and save clustered PK scan if there is one.
2803
  */
2804
  ROR_SCAN_INFO **cur_ror_scan;
2805
  ROR_SCAN_INFO *cpk_scan= NULL;
2806
  uint cpk_no;
monty@mysql.com's avatar
monty@mysql.com committed
2807
  bool cpk_scan_used= FALSE;
2808

2809 2810 2811 2812
  if (!(tree->ror_scans= (ROR_SCAN_INFO**)alloc_root(param->mem_root,
                                                     sizeof(ROR_SCAN_INFO*)*
                                                     param->keys)))
    return NULL;
2813 2814
  cpk_no= ((param->table->file->primary_key_is_clustered()) ?
           param->table->s->primary_key : MAX_KEY);
2815

2816
  for (idx= 0, cur_ror_scan= tree->ror_scans; idx < param->keys; idx++)
2817
  {
2818
    ROR_SCAN_INFO *scan;
2819
    if (!tree->ror_scans_map.is_set(idx))
2820
      continue;
2821
    if (!(scan= make_ror_scan(param, idx, tree->keys[idx])))
2822
      return NULL;
2823
    if (param->real_keynr[idx] == cpk_no)
2824
    {
2825 2826
      cpk_scan= scan;
      tree->n_ror_scans--;
2827 2828
    }
    else
2829
      *(cur_ror_scan++)= scan;
2830
  }
2831

2832
  tree->ror_scans_end= cur_ror_scan;
2833 2834
  DBUG_EXECUTE("info",print_ror_scans_arr(param->table, "original",
                                          tree->ror_scans,
2835 2836
                                          tree->ror_scans_end););
  /*
2837
    Ok, [ror_scans, ror_scans_end) is array of ptrs to initialized
2838 2839
    ROR_SCAN_INFO's.
    Step 2: Get best ROR-intersection using an approximate algorithm.
2840 2841 2842
  */
  qsort(tree->ror_scans, tree->n_ror_scans, sizeof(ROR_SCAN_INFO*),
        (qsort_cmp)cmp_ror_scan_info);
2843 2844
  DBUG_EXECUTE("info",print_ror_scans_arr(param->table, "ordered",
                                          tree->ror_scans,
2845
                                          tree->ror_scans_end););
2846

2847 2848 2849 2850 2851 2852 2853 2854 2855
  ROR_SCAN_INFO **intersect_scans; /* ROR scans used in index intersection */
  ROR_SCAN_INFO **intersect_scans_end;
  if (!(intersect_scans= (ROR_SCAN_INFO**)alloc_root(param->mem_root,
                                                     sizeof(ROR_SCAN_INFO*)*
                                                     tree->n_ror_scans)))
    return NULL;
  intersect_scans_end= intersect_scans;

  /* Create and incrementally update ROR intersection. */
2856 2857 2858
  ROR_INTERSECT_INFO *intersect, *intersect_best;
  if (!(intersect= ror_intersect_init(param)) || 
      !(intersect_best= ror_intersect_init(param)))
2859
    return NULL;
2860

2861
  /* [intersect_scans,intersect_scans_best) will hold the best intersection */
2862
  ROR_SCAN_INFO **intersect_scans_best;
2863
  cur_ror_scan= tree->ror_scans;
2864
  intersect_scans_best= intersect_scans;
2865
  while (cur_ror_scan != tree->ror_scans_end && !intersect->is_covering)
2866
  {
2867 2868 2869 2870 2871 2872 2873 2874
    /* S= S + first(R);  R= R - first(R); */
    if (!ror_intersect_add(intersect, *cur_ror_scan, false))
    {
      cur_ror_scan++;
      continue;
    }
    
    *(intersect_scans_end++)= *(cur_ror_scan++);
2875

2876
    if (intersect->total_cost < min_cost)
2877
    {
2878
      /* Local minimum found, save it */
2879
      ror_intersect_cpy(intersect_best, intersect);
2880
      intersect_scans_best= intersect_scans_end;
2881
      min_cost = intersect->total_cost;
2882 2883
    }
  }
2884

2885 2886 2887 2888 2889 2890
  if (intersect_scans_best == intersect_scans)
  {
    DBUG_PRINT("info", ("None of scans increase selectivity"));
    DBUG_RETURN(NULL);
  }
    
2891 2892 2893 2894
  DBUG_EXECUTE("info",print_ror_scans_arr(param->table,
                                          "best ROR-intersection",
                                          intersect_scans,
                                          intersect_scans_best););
2895

2896
  *are_all_covering= intersect->is_covering;
2897
  uint best_num= intersect_scans_best - intersect_scans;
2898 2899
  ror_intersect_cpy(intersect, intersect_best);

2900 2901
  /*
    Ok, found the best ROR-intersection of non-CPK key scans.
2902 2903
    Check if we should add a CPK scan. If the obtained ROR-intersection is 
    covering, it doesn't make sense to add CPK scan.
2904 2905
  */
  if (cpk_scan && !intersect->is_covering)
2906
  {
2907
    if (ror_intersect_add(intersect, cpk_scan, TRUE) && 
2908
        (intersect->total_cost < min_cost))
2909
    {
monty@mysql.com's avatar
monty@mysql.com committed
2910
      cpk_scan_used= TRUE;
2911
      intersect_best= intersect; //just set pointer here
2912 2913
    }
  }
2914

2915
  /* Ok, return ROR-intersect plan if we have found one */
2916
  TRP_ROR_INTERSECT *trp= NULL;
2917
  if (min_cost < read_time && (cpk_scan_used || best_num > 1))
2918
  {
2919 2920
    if (!(trp= new (param->mem_root) TRP_ROR_INTERSECT))
      DBUG_RETURN(trp);
2921 2922
    if (!(trp->first_scan=
           (ROR_SCAN_INFO**)alloc_root(param->mem_root,
2923 2924 2925 2926
                                       sizeof(ROR_SCAN_INFO*)*best_num)))
      DBUG_RETURN(NULL);
    memcpy(trp->first_scan, intersect_scans, best_num*sizeof(ROR_SCAN_INFO*));
    trp->last_scan=  trp->first_scan + best_num;
2927 2928 2929 2930 2931 2932
    trp->is_covering= intersect_best->is_covering;
    trp->read_cost= intersect_best->total_cost;
    /* Prevent divisons by zero */
    ha_rows best_rows = double2rows(intersect_best->out_rows);
    if (!best_rows)
      best_rows= 1;
2933
    trp->records= best_rows;
2934 2935 2936 2937 2938
    trp->index_scan_costs= intersect_best->index_scan_costs;
    trp->cpk_scan= cpk_scan_used? cpk_scan: NULL;
    DBUG_PRINT("info", ("Returning non-covering ROR-intersect plan:"
                        "cost %g, records %lu",
                        trp->read_cost, (ulong) trp->records));
2939
  }
2940
  DBUG_RETURN(trp);
2941 2942 2943 2944
}


/*
2945
  Get best covering ROR-intersection.
2946
  SYNOPSIS
2947
    get_best_covering_ror_intersect()
2948 2949 2950
      param     Parameter from test_quick_select function.
      tree      SEL_TREE with sets of intervals for different keys.
      read_time Don't return table read plans with cost > read_time.
2951

2952 2953
  RETURN
    Best covering ROR-intersection plan
2954
    NULL if no plan found.
2955 2956

  NOTES
2957
    get_best_ror_intersect must be called for a tree before calling this
2958
    function for it.
2959
    This function invalidates tree->ror_scans member values.
2960

2961 2962 2963 2964 2965 2966 2967 2968 2969 2970 2971 2972 2973
  The following approximate algorithm is used:
    I=set of all covering indexes
    F=set of all fields to cover
    S={}

    do {
      Order I by (#covered fields in F desc,
                  #components asc,
                  number of first not covered component asc);
      F=F-covered by first(I);
      S=S+first(I);
      I=I-first(I);
    } while F is not empty.
2974 2975
*/

2976
static
2977 2978
TRP_ROR_INTERSECT *get_best_covering_ror_intersect(PARAM *param,
                                                   SEL_TREE *tree,
2979
                                                   double read_time)
2980
{
2981
  ROR_SCAN_INFO **ror_scan_mark;
2982
  ROR_SCAN_INFO **ror_scans_end= tree->ror_scans_end;
2983 2984 2985 2986
  DBUG_ENTER("get_best_covering_ror_intersect");
  uint nbits= param->fields_bitmap_size*8;

  for (ROR_SCAN_INFO **scan= tree->ror_scans; scan != ror_scans_end; ++scan)
2987
    (*scan)->key_components=
2988
      param->table->key_info[(*scan)->keynr].key_parts;
2989

2990 2991
  /*
    Run covering-ROR-search algorithm.
2992
    Assume set I is [ror_scan .. ror_scans_end)
2993
  */
2994

2995 2996
  /*I=set of all covering indexes */
  ror_scan_mark= tree->ror_scans;
2997

2998 2999
  uchar buf[MAX_KEY/8+1];
  MY_BITMAP covered_fields;
monty@mysql.com's avatar
monty@mysql.com committed
3000
  if (bitmap_init(&covered_fields, buf, nbits, FALSE))
3001 3002 3003 3004 3005
    DBUG_RETURN(0);
  bitmap_clear_all(&covered_fields);

  double total_cost= 0.0f;
  ha_rows records=0;
3006 3007
  bool all_covered;

3008 3009 3010 3011 3012 3013
  DBUG_PRINT("info", ("Building covering ROR-intersection"));
  DBUG_EXECUTE("info", print_ror_scans_arr(param->table,
                                           "building covering ROR-I",
                                           ror_scan_mark, ror_scans_end););
  do {
    /*
3014
      Update changed sorting info:
3015
        #covered fields,
3016
	number of first not covered component
3017 3018 3019 3020 3021
      Calculate and save these values for each of remaining scans.
    */
    for (ROR_SCAN_INFO **scan= ror_scan_mark; scan != ror_scans_end; ++scan)
    {
      bitmap_subtract(&(*scan)->covered_fields, &covered_fields);
3022
      (*scan)->used_fields_covered=
3023
        bitmap_bits_set(&(*scan)->covered_fields);
3024
      (*scan)->first_uncovered_field=
3025 3026 3027 3028 3029 3030 3031 3032 3033
        bitmap_get_first(&(*scan)->covered_fields);
    }

    qsort(ror_scan_mark, ror_scans_end-ror_scan_mark, sizeof(ROR_SCAN_INFO*),
          (qsort_cmp)cmp_ror_scan_info_covering);

    DBUG_EXECUTE("info", print_ror_scans_arr(param->table,
                                             "remaining scans",
                                             ror_scan_mark, ror_scans_end););
3034

3035 3036 3037
    /* I=I-first(I) */
    total_cost += (*ror_scan_mark)->index_read_cost;
    records += (*ror_scan_mark)->records;
3038
    DBUG_PRINT("info", ("Adding scan on %s",
3039 3040 3041 3042 3043 3044 3045
                        param->table->key_info[(*ror_scan_mark)->keynr].name));
    if (total_cost > read_time)
      DBUG_RETURN(NULL);
    /* F=F-covered by first(I) */
    bitmap_union(&covered_fields, &(*ror_scan_mark)->covered_fields);
    all_covered= bitmap_is_subset(&param->needed_fields, &covered_fields);
  } while (!all_covered && (++ror_scan_mark < ror_scans_end));
3046

3047 3048 3049 3050 3051 3052 3053 3054 3055 3056 3057
  if (!all_covered)
    DBUG_RETURN(NULL); /* should not happen actually */

  /*
    Ok, [tree->ror_scans .. ror_scan) holds covering index_intersection with
    cost total_cost.
  */
  DBUG_PRINT("info", ("Covering ROR-intersect scans cost: %g", total_cost));
  DBUG_EXECUTE("info", print_ror_scans_arr(param->table,
                                           "creating covering ROR-intersect",
                                           tree->ror_scans, ror_scan_mark););
3058

3059
  /* Add priority queue use cost. */
3060 3061
  total_cost += rows2double(records)*
                log((double)(ror_scan_mark - tree->ror_scans)) /
3062 3063 3064 3065 3066 3067 3068 3069 3070 3071 3072 3073 3074 3075 3076 3077
                (TIME_FOR_COMPARE_ROWID * M_LN2);
  DBUG_PRINT("info", ("Covering ROR-intersect full cost: %g", total_cost));

  if (total_cost > read_time)
    DBUG_RETURN(NULL);

  TRP_ROR_INTERSECT *trp;
  if (!(trp= new (param->mem_root) TRP_ROR_INTERSECT))
    DBUG_RETURN(trp);
  uint best_num= (ror_scan_mark - tree->ror_scans);
  if (!(trp->first_scan= (ROR_SCAN_INFO**)alloc_root(param->mem_root,
                                                     sizeof(ROR_SCAN_INFO*)*
                                                     best_num)))
    DBUG_RETURN(NULL);
  memcpy(trp->first_scan, ror_scan_mark, best_num*sizeof(ROR_SCAN_INFO*));
  trp->last_scan=  trp->first_scan + best_num;
monty@mysql.com's avatar
monty@mysql.com committed
3078
  trp->is_covering= TRUE;
3079 3080 3081
  trp->read_cost= total_cost;
  trp->records= records;

3082 3083 3084
  DBUG_PRINT("info",
             ("Returning covering ROR-intersect plan: cost %g, records %lu",
              trp->read_cost, (ulong) trp->records));
3085
  DBUG_RETURN(trp);
3086 3087 3088
}


3089
/*
3090
  Get best "range" table read plan for given SEL_TREE.
3091
  Also update PARAM members and store ROR scans info in the SEL_TREE.
3092
  SYNOPSIS
3093
    get_key_scans_params
3094
      param        parameters from test_quick_select
3095
      tree         make range select for this SEL_TREE
monty@mysql.com's avatar
monty@mysql.com committed
3096
      index_read_must_be_used if TRUE, assume 'index only' option will be set
3097
                             (except for clustered PK indexes)
3098 3099
      read_time    don't create read plans with cost > read_time.
  RETURN
3100
    Best range read plan
3101
    NULL if no plan found or error occurred
3102 3103
*/

3104
static TRP_RANGE *get_key_scans_params(PARAM *param, SEL_TREE *tree,
3105
                                       bool index_read_must_be_used,
3106
                                       double read_time)
3107 3108
{
  int idx;
3109 3110 3111
  SEL_ARG **key,**end, **key_to_read= NULL;
  ha_rows best_records;
  TRP_RANGE* read_plan= NULL;
3112
  bool pk_is_clustered= param->table->file->primary_key_is_clustered();
3113 3114
  DBUG_ENTER("get_key_scans_params");
  LINT_INIT(best_records); /* protected by key_to_read */
3115
  /*
3116 3117
    Note that there may be trees that have type SEL_TREE::KEY but contain no
    key reads at all, e.g. tree for expression "key1 is not null" where key1
3118
    is defined as "not null".
3119 3120
  */
  DBUG_EXECUTE("info", print_sel_tree(param, tree, &tree->keys_map,
3121 3122 3123 3124
                                      "tree scans"););
  tree->ror_scans_map.clear_all();
  tree->n_ror_scans= 0;
  for (idx= 0,key=tree->keys, end=key+param->keys;
3125 3126 3127 3128 3129 3130 3131
       key != end ;
       key++,idx++)
  {
    ha_rows found_records;
    double found_read_time;
    if (*key)
    {
3132
      uint keynr= param->real_keynr[idx];
3133 3134
      if ((*key)->type == SEL_ARG::MAYBE_KEY ||
          (*key)->maybe_flag)
3135
        param->needed_reg->set_bit(keynr);
3136

monty@mysql.com's avatar
monty@mysql.com committed
3137 3138
      bool read_index_only= index_read_must_be_used ? TRUE :
                            (bool) param->table->used_keys.is_set(keynr);
3139

3140 3141 3142 3143 3144 3145
      found_records= check_quick_select(param, idx, *key);
      if (param->is_ror_scan)
      {
        tree->n_ror_scans++;
        tree->ror_scans_map.set_bit(idx);
      }
3146
      double cpu_cost= (double) found_records / TIME_FOR_COMPARE;
3147
      if (found_records != HA_POS_ERROR && found_records > 2 &&
sergefp@mysql.com's avatar
sergefp@mysql.com committed
3148
          read_index_only &&
monty@mysql.com's avatar
monty@mysql.com committed
3149
          (param->table->file->index_flags(keynr, param->max_key_part,1) &
monty@mysql.com's avatar
monty@mysql.com committed
3150
           HA_KEYREAD_ONLY) &&
3151
          !(pk_is_clustered && keynr == param->table->s->primary_key))
3152
        /* We can resolve this by only reading through this key. */
3153 3154
        found_read_time= get_index_only_read_time(param,found_records,keynr) +
                         cpu_cost;
3155
      else
3156
        /*
3157 3158 3159
          cost(read_through_index) = cost(disk_io) + cost(row_in_range_checks)
          The row_in_range check is in QUICK_RANGE_SELECT::cmp_next function.
        */
3160 3161 3162 3163
	found_read_time= param->table->file->read_time(keynr,
                                                       param->range_count,
                                                       found_records) +
			 cpu_cost;
3164

3165 3166 3167
      DBUG_PRINT("info",("key %s: found_read_time: %g (cur. read_time: %g)",
                         param->table->key_info[keynr].name, found_read_time,
                         read_time));
3168

3169 3170
      if (read_time > found_read_time && found_records != HA_POS_ERROR
          /*|| read_time == DBL_MAX*/ )
3171
      {
3172
        read_time=    found_read_time;
3173
        best_records= found_records;
3174 3175 3176 3177 3178 3179 3180 3181 3182 3183 3184 3185 3186 3187 3188 3189
        key_to_read=  key;
      }

    }
  }

  DBUG_EXECUTE("info", print_sel_tree(param, tree, &tree->ror_scans_map,
                                      "ROR scans"););
  if (key_to_read)
  {
    idx= key_to_read - tree->keys;
    if ((read_plan= new (param->mem_root) TRP_RANGE(*key_to_read, idx)))
    {
      read_plan->records= best_records;
      read_plan->is_ror= tree->ror_scans_map.is_set(idx);
      read_plan->read_cost= read_time;
3190 3191 3192 3193
      DBUG_PRINT("info",
                 ("Returning range plan for key %s, cost %g, records %lu",
                  param->table->key_info[param->real_keynr[idx]].name,
                  read_plan->read_cost, (ulong) read_plan->records));
3194 3195 3196 3197 3198 3199 3200 3201 3202
    }
  }
  else
    DBUG_PRINT("info", ("No 'range' table read plan found"));

  DBUG_RETURN(read_plan);
}


3203
QUICK_SELECT_I *TRP_INDEX_MERGE::make_quick(PARAM *param,
3204 3205 3206 3207 3208 3209 3210 3211 3212 3213 3214 3215 3216 3217 3218
                                            bool retrieve_full_rows,
                                            MEM_ROOT *parent_alloc)
{
  QUICK_INDEX_MERGE_SELECT *quick_imerge;
  QUICK_RANGE_SELECT *quick;
  /* index_merge always retrieves full rows, ignore retrieve_full_rows */
  if (!(quick_imerge= new QUICK_INDEX_MERGE_SELECT(param->thd, param->table)))
    return NULL;

  quick_imerge->records= records;
  quick_imerge->read_time= read_cost;
  for(TRP_RANGE **range_scan= range_scans; range_scan != range_scans_end;
      range_scan++)
  {
    if (!(quick= (QUICK_RANGE_SELECT*)
monty@mysql.com's avatar
monty@mysql.com committed
3219
           ((*range_scan)->make_quick(param, FALSE, &quick_imerge->alloc)))||
3220 3221 3222 3223 3224 3225 3226 3227 3228 3229
        quick_imerge->push_quick_back(quick))
    {
      delete quick;
      delete quick_imerge;
      return NULL;
    }
  }
  return quick_imerge;
}

3230
QUICK_SELECT_I *TRP_ROR_INTERSECT::make_quick(PARAM *param,
3231 3232 3233 3234 3235 3236 3237
                                              bool retrieve_full_rows,
                                              MEM_ROOT *parent_alloc)
{
  QUICK_ROR_INTERSECT_SELECT *quick_intrsect;
  QUICK_RANGE_SELECT *quick;
  DBUG_ENTER("TRP_ROR_INTERSECT::make_quick");
  MEM_ROOT *alloc;
3238 3239

  if ((quick_intrsect=
3240
         new QUICK_ROR_INTERSECT_SELECT(param->thd, param->table,
monty@mysql.com's avatar
monty@mysql.com committed
3241
                                        retrieve_full_rows? (!is_covering):FALSE,
3242 3243
                                        parent_alloc)))
  {
3244
    DBUG_EXECUTE("info", print_ror_scans_arr(param->table,
3245 3246 3247 3248 3249 3250 3251 3252
                                             "creating ROR-intersect",
                                             first_scan, last_scan););
    alloc= parent_alloc? parent_alloc: &quick_intrsect->alloc;
    for(; first_scan != last_scan;++first_scan)
    {
      if (!(quick= get_quick_select(param, (*first_scan)->idx,
                                    (*first_scan)->sel_arg, alloc)) ||
          quick_intrsect->push_quick_back(quick))
3253
      {
3254 3255
        delete quick_intrsect;
        DBUG_RETURN(NULL);
3256 3257
      }
    }
3258 3259 3260 3261
    if (cpk_scan)
    {
      if (!(quick= get_quick_select(param, cpk_scan->idx,
                                    cpk_scan->sel_arg, alloc)))
3262
      {
3263 3264
        delete quick_intrsect;
        DBUG_RETURN(NULL);
3265
      }
3266
      quick->file= NULL; 
3267
      quick_intrsect->cpk_quick= quick;
3268
    }
3269
    quick_intrsect->records= records;
3270
    quick_intrsect->read_time= read_cost;
3271
  }
3272 3273 3274
  DBUG_RETURN(quick_intrsect);
}

3275

3276
QUICK_SELECT_I *TRP_ROR_UNION::make_quick(PARAM *param,
3277 3278 3279 3280 3281 3282 3283
                                          bool retrieve_full_rows,
                                          MEM_ROOT *parent_alloc)
{
  QUICK_ROR_UNION_SELECT *quick_roru;
  TABLE_READ_PLAN **scan;
  QUICK_SELECT_I *quick;
  DBUG_ENTER("TRP_ROR_UNION::make_quick");
3284 3285
  /*
    It is impossible to construct a ROR-union that will not retrieve full
3286
    rows, ignore retrieve_full_rows parameter.
3287 3288 3289 3290 3291
  */
  if ((quick_roru= new QUICK_ROR_UNION_SELECT(param->thd, param->table)))
  {
    for(scan= first_ror; scan != last_ror; scan++)
    {
3292
      if (!(quick= (*scan)->make_quick(param, FALSE, &quick_roru->alloc)) ||
3293 3294 3295 3296 3297
          quick_roru->push_quick_back(quick))
        DBUG_RETURN(NULL);
    }
    quick_roru->records= records;
    quick_roru->read_time= read_cost;
3298
  }
3299
  DBUG_RETURN(quick_roru);
3300 3301
}

3302

igor@rurik.mysql.com's avatar
igor@rurik.mysql.com committed
3303 3304 3305 3306 3307 3308 3309 3310 3311 3312 3313 3314 3315 3316 3317
/*
  Build a SEL_TREE for a simple predicate
 
  SYNOPSIS
    get_func_mm_tree()
      param       PARAM from SQL_SELECT::test_quick_select
      cond_func   item for the predicate
      field       field in the predicate
      value       constant in the predicate
      cmp_type    compare type for the field

  RETURN 
    Pointer to thre built tree
*/

3318 3319 3320 3321 3322 3323 3324
static SEL_TREE *get_func_mm_tree(PARAM *param, Item_func *cond_func, 
                                  Field *field, Item *value,
                                  Item_result cmp_type)
{
  SEL_TREE *tree= 0;
  DBUG_ENTER("get_func_mm_tree");

igor@rurik.mysql.com's avatar
igor@rurik.mysql.com committed
3325 3326
  switch (cond_func->functype()) {
  case Item_func::NE_FUNC:
3327
    tree= get_mm_parts(param, cond_func, field, Item_func::LT_FUNC,
3328 3329 3330
		       value, cmp_type);
    if (tree)
    {
3331
      tree= tree_or(param, tree, get_mm_parts(param, cond_func, field,
igor@rurik.mysql.com's avatar
igor@rurik.mysql.com committed
3332 3333
					      Item_func::GT_FUNC,
					      value, cmp_type));
3334
    }
igor@rurik.mysql.com's avatar
igor@rurik.mysql.com committed
3335 3336
    break;
  case Item_func::BETWEEN:
3337
    tree= get_mm_parts(param, cond_func, field, Item_func::GE_FUNC,
3338 3339 3340
		       cond_func->arguments()[1],cmp_type);
    if (tree)
    {
3341
      tree= tree_and(param, tree, get_mm_parts(param, cond_func, field,
3342 3343 3344 3345
					       Item_func::LE_FUNC,
					       cond_func->arguments()[2],
                                               cmp_type));
    }
igor@rurik.mysql.com's avatar
igor@rurik.mysql.com committed
3346 3347
    break;
  case Item_func::IN_FUNC:
3348 3349
  {
    Item_func_in *func=(Item_func_in*) cond_func;
3350
    tree= get_mm_parts(param, cond_func, field, Item_func::EQ_FUNC,
3351 3352 3353
                       func->arguments()[1], cmp_type);
    if (tree)
    {
igor@rurik.mysql.com's avatar
igor@rurik.mysql.com committed
3354 3355 3356
      Item **arg, **end;
      for (arg= func->arguments()+2, end= arg+func->argument_count()-2;
           arg < end ; arg++)
3357
      {
3358
        tree=  tree_or(param, tree, get_mm_parts(param, cond_func, field, 
3359
                                                 Item_func::EQ_FUNC,
igor@rurik.mysql.com's avatar
igor@rurik.mysql.com committed
3360
                                                 *arg,
3361 3362 3363
                                                 cmp_type));
      }
    }
igor@rurik.mysql.com's avatar
igor@rurik.mysql.com committed
3364
    break;
3365
  }
igor@rurik.mysql.com's avatar
igor@rurik.mysql.com committed
3366
  default: 
3367
  {
igor@rurik.mysql.com's avatar
igor@rurik.mysql.com committed
3368 3369 3370 3371 3372 3373 3374
    /* 
       Here the function for the following predicates are processed:
       <, <=, =, >=, >, LIKE, IS NULL, IS NOT NULL.
       If the predicate is of the form (value op field) it is handled
       as the equivalent predicate (field rev_op value), e.g.
       2 <= a is handled as a >= 2.
    */
3375 3376 3377
    Item_func::Functype func_type=
      (value != cond_func->arguments()[0]) ? cond_func->functype() :
        ((Item_bool_func2*) cond_func)->rev_functype();
3378
    tree= get_mm_parts(param, cond_func, field, func_type, value, cmp_type);
3379
  }
igor@rurik.mysql.com's avatar
igor@rurik.mysql.com committed
3380 3381
  }

3382
  DBUG_RETURN(tree);
3383

3384 3385
}

bk@work.mysql.com's avatar
bk@work.mysql.com committed
3386 3387 3388 3389 3390
	/* make a select tree of all keys in condition */

static SEL_TREE *get_mm_tree(PARAM *param,COND *cond)
{
  SEL_TREE *tree=0;
3391 3392 3393
  SEL_TREE *ftree= 0;
  Item_field *field_item= 0;
  Item *value;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3394 3395 3396 3397 3398 3399 3400 3401 3402 3403 3404 3405 3406
  DBUG_ENTER("get_mm_tree");

  if (cond->type() == Item::COND_ITEM)
  {
    List_iterator<Item> li(*((Item_cond*) cond)->argument_list());

    if (((Item_cond*) cond)->functype() == Item_func::COND_AND_FUNC)
    {
      tree=0;
      Item *item;
      while ((item=li++))
      {
	SEL_TREE *new_tree=get_mm_tree(param,item);
3407
	if (param->thd->is_fatal_error)
3408
	  DBUG_RETURN(0);	// out of memory
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3409 3410 3411 3412 3413 3414 3415 3416 3417 3418 3419 3420 3421 3422 3423
	tree=tree_and(param,tree,new_tree);
	if (tree && tree->type == SEL_TREE::IMPOSSIBLE)
	  break;
      }
    }
    else
    {						// COND OR
      tree=get_mm_tree(param,li++);
      if (tree)
      {
	Item *item;
	while ((item=li++))
	{
	  SEL_TREE *new_tree=get_mm_tree(param,item);
	  if (!new_tree)
3424
	    DBUG_RETURN(0);	// out of memory
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3425 3426 3427 3428 3429 3430 3431 3432 3433 3434 3435 3436 3437 3438 3439
	  tree=tree_or(param,tree,new_tree);
	  if (!tree || tree->type == SEL_TREE::ALWAYS)
	    break;
	}
      }
    }
    DBUG_RETURN(tree);
  }
  /* Here when simple cond */
  if (cond->const_item())
  {
    if (cond->val_int())
      DBUG_RETURN(new SEL_TREE(SEL_TREE::ALWAYS));
    DBUG_RETURN(new SEL_TREE(SEL_TREE::IMPOSSIBLE));
  }
3440

3441 3442 3443
  table_map ref_tables= 0;
  table_map param_comp= ~(param->prev_tables | param->read_tables |
		          param->current_table);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3444 3445
  if (cond->type() != Item::FUNC_ITEM)
  {						// Should be a field
3446
    ref_tables= cond->used_tables();
3447 3448
    if ((ref_tables & param->current_table) ||
	(ref_tables & ~(param->prev_tables | param->read_tables)))
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3449 3450 3451
      DBUG_RETURN(0);
    DBUG_RETURN(new SEL_TREE(SEL_TREE::MAYBE));
  }
3452

bk@work.mysql.com's avatar
bk@work.mysql.com committed
3453 3454 3455 3456
  Item_func *cond_func= (Item_func*) cond;
  if (cond_func->select_optimize() == Item_func::OPTIMIZE_NONE)
    DBUG_RETURN(0);				// Can't be calculated

igor@rurik.mysql.com's avatar
igor@rurik.mysql.com committed
3457 3458
  param->cond= cond;

igor@rurik.mysql.com's avatar
igor@rurik.mysql.com committed
3459 3460 3461
  switch (cond_func->functype()) {
  case Item_func::BETWEEN:
    if (cond_func->arguments()[0]->type() != Item::FIELD_ITEM)
3462
      DBUG_RETURN(0);
igor@rurik.mysql.com's avatar
igor@rurik.mysql.com committed
3463 3464 3465 3466
    field_item= (Item_field*) (cond_func->arguments()[0]);
    value= NULL;
    break;
  case Item_func::IN_FUNC:
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3467 3468
  {
    Item_func_in *func=(Item_func_in*) cond_func;
igor@rurik.mysql.com's avatar
igor@rurik.mysql.com committed
3469
    if (func->key_item()->type() != Item::FIELD_ITEM)
3470
      DBUG_RETURN(0);
igor@rurik.mysql.com's avatar
igor@rurik.mysql.com committed
3471 3472 3473
    field_item= (Item_field*) (func->key_item());
    value= NULL;
    break;
3474
  }
igor@rurik.mysql.com's avatar
igor@rurik.mysql.com committed
3475
  case Item_func::MULT_EQUAL_FUNC:
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3476
  {
3477 3478
    Item_equal *item_equal= (Item_equal *) cond;    
    if (!(value= item_equal->get_const()))
igor@rurik.mysql.com's avatar
igor@rurik.mysql.com committed
3479 3480 3481 3482
      DBUG_RETURN(0);
    Item_equal_iterator it(*item_equal);
    ref_tables= value->used_tables();
    while ((field_item= it++))
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3483
    {
igor@rurik.mysql.com's avatar
igor@rurik.mysql.com committed
3484 3485 3486
      Field *field= field_item->field;
      Item_result cmp_type= field->cmp_type();
      if (!((ref_tables | field->table->map) & param_comp))
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3487
      {
3488
        tree= get_mm_parts(param, cond, field, Item_func::EQ_FUNC,
igor@rurik.mysql.com's avatar
igor@rurik.mysql.com committed
3489 3490
		           value,cmp_type);
        ftree= !ftree ? tree : tree_and(param, ftree, tree);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3491 3492
      }
    }
igor@rurik.mysql.com's avatar
igor@rurik.mysql.com committed
3493
    
