opt_range.cc 292 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.
*/

39
#ifdef USE_PRAGMA_IMPLEMENTATION
bk@work.mysql.com's avatar
bk@work.mysql.com committed
40 41 42 43 44 45 46 47 48 49 50 51
#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
static int sel_cmp(Field *f,char *a,char *b,uint8 a_flag,uint8 b_flag);

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

62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 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

/*
  A construction block of the SEL_ARG-graph.
  
  The following description only covers graphs of SEL_ARG objects with 
  sel_arg->type==KEY_RANGE:

  One SEL_ARG object represents an "elementary interval" in form
  
      min_value <=?  table.keypartX  <=? max_value
  
  The interval is a non-empty interval of any kind: with[out] minimum/maximum
  bound, [half]open/closed, single-point interval, etc.

  1. SEL_ARG GRAPH STRUCTURE
  
  SEL_ARG objects are linked together in a graph. The meaning of the graph
  is better demostrated by an example:
  
     tree->keys[i]
      | 
      |             $              $
      |    part=1   $     part=2   $    part=3
      |             $              $
      |  +-------+  $   +-------+  $   +--------+
      |  | kp1<1 |--$-->| kp2=5 |--$-->| kp3=10 |
      |  +-------+  $   +-------+  $   +--------+
      |      |      $              $       |
      |      |      $              $   +--------+
      |      |      $              $   | kp3=12 | 
      |      |      $              $   +--------+ 
      |  +-------+  $              $   
      \->| kp1=2 |--$--------------$-+ 
         +-------+  $              $ |   +--------+
             |      $              $  ==>| kp3=11 |
         +-------+  $              $ |   +--------+
         | kp1=3 |--$--------------$-+       |
         +-------+  $              $     +--------+
             |      $              $     | kp3=14 |
            ...     $              $     +--------+
 
  The entire graph is partitioned into "interval lists".

  An interval list is a sequence of ordered disjoint intervals over the same
  key part. SEL_ARG are linked via "next" and "prev" pointers. Additionally,
  all intervals in the list form an RB-tree, linked via left/right/parent 
  pointers. The RB-tree root SEL_ARG object will be further called "root of the
  interval list".
  
    In the example pic, there are 4 interval lists: 
    "kp<1 OR kp1=2 OR kp1=3", "kp2=5", "kp3=10 OR kp3=12", "kp3=11 OR kp3=13".
    The vertical lines represent SEL_ARG::next/prev pointers.
    
  In an interval list, each member X may have SEL_ARG::next_key_part pointer
  pointing to the root of another interval list Y. The pointed interval list
  must cover a key part with greater number (i.e. Y->part > X->part).
    
    In the example pic, the next_key_part pointers are represented by
    horisontal lines.

  2. SEL_ARG GRAPH SEMANTICS

  It represents a condition in a special form (we don't have a name for it ATM)
  The SEL_ARG::next/prev is "OR", and next_key_part is "AND".
  
  For example, the picture represents the condition in form:
   (kp1 < 1 AND kp2=5 AND (kp3=10 OR kp3=12)) OR 
   (kp1=2 AND (kp3=11 OR kp3=14)) OR 
   (kp1=3 AND (kp3=11 OR kp3=14))


  3. SEL_ARG GRAPH USE

  Use get_mm_tree() to construct SEL_ARG graph from WHERE condition.
  Then walk the SEL_ARG graph and get a list of dijsoint ordered key
  intervals (i.e. intervals in form
  
   (constA1, .., const1_K) < (keypart1,.., keypartK) < (constB1, .., constB_K)

  Those intervals can be used to access the index. The uses are in:
   - check_quick_select() - Walk the SEL_ARG graph and find an estimate of
                            how many table records are contained within all
                            intervals.
   - get_quick_select()   - Walk the SEL_ARG, materialize the key intervals,
                            and create QUICK_RANGE_SELECT object that will
                            read records within these intervals.
*/

bk@work.mysql.com's avatar
bk@work.mysql.com committed
150 151 152 153 154 155
class SEL_ARG :public Sql_alloc
{
public:
  uint8 min_flag,max_flag,maybe_flag;
  uint8 part;					// Which key part
  uint8 maybe_null;
156 157 158 159 160 161 162 163 164 165 166 167
  /* 
    Number of children of this element in the RB-tree, plus 1 for this
    element itself.
  */
  uint16 elements;
  /*
    Valid only for elements which are RB-tree roots: Number of times this
    RB-tree is referred to (it is referred by SEL_ARG::next_key_part or by
    SEL_TREE::keys[i] or by a temporary SEL_ARG* variable)
  */
  ulong use_count;

bk@work.mysql.com's avatar
bk@work.mysql.com committed
168 169 170
  Field *field;
  char *min_value,*max_value;			// Pointer to range

171 172 173 174
  SEL_ARG *left,*right;   /* R-B tree children */
  SEL_ARG *next,*prev;    /* Links for bi-directional interval list */
  SEL_ARG *parent;        /* R-B tree parent */
  SEL_ARG *next_key_part; 
bk@work.mysql.com's avatar
bk@work.mysql.com committed
175 176 177 178 179 180 181 182 183
  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)
184 185
    :min_flag(0),elements(1),use_count(1),left(0),next_key_part(0),
    color(BLACK), type(type_arg)
186
  {}
bk@work.mysql.com's avatar
bk@work.mysql.com committed
187 188
  inline bool is_same(SEL_ARG *arg)
  {
189
    if (type != arg->type || part != arg->part)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287
      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;
  }
288
  void store_min(uint length,char **min_key,uint min_key_flag)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
289
  {
290 291 292
    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
293 294 295 296
    {
      if (maybe_null && *min_value)
      {
	**min_key=1;
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
297
	bzero(*min_key+1,length-1);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
298 299
      }
      else
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
300 301
	memcpy(*min_key,min_value,length);
      (*min_key)+= length;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
302
    }
303
  }
bk@work.mysql.com's avatar
bk@work.mysql.com committed
304 305 306
  void store(uint length,char **min_key,uint min_key_flag,
	     char **max_key, uint max_key_flag)
  {
307
    store_min(length, min_key, min_key_flag);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
308 309 310 311 312 313
    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
314
	bzero(*max_key+1,length-1);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
315 316
      }
      else
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
317 318
	memcpy(*max_key,max_value,length);
      (*max_key)+= length;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
319 320 321 322 323 324
    }
  }

  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
325
    key_tree->store(key[key_tree->part].store_length,
bk@work.mysql.com's avatar
bk@work.mysql.com committed
326 327 328 329 330 331 332 333 334 335 336 337
		    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
338
    key_tree->store(key[key_tree->part].store_length,
bk@work.mysql.com's avatar
bk@work.mysql.com committed
339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391
		    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();
};

392
class SEL_IMERGE;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
393

394

bk@work.mysql.com's avatar
bk@work.mysql.com committed
395 396 397 398 399
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) {}
400
  SEL_TREE() :type(KEY)
401
  {
402
    keys_map.clear_all();
403 404
    bzero((char*) keys,sizeof(keys));
  }
bk@work.mysql.com's avatar
bk@work.mysql.com committed
405
  SEL_ARG *keys[MAX_KEY];
406 407
  key_map keys_map;        /* bitmask of non-NULL elements in keys */

408 409
  /*
    Possible ways to read rows using index_merge. The list is non-empty only
410 411 412
    if type==KEY. Currently can be non empty only if keys_map.is_clear_all().
  */
  List<SEL_IMERGE> merges;
413

414 415
  /* 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 */
416
  uint    n_ror_scans;     /* number of set bits in ror_scans_map */
417 418 419 420

  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
421 422 423 424
};


typedef struct st_qsel_param {
425
  THD	*thd;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
426
  TABLE *table;
427 428
  KEY_PART *key_parts,*key_parts_end;
  KEY_PART *key[MAX_KEY]; /* First key parts of keys used in the query */
429
  MEM_ROOT *mem_root, *old_root;
430
  table_map prev_tables,read_tables,current_table;
431
  uint baseflag, max_key_part, range_count;
432

433 434 435
  uint keys; /* number of keys used in the query */

  /* used_key_no -> table_key_no translation table */
436
  uint real_keynr[MAX_KEY];
437

bk@work.mysql.com's avatar
bk@work.mysql.com committed
438 439
  char min_key[MAX_KEY_LENGTH+MAX_FIELD_WIDTH],
    max_key[MAX_KEY_LENGTH+MAX_FIELD_WIDTH];
440
  bool quick;				// Don't calulate possible keys
441
  COND *cond;
442

443
  uint fields_bitmap_size;
444
  MY_BITMAP needed_fields;    /* bitmask of fields needed by the query */
445
  MY_BITMAP tmp_covered_fields;
446 447 448

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

449 450
  uint *imerge_cost_buff;     /* buffer for index_merge cost estimates */
  uint imerge_cost_buff_size; /* size of the buffer */
451

452
  /* TRUE if last checked tree->key can be used for ROR-scan */
453
  bool is_ror_scan;
454 455
  /* Number of ranges in the last checked tree->key */
  uint n_ranges;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
456 457
} PARAM;

458 459 460 461 462
class TABLE_READ_PLAN;
  class TRP_RANGE;
  class TRP_ROR_INTERSECT;
  class TRP_ROR_UNION;
  class TRP_ROR_INDEX_MERGE;
463
  class TRP_GROUP_MIN_MAX;
464 465 466

struct st_ror_scan_info;

467
static SEL_TREE * get_mm_parts(PARAM *param,COND *cond_func,Field *field,
bk@work.mysql.com's avatar
bk@work.mysql.com committed
468 469
			       Item_func::Functype type,Item *value,
			       Item_result cmp_type);
470 471
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
472 473
			    Item_func::Functype type,Item *value);
static SEL_TREE *get_mm_tree(PARAM *param,COND *cond);
474 475

static bool is_key_scan_ror(PARAM *param, uint keynr, uint8 nparts);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
476 477 478 479 480
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);

481
QUICK_RANGE_SELECT *get_quick_select(PARAM *param,uint index,
482
                                     SEL_ARG *key_tree,
483
                                     MEM_ROOT *alloc = NULL);
484
static TRP_RANGE *get_key_scans_params(PARAM *param, SEL_TREE *tree,
485
                                       bool index_read_must_be_used,
486 487 488 489 490 491
                                       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
492 493
TRP_ROR_INTERSECT *get_best_covering_ror_intersect(PARAM *param,
                                                   SEL_TREE *tree,
494 495 496 497
                                                   double read_time);
static
TABLE_READ_PLAN *get_best_disjunct_quick(PARAM *param, SEL_IMERGE *imerge,
                                         double read_time);
498 499
static
TRP_GROUP_MIN_MAX *get_best_group_min_max(PARAM *param, SEL_TREE *tree);
500
static int get_index_merge_params(PARAM *param, key_map& needed_reg,
501
                           SEL_IMERGE *imerge, double *read_time,
502
                           ha_rows* imerge_rows);
503
static double get_index_only_read_time(const PARAM* param, ha_rows records,
504 505
                                       int keynr);

bk@work.mysql.com's avatar
bk@work.mysql.com committed
506
#ifndef DBUG_OFF
507 508
static void print_sel_tree(PARAM *param, SEL_TREE *tree, key_map *tree_map,
                           const char *msg);
509 510
static void print_ror_scans_arr(TABLE *table, const char *msg,
                                struct st_ror_scan_info **start,
511 512 513
                                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
514
#endif
515

bk@work.mysql.com's avatar
bk@work.mysql.com committed
516 517 518 519 520 521
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);
522
bool get_quick_keys(PARAM *param,QUICK_RANGE_SELECT *quick,KEY_PART *key,
bk@work.mysql.com's avatar
bk@work.mysql.com committed
523 524 525 526 527
			   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);
528
static bool null_part_in_key(KEY_PART *key_part, const char *key,
529
                             uint length);
530 531 532 533
bool sel_trees_can_be_ored(SEL_TREE *tree1, SEL_TREE *tree2, PARAM* param);


/*
534
  SEL_IMERGE is a list of possible ways to do index merge, i.e. it is
535
  a condition in the following form:
536
   (t_1||t_2||...||t_N) && (next)
537

538
  where all t_i are SEL_TREEs, next is another SEL_IMERGE and no pair
539 540 541 542 543 544 545 546 547 548 549
  (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:
550
  SEL_TREE *trees_prealloced[PREALLOCED_TREES];
551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567
  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);
};


568
/*
569 570
  Add SEL_TREE to this index_merge without any checks,

571 572
  NOTES
    This function implements the following:
573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604
      (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.
605

606 607 608 609 610
  SYNOPSIS
    or_sel_tree_with_checks()
      param    PARAM from SQL_SELECT::test_quick_select
      new_tree SEL_TREE with type KEY or KEY_SMALLER.