3494
    DBUG_RETURN(ftree);
3495 3496
  }
  default:
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3497 3498
    if (cond_func->arguments()[0]->type() == Item::FIELD_ITEM)
    {
3499 3500
      field_item= (Item_field*) (cond_func->arguments()[0]);
      value= cond_func->arg_count > 1 ? cond_func->arguments()[1] : 0;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3501
    }
3502 3503 3504 3505 3506 3507 3508 3509
    else if (cond_func->have_rev_func() &&
             cond_func->arguments()[1]->type() == Item::FIELD_ITEM)
    {
      field_item= (Item_field*) (cond_func->arguments()[1]);
      value= cond_func->arguments()[0];
    }
    else
      DBUG_RETURN(0);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3510
  }
3511 3512 3513 3514 3515 3516 3517 3518 3519 3520 3521 3522 3523 3524 3525 3526 3527 3528 3529 3530 3531 3532 3533 3534 3535 3536 3537 3538 3539

  /* 
     If the where condition contains a predicate (ti.field op const),
     then not only SELL_TREE for this predicate is built, but
     the trees for the results of substitution of ti.field for
     each tj.field belonging to the same multiple equality as ti.field
     are built as well.
     E.g. for WHERE t1.a=t2.a AND t2.a > 10 
     a SEL_TREE for t2.a > 10 will be built for quick select from t2
     and   
     a SEL_TREE for t1.a > 10 will be built for quick select from t1.
  */
     
  for (uint i= 0; i < cond_func->arg_count; i++)
  {
    Item *arg= cond_func->arguments()[i];
    if (arg != field_item)
      ref_tables|= arg->used_tables();
  }
  Field *field= field_item->field;
  Item_result cmp_type= field->cmp_type();
  if (!((ref_tables | field->table->map) & param_comp))
    ftree= get_func_mm_tree(param, cond_func, field, value, cmp_type);
  Item_equal *item_equal= field_item->item_equal;
  if (item_equal)
  {
    Item_equal_iterator it(*item_equal);
    Item_field *item;
    while ((item= it++))
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3540
    {
3541 3542 3543 3544
      Field *f= item->field;
      if (field->eq(f))
        continue;
      if (!((ref_tables | f->table->map) & param_comp))
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3545
      {
3546 3547
        tree= get_func_mm_tree(param, cond_func, f, value, cmp_type);
        ftree= !ftree ? tree : tree_and(param, ftree, tree);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3548 3549 3550
      }
    }
  }
3551
  DBUG_RETURN(ftree);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3552 3553 3554 3555
}


static SEL_TREE *
3556
get_mm_parts(PARAM *param, COND *cond_func, Field *field,
3557
	     Item_func::Functype type,
3558
	     Item *value, Item_result cmp_type)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3559 3560 3561 3562 3563
{
  DBUG_ENTER("get_mm_parts");
  if (field->table != param->table)
    DBUG_RETURN(0);

3564 3565
  KEY_PART *key_part = param->key_parts;
  KEY_PART *end = param->key_parts_end;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3566 3567 3568 3569
  SEL_TREE *tree=0;
  if (value &&
      value->used_tables() & ~(param->prev_tables | param->read_tables))
    DBUG_RETURN(0);
3570
  for (; key_part != end ; key_part++)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3571 3572 3573 3574
  {
    if (field->eq(key_part->field))
    {
      SEL_ARG *sel_arg=0;
3575
      if (!tree && !(tree=new SEL_TREE()))
3576
	DBUG_RETURN(0);				// OOM
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3577 3578
      if (!value || !(value->used_tables() & ~param->read_tables))
      {
3579 3580
	sel_arg=get_mm_leaf(param,cond_func,
			    key_part->field,key_part,type,value);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3581 3582 3583 3584 3585 3586 3587 3588
	if (!sel_arg)
	  continue;
	if (sel_arg->type == SEL_ARG::IMPOSSIBLE)
	{
	  tree->type=SEL_TREE::IMPOSSIBLE;
	  DBUG_RETURN(tree);
	}
      }
3589 3590
      else
      {
3591
	// This key may be used later
3592
	if (!(sel_arg= new SEL_ARG(SEL_ARG::MAYBE_KEY)))
3593
	  DBUG_RETURN(0);			// OOM
3594
      }
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3595 3596
      sel_arg->part=(uchar) key_part->part;
      tree->keys[key_part->key]=sel_add(tree->keys[key_part->key],sel_arg);
sergefp@mysql.com's avatar
sergefp@mysql.com committed
3597
      tree->keys_map.set_bit(key_part->key);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3598 3599
    }
  }
3600

bk@work.mysql.com's avatar
bk@work.mysql.com committed
3601 3602 3603 3604 3605
  DBUG_RETURN(tree);
}


static SEL_ARG *
3606
get_mm_leaf(PARAM *param, COND *conf_func, Field *field, KEY_PART *key_part,
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3607 3608
	    Item_func::Functype type,Item *value)
{
3609
  uint maybe_null=(uint) field->real_maybe_null();
3610
  bool optimize_range;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3611
  SEL_ARG *tree;
3612
  char *str;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3613 3614
  DBUG_ENTER("get_mm_leaf");

3615 3616
  if (!value)					// IS NULL or IS NOT NULL
  {
3617
    if (field->table->maybe_null)		// Can't use a key on this
3618 3619 3620 3621 3622 3623 3624 3625 3626 3627 3628 3629 3630 3631
      DBUG_RETURN(0);
    if (!maybe_null)				// Not null field
      DBUG_RETURN(type == Item_func::ISNULL_FUNC ? &null_element : 0);
    if (!(tree=new SEL_ARG(field,is_null_string,is_null_string)))
      DBUG_RETURN(0);		// out of memory
    if (type == Item_func::ISNOTNULL_FUNC)
    {
      tree->min_flag=NEAR_MIN;		    /* IS NOT NULL ->  X > NULL */
      tree->max_flag=NO_MAX_RANGE;
    }
    DBUG_RETURN(tree);
  }

  /*
3632 3633 3634 3635 3636 3637 3638 3639 3640 3641
    1. Usually we can't use an index if the column collation
       differ from the operation collation.

    2. However, we can reuse a case insensitive index for
       the binary searches:

       WHERE latin1_swedish_ci_column = 'a' COLLATE lati1_bin;

       WHERE latin1_swedish_ci_colimn = BINARY 'a '

3642 3643 3644 3645
  */
  if (field->result_type() == STRING_RESULT &&
      value->result_type() == STRING_RESULT &&
      key_part->image_type == Field::itRAW &&
3646 3647
      ((Field_str*)field)->charset() != conf_func->compare_collation() &&
      !(conf_func->compare_collation()->state & MY_CS_BINSORT))
3648 3649
    DBUG_RETURN(0);

3650 3651 3652
  optimize_range= field->optimize_range(param->real_keynr[key_part->key],
                                        key_part->part);

bk@work.mysql.com's avatar
bk@work.mysql.com committed
3653 3654 3655 3656
  if (type == Item_func::LIKE_FUNC)
  {
    bool like_error;
    char buff1[MAX_FIELD_WIDTH],*min_str,*max_str;
3657
    String tmp(buff1,sizeof(buff1),value->collation.collation),*res;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3658
    uint length,offset,min_length,max_length;
3659
    uint field_length= field->pack_length()+maybe_null;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3660

3661
    if (!optimize_range)
3662
      DBUG_RETURN(0);				// Can't optimize this
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3663 3664 3665
    if (!(res= value->val_str(&tmp)))
      DBUG_RETURN(&null_element);

3666 3667 3668 3669 3670
    /*
      TODO:
      Check if this was a function. This should have be optimized away
      in the sql_select.cc
    */
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3671 3672 3673 3674 3675 3676 3677 3678 3679
    if (res != &tmp)
    {
      tmp.copy(*res);				// Get own copy
      res= &tmp;
    }
    if (field->cmp_type() != STRING_RESULT)
      DBUG_RETURN(0);				// Can only optimize strings

    offset=maybe_null;
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
3680 3681 3682
    length=key_part->store_length;

    if (length != key_part->length  + maybe_null)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3683
    {
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
3684 3685 3686
      /* key packed with length prefix */
      offset+= HA_KEY_BLOB_LENGTH;
      field_length= length - HA_KEY_BLOB_LENGTH;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3687 3688 3689
    }
    else
    {
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
3690 3691 3692 3693 3694 3695 3696 3697
      if (unlikely(length < field_length))
      {
	/*
	  This can only happen in a table created with UNIREG where one key
	  overlaps many fields
	*/
	length= field_length;
      }
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3698
      else
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
3699
	field_length= length;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3700 3701
    }
    length+=offset;
3702
    if (!(min_str= (char*) alloc_root(param->mem_root, length*2)))
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3703
      DBUG_RETURN(0);
3704

bk@work.mysql.com's avatar
bk@work.mysql.com committed
3705 3706 3707
    max_str=min_str+length;
    if (maybe_null)
      max_str[0]= min_str[0]=0;
3708

3709
    field_length-= maybe_null;
3710
    like_error= my_like_range(field->charset(),
monty@mysql.com's avatar
monty@mysql.com committed
3711
			      res->ptr(), res->length(),
monty@mysql.com's avatar
monty@mysql.com committed
3712 3713
			      ((Item_func_like*)(param->cond))->escape,
			      wild_one, wild_many,
3714
			      field_length,
monty@mysql.com's avatar
monty@mysql.com committed
3715 3716
			      min_str+offset, max_str+offset,
			      &min_length, &max_length);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3717 3718
    if (like_error)				// Can't optimize with LIKE
      DBUG_RETURN(0);
monty@mysql.com's avatar
monty@mysql.com committed
3719

3720
    if (offset != maybe_null)			// BLOB or VARCHAR
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3721 3722 3723 3724 3725 3726 3727
    {
      int2store(min_str+maybe_null,min_length);
      int2store(max_str+maybe_null,max_length);
    }
    DBUG_RETURN(new SEL_ARG(field,min_str,max_str));
  }

3728
  if (!optimize_range &&
3729
      type != Item_func::EQ_FUNC &&
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3730 3731 3732
      type != Item_func::EQUAL_FUNC)
    DBUG_RETURN(0);				// Can't optimize this

3733 3734 3735 3736
  /*
    We can't always use indexes when comparing a string index to a number
    cmp_type() is checked to allow compare of dates to numbers
  */
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3737 3738 3739 3740
  if (field->result_type() == STRING_RESULT &&
      value->result_type() != STRING_RESULT &&
      field->cmp_type() != value->result_type())
    DBUG_RETURN(0);
3741

3742
  if (value->save_in_field_no_warnings(field, 1) < 0)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3743
  {
3744
    /* This happens when we try to insert a NULL field in a not null column */
monty@mysql.com's avatar
monty@mysql.com committed
3745
    DBUG_RETURN(&null_element);			// cmp with NULL is never TRUE
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3746
  }
3747
  str= (char*) alloc_root(param->mem_root, key_part->store_length+1);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3748 3749 3750
  if (!str)
    DBUG_RETURN(0);
  if (maybe_null)
3751
    *str= (char) field->is_real_null();		// Set to 1 if null
3752
  field->get_key_image(str+maybe_null, key_part->length, key_part->image_type);
3753
  if (!(tree=new SEL_ARG(field,str,str)))
3754
    DBUG_RETURN(0);		// out of memory
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3755 3756 3757 3758 3759 3760 3761 3762 3763 3764 3765 3766 3767 3768 3769 3770 3771 3772 3773 3774 3775 3776

  switch (type) {
  case Item_func::LT_FUNC:
    if (field_is_equal_to_item(field,value))
      tree->max_flag=NEAR_MAX;
    /* fall through */
  case Item_func::LE_FUNC:
    if (!maybe_null)
      tree->min_flag=NO_MIN_RANGE;		/* From start */
    else
    {						// > NULL
      tree->min_value=is_null_string;
      tree->min_flag=NEAR_MIN;
    }
    break;
  case Item_func::GT_FUNC:
    if (field_is_equal_to_item(field,value))
      tree->min_flag=NEAR_MIN;
    /* fall through */
  case Item_func::GE_FUNC:
    tree->max_flag=NO_MAX_RANGE;
    break;
3777
  case Item_func::SP_EQUALS_FUNC:
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
3778 3779 3780
    tree->min_flag=GEOM_FLAG | HA_READ_MBR_EQUAL;// NEAR_MIN;//512;
    tree->max_flag=NO_MAX_RANGE;
    break;
3781
  case Item_func::SP_DISJOINT_FUNC:
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
3782 3783 3784
    tree->min_flag=GEOM_FLAG | HA_READ_MBR_DISJOINT;// NEAR_MIN;//512;
    tree->max_flag=NO_MAX_RANGE;
    break;
3785
  case Item_func::SP_INTERSECTS_FUNC:
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
3786 3787 3788
    tree->min_flag=GEOM_FLAG | HA_READ_MBR_INTERSECT;// NEAR_MIN;//512;
    tree->max_flag=NO_MAX_RANGE;
    break;
3789
  case Item_func::SP_TOUCHES_FUNC:
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
3790 3791 3792
    tree->min_flag=GEOM_FLAG | HA_READ_MBR_INTERSECT;// NEAR_MIN;//512;
    tree->max_flag=NO_MAX_RANGE;
    break;
3793 3794

  case Item_func::SP_CROSSES_FUNC:
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
3795 3796 3797
    tree->min_flag=GEOM_FLAG | HA_READ_MBR_INTERSECT;// NEAR_MIN;//512;
    tree->max_flag=NO_MAX_RANGE;
    break;
3798
  case Item_func::SP_WITHIN_FUNC:
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
3799 3800 3801
    tree->min_flag=GEOM_FLAG | HA_READ_MBR_WITHIN;// NEAR_MIN;//512;
    tree->max_flag=NO_MAX_RANGE;
    break;
3802 3803

  case Item_func::SP_CONTAINS_FUNC:
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
3804 3805 3806
    tree->min_flag=GEOM_FLAG | HA_READ_MBR_CONTAIN;// NEAR_MIN;//512;
    tree->max_flag=NO_MAX_RANGE;
    break;
3807
  case Item_func::SP_OVERLAPS_FUNC:
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
3808 3809 3810
    tree->min_flag=GEOM_FLAG | HA_READ_MBR_INTERSECT;// NEAR_MIN;//512;
    tree->max_flag=NO_MAX_RANGE;
    break;
3811

bk@work.mysql.com's avatar
bk@work.mysql.com committed
3812 3813 3814 3815 3816 3817 3818 3819 3820 3821 3822 3823
  default:
    break;
  }
  DBUG_RETURN(tree);
}


/******************************************************************************
** Tree manipulation functions
** If tree is 0 it means that the condition can't be tested. It refers
** to a non existent table or to a field in current table with isn't a key.
** The different tree flags:
monty@mysql.com's avatar
monty@mysql.com committed
3824 3825
** IMPOSSIBLE:	 Condition is never TRUE
** ALWAYS:	 Condition is always TRUE
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3826 3827 3828 3829 3830 3831
** MAYBE:	 Condition may exists when tables are read
** MAYBE_KEY:	 Condition refers to a key that may be used in join loop
** KEY_RANGE:	 Condition uses a key
******************************************************************************/

/*
3832 3833
  Add a new key test to a key when scanning through all keys
  This will never be called for same key parts.
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3834 3835 3836 3837 3838 3839 3840 3841 3842 3843 3844 3845 3846 3847 3848 3849 3850 3851 3852 3853 3854 3855 3856 3857 3858 3859 3860 3861 3862 3863 3864 3865 3866 3867 3868 3869 3870 3871 3872 3873 3874 3875 3876 3877 3878 3879 3880 3881 3882 3883 3884 3885 3886 3887 3888 3889 3890 3891 3892 3893 3894
*/

static SEL_ARG *
sel_add(SEL_ARG *key1,SEL_ARG *key2)
{
  SEL_ARG *root,**key_link;

  if (!key1)
    return key2;
  if (!key2)
    return key1;

  key_link= &root;
  while (key1 && key2)
  {
    if (key1->part < key2->part)
    {
      *key_link= key1;
      key_link= &key1->next_key_part;
      key1=key1->next_key_part;
    }
    else
    {
      *key_link= key2;
      key_link= &key2->next_key_part;
      key2=key2->next_key_part;
    }
  }
  *key_link=key1 ? key1 : key2;
  return root;
}

#define CLONE_KEY1_MAYBE 1
#define CLONE_KEY2_MAYBE 2
#define swap_clone_flag(A) ((A & 1) << 1) | ((A & 2) >> 1)


static SEL_TREE *
tree_and(PARAM *param,SEL_TREE *tree1,SEL_TREE *tree2)
{
  DBUG_ENTER("tree_and");
  if (!tree1)
    DBUG_RETURN(tree2);
  if (!tree2)
    DBUG_RETURN(tree1);
  if (tree1->type == SEL_TREE::IMPOSSIBLE || tree2->type == SEL_TREE::ALWAYS)
    DBUG_RETURN(tree1);
  if (tree2->type == SEL_TREE::IMPOSSIBLE || tree1->type == SEL_TREE::ALWAYS)
    DBUG_RETURN(tree2);
  if (tree1->type == SEL_TREE::MAYBE)
  {
    if (tree2->type == SEL_TREE::KEY)
      tree2->type=SEL_TREE::KEY_SMALLER;
    DBUG_RETURN(tree2);
  }
  if (tree2->type == SEL_TREE::MAYBE)
  {
    tree1->type=SEL_TREE::KEY_SMALLER;
    DBUG_RETURN(tree1);
  }

sergefp@mysql.com's avatar
sergefp@mysql.com committed
3895 3896
  key_map  result_keys;
  result_keys.clear_all();
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3897 3898 3899 3900 3901 3902 3903 3904 3905 3906 3907 3908 3909
  /* Join the trees key per key */
  SEL_ARG **key1,**key2,**end;
  for (key1= tree1->keys,key2= tree2->keys,end=key1+param->keys ;
       key1 != end ; key1++,key2++)
  {
    uint flag=0;
    if (*key1 || *key2)
    {
      if (*key1 && !(*key1)->simple_key())
	flag|=CLONE_KEY1_MAYBE;
      if (*key2 && !(*key2)->simple_key())
	flag|=CLONE_KEY2_MAYBE;
      *key1=key_and(*key1,*key2,flag);
3910
      if (*key1 && (*key1)->type == SEL_ARG::IMPOSSIBLE)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3911 3912
      {
	tree1->type= SEL_TREE::IMPOSSIBLE;
3913
        DBUG_RETURN(tree1);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3914
      }
sergefp@mysql.com's avatar
sergefp@mysql.com committed
3915
      result_keys.set_bit(key1 - tree1->keys);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3916
#ifdef EXTRA_DEBUG
3917 3918
      if (*key1)
        (*key1)->test_use_count(*key1);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3919 3920 3921
#endif
    }
  }
3922 3923
  tree1->keys_map= result_keys;
  /* dispose index_merge if there is a "range" option */
sergefp@mysql.com's avatar
sergefp@mysql.com committed
3924
  if (!result_keys.is_clear_all())
3925 3926 3927 3928 3929 3930 3931
  {
    tree1->merges.empty();
    DBUG_RETURN(tree1);
  }

  /* ok, both trees are index_merge trees */
  imerge_list_and_list(&tree1->merges, &tree2->merges);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3932 3933 3934 3935
  DBUG_RETURN(tree1);
}


3936
/*
3937 3938
  Check if two SEL_TREES can be combined into one (i.e. a single key range
  read can be constructed for "cond_of_tree1 OR cond_of_tree2" ) without
3939
  using index_merge.
3940 3941 3942 3943
*/

bool sel_trees_can_be_ored(SEL_TREE *tree1, SEL_TREE *tree2, PARAM* param)
{
sergefp@mysql.com's avatar
sergefp@mysql.com committed
3944
  key_map common_keys= tree1->keys_map;
3945
  DBUG_ENTER("sel_trees_can_be_ored");
3946
  common_keys.intersect(tree2->keys_map);
3947

sergefp@mysql.com's avatar
sergefp@mysql.com committed
3948
  if (common_keys.is_clear_all())
monty@mysql.com's avatar
monty@mysql.com committed
3949
    DBUG_RETURN(FALSE);
3950 3951

  /* trees have a common key, check if they refer to same key part */
3952
  SEL_ARG **key1,**key2;
sergefp@mysql.com's avatar
sergefp@mysql.com committed
3953
  for (uint key_no=0; key_no < param->keys; key_no++)
3954
  {
sergefp@mysql.com's avatar
sergefp@mysql.com committed
3955
    if (common_keys.is_set(key_no))
3956 3957 3958 3959 3960
    {
      key1= tree1->keys + key_no;
      key2= tree2->keys + key_no;
      if ((*key1)->part == (*key2)->part)
      {
monty@mysql.com's avatar
monty@mysql.com committed
3961
        DBUG_RETURN(TRUE);
3962 3963 3964
      }
    }
  }
monty@mysql.com's avatar
monty@mysql.com committed
3965
  DBUG_RETURN(FALSE);
3966
}
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3967 3968 3969 3970 3971 3972 3973 3974 3975 3976 3977 3978 3979 3980 3981 3982

static SEL_TREE *
tree_or(PARAM *param,SEL_TREE *tree1,SEL_TREE *tree2)
{
  DBUG_ENTER("tree_or");
  if (!tree1 || !tree2)
    DBUG_RETURN(0);
  if (tree1->type == SEL_TREE::IMPOSSIBLE || tree2->type == SEL_TREE::ALWAYS)
    DBUG_RETURN(tree2);
  if (tree2->type == SEL_TREE::IMPOSSIBLE || tree1->type == SEL_TREE::ALWAYS)
    DBUG_RETURN(tree1);
  if (tree1->type == SEL_TREE::MAYBE)
    DBUG_RETURN(tree1);				// Can't use this
  if (tree2->type == SEL_TREE::MAYBE)
    DBUG_RETURN(tree2);

3983
  SEL_TREE *result= 0;
sergefp@mysql.com's avatar
sergefp@mysql.com committed
3984 3985
  key_map  result_keys;
  result_keys.clear_all();
3986
  if (sel_trees_can_be_ored(tree1, tree2, param))
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3987
  {
3988 3989 3990 3991
    /* Join the trees key per key */
    SEL_ARG **key1,**key2,**end;
    for (key1= tree1->keys,key2= tree2->keys,end= key1+param->keys ;
         key1 != end ; key1++,key2++)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3992
    {
3993 3994 3995 3996
      *key1=key_or(*key1,*key2);
      if (*key1)
      {
        result=tree1;				// Added to tree1
sergefp@mysql.com's avatar
sergefp@mysql.com committed
3997
        result_keys.set_bit(key1 - tree1->keys);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3998
#ifdef EXTRA_DEBUG
3999
        (*key1)->test_use_count(*key1);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4000
#endif
4001 4002 4003 4004 4005 4006 4007 4008 4009 4010 4011 4012 4013 4014 4015 4016 4017 4018 4019 4020 4021 4022 4023 4024 4025 4026 4027 4028 4029 4030 4031
      }
    }
    if (result)
      result->keys_map= result_keys;
  }
  else
  {
    /* ok, two trees have KEY type but cannot be used without index merge */
    if (tree1->merges.is_empty() && tree2->merges.is_empty())
    {
      SEL_IMERGE *merge;
      /* both trees are "range" trees, produce new index merge structure */
      if (!(result= new SEL_TREE()) || !(merge= new SEL_IMERGE()) ||
          (result->merges.push_back(merge)) ||
          (merge->or_sel_tree(param, tree1)) ||
          (merge->or_sel_tree(param, tree2)))
        result= NULL;
      else
        result->type= tree1->type;
    }
    else if (!tree1->merges.is_empty() && !tree2->merges.is_empty())
    {
      if (imerge_list_or_list(param, &tree1->merges, &tree2->merges))
        result= new SEL_TREE(SEL_TREE::ALWAYS);
      else
        result= tree1;
    }
    else
    {
      /* one tree is index merge tree and another is range tree */
      if (tree1->merges.is_empty())
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
4032
        swap_variables(SEL_TREE*, tree1, tree2);
4033 4034 4035 4036 4037 4038

      /* add tree2 to tree1->merges, checking if it collapses to ALWAYS */
      if (imerge_list_or_tree(param, &tree1->merges, tree2))
        result= new SEL_TREE(SEL_TREE::ALWAYS);
      else
        result= tree1;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4039 4040 4041 4042 4043 4044 4045 4046 4047 4048 4049 4050 4051 4052 4053 4054 4055 4056 4057 4058 4059
    }
  }
  DBUG_RETURN(result);
}


/* And key trees where key1->part < key2 -> part */

static SEL_ARG *
and_all_keys(SEL_ARG *key1,SEL_ARG *key2,uint clone_flag)
{
  SEL_ARG *next;
  ulong use_count=key1->use_count;

  if (key1->elements != 1)
  {
    key2->use_count+=key1->elements-1;
    key2->increment_use_count((int) key1->elements-1);
  }
  if (key1->type == SEL_ARG::MAYBE_KEY)
  {
4060 4061
    key1->right= key1->left= &null_element;
    key1->next= key1->prev= 0;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4062 4063 4064 4065 4066 4067 4068 4069 4070 4071 4072 4073 4074 4075 4076 4077 4078 4079 4080 4081 4082 4083 4084 4085 4086 4087 4088 4089 4090 4091 4092 4093 4094 4095 4096 4097
  }
  for (next=key1->first(); next ; next=next->next)
  {
    if (next->next_key_part)
    {
      SEL_ARG *tmp=key_and(next->next_key_part,key2,clone_flag);
      if (tmp && tmp->type == SEL_ARG::IMPOSSIBLE)
      {
	key1=key1->tree_delete(next);
	continue;
      }
      next->next_key_part=tmp;
      if (use_count)
	next->increment_use_count(use_count);
    }
    else
      next->next_key_part=key2;
  }
  if (!key1)
    return &null_element;			// Impossible ranges
  key1->use_count++;
  return key1;
}


static SEL_ARG *
key_and(SEL_ARG *key1,SEL_ARG *key2,uint clone_flag)
{
  if (!key1)
    return key2;
  if (!key2)
    return key1;
  if (key1->part != key2->part)
  {
    if (key1->part > key2->part)
    {
4098
      swap_variables(SEL_ARG *, key1, key2);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4099 4100 4101 4102 4103
      clone_flag=swap_clone_flag(clone_flag);
    }
    // key1->part < key2->part
    key1->use_count--;
    if (key1->use_count > 0)
4104 4105
      if (!(key1= key1->clone_tree()))
	return 0;				// OOM
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4106 4107 4108 4109
    return and_all_keys(key1,key2,clone_flag);
  }

  if (((clone_flag & CLONE_KEY2_MAYBE) &&
4110 4111
       !(clone_flag & CLONE_KEY1_MAYBE) &&
       key2->type != SEL_ARG::MAYBE_KEY) ||
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4112 4113
      key1->type == SEL_ARG::MAYBE_KEY)
  {						// Put simple key in key2
4114
    swap_variables(SEL_ARG *, key1, key2);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4115 4116 4117 4118 4119 4120 4121 4122 4123
    clone_flag=swap_clone_flag(clone_flag);
  }

  // If one of the key is MAYBE_KEY then the found region may be smaller
  if (key2->type == SEL_ARG::MAYBE_KEY)
  {
    if (key1->use_count > 1)
    {
      key1->use_count--;
4124 4125
      if (!(key1=key1->clone_tree()))
	return 0;				// OOM
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4126 4127 4128 4129 4130 4131 4132 4133 4134 4135 4136 4137 4138 4139
      key1->use_count++;
    }
    if (key1->type == SEL_ARG::MAYBE_KEY)
    {						// Both are maybe key
      key1->next_key_part=key_and(key1->next_key_part,key2->next_key_part,
				 clone_flag);
      if (key1->next_key_part &&
	  key1->next_key_part->type == SEL_ARG::IMPOSSIBLE)
	return key1;
    }
    else
    {
      key1->maybe_smaller();
      if (key2->next_key_part)
4140 4141
      {
	key1->use_count--;			// Incremented in and_all_keys
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4142
	return and_all_keys(key1,key2,clone_flag);
4143
      }
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4144 4145 4146 4147 4148
      key2->use_count--;			// Key2 doesn't have a tree
    }
    return key1;
  }

4149 4150 4151 4152 4153 4154 4155
  if ((key1->min_flag | key2->min_flag) & GEOM_FLAG)
  {
    key1->free_tree();
    key2->free_tree();
    return 0;					// Can't optimize this
  }

4156 4157 4158
  if ((key1->min_flag | key2->min_flag) & GEOM_FLAG)
  {
    key1->free_tree();
4159 4160 4161 4162
    key2->free_tree();
    return 0;					// Can't optimize this
  }

bk@work.mysql.com's avatar
bk@work.mysql.com committed
4163 4164 4165 4166 4167 4168 4169 4170 4171 4172 4173 4174 4175 4176 4177 4178 4179 4180 4181 4182
  key1->use_count--;
  key2->use_count--;
  SEL_ARG *e1=key1->first(), *e2=key2->first(), *new_tree=0;

  while (e1 && e2)
  {
    int cmp=e1->cmp_min_to_min(e2);
    if (cmp < 0)
    {
      if (get_range(&e1,&e2,key1))
	continue;
    }
    else if (get_range(&e2,&e1,key2))
      continue;
    SEL_ARG *next=key_and(e1->next_key_part,e2->next_key_part,clone_flag);
    e1->increment_use_count(1);
    e2->increment_use_count(1);
    if (!next || next->type != SEL_ARG::IMPOSSIBLE)
    {
      SEL_ARG *new_arg= e1->clone_and(e2);
4183 4184
      if (!new_arg)
	return &null_element;			// End of memory
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4185 4186 4187 4188 4189 4190 4191 4192 4193 4194 4195 4196 4197 4198 4199 4200 4201 4202 4203 4204 4205 4206 4207 4208 4209 4210 4211 4212 4213 4214 4215 4216 4217 4218 4219 4220 4221 4222 4223 4224 4225 4226 4227 4228 4229 4230 4231 4232 4233 4234 4235
      new_arg->next_key_part=next;
      if (!new_tree)
      {
	new_tree=new_arg;
      }
      else
	new_tree=new_tree->insert(new_arg);
    }
    if (e1->cmp_max_to_max(e2) < 0)
      e1=e1->next;				// e1 can't overlapp next e2
    else
      e2=e2->next;
  }
  key1->free_tree();
  key2->free_tree();
  if (!new_tree)
    return &null_element;			// Impossible range
  return new_tree;
}


static bool
get_range(SEL_ARG **e1,SEL_ARG **e2,SEL_ARG *root1)
{
  (*e1)=root1->find_range(*e2);			// first e1->min < e2->min
  if ((*e1)->cmp_max_to_min(*e2) < 0)
  {
    if (!((*e1)=(*e1)->next))
      return 1;
    if ((*e1)->cmp_min_to_max(*e2) > 0)
    {
      (*e2)=(*e2)->next;
      return 1;
    }
  }
  return 0;
}


static SEL_ARG *
key_or(SEL_ARG *key1,SEL_ARG *key2)
{
  if (!key1)
  {
    if (key2)
    {
      key2->use_count--;
      key2->free_tree();
    }
    return 0;
  }
4236
  if (!key2)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4237 4238 4239 4240 4241 4242 4243 4244
  {
    key1->use_count--;
    key1->free_tree();
    return 0;
  }
  key1->use_count--;
  key2->use_count--;

4245 4246
  if (key1->part != key2->part || 
      (key1->min_flag | key2->min_flag) & GEOM_FLAG)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4247 4248 4249 4250 4251 4252 4253 4254 4255 4256 4257 4258 4259 4260 4261 4262 4263 4264 4265 4266 4267 4268 4269 4270
  {
    key1->free_tree();
    key2->free_tree();
    return 0;					// Can't optimize this
  }

  // If one of the key is MAYBE_KEY then the found region may be bigger
  if (key1->type == SEL_ARG::MAYBE_KEY)
  {
    key2->free_tree();
    key1->use_count++;
    return key1;
  }
  if (key2->type == SEL_ARG::MAYBE_KEY)
  {
    key1->free_tree();
    key2->use_count++;
    return key2;
  }

  if (key1->use_count > 0)
  {
    if (key2->use_count == 0 || key1->elements > key2->elements)
    {
4271
      swap_variables(SEL_ARG *,key1,key2);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4272
    }
4273
    if (key1->use_count > 0 || !(key1=key1->clone_tree()))
4274
      return 0;					// OOM
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4275 4276 4277 4278 4279 4280 4281 4282 4283 4284 4285 4286 4287 4288 4289 4290 4291 4292 4293 4294 4295 4296 4297 4298 4299
  }

  // Add tree at key2 to tree at key1
  bool key2_shared=key2->use_count != 0;
  key1->maybe_flag|=key2->maybe_flag;

  for (key2=key2->first(); key2; )
  {
    SEL_ARG *tmp=key1->find_range(key2);	// Find key1.min <= key2.min
    int cmp;

    if (!tmp)
    {
      tmp=key1->first();			// tmp.min > key2.min
      cmp= -1;
    }
    else if ((cmp=tmp->cmp_max_to_min(key2)) < 0)
    {						// Found tmp.max < key2.min
      SEL_ARG *next=tmp->next;
      if (cmp == -2 && eq_tree(tmp->next_key_part,key2->next_key_part))
      {
	// Join near ranges like tmp.max < 0 and key2.min >= 0
	SEL_ARG *key2_next=key2->next;
	if (key2_shared)
	{
monty@narttu.mysql.fi's avatar
monty@narttu.mysql.fi committed
4300
	  if (!(key2=new SEL_ARG(*key2)))
4301
	    return 0;		// out of memory
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4302 4303 4304 4305 4306 4307 4308 4309 4310 4311 4312 4313 4314 4315 4316 4317 4318 4319 4320 4321 4322 4323 4324 4325 4326 4327 4328 4329 4330 4331 4332 4333 4334 4335 4336 4337 4338 4339 4340 4341
	  key2->increment_use_count(key1->use_count+1);
	  key2->next=key2_next;			// New copy of key2
	}
	key2->copy_min(tmp);
	if (!(key1=key1->tree_delete(tmp)))
	{					// Only one key in tree
	  key1=key2;
	  key1->make_root();
	  key2=key2_next;
	  break;
	}
      }
      if (!(tmp=next))				// tmp.min > key2.min
	break;					// Copy rest of key2
    }
    if (cmp < 0)
    {						// tmp.min > key2.min
      int tmp_cmp;
      if ((tmp_cmp=tmp->cmp_min_to_max(key2)) > 0) // if tmp.min > key2.max
      {
	if (tmp_cmp == 2 && eq_tree(tmp->next_key_part,key2->next_key_part))
	{					// ranges are connected
	  tmp->copy_min_to_min(key2);
	  key1->merge_flags(key2);
	  if (tmp->min_flag & NO_MIN_RANGE &&
	      tmp->max_flag & NO_MAX_RANGE)
	  {
	    if (key1->maybe_flag)
	      return new SEL_ARG(SEL_ARG::MAYBE_KEY);
	    return 0;
	  }
	  key2->increment_use_count(-1);	// Free not used tree
	  key2=key2->next;
	  continue;
	}
	else
	{
	  SEL_ARG *next=key2->next;		// Keys are not overlapping
	  if (key2_shared)
	  {
4342 4343
	    SEL_ARG *cpy= new SEL_ARG(*key2);	// Must make copy
	    if (!cpy)
4344
	      return 0;				// OOM
4345
	    key1=key1->insert(cpy);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4346 4347 4348 4349 4350 4351 4352 4353 4354 4355 4356 4357 4358 4359 4360 4361 4362 4363 4364 4365 4366 4367 4368 4369 4370 4371 4372 4373
	    key2->increment_use_count(key1->use_count+1);
	  }
	  else
	    key1=key1->insert(key2);		// Will destroy key2_root
	  key2=next;
	  continue;
	}
      }
    }

    // tmp.max >= key2.min && tmp.min <= key.max  (overlapping ranges)
    if (eq_tree(tmp->next_key_part,key2->next_key_part))
    {
      if (tmp->is_same(key2))
      {
	tmp->merge_flags(key2);			// Copy maybe flags
	key2->increment_use_count(-1);		// Free not used tree
      }
      else
      {
	SEL_ARG *last=tmp;
	while (last->next && last->next->cmp_min_to_max(key2) <= 0 &&
	       eq_tree(last->next->next_key_part,key2->next_key_part))
	{
	  SEL_ARG *save=last;
	  last=last->next;
	  key1=key1->tree_delete(save);
	}
sergefp@mysql.com's avatar
sergefp@mysql.com committed
4374
        last->copy_min(tmp);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4375 4376 4377 4378 4379 4380 4381 4382 4383 4384 4385 4386 4387 4388 4389 4390 4391
	if (last->copy_min(key2) || last->copy_max(key2))
	{					// Full range
	  key1->free_tree();
	  for (; key2 ; key2=key2->next)
	    key2->increment_use_count(-1);	// Free not used tree
	  if (key1->maybe_flag)
	    return new SEL_ARG(SEL_ARG::MAYBE_KEY);
	  return 0;
	}
      }
      key2=key2->next;
      continue;
    }

    if (cmp >= 0 && tmp->cmp_min_to_min(key2) < 0)
    {						// tmp.min <= x < key2.min
      SEL_ARG *new_arg=tmp->clone_first(key2);
4392 4393
      if (!new_arg)
	return 0;				// OOM
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4394 4395 4396 4397 4398 4399 4400 4401 4402 4403 4404 4405 4406
      if ((new_arg->next_key_part= key1->next_key_part))
	new_arg->increment_use_count(key1->use_count+1);
      tmp->copy_min_to_min(key2);
      key1=key1->insert(new_arg);
    }

    // tmp.min >= key2.min && tmp.min <= key2.max
    SEL_ARG key(*key2);				// Get copy we can modify
    for (;;)
    {
      if (tmp->cmp_min_to_min(&key) > 0)
      {						// key.min <= x < tmp.min
	SEL_ARG *new_arg=key.clone_first(tmp);
4407 4408
	if (!new_arg)
	  return 0;				// OOM
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4409 4410 4411 4412 4413 4414 4415 4416 4417 4418 4419 4420 4421 4422
	if ((new_arg->next_key_part=key.next_key_part))
	  new_arg->increment_use_count(key1->use_count+1);
	key1=key1->insert(new_arg);
      }
      if ((cmp=tmp->cmp_max_to_max(&key)) <= 0)
      {						// tmp.min. <= x <= tmp.max
	tmp->maybe_flag|= key.maybe_flag;
	key.increment_use_count(key1->use_count+1);
	tmp->next_key_part=key_or(tmp->next_key_part,key.next_key_part);
	if (!cmp)				// Key2 is ready
	  break;
	key.copy_max_to_min(tmp);
	if (!(tmp=tmp->next))
	{
4423 4424 4425 4426
	  SEL_ARG *tmp2= new SEL_ARG(key);
	  if (!tmp2)
	    return 0;				// OOM
	  key1=key1->insert(tmp2);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4427 4428 4429 4430 4431
	  key2=key2->next;
	  goto end;
	}
	if (tmp->cmp_min_to_max(&key) > 0)
	{
4432 4433 4434 4435
	  SEL_ARG *tmp2= new SEL_ARG(key);
	  if (!tmp2)
	    return 0;				// OOM
	  key1=key1->insert(tmp2);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4436 4437 4438 4439 4440 4441
	  break;
	}
      }
      else
      {
	SEL_ARG *new_arg=tmp->clone_last(&key); // tmp.min <= x <= key.max
4442 4443
	if (!new_arg)
	  return 0;				// OOM
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4444 4445
	tmp->copy_max_to_min(&key);
	tmp->increment_use_count(key1->use_count+1);
4446 4447
	/* Increment key count as it may be used for next loop */
	key.increment_use_count(1);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4448 4449 4450 4451 4452 4453 4454 4455 4456 4457 4458 4459 4460 4461
	new_arg->next_key_part=key_or(tmp->next_key_part,key.next_key_part);
	key1=key1->insert(new_arg);
	break;
      }
    }
    key2=key2->next;
  }

end:
  while (key2)
  {
    SEL_ARG *next=key2->next;
    if (key2_shared)
    {
4462 4463 4464
      SEL_ARG *tmp=new SEL_ARG(*key2);		// Must make copy
      if (!tmp)
	return 0;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4465
      key2->increment_use_count(key1->use_count+1);
4466
      key1=key1->insert(tmp);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4467 4468 4469 4470 4471 4472 4473 4474 4475 4476 4477 4478 4479 4480 4481 4482 4483 4484 4485 4486 4487 4488 4489 4490 4491 4492 4493 4494 4495 4496 4497 4498 4499 4500 4501 4502 4503 4504 4505 4506 4507 4508 4509 4510 4511 4512 4513
    }
    else
      key1=key1->insert(key2);			// Will destroy key2_root
    key2=next;
  }
  key1->use_count++;
  return key1;
}


/* Compare if two trees are equal */

static bool eq_tree(SEL_ARG* a,SEL_ARG *b)
{
  if (a == b)
    return 1;
  if (!a || !b || !a->is_same(b))
    return 0;
  if (a->left != &null_element && b->left != &null_element)
  {
    if (!eq_tree(a->left,b->left))
      return 0;
  }
  else if (a->left != &null_element || b->left != &null_element)
    return 0;
  if (a->right != &null_element && b->right != &null_element)
  {
    if (!eq_tree(a->right,b->right))
      return 0;
  }
  else if (a->right != &null_element || b->right != &null_element)
    return 0;
  if (a->next_key_part != b->next_key_part)
  {						// Sub range
    if (!a->next_key_part != !b->next_key_part ||
	!eq_tree(a->next_key_part, b->next_key_part))
      return 0;
  }
  return 1;
}


SEL_ARG *
SEL_ARG::insert(SEL_ARG *key)
{
  SEL_ARG *element,**par,*last_element;
  LINT_INIT(par); LINT_INIT(last_element);
4514

bk@work.mysql.com's avatar
bk@work.mysql.com committed
4515 4516 4517 4518 4519 4520 4521 4522 4523 4524 4525 4526 4527 4528 4529 4530 4531 4532 4533 4534 4535 4536 4537 4538 4539 4540 4541 4542 4543 4544 4545 4546 4547 4548 4549 4550 4551 4552 4553 4554 4555 4556 4557 4558 4559 4560 4561 4562 4563 4564 4565 4566 4567 4568 4569 4570 4571 4572 4573 4574 4575 4576 4577 4578 4579 4580 4581
  for (element= this; element != &null_element ; )
  {
    last_element=element;
    if (key->cmp_min_to_min(element) > 0)
    {
      par= &element->right; element= element->right;
    }
    else
    {
      par = &element->left; element= element->left;
    }
  }
  *par=key;
  key->parent=last_element;
	/* Link in list */
  if (par == &last_element->left)
  {
    key->next=last_element;
    if ((key->prev=last_element->prev))
      key->prev->next=key;
    last_element->prev=key;
  }
  else
  {
    if ((key->next=last_element->next))
      key->next->prev=key;
    key->prev=last_element;
    last_element->next=key;
  }
  key->left=key->right= &null_element;
  SEL_ARG *root=rb_insert(key);			// rebalance tree
  root->use_count=this->use_count;		// copy root info
  root->elements= this->elements+1;
  root->maybe_flag=this->maybe_flag;
  return root;
}


/*
** Find best key with min <= given key
** Because the call context this should never return 0 to get_range
*/

SEL_ARG *
SEL_ARG::find_range(SEL_ARG *key)
{
  SEL_ARG *element=this,*found=0;

  for (;;)
  {
    if (element == &null_element)
      return found;
    int cmp=element->cmp_min_to_min(key);
    if (cmp == 0)
      return element;
    if (cmp < 0)
    {
      found=element;
      element=element->right;
    }
    else
      element=element->left;
  }
}


/*
4582 4583 4584 4585 4586
  Remove a element from the tree

  SYNOPSIS
    tree_delete()
    key		Key that is to be deleted from tree (this)
4587

4588 4589 4590 4591 4592
  NOTE
    This also frees all sub trees that is used by the element

  RETURN
    root of new tree (with key deleted)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4593 4594 4595 4596 4597 4598 4599
*/

SEL_ARG *
SEL_ARG::tree_delete(SEL_ARG *key)
{
  enum leaf_color remove_color;
  SEL_ARG *root,*nod,**par,*fix_par;
4600 4601 4602 4603
  DBUG_ENTER("tree_delete");

  root=this;
  this->parent= 0;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4604 4605 4606 4607 4608 4609 4610 4611 4612 4613 4614 4615 4616 4617 4618 4619 4620 4621 4622 4623 4624 4625 4626 4627 4628 4629 4630 4631 4632 4633 4634 4635 4636 4637 4638 4639 4640 4641 4642 4643 4644 4645 4646 4647 4648 4649

  /* Unlink from list */
  if (key->prev)
    key->prev->next=key->next;
  if (key->next)
    key->next->prev=key->prev;
  key->increment_use_count(-1);
  if (!key->parent)
    par= &root;
  else
    par=key->parent_ptr();

  if (key->left == &null_element)
  {
    *par=nod=key->right;
    fix_par=key->parent;
    if (nod != &null_element)
      nod->parent=fix_par;
    remove_color= key->color;
  }
  else if (key->right == &null_element)
  {
    *par= nod=key->left;
    nod->parent=fix_par=key->parent;
    remove_color= key->color;
  }
  else
  {
    SEL_ARG *tmp=key->next;			// next bigger key (exist!)
    nod= *tmp->parent_ptr()= tmp->right;	// unlink tmp from tree
    fix_par=tmp->parent;
    if (nod != &null_element)
      nod->parent=fix_par;
    remove_color= tmp->color;

    tmp->parent=key->parent;			// Move node in place of key
    (tmp->left=key->left)->parent=tmp;
    if ((tmp->right=key->right) != &null_element)
      tmp->right->parent=tmp;
    tmp->color=key->color;
    *par=tmp;
    if (fix_par == key)				// key->right == key->next
      fix_par=tmp;				// new parent of nod
  }

  if (root == &null_element)
4650
    DBUG_RETURN(0);				// Maybe root later
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4651 4652 4653 4654 4655 4656 4657
  if (remove_color == BLACK)
    root=rb_delete_fixup(root,nod,fix_par);
  test_rb_tree(root,root->parent);

  root->use_count=this->use_count;		// Fix root counters
  root->elements=this->elements-1;
  root->maybe_flag=this->maybe_flag;
4658
  DBUG_RETURN(root);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4659 4660 4661 4662 4663 4664 4665 4666 4667 4668 4669 4670 4671 4672 4673 4674 4675 4676 4677 4678 4679 4680 4681 4682 4683 4684 4685 4686 4687 4688 4689 4690 4691 4692 4693 4694 4695 4696 4697 4698 4699 4700 4701 4702 4703 4704 4705 4706 4707 4708 4709 4710 4711 4712 4713 4714 4715 4716 4717 4718 4719 4720 4721 4722 4723 4724 4725 4726 4727 4728 4729 4730 4731 4732 4733 4734 4735 4736 4737 4738 4739 4740 4741 4742 4743 4744 4745 4746 4747 4748 4749 4750 4751 4752 4753 4754 4755 4756 4757 4758 4759 4760 4761 4762 4763 4764 4765 4766 4767 4768 4769 4770 4771 4772 4773 4774 4775 4776 4777 4778 4779 4780 4781 4782 4783 4784 4785 4786 4787 4788 4789 4790 4791 4792 4793 4794 4795 4796 4797 4798 4799 4800 4801 4802 4803 4804 4805 4806 4807 4808 4809 4810 4811 4812 4813 4814 4815 4816 4817 4818 4819 4820 4821 4822 4823 4824 4825 4826 4827 4828 4829 4830 4831 4832 4833
}


	/* Functions to fix up the tree after insert and delete */

static void left_rotate(SEL_ARG **root,SEL_ARG *leaf)
{
  SEL_ARG *y=leaf->right;
  leaf->right=y->left;
  if (y->left != &null_element)
    y->left->parent=leaf;
  if (!(y->parent=leaf->parent))
    *root=y;
  else
    *leaf->parent_ptr()=y;
  y->left=leaf;
  leaf->parent=y;
}

static void right_rotate(SEL_ARG **root,SEL_ARG *leaf)
{
  SEL_ARG *y=leaf->left;
  leaf->left=y->right;
  if (y->right != &null_element)
    y->right->parent=leaf;
  if (!(y->parent=leaf->parent))
    *root=y;
  else
    *leaf->parent_ptr()=y;
  y->right=leaf;
  leaf->parent=y;
}


SEL_ARG *
SEL_ARG::rb_insert(SEL_ARG *leaf)
{
  SEL_ARG *y,*par,*par2,*root;
  root= this; root->parent= 0;

  leaf->color=RED;
  while (leaf != root && (par= leaf->parent)->color == RED)
  {					// This can't be root or 1 level under
    if (par == (par2= leaf->parent->parent)->left)
    {
      y= par2->right;
      if (y->color == RED)
      {
	par->color=BLACK;
	y->color=BLACK;
	leaf=par2;
	leaf->color=RED;		/* And the loop continues */
      }
      else
      {
	if (leaf == par->right)
	{
	  left_rotate(&root,leaf->parent);
	  par=leaf;			/* leaf is now parent to old leaf */
	}
	par->color=BLACK;
	par2->color=RED;
	right_rotate(&root,par2);
	break;
      }
    }
    else
    {
      y= par2->left;
      if (y->color == RED)
      {
	par->color=BLACK;
	y->color=BLACK;
	leaf=par2;
	leaf->color=RED;		/* And the loop continues */
      }
      else
      {
	if (leaf == par->left)
	{
	  right_rotate(&root,par);
	  par=leaf;
	}
	par->color=BLACK;
	par2->color=RED;
	left_rotate(&root,par2);
	break;
      }
    }
  }
  root->color=BLACK;
  test_rb_tree(root,root->parent);
  return root;
}


SEL_ARG *rb_delete_fixup(SEL_ARG *root,SEL_ARG *key,SEL_ARG *par)
{
  SEL_ARG *x,*w;
  root->parent=0;

  x= key;
  while (x != root && x->color == SEL_ARG::BLACK)
  {
    if (x == par->left)
    {
      w=par->right;
      if (w->color == SEL_ARG::RED)
      {
	w->color=SEL_ARG::BLACK;
	par->color=SEL_ARG::RED;
	left_rotate(&root,par);
	w=par->right;
      }
      if (w->left->color == SEL_ARG::BLACK && w->right->color == SEL_ARG::BLACK)
      {
	w->color=SEL_ARG::RED;
	x=par;
      }
      else
      {
	if (w->right->color == SEL_ARG::BLACK)
	{
	  w->left->color=SEL_ARG::BLACK;
	  w->color=SEL_ARG::RED;
	  right_rotate(&root,w);
	  w=par->right;
	}
	w->color=par->color;
	par->color=SEL_ARG::BLACK;
	w->right->color=SEL_ARG::BLACK;
	left_rotate(&root,par);
	x=root;
	break;
      }
    }
    else
    {
      w=par->left;
      if (w->color == SEL_ARG::RED)
      {
	w->color=SEL_ARG::BLACK;
	par->color=SEL_ARG::RED;
	right_rotate(&root,par);
	w=par->left;
      }
      if (w->right->color == SEL_ARG::BLACK && w->left->color == SEL_ARG::BLACK)
      {
	w->color=SEL_ARG::RED;
	x=par;
      }
      else
      {
	if (w->left->color == SEL_ARG::BLACK)
	{
	  w->right->color=SEL_ARG::BLACK;
	  w->color=SEL_ARG::RED;
	  left_rotate(&root,w);
	  w=par->left;
	}
	w->color=par->color;
	par->color=SEL_ARG::BLACK;
	w->left->color=SEL_ARG::BLACK;
	right_rotate(&root,par);
	x=root;
	break;
      }
    }
    par=x->parent;
  }
  x->color=SEL_ARG::BLACK;
  return root;
}


4834
	/* Test that the properties for a red-black tree hold */
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4835 4836 4837 4838 4839 4840 4841 4842 4843 4844 4845 4846 4847 4848 4849 4850 4851 4852 4853 4854 4855 4856 4857 4858 4859 4860 4861 4862 4863 4864 4865 4866 4867 4868 4869 4870 4871 4872 4873 4874 4875 4876 4877 4878 4879 4880 4881 4882 4883 4884 4885 4886 4887 4888 4889 4890

#ifdef EXTRA_DEBUG
int test_rb_tree(SEL_ARG *element,SEL_ARG *parent)
{
  int count_l,count_r;

  if (element == &null_element)
    return 0;					// Found end of tree
  if (element->parent != parent)
  {
    sql_print_error("Wrong tree: Parent doesn't point at parent");
    return -1;
  }
  if (element->color == SEL_ARG::RED &&
      (element->left->color == SEL_ARG::RED ||
       element->right->color == SEL_ARG::RED))
  {
    sql_print_error("Wrong tree: Found two red in a row");
    return -1;
  }
  if (element->left == element->right && element->left != &null_element)
  {						// Dummy test
    sql_print_error("Wrong tree: Found right == left");
    return -1;
  }
  count_l=test_rb_tree(element->left,element);
  count_r=test_rb_tree(element->right,element);
  if (count_l >= 0 && count_r >= 0)
  {
    if (count_l == count_r)
      return count_l+(element->color == SEL_ARG::BLACK);
    sql_print_error("Wrong tree: Incorrect black-count: %d - %d",
	    count_l,count_r);
  }
  return -1;					// Error, no more warnings
}

static ulong count_key_part_usage(SEL_ARG *root, SEL_ARG *key)
{
  ulong count= 0;
  for (root=root->first(); root ; root=root->next)
  {
    if (root->next_key_part)
    {
      if (root->next_key_part == key)
	count++;
      if (root->next_key_part->part < key->part)
	count+=count_key_part_usage(root->next_key_part,key);
    }
  }
  return count;
}


void SEL_ARG::test_use_count(SEL_ARG *root)
{
4891
  uint e_count=0;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4892 4893
  if (this == root && use_count != 1)
  {
monty@mysql.com's avatar
monty@mysql.com committed
4894
    sql_print_information("Use_count: Wrong count %lu for root",use_count);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4895 4896 4897 4898 4899 4900 4901 4902 4903 4904 4905 4906
    return;
  }
  if (this->type != SEL_ARG::KEY_RANGE)
    return;
  for (SEL_ARG *pos=first(); pos ; pos=pos->next)
  {
    e_count++;
    if (pos->next_key_part)
    {
      ulong count=count_key_part_usage(root,pos->next_key_part);
      if (count > pos->next_key_part->use_count)
      {
monty@mysql.com's avatar
monty@mysql.com committed
4907
	sql_print_information("Use_count: Wrong count for key at 0x%lx, %lu should be %lu",
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4908 4909 4910 4911 4912 4913 4914
			pos,pos->next_key_part->use_count,count);
	return;
      }
      pos->next_key_part->test_use_count(root);
    }
  }
  if (e_count != elements)
monty@mysql.com's avatar
monty@mysql.com committed
4915
    sql_print_warning("Wrong use count: %u (should be %u) for tree at 0x%lx",
4916
		    e_count, elements, (gptr) this);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4917 4918 4919 4920 4921
}

#endif


4922 4923 4924 4925 4926 4927 4928 4929 4930 4931
/*
  Calculate estimate of number records that will be retrieved by a range
  scan on given index using given SEL_ARG intervals tree.
  SYNOPSIS
    check_quick_select
      param  Parameter from test_quick_select
      idx    Number of index to use in PARAM::key SEL_TREE::key
      tree   Transformed selection condition, tree->key[idx] holds intervals
             tree to be used for scanning.
  NOTES
4932
    param->is_ror_scan is set to reflect if the key scan is a ROR (see
4933
    is_key_scan_ror function for more info)
4934
    param->table->quick_*, param->range_count (and maybe others) are
4935
    updated with data of given key scan, see check_quick_keys for details.
4936 4937

  RETURN
4938
    Estimate # of records to be retrieved.
4939
    HA_POS_ERROR if estimate calculation failed due to table handler problems.
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4940

4941
*/
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4942 4943 4944 4945 4946

static ha_rows
check_quick_select(PARAM *param,uint idx,SEL_ARG *tree)
{
  ha_rows records;
4947 4948
  bool    cpk_scan;
  uint key;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4949
  DBUG_ENTER("check_quick_select");
4950

monty@mysql.com's avatar
monty@mysql.com committed
4951
  param->is_ror_scan= FALSE;
4952

bk@work.mysql.com's avatar
bk@work.mysql.com committed
4953 4954
  if (!tree)
    DBUG_RETURN(HA_POS_ERROR);			// Can't use it
monty@narttu.mysql.fi's avatar
monty@narttu.mysql.fi committed
4955 4956
  param->max_key_part=0;
  param->range_count=0;
4957 4958
  key= param->real_keynr[idx];

bk@work.mysql.com's avatar
bk@work.mysql.com committed
4959 4960 4961 4962
  if (tree->type == SEL_ARG::IMPOSSIBLE)
    DBUG_RETURN(0L);				// Impossible select. return
  if (tree->type != SEL_ARG::KEY_RANGE || tree->part != 0)
    DBUG_RETURN(HA_POS_ERROR);				// Don't use tree
4963 4964 4965 4966 4967

  enum ha_key_alg key_alg= param->table->key_info[key].algorithm;
  if ((key_alg != HA_KEY_ALG_BTREE) && (key_alg!= HA_KEY_ALG_UNDEF))
  {
    /* Records are not ordered by rowid for other types of indexes. */
monty@mysql.com's avatar
monty@mysql.com committed
4968
    cpk_scan= FALSE;
4969 4970 4971 4972 4973 4974 4975
  }
  else
  {
    /*
      Clustered PK scan is a special case, check_quick_keys doesn't recognize
      CPK scans as ROR scans (while actually any CPK scan is a ROR scan).
    */
4976 4977
    cpk_scan= ((param->table->s->primary_key == param->real_keynr[idx]) &&
               param->table->file->primary_key_is_clustered());
4978
    param->is_ror_scan= !cpk_scan;
4979 4980
  }

bk@work.mysql.com's avatar
bk@work.mysql.com committed
4981 4982
  records=check_quick_keys(param,idx,tree,param->min_key,0,param->max_key,0);
  if (records != HA_POS_ERROR)
4983
  {
4984
    param->table->quick_keys.set_bit(key);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4985 4986
    param->table->quick_rows[key]=records;
    param->table->quick_key_parts[key]=param->max_key_part+1;
4987

4988
    if (cpk_scan)
monty@mysql.com's avatar
monty@mysql.com committed
4989
      param->is_ror_scan= TRUE;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4990
  }
4991
  DBUG_PRINT("exit", ("Records: %lu", (ulong) records));
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4992 4993 4994 4995
  DBUG_RETURN(records);
}


4996
/*
4997 4998
  Recursively calculate estimate of # rows that will be retrieved by
  key scan on key idx.
4999 5000
  SYNOPSIS
    check_quick_keys()
5001
      param         Parameter from test_quick select function.
5002
      idx           Number of key to use in PARAM::keys in list of used keys
5003 5004 5005
                    (param->real_keynr[idx] holds the key number in table)
      key_tree      SEL_ARG tree being examined.
      min_key       Buffer with partial min key value tuple
5006
      min_key_flag
5007
      max_key       Buffer with partial max key value tuple
5008 5009
      max_key_flag

5010
  NOTES
5011 5012
    The function does the recursive descent on the tree via SEL_ARG::left,
    SEL_ARG::right, and SEL_ARG::next_key_part edges. The #rows estimates
5013 5014
    are calculated using records_in_range calls at the leaf nodes and then
    summed.
5015

5016 5017
    param->min_key and param->max_key are used to hold prefixes of key value
    tuples.
5018 5019

    The side effects are:
5020

5021 5022
    param->max_key_part is updated to hold the maximum number of key parts used
      in scan minus 1.
5023 5024

    param->range_count is incremented if the function finds a range that
5025
      wasn't counted by the caller.
5026

5027 5028 5029
    param->is_ror_scan is cleared if the function detects that the key scan is
      not a Rowid-Ordered Retrieval scan ( see comments for is_key_scan_ror
      function for description of which key scans are ROR scans)
5030 5031
*/

bk@work.mysql.com's avatar
bk@work.mysql.com committed
5032 5033 5034 5035 5036
static ha_rows
check_quick_keys(PARAM *param,uint idx,SEL_ARG *key_tree,
		 char *min_key,uint min_key_flag, char *max_key,
		 uint max_key_flag)
{
monty@mysql.com's avatar
monty@mysql.com committed
5037 5038 5039
  ha_rows records=0, tmp;
  uint tmp_min_flag, tmp_max_flag, keynr, min_key_length, max_key_length;
  char *tmp_min_key, *tmp_max_key;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
5040 5041 5042 5043

  param->max_key_part=max(param->max_key_part,key_tree->part);
  if (key_tree->left != &null_element)
  {
5044 5045 5046 5047 5048 5049
    /*
      There are at least two intervals for current key part, i.e. condition
      was converted to something like
        (keyXpartY less/equals c1) OR (keyXpartY more/equals c2).
      This is not a ROR scan if the key is not Clustered Primary Key.
    */
monty@mysql.com's avatar
monty@mysql.com committed
5050
    param->is_ror_scan= FALSE;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
5051 5052 5053 5054 5055 5056
    records=check_quick_keys(param,idx,key_tree->left,min_key,min_key_flag,
			     max_key,max_key_flag);
    if (records == HA_POS_ERROR)			// Impossible
      return records;
  }

monty@mysql.com's avatar
monty@mysql.com committed
5057 5058
  tmp_min_key= min_key;
  tmp_max_key= max_key;
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
5059
  key_tree->store(param->key[idx][key_tree->part].store_length,
bk@work.mysql.com's avatar
bk@work.mysql.com committed
5060
		  &tmp_min_key,min_key_flag,&tmp_max_key,max_key_flag);
monty@mysql.com's avatar
monty@mysql.com committed
5061 5062
  min_key_length= (uint) (tmp_min_key- param->min_key);
  max_key_length= (uint) (tmp_max_key- param->max_key);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
5063