611
  NOTES
612
    This does the following:
613 614
    (t_1||...||t_k)||new_tree =
     either
615 616 617
       = (t_1||...||t_k||new_tree)
     or
       = (t_1||....||(t_j|| new_tree)||...||t_k),
618

619
     where t_i, y are SEL_TREEs.
620 621
    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
622 623
    read may depend on the order of conditions in WHERE part of the query.

624
  RETURN
625
    0  OK
626
    1  One of the trees was combined with new_tree to SEL_TREE::ALWAYS,
627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649
       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;
    }
  }

650
  /* New tree cannot be combined with any of existing trees. */
651 652 653 654 655 656 657 658 659
  return or_sel_tree(param, new_tree);
}


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

  RETURN
    0 - OK
660
    1 - One of conditions in result is always TRUE and this SEL_IMERGE
661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677
        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;
}


678
/*
679
  Perform AND operation on two index_merge lists and store result in *im1.
680 681 682 683 684 685 686 687 688 689 690
*/

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.

691
  NOTES
692 693 694
    The following conversion is implemented:
     (a_1 &&...&& a_N)||(b_1 &&...&& b_K) = AND_i,j(a_i || b_j) =>
      => (a_1||b_1).
695 696

    i.e. all conjuncts except the first one are currently dropped.
697 698
    This is done to avoid producing N*K ways to do index_merge.

monty@mysql.com's avatar
monty@mysql.com committed
699
    If (a_1||b_1) produce a condition that is always TRUE, NULL is returned
700
    and index_merge is discarded (while it is actually possible to try
701
    harder).
702

703 704
    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.
705 706

  RETURN
707
    0     OK, result is stored in *im1
708 709 710
    other Error, both passed lists are unusable
*/

711
int imerge_list_or_list(PARAM *param,
712 713 714 715 716 717
                        List<SEL_IMERGE> *im1,
                        List<SEL_IMERGE> *im2)
{
  SEL_IMERGE *imerge= im1->head();
  im1->empty();
  im1->push_back(imerge);
718

719 720 721 722 723 724 725 726
  return imerge->or_sel_imerge_with_checks(param, im2->head());
}


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

  RETURN
727
    0     OK, result is stored in *im1.
728 729 730
    other Error
*/

731
int imerge_list_or_tree(PARAM *param,
732 733 734 735 736
                        List<SEL_IMERGE> *im1,
                        SEL_TREE *tree)
{
  SEL_IMERGE *imerge;
  List_iterator<SEL_IMERGE> it(*im1);
monty@mishka.local's avatar
monty@mishka.local committed
737
  while ((imerge= it++))
738 739 740 741 742 743
  {
    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
744 745

/***************************************************************************
746
** Basic functions for SQL_SELECT and QUICK_RANGE_SELECT
bk@work.mysql.com's avatar
bk@work.mysql.com committed
747 748 749 750 751 752 753 754 755
***************************************************************************/

	/* 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
756 757 758
			table_map read_tables, COND *conds,
                        bool allow_null_cond,
                        int *error)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
759 760 761 762 763
{
  SQL_SELECT *select;
  DBUG_ENTER("make_select");

  *error=0;
764 765

  if (!conds && !allow_null_cond)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
766 767 768
    DBUG_RETURN(0);
  if (!(select= new SQL_SELECT))
  {
769 770
    *error= 1;			// out of memory
    DBUG_RETURN(0);		/* purecov: inspected */
bk@work.mysql.com's avatar
bk@work.mysql.com committed
771 772 773 774 775 776
  }
  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
777
  if (head->sort.io_cache)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
778
  {
igor@hundin.mysql.fi's avatar
igor@hundin.mysql.fi committed
779
    select->file= *head->sort.io_cache;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
780 781
    select->records=(ha_rows) (select->file.end_of_file/
			       head->file->ref_length);
igor@hundin.mysql.fi's avatar
igor@hundin.mysql.fi committed
782 783
    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
784 785 786 787 788 789 790
  }
  DBUG_RETURN(select);
}


SQL_SELECT::SQL_SELECT() :quick(0),cond(0),free_cond(0)
{
serg@serg.mylan's avatar
serg@serg.mylan committed
791
  quick_keys.clear_all(); needed_reg.clear_all();
bk@work.mysql.com's avatar
bk@work.mysql.com committed
792 793 794 795
  my_b_clear(&file);
}


796
void SQL_SELECT::cleanup()
bk@work.mysql.com's avatar
bk@work.mysql.com committed
797 798
{
  delete quick;
799
  quick= 0;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
800
  if (free_cond)
801 802
  {
    free_cond=0;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
803
    delete cond;
804
    cond= 0;
805
  }
bk@work.mysql.com's avatar
bk@work.mysql.com committed
806 807 808
  close_cached_file(&file);
}

809 810 811 812 813 814

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

815
#undef index					// Fix for Unixware 7
bk@work.mysql.com's avatar
bk@work.mysql.com committed
816

sergefp@mysql.com's avatar
sergefp@mysql.com committed
817 818
QUICK_SELECT_I::QUICK_SELECT_I()
  :max_used_key_length(0),
819
   used_key_parts(0)
sergefp@mysql.com's avatar
sergefp@mysql.com committed
820 821
{}

822
QUICK_RANGE_SELECT::QUICK_RANGE_SELECT(THD *thd, TABLE *table, uint key_nr,
sergefp@mysql.com's avatar
sergefp@mysql.com committed
823
                                       bool no_alloc, MEM_ROOT *parent_alloc)
824
  :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
825
{
monty@mysql.com's avatar
monty@mysql.com committed
826
  sorted= 0;
sergefp@mysql.com's avatar
sergefp@mysql.com committed
827 828
  index= key_nr;
  head=  table;
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
829
  key_part_info= head->key_info[index].key_part;
830
  my_init_dynamic_array(&ranges, sizeof(QUICK_RANGE*), 16, 16);
sergefp@mysql.com's avatar
sergefp@mysql.com committed
831

sergefp@mysql.com's avatar
sergefp@mysql.com committed
832
  /* 'thd' is not accessible in QUICK_RANGE_SELECT::reset(). */
ingo@mysql.com's avatar
ingo@mysql.com committed
833 834 835 836 837 838
  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
839
  if (!no_alloc && !parent_alloc)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
840
  {
841 842
    // Allocates everything through the internal memroot
    init_sql_alloc(&alloc, thd->variables.range_alloc_block_size, 0);
843
    thd->mem_root= &alloc;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
844 845 846
  }
  else
    bzero((char*) &alloc,sizeof(alloc));
847 848
  file= head->file;
  record= head->record[0];
bk@work.mysql.com's avatar
bk@work.mysql.com committed
849 850
}

monty@mysql.com's avatar
monty@mysql.com committed
851

852 853
int QUICK_RANGE_SELECT::init()
{
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
854
  DBUG_ENTER("QUICK_RANGE_SELECT::init");
ingo@mysql.com's avatar
ingo@mysql.com committed
855

856 857 858
  if (file->inited != handler::NONE)
    file->ha_index_or_rnd_end();
  DBUG_RETURN(error= file->ha_index_init(index));
monty@mysql.com's avatar
monty@mysql.com committed
859 860 861 862 863 864
}


void QUICK_RANGE_SELECT::range_end()
{
  if (file->inited != handler::NONE)
865
    file->ha_index_or_rnd_end();
bk@work.mysql.com's avatar
bk@work.mysql.com committed
866 867
}

monty@mysql.com's avatar
monty@mysql.com committed
868

869
QUICK_RANGE_SELECT::~QUICK_RANGE_SELECT()
870
{
871
  DBUG_ENTER("QUICK_RANGE_SELECT::~QUICK_RANGE_SELECT");
872 873
  if (!dont_free)
  {
874 875
    /* file is NULL for CPK scan on covering ROR-intersection */
    if (file) 
876
    {
877 878 879 880 881 882 883
      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();
884
        file->external_lock(current_thd, F_UNLCK);
885 886
        file->close();
      }
887
    }
888
    delete_dynamic(&ranges); /* ranges are allocated in alloc */
889 890
    free_root(&alloc,MYF(0));
  }
ingo@mysql.com's avatar
ingo@mysql.com committed
891 892 893 894
  if (multi_range)
    my_free((char*) multi_range, MYF(0));
  if (multi_range_buff)
    my_free((char*) multi_range_buff, MYF(0));
895
  DBUG_VOID_RETURN;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
896 897
}

898

899
QUICK_INDEX_MERGE_SELECT::QUICK_INDEX_MERGE_SELECT(THD *thd_param,
900
                                                   TABLE *table)
sergefp@mysql.com's avatar
sergefp@mysql.com committed
901
  :pk_quick_select(NULL), thd(thd_param)
902
{
903
  DBUG_ENTER("QUICK_INDEX_MERGE_SELECT::QUICK_INDEX_MERGE_SELECT");
904 905
  index= MAX_KEY;
  head= table;
906
  bzero(&read_record, sizeof(read_record));
907
  init_sql_alloc(&alloc, thd->variables.range_alloc_block_size, 0);
908
  DBUG_VOID_RETURN;
909 910 911 912
}

int QUICK_INDEX_MERGE_SELECT::init()
{
sergefp@mysql.com's avatar
sergefp@mysql.com committed
913 914
  DBUG_ENTER("QUICK_INDEX_MERGE_SELECT::init");
  DBUG_RETURN(0);
915 916
}

917
int QUICK_INDEX_MERGE_SELECT::reset()
918
{
919
  DBUG_ENTER("QUICK_INDEX_MERGE_SELECT::reset");
sergefp@mysql.com's avatar
sergefp@mysql.com committed
920
  DBUG_RETURN(read_keys_and_merge());
921 922
}

923
bool
924 925
QUICK_INDEX_MERGE_SELECT::push_quick_back(QUICK_RANGE_SELECT *quick_sel_range)
{
926 927
  /*
    Save quick_select that does scan on clustered primary key as it will be
928
    processed separately.
929
  */
930
  if (head->file->primary_key_is_clustered() &&
931
      quick_sel_range->index == head->s->primary_key)
932 933 934 935
    pk_quick_select= quick_sel_range;
  else
    return quick_selects.push_back(quick_sel_range);
  return 0;
936 937 938 939
}

QUICK_INDEX_MERGE_SELECT::~QUICK_INDEX_MERGE_SELECT()
{
sergefp@mysql.com's avatar
sergefp@mysql.com committed
940 941
  List_iterator_fast<QUICK_RANGE_SELECT> quick_it(quick_selects);
  QUICK_RANGE_SELECT* quick;
942
  DBUG_ENTER("QUICK_INDEX_MERGE_SELECT::~QUICK_INDEX_MERGE_SELECT");
sergefp@mysql.com's avatar
sergefp@mysql.com committed
943 944 945
  quick_it.rewind();
  while ((quick= quick_it++))
    quick->file= NULL;
946
  quick_selects.delete_elements();
947
  delete pk_quick_select;
948
  free_root(&alloc,MYF(0));
949
  DBUG_VOID_RETURN;
950 951
}

952 953 954 955 956

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
957
  : cpk_quick(NULL), thd(thd_param), need_to_fetch_row(retrieve_full_rows),
958
    scans_inited(FALSE)
959 960
{
  index= MAX_KEY;
961
  head= table;
962 963
  record= head->record[0];
  if (!parent_alloc)
964
    init_sql_alloc(&alloc, thd->variables.range_alloc_block_size, 0);
965 966
  else
    bzero(&alloc, sizeof(MEM_ROOT));
967
  last_rowid= (byte*)alloc_root(parent_alloc? parent_alloc : &alloc,
968 969 970
                                head->file->ref_length);
}

971

972
/*
973 974 975
  Do post-constructor initialization.
  SYNOPSIS
    QUICK_ROR_INTERSECT_SELECT::init()
976

977 978 979 980 981
  RETURN
    0      OK
    other  Error code
*/

982 983
int QUICK_ROR_INTERSECT_SELECT::init()
{
sergefp@mysql.com's avatar
sergefp@mysql.com committed
984 985 986
  DBUG_ENTER("QUICK_ROR_INTERSECT_SELECT::init");
 /* Check if last_rowid was successfully allocated in ctor */
  DBUG_RETURN(!last_rowid);
987 988 989 990
}


/*
991 992 993 994
  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
995
      reuse_handler If TRUE, use head->file, otherwise create a separate
996 997 998 999
                    handler object

  NOTES
    This function creates and prepares for subsequent use a separate handler
1000
    object if it can't reuse head->file. The reason for this is that during
1001 1002 1003
    ROR-merge several key scans are performed simultaneously, and a single
    handler is only capable of preserving context of a single key scan.