5064 5065
  if (param->is_ror_scan)
  {
5066
    /*
5067
      If the index doesn't cover entire key, mark the scan as non-ROR scan.
5068
      Actually we're cutting off some ROR scans here.
5069 5070 5071
    */
    uint16 fieldnr= param->table->key_info[param->real_keynr[idx]].
                    key_part[key_tree->part].fieldnr - 1;
5072
    if (param->table->field[fieldnr]->key_length() !=
5073
        param->key[idx][key_tree->part].length)
monty@mysql.com's avatar
monty@mysql.com committed
5074
      param->is_ror_scan= FALSE;
5075 5076
  }

bk@work.mysql.com's avatar
bk@work.mysql.com committed
5077 5078 5079 5080 5081 5082 5083 5084 5085 5086 5087 5088 5089
  if (key_tree->next_key_part &&
      key_tree->next_key_part->part == key_tree->part+1 &&
      key_tree->next_key_part->type == SEL_ARG::KEY_RANGE)
  {						// const key as prefix
    if (min_key_length == max_key_length &&
	!memcmp(min_key,max_key, (uint) (tmp_max_key - max_key)) &&
	!key_tree->min_flag && !key_tree->max_flag)
    {
      tmp=check_quick_keys(param,idx,key_tree->next_key_part,
			   tmp_min_key, min_key_flag | key_tree->min_flag,
			   tmp_max_key, max_key_flag | key_tree->max_flag);
      goto end;					// Ugly, but efficient
    }
5090
    else
5091 5092
    {
      /* The interval for current key part is not c1 <= keyXpartY <= c1 */
monty@mysql.com's avatar
monty@mysql.com committed
5093
      param->is_ror_scan= FALSE;
5094
    }
5095

bk@work.mysql.com's avatar
bk@work.mysql.com committed
5096 5097 5098 5099 5100 5101 5102 5103 5104 5105 5106 5107 5108 5109 5110 5111 5112 5113
    tmp_min_flag=key_tree->min_flag;
    tmp_max_flag=key_tree->max_flag;
    if (!tmp_min_flag)
      key_tree->next_key_part->store_min_key(param->key[idx], &tmp_min_key,
					     &tmp_min_flag);
    if (!tmp_max_flag)
      key_tree->next_key_part->store_max_key(param->key[idx], &tmp_max_key,
					     &tmp_max_flag);
    min_key_length= (uint) (tmp_min_key- param->min_key);
    max_key_length= (uint) (tmp_max_key- param->max_key);
  }
  else
  {
    tmp_min_flag=min_key_flag | key_tree->min_flag;
    tmp_max_flag=max_key_flag | key_tree->max_flag;
  }

  keynr=param->real_keynr[idx];
monty@narttu.mysql.fi's avatar
monty@narttu.mysql.fi committed
5114
  param->range_count++;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
5115 5116
  if (!tmp_min_flag && ! tmp_max_flag &&
      (uint) key_tree->part+1 == param->table->key_info[keynr].key_parts &&
5117 5118
      (param->table->key_info[keynr].flags & (HA_NOSAME | HA_END_SPACE_KEY)) ==
      HA_NOSAME &&
bk@work.mysql.com's avatar
bk@work.mysql.com committed
5119 5120 5121 5122
      min_key_length == max_key_length &&
      !memcmp(param->min_key,param->max_key,min_key_length))
    tmp=1;					// Max one record
  else
5123
  {
5124 5125
    if (param->is_ror_scan)
    {
5126 5127 5128 5129 5130 5131 5132 5133 5134
      /*
        If we get here, the condition on the key was converted to form
        "(keyXpart1 = c1) AND ... AND (keyXpart{key_tree->part - 1} = cN) AND
          somecond(keyXpart{key_tree->part})"
        Check if
          somecond is "keyXpart{key_tree->part} = const" and
          uncovered "tail" of KeyX parts is either empty or is identical to
          first members of clustered primary key.
      */
5135 5136
      if (!(min_key_length == max_key_length &&
            !memcmp(min_key,max_key, (uint) (tmp_max_key - max_key)) &&
5137
            !key_tree->min_flag && !key_tree->max_flag &&
5138
            is_key_scan_ror(param, keynr, key_tree->part + 1)))
monty@mysql.com's avatar
monty@mysql.com committed
5139
        param->is_ror_scan= FALSE;
5140 5141
    }

5142
    if (tmp_min_flag & GEOM_FLAG)
5143
    {
5144 5145 5146 5147 5148 5149 5150 5151
      key_range min_range;
      min_range.key=    (byte*) param->min_key;
      min_range.length= min_key_length;
      /* In this case tmp_min_flag contains the handler-read-function */
      min_range.flag=   (ha_rkey_function) (tmp_min_flag ^ GEOM_FLAG);

      tmp= param->table->file->records_in_range(keynr, &min_range,
                                                (key_range*) 0);
5152 5153 5154
    }
    else
    {
5155 5156 5157 5158 5159 5160
      key_range min_range, max_range;

      min_range.key=    (byte*) param->min_key;
      min_range.length= min_key_length;
      min_range.flag=   (tmp_min_flag & NEAR_MIN ? HA_READ_AFTER_KEY :
                         HA_READ_KEY_EXACT);
monty@mysql.com's avatar
monty@mysql.com committed
5161
      max_range.key=    (byte*) param->max_key;
5162 5163 5164 5165 5166 5167 5168 5169
      max_range.length= max_key_length;
      max_range.flag=   (tmp_max_flag & NEAR_MAX ?
                         HA_READ_BEFORE_KEY : HA_READ_AFTER_KEY);
      tmp=param->table->file->records_in_range(keynr,
                                               (min_key_length ? &min_range :
                                                (key_range*) 0),
                                               (max_key_length ? &max_range :
                                                (key_range*) 0));
5170 5171
    }
  }
bk@work.mysql.com's avatar
bk@work.mysql.com committed
5172 5173 5174 5175 5176 5177
 end:
  if (tmp == HA_POS_ERROR)			// Impossible range
    return tmp;
  records+=tmp;
  if (key_tree->right != &null_element)
  {
5178 5179 5180 5181 5182 5183
    /*
      There are at least two intervals for current key part, i.e. condition
      was converted to something like
        (keyXpartY less/equals c1) OR (keyXpartY more/equals c2).
      This is not a ROR scan if the key is not Clustered Primary Key.
    */
monty@mysql.com's avatar
monty@mysql.com committed
5184
    param->is_ror_scan= FALSE;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
5185 5186 5187 5188 5189 5190 5191 5192 5193
    tmp=check_quick_keys(param,idx,key_tree->right,min_key,min_key_flag,
			 max_key,max_key_flag);
    if (tmp == HA_POS_ERROR)
      return tmp;
    records+=tmp;
  }
  return records;
}

5194

5195
/*
5196
  Check if key scan on given index with equality conditions on first n key
5197 5198 5199 5200
  parts is a ROR scan.

  SYNOPSIS
    is_key_scan_ror()
5201
      param  Parameter from test_quick_select
5202 5203 5204 5205
      keynr  Number of key in the table. The key must not be a clustered
             primary key.
      nparts Number of first key parts for which equality conditions
             are present.
5206

5207 5208 5209
  NOTES
    ROR (Rowid Ordered Retrieval) key scan is a key scan that produces
    ordered sequence of rowids (ha_xxx::cmp_ref is the comparison function)
5210

5211 5212 5213
    An index scan is a ROR scan if it is done using a condition in form

        "key1_1=c_1 AND ... AND key1_n=c_n"  (1)
5214

5215 5216
    where the index is defined on (key1_1, ..., key1_N [,a_1, ..., a_n])

5217
    and the table has a clustered Primary Key
5218

5219
    PRIMARY KEY(a_1, ..., a_n, b1, ..., b_k) with first key parts being
5220
    identical to uncovered parts ot the key being scanned (2)
5221 5222

    Scans on HASH indexes are not ROR scans,
5223 5224 5225 5226 5227 5228
    any range scan on clustered primary key is ROR scan  (3)

    Check (1) is made in check_quick_keys()
    Check (3) is made check_quick_select()
    Check (2) is made by this function.

5229
  RETURN
monty@mysql.com's avatar
monty@mysql.com committed
5230 5231
    TRUE  If the scan is ROR-scan
    FALSE otherwise
5232
*/
5233

5234 5235 5236 5237
static bool is_key_scan_ror(PARAM *param, uint keynr, uint8 nparts)
{
  KEY *table_key= param->table->key_info + keynr;
  KEY_PART_INFO *key_part= table_key->key_part + nparts;
5238 5239 5240
  KEY_PART_INFO *key_part_end= (table_key->key_part +
                                table_key->key_parts);
  uint pk_number;
5241

5242
  if (key_part == key_part_end)
monty@mysql.com's avatar
monty@mysql.com committed
5243
    return TRUE;
5244
  pk_number= param->table->s->primary_key;
5245
  if (!param->table->file->primary_key_is_clustered() || pk_number == MAX_KEY)
monty@mysql.com's avatar
monty@mysql.com committed
5246
    return FALSE;
5247 5248

  KEY_PART_INFO *pk_part= param->table->key_info[pk_number].key_part;
5249
  KEY_PART_INFO *pk_part_end= pk_part +
5250
                              param->table->key_info[pk_number].key_parts;
5251
  for(;(key_part!=key_part_end) && (pk_part != pk_part_end);
5252 5253
      ++key_part, ++pk_part)
  {
5254
    if ((key_part->field != pk_part->field) ||
5255
        (key_part->length != pk_part->length))
5256
      return FALSE;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
5257
  }
5258
  return (key_part == key_part_end);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
5259 5260 5261
}


5262 5263
/*
  Create a QUICK_RANGE_SELECT from given key and SEL_ARG tree for that key.
5264

5265 5266
  SYNOPSIS
    get_quick_select()
5267
      param
5268
      idx          Index of used key in param->key.
5269 5270
      key_tree     SEL_ARG tree for the used key
      parent_alloc If not NULL, use it to allocate memory for
5271
                   quick select data. Otherwise use quick->alloc.
5272
  NOTES
5273
    The caller must call QUICK_SELECT::init for returned quick select
5274

5275
    CAUTION! This function may change thd->mem_root to a MEM_ROOT which will be
5276
    deallocated when the returned quick select is deleted.
5277 5278 5279 5280

  RETURN
    NULL on error
    otherwise created quick select
5281
*/
5282

5283 5284 5285
QUICK_RANGE_SELECT *
get_quick_select(PARAM *param,uint idx,SEL_ARG *key_tree,
                 MEM_ROOT *parent_alloc)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
5286
{
5287
  QUICK_RANGE_SELECT *quick;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
5288
  DBUG_ENTER("get_quick_select");
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
5289 5290 5291 5292 5293 5294 5295 5296 5297

  if (param->table->key_info[param->real_keynr[idx]].flags & HA_SPATIAL)
    quick=new QUICK_RANGE_SELECT_GEOM(param->thd, param->table,
                                      param->real_keynr[idx],
                                      test(parent_alloc),
                                      parent_alloc);
  else
    quick=new QUICK_RANGE_SELECT(param->thd, param->table,
                                 param->real_keynr[idx],
monty@mysql.com's avatar
monty@mysql.com committed
5298
                                 test(parent_alloc));
5299

pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
5300
  if (quick)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
5301 5302 5303 5304 5305 5306 5307 5308 5309 5310 5311
  {
    if (quick->error ||
	get_quick_keys(param,quick,param->key[idx],key_tree,param->min_key,0,
		       param->max_key,0))
    {
      delete quick;
      quick=0;
    }
    else
    {
      quick->key_parts=(KEY_PART*)
5312 5313 5314 5315
        memdup_root(parent_alloc? parent_alloc : &quick->alloc,
                    (char*) param->key[idx],
                    sizeof(KEY_PART)*
                    param->table->key_info[param->real_keynr[idx]].key_parts);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
5316
    }
5317
  }
bk@work.mysql.com's avatar
bk@work.mysql.com committed
5318 5319 5320 5321 5322 5323 5324
  DBUG_RETURN(quick);
}


/*
** Fix this to get all possible sub_ranges
*/
5325 5326
bool
get_quick_keys(PARAM *param,QUICK_RANGE_SELECT *quick,KEY_PART *key,
bk@work.mysql.com's avatar
bk@work.mysql.com committed
5327 5328 5329 5330 5331 5332 5333 5334 5335 5336 5337 5338 5339
	       SEL_ARG *key_tree,char *min_key,uint min_key_flag,
	       char *max_key, uint max_key_flag)
{
  QUICK_RANGE *range;
  uint flag;

  if (key_tree->left != &null_element)
  {
    if (get_quick_keys(param,quick,key,key_tree->left,
		       min_key,min_key_flag, max_key, max_key_flag))
      return 1;
  }
  char *tmp_min_key=min_key,*tmp_max_key=max_key;
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
5340
  key_tree->store(key[key_tree->part].store_length,
bk@work.mysql.com's avatar
bk@work.mysql.com committed
5341 5342 5343 5344 5345 5346 5347 5348 5349 5350 5351 5352 5353 5354 5355 5356 5357 5358 5359 5360 5361 5362 5363 5364 5365 5366 5367 5368
		  &tmp_min_key,min_key_flag,&tmp_max_key,max_key_flag);

  if (key_tree->next_key_part &&
      key_tree->next_key_part->part == key_tree->part+1 &&
      key_tree->next_key_part->type == SEL_ARG::KEY_RANGE)
  {						  // const key as prefix
    if (!((tmp_min_key - min_key) != (tmp_max_key - max_key) ||
	  memcmp(min_key,max_key, (uint) (tmp_max_key - max_key)) ||
	  key_tree->min_flag || key_tree->max_flag))
    {
      if (get_quick_keys(param,quick,key,key_tree->next_key_part,
			 tmp_min_key, min_key_flag | key_tree->min_flag,
			 tmp_max_key, max_key_flag | key_tree->max_flag))
	return 1;
      goto end;					// Ugly, but efficient
    }
    {
      uint tmp_min_flag=key_tree->min_flag,tmp_max_flag=key_tree->max_flag;
      if (!tmp_min_flag)
	key_tree->next_key_part->store_min_key(key, &tmp_min_key,
					       &tmp_min_flag);
      if (!tmp_max_flag)
	key_tree->next_key_part->store_max_key(key, &tmp_max_key,
					       &tmp_max_flag);
      flag=tmp_min_flag | tmp_max_flag;
    }
  }
  else
5369 5370 5371 5372
  {
    flag = (key_tree->min_flag & GEOM_FLAG) ?
      key_tree->min_flag : key_tree->min_flag | key_tree->max_flag;
  }
bk@work.mysql.com's avatar
bk@work.mysql.com committed
5373

5374 5375 5376 5377 5378
  /*
    Ensure that some part of min_key and max_key are used.  If not,
    regard this as no lower/upper range
  */
  if ((flag & GEOM_FLAG) == 0)
5379 5380 5381 5382 5383 5384 5385 5386 5387 5388
  {
    if (tmp_min_key != param->min_key)
      flag&= ~NO_MIN_RANGE;
    else
      flag|= NO_MIN_RANGE;
    if (tmp_max_key != param->max_key)
      flag&= ~NO_MAX_RANGE;
    else
      flag|= NO_MAX_RANGE;
  }
bk@work.mysql.com's avatar
bk@work.mysql.com committed
5389 5390 5391 5392 5393 5394 5395 5396
  if (flag == 0)
  {
    uint length= (uint) (tmp_min_key - param->min_key);
    if (length == (uint) (tmp_max_key - param->max_key) &&
	!memcmp(param->min_key,param->max_key,length))
    {
      KEY *table_key=quick->head->key_info+quick->index;
      flag=EQ_RANGE;
5397 5398
      if ((table_key->flags & (HA_NOSAME | HA_END_SPACE_KEY)) == HA_NOSAME &&
	  key->part == table_key->key_parts-1)
5399 5400 5401 5402 5403 5404 5405 5406 5407
      {
	if (!(table_key->flags & HA_NULL_PART_KEY) ||
	    !null_part_in_key(key,
			      param->min_key,
			      (uint) (tmp_min_key - param->min_key)))
	  flag|= UNIQUE_RANGE;
	else
	  flag|= NULL_RANGE;
      }
bk@work.mysql.com's avatar
bk@work.mysql.com committed
5408 5409 5410 5411
    }
  }

  /* Get range for retrieving rows in QUICK_SELECT::get_next */
5412
  if (!(range= new QUICK_RANGE((const char *) param->min_key,
5413
			       (uint) (tmp_min_key - param->min_key),
5414
			       (const char *) param->max_key,
5415 5416
			       (uint) (tmp_max_key - param->max_key),
			       flag)))
5417 5418
    return 1;			// out of memory

bk@work.mysql.com's avatar
bk@work.mysql.com committed
5419 5420
  set_if_bigger(quick->max_used_key_length,range->min_length);
  set_if_bigger(quick->max_used_key_length,range->max_length);
5421
  set_if_bigger(quick->used_key_parts, (uint) key_tree->part+1);
5422 5423 5424
  if (insert_dynamic(&quick->ranges, (gptr)&range))
    return 1;

bk@work.mysql.com's avatar
bk@work.mysql.com committed
5425 5426 5427 5428 5429 5430 5431 5432 5433 5434 5435 5436
 end:
  if (key_tree->right != &null_element)
    return get_quick_keys(param,quick,key,key_tree->right,
			  min_key,min_key_flag,
			  max_key,max_key_flag);
  return 0;
}

/*
  Return 1 if there is only one range and this uses the whole primary key
*/

5437
bool QUICK_RANGE_SELECT::unique_key_range()
bk@work.mysql.com's avatar
bk@work.mysql.com committed
5438 5439 5440
{
  if (ranges.elements == 1)
  {
5441 5442
    QUICK_RANGE *tmp= *((QUICK_RANGE**)ranges.buffer);
    if ((tmp->flag & (EQ_RANGE | NULL_RANGE)) == EQ_RANGE)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
5443 5444
    {
      KEY *key=head->key_info+index;
5445
      return ((key->flags & (HA_NOSAME | HA_END_SPACE_KEY)) == HA_NOSAME &&
bk@work.mysql.com's avatar
bk@work.mysql.com committed
5446 5447 5448 5449 5450 5451
	      key->key_length == tmp->min_length);
    }
  }
  return 0;
}

5452

monty@mysql.com's avatar
monty@mysql.com committed
5453
/* Returns TRUE if any part of the key is NULL */
5454 5455 5456

static bool null_part_in_key(KEY_PART *key_part, const char *key, uint length)
{
5457
  for (const char *end=key+length ;
5458
       key < end;
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
5459
       key+= key_part++->store_length)
5460
  {
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
5461 5462
    if (key_part->null_bit && *key)
      return 1;
5463 5464 5465 5466
  }
  return 0;
}

pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
5467

5468 5469
bool QUICK_SELECT_I::check_if_keys_used(List<Item> *fields)
{
5470
  return check_if_key_used(head, index, *fields);
5471 5472 5473 5474 5475 5476 5477 5478 5479 5480 5481 5482 5483 5484 5485 5486 5487 5488 5489 5490 5491 5492 5493 5494 5495 5496 5497 5498 5499 5500 5501 5502 5503 5504 5505 5506 5507 5508
}

bool QUICK_INDEX_MERGE_SELECT::check_if_keys_used(List<Item> *fields)
{
  QUICK_RANGE_SELECT *quick;
  List_iterator_fast<QUICK_RANGE_SELECT> it(quick_selects);
  while ((quick= it++))
  {
    if (check_if_key_used(head, quick->index, *fields))
      return 1;
  }
  return 0;
}

bool QUICK_ROR_INTERSECT_SELECT::check_if_keys_used(List<Item> *fields)
{
  QUICK_RANGE_SELECT *quick;
  List_iterator_fast<QUICK_RANGE_SELECT> it(quick_selects);
  while ((quick= it++))
  {
    if (check_if_key_used(head, quick->index, *fields))
      return 1;
  }
  return 0;
}

bool QUICK_ROR_UNION_SELECT::check_if_keys_used(List<Item> *fields)
{
  QUICK_SELECT_I *quick;
  List_iterator_fast<QUICK_SELECT_I> it(quick_selects);
  while ((quick= it++))
  {
    if (quick->check_if_keys_used(fields))
      return 1;
  }
  return 0;
}

monty@mysql.com's avatar
monty@mysql.com committed
5509

bk@work.mysql.com's avatar
bk@work.mysql.com committed
5510
/****************************************************************************
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
5511
  Create a QUICK RANGE based on a key
monty@mysql.com's avatar
monty@mysql.com committed
5512 5513
  This allocates things in a new memory root, as this may be called many times
  during a query.
bk@work.mysql.com's avatar
bk@work.mysql.com committed
5514 5515
****************************************************************************/

5516
QUICK_RANGE_SELECT *get_quick_select_for_ref(THD *thd, TABLE *table,
sergefp@mysql.com's avatar
sergefp@mysql.com committed
5517
                                             TABLE_REF *ref)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
5518
{
5519 5520
  MEM_ROOT *old_root= thd->mem_root;
  /* The following call may change thd->mem_root */
monty@mysql.com's avatar
monty@mysql.com committed
5521
  QUICK_RANGE_SELECT *quick= new QUICK_RANGE_SELECT(thd, table, ref->key, 0);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
5522 5523
  KEY *key_info = &table->key_info[ref->key];
  KEY_PART *key_part;
serg@serg.mylan's avatar
serg@serg.mylan committed
5524
  QUICK_RANGE *range;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
5525 5526 5527
  uint part;

  if (!quick)
5528
    return 0;			/* no ranges found */
sergefp@mysql.com's avatar
sergefp@mysql.com committed
5529
  if (quick->init())
5530 5531
  {
    delete quick;
monty@mysql.com's avatar
monty@mysql.com committed
5532
    goto err;
5533
  }
5534

monty@mysql.com's avatar
monty@mysql.com committed
5535 5536 5537
  if (cp_buffer_from_ref(ref) && thd->is_fatal_error ||
      !(range= new QUICK_RANGE()))
    goto err;                                   // out of memory
5538

bk@work.mysql.com's avatar
bk@work.mysql.com committed
5539 5540 5541
  range->min_key=range->max_key=(char*) ref->key_buff;
  range->min_length=range->max_length=ref->key_length;
  range->flag= ((ref->key_length == key_info->key_length &&
5542 5543
		 (key_info->flags & (HA_NOSAME | HA_END_SPACE_KEY)) ==
		 HA_NOSAME) ? EQ_RANGE : 0);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
5544 5545

  if (!(quick->key_parts=key_part=(KEY_PART *)
5546
	alloc_root(&quick->alloc,sizeof(KEY_PART)*ref->key_parts)))
bk@work.mysql.com's avatar
bk@work.mysql.com committed
5547 5548 5549 5550 5551 5552
    goto err;

  for (part=0 ; part < ref->key_parts ;part++,key_part++)
  {
    key_part->part=part;
    key_part->field=        key_info->key_part[part].field;
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
5553 5554
    key_part->length=  	    key_info->key_part[part].length;
    key_part->store_length= key_info->key_part[part].store_length;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
5555 5556
    key_part->null_bit=     key_info->key_part[part].null_bit;
  }
pem@mysql.com's avatar
pem@mysql.com committed
5557
  if (insert_dynamic(&quick->ranges,(gptr)&range))
5558 5559
    goto err;

5560
  /*
5561 5562 5563 5564 5565
     Add a NULL range if REF_OR_NULL optimization is used.
     For example:
       if we have "WHERE A=2 OR A IS NULL" we created the (A=2) range above
       and have ref->null_ref_key set. Will create a new NULL range here.
  */
5566 5567 5568 5569 5570
  if (ref->null_ref_key)
  {
    QUICK_RANGE *null_range;

    *ref->null_ref_key= 1;		// Set null byte then create a range
5571 5572
    if (!(null_range= new QUICK_RANGE((char*)ref->key_buff, ref->key_length,
				      (char*)ref->key_buff, ref->key_length,
5573 5574 5575
				      EQ_RANGE)))
      goto err;
    *ref->null_ref_key= 0;		// Clear null byte
pem@mysql.com's avatar
pem@mysql.com committed
5576
    if (insert_dynamic(&quick->ranges,(gptr)&null_range))
5577 5578 5579
      goto err;
  }

5580
  thd->mem_root= old_root;
5581
  return quick;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
5582 5583

err:
5584
  thd->mem_root= old_root;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
5585 5586 5587 5588
  delete quick;
  return 0;
}

5589 5590

/*
sergefp@mysql.com's avatar
sergefp@mysql.com committed
5591 5592 5593 5594 5595 5596
  Perform key scans for all used indexes (except CPK), get rowids and merge 
  them into an ordered non-recurrent sequence of rowids.
  
  The merge/duplicate removal is performed using Unique class. We put all
  rowids into Unique, get the sorted sequence and destroy the Unique.
  
5597
  If table has a clustered primary key that covers all rows (TRUE for bdb
5598
     and innodb currently) and one of the index_merge scans is a scan on PK,
5599
  then
sergefp@mysql.com's avatar
sergefp@mysql.com committed
5600 5601
    rows that will be retrieved by PK scan are not put into Unique and 
    primary key scan is not performed here, it is performed later separately.
5602

5603 5604 5605
  RETURN
    0     OK
    other error
5606
*/
5607

sergefp@mysql.com's avatar
sergefp@mysql.com committed
5608
int QUICK_INDEX_MERGE_SELECT::read_keys_and_merge()
5609
{
sergefp@mysql.com's avatar
sergefp@mysql.com committed
5610 5611
  List_iterator_fast<QUICK_RANGE_SELECT> cur_quick_it(quick_selects);
  QUICK_RANGE_SELECT* cur_quick;
5612
  int result;
sergefp@mysql.com's avatar
sergefp@mysql.com committed
5613
  Unique *unique;
5614
  DBUG_ENTER("QUICK_INDEX_MERGE_SELECT::prepare_unique");
5615

5616
  /* We're going to just read rowids. */
5617 5618
  if (head->file->extra(HA_EXTRA_KEYREAD))
    DBUG_RETURN(1);
5619

5620 5621
  /*
    Make innodb retrieve all PK member fields, so
5622
     * ha_innobase::position (which uses them) call works.
5623
     * We can filter out rows that will be retrieved by clustered PK.
5624
    (This also creates a deficiency - it is possible that we will retrieve
5625
     parts of key that are not used by current query at all.)
5626
  */
5627 5628
  if (head->file->extra(HA_EXTRA_RETRIEVE_PRIMARY_KEY))
    DBUG_RETURN(1);
5629

sergefp@mysql.com's avatar
sergefp@mysql.com committed
5630 5631
  cur_quick_it.rewind();
  cur_quick= cur_quick_it++;
5632
  DBUG_ASSERT(cur_quick != 0);
sergefp@mysql.com's avatar
sergefp@mysql.com committed
5633 5634 5635 5636 5637
  
  /*
    We reuse the same instance of handler so we need to call both init and 
    reset here.
  */
sergefp@mysql.com's avatar
sergefp@mysql.com committed
5638
  if (cur_quick->init() || cur_quick->reset())
sergefp@mysql.com's avatar
sergefp@mysql.com committed
5639
    DBUG_RETURN(1);
5640

5641
  unique= new Unique(refpos_order_cmp, (void *)head->file,
5642
                     head->file->ref_length,
5643
                     thd->variables.sortbuff_size);
5644 5645
  if (!unique)
    DBUG_RETURN(1);
monty@mysql.com's avatar
monty@mysql.com committed
5646
  for (;;)
5647
  {
sergefp@mysql.com's avatar
sergefp@mysql.com committed
5648
    while ((result= cur_quick->get_next()) == HA_ERR_END_OF_FILE)
5649
    {
sergefp@mysql.com's avatar
sergefp@mysql.com committed
5650 5651 5652
      cur_quick->range_end();
      cur_quick= cur_quick_it++;
      if (!cur_quick)
5653
        break;
5654

sergefp@mysql.com's avatar
sergefp@mysql.com committed
5655 5656
      if (cur_quick->file->inited != handler::NONE) 
        cur_quick->file->ha_index_end();
sergefp@mysql.com's avatar
sergefp@mysql.com committed
5657
      if (cur_quick->init() || cur_quick->reset())
5658
        DBUG_RETURN(1);
5659 5660 5661
    }

    if (result)
5662
    {
5663
      if (result != HA_ERR_END_OF_FILE)
sergefp@mysql.com's avatar
sergefp@mysql.com committed
5664 5665
      {
        cur_quick->range_end();
5666
        DBUG_RETURN(result);
sergefp@mysql.com's avatar
sergefp@mysql.com committed
5667
      }
5668
      break;
5669
    }
5670

5671 5672
    if (thd->killed)
      DBUG_RETURN(1);
5673

5674
    /* skip row if it will be retrieved by clustered PK scan */
5675 5676
    if (pk_quick_select && pk_quick_select->row_in_ranges())
      continue;
5677

sergefp@mysql.com's avatar
sergefp@mysql.com committed
5678 5679
    cur_quick->file->position(cur_quick->record);
    result= unique->unique_add((char*)cur_quick->file->ref);
5680
    if (result)
5681 5682
      DBUG_RETURN(1);

monty@mysql.com's avatar
monty@mysql.com committed
5683
  }
5684

5685 5686
  /* ok, all row ids are in Unique */
  result= unique->get(head);
sergefp@mysql.com's avatar
sergefp@mysql.com committed
5687
  delete unique;
monty@mysql.com's avatar
monty@mysql.com committed
5688
  doing_pk_scan= FALSE;
monty@mysql.com's avatar
monty@mysql.com committed
5689 5690
  /* start table scan */
  init_read_record(&read_record, thd, head, (SQL_SELECT*) 0, 1, 1);
5691 5692
  /* index_merge currently doesn't support "using index" at all */
  head->file->extra(HA_EXTRA_NO_KEYREAD);
5693

5694 5695 5696
  DBUG_RETURN(result);
}

5697

5698 5699 5700
/*
  Get next row for index_merge.
  NOTES
5701 5702 5703 5704
    The rows are read from
      1. rowids stored in Unique.
      2. QUICK_RANGE_SELECT with clustered primary key (if any).
    The sets of rows retrieved in 1) and 2) are guaranteed to be disjoint.
5705
*/
5706

5707 5708
int QUICK_INDEX_MERGE_SELECT::get_next()
{
5709
  int result;
5710
  DBUG_ENTER("QUICK_INDEX_MERGE_SELECT::get_next");
5711

5712 5713 5714 5715 5716 5717 5718 5719 5720
  if (doing_pk_scan)
    DBUG_RETURN(pk_quick_select->get_next());

  result= read_record.read_record(&read_record);

  if (result == -1)
  {
    result= HA_ERR_END_OF_FILE;
    end_read_record(&read_record);
5721
    /* All rows from Unique have been retrieved, do a clustered PK scan */
monty@mysql.com's avatar
monty@mysql.com committed
5722
    if (pk_quick_select)
5723
    {
monty@mysql.com's avatar
monty@mysql.com committed
5724
      doing_pk_scan= TRUE;
sergefp@mysql.com's avatar
sergefp@mysql.com committed
5725
      if ((result= pk_quick_select->init()) || (result= pk_quick_select->reset()))
5726 5727 5728 5729 5730 5731
        DBUG_RETURN(result);
      DBUG_RETURN(pk_quick_select->get_next());
    }
  }

  DBUG_RETURN(result);
5732 5733
}

5734 5735

/*
5736
  Retrieve next record.
5737
  SYNOPSIS
5738 5739
     QUICK_ROR_INTERSECT_SELECT::get_next()

5740
  NOTES
5741 5742
    Invariant on enter/exit: all intersected selects have retrieved all index
    records with rowid <= some_rowid_val and no intersected select has
5743 5744 5745 5746
    retrieved any index records with rowid > some_rowid_val.
    We start fresh and loop until we have retrieved the same rowid in each of
    the key scans or we got an error.

5747
    If a Clustered PK scan is present, it is used only to check if row
5748 5749 5750 5751 5752
    satisfies its condition (and never used for row retrieval).

  RETURN
   0     - Ok
   other - Error code if any error occurred.
5753 5754 5755 5756 5757 5758 5759 5760 5761
*/

int QUICK_ROR_INTERSECT_SELECT::get_next()
{
  List_iterator_fast<QUICK_RANGE_SELECT> quick_it(quick_selects);
  QUICK_RANGE_SELECT* quick;
  int error, cmp;
  uint last_rowid_count=0;
  DBUG_ENTER("QUICK_ROR_INTERSECT_SELECT::get_next");
5762

5763 5764 5765 5766 5767 5768 5769 5770 5771 5772
  /* Get a rowid for first quick and save it as a 'candidate' */
  quick= quick_it++;
  if (cpk_quick)
  {
    do {
      error= quick->get_next();
    }while (!error && !cpk_quick->row_in_ranges());
  }
  else
    error= quick->get_next();
5773

5774 5775 5776 5777 5778 5779
  if (error)
    DBUG_RETURN(error);

  quick->file->position(quick->record);
  memcpy(last_rowid, quick->file->ref, head->file->ref_length);
  last_rowid_count= 1;
5780

5781 5782 5783 5784 5785 5786 5787
  while (last_rowid_count < quick_selects.elements)
  {
    if (!(quick= quick_it++))
    {
      quick_it.rewind();
      quick= quick_it++;
    }
5788

5789 5790 5791 5792
    do {
      if ((error= quick->get_next()))
        DBUG_RETURN(error);
      quick->file->position(quick->record);
5793
      cmp= head->file->cmp_ref(quick->file->ref, last_rowid);
5794 5795 5796 5797 5798 5799 5800 5801 5802 5803 5804 5805 5806 5807 5808
    } while (cmp < 0);

    /* Ok, current select 'caught up' and returned ref >= cur_ref */
    if (cmp > 0)
    {
      /* Found a row with ref > cur_ref. Make it a new 'candidate' */
      if (cpk_quick)
      {
        while (!cpk_quick->row_in_ranges())
        {
          if ((error= quick->get_next()))
            DBUG_RETURN(error);
        }
      }
      memcpy(last_rowid, quick->file->ref, head->file->ref_length);
5809
      last_rowid_count= 1;
5810 5811 5812 5813 5814 5815 5816 5817 5818 5819 5820 5821 5822 5823 5824
    }
    else
    {
      /* current 'candidate' row confirmed by this select */
      last_rowid_count++;
    }
  }

  /* We get here iff we got the same row ref in all scans. */
  if (need_to_fetch_row)
    error= head->file->rnd_pos(head->record[0], last_rowid);
  DBUG_RETURN(error);
}


5825 5826
/*
  Retrieve next record.
5827 5828
  SYNOPSIS
    QUICK_ROR_UNION_SELECT::get_next()
5829

5830
  NOTES
5831 5832
    Enter/exit invariant:
    For each quick select in the queue a {key,rowid} tuple has been
5833
    retrieved but the corresponding row hasn't been passed to output.
5834

5835
  RETURN
5836 5837
   0     - Ok
   other - Error code if any error occurred.
5838 5839 5840 5841 5842 5843 5844 5845
*/

int QUICK_ROR_UNION_SELECT::get_next()
{
  int error, dup_row;
  QUICK_SELECT_I *quick;
  byte *tmp;
  DBUG_ENTER("QUICK_ROR_UNION_SELECT::get_next");
5846

5847 5848 5849 5850
  do
  {
    if (!queue.elements)
      DBUG_RETURN(HA_ERR_END_OF_FILE);
5851
    /* Ok, we have a queue with >= 1 scans */
5852 5853 5854 5855 5856 5857 5858 5859 5860 5861 5862 5863 5864 5865 5866 5867

    quick= (QUICK_SELECT_I*)queue_top(&queue);
    memcpy(cur_rowid, quick->last_rowid, rowid_length);

    /* put into queue rowid from the same stream as top element */
    if ((error= quick->get_next()))
    {
      if (error != HA_ERR_END_OF_FILE)
        DBUG_RETURN(error);
      queue_remove(&queue, 0);
    }
    else
    {
      quick->save_last_pos();
      queue_replaced(&queue);
    }
5868

5869 5870 5871
    if (!have_prev_rowid)
    {
      /* No rows have been returned yet */
monty@mysql.com's avatar
monty@mysql.com committed
5872 5873
      dup_row= FALSE;
      have_prev_rowid= TRUE;
5874 5875 5876 5877
    }
    else
      dup_row= !head->file->cmp_ref(cur_rowid, prev_rowid);
  }while (dup_row);
5878

5879 5880 5881 5882 5883 5884 5885 5886
  tmp= cur_rowid;
  cur_rowid= prev_rowid;
  prev_rowid= tmp;

  error= head->file->rnd_pos(quick->record, prev_rowid);
  DBUG_RETURN(error);
}

sergefp@mysql.com's avatar
sergefp@mysql.com committed
5887
int QUICK_RANGE_SELECT::reset()
ingo@mysql.com's avatar
ingo@mysql.com committed
5888 5889 5890
{
  uint  mrange_bufsiz;
  byte  *mrange_buff;
sergefp@mysql.com's avatar
sergefp@mysql.com committed
5891 5892 5893 5894 5895
  DBUG_ENTER("QUICK_RANGE_SELECT::reset");
  next=0;
  range= NULL;
  cur_range= (QUICK_RANGE**) ranges.buffer;
  
ingo@mysql.com's avatar
ingo@mysql.com committed
5896 5897 5898 5899 5900 5901 5902
  /* Do not allocate the buffers twice. */
  if (multi_range_length)
  {
    DBUG_ASSERT(multi_range_length == min(multi_range_count, ranges.elements));
    DBUG_RETURN(0);
  }

sergefp@mysql.com's avatar
sergefp@mysql.com committed
5903 5904
  /* Allocate the ranges array. */
  DBUG_ASSERT(ranges.elements);
ingo@mysql.com's avatar
ingo@mysql.com committed
5905 5906 5907 5908 5909 5910 5911 5912 5913 5914 5915 5916 5917 5918 5919 5920
  multi_range_length= min(multi_range_count, ranges.elements);
  DBUG_ASSERT(multi_range_length > 0);
  while (multi_range_length && ! (multi_range= (KEY_MULTI_RANGE*)
                                  my_malloc(multi_range_length *
                                            sizeof(KEY_MULTI_RANGE),
                                            MYF(MY_WME))))
  {
    /* Try to shrink the buffers until it is 0. */
    multi_range_length/= 2;
  }
  if (! multi_range)
  {
    multi_range_length= 0;
    DBUG_RETURN(HA_ERR_OUT_OF_MEM);
  }

sergefp@mysql.com's avatar
sergefp@mysql.com committed
5921
  /* Allocate the handler buffer if necessary.  */
ingo@mysql.com's avatar
ingo@mysql.com committed
5922 5923 5924
  if (file->table_flags() & HA_NEED_READ_RANGE_BUFFER)
  {
    mrange_bufsiz= min(multi_range_bufsiz,
joreland@mysql.com's avatar
merge  
joreland@mysql.com committed
5925
                       (QUICK_SELECT_I::records + 1)* head->s->reclength);
ingo@mysql.com's avatar
ingo@mysql.com committed
5926 5927 5928 5929 5930 5931 5932 5933 5934 5935 5936 5937 5938 5939 5940 5941 5942 5943 5944 5945 5946 5947 5948 5949 5950 5951 5952 5953 5954 5955 5956 5957 5958 5959 5960 5961 5962 5963 5964 5965 5966

    while (mrange_bufsiz &&
           ! my_multi_malloc(MYF(MY_WME),
                             &multi_range_buff, sizeof(*multi_range_buff),
                             &mrange_buff, mrange_bufsiz,
                             NullS))
    {
      /* Try to shrink the buffers until both are 0. */
      mrange_bufsiz/= 2;
    }
    if (! multi_range_buff)
    {
      my_free((char*) multi_range, MYF(0));
      multi_range= NULL;
      multi_range_length= 0;
      DBUG_RETURN(HA_ERR_OUT_OF_MEM);
    }

    /* Initialize the handler buffer. */
    multi_range_buff->buffer= mrange_buff;
    multi_range_buff->buffer_end= mrange_buff + mrange_bufsiz;
    multi_range_buff->end_of_used_area= mrange_buff;
  }
  DBUG_RETURN(0);
}


/*
  Get next possible record using quick-struct.

  SYNOPSIS
    QUICK_RANGE_SELECT::get_next()

  NOTES
    Record is read into table->record[0]

  RETURN
    0			Found row
    HA_ERR_END_OF_FILE	No (more) rows in range
    #			Error code
*/
bk@work.mysql.com's avatar
bk@work.mysql.com committed
5967

5968
int QUICK_RANGE_SELECT::get_next()
bk@work.mysql.com's avatar
bk@work.mysql.com committed
5969
{
ingo@mysql.com's avatar
ingo@mysql.com committed
5970 5971 5972 5973
  int             result;
  KEY_MULTI_RANGE *mrange;
  key_range       *start_key;
  key_range       *end_key;
5974
  DBUG_ENTER("QUICK_RANGE_SELECT::get_next");
ingo@mysql.com's avatar
ingo@mysql.com committed
5975 5976 5977
  DBUG_ASSERT(multi_range_length && multi_range &&
              (cur_range >= (QUICK_RANGE**) ranges.buffer) &&
              (cur_range <= (QUICK_RANGE**) ranges.buffer + ranges.elements));
bk@work.mysql.com's avatar
bk@work.mysql.com committed
5978 5979 5980

  for (;;)
  {
ingo@mysql.com's avatar
ingo@mysql.com committed
5981
    if (in_range)
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
5982
    {
ingo@mysql.com's avatar
ingo@mysql.com committed
5983 5984
      /* We did already start to read this key. */
      result= file->read_multi_range_next(&mrange);
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
5985
      if (result != HA_ERR_END_OF_FILE)
ingo@mysql.com's avatar
ingo@mysql.com committed
5986 5987
      {
        in_range= ! result;
5988
	DBUG_RETURN(result);
ingo@mysql.com's avatar
ingo@mysql.com committed
5989
      }
bk@work.mysql.com's avatar
bk@work.mysql.com committed
5990
    }
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
5991

ingo@mysql.com's avatar
ingo@mysql.com committed
5992 5993 5994 5995 5996 5997 5998 5999 6000 6001 6002 6003 6004 6005 6006 6007 6008 6009 6010 6011 6012 6013 6014 6015 6016 6017 6018 6019 6020 6021
    uint count= min(multi_range_length, ranges.elements -
                    (cur_range - (QUICK_RANGE**) ranges.buffer));
    if (count == 0)
    {
      /* Ranges have already been used up before. None is left for read. */
      in_range= FALSE;
      DBUG_RETURN(HA_ERR_END_OF_FILE);
    }
    KEY_MULTI_RANGE *mrange_slot, *mrange_end;
    for (mrange_slot= multi_range, mrange_end= mrange_slot+count;
         mrange_slot < mrange_end;
         mrange_slot++)
    {
      start_key= &mrange_slot->start_key;
      end_key= &mrange_slot->end_key;
      range= *(cur_range++);

      start_key->key=    (const byte*) range->min_key;
      start_key->length= range->min_length;
      start_key->flag=   ((range->flag & NEAR_MIN) ? HA_READ_AFTER_KEY :
                          (range->flag & EQ_RANGE) ?
                          HA_READ_KEY_EXACT : HA_READ_KEY_OR_NEXT);
      end_key->key=      (const byte*) range->max_key;
      end_key->length=   range->max_length;
      /*
        We use HA_READ_AFTER_KEY here because if we are reading on a key
        prefix. We want to find all keys with this prefix.
      */
      end_key->flag=     (range->flag & NEAR_MAX ? HA_READ_BEFORE_KEY :
                          HA_READ_AFTER_KEY);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
6022

ingo@mysql.com's avatar
ingo@mysql.com committed
6023 6024
      mrange_slot->range_flag= range->flag;
    }
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
6025

ingo@mysql.com's avatar
ingo@mysql.com committed
6026 6027
    result= file->read_multi_range_first(&mrange, multi_range, count,
                                         sorted, multi_range_buff);
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
6028
    if (result != HA_ERR_END_OF_FILE)
ingo@mysql.com's avatar
ingo@mysql.com committed
6029 6030
    {
      in_range= ! result;
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
6031
      DBUG_RETURN(result);
ingo@mysql.com's avatar
ingo@mysql.com committed
6032 6033
    }
    in_range= FALSE; /* No matching rows; go to next set of ranges. */
bk@work.mysql.com's avatar
bk@work.mysql.com committed
6034 6035 6036
  }
}

6037

6038 6039 6040 6041 6042 6043
/*
  Get the next record with a different prefix.

  SYNOPSIS
    QUICK_RANGE_SELECT::get_next_prefix()
    prefix_length  length of cur_prefix
6044
    cur_prefix     prefix of a key to be searched for
6045 6046 6047 6048 6049 6050 6051 6052 6053 6054 6055 6056 6057 6058 6059 6060 6061 6062 6063 6064 6065 6066 6067 6068 6069 6070 6071 6072 6073 6074 6075 6076

  DESCRIPTION
    Each subsequent call to the method retrieves the first record that has a
    prefix with length prefix_length different from cur_prefix, such that the
    record with the new prefix is within the ranges described by
    this->ranges. The record found is stored into the buffer pointed by
    this->record.
    The method is useful for GROUP-BY queries with range conditions to
    discover the prefix of the next group that satisfies the range conditions.

  TODO
    This method is a modified copy of QUICK_RANGE_SELECT::get_next(), so both
    methods should be unified into a more general one to reduce code
    duplication.

  RETURN
    0                  on success
    HA_ERR_END_OF_FILE if returned all keys
    other              if some error occurred
*/

int QUICK_RANGE_SELECT::get_next_prefix(uint prefix_length, byte *cur_prefix)
{
  DBUG_ENTER("QUICK_RANGE_SELECT::get_next_prefix");

  for (;;)
  {
    int result;
    key_range start_key, end_key;
    if (range)
    {
      /* Read the next record in the same range with prefix after cur_prefix. */
6077
      DBUG_ASSERT(cur_prefix != 0);
6078 6079 6080 6081 6082 6083
      result= file->index_read(record, cur_prefix, prefix_length,
                               HA_READ_AFTER_KEY);
      if (result || (file->compare_key(file->end_range) <= 0))
        DBUG_RETURN(result);
    }

ingo@mysql.com's avatar
ingo@mysql.com committed
6084 6085 6086 6087 6088 6089 6090 6091
    uint count= ranges.elements - (cur_range - (QUICK_RANGE**) ranges.buffer);
    if (count == 0)
    {
      /* Ranges have already been used up before. None is left for read. */
      range= 0;
      DBUG_RETURN(HA_ERR_END_OF_FILE);
    }
    range= *(cur_range++);
6092 6093 6094 6095 6096 6097 6098 6099 6100 6101 6102 6103 6104 6105 6106 6107 6108 6109 6110 6111 6112 6113 6114 6115 6116 6117 6118 6119 6120

    start_key.key=    (const byte*) range->min_key;
    start_key.length= min(range->min_length, prefix_length);
    start_key.flag=   ((range->flag & NEAR_MIN) ? HA_READ_AFTER_KEY :
		       (range->flag & EQ_RANGE) ?
		       HA_READ_KEY_EXACT : HA_READ_KEY_OR_NEXT);
    end_key.key=      (const byte*) range->max_key;
    end_key.length=   min(range->max_length, prefix_length);
    /*
      We use READ_AFTER_KEY here because if we are reading on a key
      prefix we want to find all keys with this prefix
    */
    end_key.flag=     (range->flag & NEAR_MAX ? HA_READ_BEFORE_KEY :
		       HA_READ_AFTER_KEY);

    result= file->read_range_first(range->min_length ? &start_key : 0,
				   range->max_length ? &end_key : 0,
                                   test(range->flag & EQ_RANGE),
				   sorted);
    if (range->flag == (UNIQUE_RANGE | EQ_RANGE))
      range=0;				// Stop searching

    if (result != HA_ERR_END_OF_FILE)
      DBUG_RETURN(result);
    range=0;				// No matching rows; go to next range
  }
}


pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
6121
/* Get next for geometrical indexes */
bk@work.mysql.com's avatar
bk@work.mysql.com committed
6122

pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
6123
int QUICK_RANGE_SELECT_GEOM::get_next()
bk@work.mysql.com's avatar
bk@work.mysql.com committed
6124
{
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
6125
  DBUG_ENTER("QUICK_RANGE_SELECT_GEOM::get_next");
bk@work.mysql.com's avatar
bk@work.mysql.com committed
6126

pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
6127
  for (;;)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
6128
  {
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
6129 6130
    int result;
    if (range)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
6131
    {
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
6132 6133 6134 6135 6136
      // Already read through key
      result= file->index_next_same(record, (byte*) range->min_key,
				    range->min_length);
      if (result != HA_ERR_END_OF_FILE)
	DBUG_RETURN(result);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
6137
    }
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
6138

ingo@mysql.com's avatar
ingo@mysql.com committed
6139 6140 6141 6142 6143 6144 6145 6146
    uint count= ranges.elements - (cur_range - (QUICK_RANGE**) ranges.buffer);
    if (count == 0)
    {
      /* Ranges have already been used up before. None is left for read. */
      range= 0;
      DBUG_RETURN(HA_ERR_END_OF_FILE);
    }
    range= *(cur_range++);
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
6147 6148 6149 6150 6151 6152 6153 6154

    result= file->index_read(record,
			     (byte*) range->min_key,
			     range->min_length,
			     (ha_rkey_function)(range->flag ^ GEOM_FLAG));
    if (result != HA_ERR_KEY_NOT_FOUND)
      DBUG_RETURN(result);
    range=0;				// Not found, to next range
bk@work.mysql.com's avatar
bk@work.mysql.com committed
6155 6156 6157
  }
}

6158

6159 6160 6161 6162
/*
  Check if current row will be retrieved by this QUICK_RANGE_SELECT

  NOTES
6163 6164
    It is assumed that currently a scan is being done on another index
    which reads all necessary parts of the index that is scanned by this
6165
    quick select.
6166
    The implementation does a binary search on sorted array of disjoint
6167 6168
    ranges, without taking size of range into account.

6169
    This function is used to filter out clustered PK scan rows in
6170 6171
    index_merge quick select.

6172
  RETURN
monty@mysql.com's avatar
monty@mysql.com committed
6173 6174
    TRUE  if current row will be retrieved by this quick select
    FALSE if not
6175 6176 6177 6178 6179 6180 6181 6182 6183 6184
*/

bool QUICK_RANGE_SELECT::row_in_ranges()
{
  QUICK_RANGE *range;
  uint min= 0;
  uint max= ranges.elements - 1;
  uint mid= (max + min)/2;

  while (min != max)
6185
  {
6186 6187 6188 6189 6190 6191 6192 6193 6194 6195 6196 6197 6198
    if (cmp_next(*(QUICK_RANGE**)dynamic_array_ptr(&ranges, mid)))
    {
      /* current row value > mid->max */
      min= mid + 1;
    }
    else
      max= mid;
    mid= (min + max) / 2;
  }
  range= *(QUICK_RANGE**)dynamic_array_ptr(&ranges, mid);
  return (!cmp_next(range) && !cmp_prev(range));
}

6199
/*
6200 6201 6202 6203 6204 6205 6206
  This is a hack: we inherit from QUICK_SELECT so that we can use the
  get_next() interface, but we have to hold a pointer to the original
  QUICK_SELECT because its data are used all over the place.  What
  should be done is to factor out the data that is needed into a base
  class (QUICK_SELECT), and then have two subclasses (_ASC and _DESC)
  which handle the ranges and implement the get_next() function.  But
  for now, this seems to work right at least.
6207
 */
6208

6209
QUICK_SELECT_DESC::QUICK_SELECT_DESC(QUICK_RANGE_SELECT *q,
6210 6211
                                     uint used_key_parts)
 : QUICK_RANGE_SELECT(*q), rev_it(rev_ranges)
6212
{
monty@hundin.mysql.fi's avatar
monty@hundin.mysql.fi committed
6213
  QUICK_RANGE *r;
6214

6215 6216
  QUICK_RANGE **pr= (QUICK_RANGE**)ranges.buffer;
  QUICK_RANGE **last_range= pr + ranges.elements;
monty@mysql.com's avatar
monty@mysql.com committed
6217 6218
  for (; pr!=last_range; pr++)
    rev_ranges.push_front(*pr);
monty@hundin.mysql.fi's avatar
monty@hundin.mysql.fi committed
6219

6220
  /* Remove EQ_RANGE flag for keys that are not using the full key */
monty@hundin.mysql.fi's avatar
monty@hundin.mysql.fi committed
6221
  for (r = rev_it++; r; r = rev_it++)
6222 6223 6224 6225 6226 6227 6228 6229
  {
    if ((r->flag & EQ_RANGE) &&
	head->key_info[index].key_length != r->max_length)
      r->flag&= ~EQ_RANGE;
  }
  rev_it.rewind();
  q->dont_free=1;				// Don't free shared mem
  delete q;
6230 6231
}

6232

6233 6234 6235 6236 6237 6238
int QUICK_SELECT_DESC::get_next()
{
  DBUG_ENTER("QUICK_SELECT_DESC::get_next");

  /* The max key is handled as follows:
   *   - if there is NO_MAX_RANGE, start at the end and move backwards
6239 6240
   *   - if it is an EQ_RANGE, which means that max key covers the entire
   *     key, go directly to the key and read through it (sorting backwards is
6241 6242 6243 6244 6245 6246 6247 6248 6249 6250 6251 6252
   *     same as sorting forwards)
   *   - if it is NEAR_MAX, go to the key or next, step back once, and
   *     move backwards
   *   - otherwise (not NEAR_MAX == include the key), go after the key,
   *     step back once, and move backwards
   */

  for (;;)
  {
    int result;
    if (range)
    {						// Already read through key
6253 6254 6255
      result = ((range->flag & EQ_RANGE)
		? file->index_next_same(record, (byte*) range->min_key,
					range->min_length) :
6256 6257 6258 6259 6260 6261 6262 6263 6264 6265 6266 6267 6268 6269 6270
		file->index_prev(record));
      if (!result)
      {
	if (cmp_prev(*rev_it.ref()) == 0)
	  DBUG_RETURN(0);
      }
      else if (result != HA_ERR_END_OF_FILE)
	DBUG_RETURN(result);
    }

    if (!(range=rev_it++))
      DBUG_RETURN(HA_ERR_END_OF_FILE);		// All ranges used

    if (range->flag & NO_MAX_RANGE)		// Read last record
    {
6271 6272 6273
      int local_error;
      if ((local_error=file->index_last(record)))
	DBUG_RETURN(local_error);		// Empty table
6274 6275 6276 6277 6278 6279
      if (cmp_prev(range) == 0)
	DBUG_RETURN(0);
      range=0;			// No matching records; go to next range
      continue;
    }

6280
    if (range->flag & EQ_RANGE)
6281 6282 6283 6284 6285 6286
    {
      result = file->index_read(record, (byte*) range->max_key,
				range->max_length, HA_READ_KEY_EXACT);
    }
    else
    {
6287 6288 6289 6290 6291
      DBUG_ASSERT(range->flag & NEAR_MAX || range_reads_after_key(range));
      result=file->index_read(record, (byte*) range->max_key,
			      range->max_length,
			      ((range->flag & NEAR_MAX) ?
			       HA_READ_BEFORE_KEY : HA_READ_PREFIX_LAST_OR_PREV));
6292 6293 6294 6295 6296 6297 6298 6299 6300 6301 6302 6303 6304 6305 6306 6307 6308 6309
    }
    if (result)
    {
      if (result != HA_ERR_KEY_NOT_FOUND)
	DBUG_RETURN(result);
      range=0;					// Not found, to next range
      continue;
    }
    if (cmp_prev(range) == 0)
    {
      if (range->flag == (UNIQUE_RANGE | EQ_RANGE))
	range = 0;				// Stop searching
      DBUG_RETURN(0);				// Found key is in range
    }
    range = 0;					// To next range
  }
}

6310

pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
6311 6312 6313 6314 6315 6316 6317 6318 6319 6320 6321 6322 6323 6324 6325 6326 6327 6328 6329 6330 6331 6332 6333 6334 6335 6336 6337 6338 6339 6340 6341 6342 6343 6344 6345 6346 6347 6348 6349 6350 6351
/*
  Compare if found key is over max-value
  Returns 0 if key <= range->max_key
*/

int QUICK_RANGE_SELECT::cmp_next(QUICK_RANGE *range_arg)
{
  if (range_arg->flag & NO_MAX_RANGE)
    return 0;                                   /* key can't be to large */

  KEY_PART *key_part=key_parts;
  uint store_length;

  for (char *key=range_arg->max_key, *end=key+range_arg->max_length;
       key < end;
       key+= store_length, key_part++)
  {
    int cmp;
    store_length= key_part->store_length;
    if (key_part->null_bit)
    {
      if (*key)
      {
        if (!key_part->field->is_null())
          return 1;
        continue;
      }
      else if (key_part->field->is_null())
        return 0;
      key++;					// Skip null byte
      store_length--;
    }
    if ((cmp=key_part->field->key_cmp((byte*) key, key_part->length)) < 0)
      return 0;
    if (cmp > 0)
      return 1;
  }
  return (range_arg->flag & NEAR_MAX) ? 1 : 0;          // Exact match
}


6352
/*
6353 6354 6355
  Returns 0 if found key is inside range (found key >= range->min_key).
*/

6356
int QUICK_RANGE_SELECT::cmp_prev(QUICK_RANGE *range_arg)
6357
{
6358
  int cmp;
6359
  if (range_arg->flag & NO_MIN_RANGE)
6360
    return 0;					/* key can't be to small */
6361

monty@mysql.com's avatar
monty@mysql.com committed
6362 6363
  cmp= key_cmp(key_part_info, (byte*) range_arg->min_key,
               range_arg->min_length);
6364 6365 6366
  if (cmp > 0 || cmp == 0 && !(range_arg->flag & NEAR_MIN))
    return 0;
  return 1;                                     // outside of range
6367 6368
}

6369

6370
/*
monty@mysql.com's avatar
monty@mysql.com committed
6371
 * TRUE if this range will require using HA_READ_AFTER_KEY
6372
   See comment in get_next() about this
6373
 */
6374

6375
bool QUICK_SELECT_DESC::range_reads_after_key(QUICK_RANGE *range_arg)
6376
{
6377
  return ((range_arg->flag & (NO_MAX_RANGE | NEAR_MAX)) ||
6378
	  !(range_arg->flag & EQ_RANGE) ||
6379
	  head->key_info[index].key_length != range_arg->max_length) ? 1 : 0;
6380 6381
}

6382

monty@mysql.com's avatar
monty@mysql.com committed
6383
/* TRUE if we are reading over a key that may have a NULL value */
6384

6385
#ifdef NOT_USED
6386
bool QUICK_SELECT_DESC::test_if_null_range(QUICK_RANGE *range_arg,
6387 6388
					   uint used_key_parts)
{
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
6389
  uint offset, end;
6390 6391 6392
  KEY_PART *key_part = key_parts,
           *key_part_end= key_part+used_key_parts;

6393
  for (offset= 0,  end = min(range_arg->min_length, range_arg->max_length) ;
6394
       offset < end && key_part != key_part_end ;
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
6395
       offset+= key_part++->store_length)
6396
  {
6397 6398
    if (!memcmp((char*) range_arg->min_key+offset,
		(char*) range_arg->max_key+offset,
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
6399
		key_part->store_length))
6400
      continue;
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
6401 6402

    if (key_part->null_bit && range_arg->min_key[offset])
6403 6404 6405 6406 6407 6408 6409 6410 6411 6412 6413 6414
      return 1;				// min_key is null and max_key isn't
    // Range doesn't cover NULL. This is ok if there is no more null parts
    break;
  }
  /*
    If the next min_range is > NULL, then we can use this, even if
    it's a NULL key
    Example:  SELECT * FROM t1 WHERE a = 2 AND b >0 ORDER BY a DESC,b DESC;

  */
  if (key_part != key_part_end && key_part->null_bit)
  {
6415
    if (offset >= range_arg->min_length || range_arg->min_key[offset])
6416 6417 6418 6419 6420 6421 6422 6423 6424 6425 6426 6427
      return 1;					// Could be null
    key_part++;
  }
  /*
    If any of the key parts used in the ORDER BY could be NULL, we can't
    use the key to sort the data.
  */
  for (; key_part != key_part_end ; key_part++)
    if (key_part->null_bit)
      return 1;					// Covers null part
  return 0;
}
6428
#endif
6429 6430


6431 6432 6433 6434 6435 6436 6437 6438 6439
void QUICK_RANGE_SELECT::add_info_string(String *str)
{
  KEY *key_info= head->key_info + index;
  str->append(key_info->name);
}

void QUICK_INDEX_MERGE_SELECT::add_info_string(String *str)
{
  QUICK_RANGE_SELECT *quick;
monty@mysql.com's avatar
monty@mysql.com committed
6440
  bool first= TRUE;
6441 6442 6443 6444 6445 6446 6447
  List_iterator_fast<QUICK_RANGE_SELECT> it(quick_selects);
  str->append("sort_union(");
  while ((quick= it++))
  {
    if (!first)
      str->append(',');
    else
monty@mysql.com's avatar
monty@mysql.com committed
6448
      first= FALSE;
6449 6450 6451 6452 6453 6454 6455 6456 6457 6458 6459 6460
    quick->add_info_string(str);
  }
  if (pk_quick_select)
  {
    str->append(',');
    pk_quick_select->add_info_string(str);
  }
  str->append(')');
}

void QUICK_ROR_INTERSECT_SELECT::add_info_string(String *str)
{
6461
  bool first= TRUE;
6462 6463 6464 6465 6466 6467 6468 6469
  QUICK_RANGE_SELECT *quick;
  List_iterator_fast<QUICK_RANGE_SELECT> it(quick_selects);
  str->append("intersect(");
  while ((quick= it++))
  {
    KEY *key_info= head->key_info + quick->index;
    if (!first)
      str->append(',');
6470
    else
monty@mysql.com's avatar
monty@mysql.com committed
6471
      first= FALSE;
6472 6473 6474 6475 6476 6477 6478 6479 6480 6481 6482 6483 6484
    str->append(key_info->name);
  }
  if (cpk_quick)
  {
    KEY *key_info= head->key_info + cpk_quick->index;
    str->append(',');
    str->append(key_info->name);
  }
  str->append(')');
}

void QUICK_ROR_UNION_SELECT::add_info_string(String *str)
{
6485
  bool first= TRUE;
6486 6487 6488 6489 6490 6491 6492 6493
  QUICK_SELECT_I *quick;
  List_iterator_fast<QUICK_SELECT_I> it(quick_selects);
  str->append("union(");
  while ((quick= it++))
  {
    if (!first)
      str->append(',');
    else
monty@mysql.com's avatar
monty@mysql.com committed
6494
      first= FALSE;
6495 6496 6497 6498 6499 6500
    quick->add_info_string(str);
  }
  str->append(')');
}


6501
void QUICK_RANGE_SELECT::add_keys_and_lengths(String *key_names,
6502
                                              String *used_lengths)
6503 6504 6505 6506 6507 6508 6509 6510 6511
{
  char buf[64];
  uint length;
  KEY *key_info= head->key_info + index;
  key_names->append(key_info->name);
  length= longlong2str(max_used_key_length, buf, 10) - buf;
  used_lengths->append(buf, length);
}

6512 6513
void QUICK_INDEX_MERGE_SELECT::add_keys_and_lengths(String *key_names,
                                                    String *used_lengths)
6514 6515 6516
{
  char buf[64];
  uint length;
monty@mysql.com's avatar
monty@mysql.com committed
6517
  bool first= TRUE;
6518
  QUICK_RANGE_SELECT *quick;
6519

6520 6521 6522
  List_iterator_fast<QUICK_RANGE_SELECT> it(quick_selects);
  while ((quick= it++))
  {
6523
    if (first)
monty@mysql.com's avatar
monty@mysql.com committed
6524
      first= FALSE;
6525 6526
    else
    {
6527 6528
      key_names->append(',');
      used_lengths->append(',');
6529
    }
6530

6531 6532
    KEY *key_info= head->key_info + quick->index;
    key_names->append(key_info->name);
6533 6534 6535 6536 6537 6538 6539 6540 6541 6542 6543 6544 6545 6546
    length= longlong2str(quick->max_used_key_length, buf, 10) - buf;
    used_lengths->append(buf, length);
  }
  if (pk_quick_select)
  {
    KEY *key_info= head->key_info + pk_quick_select->index;
    key_names->append(',');
    key_names->append(key_info->name);
    length= longlong2str(pk_quick_select->max_used_key_length, buf, 10) - buf;
    used_lengths->append(',');
    used_lengths->append(buf, length);
  }
}

6547 6548
void QUICK_ROR_INTERSECT_SELECT::add_keys_and_lengths(String *key_names,
                                                      String *used_lengths)
6549 6550 6551
{
  char buf[64];
  uint length;
6552
  bool first= TRUE;
6553 6554 6555 6556 6557 6558
  QUICK_RANGE_SELECT *quick;
  List_iterator_fast<QUICK_RANGE_SELECT> it(quick_selects);
  while ((quick= it++))
  {
    KEY *key_info= head->key_info + quick->index;
    if (first)
monty@mysql.com's avatar
monty@mysql.com committed
6559
      first= FALSE;
6560
    else
6561 6562
    {
      key_names->append(',');
6563
      used_lengths->append(',');
6564 6565
    }
    key_names->append(key_info->name);
6566 6567 6568
    length= longlong2str(quick->max_used_key_length, buf, 10) - buf;
    used_lengths->append(buf, length);
  }
6569

6570 6571 6572 6573 6574 6575 6576 6577 6578 6579 6580
  if (cpk_quick)
  {
    KEY *key_info= head->key_info + cpk_quick->index;
    key_names->append(',');
    key_names->append(key_info->name);
    length= longlong2str(cpk_quick->max_used_key_length, buf, 10) - buf;
    used_lengths->append(',');
    used_lengths->append(buf, length);
  }
}

6581 6582
void QUICK_ROR_UNION_SELECT::add_keys_and_lengths(String *key_names,
                                                  String *used_lengths)
6583
{
6584
  bool first= TRUE;
6585 6586 6587 6588 6589
  QUICK_SELECT_I *quick;
  List_iterator_fast<QUICK_SELECT_I> it(quick_selects);
  while ((quick= it++))
  {
    if (first)
monty@mysql.com's avatar
monty@mysql.com committed
6590
      first= FALSE;
6591
    else
6592
    {
6593 6594 6595
      used_lengths->append(',');
      key_names->append(',');
    }
6596
    quick->add_keys_and_lengths(key_names, used_lengths);
6597 6598 6599
  }
}

6600 6601 6602 6603 6604 6605 6606 6607 6608

/*******************************************************************************
* Implementation of QUICK_GROUP_MIN_MAX_SELECT
*******************************************************************************/

static inline uint get_field_keypart(KEY *index, Field *field);
static inline SEL_ARG * get_index_range_tree(uint index, SEL_TREE* range_tree,
                                             PARAM *param, uint *param_idx);
static bool
6609
get_constant_key_infix(KEY *index_info, SEL_ARG *index_range_tree,
6610
                       KEY_PART_INFO *first_non_group_part,
6611 6612 6613 6614
                       KEY_PART_INFO *min_max_arg_part,
                       KEY_PART_INFO *last_part, THD *thd,
                       byte *key_infix, uint *key_infix_len,
                       KEY_PART_INFO **first_non_infix_part);
6615
static bool
6616 6617
check_group_min_max_predicates(COND *cond, Item_field *min_max_arg_item,
                               Field::imagetype image_type);
6618

6619 6620 6621 6622 6623 6624
static void
cost_group_min_max(TABLE* table, KEY *index_info, uint used_key_parts,
                   uint group_key_parts, SEL_TREE *range_tree,
                   SEL_ARG *index_tree, ha_rows quick_prefix_records,
                   bool have_min, bool have_max,
                   double *read_cost, ha_rows *records);
6625 6626 6627 6628 6629 6630 6631 6632 6633 6634 6635 6636 6637 6638 6639 6640 6641 6642 6643 6644 6645 6646 6647 6648 6649 6650

/*
  Test if this access method is applicable to a GROUP query with MIN/MAX
  functions, and if so, construct a new TRP object.