1004
    In ROR-merge the quick select doing merge does full records retrieval,
1005
    merged quick selects read only keys.
1006 1007

  RETURN
1008 1009 1010 1011
    0  ROR child scan initialized, ok to use.
    1  error
*/

1012
int QUICK_RANGE_SELECT::init_ror_merged_scan(bool reuse_handler)
1013 1014
{
  handler *save_file= file;
1015
  DBUG_ENTER("QUICK_RANGE_SELECT::init_ror_merged_scan");
1016

1017 1018 1019 1020
  if (reuse_handler)
  {
    DBUG_PRINT("info", ("Reusing handler %p", file));
    if (file->extra(HA_EXTRA_KEYREAD) ||
1021
        file->extra(HA_EXTRA_RETRIEVE_PRIMARY_KEY) ||
1022 1023 1024 1025
        init() || reset())
    {
      DBUG_RETURN(1);
    }
monty@mysql.com's avatar
monty@mysql.com committed
1026
    DBUG_RETURN(0);
1027 1028 1029 1030 1031 1032 1033 1034
  }

  /* Create a separate handler object for this quick select */
  if (free_file)
  {
    /* already have own 'handler' object. */
    DBUG_RETURN(0);
  }
1035

1036
  THD *thd= current_thd;
1037
  if (!(file= head->file->clone(thd->mem_root)))
1038
  {
1039
    /* Caller will free the memory */
1040 1041
    goto failure;
  }
1042 1043
  if (file->external_lock(thd, F_RDLCK))
    goto failure;
1044 1045

  if (file->extra(HA_EXTRA_KEYREAD) ||
1046
      file->extra(HA_EXTRA_RETRIEVE_PRIMARY_KEY) ||
1047 1048
      init() || reset())
  {
1049
    file->external_lock(thd, F_UNLCK);
1050 1051 1052
    file->close();
    goto failure;
  }
monty@mysql.com's avatar
monty@mysql.com committed
1053
  free_file= TRUE;
1054 1055 1056 1057 1058 1059 1060 1061
  last_rowid= file->ref;
  DBUG_RETURN(0);

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

1062 1063 1064 1065 1066

/*
  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
1067
      reuse_handler If TRUE, use head->file, otherwise create separate
1068
                    handler object.
1069
  RETURN
1070 1071 1072 1073
    0     OK
    other error code
*/
int QUICK_ROR_INTERSECT_SELECT::init_ror_merged_scan(bool reuse_handler)
1074 1075 1076
{
  List_iterator_fast<QUICK_RANGE_SELECT> quick_it(quick_selects);
  QUICK_RANGE_SELECT* quick;
1077
  DBUG_ENTER("QUICK_ROR_INTERSECT_SELECT::init_ror_merged_scan");
1078 1079

  /* Initialize all merged "children" quick selects */
sergefp@mysql.com's avatar
sergefp@mysql.com committed
1080
  DBUG_ASSERT(!need_to_fetch_row || reuse_handler);
1081 1082 1083
  if (!need_to_fetch_row && reuse_handler)
  {
    quick= quick_it++;
1084
    /*
1085
      There is no use of this->file. Use it for the first of merged range
1086 1087
      selects.
    */
monty@mysql.com's avatar
monty@mysql.com committed
1088
    if (quick->init_ror_merged_scan(TRUE))
1089 1090 1091
      DBUG_RETURN(1);
    quick->file->extra(HA_EXTRA_KEYREAD_PRESERVE_FIELDS);
  }
monty@mishka.local's avatar
monty@mishka.local committed
1092
  while ((quick= quick_it++))
1093
  {
monty@mysql.com's avatar
monty@mysql.com committed
1094
    if (quick->init_ror_merged_scan(FALSE))
1095 1096
      DBUG_RETURN(1);
    quick->file->extra(HA_EXTRA_KEYREAD_PRESERVE_FIELDS);
1097
    /* All merged scans share the same record buffer in intersection. */
1098 1099 1100
    quick->record= head->record[0];
  }

monty@mysql.com's avatar
monty@mysql.com committed
1101
  if (need_to_fetch_row && head->file->ha_rnd_init(1))
1102 1103 1104 1105 1106 1107 1108
  {
    DBUG_PRINT("error", ("ROR index_merge rnd_init call failed"));
    DBUG_RETURN(1);
  }
  DBUG_RETURN(0);
}

1109

1110
/*
1111 1112 1113 1114 1115 1116 1117 1118
  Initialize quick select for row retrieval.
  SYNOPSIS
    reset()
  RETURN
    0      OK
    other  Error code
*/

1119 1120 1121
int QUICK_ROR_INTERSECT_SELECT::reset()
{
  DBUG_ENTER("QUICK_ROR_INTERSECT_SELECT::reset");
sergefp@mysql.com's avatar
sergefp@mysql.com committed
1122 1123
  if (!scans_inited && init_ror_merged_scan(TRUE))
    DBUG_RETURN(1);
1124
  scans_inited= TRUE;
sergefp@mysql.com's avatar
sergefp@mysql.com committed
1125 1126 1127 1128 1129
  List_iterator_fast<QUICK_RANGE_SELECT> it(quick_selects);
  QUICK_RANGE_SELECT *quick;
  while ((quick= it++))
    quick->reset();
  DBUG_RETURN(0);
1130 1131
}

1132 1133 1134

/*
  Add a merged quick select to this ROR-intersection quick select.
1135

1136 1137 1138 1139 1140 1141
  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.
1142

1143
  RETURN
1144
    FALSE OK
monty@mysql.com's avatar
monty@mysql.com committed
1145
    TRUE  Out of memory.
1146 1147
*/

1148
bool
1149 1150
QUICK_ROR_INTERSECT_SELECT::push_quick_back(QUICK_RANGE_SELECT *quick)
{
1151
  return quick_selects.push_back(quick);
1152 1153 1154
}

QUICK_ROR_INTERSECT_SELECT::~QUICK_ROR_INTERSECT_SELECT()
1155
{
1156
  DBUG_ENTER("QUICK_ROR_INTERSECT_SELECT::~QUICK_ROR_INTERSECT_SELECT");
1157
  quick_selects.delete_elements();
1158 1159
  delete cpk_quick;
  free_root(&alloc,MYF(0));
monty@mysql.com's avatar
monty@mysql.com committed
1160 1161
  if (need_to_fetch_row && head->file->inited != handler::NONE)
    head->file->ha_rnd_end();
1162 1163 1164
  DBUG_VOID_RETURN;
}

monty@mysql.com's avatar
monty@mysql.com committed
1165

1166 1167
QUICK_ROR_UNION_SELECT::QUICK_ROR_UNION_SELECT(THD *thd_param,
                                               TABLE *table)
1168
  : thd(thd_param), scans_inited(FALSE)
1169 1170 1171 1172 1173 1174
{
  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
1175
  thd_param->mem_root= &alloc;
1176 1177
}

1178 1179 1180 1181 1182

/*
  Do post-constructor initialization.
  SYNOPSIS
    QUICK_ROR_UNION_SELECT::init()
1183

1184 1185 1186 1187 1188
  RETURN
    0      OK
    other  Error code
*/

1189 1190
int QUICK_ROR_UNION_SELECT::init()
{
sergefp@mysql.com's avatar
sergefp@mysql.com committed
1191
  DBUG_ENTER("QUICK_ROR_UNION_SELECT::init");
1192
  if (init_queue(&queue, quick_selects.elements, 0,
monty@mysql.com's avatar
monty@mysql.com committed
1193
                 FALSE , QUICK_ROR_UNION_SELECT::queue_cmp,
1194 1195 1196
                 (void*) this))
  {
    bzero(&queue, sizeof(QUEUE));
sergefp@mysql.com's avatar
sergefp@mysql.com committed
1197
    DBUG_RETURN(1);
1198
  }
1199

1200
  if (!(cur_rowid= (byte*)alloc_root(&alloc, 2*head->file->ref_length)))
sergefp@mysql.com's avatar
sergefp@mysql.com committed
1201
    DBUG_RETURN(1);
1202
  prev_rowid= cur_rowid + head->file->ref_length;
sergefp@mysql.com's avatar
sergefp@mysql.com committed
1203
  DBUG_RETURN(0);
1204 1205
}

1206

1207
/*
1208
  Comparison function to be used QUICK_ROR_UNION_SELECT::queue priority
1209 1210
  queue.

1211 1212 1213 1214 1215 1216
  SYNPOSIS
    QUICK_ROR_UNION_SELECT::queue_cmp()
      arg   Pointer to QUICK_ROR_UNION_SELECT
      val1  First merged select
      val2  Second merged select
*/
1217

1218 1219
int QUICK_ROR_UNION_SELECT::queue_cmp(void *arg, byte *val1, byte *val2)
{
1220
  QUICK_ROR_UNION_SELECT *self= (QUICK_ROR_UNION_SELECT*)arg;
1221 1222 1223 1224
  return self->head->file->cmp_ref(((QUICK_SELECT_I*)val1)->last_rowid,
                                   ((QUICK_SELECT_I*)val2)->last_rowid);
}

1225

1226
/*
1227 1228 1229
  Initialize quick select for row retrieval.
  SYNOPSIS
    reset()
1230

1231 1232 1233 1234 1235
  RETURN
    0      OK
    other  Error code
*/

1236 1237 1238 1239 1240
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
1241
  have_prev_rowid= FALSE;
sergefp@mysql.com's avatar
sergefp@mysql.com committed
1242 1243 1244 1245 1246 1247 1248 1249 1250
  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);
    }
1251
    scans_inited= TRUE;
sergefp@mysql.com's avatar
sergefp@mysql.com committed
1252 1253
  }
  queue_remove_all(&queue);
1254 1255
  /*
    Initialize scans for merged quick selects and put all merged quick
1256 1257 1258 1259 1260
    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
1261
    if (quick->reset())
1262
      DBUG_RETURN(1);
1263 1264 1265 1266
    if ((error= quick->get_next()))
    {
      if (error == HA_ERR_END_OF_FILE)
        continue;
monty@mysql.com's avatar
monty@mysql.com committed
1267
      DBUG_RETURN(error);
1268 1269 1270 1271 1272
    }
    quick->save_last_pos();
    queue_insert(&queue, (byte*)quick);
  }

monty@mysql.com's avatar
monty@mysql.com committed
1273
  if (head->file->ha_rnd_init(1))
1274 1275 1276 1277 1278 1279 1280 1281 1282
  {
    DBUG_PRINT("error", ("ROR index_merge rnd_init call failed"));
    DBUG_RETURN(1);
  }