  SYNOPSIS
    get_best_group_min_max()
    param    Parameter from test_quick_select
    sel_tree Range tree generated by get_mm_tree

  DESCRIPTION
    Test whether a query can be computed via a QUICK_GROUP_MIN_MAX_SELECT.
    Queries computable via a QUICK_GROUP_MIN_MAX_SELECT must satisfy the
    following conditions:
    A) Table T has at least one compound index I of the form:
       I = <A_1, ...,A_k, [B_1,..., B_m], C, [D_1,...,D_n]>
    B) Query conditions:
    B0. Q is over a single table T.
    B1. The attributes referenced by Q are a subset of the attributes of I.
    B2. All attributes QA in Q can be divided into 3 overlapping groups:
        - SA = {S_1, ..., S_l, [C]} - from the SELECT clause, where C is
          referenced by any number of MIN and/or MAX functions if present.
        - WA = {W_1, ..., W_p} - from the WHERE clause
        - GA = <G_1, ..., G_k> - from the GROUP BY clause (if any)
             = SA              - if Q is a DISTINCT query (based on the
                                 equivalence of DISTINCT and GROUP queries.
monty@mysql.com's avatar
monty@mysql.com committed
6651 6652
        - NGA = QA - (GA union C) = {NG_1, ..., NG_m} - the ones not in
          GROUP BY and not referenced by MIN/MAX functions.
6653 6654 6655 6656 6657 6658 6659 6660 6661 6662 6663
        with the following properties specified below.

    SA1. There is at most one attribute in SA referenced by any number of
         MIN and/or MAX functions which, which if present, is denoted as C.
    SA2. The position of the C attribute in the index is after the last A_k.
    SA3. The attribute C can be referenced in the WHERE clause only in
         predicates of the forms:
         - (C {< | <= | > | >= | =} const)
         - (const {< | <= | > | >= | =} C)
         - (C between const_i and const_j)
         - C IS NULL
6664 6665
         - C IS NOT NULL
         - C != const
6666 6667 6668
    SA4. If Q has a GROUP BY clause, there are no other aggregate functions
         except MIN and MAX. For queries with DISTINCT, aggregate functions
         are allowed.
6669
    SA5. The select list in DISTINCT queries should not contain expressions.
6670 6671 6672 6673 6674 6675 6676 6677 6678 6679 6680 6681 6682 6683 6684 6685 6686 6687 6688 6689 6690 6691 6692 6693 6694 6695 6696 6697 6698
    GA1. If Q has a GROUP BY clause, then GA is a prefix of I. That is, if
         G_i = A_j => i = j.
    GA2. If Q has a DISTINCT clause, then there is a permutation of SA that
         forms a prefix of I. This permutation is used as the GROUP clause
         when the DISTINCT query is converted to a GROUP query.
    GA3. The attributes in GA may participate in arbitrary predicates, divided
         into two groups:
         - RNG(G_1,...,G_q ; where q <= k) is a range condition over the
           attributes of a prefix of GA
         - PA(G_i1,...G_iq) is an arbitrary predicate over an arbitrary subset
           of GA. Since P is applied to only GROUP attributes it filters some
           groups, and thus can be applied after the grouping.
    GA4. There are no expressions among G_i, just direct column references.
    NGA1.If in the index I there is a gap between the last GROUP attribute G_k,
         and the MIN/MAX attribute C, then NGA must consist of exactly the index
         attributes that constitute the gap. As a result there is a permutation
         of NGA that coincides with the gap in the index <B_1, ..., B_m>.
    NGA2.If BA <> {}, then the WHERE clause must contain a conjunction EQ of
         equality conditions for all NG_i of the form (NG_i = const) or
         (const = NG_i), such that each NG_i is referenced in exactly one
         conjunct. Informally, the predicates provide constants to fill the
         gap in the index.
    WA1. There are no other attributes in the WHERE clause except the ones
         referenced in predicates RNG, PA, PC, EQ defined above. Therefore
         WA is subset of (GA union NGA union C) for GA,NGA,C that pass the above
         tests. By transitivity then it also follows that each WA_i participates
         in the index I (if this was already tested for GA, NGA and C).

    C) Overall query form:
6699 6700 6701 6702
       SELECT EXPR([A_1,...,A_k], [B_1,...,B_m], [MIN(C)], [MAX(C)])
         FROM T
        WHERE [RNG(A_1,...,A_p ; where p <= k)]
         [AND EQ(B_1,...,B_m)]
6703 6704
         [AND PC(C)]
         [AND PA(A_i1,...,A_iq)]
6705 6706 6707 6708
       GROUP BY A_1,...,A_k
       [HAVING PH(A_1, ..., B_1,..., C)]
    where EXPR(...) is an arbitrary expression over some or all SELECT fields,
    or:
6709 6710 6711 6712 6713 6714 6715 6716 6717 6718 6719 6720 6721 6722 6723 6724 6725 6726 6727 6728 6729 6730 6731 6732 6733 6734 6735 6736 6737 6738 6739 6740 6741 6742 6743 6744 6745 6746 6747 6748 6749 6750 6751 6752 6753 6754 6755 6756 6757 6758 6759 6760 6761 6762 6763 6764 6765 6766 6767 6768 6769 6770 6771 6772 6773 6774 6775 6776 6777 6778
       SELECT DISTINCT A_i1,...,A_ik
         FROM T
        WHERE [RNG(A_1,...,A_p ; where p <= k)]
         [AND PA(A_i1,...,A_iq)];

  NOTES
    If the current query satisfies the conditions above, and if
    (mem_root! = NULL), then the function constructs and returns a new TRP
    object, that is later used to construct a new QUICK_GROUP_MIN_MAX_SELECT.
    If (mem_root == NULL), then the function only tests whether the current
    query satisfies the conditions above, and, if so, sets
    is_applicable = TRUE.

    Queries with DISTINCT for which index access can be used are transformed
    into equivalent group-by queries of the form:

    SELECT A_1,...,A_k FROM T
     WHERE [RNG(A_1,...,A_p ; where p <= k)]
      [AND PA(A_i1,...,A_iq)]
    GROUP BY A_1,...,A_k;

    The group-by list is a permutation of the select attributes, according
    to their order in the index.

  TODO
  - What happens if the query groups by the MIN/MAX field, and there is no
    other field as in: "select min(a) from t1 group by a" ?
  - We assume that the general correctness of the GROUP-BY query was checked
    before this point. Is this correct, or do we have to check it completely?

  RETURN
    If mem_root != NULL
    - valid TRP_GROUP_MIN_MAX object if this QUICK class can be used for
      the query
    -  NULL o/w.
    If mem_root == NULL
    - NULL
*/

static TRP_GROUP_MIN_MAX *
get_best_group_min_max(PARAM *param, SEL_TREE *tree)
{
  THD *thd= param->thd;
  JOIN *join= thd->lex->select_lex.join;
  TABLE *table= param->table;
  bool have_min= FALSE;              /* TRUE if there is a MIN function. */
  bool have_max= FALSE;              /* TRUE if there is a MAX function. */
  Item_field *min_max_arg_item= NULL;/* The argument of all MIN/MAX functions.*/
  KEY_PART_INFO *min_max_arg_part= NULL; /* The corresponding keypart. */
  uint group_prefix_len= 0; /* Length (in bytes) of the key prefix. */
  KEY *index_info= NULL;    /* The index chosen for data access. */
  uint index= 0;            /* The id of the chosen index. */
  uint group_key_parts= 0;  /* Number of index key parts in the group prefix. */
  uint used_key_parts= 0;   /* Number of index key parts used for access. */
  byte key_infix[MAX_KEY_LENGTH]; /* Constants from equality predicates.*/
  uint key_infix_len= 0;          /* Length of key_infix. */
  TRP_GROUP_MIN_MAX *read_plan= NULL; /* The eventually constructed TRP. */
  uint key_part_nr;
  ORDER *tmp_group;
  Item *item;
  Item_field *item_field;
  DBUG_ENTER("get_best_group_min_max");

  /* Perform few 'cheap' tests whether this access method is applicable. */
  if (!join || (thd->lex->sql_command != SQLCOM_SELECT))
    DBUG_RETURN(NULL);        /* This is not a select statement. */
  if ((join->tables != 1) ||  /* The query must reference one table. */
      ((!join->group_list) && /* Neither GROUP BY nor a DISTINCT query. */
       (!join->select_distinct)))
    DBUG_RETURN(NULL);
6779
  if (table->s->keys == 0)        /* There are no indexes to use. */
6780 6781 6782
    DBUG_RETURN(NULL);

  /* Analyze the query in more detail. */
6783
  List_iterator<Item> select_items_it(join->fields_list);
6784

6785 6786 6787 6788
  /* Check (SA1,SA4) and store the only MIN/MAX argument - the C attribute.*/
  if(join->make_sum_func_list(join->all_fields, join->fields_list, 1))
    DBUG_RETURN(NULL);
  if (join->sum_funcs[0])
6789
  {
6790 6791 6792
    Item_sum *min_max_item;
    Item_sum **func_ptr= join->sum_funcs;
    while ((min_max_item= *(func_ptr++)))
6793
    {
6794 6795 6796 6797 6798
      if (min_max_item->sum_func() == Item_sum::MIN_FUNC)
        have_min= TRUE;
      else if (min_max_item->sum_func() == Item_sum::MAX_FUNC)
        have_max= TRUE;
      else
6799 6800
        DBUG_RETURN(NULL);

6801 6802
      Item *expr= min_max_item->args[0];    /* The argument of MIN/MAX. */
      if (expr->type() == Item::FIELD_ITEM) /* Is it an attribute? */
6803
      {
6804 6805 6806 6807
        if (! min_max_arg_item)
          min_max_arg_item= (Item_field*) expr;
        else if (! min_max_arg_item->eq(expr, 1))
          DBUG_RETURN(NULL);
6808
      }
6809 6810
      else
        DBUG_RETURN(NULL);
6811
    }
6812
  }
6813

6814 6815 6816 6817
  /* Check (SA5). */
  if (join->select_distinct)
  {
    while ((item= select_items_it++))
6818
    {
6819 6820
      if (item->type() != Item::FIELD_ITEM)
        DBUG_RETURN(NULL);
6821 6822 6823 6824 6825 6826 6827 6828 6829 6830 6831 6832 6833 6834 6835 6836
    }
  }

  /* Check (GA4) - that there are no expressions among the group attributes. */
  for (tmp_group= join->group_list; tmp_group; tmp_group= tmp_group->next)
  {
    if ((*tmp_group->item)->type() != Item::FIELD_ITEM)
      DBUG_RETURN(NULL);
  }

  /*
    Check that table has at least one compound index such that the conditions
    (GA1,GA2) are all TRUE. If there is more than one such index, select the
    first one. Here we set the variables: group_prefix_len and index_info.
  */
  KEY *cur_index_info= table->key_info;
6837
  KEY *cur_index_info_end= cur_index_info + table->s->keys;
6838
  KEY_PART_INFO *cur_part= NULL;
6839 6840 6841 6842 6843 6844 6845 6846 6847 6848 6849 6850 6851 6852 6853 6854 6855 6856 6857
  KEY_PART_INFO *end_part; /* Last part for loops. */
  /* Last index part. */
  KEY_PART_INFO *last_part= NULL;
  KEY_PART_INFO *first_non_group_part= NULL;
  KEY_PART_INFO *first_non_infix_part= NULL;
  uint key_infix_parts= 0;
  uint cur_group_key_parts= 0;
  uint cur_group_prefix_len= 0;
  /* Cost-related variables for the best index so far. */
  double best_read_cost= DBL_MAX;
  ha_rows best_records= 0;
  SEL_ARG *best_index_tree= NULL;
  ha_rows best_quick_prefix_records= 0;
  uint best_param_idx= 0;
  double cur_read_cost= DBL_MAX;
  ha_rows cur_records;
  SEL_ARG *cur_index_tree= NULL;
  ha_rows cur_quick_prefix_records= 0;
  uint cur_param_idx;
timour@mysql.com's avatar
timour@mysql.com committed
6858
  key_map cur_used_key_parts;
6859 6860 6861 6862 6863 6864 6865

  for (uint cur_index= 0 ; cur_index_info != cur_index_info_end ;
       cur_index_info++, cur_index++)
  {
    /* Check (B1) - if current index is covering. */
    if (!table->used_keys.is_set(cur_index))
      goto next_index;
6866 6867 6868 6869 6870 6871 6872 6873 6874 6875 6876 6877 6878 6879 6880 6881 6882 6883 6884 6885 6886 6887

    /*
      Check (GA1) for GROUP BY queries.
    */
    if (join->group_list)
    {
      cur_part= cur_index_info->key_part;
      end_part= cur_part + cur_index_info->key_parts;
      /* Iterate in parallel over the GROUP list and the index parts. */
      for (tmp_group= join->group_list; tmp_group && (cur_part != end_part);
           tmp_group= tmp_group->next, cur_part++)
      {
        /*
          TODO:
          tmp_group::item is an array of Item, is it OK to consider only the
          first Item? If so, then why? What is the array for?
        */
        /* Above we already checked that all group items are fields. */
        DBUG_ASSERT((*tmp_group->item)->type() == Item::FIELD_ITEM);
        Item_field *group_field= (Item_field *) (*tmp_group->item);
        if (group_field->field->eq(cur_part->field))
        {
6888 6889
          cur_group_prefix_len+= cur_part->store_length;
          ++cur_group_key_parts;
6890 6891 6892 6893 6894 6895 6896 6897 6898 6899 6900 6901 6902 6903 6904
        }
        else
          goto next_index;
      }
    }
    /*
      Check (GA2) if this is a DISTINCT query.
      If GA2, then Store a new ORDER object in group_fields_array at the
      position of the key part of item_field->field. Thus we get the ORDER
      objects for each field ordered as the corresponding key parts.
      Later group_fields_array of ORDER objects is used to convert the query
      to a GROUP query.
    */
    else if (join->select_distinct)
    {
6905
      select_items_it.rewind();
timour@mysql.com's avatar
timour@mysql.com committed
6906
      cur_used_key_parts.clear_all();
6907
      while ((item= select_items_it++))
6908
      {
6909
        item_field= (Item_field*) item; /* (SA5) already checked above. */
6910 6911
        /* Find the order of the key part in the index. */
        key_part_nr= get_field_keypart(cur_index_info, item_field->field);
timour@mysql.com's avatar
timour@mysql.com committed
6912 6913 6914 6915 6916 6917
        /*
          Check if this attribute was already present in the select list.
          If it was present, then its corresponding key part was alredy used.
        */
        if (cur_used_key_parts.is_set(key_part_nr))
          continue;
6918
        if (key_part_nr < 1 || key_part_nr > join->fields_list.elements)
6919 6920
          goto next_index;
        cur_part= cur_index_info->key_part + key_part_nr - 1;
6921
        cur_group_prefix_len+= cur_part->store_length;
timour@mysql.com's avatar
timour@mysql.com committed
6922 6923
        cur_used_key_parts.set_bit(key_part_nr);
        ++cur_group_key_parts;
6924 6925 6926 6927 6928 6929 6930 6931 6932
      }
    }
    else
      DBUG_ASSERT(FALSE);

    /* Check (SA2). */
    if (min_max_arg_item)
    {
      key_part_nr= get_field_keypart(cur_index_info, min_max_arg_item->field);
6933
      if (key_part_nr <= cur_group_key_parts)
6934 6935 6936 6937 6938 6939 6940 6941
        goto next_index;
      min_max_arg_part= cur_index_info->key_part + key_part_nr - 1;
    }

    /*
      Check (NGA1, NGA2) and extract a sequence of constants to be used as part
      of all search keys.
    */
6942 6943 6944 6945 6946 6947 6948 6949 6950 6951 6952 6953 6954 6955 6956 6957 6958 6959 6960 6961 6962 6963

    /*
      If there is MIN/MAX, each keypart between the last group part and the
      MIN/MAX part must participate in one equality with constants, and all
      keyparts after the MIN/MAX part must not be referenced in the query.

      If there is no MIN/MAX, the keyparts after the last group part can be
      referenced only in equalities with constants, and the referenced keyparts
      must form a sequence without any gaps that starts immediately after the
      last group keypart.
    */
    last_part= cur_index_info->key_part + cur_index_info->key_parts;
    first_non_group_part= (cur_group_key_parts < cur_index_info->key_parts) ?
                          cur_index_info->key_part + cur_group_key_parts :
                          NULL;
    first_non_infix_part= min_max_arg_part ?
                          (min_max_arg_part < last_part) ?
                             min_max_arg_part + 1 :
                             NULL :
                           NULL;
    if (first_non_group_part &&
        (!min_max_arg_part || (min_max_arg_part - first_non_group_part > 0)))
6964
    {
6965 6966 6967 6968 6969 6970 6971 6972 6973 6974 6975 6976 6977 6978 6979 6980 6981
      if (tree)
      {
        uint dummy;
        SEL_ARG *index_range_tree= get_index_range_tree(cur_index, tree, param,
                                                        &dummy);
        if (!get_constant_key_infix(cur_index_info, index_range_tree,
                                    first_non_group_part, min_max_arg_part,
                                    last_part, thd, key_infix, &key_infix_len,
                                    &first_non_infix_part))
          goto next_index;
      }
      else if (min_max_arg_part &&
               (min_max_arg_part - first_non_group_part > 0))
        /*
          There is a gap but no range tree, thus no predicates at all for the
          non-group keyparts.
        */
6982 6983 6984
        goto next_index;
    }

6985 6986 6987 6988 6989 6990 6991 6992 6993 6994 6995 6996 6997
    /*
      Test (WA1) partially - that no other keypart after the last infix part is
      referenced in the query.
    */
    if (first_non_infix_part)
    {
      for (cur_part= first_non_infix_part; cur_part != last_part; cur_part++)
      {
        if (cur_part->field->query_id == thd->query_id)
          goto next_index;
      }
    }

6998
    /* If we got to this point, cur_index_info passes the test. */
6999 7000 7001
    key_infix_parts= key_infix_len ?
                     (first_non_infix_part - first_non_group_part) : 0;
    used_key_parts= cur_group_key_parts + key_infix_parts;
7002

7003 7004 7005 7006 7007 7008 7009 7010 7011 7012 7013 7014 7015 7016
    /* Compute the cost of using this index. */
    if (tree)
    {
      /* Find the SEL_ARG sub-tree that corresponds to the chosen index. */
      cur_index_tree= get_index_range_tree(cur_index, tree, param,
                                           &cur_param_idx);
      /* Check if this range tree can be used for prefix retrieval. */
      cur_quick_prefix_records= check_quick_select(param, cur_param_idx,
                                                    cur_index_tree);
    }
    cost_group_min_max(table, cur_index_info, used_key_parts,
                       cur_group_key_parts, tree, cur_index_tree,
                       cur_quick_prefix_records, have_min, have_max,
                       &cur_read_cost, &cur_records);
timour@mysql.com's avatar
timour@mysql.com committed
7017 7018 7019 7020 7021 7022
    /*
      If cur_read_cost is lower than best_read_cost use cur_index.
      Do not compare doubles directly because they may have different
      representations (64 vs. 80 bits).
    */
    if (cur_read_cost < best_read_cost - (DBL_EPSILON * cur_read_cost))
7023 7024 7025 7026 7027 7028 7029 7030 7031 7032 7033
    {
      index_info= cur_index_info;
      index= cur_index;
      best_read_cost= cur_read_cost;
      best_records= cur_records;
      best_index_tree= cur_index_tree;
      best_quick_prefix_records= cur_quick_prefix_records;
      best_param_idx= cur_param_idx;
      group_key_parts= cur_group_key_parts;
      group_prefix_len= cur_group_prefix_len;
    }
7034 7035

  next_index:
7036 7037
    cur_group_key_parts= 0;
    cur_group_prefix_len= 0;
7038 7039 7040 7041
  }
  if (!index_info) /* No usable index found. */
    DBUG_RETURN(NULL);

7042 7043 7044
  /* Check (SA3) for the where clause. */
  if (join->conds && min_max_arg_item &&
      !check_group_min_max_predicates(join->conds, min_max_arg_item,
7045 7046
                                      (index_info->flags & HA_SPATIAL) ?
                                      Field::itMBR : Field::itRAW))
7047 7048 7049 7050
    DBUG_RETURN(NULL);

  /* The query passes all tests, so construct a new TRP object. */
  read_plan= new (param->mem_root)
7051 7052 7053 7054
                 TRP_GROUP_MIN_MAX(have_min, have_max, min_max_arg_part,
                                   group_prefix_len, used_key_parts,
                                   group_key_parts, index_info, index,
                                   key_infix_len,
7055
                                   (key_infix_len > 0) ? key_infix : NULL,
7056
                                   tree, best_index_tree, best_param_idx,
7057
                                   best_quick_prefix_records);
7058 7059 7060 7061 7062
  if (read_plan)
  {
    if (tree && read_plan->quick_prefix_records == 0)
      DBUG_RETURN(NULL);

7063 7064 7065
    read_plan->read_cost= best_read_cost;
    read_plan->records=   best_records;

7066 7067 7068 7069 7070 7071 7072 7073 7074 7075
    DBUG_PRINT("info",
               ("Returning group min/max plan: cost: %g, records: %lu",
                read_plan->read_cost, (ulong) read_plan->records));
  }

  DBUG_RETURN(read_plan);
}


/*
7076 7077
  Check that the MIN/MAX attribute participates only in range predicates
  with constants.
7078 7079 7080 7081 7082 7083

  SYNOPSIS
    check_group_min_max_predicates()
    cond              tree (or subtree) describing all or part of the WHERE
                      clause being analyzed
    min_max_arg_item  the field referenced by the MIN/MAX function(s)
7084
    min_max_arg_part  the keypart of the MIN/MAX argument if any
7085 7086 7087

  DESCRIPTION
    The function walks recursively over the cond tree representing a WHERE
7088
    clause, and checks condition (SA3) - if a field is referenced by a MIN/MAX
7089 7090
    aggregate function, it is referenced only by one of the following
    predicates: {=, !=, <, <=, >, >=, between, is null, is not null}.
7091 7092 7093 7094 7095 7096 7097

  RETURN
    TRUE  if cond passes the test
    FALSE o/w
*/

static bool
7098 7099
check_group_min_max_predicates(COND *cond, Item_field *min_max_arg_item,
                               Field::imagetype image_type)
7100 7101
{
  DBUG_ENTER("check_group_min_max_predicates");
7102
  DBUG_ASSERT(cond && min_max_arg_item);
7103 7104 7105 7106 7107 7108 7109 7110 7111

  Item::Type cond_type= cond->type();
  if (cond_type == Item::COND_ITEM) /* 'AND' or 'OR' */
  {
    DBUG_PRINT("info", ("Analyzing: %s", ((Item_func*) cond)->func_name()));
    List_iterator_fast<Item> li(*((Item_cond*) cond)->argument_list());
    Item *and_or_arg;
    while ((and_or_arg= li++))
    {
7112 7113
      if(!check_group_min_max_predicates(and_or_arg, min_max_arg_item,
                                         image_type))
7114 7115 7116 7117 7118
        DBUG_RETURN(FALSE);
    }
    DBUG_RETURN(TRUE);
  }

7119 7120 7121 7122 7123 7124 7125 7126 7127 7128 7129 7130 7131
  /*
    TODO:
    This is a very crude fix to handle sub-selects in the WHERE clause
    (Item_subselect objects). With the test below we rule out from the
    optimization all queries with subselects in the WHERE clause. What has to
    be done, is that here we should analyze whether the subselect references
    the MIN/MAX argument field, and disallow the optimization only if this is
    so.
  */
  if (cond_type == Item::SUBSELECT_ITEM)
    DBUG_RETURN(FALSE);
  
  /* We presume that at this point there are no other Items than functions. */
7132 7133 7134 7135 7136 7137 7138 7139 7140 7141 7142 7143 7144
  DBUG_ASSERT(cond_type == Item::FUNC_ITEM);

  /* Test if cond references only group-by or non-group fields. */
  Item_func *pred= (Item_func*) cond;
  Item **arguments= pred->arguments();
  Item *cur_arg;
  DBUG_PRINT("info", ("Analyzing: %s", pred->func_name()));
  for (uint arg_idx= 0; arg_idx < pred->argument_count (); arg_idx++)
  {
    cur_arg= arguments[arg_idx];
    DBUG_PRINT("info", ("cur_arg: %s", cur_arg->full_name()));
    if (cur_arg->type() == Item::FIELD_ITEM)
    {
7145
      if (min_max_arg_item->eq(cur_arg, 1)) 
7146 7147 7148
      {
       /*
         If pred references the MIN/MAX argument, check whether pred is a range
7149
         condition that compares the MIN/MAX argument with a constant.
7150 7151
       */
        Item_func::Functype pred_type= pred->functype();
7152 7153 7154 7155 7156 7157 7158 7159 7160 7161
        if (pred_type != Item_func::EQUAL_FUNC     &&
            pred_type != Item_func::LT_FUNC        &&
            pred_type != Item_func::LE_FUNC        &&
            pred_type != Item_func::GT_FUNC        &&
            pred_type != Item_func::GE_FUNC        &&
            pred_type != Item_func::BETWEEN        &&
            pred_type != Item_func::ISNULL_FUNC    &&
            pred_type != Item_func::ISNOTNULL_FUNC &&
            pred_type != Item_func::EQ_FUNC        &&
            pred_type != Item_func::NE_FUNC)
7162 7163 7164 7165
          DBUG_RETURN(FALSE);

        /* Check that pred compares min_max_arg_item with a constant. */
        Item *args[3];
7166
        bzero(args, 3 * sizeof(Item*));
7167 7168 7169 7170
        bool inv;
        /* Test if this is a comparison of a field and a constant. */
        if (!simple_pred(pred, args, &inv))
          DBUG_RETURN(FALSE);
7171 7172 7173 7174 7175 7176 7177 7178 7179 7180 7181 7182 7183 7184 7185 7186 7187 7188 7189

        /* Check for compatible string comparisons - similar to get_mm_leaf. */
        if (args[0] && args[1] && !args[2] && // this is a binary function
            min_max_arg_item->result_type() == STRING_RESULT &&
            /*
              Don't use an index when comparing strings of different collations.
            */
            ((args[1]->result_type() == STRING_RESULT &&
              image_type == Field::itRAW &&
              ((Field_str*) min_max_arg_item->field)->charset() !=
              pred->compare_collation())
             ||
             /*
               We can't always use indexes when comparing a string index to a
               number.
             */
             (args[1]->result_type() != STRING_RESULT &&
              min_max_arg_item->field->cmp_type() != args[1]->result_type())))
          DBUG_RETURN(FALSE);
7190 7191 7192 7193
      }
    }
    else if (cur_arg->type() == Item::FUNC_ITEM)
    {
7194 7195
      if(!check_group_min_max_predicates(cur_arg, min_max_arg_item,
                                         image_type))
7196 7197 7198 7199 7200 7201 7202 7203 7204 7205 7206 7207 7208 7209 7210 7211 7212 7213 7214
        DBUG_RETURN(FALSE);
    }
    else if (cur_arg->const_item())
    {
      DBUG_RETURN(TRUE);
    }
    else
      DBUG_RETURN(FALSE);
  }

  DBUG_RETURN(TRUE);
}


/*
  Extract a sequence of constants from a conjunction of equality predicates.

  SYNOPSIS
    get_constant_key_infix()
7215 7216 7217 7218 7219 7220 7221 7222 7223
    index_info             [in]  Descriptor of the chosen index.
    index_range_tree       [in]  Range tree for the chosen index
    first_non_group_part   [in]  First index part after group attribute parts
    min_max_arg_part       [in]  The keypart of the MIN/MAX argument if any
    last_part              [in]  Last keypart of the index
    thd                    [in]  Current thread
    key_infix              [out] Infix of constants to be used for index lookup
    key_infix_len          [out] Lenghth of the infix
    first_non_infix_part   [out] The first keypart after the infix (if any)
7224 7225 7226
    
  DESCRIPTION
    Test conditions (NGA1, NGA2) from get_best_group_min_max(). Namely,
7227 7228 7229 7230
    for each keypart field NGF_i not in GROUP-BY, check that there is a constant
    equality predicate among conds with the form (NGF_i = const_ci) or
    (const_ci = NGF_i).
    Thus all the NGF_i attributes must fill the 'gap' between the last group-by
7231 7232 7233 7234 7235 7236
    attribute and the MIN/MAX attribute in the index (if present). If these
    conditions hold, copy each constant from its corresponding predicate into
    key_infix, in the order its NG_i attribute appears in the index, and update
    key_infix_len with the total length of the key parts in key_infix.