  DBUG_RETURN(0);
}


1283
bool
1284 1285 1286 1287 1288 1289 1290 1291 1292
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);
1293
  quick_selects.delete_elements();
1294 1295
  if (head->file->inited != handler::NONE)
    head->file->ha_rnd_end();
1296 1297
  free_root(&alloc,MYF(0));
  DBUG_VOID_RETURN;
1298 1299
}

1300

bk@work.mysql.com's avatar
bk@work.mysql.com committed
1301 1302 1303 1304 1305 1306 1307 1308 1309 1310 1311 1312 1313 1314 1315 1316 1317 1318 1319 1320 1321 1322 1323 1324 1325 1326 1327 1328 1329 1330 1331 1332 1333 1334 1335 1336 1337 1338 1339 1340 1341 1342 1343 1344 1345 1346 1347 1348 1349 1350 1351 1352 1353
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)
  {
1354 1355
    if (!(tmp= new SEL_ARG(type)))
      return 0;					// out of memory
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1356 1357 1358 1359 1360 1361
    tmp->prev= *next_arg;			// Link into next/prev chain
    (*next_arg)->next=tmp;
    (*next_arg)= tmp;
  }
  else
  {
1362 1363 1364
    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
1365 1366 1367 1368 1369 1370 1371 1372 1373 1374
    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)
1375 1376
      if (!(tmp->right= right->clone(tmp,next_arg)))
	return 0;				// OOM
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1377 1378
  }
  increment_use_count(1);
1379
  tmp->color= color;
1380
  tmp->elements= this->elements;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1381 1382 1383 1384 1385 1386 1387 1388 1389 1390 1391 1392 1393 1394 1395 1396 1397 1398 1399 1400 1401 1402 1403
  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;
}

1404

bk@work.mysql.com's avatar
bk@work.mysql.com committed
1405 1406 1407
/*
  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
1408
*/
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1409 1410 1411 1412 1413 1414 1415 1416 1417 1418 1419 1420 1421 1422 1423 1424 1425 1426 1427 1428 1429 1430 1431

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
1432
    a++; b++;					// Skip NULL marker
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1433 1434 1435 1436 1437 1438 1439 1440 1441 1442 1443 1444 1445 1446 1447 1448 1449 1450 1451 1452 1453 1454 1455 1456
  }
  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;
1457
  root= clone((SEL_ARG *) 0, &next_arg);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1458 1459
  next_arg->next=0;				// Fix last link
  tmp_link.next->prev=0;			// Fix first link
1460 1461
  if (root)					// If not OOM
    root->use_count= 0;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1462 1463 1464
  return root;
}

1465

1466
/*
1467
  Find the best index to retrieve first N records in given order
1468 1469 1470 1471 1472 1473 1474 1475

  SYNOPSIS
    get_index_for_order()
      table  Table to be accessed
      order  Required ordering
      limit  Number of records that will be retrieved

  DESCRIPTION
1476 1477 1478 1479
    Find the best index that allows to retrieve first #limit records in the 
    given order cheaper then one would retrieve them using full table scan.

  IMPLEMENTATION
1480
    Run through all table indexes and find the shortest index that allows
1481 1482
    records to be retrieved in given order. We look for the shortest index
    as we will have fewer index pages to read with it.
1483 1484 1485 1486 1487 1488 1489 1490 1491 1492 1493 1494 1495 1496 1497 1498 1499 1500 1501 1502 1503 1504

    This function is used only by UPDATE/DELETE, so we take into account how
    the UPDATE/DELETE code will work:
     * index can only be scanned in forward direction
     * HA_EXTRA_KEYREAD will not be used
    Perhaps these assumptions could be relaxed

  RETURN
    index number
    MAX_KEY if no such index was found.
*/

uint get_index_for_order(TABLE *table, ORDER *order, ha_rows limit)
{
  uint idx;
  uint match_key= MAX_KEY, match_key_len= MAX_KEY_LENGTH + 1;
  ORDER *ord;
  
  for (ord= order; ord; ord= ord->next)
    if (!ord->asc)
      return MAX_KEY;

sergefp@mysql.com's avatar
sergefp@mysql.com committed
1505
  for (idx= 0; idx < table->s->keys; idx++)
1506 1507 1508 1509 1510 1511 1512 1513 1514 1515 1516 1517 1518 1519 1520 1521 1522 1523 1524 1525 1526 1527 1528 1529 1530 1531 1532 1533 1534 1535 1536 1537 1538 1539 1540 1541 1542 1543 1544 1545 1546 1547 1548 1549 1550 1551 1552 1553 1554
  {
    if (!(table->keys_in_use_for_query.is_set(idx)))
      continue;
    KEY_PART_INFO *keyinfo= table->key_info[idx].key_part;
    uint partno= 0;
    
    /* 
      The below check is sufficient considering we now have either BTREE 
      indexes (records are returned in order for any index prefix) or HASH 
      indexes (records are not returned in order for any index prefix).
    */
    if (!(table->file->index_flags(idx, 0, 1) & HA_READ_ORDER))
      continue;
    for (ord= order; ord; ord= ord->next, partno++)
    {
      Item *item= order->item[0];
      if (!(item->type() == Item::FIELD_ITEM &&
           ((Item_field*)item)->field->eq(keyinfo[partno].field)))
        break;
    }
    
    if (!ord && table->key_info[idx].key_length < match_key_len)
    {
      /* 
        Ok, the ordering is compatible and this key is shorter then
        previous match (we want shorter keys as we'll have to read fewer
        index pages for the same number of records)
      */
      match_key= idx;
      match_key_len= table->key_info[idx].key_length;
    }
  }

  if (match_key != MAX_KEY)
  {
    /* 
      Found an index that allows records to be retrieved in the requested 
      order. Now we'll check if using the index is cheaper then doing a table
      scan.
    */
    double full_scan_time= table->file->scan_time();
    double index_scan_time= table->file->read_time(match_key, 1, limit);
    if (index_scan_time > full_scan_time)
      match_key= MAX_KEY;
  }
  return match_key;
}


serg@serg.mylan's avatar
serg@serg.mylan committed
1555
/*
1556
  Table rows retrieval plan. Range optimizer creates QUICK_SELECT_I-derived
1557 1558 1559 1560 1561
  objects from table read plans.
*/
class TABLE_READ_PLAN
{
public:
1562 1563
  /*
    Plan read cost, with or without cost of full row retrieval, depending
1564 1565
    on plan creation parameters.
  */
1566
  double read_cost;
1567
  ha_rows records; /* estimate of #rows to be examined */
serg@serg.mylan's avatar
serg@serg.mylan committed
1568

1569 1570
  /*
    If TRUE, the scan returns rows in rowid order. This is used only for
1571 1572
    scans that can be both ROR and non-ROR.
  */
1573
  bool is_ror;
1574

1575 1576 1577 1578 1579
  /*
    Create quick select for this plan.
    SYNOPSIS
     make_quick()
       param               Parameter from test_quick_select
monty@mysql.com's avatar
monty@mysql.com committed
1580
       retrieve_full_rows  If TRUE, created quick select will do full record
1581 1582
                           retrieval.
       parent_alloc        Memory pool to use, if any.
1583

1584 1585
    NOTES
      retrieve_full_rows is ignored by some implementations.
1586 1587

    RETURN
1588 1589 1590
      created quick select
      NULL on any error.
  */
1591 1592 1593 1594
  virtual QUICK_SELECT_I *make_quick(PARAM *param,
                                     bool retrieve_full_rows,
                                     MEM_ROOT *parent_alloc=NULL) = 0;

1595
  /* Table read plans are allocated on MEM_ROOT and are never deleted */
1596 1597
  static void *operator new(size_t size, MEM_ROOT *mem_root)
  { return (void*) alloc_root(mem_root, (uint) size); }
1598
  static void operator delete(void *ptr,size_t size) { TRASH(ptr, size); }
1599
  static void operator delete(void *ptr, MEM_ROOT *mem_root) { /* Never called */ }
1600 1601
  virtual ~TABLE_READ_PLAN() {}               /* Remove gcc warning */

1602 1603 1604 1605 1606 1607 1608
};

class TRP_ROR_INTERSECT;
class TRP_ROR_UNION;
class TRP_INDEX_MERGE;


1609
/*
1610
  Plan for a QUICK_RANGE_SELECT scan.
1611 1612 1613
  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
1614
*/
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1615

1616
class TRP_RANGE : public TABLE_READ_PLAN
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1617
{
1618
public:
1619 1620
  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
1621

1622
  TRP_RANGE(SEL_ARG *key_arg, uint idx_arg)
1623 1624
   : key(key_arg), key_idx(idx_arg)
  {}
1625
  virtual ~TRP_RANGE() {}                     /* Remove gcc warning */
1626

1627 1628 1629 1630 1631 1632 1633 1634 1635 1636 1637 1638 1639
  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
1640 1641


1642 1643
/* Plan for QUICK_ROR_INTERSECT_SELECT scan. */

1644 1645 1646
class TRP_ROR_INTERSECT : public TABLE_READ_PLAN
{
public:
1647 1648
  TRP_ROR_INTERSECT() {}                      /* Remove gcc warning */
  virtual ~TRP_ROR_INTERSECT() {}             /* Remove gcc warning */
1649
  QUICK_SELECT_I *make_quick(PARAM *param, bool retrieve_full_rows,
1650
                             MEM_ROOT *parent_alloc);
1651

1652
  /* Array of pointers to ROR range scans used in this intersection */
1653
  struct st_ror_scan_info **first_scan;
1654 1655
  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
1656
  bool is_covering; /* TRUE if no row retrieval phase is necessary */
1657
  double index_scan_costs; /* SUM(cost(index_scan)) */
1658 1659
};

1660

1661
/*
1662 1663
  Plan for QUICK_ROR_UNION_SELECT scan.
  QUICK_ROR_UNION_SELECT always retrieves full rows, so retrieve_full_rows
1664
  is ignored by make_quick.
1665
*/
1666

1667 1668 1669
class TRP_ROR_UNION : public TABLE_READ_PLAN
{
public:
1670 1671
  TRP_ROR_UNION() {}                          /* Remove gcc warning */
  virtual ~TRP_ROR_UNION() {}                 /* Remove gcc warning */
1672
  QUICK_SELECT_I *make_quick(PARAM *param, bool retrieve_full_rows,
1673
                             MEM_ROOT *parent_alloc);
1674 1675
  TABLE_READ_PLAN **first_ror; /* array of ptrs to plans for merged scans */
  TABLE_READ_PLAN **last_ror;  /* end of the above array */
1676 1677
};

1678 1679 1680 1681

/*
  Plan for QUICK_INDEX_MERGE_SELECT scan.
  QUICK_ROR_INTERSECT_SELECT always retrieves full rows, so retrieve_full_rows
1682
  is ignored by make_quick.
1683 1684
*/

1685 1686 1687
class TRP_INDEX_MERGE : public TABLE_READ_PLAN
{
public:
1688 1689
  TRP_INDEX_MERGE() {}                        /* Remove gcc warning */
  virtual ~TRP_INDEX_MERGE() {}               /* Remove gcc warning */
1690
  QUICK_SELECT_I *make_quick(PARAM *param, bool retrieve_full_rows,
1691
                             MEM_ROOT *parent_alloc);
1692 1693
  TRP_RANGE **range_scans; /* array of ptrs to plans of merged scans */
  TRP_RANGE **range_scans_end; /* end of the array */
1694 1695 1696
};


1697 1698 1699 1700 1701 1702 1703 1704 1705 1706 1707 1708 1709 1710 1711 1712 1713 1714 1715 1716 1717 1718 1719
/*
  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:
1720 1721 1722 1723
  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,
1724 1725
                    uint index_arg, uint key_infix_len_arg,
                    byte *key_infix_arg,
1726 1727 1728 1729 1730 1731 1732 1733 1734
                    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)
1735 1736 1737 1738
    {
      if (key_infix_len)
        memcpy(this->key_infix, key_infix_arg, key_infix_len);
    }
1739
  virtual ~TRP_GROUP_MIN_MAX() {}             /* Remove gcc warning */
1740 1741 1742 1743 1744 1745

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


1746
/*
1747
  Fill param->needed_fields with bitmap of fields used in the query.
1748
  SYNOPSIS
1749 1750
    fill_used_fields_bitmap()
      param Parameter from test_quick_select function.
1751

1752 1753 1754
  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).
1755 1756 1757
  RETURN
    0  Ok
    1  Out of memory.
1758 1759 1760 1761 1762
*/

static int fill_used_fields_bitmap(PARAM *param)
{
  TABLE *table= param->table;
1763
  param->fields_bitmap_size= (table->s->fields/8 + 1);
1764 1765
  uchar *tmp;
  uint pk;
1766
  param->tmp_covered_fields.bitmap= 0;
1767
  if (!(tmp= (uchar*)alloc_root(param->mem_root,param->fields_bitmap_size)) ||
1768
      bitmap_init(&param->needed_fields, tmp, param->fields_bitmap_size*8,
monty@mysql.com's avatar
monty@mysql.com committed
1769
                  FALSE))
1770
    return 1;
1771

1772
  bitmap_clear_all(&param->needed_fields);
1773
  for (uint i= 0; i < table->s->fields; i++)
1774 1775 1776 1777 1778
  {
    if (param->thd->query_id == table->field[i]->query_id)
      bitmap_set_bit(&param->needed_fields, i+1);
  }

1779
  pk= param->table->s->primary_key;
1780 1781
  if (param->table->file->primary_key_is_clustered() && pk != MAX_KEY)
  {
1782
    /* The table uses clustered PK and it is not internally generated */
1783
    KEY_PART_INFO *key_part= param->table->key_info[pk].key_part;
1784
    KEY_PART_INFO *key_part_end= key_part +
1785
                                 param->table->key_info[pk].key_parts;
1786
    for (;key_part != key_part_end; ++key_part)
1787 1788 1789 1790 1791 1792 1793 1794
    {
      bitmap_clear_bit(&param->needed_fields, key_part->fieldnr);
    }
  }
  return 0;
}


serg@serg.mylan's avatar
serg@serg.mylan committed
1795
/*
1796
  Test if a key can be used in different ranges
serg@serg.mylan's avatar
serg@serg.mylan committed
1797 1798

  SYNOPSIS
1799 1800 1801 1802 1803
    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)
1804 1805 1806
      limit             Query limit
      force_quick_range Prefer to use range (instead of full table scan) even
                        if it is more expensive.
1807 1808 1809 1810 1811

  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.
1812

1813 1814 1815
    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
1816

1817 1818 1819 1820
  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.

1821
   In addition to force_quick_range other means can be (an usually are) used
1822 1823
   to make this function prefer range over full table scan. Figure out if
   force_quick_range is really needed.
1824

1825 1826 1827 1828
  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
1829
*/
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1830

1831 1832
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
1833 1834 1835 1836
				  ha_rows limit, bool force_quick_range)
{
  uint idx;
  double scan_time;
1837
  DBUG_ENTER("SQL_SELECT::test_quick_select");
serg@serg.mylan's avatar
serg@serg.mylan committed
1838 1839 1840
  DBUG_PRINT("enter",("keys_to_use: %lu  prev_tables: %lu  const_tables: %lu",
		      keys_to_use.to_ulonglong(), (ulong) prev_tables,
		      (ulong) const_tables));
1841
  DBUG_PRINT("info", ("records=%lu", (ulong)head->file->records));
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1842 1843
  delete quick;
  quick=0;
1844 1845 1846
  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
1847 1848
      !limit)
    DBUG_RETURN(0); /* purecov: inspected */
1849 1850
  if (keys_to_use.is_clear_all())
    DBUG_RETURN(0);
1851
  records= head->file->records;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1852 1853
  if (!records)
    records++;					/* purecov: inspected */
1854 1855
  scan_time= (double) records / TIME_FOR_COMPARE + 1;
  read_time= (double) head->file->scan_time() + scan_time + 1.1;
1856 1857
  if (head->force_index)
    scan_time= read_time= DBL_MAX;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1858
  if (limit < records)
1859
    read_time= (double) records + scan_time + 1; // Force to use index
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1860
  else if (read_time <= 2.0 && !force_quick_range)
1861
    DBUG_RETURN(0);				/* No need for quick select */
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1862

1863
  DBUG_PRINT("info",("Time to scan table: %g", read_time));
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1864

1865 1866
  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
1867
  {
1868
    MEM_ROOT alloc;
1869
    SEL_TREE *tree= NULL;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1870
    KEY_PART *key_parts;
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
1871
    KEY *key_info;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1872
    PARAM param;
serg@serg.mylan's avatar
serg@serg.mylan committed
1873

bk@work.mysql.com's avatar
bk@work.mysql.com committed
1874
    /* set up parameter that is passed to all functions */
1875
    param.thd= thd;
monty@mysql.com's avatar
monty@mysql.com committed
1876
    param.baseflag=head->file->table_flags();
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1877 1878 1879 1880 1881
    param.prev_tables=prev_tables | const_tables;
    param.read_tables=read_tables;
    param.current_table= head->map;
    param.table=head;
    param.keys=0;
1882
    param.mem_root= &alloc;
1883
    param.old_root= thd->mem_root;
1884
    param.needed_reg= &needed_reg;
1885
    param.imerge_cost_buff_size= 0;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1886

monty@mashka.mysql.fi's avatar
monty@mashka.mysql.fi committed
1887
    thd->no_errors=1;				// Don't warn about NULL
1888
    init_sql_alloc(&alloc, thd->variables.range_alloc_block_size, 0);
1889 1890 1891 1892
    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
1893
    {
monty@mashka.mysql.fi's avatar
monty@mashka.mysql.fi committed
1894
      thd->no_errors=0;
1895
      free_root(&alloc,MYF(0));			// Return memory & allocator
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1896 1897 1898
      DBUG_RETURN(0);				// Can't use range
    }
    key_parts= param.key_parts;
1899
    thd->mem_root= &alloc;
1900 1901 1902 1903

    /*
      Make an array with description of all key parts of all table keys.
      This is used in get_mm_parts function.
1904
    */
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
1905
    key_info= head->key_info;
1906
    for (idx=0 ; idx < head->s->keys ; idx++, key_info++)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1907
    {
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
1908
      KEY_PART_INFO *key_part_info;
1909
      if (!keys_to_use.is_set(idx))
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1910 1911 1912 1913 1914
	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
1915 1916 1917
      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
1918
      {
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
1919 1920 1921 1922 1923 1924
	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;
1925
        key_parts->image_type =
1926
          (key_info->flags & HA_SPATIAL) ? Field::itMBR : Field::itRAW;
1927
        key_parts->flag=         key_part_info->key_part_flag;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1928 1929 1930 1931 1932
      }
      param.real_keynr[param.keys++]=idx;
    }
    param.key_parts_end=key_parts;

sergefp@mysql.com's avatar
sergefp@mysql.com committed
1933 1934 1935 1936
    /* 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);
1937 1938 1939
      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
1940 1941 1942 1943 1944
      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;
    }
1945

1946 1947 1948 1949 1950
    TABLE_READ_PLAN *best_trp= NULL;
    TRP_GROUP_MIN_MAX *group_trp;
    double best_read_time= read_time;

    if (cond)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1951
    {
1952 1953 1954 1955 1956 1957 1958 1959 1960 1961 1962 1963
      if ((tree= get_mm_tree(&param,cond)))
      {
        if (tree->type == SEL_TREE::IMPOSSIBLE)
        {
          records=0L;                      /* Return -1 from this function. */
          read_time= (double) HA_POS_ERROR;
          goto free_mem;
        }
        if (tree->type != SEL_TREE::KEY &&
            tree->type != SEL_TREE::KEY_SMALLER)
          goto free_mem;
      }
1964 1965 1966 1967 1968 1969 1970 1971 1972 1973 1974 1975 1976 1977
    }

    /*
      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
1978
    {
monty@mysql.com's avatar
monty@mysql.com committed
1979 1980 1981
      /*
        It is possible to use a range-based quick select (but it might be
        slower than 'all' table scan).
1982 1983
      */
      if (tree->merges.is_empty())
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1984
      {
1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996
        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;
        }

1997
        /*
1998 1999 2000
          Simultaneous key scans and row deletes on several handler
          objects are not allowed so don't use ROR-intersection for
          table deletes.
2001
        */
2002 2003 2004 2005
        if ((thd->lex->sql_command != SQLCOM_DELETE))
#ifdef NOT_USED
          if ((thd->lex->sql_command != SQLCOM_UPDATE))
#endif
2006
        {
2007
          /*
2008 2009
            Get best non-covering ROR-intersection plan and prepare data for
            building covering ROR-intersection.
2010
          */
2011 2012
          if ((rori_trp= get_best_ror_intersect(&param, tree, best_read_time,
                                                &can_build_covering)))
2013
          {
2014 2015
            best_trp= rori_trp;
            best_read_time= best_trp->read_cost;
2016 2017
            /*
              Try constructing covering ROR-intersect only if it looks possible
2018 2019
              and worth doing.
            */
2020 2021 2022 2023
            if (!rori_trp->is_covering && can_build_covering &&
                (rori_trp= get_best_covering_ror_intersect(&param, tree,
                                                           best_read_time)))
              best_trp= rori_trp;
2024 2025
          }
        }
2026 2027 2028 2029 2030 2031 2032 2033 2034 2035 2036 2037
      }
      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++))
2038
        {
2039 2040 2041 2042
          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;
2043
        }
2044 2045 2046 2047
        if (best_conj_trp)
          best_trp= best_conj_trp;
      }
    }
2048

2049
    thd->mem_root= param.old_root;
2050 2051 2052 2053 2054 2055 2056 2057 2058

    /* 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
2059 2060
      }
    }
2061 2062

  free_mem:
2063
    free_root(&alloc,MYF(0));			// Return memory & allocator
2064
    thd->mem_root= param.old_root;
monty@mashka.mysql.fi's avatar
monty@mashka.mysql.fi committed
2065
    thd->no_errors=0;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2066
  }
2067

2068
  DBUG_EXECUTE("info", print_quick(quick, &needed_reg););
2069

bk@work.mysql.com's avatar
bk@work.mysql.com committed
2070 2071 2072 2073 2074 2075 2076
  /*
    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);
}

2077

2078
/*
2079 2080 2081 2082
  Get cost of 'sweep' full records retrieval.
  SYNOPSIS
    get_sweep_read_cost()
      param            Parameter from test_quick_select
2083
      records          # of records to be retrieved
2084
  RETURN
2085
    cost of sweep
2086
*/
2087