  RETURN
7237
    TRUE  if the index passes the test
7238 7239 7240 7241
    FALSE o/w
*/

static bool
7242
get_constant_key_infix(KEY *index_info, SEL_ARG *index_range_tree,
7243
                       KEY_PART_INFO *first_non_group_part,
7244 7245 7246 7247
                       KEY_PART_INFO *min_max_arg_part,
                       KEY_PART_INFO *last_part, THD *thd,
                       byte *key_infix, uint *key_infix_len,
                       KEY_PART_INFO **first_non_infix_part)
7248 7249 7250
{
  SEL_ARG       *cur_range;
  KEY_PART_INFO *cur_part;
7251 7252
  /* End part for the first loop below. */
  KEY_PART_INFO *end_part= min_max_arg_part ? min_max_arg_part : last_part;
7253 7254 7255 7256 7257 7258 7259 7260 7261 7262 7263 7264 7265 7266 7267 7268 7269

  *key_infix_len= 0;
  byte *key_ptr= key_infix;
  for (cur_part= first_non_group_part; cur_part != end_part; cur_part++)
  {
    /*
      Find the range tree for the current keypart. We assume that
      index_range_tree points to the leftmost keypart in the index.
    */
    for (cur_range= index_range_tree; cur_range;
         cur_range= cur_range->next_key_part)
    {
      if (cur_range->field->eq(cur_part->field))
        break;
    }
    if (!cur_range)
    {
7270 7271 7272 7273 7274 7275 7276
      if (min_max_arg_part)
        return FALSE; /* The current keypart has no range predicates at all. */
      else
      {
        *first_non_infix_part= cur_part;
        return TRUE;
      }
7277 7278 7279 7280 7281 7282 7283 7284 7285 7286 7287 7288 7289 7290 7291 7292 7293 7294 7295 7296 7297 7298 7299 7300
    }

    /* Check that the current range tree is a single point interval. */
    if (cur_range->prev || cur_range->next)
      return FALSE; /* This is not the only range predicate for the field. */
    if ((cur_range->min_flag & NO_MIN_RANGE) ||
        (cur_range->max_flag & NO_MAX_RANGE) ||
        (cur_range->min_flag & NEAR_MIN) || (cur_range->max_flag & NEAR_MAX))
      return FALSE;

    uint field_length= cur_part->store_length;
    if ((cur_range->maybe_null &&
         cur_range->min_value[0] && cur_range->max_value[0])
        ||
        (memcmp(cur_range->min_value, cur_range->max_value, field_length) == 0))
    { /* cur_range specifies 'IS NULL' or an equality condition. */
      memcpy(key_ptr, cur_range->min_value, field_length);
      key_ptr+= field_length;
      *key_infix_len+= field_length;
    }
    else
      return FALSE;
  }

7301 7302 7303
  if (!min_max_arg_part && (cur_part == last_part))
    *first_non_infix_part= last_part;

7304 7305 7306 7307
  return TRUE;
}


7308 7309 7310 7311 7312 7313 7314 7315 7316 7317 7318 7319 7320 7321 7322 7323 7324 7325 7326 7327
/*
  Find the key part referenced by a field.

  SYNOPSIS
    get_field_keypart()
    index  descriptor of an index
    field  field that possibly references some key part in index

  NOTES
    The return value can be used to get a KEY_PART_INFO pointer by
    part= index->key_part + get_field_keypart(...) - 1;

  RETURN
    Positive number which is the consecutive number of the key part, or
    0 if field does not reference any index field.
*/

static inline uint
get_field_keypart(KEY *index, Field *field)
{
7328
  KEY_PART_INFO *part, *end;
7329

7330
  for (part= index->key_part, end= part + index->key_parts; part < end; part++)
7331 7332
  {
    if (field->eq(part->field))
ram@gw.mysql.r18.ru's avatar
ram@gw.mysql.r18.ru committed
7333
      return part - index->key_part + 1;
7334
  }
7335
  return 0;
7336 7337 7338 7339 7340 7341 7342 7343 7344 7345 7346 7347 7348 7349 7350 7351 7352 7353 7354 7355 7356 7357 7358 7359 7360 7361 7362 7363 7364 7365 7366 7367 7368 7369 7370 7371 7372 7373 7374 7375 7376
}


/*
  Find the SEL_ARG sub-tree that corresponds to the chosen index.

  SYNOPSIS
    get_index_range_tree()
    index     [in]  The ID of the index being looked for
    range_tree[in]  Tree of ranges being searched
    param     [in]  PARAM from SQL_SELECT::test_quick_select
    param_idx [out] Index in the array PARAM::key that corresponds to 'index'

  DESCRIPTION

    A SEL_TREE contains range trees for all usable indexes. This procedure
    finds the SEL_ARG sub-tree for 'index'. The members of a SEL_TREE are
    ordered in the same way as the members of PARAM::key, thus we first find
    the corresponding index in the array PARAM::key. This index is returned
    through the variable param_idx, to be used later as argument of
    check_quick_select().

  RETURN
    Pointer to the SEL_ARG subtree that corresponds to index.
*/

SEL_ARG * get_index_range_tree(uint index, SEL_TREE* range_tree, PARAM *param,
                               uint *param_idx)
{
  uint idx= 0; /* Index nr in param->key_parts */
  while (idx < param->keys)
  {
    if (index == param->real_keynr[idx])
      break;
    idx++;
  }
  *param_idx= idx;
  return(range_tree->keys[idx]);
}


7377
/*
7378
  Compute the cost of a quick_group_min_max_select for a particular index.
7379 7380

  SYNOPSIS
7381 7382 7383 7384 7385 7386 7387
    cost_group_min_max()
    table                [in] The table being accessed
    index_info           [in] The index used to access the table
    used_key_parts       [in] Number of key parts used to access the index
    group_key_parts      [in] Number of index key parts in the group prefix
    range_tree           [in] Tree of ranges for all indexes
    index_tree           [in] The range tree for the current index
monty@mysql.com's avatar
monty@mysql.com committed
7388 7389
    quick_prefix_records [in] Number of records retrieved by the internally
			      used quick range select if any
7390 7391 7392 7393
    have_min             [in] True if there is a MIN function
    have_max             [in] True if there is a MAX function
    read_cost           [out] The cost to retrieve rows via this quick select
    records             [out] The number of rows retrieved
7394 7395

  DESCRIPTION
monty@mysql.com's avatar
monty@mysql.com committed
7396 7397
    This method computes the access cost of a TRP_GROUP_MIN_MAX instance and
    the number of rows returned. It updates this->read_cost and this->records.
7398 7399 7400 7401 7402 7403 7404 7405 7406 7407 7408 7409 7410 7411 7412 7413 7414 7415 7416 7417 7418 7419 7420 7421 7422 7423 7424 7425 7426 7427 7428 7429 7430 7431 7432 7433 7434 7435 7436

  NOTES
    The cost computation distinguishes several cases:
    1) No equality predicates over non-group attributes (thus no key_infix).
       If groups are bigger than blocks on the average, then we assume that it
       is very unlikely that block ends are aligned with group ends, thus even
       if we look for both MIN and MAX keys, all pairs of neighbor MIN/MAX
       keys, except for the first MIN and the last MAX keys, will be in the
       same block.  If groups are smaller than blocks, then we are going to
       read all blocks.
    2) There are equality predicates over non-group attributes.
       In this case the group prefix is extended by additional constants, and
       as a result the min/max values are inside sub-groups of the original
       groups. The number of blocks that will be read depends on whether the
       ends of these sub-groups will be contained in the same or in different
       blocks. We compute the probability for the two ends of a subgroup to be
       in two different blocks as the ratio of:
       - the number of positions of the left-end of a subgroup inside a group,
         such that the right end of the subgroup is past the end of the buffer
         containing the left-end, and
       - the total number of possible positions for the left-end of the
         subgroup, which is the number of keys in the containing group.
       We assume it is very unlikely that two ends of subsequent subgroups are
       in the same block.
    3) The are range predicates over the group attributes.
       Then some groups may be filtered by the range predicates. We use the
       selectivity of the range predicates to decide how many groups will be
       filtered.

  TODO
     - Take into account the optional range predicates over the MIN/MAX
       argument.
     - Check if we have a PK index and we use all cols - then each key is a
       group, and it will be better to use an index scan.

  RETURN
    None
*/

7437 7438 7439 7440 7441
void cost_group_min_max(TABLE* table, KEY *index_info, uint used_key_parts,
                        uint group_key_parts, SEL_TREE *range_tree,
                        SEL_ARG *index_tree, ha_rows quick_prefix_records,
                        bool have_min, bool have_max,
                        double *read_cost, ha_rows *records)
7442 7443 7444 7445 7446 7447 7448 7449 7450 7451 7452 7453
{
  uint table_records;
  uint num_groups;
  uint num_blocks;
  uint keys_per_block;
  uint keys_per_group;
  uint keys_per_subgroup; /* Average number of keys in sub-groups */
                          /* formed by a key infix. */
  double p_overlap; /* Probability that a sub-group overlaps two blocks. */
  double quick_prefix_selectivity;
  double io_cost;
  double cpu_cost= 0; /* TODO: CPU cost of index_read calls? */
timour@mysql.com's avatar
timour@mysql.com committed
7454
  DBUG_ENTER("cost_group_min_max");
monty@mysql.com's avatar
monty@mysql.com committed
7455

7456 7457 7458 7459 7460 7461 7462 7463 7464 7465 7466 7467 7468 7469 7470 7471 7472 7473
  table_records= table->file->records;
  keys_per_block= (table->file->block_size / 2 /
                   (index_info->key_length + table->file->ref_length)
                        + 1);
  num_blocks= (table_records / keys_per_block) + 1;

  /* Compute the number of keys in a group. */
  keys_per_group= index_info->rec_per_key[group_key_parts - 1];
  if (keys_per_group == 0) /* If there is no statistics try to guess */
    /* each group contains 10% of all records */
    keys_per_group= (table_records / 10) + 1;
  num_groups= (table_records / keys_per_group) + 1;

  /* Apply the selectivity of the quick select for group prefixes. */
  if (range_tree && (quick_prefix_records != HA_POS_ERROR))
  {
    quick_prefix_selectivity= (double) quick_prefix_records /
                              (double) table_records;
serg@serg.mylan's avatar
serg@serg.mylan committed
7474
    num_groups= (uint) rint(num_groups * quick_prefix_selectivity);
7475 7476 7477 7478 7479 7480 7481 7482 7483 7484 7485 7486 7487 7488 7489 7490 7491 7492 7493 7494 7495 7496 7497 7498 7499 7500 7501 7502 7503 7504 7505
  }

  if (used_key_parts > group_key_parts)
  { /*
      Compute the probability that two ends of a subgroup are inside
      different blocks.
    */
    keys_per_subgroup= index_info->rec_per_key[used_key_parts - 1];
    if (keys_per_subgroup >= keys_per_block) /* If a subgroup is bigger than */
      p_overlap= 1.0;       /* a block, it will overlap at least two blocks. */
    else
    {
      double blocks_per_group= (double) num_blocks / (double) num_groups;
      p_overlap= (blocks_per_group * (keys_per_subgroup - 1)) / keys_per_group;
      p_overlap= min(p_overlap, 1.0);
    }
    io_cost= (double) min(num_groups * (1 + p_overlap), num_blocks);
  }
  else
    io_cost= (keys_per_group > keys_per_block) ?
             (have_min && have_max) ? (double) (num_groups + 1) :
                                      (double) num_groups :
             (double) num_blocks;

  /*
    TODO: If there is no WHERE clause and no other expressions, there should be
    no CPU cost. We leave it here to make this cost comparable to that of index
    scan as computed in SQL_SELECT::test_quick_select().
  */
  cpu_cost= (double) num_groups / TIME_FOR_COMPARE;

7506
  *read_cost= io_cost + cpu_cost;
7507
  *records= num_groups;
7508 7509 7510 7511 7512 7513 7514 7515 7516 7517 7518 7519 7520 7521 7522 7523 7524 7525 7526 7527 7528 7529 7530 7531 7532 7533 7534 7535 7536 7537 7538 7539 7540 7541 7542 7543 7544

  DBUG_PRINT("info",
             ("records=%u, keys/block=%u, keys/group=%u, records=%u, blocks=%u",
              table_records, keys_per_block, keys_per_group, records,
              num_blocks));
  DBUG_VOID_RETURN;
}


/*
  Construct a new quick select object for queries with group by with min/max.

  SYNOPSIS
    TRP_GROUP_MIN_MAX::make_quick()
    param              Parameter from test_quick_select
    retrieve_full_rows ignored
    parent_alloc       Memory pool to use, if any.

  NOTES
    Make_quick ignores the retrieve_full_rows parameter because
    QUICK_GROUP_MIN_MAX_SELECT always performs 'index only' scans.
    The other parameter are ignored as well because all necessary
    data to create the QUICK object is computed at this TRP creation
    time.

  RETURN
    New QUICK_GROUP_MIN_MAX_SELECT object if successfully created,
    NULL o/w.
*/

QUICK_SELECT_I *
TRP_GROUP_MIN_MAX::make_quick(PARAM *param, bool retrieve_full_rows,
                              MEM_ROOT *parent_alloc)
{
  QUICK_GROUP_MIN_MAX_SELECT *quick;
  DBUG_ENTER("TRP_GROUP_MIN_MAX::make_quick");

7545 7546 7547 7548 7549
  quick= new QUICK_GROUP_MIN_MAX_SELECT(param->table,
                                        param->thd->lex->select_lex.join,
                                        have_min, have_max, min_max_arg_part,
                                        group_prefix_len, used_key_parts,
                                        index_info, index, read_cost, records,
monty@mysql.com's avatar
monty@mysql.com committed
7550 7551
                                        key_infix_len, key_infix,
                                        parent_alloc);
7552 7553 7554 7555 7556 7557 7558 7559 7560 7561 7562 7563 7564 7565 7566 7567
  if (!quick)
    DBUG_RETURN(NULL);

  if (quick->init())
  {
    delete quick;
    DBUG_RETURN(NULL);
  }

  if (range_tree)
  {
    DBUG_ASSERT(quick_prefix_records > 0);
    if (quick_prefix_records == HA_POS_ERROR)
      quick->quick_prefix_select= NULL; /* Can't construct a quick select. */
    else
      /* Make a QUICK_RANGE_SELECT to be used for group prefix retrieval. */
7568 7569
      quick->quick_prefix_select= get_quick_select(param, param_idx,
                                                   index_tree,
7570 7571 7572 7573 7574 7575 7576 7577 7578 7579 7580 7581 7582 7583 7584 7585 7586 7587 7588 7589 7590 7591
                                                   &quick->alloc);

    /*
      Extract the SEL_ARG subtree that contains only ranges for the MIN/MAX
      attribute, and create an array of QUICK_RANGES to be used by the
      new quick select.
    */
    if (min_max_arg_part)
    {
      SEL_ARG *min_max_range= index_tree;
      while (min_max_range) /* Find the tree for the MIN/MAX key part. */
      {
        if (min_max_range->field->eq(min_max_arg_part->field))
          break;
        min_max_range= min_max_range->next_key_part;
      }
      /* Scroll to the leftmost interval for the MIN/MAX argument. */
      while (min_max_range && min_max_range->prev)
        min_max_range= min_max_range->prev;
      /* Create an array of QUICK_RANGEs for the MIN/MAX argument. */
      while (min_max_range)
      {
7592
        if (quick->add_range(min_max_range))
7593 7594 7595 7596 7597 7598 7599 7600 7601 7602 7603 7604 7605 7606 7607 7608 7609 7610 7611 7612 7613 7614 7615 7616 7617 7618 7619 7620 7621 7622 7623 7624 7625 7626 7627 7628 7629 7630 7631 7632 7633 7634
        {
          delete quick;
          quick= NULL;
          DBUG_RETURN(NULL);
        }
        min_max_range= min_max_range->next;
      }
    }
  }
  else
    quick->quick_prefix_select= NULL;

  quick->update_key_stat();

  DBUG_RETURN(quick);
}


/*
  Construct new quick select for group queries with min/max.

  SYNOPSIS
    QUICK_GROUP_MIN_MAX_SELECT::QUICK_GROUP_MIN_MAX_SELECT()
    table             The table being accessed
    join              Descriptor of the current query
    have_min          TRUE if the query selects a MIN function
    have_max          TRUE if the query selects a MAX function
    min_max_arg_part  The only argument field of all MIN/MAX functions
    group_prefix_len  Length of all key parts in the group prefix
    prefix_key_parts  All key parts in the group prefix
    index_info        The index chosen for data access
    use_index         The id of index_info
    read_cost         Cost of this access method
    records           Number of records returned
    key_infix_len     Length of the key infix appended to the group prefix
    key_infix         Infix of constants from equality predicates
    parent_alloc      Memory pool for this and quick_prefix_select data

  RETURN
    None
*/

monty@mysql.com's avatar
monty@mysql.com committed
7635 7636 7637 7638 7639 7640 7641 7642 7643 7644 7645 7646 7647
QUICK_GROUP_MIN_MAX_SELECT::
QUICK_GROUP_MIN_MAX_SELECT(TABLE *table, JOIN *join_arg, bool have_min_arg,
                           bool have_max_arg,
                           KEY_PART_INFO *min_max_arg_part_arg,
                           uint group_prefix_len_arg,
                           uint used_key_parts_arg, KEY *index_info_arg,
                           uint use_index, double read_cost_arg,
                           ha_rows records_arg, uint key_infix_len_arg,
                           byte *key_infix_arg, MEM_ROOT *parent_alloc)
  :join(join_arg), index_info(index_info_arg),
   group_prefix_len(group_prefix_len_arg), have_min(have_min_arg),
   have_max(have_max_arg), seen_first_key(FALSE),
   min_max_arg_part(min_max_arg_part_arg), key_infix(key_infix_arg),
7648 7649
   key_infix_len(key_infix_len_arg), min_functions_it(NULL),
   max_functions_it(NULL)
7650 7651 7652 7653 7654 7655
{
  head=       table;
  file=       head->file;
  index=      use_index;
  record=     head->record[0];
  tmp_record= head->record[1];
7656 7657 7658
  read_time= read_cost_arg;
  records= records_arg;
  used_key_parts= used_key_parts_arg;
7659 7660 7661
  real_prefix_len= group_prefix_len + key_infix_len;
  group_prefix= NULL;
  min_max_arg_len= min_max_arg_part ? min_max_arg_part->store_length : 0;
monty@mysql.com's avatar
monty@mysql.com committed
7662 7663 7664 7665 7666 7667

  /*
    We can't have parent_alloc set as the init function can't handle this case
    yet.
  */
  DBUG_ASSERT(!parent_alloc);
7668 7669 7670
  if (!parent_alloc)
  {
    init_sql_alloc(&alloc, join->thd->variables.range_alloc_block_size, 0);
monty@mysql.com's avatar
monty@mysql.com committed
7671
    join->thd->mem_root= &alloc;
7672 7673
  }
  else
7674
    bzero(&alloc, sizeof(MEM_ROOT));            // ensure that it's not used
7675 7676 7677 7678 7679 7680 7681 7682 7683
}


/*
  Do post-constructor initialization.

  SYNOPSIS
    QUICK_GROUP_MIN_MAX_SELECT::init()
  
7684 7685 7686 7687 7688 7689
  DESCRIPTION
    The method performs initialization that cannot be done in the constructor
    such as memory allocations that may fail. It allocates memory for the
    group prefix and inifix buffers, and for the lists of MIN/MAX item to be
    updated during execution.

7690 7691 7692 7693 7694 7695 7696 7697 7698 7699 7700 7701 7702 7703 7704 7705 7706 7707 7708 7709 7710 7711 7712 7713 7714 7715 7716 7717 7718 7719 7720 7721 7722 7723 7724 7725 7726 7727
  RETURN
    0      OK
    other  Error code
*/

int QUICK_GROUP_MIN_MAX_SELECT::init()
{
  if (group_prefix) /* Already initialized. */
    return 0;

  if (!(last_prefix= (byte*) alloc_root(&alloc, group_prefix_len)))
      return 1;
  /*
    We may use group_prefix to store keys with all select fields, so allocate
    enough space for it.
  */
  if (!(group_prefix= (byte*) alloc_root(&alloc,
                                         real_prefix_len + min_max_arg_len)))
    return 1;

  if (key_infix_len > 0)
  {
    /*
      The memory location pointed to by key_infix will be deleted soon, so
      allocate a new buffer and copy the key_infix into it.
    */
    byte *tmp_key_infix= (byte*) alloc_root(&alloc, key_infix_len);
    if (!tmp_key_infix)
      return 1;
    memcpy(tmp_key_infix, this->key_infix, key_infix_len);
    this->key_infix= tmp_key_infix;
  }

  if (min_max_arg_part)
  {
    if(my_init_dynamic_array(&min_max_ranges, sizeof(QUICK_RANGE*), 16, 16))
      return 1;

7728 7729 7730 7731 7732 7733 7734 7735 7736 7737 7738 7739 7740 7741
    if (have_min)
    {
      if(!(min_functions= new List<Item_sum>))
        return 1;
    }
    else
      min_functions= NULL;
    if (have_max)
    {
      if(!(max_functions= new List<Item_sum>))
        return 1;
    }
    else
      max_functions= NULL;
7742

7743 7744 7745
    Item_sum *min_max_item;
    Item_sum **func_ptr= join->sum_funcs;
    while ((min_max_item= *(func_ptr++)))
7746
    {
7747 7748 7749 7750
      if (have_min && (min_max_item->sum_func() == Item_sum::MIN_FUNC))
        min_functions->push_back(min_max_item);
      else if (have_max && (min_max_item->sum_func() == Item_sum::MAX_FUNC))
        max_functions->push_back(min_max_item);
7751 7752
    }

7753 7754 7755 7756 7757 7758 7759 7760 7761 7762 7763
    if (have_min)
    {
      if (!(min_functions_it= new List_iterator<Item_sum>(*min_functions)))
        return 1;
    }

    if (have_max)
    {
      if (!(max_functions_it= new List_iterator<Item_sum>(*max_functions)))
        return 1;
    }
7764
  }
igor@rurik.mysql.com's avatar
igor@rurik.mysql.com committed
7765 7766
  else
    min_max_ranges.elements= 0;
7767 7768 7769 7770 7771 7772 7773 7774 7775 7776 7777 7778 7779

  return 0;
}


QUICK_GROUP_MIN_MAX_SELECT::~QUICK_GROUP_MIN_MAX_SELECT()
{
  DBUG_ENTER("QUICK_GROUP_MIN_MAX_SELECT::~QUICK_GROUP_MIN_MAX_SELECT");
  if (file->inited != handler::NONE) 
    file->ha_index_end();
  if (min_max_arg_part)
    delete_dynamic(&min_max_ranges);
  free_root(&alloc,MYF(0));
7780 7781
  delete min_functions_it;
  delete max_functions_it;
7782 7783 7784 7785 7786 7787 7788 7789 7790 7791 7792 7793 7794 7795 7796 7797 7798 7799 7800
  delete quick_prefix_select;
  DBUG_VOID_RETURN; 
}


/*
  Eventually create and add a new quick range object.

  SYNOPSIS
    QUICK_GROUP_MIN_MAX_SELECT::add_range()
    sel_range  Range object from which a 

  NOTES
    Construct a new QUICK_RANGE object from a SEL_ARG object, and
    add it to the array min_max_ranges. If sel_arg is an infinite
    range, e.g. (x < 5 or x > 4), then skip it and do not construct
    a quick range.

  RETURN
7801 7802
    FALSE on success
    TRUE  otherwise
7803 7804 7805 7806 7807 7808 7809 7810 7811
*/

bool QUICK_GROUP_MIN_MAX_SELECT::add_range(SEL_ARG *sel_range)
{
  QUICK_RANGE *range;
  uint range_flag= sel_range->min_flag | sel_range->max_flag;

  /* Skip (-inf,+inf) ranges, e.g. (x < 5 or x > 4). */
  if((range_flag & NO_MIN_RANGE) && (range_flag & NO_MAX_RANGE))
7812
    return FALSE;
7813 7814 7815 7816 7817 7818 7819 7820 7821 7822 7823 7824 7825 7826 7827

  if (!(sel_range->min_flag & NO_MIN_RANGE) &&
      !(sel_range->max_flag & NO_MAX_RANGE))
  {
    if (sel_range->maybe_null &&
        sel_range->min_value[0] && sel_range->max_value[0])
      range_flag|= NULL_RANGE; /* IS NULL condition */
    else if (memcmp(sel_range->min_value, sel_range->max_value,
                    min_max_arg_len) == 0)
      range_flag|= EQ_RANGE;  /* equality condition */
  }
  range= new QUICK_RANGE(sel_range->min_value, min_max_arg_len,
                         sel_range->max_value, min_max_arg_len,
                         range_flag);
  if (!range)
7828
    return TRUE;
7829
  if (insert_dynamic(&min_max_ranges, (gptr)&range))
7830 7831
    return TRUE;
  return FALSE;
7832 7833 7834 7835 7836 7837 7838 7839 7840 7841 7842 7843 7844 7845 7846 7847 7848 7849 7850 7851 7852 7853 7854 7855 7856 7857 7858 7859 7860
}


/*
  Determine the total number and length of the keys that will be used for
  index lookup.

  SYNOPSIS
    QUICK_GROUP_MIN_MAX_SELECT::update_key_stat()

  DESCRIPTION
    The total length of the keys used for index lookup depends on whether
    there are any predicates referencing the min/max argument, and/or if
    the min/max argument field can be NULL.
    This function does an optimistic analysis whether the search key might
    be extended by a constant for the min/max keypart. It is 'optimistic'
    because during actual execution it may happen that a particular range
    is skipped, and then a shorter key will be used. However this is data
    dependent and can't be easily estimated here.

  RETURN
    None
*/

void QUICK_GROUP_MIN_MAX_SELECT::update_key_stat()
{
  max_used_key_length= real_prefix_len;
  if (min_max_ranges.elements > 0)
  {
igor@rurik.mysql.com's avatar
igor@rurik.mysql.com committed
7861
    QUICK_RANGE *cur_range= 0;
7862 7863 7864 7865 7866 7867 7868 7869 7870 7871 7872 7873 7874 7875 7876 7877 7878 7879 7880 7881 7882 7883 7884 7885 7886 7887 7888 7889 7890 7891 7892 7893 7894 7895 7896 7897
    if (have_min)
    { /* Check if the right-most range has a lower boundary. */
      get_dynamic(&min_max_ranges, (gptr)&cur_range,
                  min_max_ranges.elements - 1);
      if (!(cur_range->flag & NO_MIN_RANGE))
      {
        max_used_key_length+= min_max_arg_len;
        ++used_key_parts;
        return;
      }
    }
    if (have_max)
    { /* Check if the left-most range has an upper boundary. */
      get_dynamic(&min_max_ranges, (gptr)&cur_range, 0);
      if (!(cur_range->flag & NO_MAX_RANGE))
      {
        max_used_key_length+= min_max_arg_len;
        ++used_key_parts;
        return;
      }
    }
  }
  else if (have_min && min_max_arg_part && min_max_arg_part->field->is_null())
  {
    max_used_key_length+= min_max_arg_len;
    ++used_key_parts;
  }
}


/*
  Initialize a quick group min/max select for key retrieval.

  SYNOPSIS
    QUICK_GROUP_MIN_MAX_SELECT::reset()

7898 7899 7900 7901
  DESCRIPTION
    Initialize the index chosen for access and find and store the prefix
    of the last group. The method is expensive since it performs disk access.

7902 7903 7904 7905 7906 7907 7908 7909 7910 7911 7912 7913 7914 7915 7916
  RETURN
    0      OK
    other  Error code
*/

int QUICK_GROUP_MIN_MAX_SELECT::reset(void)
{
  int result;
  DBUG_ENTER("QUICK_GROUP_MIN_MAX_SELECT::reset");

  file->extra(HA_EXTRA_KEYREAD); /* We need only the key attributes */
  result= file->ha_index_init(index);
  result= file->index_last(record);
  if (result)
    DBUG_RETURN(result);
sergefp@mysql.com's avatar
sergefp@mysql.com committed
7917 7918
  if (quick_prefix_select && quick_prefix_select->reset())
    DBUG_RETURN(1);
7919 7920 7921 7922 7923 7924 7925 7926 7927 7928 7929 7930 7931 7932 7933 7934 7935 7936 7937 7938 7939 7940 7941 7942 7943 7944 7945 7946 7947 7948 7949 7950 7951 7952 7953 7954 7955 7956 7957
  /* Save the prefix of the last group. */
  key_copy(last_prefix, record, index_info, group_prefix_len);

  DBUG_RETURN(0);
}



/* 
  Get the next key containing the MIN and/or MAX key for the next group.

  SYNOPSIS
    QUICK_GROUP_MIN_MAX_SELECT::get_next()

  DESCRIPTION
    The method finds the next subsequent group of records that satisfies the
    query conditions and finds the keys that contain the MIN/MAX values for
    the key part referenced by the MIN/MAX function(s). Once a group and its
    MIN/MAX values are found, store these values in the Item_sum objects for
    the MIN/MAX functions. The rest of the values in the result row are stored
    in the Item_field::result_field of each select field. If the query does
    not contain MIN and/or MAX functions, then the function only finds the
    group prefix, which is a query answer itself.

  NOTES
    If both MIN and MAX are computed, then we use the fact that if there is
    no MIN key, there can't be a MAX key as well, so we can skip looking
    for a MAX key in this case.

  RETURN
    0                  on success
    HA_ERR_END_OF_FILE if returned all keys
    other              if some error occurred
*/

int QUICK_GROUP_MIN_MAX_SELECT::get_next()
{
  int min_res= 0;
  int max_res= 0;
timour@mysql.com's avatar
timour@mysql.com committed
7958 7959 7960 7961 7962 7963 7964
#ifdef HPUX11
  /*
    volatile is required by a bug in the HP compiler due to which the
    last test of result fails.
  */
  volatile int result;
#else
7965
  int result;
timour@mysql.com's avatar
timour@mysql.com committed
7966
#endif
7967 7968 7969 7970 7971 7972 7973 7974 7975 7976 7977 7978 7979 7980 7981 7982 7983 7984 7985 7986 7987 7988 7989 7990 7991 7992 7993 7994 7995 7996 7997 7998 7999 8000 8001 8002 8003 8004 8005 8006 8007 8008 8009 8010 8011 8012 8013 8014 8015 8016 8017 8018 8019 8020 8021 8022 8023 8024 8025 8026 8027 8028 8029 8030 8031 8032 8033 8034 8035 8036 8037 8038 8039 8040 8041 8042 8043 8044 8045 8046 8047 8048 8049 8050 8051 8052 8053 8054 8055 8056 8057 8058 8059 8060 8061 8062 8063 8064 8065 8066 8067 8068 8069 8070 8071 8072 8073 8074 8075 8076 8077 8078 8079 8080 8081 8082 8083 8084 8085 8086 8087 8088 8089 8090 8091 8092 8093 8094 8095 8096 8097 8098 8099 8100 8101 8102 8103 8104 8105 8106 8107 8108 8109 8110 8111 8112 8113 8114 8115 8116 8117 8118 8119 8120 8121 8122 8123 8124 8125 8126 8127 8128 8129 8130 8131 8132 8133 8134 8135 8136 8137 8138 8139 8140 8141 8142 8143 8144 8145 8146 8147 8148 8149 8150 8151 8152 8153 8154 8155 8156 8157 8158 8159 8160 8161 8162 8163 8164 8165 8166 8167 8168 8169 8170 8171 8172 8173 8174 8175 8176 8177 8178 8179 8180 8181 8182 8183 8184 8185 8186 8187 8188 8189 8190 8191 8192 8193 8194 8195 8196 8197 8198 8199 8200 8201 8202 8203 8204 8205 8206 8207 8208 8209 8210 8211 8212 8213 8214 8215 8216 8217 8218 8219 8220 8221 8222 8223 8224 8225 8226 8227 8228 8229 8230 8231 8232 8233 8234 8235 8236 8237 8238 8239 8240 8241 8242 8243 8244 8245 8246
  int is_last_prefix;

  DBUG_ENTER("QUICK_GROUP_MIN_MAX_SELECT::get_next");

  /*
    Loop until a group is found that satisfies all query conditions or the last
    group is reached.
  */
  do
  {
    result= next_prefix();
    /*
      Check if this is the last group prefix. Notice that at this point
      this->record contains the current prefix in record format.
    */
    is_last_prefix= key_cmp(index_info->key_part, last_prefix,
                            group_prefix_len);
    DBUG_ASSERT(is_last_prefix <= 0);
    if (result == HA_ERR_KEY_NOT_FOUND)
      continue;
    else if (result)
      break;

    if (have_min)
    {
      min_res= next_min();
      if (min_res == 0)
        update_min_result();
    }
    /* If there is no MIN in the group, there is no MAX either. */
    if ((have_max && !have_min) ||
        (have_max && have_min && (min_res == 0)))
    {
      max_res= next_max();
      if (max_res == 0)
        update_max_result();
      /* If a MIN was found, a MAX must have been found as well. */
      DBUG_ASSERT((have_max && !have_min) ||
                  (have_max && have_min && (max_res == 0)));
    }
    result= have_min ? min_res : have_max ? max_res : result;
  }
  while (result == HA_ERR_KEY_NOT_FOUND && is_last_prefix != 0);

  if (result == 0)
    /*
      Partially mimic the behavior of end_select_send. Copy the
      field data from Item_field::field into Item_field::result_field
      of each non-aggregated field (the group fields, and optionally
      other fields in non-ANSI SQL mode).
    */
    copy_fields(&join->tmp_table_param);
  else if (result == HA_ERR_KEY_NOT_FOUND)
    result= HA_ERR_END_OF_FILE;

  DBUG_RETURN(result);
}


/*
  Retrieve the minimal key in the next group.