2088
double get_sweep_read_cost(const PARAM *param, ha_rows records)
2089
{
2090
  double result;
2091
  DBUG_ENTER("get_sweep_read_cost");
2092 2093
  if (param->table->file->primary_key_is_clustered())
  {
2094
    result= param->table->file->read_time(param->table->s->primary_key,
2095
                                          records, records);
2096 2097
  }
  else
2098
  {
2099
    double n_blocks=
2100
      ceil(ulonglong2double(param->table->file->data_file_length) / IO_SIZE);
2101 2102 2103 2104
    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;
2105
    DBUG_PRINT("info",("sweep: nblocks=%g, busy_blocks=%g", n_blocks,
2106
                       busy_blocks));
2107
    /*
2108
      Disabled: Bail out if # of blocks to read is bigger than # of blocks in
2109 2110 2111 2112 2113 2114 2115 2116
      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' */
2117
      result= busy_blocks*(DISK_SEEK_BASE_COST +
2118 2119 2120 2121
                          DISK_SEEK_PROP_COST*n_blocks/busy_blocks);
    }
    else
    {
2122
      /*
2123 2124 2125
        Possibly this is a join with source table being non-last table, so
        assume that disk seeks are random here.
      */
2126
      result= busy_blocks;
2127 2128
    }
  }
2129
  DBUG_PRINT("info",("returning cost=%g", result));
2130
  DBUG_RETURN(result);
2131
}
2132 2133


2134 2135 2136 2137
/*
  Get best plan for a SEL_IMERGE disjunctive expression.
  SYNOPSIS
    get_best_disjunct_quick()
2138 2139
      param     Parameter from check_quick_select function
      imerge    Expression to use
2140
      read_time Don't create scans with cost > read_time
2141

2142
  NOTES
2143
    index_merge cost is calculated as follows:
2144
    index_merge_cost =
2145 2146 2147 2148 2149
      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))
2150 2151
       For non-CPK scans,
         cost(index_read_i) = {cost of ordinary 'index only' scan}
2152 2153 2154 2155 2156
       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
2157
        cost(rowid_to_row_scan) =
2158
          {cost of ordinary clustered PK scan with n_ranges=n_rows}
2159 2160

      Otherwise, we use the following model to calculate costs:
2161
      We need to retrieve n_rows rows from file that occupies n_blocks blocks.
2162
      We assume that offsets of rows we need are independent variates with
2163
      uniform distribution in [0..max_file_offset] range.
2164

2165 2166
      We'll denote block as "busy" if it contains row(s) we need to retrieve
      and "empty" if doesn't contain rows we need.
2167

2168
      Probability that a block is empty is (1 - 1/n_blocks)^n_rows (this
2169
      applies to any block in file). Let x_i be a variate taking value 1 if
2170
      block #i is empty and 0 otherwise.
2171

2172 2173
      Then E(x_i) = (1 - 1/n_blocks)^n_rows;

2174 2175
      E(n_empty_blocks) = E(sum(x_i)) = sum(E(x_i)) =
        = n_blocks * ((1 - 1/n_blocks)^n_rows) =
2176 2177 2178 2179
       ~= 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)).
2180

2181 2182
      Average size of "hole" between neighbor non-empty blocks is
           E(hole_size) = n_blocks/E(n_busy_blocks).
2183

2184 2185 2186 2187 2188 2189
      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.
2190 2191 2192 2193 2194

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

  RETURN
2195 2196
    Created read plan
    NULL - Out of memory or no read scan could be built.
2197
*/
2198

2199 2200
static
TABLE_READ_PLAN *get_best_disjunct_quick(PARAM *param, SEL_IMERGE *imerge,
2201
                                         double read_time)
2202 2203 2204 2205 2206 2207 2208
{
  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
2209
  bool imerge_too_expensive= FALSE;
2210 2211 2212 2213
  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
2214 2215
  bool all_scans_ror_able= TRUE;
  bool all_scans_rors= TRUE;
2216 2217 2218 2219 2220 2221 2222 2223 2224
  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));

2225
  if (!(range_scans= (TRP_RANGE**)alloc_root(param->mem_root,
2226 2227 2228
                                             sizeof(TRP_RANGE*)*
                                             n_child_scans)))
    DBUG_RETURN(NULL);
2229
  /*
2230 2231 2232
    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.
2233
  */
2234
  for (ptree= imerge->trees, cur_child= range_scans;
2235
       ptree != imerge->trees_next;
2236
       ptree++, cur_child++)
2237
  {
2238 2239
    DBUG_EXECUTE("info", print_sel_tree(param, *ptree, &(*ptree)->keys_map,
                                        "tree in SEL_IMERGE"););
monty@mysql.com's avatar
monty@mysql.com committed
2240
    if (!(*cur_child= get_key_scans_params(param, *ptree, TRUE, read_time)))
2241 2242
    {
      /*
2243
        One of index scans in this index_merge is more expensive than entire
2244 2245 2246
        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.
2247
      */
monty@mysql.com's avatar
monty@mysql.com committed
2248
      imerge_too_expensive= TRUE;
2249 2250 2251
    }
    if (imerge_too_expensive)
      continue;
2252

2253 2254 2255
    imerge_cost += (*cur_child)->read_cost;
    all_scans_ror_able &= ((*ptree)->n_ror_scans > 0);
    all_scans_rors &= (*cur_child)->is_ror;
2256
    if (pk_is_clustered &&
2257 2258
        param->real_keynr[(*cur_child)->key_idx] ==
        param->table->s->primary_key)
2259
    {
2260 2261
      cpk_scan= cur_child;
      cpk_scan_records= (*cur_child)->records;
2262 2263
    }
    else
2264
      non_cpk_scan_records += (*cur_child)->records;
2265
  }
2266

2267
  DBUG_PRINT("info", ("index_merge scans cost=%g", imerge_cost));
2268
  if (imerge_too_expensive || (imerge_cost > read_time) ||
2269 2270
      (non_cpk_scan_records+cpk_scan_records >= param->table->file->records) &&
      read_time != DBL_MAX)
2271
  {
2272 2273
    /*
      Bail out if it is obvious that both index_merge and ROR-union will be
2274
      more expensive
2275
    */
2276 2277
    DBUG_PRINT("info", ("Sum of index_merge scans is more expensive than "
                        "full table scan, bailing out"));
2278
    DBUG_RETURN(NULL);
2279
  }
2280
  if (all_scans_rors)
2281
  {
2282 2283
    roru_read_plans= (TABLE_READ_PLAN**)range_scans;
    goto skip_to_ror_scan;
2284
  }
2285 2286
  if (cpk_scan)
  {
2287 2288
    /*
      Add one ROWID comparison for each row retrieved on non-CPK scan.  (it
2289 2290 2291
      is done in QUICK_RANGE_SELECT::row_in_ranges)
     */
    imerge_cost += non_cpk_scan_records / TIME_FOR_COMPARE_ROWID;
2292 2293 2294
  }

  /* Calculate cost(rowid_to_row_scan) */
2295
  imerge_cost += get_sweep_read_cost(param, non_cpk_scan_records);
2296
  DBUG_PRINT("info",("index_merge cost with rowid-to-row scan: %g",
2297
                     imerge_cost));
2298 2299
  if (imerge_cost > read_time)
    goto build_ror_index_merge;
2300 2301

  /* Add Unique operations cost */
2302 2303
  unique_calc_buff_size=
    Unique::get_cost_calc_buff_size(non_cpk_scan_records,
2304 2305 2306 2307 2308 2309
                                    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)))
2310
      DBUG_RETURN(NULL);
2311 2312 2313
    param->imerge_cost_buff_size= unique_calc_buff_size;
  }

2314
  imerge_cost +=
2315
    Unique::get_use_cost(param->imerge_cost_buff, non_cpk_scan_records,
2316 2317
                         param->table->file->ref_length,
                         param->thd->variables.sortbuff_size);
2318
  DBUG_PRINT("info",("index_merge total cost: %g (wanted: less then %g)",
2319 2320 2321 2322 2323 2324 2325
                     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;
2326
      imerge_trp->records= min(imerge_trp->records,
2327 2328 2329 2330 2331 2332
                               param->table->file->records);
      imerge_trp->range_scans= range_scans;
      imerge_trp->range_scans_end= range_scans + n_child_scans;
      read_time= imerge_cost;
    }
  }
2333

2334
build_ror_index_merge:
2335 2336
  if (!all_scans_ror_able || param->thd->lex->sql_command == SQLCOM_DELETE)
    DBUG_RETURN(imerge_trp);
2337

2338 2339
  /* Ok, it is possible to build a ROR-union, try it. */
  bool dummy;
2340
  if (!(roru_read_plans=
2341 2342 2343 2344 2345 2346 2347 2348 2349 2350 2351 2352 2353
          (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++)
2354
  {
2355 2356
    /*
      Assume the best ROR scan is the one that has cheapest full-row-retrieval
2357 2358
      scan cost.
      Also accumulate index_only scan costs as we'll need them to calculate
2359 2360 2361 2362 2363 2364 2365
      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->
2366
              read_time(param->real_keynr[(*cur_child)->key_idx], 1,
2367 2368 2369 2370 2371 2372 2373
                        (*cur_child)->records) +
              rows2double((*cur_child)->records) / TIME_FOR_COMPARE;
    }
    else
      cost= read_time;

    TABLE_READ_PLAN *prev_plan= *cur_child;
2374
    if (!(*cur_roru_plan= get_best_ror_intersect(param, *ptree, cost,
2375 2376 2377 2378 2379 2380 2381 2382 2383
                                                 &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
2384 2385
      roru_index_costs +=
        ((TRP_ROR_INTERSECT*)(*cur_roru_plan))->index_scan_costs;
2386
    roru_total_records += (*cur_roru_plan)->records;
2387
    roru_intersect_part *= (*cur_roru_plan)->records /
2388
                           param->table->file->records;
2389
  }
2390

2391 2392
  /*
    rows to retrieve=
2393
      SUM(rows_in_scan_i) - table_rows * PROD(rows_in_scan_i / table_rows).
2394
    This is valid because index_merge construction guarantees that conditions
2395 2396 2397
    in disjunction do not share key parts.
  */
  roru_total_records -= (ha_rows)(roru_intersect_part*
2398 2399 2400
                                  param->table->file->records);
  /* ok, got a ROR read plan for each of the disjuncts
    Calculate cost:
2401 2402 2403 2404 2405 2406
    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.
  */
2407

2408
  double roru_total_cost;
2409 2410 2411
  roru_total_cost= roru_index_costs +
                   rows2double(roru_total_records)*log((double)n_child_scans) /
                   (TIME_FOR_COMPARE_ROWID * M_LN2) +
2412 2413
                   get_sweep_read_cost(param, roru_total_records);

2414
  DBUG_PRINT("info", ("ROR-union: cost %g, %d members", roru_total_cost,
2415 2416 2417 2418 2419 2420 2421 2422 2423 2424 2425 2426 2427 2428
                      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);
2429 2430 2431 2432 2433 2434 2435
}


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

  SYNOPSIS
2436
    get_index_only_read_time()
2437 2438 2439 2440 2441
      param    parameters structure
      records  #of records to read
      keynr    key to read

  NOTES
2442
    It is assumed that we will read trough the whole key range and that all
2443 2444 2445 2446
    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.
2447 2448 2449 2450 2451 2452

  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)
2453 2454
*/

2455
static double get_index_only_read_time(const PARAM* param, ha_rows records,
2456
                                       int keynr)
2457 2458 2459 2460 2461 2462 2463
{
  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);
2464
  return read_time;
2465 2466
}

2467

2468 2469
typedef struct st_ror_scan_info
{
2470 2471 2472 2473 2474
  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. */
2475
  SEL_ARG   *sel_arg;
2476 2477

  /* Fields used in the query and covered by this ROR scan. */
2478 2479
  MY_BITMAP covered_fields;
  uint      used_fields_covered; /* # of set bits in covered_fields */
2480
  int       key_rec_length; /* length of key record (including rowid) */
2481 2482

  /*
2483 2484
    Cost of reading all index records with values in sel_arg intervals set
    (assuming there is no need to access full table records)
2485 2486
  */
  double    index_read_cost;
2487 2488 2489
  uint      first_uncovered_field; /* first unused bit in covered_fields */
  uint      key_components; /* # of parts in the key */
} ROR_SCAN_INFO;
2490 2491 2492


/*
2493
  Create ROR_SCAN_INFO* structure with a single ROR scan on index idx using
2494
  sel_arg set of intervals.
2495