  SYNOPSIS
    QUICK_GROUP_MIN_MAX_SELECT::next_min()

  DESCRIPTION
    Load the prefix of the next group into group_prefix and find the minimal
    key within this group such that the key satisfies the query conditions and
    NULL semantics. The found key is loaded into this->record.

  IMPLEMENTATION
    Depending on the values of min_max_ranges.elements, key_infix_len, and
    whether there is a  NULL in the MIN field, this function may directly
    return without any data access. In this case we use the key loaded into
    this->record by the call to this->next_prefix() just before this call.

  RETURN
    0                    on success
    HA_ERR_KEY_NOT_FOUND if no MIN key was found that fulfills all conditions.
    other                if some error occurred
*/

int QUICK_GROUP_MIN_MAX_SELECT::next_min()
{
  int result= 0;
  DBUG_ENTER("QUICK_GROUP_MIN_MAX_SELECT::next_min");

  /* Find the MIN key using the eventually extended group prefix. */
  if (min_max_ranges.elements > 0)
  {
    if ((result= next_min_in_range()))
      DBUG_RETURN(result);
  }
  else
  {
    /* Apply the constant equality conditions to the non-group select fields. */
    if (key_infix_len > 0)
    {
      if ((result= file->index_read(record, group_prefix, real_prefix_len,
                                    HA_READ_KEY_EXACT)))
        DBUG_RETURN(result);
    }

    /*
      If the min/max argument field is NULL, skip subsequent rows in the same
      group with NULL in it. Notice that:
      - if the first row in a group doesn't have a NULL in the field, no row
      in the same group has (because NULL < any other value),
      - min_max_arg_part->field->ptr points to some place in 'record'.
    */
    if (min_max_arg_part && min_max_arg_part->field->is_null())
    {
      /* Find the first subsequent record without NULL in the MIN/MAX field. */
      key_copy(tmp_record, record, index_info, 0);
      result= file->index_read(record, tmp_record,
                               real_prefix_len + min_max_arg_len,
                               HA_READ_AFTER_KEY);
      /*
        Check if the new record belongs to the current group by comparing its
        prefix with the group's prefix. If it is from the next group, then the
        whole group has NULLs in the MIN/MAX field, so use the first record in
        the group as a result.
        TODO:
        It is possible to reuse this new record as the result candidate for the
        next call to next_min(), and to save one lookup in the next call. For
        this add a new member 'this->next_group_prefix'.
      */
      if (!result)
      {
        if(key_cmp(index_info->key_part, group_prefix, real_prefix_len))
          key_restore(record, tmp_record, index_info, 0);
      } else if (result == HA_ERR_KEY_NOT_FOUND) 
        result= 0; /* There is a result in any case. */
    }
  }

  /*
    If the MIN attribute is non-nullable, this->record already contains the
    MIN key in the group, so just return.
  */
  DBUG_RETURN(result);
}


/* 
  Retrieve the maximal key in the next group.

  SYNOPSIS
    QUICK_GROUP_MIN_MAX_SELECT::next_max()

  DESCRIPTION
    If there was no previous next_min call to determine the next group prefix,
    then load the next prefix into group_prefix, then lookup the maximal key of
    the group, and store it into this->record.

  RETURN
    0                    on success
    HA_ERR_KEY_NOT_FOUND if no MAX key was found that fulfills all conditions.
    other                if some error occurred
*/

int QUICK_GROUP_MIN_MAX_SELECT::next_max()
{
  int result;

  DBUG_ENTER("QUICK_GROUP_MIN_MAX_SELECT::next_max");

  /* Get the last key in the (possibly extended) group. */
  if (min_max_ranges.elements > 0)
    result= next_max_in_range();
  else
    result= file->index_read(record, group_prefix, real_prefix_len,
                             HA_READ_PREFIX_LAST);
  DBUG_RETURN(result);
}


/*
  Determine the prefix of the next group.

  SYNOPSIS
    QUICK_GROUP_MIN_MAX_SELECT::next_prefix()

  DESCRIPTION
    Determine the prefix of the next group that satisfies the query conditions.
    If there is a range condition referencing the group attributes, use a
    QUICK_RANGE_SELECT object to retrieve the *first* key that satisfies the
    condition. If there is a key infix of constants, append this infix
    immediately after the group attributes. The possibly extended prefix is
    stored in this->group_prefix. The first key of the found group is stored in
    this->record, on which relies this->next_min().

  RETURN
    0                    on success
    HA_ERR_KEY_NOT_FOUND if there is no key with the formed prefix
    HA_ERR_END_OF_FILE   if there are no more keys
    other                if some error occurred
*/
int QUICK_GROUP_MIN_MAX_SELECT::next_prefix()
{
  int result;
  DBUG_ENTER("QUICK_GROUP_MIN_MAX_SELECT::next_prefix");

  if (quick_prefix_select)
  {
    byte *cur_prefix= seen_first_key ? group_prefix : NULL;
    if ((result= quick_prefix_select->get_next_prefix(group_prefix_len,
                                                      cur_prefix)))
      DBUG_RETURN(result);
    seen_first_key= TRUE;
  }
  else
  {
    if (!seen_first_key)
    {
      result= file->index_first(record);
      if (result)
        DBUG_RETURN(result);
      seen_first_key= TRUE;
    }
    else
    {
      /* Load the first key in this group into record. */
      result= file->index_read(record, group_prefix, group_prefix_len,
                               HA_READ_AFTER_KEY);
      if (result)
        DBUG_RETURN(result);
    }
  }

  /* Save the prefix of this group for subsequent calls. */
  key_copy(group_prefix, record, index_info, group_prefix_len);
  /* Append key_infix to group_prefix. */
  if (key_infix_len > 0)
    memcpy(group_prefix + group_prefix_len,
           key_infix, key_infix_len);

  DBUG_RETURN(0);
}


/*
  Find the minimal key in a group that satisfies some range conditions for the
  min/max argument field.

  SYNOPSIS
    QUICK_GROUP_MIN_MAX_SELECT::next_min_in_range()

  DESCRIPTION
    Given the sequence of ranges min_max_ranges, find the minimal key that is
    in the left-most possible range. If there is no such key, then the current
    group does not have a MIN key that satisfies the WHERE clause. If a key is
    found, its value is stored in this->record.

  RETURN
    0                    on success
    HA_ERR_KEY_NOT_FOUND if there is no key with the given prefix in any of
                         the ranges
    other                if some error
*/

int QUICK_GROUP_MIN_MAX_SELECT::next_min_in_range()
{
  ha_rkey_function find_flag;
  uint search_prefix_len;
  QUICK_RANGE *cur_range;
  bool found_null= FALSE;
  int result= HA_ERR_KEY_NOT_FOUND;

  DBUG_ASSERT(min_max_ranges.elements > 0);

  for (uint range_idx= 0; range_idx < min_max_ranges.elements; range_idx++)
  { /* Search from the left-most range to the right. */
    get_dynamic(&min_max_ranges, (gptr)&cur_range, range_idx);

    /*
      If the current value for the min/max argument is bigger than the right
      boundary of cur_range, there is no need to check this range.
    */
    if (range_idx != 0 && !(cur_range->flag & NO_MAX_RANGE) &&
8247
        (key_cmp(min_max_arg_part, (const byte*) cur_range->max_key,
8248
                 min_max_arg_len) == 1))
8249 8250 8251 8252 8253 8254 8255 8256 8257 8258 8259 8260 8261 8262 8263 8264 8265 8266 8267 8268 8269 8270 8271 8272
      continue;

    if (cur_range->flag & NO_MIN_RANGE)
    {
      find_flag= HA_READ_KEY_EXACT;
      search_prefix_len= real_prefix_len;
    }
    else
    {
      /* Extend the search key with the lower boundary for this range. */
      memcpy(group_prefix + real_prefix_len, cur_range->min_key,
             cur_range->min_length);
      search_prefix_len= real_prefix_len + min_max_arg_len;
      find_flag= (cur_range->flag & (EQ_RANGE | NULL_RANGE)) ?
                 HA_READ_KEY_EXACT : (cur_range->flag & NEAR_MIN) ?
                 HA_READ_AFTER_KEY : HA_READ_KEY_OR_NEXT;
    }

    result= file->index_read(record, group_prefix, search_prefix_len,
                             find_flag);
    if ((result == HA_ERR_KEY_NOT_FOUND) &&
        (cur_range->flag & (EQ_RANGE | NULL_RANGE)))
        continue; /* Check the next range. */
    else if (result)
8273 8274 8275 8276 8277 8278
    {
      /*
        In all other cases (HA_ERR_*, HA_READ_KEY_EXACT with NO_MIN_RANGE,
        HA_READ_AFTER_KEY, HA_READ_KEY_OR_NEXT) if the lookup failed for this
        range, it can't succeed for any other subsequent range.
      */
8279
      break;
8280
    }
8281 8282 8283 8284 8285 8286

    /* A key was found. */
    if (cur_range->flag & EQ_RANGE)
      break; /* No need to perform the checks below for equal keys. */

    if (cur_range->flag & NULL_RANGE)
8287 8288 8289 8290 8291 8292
    {
      /*
        Remember this key, and continue looking for a non-NULL key that
        satisfies some other condition.
      */
      memcpy(tmp_record, record, head->s->rec_buff_length);
8293 8294 8295 8296 8297 8298 8299 8300 8301 8302 8303 8304 8305 8306 8307 8308 8309 8310 8311 8312 8313 8314 8315 8316 8317 8318 8319 8320 8321 8322 8323 8324 8325 8326 8327 8328 8329 8330 8331 8332
      found_null= TRUE;
      continue;
    }

    /* Check if record belongs to the current group. */
    if (key_cmp(index_info->key_part, group_prefix, real_prefix_len))
    {
      result = HA_ERR_KEY_NOT_FOUND;
      continue;
    }

    /* If there is an upper limit, check if the found key is in the range. */
    if ( !(cur_range->flag & NO_MAX_RANGE) )
    {
      /* Compose the MAX key for the range. */
      byte *max_key= (byte*) my_alloca(real_prefix_len + min_max_arg_len);
      memcpy(max_key, group_prefix, real_prefix_len);
      memcpy(max_key + real_prefix_len, cur_range->max_key,
             cur_range->max_length);
      /* Compare the found key with max_key. */
      int cmp_res= key_cmp(index_info->key_part, max_key,
                           real_prefix_len + min_max_arg_len);
      if (!((cur_range->flag & NEAR_MAX) && (cmp_res == -1) ||
            (cmp_res <= 0)))
      {
        result = HA_ERR_KEY_NOT_FOUND;
        continue;
      }
    }
    /* If we got to this point, the current key qualifies as MIN. */
    DBUG_ASSERT(result == 0);
    break;
  }
  /*
    If there was a key with NULL in the MIN/MAX field, and there was no other
    key without NULL from the same group that satisfies some other condition,
    then use the key with the NULL.
  */
  if (found_null && result)
  {
8333
    memcpy(record, tmp_record, head->s->rec_buff_length);
8334 8335 8336 8337 8338 8339 8340 8341 8342 8343 8344 8345 8346 8347 8348 8349 8350 8351 8352 8353 8354 8355 8356 8357 8358 8359 8360 8361 8362 8363 8364 8365 8366 8367 8368 8369 8370 8371 8372 8373 8374 8375 8376 8377 8378
    result= 0;
  }
  return result;
}


/*
  Find the maximal key in a group that satisfies some range conditions for the
  min/max argument field.

  SYNOPSIS
    QUICK_GROUP_MIN_MAX_SELECT::next_max_in_range()

  DESCRIPTION
    Given the sequence of ranges min_max_ranges, find the maximal key that is
    in the right-most possible range. If there is no such key, then the current
    group does not have a MAX key that satisfies the WHERE clause. If a key is
    found, its value is stored in this->record.

  RETURN
    0                    on success
    HA_ERR_KEY_NOT_FOUND if there is no key with the given prefix in any of
                         the ranges
    other                if some error
*/

int QUICK_GROUP_MIN_MAX_SELECT::next_max_in_range()
{
  ha_rkey_function find_flag;
  uint search_prefix_len;
  QUICK_RANGE *cur_range;
  int result;

  DBUG_ASSERT(min_max_ranges.elements > 0);

  for (uint range_idx= min_max_ranges.elements; range_idx > 0; range_idx--)
  { /* Search from the right-most range to the left. */
    get_dynamic(&min_max_ranges, (gptr)&cur_range, range_idx - 1);

    /*
      If the current value for the min/max argument is smaller than the left
      boundary of cur_range, there is no need to check this range.
    */
    if (range_idx != min_max_ranges.elements &&
        !(cur_range->flag & NO_MIN_RANGE) &&
8379
        (key_cmp(min_max_arg_part, (const byte*) cur_range->min_key,
8380
                 min_max_arg_len) == -1))
8381 8382 8383 8384 8385 8386 8387 8388 8389 8390 8391 8392 8393 8394 8395 8396 8397 8398 8399 8400 8401 8402 8403 8404 8405 8406 8407 8408 8409 8410 8411 8412 8413 8414 8415 8416 8417 8418 8419 8420 8421 8422 8423 8424 8425 8426 8427 8428 8429 8430 8431 8432 8433 8434 8435 8436 8437 8438 8439 8440 8441 8442 8443 8444 8445 8446 8447 8448 8449 8450 8451 8452 8453 8454 8455 8456 8457 8458 8459 8460 8461 8462 8463 8464 8465 8466 8467 8468 8469 8470 8471 8472 8473 8474 8475 8476 8477 8478 8479 8480 8481 8482 8483 8484 8485 8486 8487 8488 8489 8490 8491 8492 8493 8494 8495 8496 8497 8498 8499 8500 8501 8502 8503 8504 8505 8506 8507 8508
      continue;

    if (cur_range->flag & NO_MAX_RANGE)
    {
      find_flag= HA_READ_PREFIX_LAST;
      search_prefix_len= real_prefix_len;
    }
    else
    {
      /* Extend the search key with the upper boundary for this range. */
      memcpy(group_prefix + real_prefix_len, cur_range->max_key,
             cur_range->max_length);
      search_prefix_len= real_prefix_len + min_max_arg_len;
      find_flag= (cur_range->flag & EQ_RANGE) ?
                 HA_READ_KEY_EXACT : (cur_range->flag & NEAR_MAX) ?
                 HA_READ_BEFORE_KEY : HA_READ_PREFIX_LAST_OR_PREV;
    }

    result= file->index_read(record, group_prefix, search_prefix_len,
                             find_flag);

    if ((result == HA_ERR_KEY_NOT_FOUND) && (cur_range->flag & EQ_RANGE))
      continue; /* Check the next range. */
    else if (result)
      /*
        In no key was found with this upper bound, there certainly are no keys
        in the ranges to the left.
      */
      return result;

    /* A key was found. */
    if (cur_range->flag & EQ_RANGE)
      return result; /* No need to perform the checks below for equal keys. */

    /* Check if record belongs to the current group. */
    if (key_cmp(index_info->key_part, group_prefix, real_prefix_len))
    {
      result = HA_ERR_KEY_NOT_FOUND;
      continue;
    }

    /* If there is a lower limit, check if the found key is in the range. */
    if ( !(cur_range->flag & NO_MIN_RANGE) )
    {
      /* Compose the MIN key for the range. */
      byte *min_key= (byte*) my_alloca(real_prefix_len + min_max_arg_len);
      memcpy(min_key, group_prefix, real_prefix_len);
      memcpy(min_key + real_prefix_len, cur_range->min_key,
             cur_range->min_length);
      /* Compare the found key with min_key. */
      int cmp_res= key_cmp(index_info->key_part, min_key,
                           real_prefix_len + min_max_arg_len);
      if (!((cur_range->flag & NEAR_MIN) && (cmp_res == 1) ||
            (cmp_res >= 0)))
        continue;
    }
    /* If we got to this point, the current key qualifies as MAX. */
    return result;
  }
  return HA_ERR_KEY_NOT_FOUND;
}


/*
  Update all MIN function results with the newly found value.

  SYNOPSIS
    QUICK_GROUP_MIN_MAX_SELECT::update_min_result()

  DESCRIPTION
    The method iterates through all MIN functions and updates the result value
    of each function by calling Item_sum::reset(), which in turn picks the new
    result value from this->head->record[0], previously updated by
    next_min(). The updated value is stored in a member variable of each of the
    Item_sum objects, depending on the value type.

  IMPLEMENTATION
    The update must be done separately for MIN and MAX, immediately after
    next_min() was called and before next_max() is called, because both MIN and
    MAX take their result value from the same buffer this->head->record[0]
    (i.e.  this->record).

  RETURN
    None
*/

void QUICK_GROUP_MIN_MAX_SELECT::update_min_result()
{
  Item_sum *min_func;

  min_functions_it->rewind();
  while ((min_func= (*min_functions_it)++))
    min_func->reset();
}


/*
  Update all MAX function results with the newly found value.

  SYNOPSIS
    QUICK_GROUP_MIN_MAX_SELECT::update_max_result()

  DESCRIPTION
    The method iterates through all MAX functions and updates the result value
    of each function by calling Item_sum::reset(), which in turn picks the new
    result value from this->head->record[0], previously updated by
    next_max(). The updated value is stored in a member variable of each of the
    Item_sum objects, depending on the value type.

  IMPLEMENTATION
    The update must be done separately for MIN and MAX, immediately after
    next_max() was called, because both MIN and MAX take their result value
    from the same buffer this->head->record[0] (i.e.  this->record).

  RETURN
    None
*/

void QUICK_GROUP_MIN_MAX_SELECT::update_max_result()
{
  Item_sum *max_func;

  max_functions_it->rewind();
  while ((max_func= (*max_functions_it)++))
    max_func->reset();
}


8509 8510 8511 8512 8513 8514 8515 8516 8517 8518 8519 8520 8521 8522 8523
/*
  Append comma-separated list of keys this quick select uses to key_names;
  append comma-separated list of corresponding used lengths to used_lengths.

  SYNOPSIS
    QUICK_GROUP_MIN_MAX_SELECT::add_keys_and_lengths()
    key_names    [out] Names of used indexes
    used_lengths [out] Corresponding lengths of the index names

  DESCRIPTION
    This method is used by select_describe to extract the names of the
    indexes used by a quick select.

*/

8524 8525 8526 8527 8528 8529 8530 8531 8532 8533 8534
void QUICK_GROUP_MIN_MAX_SELECT::add_keys_and_lengths(String *key_names,
                                                      String *used_lengths)
{
  char buf[64];
  uint length;
  key_names->append(index_info->name);
  length= longlong2str(max_used_key_length, buf, 10) - buf;
  used_lengths->append(buf, length);
}


8535
#ifndef DBUG_OFF
8536

8537 8538 8539 8540 8541 8542 8543 8544 8545
static void print_sel_tree(PARAM *param, SEL_TREE *tree, key_map *tree_map,
                           const char *msg)
{
  SEL_ARG **key,**end;
  int idx;
  char buff[1024];
  DBUG_ENTER("print_sel_tree");
  if (! _db_on_)
    DBUG_VOID_RETURN;
8546

8547 8548 8549 8550 8551 8552 8553 8554 8555 8556 8557 8558 8559 8560 8561 8562
  String tmp(buff,sizeof(buff),&my_charset_bin);
  tmp.length(0);
  for (idx= 0,key=tree->keys, end=key+param->keys ;
       key != end ;
       key++,idx++)
  {
    if (tree_map->is_set(idx))
    {
      uint keynr= param->real_keynr[idx];
      if (tmp.length())
        tmp.append(',');
      tmp.append(param->table->key_info[keynr].name);
    }
  }
  if (!tmp.length())
    tmp.append("(empty)");
8563

8564
  DBUG_PRINT("info", ("SEL_TREE %p (%s) scans:%s", tree, msg, tmp.ptr()));
8565

8566 8567
  DBUG_VOID_RETURN;
}
8568

8569 8570 8571 8572

static void print_ror_scans_arr(TABLE *table, const char *msg,
                                struct st_ror_scan_info **start,
                                struct st_ror_scan_info **end)
8573
{
8574 8575 8576 8577 8578 8579 8580 8581
  DBUG_ENTER("print_ror_scans");
  if (! _db_on_)
    DBUG_VOID_RETURN;

  char buff[1024];
  String tmp(buff,sizeof(buff),&my_charset_bin);
  tmp.length(0);
  for(;start != end; start++)
8582
  {
8583 8584 8585
    if (tmp.length())
      tmp.append(',');
    tmp.append(table->key_info[(*start)->keynr].name);
8586
  }
8587 8588 8589 8590
  if (!tmp.length())
    tmp.append("(empty)");
  DBUG_PRINT("info", ("ROR key scans (%s): %s", msg, tmp.ptr()));
  DBUG_VOID_RETURN;
8591 8592 8593
}


bk@work.mysql.com's avatar
bk@work.mysql.com committed
8594 8595 8596 8597 8598 8599 8600 8601 8602 8603 8604
/*****************************************************************************
** Print a quick range for debugging
** TODO:
** This should be changed to use a String to store each row instead
** of locking the DEBUG stream !
*****************************************************************************/

static void
print_key(KEY_PART *key_part,const char *key,uint used_length)
{
  char buff[1024];
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
8605
  const char *key_end= key+used_length;
8606
  String tmp(buff,sizeof(buff),&my_charset_bin);
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
8607
  uint store_length;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
8608

pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
8609
  for (; key < key_end; key+=store_length, key_part++)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
8610
  {
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
8611 8612 8613
    Field *field=      key_part->field;
    store_length= key_part->store_length;

bk@work.mysql.com's avatar
bk@work.mysql.com committed
8614 8615
    if (field->real_maybe_null())
    {
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
8616
      if (*key)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
8617 8618 8619 8620
      {
	fwrite("NULL",sizeof(char),4,DBUG_FILE);
	continue;
      }
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
8621 8622
      key++;					// Skip null byte
      store_length--;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
8623
    }
8624
    field->set_key_image((char*) key, key_part->length);
monty@mysql.com's avatar
monty@mysql.com committed
8625 8626 8627 8628
    if (field->type() == MYSQL_TYPE_BIT)
      (void) field->val_int_as_str(&tmp, 1);
    else
      field->val_str(&tmp);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
8629
    fwrite(tmp.ptr(),sizeof(char),tmp.length(),DBUG_FILE);
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
8630 8631
    if (key+store_length < key_end)
      fputc('/',DBUG_FILE);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
8632 8633 8634
  }
}

pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
8635

8636
static void print_quick(QUICK_SELECT_I *quick, const key_map *needed_reg)
8637
{
8638
  char buf[MAX_KEY/8+1];
8639
  DBUG_ENTER("print_quick");
8640 8641
  if (! _db_on_ || !quick)
    DBUG_VOID_RETURN;
8642
  DBUG_LOCK_FILE;
8643

monty@mysql.com's avatar
monty@mysql.com committed
8644
  quick->dbug_dump(0, TRUE);
8645
  fprintf(DBUG_FILE,"other_keys: 0x%s:\n", needed_reg->print(buf));
8646

8647
  DBUG_UNLOCK_FILE;
8648 8649 8650
  DBUG_VOID_RETURN;
}

pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
8651

8652
static void print_rowid(byte* val, int len)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
8653
{
8654
  byte *pb;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
8655
  DBUG_LOCK_FILE;
8656 8657 8658 8659 8660 8661 8662 8663 8664 8665
  fputc('\"', DBUG_FILE);
  for (pb= val; pb!= val + len; ++pb)
    fprintf(DBUG_FILE, "%c", *pb);
  fprintf(DBUG_FILE, "\", hex: ");

  for (pb= val; pb!= val + len; ++pb)
    fprintf(DBUG_FILE, "%x ", *pb);
  fputc('\n', DBUG_FILE);
  DBUG_UNLOCK_FILE;
}
8666

8667 8668 8669 8670
void QUICK_RANGE_SELECT::dbug_dump(int indent, bool verbose)
{
  fprintf(DBUG_FILE, "%*squick range select, key %s, length: %d\n",
	  indent, "", head->key_info[index].name, max_used_key_length);
8671

8672
  if (verbose)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
8673
  {
8674 8675
    QUICK_RANGE *range;
    QUICK_RANGE **pr= (QUICK_RANGE**)ranges.buffer;
8676
    QUICK_RANGE **last_range= pr + ranges.elements;
8677
    for (; pr!=last_range; ++pr)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
8678
    {
8679 8680 8681 8682 8683 8684 8685 8686 8687 8688 8689
      fprintf(DBUG_FILE, "%*s", indent + 2, "");
      range= *pr;
      if (!(range->flag & NO_MIN_RANGE))
      {
        print_key(key_parts,range->min_key,range->min_length);
        if (range->flag & NEAR_MIN)
	  fputs(" < ",DBUG_FILE);
        else
	  fputs(" <= ",DBUG_FILE);
      }
      fputs("X",DBUG_FILE);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
8690

8691 8692 8693 8694 8695 8696 8697 8698 8699
      if (!(range->flag & NO_MAX_RANGE))
      {
        if (range->flag & NEAR_MAX)
	  fputs(" < ",DBUG_FILE);
        else
	  fputs(" <= ",DBUG_FILE);
        print_key(key_parts,range->max_key,range->max_length);
      }
      fputs("\n",DBUG_FILE);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
8700 8701
    }
  }
8702 8703 8704 8705 8706 8707 8708 8709 8710 8711 8712 8713
}

void QUICK_INDEX_MERGE_SELECT::dbug_dump(int indent, bool verbose)
{
  List_iterator_fast<QUICK_RANGE_SELECT> it(quick_selects);
  QUICK_RANGE_SELECT *quick;
  fprintf(DBUG_FILE, "%*squick index_merge select\n", indent, "");
  fprintf(DBUG_FILE, "%*smerged scans {\n", indent, "");
  while ((quick= it++))
    quick->dbug_dump(indent+2, verbose);
  if (pk_quick_select)
  {
8714
    fprintf(DBUG_FILE, "%*sclustered PK quick:\n", indent, "");
8715 8716 8717 8718 8719 8720 8721 8722 8723
    pk_quick_select->dbug_dump(indent+2, verbose);
  }
  fprintf(DBUG_FILE, "%*s}\n", indent, "");
}

void QUICK_ROR_INTERSECT_SELECT::dbug_dump(int indent, bool verbose)
{
  List_iterator_fast<QUICK_RANGE_SELECT> it(quick_selects);
  QUICK_RANGE_SELECT *quick;
8724
  fprintf(DBUG_FILE, "%*squick ROR-intersect select, %scovering\n",
8725 8726 8727
          indent, "", need_to_fetch_row? "":"non-");
  fprintf(DBUG_FILE, "%*smerged scans {\n", indent, "");
  while ((quick= it++))
8728
    quick->dbug_dump(indent+2, verbose);
8729 8730
  if (cpk_quick)
  {
8731
    fprintf(DBUG_FILE, "%*sclustered PK quick:\n", indent, "");
8732 8733 8734 8735 8736 8737 8738 8739 8740 8741 8742 8743 8744 8745
    cpk_quick->dbug_dump(indent+2, verbose);
  }
  fprintf(DBUG_FILE, "%*s}\n", indent, "");
}

void QUICK_ROR_UNION_SELECT::dbug_dump(int indent, bool verbose)
{
  List_iterator_fast<QUICK_SELECT_I> it(quick_selects);
  QUICK_SELECT_I *quick;
  fprintf(DBUG_FILE, "%*squick ROR-union select\n", indent, "");
  fprintf(DBUG_FILE, "%*smerged scans {\n", indent, "");
  while ((quick= it++))
    quick->dbug_dump(indent+2, verbose);
  fprintf(DBUG_FILE, "%*s}\n", indent, "");
bk@work.mysql.com's avatar
bk@work.mysql.com committed
8746 8747
}

8748 8749 8750 8751 8752 8753 8754 8755 8756 8757 8758 8759 8760 8761 8762 8763 8764 8765 8766 8767 8768 8769 8770 8771 8772 8773 8774 8775 8776 8777 8778 8779 8780 8781 8782 8783 8784 8785 8786 8787 8788 8789 8790 8791

/*
  Print quick select information to DBUG_FILE.

  SYNOPSIS
    QUICK_GROUP_MIN_MAX_SELECT::dbug_dump()
    indent  Indentation offset
    verbose If TRUE show more detailed output.

  DESCRIPTION
    Print the contents of this quick select to DBUG_FILE. The method also
    calls dbug_dump() for the used quick select if any.

  IMPLEMENTATION
    Caller is responsible for locking DBUG_FILE before this call and unlocking
    it afterwards.

  RETURN
    None
*/

void QUICK_GROUP_MIN_MAX_SELECT::dbug_dump(int indent, bool verbose)
{
  fprintf(DBUG_FILE,
          "%*squick_group_min_max_select: index %s (%d), length: %d\n",
	  indent, "", index_info->name, index, max_used_key_length);
  if (key_infix_len > 0)
  {
    fprintf(DBUG_FILE, "%*susing key_infix with length %d:\n",
            indent, "", key_infix_len);
  }
  if (quick_prefix_select)
  {
    fprintf(DBUG_FILE, "%*susing quick_range_select:\n", indent, "");
    quick_prefix_select->dbug_dump(indent + 2, verbose);
  }
  if (min_max_ranges.elements > 0)
  {
    fprintf(DBUG_FILE, "%*susing %d quick_ranges for MIN/MAX:\n",
            indent, "", min_max_ranges.elements);
  }
}


monty@mysql.com's avatar
monty@mysql.com committed
8792
#endif /* NOT_USED */
bk@work.mysql.com's avatar
bk@work.mysql.com committed
8793 8794

/*****************************************************************************
8795
** Instantiate templates
bk@work.mysql.com's avatar
bk@work.mysql.com committed
8796 8797 8798 8799 8800 8801
*****************************************************************************/

#ifdef __GNUC__
template class List<QUICK_RANGE>;
template class List_iterator<QUICK_RANGE>;
#endif