2496 2497
  SYNOPSIS
    make_ror_scan()
2498 2499 2500
      param    Parameter from test_quick_select function
      idx      Index of key in param->keys
      sel_arg  Set of intervals for a given key
2501

2502
  RETURN
2503
    NULL - out of memory
2504
    ROR scan structure containing a scan for {idx, sel_arg}
2505 2506 2507 2508 2509 2510 2511 2512 2513
*/

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");
2514

2515 2516 2517 2518 2519 2520
  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];
2521 2522
  ror_scan->key_rec_length= (param->table->key_info[keynr].key_length +
                             param->table->file->ref_length);
2523 2524
  ror_scan->sel_arg= sel_arg;
  ror_scan->records= param->table->quick_rows[keynr];
2525 2526

  if (!(bitmap_buf= (uchar*)alloc_root(param->mem_root,
2527
                                       param->fields_bitmap_size)))
2528
    DBUG_RETURN(NULL);
2529

2530
  if (bitmap_init(&ror_scan->covered_fields, bitmap_buf,
monty@mysql.com's avatar
monty@mysql.com committed
2531
                  param->fields_bitmap_size*8, FALSE))
2532 2533
    DBUG_RETURN(NULL);
  bitmap_clear_all(&ror_scan->covered_fields);
2534

2535
  KEY_PART_INFO *key_part= param->table->key_info[keynr].key_part;
2536
  KEY_PART_INFO *key_part_end= key_part +
2537 2538 2539 2540 2541 2542
                               param->table->key_info[keynr].key_parts;
  for (;key_part != key_part_end; ++key_part)
  {
    if (bitmap_is_set(&param->needed_fields, key_part->fieldnr))
      bitmap_set_bit(&ror_scan->covered_fields, key_part->fieldnr);
  }
2543
  ror_scan->index_read_cost=
2544 2545 2546 2547 2548 2549
    get_index_only_read_time(param, param->table->quick_rows[ror_scan->keynr],
                             ror_scan->keynr);
  DBUG_RETURN(ror_scan);
}


2550
/*
2551 2552 2553 2554 2555 2556 2557
  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
2558
   -1 a < b
2559 2560
    0 a = b
    1 a > b
2561
*/
2562

2563
static int cmp_ror_scan_info(ROR_SCAN_INFO** a, ROR_SCAN_INFO** b)
2564 2565 2566 2567 2568 2569 2570
{
  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;
}

/*
2571 2572 2573
  Compare two ROR_SCAN_INFO** by
   (#covered fields in F desc,
    #components asc,
2574
    number of first not covered component asc)
2575 2576 2577 2578 2579 2580 2581

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

  RETURN
2582
   -1 a < b
2583 2584
    0 a = b
    1 a > b
2585
*/
2586

2587
static int cmp_ror_scan_info_covering(ROR_SCAN_INFO** a, ROR_SCAN_INFO** b)
2588 2589 2590 2591 2592 2593 2594 2595 2596 2597 2598 2599 2600 2601 2602 2603
{
  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;
}

2604

2605
/* Auxiliary structure for incremental ROR-intersection creation */
2606
typedef struct
2607 2608 2609
{
  const PARAM *param;
  MY_BITMAP covered_fields; /* union of fields covered by all scans */
2610
  /*
2611
    Fraction of table records that satisfies conditions of all scans.
2612
    This is the number of full records that will be retrieved if a
2613 2614
    non-index_only index intersection will be employed.
  */
2615 2616 2617 2618
  double out_rows;
  /* TRUE if covered_fields is a superset of needed_fields */
  bool is_covering;

2619
  ha_rows index_records; /* sum(#records to look in indexes) */
2620 2621
  double index_scan_costs; /* SUM(cost of 'index-only' scans) */
  double total_cost;
2622
} ROR_INTERSECT_INFO;
2623 2624


2625 2626 2627 2628
/*
  Allocate a ROR_INTERSECT_INFO and initialize it to contain zero scans.

  SYNOPSIS
2629 2630 2631
    ror_intersect_init()
      param         Parameter from test_quick_select

2632 2633 2634 2635 2636 2637
  RETURN
    allocated structure
    NULL on error
*/

static
2638
ROR_INTERSECT_INFO* ror_intersect_init(const PARAM *param)
2639 2640 2641
{
  ROR_INTERSECT_INFO *info;
  uchar* buf;
2642
  if (!(info= (ROR_INTERSECT_INFO*)alloc_root(param->mem_root,
2643 2644 2645 2646 2647 2648
                                              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
2649
                  FALSE))
2650
    return NULL;
2651
  info->is_covering= FALSE;
2652
  info->index_scan_costs= 0.0;
2653 2654 2655
  info->index_records= 0;
  info->out_rows= param->table->file->records;
  bitmap_clear_all(&info->covered_fields);
2656 2657 2658
  return info;
}

2659 2660 2661 2662 2663 2664 2665 2666 2667 2668 2669
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;
}
2670 2671


2672
/*
2673
  Get selectivity of a ROR scan wrt ROR-intersection.
2674

2675
  SYNOPSIS
2676 2677 2678 2679
    ror_scan_selectivity()
      info  ROR-interection 
      scan  ROR scan
      
2680
  NOTES
2681
    Suppose we have a condition on several keys
2682 2683
    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
2684
          ...
2685
         k_n1=c_n1 AND k_n3=c_n3 AND ...  (1) //parts of the key used by *scan
2686

2687 2688
    where k_ij may be the same as any k_pq (i.e. keys may have common parts).

2689
    A full row is retrieved if entire condition holds.
2690 2691

    The recursive procedure for finding P(cond) is as follows:
2692

2693
    First step:
2694
    Pick 1st part of 1st key and break conjunction (1) into two parts:
2695 2696
      cond= (k_11=c_11 AND R)

2697
    Here R may still contain condition(s) equivalent to k_11=c_11.
2698 2699
    Nevertheless, the following holds:

2700
      P(k_11=c_11 AND R) = P(k_11=c_11) * P(R | k_11=c_11).
2701 2702 2703 2704 2705

    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:
2706
    We have a set of fixed fields/satisfied conditions) F, probability P(F),
2707 2708 2709
    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).
2710
    Lets denote k_ij as t,  R = t AND R1, where R1 may still contain t. Then
2711

2712
     P((t AND R1)|F) = P(t|F) * P(R1|t|F) = P(t|F) * P(R1|(t AND F)) (2)
2713 2714 2715 2716 2717 2718 2719

    (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

2720 2721
    b) F doesn't contain condition on field used in t. Then F and t are
     considered independent.
2722

2723
     P(t|F) = P(t|(fields_before_t_in_key AND other_fields)) =
2724 2725
          = P(t|fields_before_t_in_key).

2726 2727
     P(t|fields_before_t_in_key) = #records(fields_before_t_in_key) /
                                   #records(fields_before_t_in_key, t)
2728 2729

    The second multiplier is calculated by applying this step recursively.
2730

2731 2732 2733 2734 2735
  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.

2736
    The calculation is conducted as follows:
2737
    Lets denote #records(keypart1, ... keypartK) as n_k. We need to calculate
2738

2739 2740
     n_{k1}      n_{k_2}
    --------- * ---------  * .... (3)
2741
     n_{k1-1}    n_{k2_1}
2742

2743 2744 2745 2746
    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
2747
    as fixed, we calculate (3) as
2748 2749 2750

                                  n_{i1}      n_{i_2}
    (3) = n_{max_key_part}  / (   --------- * ---------  * ....  )
2751 2752 2753 2754
                                  n_{i1-1}    n_{i2_1}

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

2755 2756
    In order to minimize number of expensive records_in_range calls we group
    and reduce adjacent fractions.
2757

2758
  RETURN
2759 2760
    Selectivity of given ROR scan.
    
2761 2762
*/

2763 2764
static double ror_scan_selectivity(const ROR_INTERSECT_INFO *info, 
                                   const ROR_SCAN_INFO *scan)
2765 2766
{
  double selectivity_mult= 1.0;
2767
  KEY_PART_INFO *key_part= info->param->table->key_info[scan->keynr].key_part;
sergefp@mysql.com's avatar
sergefp@mysql.com committed
2768
  byte key_val[MAX_KEY_LENGTH+MAX_FIELD_WIDTH]; /* key values tuple */
2769
  char *key_ptr= (char*) key_val;
2770 2771
  SEL_ARG *sel_arg, *tuple_arg= NULL;
  bool cur_covered;
2772 2773
  bool prev_covered= test(bitmap_is_set(&info->covered_fields,
                                        key_part->fieldnr));
sergefp@mysql.com's avatar
sergefp@mysql.com committed
2774 2775 2776 2777 2778 2779
  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;
2780 2781
  ha_rows prev_records= info->param->table->file->records;
  DBUG_ENTER("ror_intersect_selectivity");
2782 2783 2784

  for (sel_arg= scan->sel_arg; sel_arg;
       sel_arg= sel_arg->next_key_part)
2785
  {
2786
    DBUG_PRINT("info",("sel_arg step"));
2787
    cur_covered= test(bitmap_is_set(&info->covered_fields,
2788
                                    key_part[sel_arg->part].fieldnr));
2789
    if (cur_covered != prev_covered)
2790
    {
2791
      /* create (part1val, ..., part{n-1}val) tuple. */
2792 2793
      ha_rows records;
      if (!tuple_arg)
2794
      {
2795 2796
        tuple_arg= scan->sel_arg;
        /* Here we use the length of the first key part */
2797
        tuple_arg->store_min(key_part->store_length, &key_ptr, 0);
2798 2799 2800 2801
      }
      while (tuple_arg->next_key_part != sel_arg)
      {
        tuple_arg= tuple_arg->next_key_part;
2802
        tuple_arg->store_min(key_part[tuple_arg->part].store_length, &key_ptr, 0);
2803
      }
2804
      min_range.length= max_range.length= ((char*) key_ptr - (char*) key_val);
2805 2806
      records= (info->param->table->file->
                records_in_range(scan->keynr, &min_range, &max_range));
2807 2808 2809 2810 2811 2812 2813 2814 2815 2816 2817
      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 */
2818
        prev_records= records;
2819
      }
2820
    }
2821 2822 2823 2824
    prev_covered= cur_covered;
  }
  if (!prev_covered)
  {
2825
    double tmp= rows2double(info->param->table->quick_rows[scan->keynr]) /
2826 2827
                rows2double(prev_records);
    DBUG_PRINT("info", ("Selectivity multiplier: %g", tmp));
2828
    selectivity_mult *= tmp;
2829
  }
2830 2831 2832
  DBUG_PRINT("info", ("Returning multiplier: %g", selectivity_mult));
  DBUG_RETURN(selectivity_mult);
}
2833

2834

2835 2836 2837 2838 2839 2840 2841 2842 2843 2844 2845 2846 2847 2848 2849 2850 2851 2852 2853 2854 2855 2856 2857 2858 2859 2860 2861 2862 2863 2864 2865 2866 2867 2868 2869 2870 2871
/*
  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,
2872
                              ROR_SCAN_INFO* ror_scan, bool is_cpk_scan)
2873 2874 2875 2876 2877 2878 2879 2880 2881 2882
{
  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);
2883 2884 2885
  if (selectivity_mult == 1.0)
  {
    /* Don't add this scan if it doesn't improve selectivity. */
2886
    DBUG_PRINT("info", ("The scan doesn't improve selectivity."));
2887
    DBUG_RETURN(FALSE);
2888
  }
2889 2890 2891 2892
  
  info->out_rows *= selectivity_mult;
  DBUG_PRINT("info", ("info->total_cost= %g", info->total_cost));
  
2893
  if (is_cpk_scan)
2894
  {
2895 2896 2897 2898 2899 2900
    /*
      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) / 
2901 2902 2903 2904
                              TIME_FOR_COMPARE_ROWID;
  }
  else
  {
2905
    info->index_records += info->param->table->quick_rows[ror_scan->keynr];
2906 2907
    info->index_scan_costs += ror_scan->index_read_cost;
    bitmap_union(&info->covered_fields, &ror_scan->covered_fields);
2908 2909 2910 2911 2912 2913
    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;
    }
2914
  }
2915

2916
  info->total_cost= info->index_scan_costs;
2917
  DBUG_PRINT("info", ("info->total_cost= %g", info->total_cost));
2918 2919
  if (!info->is_covering)
  {
2920 2921 2922
    info->total_cost += 
      get_sweep_read_cost(info->param, double2rows(info->out_rows));
    DBUG_PRINT("info", ("info->total_cost= %g", info->total_cost));
2923
  }
2924
  DBUG_PRINT("info", ("New out_rows= %g", info->out_rows));
2925
  DBUG_PRINT("info", ("New cost= %g, %scovering", info->total_cost,
2926
                      info->is_covering?"" : "non-"));
2927
  DBUG_RETURN(TRUE);
2928 2929
}

2930

2931 2932
/*
  Get best ROR-intersection plan using non-covering ROR-intersection search
2933 2934 2935 2936
  algorithm. The returned plan may be covering.

  SYNOPSIS
    get_best_ror_intersect()
2937 2938 2939
      param            Parameter from test_quick_select function.
      tree             Transformed restriction condition to be used to look
                       for ROR scans.
2940
      read_time        Do not return read plans with cost > read_time.
2941
      are_all_covering [out] set to TRUE if union of all scans covers all
2942 2943
                       fields needed by the query (and it is possible to build
                       a covering ROR-intersection)
2944

2945
  NOTES
2946 2947 2948 2949 2950
    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.
2951

2952
  IMPLEMENTATION
2953
    The approximate best non-covering plan search algorithm is as follows:
2954

2955 2956 2957 2958
    find_min_ror_intersection_scan()
    {
      R= select all ROR scans;
      order R by (E(#records_matched) * key_record_length).
2959

2960 2961 2962 2963 2964 2965
      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)
      {
2966 2967
        firstR= R - first(R);
        if (!selectivity(S + firstR < selectivity(S)))
2968
          continue;
2969
          
2970 2971 2972 2973 2974 2975 2976 2977 2978
        S= S + first(R);
        if (cost(S) < min_cost)
        {
          min_cost= cost(S);
          min_scan= make_scan(S);
        }
      }
      return min_scan;
    }
2979

2980
    See ror_intersect_add function for ROR intersection costs.
2981

2982
    Special handling for Clustered PK scans
2983 2984
    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
2985 2986
    expensive in this case.
    Clustered PK scan has special handling in ROR-intersection: it is not used
2987
    to retrieve rows, instead its condition is used to filter row references
2988
    we get from scans on other keys.
2989 2990

  RETURN
2991
    ROR-intersection table read plan
2992
    NULL if out of memory or no suitable plan found.
2993 2994
*/

2995 2996 2997 2998 2999 3000
static
TRP_ROR_INTERSECT *get_best_ror_intersect(const PARAM *param, SEL_TREE *tree,
                                          double read_time,
                                          bool *are_all_covering)
{
  uint idx;
3001
  double min_cost= DBL_MAX;
3002
  DBUG_ENTER("get_best_ror_intersect");
3003

3004
  if ((tree->n_ror_scans < 2) || !param->table->file->records)
3005
    DBUG_RETURN(NULL);
3006 3007

  /*
3008 3009
    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.
3010
  */
3011
  ROR_SCAN_INFO **cur_ror_scan;
3012
  ROR_SCAN_INFO *cpk_scan= NULL;
3013
  uint cpk_no;
monty@mysql.com's avatar
monty@mysql.com committed
3014
  bool cpk_scan_used= FALSE;
3015

3016 3017 3018 3019
  if (!(tree->ror_scans= (ROR_SCAN_INFO**)alloc_root(param->mem_root,
                                                     sizeof(ROR_SCAN_INFO*)*
                                                     param->keys)))
    return NULL;
3020 3021
  cpk_no= ((param->table->file->primary_key_is_clustered()) ?
           param->table->s->primary_key : MAX_KEY);
3022

3023
  for (idx= 0, cur_ror_scan= tree->ror_scans; idx < param->keys; idx++)
3024
  {
3025
    ROR_SCAN_INFO *scan;
3026
    if (!tree->ror_scans_map.is_set(idx))
3027
      continue;
3028
    if (!(scan= make_ror_scan(param, idx, tree->keys[idx])))
3029
      return NULL;
3030
    if (param->real_keynr[idx] == cpk_no)
3031
    {
3032 3033
      cpk_scan= scan;
      tree->n_ror_scans--;
3034 3035
    }
    else
3036
      *(cur_ror_scan++)= scan;
3037
  }
3038

3039
  tree->ror_scans_end= cur_ror_scan;
3040 3041
  DBUG_EXECUTE("info",print_ror_scans_arr(param->table, "original",
                                          tree->ror_scans,
3042 3043
                                          tree->ror_scans_end););
  /*
3044
    Ok, [ror_scans, ror_scans_end) is array of ptrs to initialized
3045 3046
    ROR_SCAN_INFO's.
    Step 2: Get best ROR-intersection using an approximate algorithm.
3047 3048 3049
  */
  qsort(tree->ror_scans, tree->n_ror_scans, sizeof(ROR_SCAN_INFO*),
        (qsort_cmp)cmp_ror_scan_info);
3050 3051
  DBUG_EXECUTE("info",print_ror_scans_arr(param->table, "ordered",
                                          tree->ror_scans,
3052
                                          tree->ror_scans_end););
3053

3054 3055 3056 3057 3058 3059 3060 3061 3062
  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. */
3063 3064 3065
  ROR_INTERSECT_INFO *intersect, *intersect_best;
  if (!(intersect= ror_intersect_init(param)) || 
      !(intersect_best= ror_intersect_init(param)))
3066
    return NULL;
3067

3068
  /* [intersect_scans,intersect_scans_best) will hold the best intersection */
3069
  ROR_SCAN_INFO **intersect_scans_best;
3070
  cur_ror_scan= tree->ror_scans;
3071
  intersect_scans_best= intersect_scans;
3072
  while (cur_ror_scan != tree->ror_scans_end && !intersect->is_covering)
3073
  {
3074
    /* S= S + first(R);  R= R - first(R); */
3075
    if (!ror_intersect_add(intersect, *cur_ror_scan, FALSE))
3076 3077 3078 3079 3080 3081
    {
      cur_ror_scan++;
      continue;
    }
    
    *(intersect_scans_end++)= *(cur_ror_scan++);
3082

3083
    if (intersect->total_cost < min_cost)
3084
    {
3085
      /* Local minimum found, save it */
3086
      ror_intersect_cpy(intersect_best, intersect);
3087
      intersect_scans_best= intersect_scans_end;
3088
      min_cost = intersect->total_cost;
3089 3090
    }
  }
3091

3092 3093 3094 3095 3096 3097
  if (intersect_scans_best == intersect_scans)
  {
    DBUG_PRINT("info", ("None of scans increase selectivity"));
    DBUG_RETURN(NULL);
  }
    
3098 3099 3100 3101
  DBUG_EXECUTE("info",print_ror_scans_arr(param->table,
                                          "best ROR-intersection",
                                          intersect_scans,
                                          intersect_scans_best););
3102

3103
  *are_all_covering= intersect->is_covering;
3104
  uint best_num= intersect_scans_best - intersect_scans;
3105 3106
  ror_intersect_cpy(intersect, intersect_best);

3107 3108
  /*
    Ok, found the best ROR-intersection of non-CPK key scans.
3109 3110
    Check if we should add a CPK scan. If the obtained ROR-intersection is 
    covering, it doesn't make sense to add CPK scan.
3111 3112
  */
  if (cpk_scan && !intersect->is_covering)
3113
  {
3114
    if (ror_intersect_add(intersect, cpk_scan, TRUE) && 
3115
        (intersect->total_cost < min_cost))
3116
    {
monty@mysql.com's avatar
monty@mysql.com committed
3117
      cpk_scan_used= TRUE;
3118
      intersect_best= intersect; //just set pointer here
3119 3120
    }
  }
3121

3122
  /* Ok, return ROR-intersect plan if we have found one */
3123
  TRP_ROR_INTERSECT *trp= NULL;
3124
  if (min_cost < read_time && (cpk_scan_used || best_num > 1))
3125
  {
3126 3127
    if (!(trp= new (param->mem_root) TRP_ROR_INTERSECT))
      DBUG_RETURN(trp);
3128 3129
    if (!(trp->first_scan=
           (ROR_SCAN_INFO**)alloc_root(param->mem_root,
3130 3131 3132 3133
                                       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;
3134 3135 3136 3137 3138 3139
    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;
3140
    trp->records= best_rows;
3141 3142 3143 3144 3145
    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));
3146
  }
3147
  DBUG_RETURN(trp);
3148 3149 3150 3151
}


/*
3152
  Get best covering ROR-intersection.
3153
  SYNOPSIS
3154
    get_best_covering_ror_intersect()
3155 3156 3157
      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.
3158

3159 3160
  RETURN
    Best covering ROR-intersection plan
3161
    NULL if no plan found.
3162 3163

  NOTES
3164
    get_best_ror_intersect must be called for a tree before calling this
3165
    function for it.
3166
    This function invalidates tree->ror_scans member values.
3167

3168 3169 3170 3171 3172 3173 3174 3175 3176 3177 3178 3179 3180
  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.
3181 3182
*/

3183
static
3184 3185
TRP_ROR_INTERSECT *get_best_covering_ror_intersect(PARAM *param,
                                                   SEL_TREE *tree,
3186
                                                   double read_time)
3187
{
3188
  ROR_SCAN_INFO **ror_scan_mark;
3189
  ROR_SCAN_INFO **ror_scans_end= tree->ror_scans_end;
3190 3191 3192 3193
  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)
3194
    (*scan)->key_components=
3195
      param->table->key_info[(*scan)->keynr].key_parts;
3196

3197 3198
  /*
    Run covering-ROR-search algorithm.
3199
    Assume set I is [ror_scan .. ror_scans_end)
3200
  */
3201

3202 3203
  /*I=set of all covering indexes */
  ror_scan_mark= tree->ror_scans;
3204

3205 3206 3207 3208 3209 3210
  MY_BITMAP *covered_fields= &param->tmp_covered_fields;
  if (!covered_fields->bitmap) 
    covered_fields->bitmap= (uchar*)alloc_root(param->mem_root,
                                               param->fields_bitmap_size);
  if (!covered_fields->bitmap ||
      bitmap_init(covered_fields, covered_fields->bitmap, nbits, FALSE))
3211
    DBUG_RETURN(0);
3212
  bitmap_clear_all(covered_fields);
3213 3214 3215

  double total_cost= 0.0f;
  ha_rows records=0;
3216 3217
  bool all_covered;

3218 3219 3220 3221 3222 3223
  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 {
    /*
3224
      Update changed sorting info:
3225
        #covered fields,
3226
	number of first not covered component
3227 3228 3229 3230
      Calculate and save these values for each of remaining scans.
    */
    for (ROR_SCAN_INFO **scan= ror_scan_mark; scan != ror_scans_end; ++scan)
    {
3231
      bitmap_subtract(&(*scan)->covered_fields, covered_fields);
3232
      (*scan)->used_fields_covered=
3233
        bitmap_bits_set(&(*scan)->covered_fields);
3234
      (*scan)->first_uncovered_field=
3235 3236 3237 3238 3239 3240 3241 3242 3243
        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););
3244

3245 3246 3247
    /* I=I-first(I) */
    total_cost += (*ror_scan_mark)->index_read_cost;
    records += (*ror_scan_mark)->records;
3248
    DBUG_PRINT("info", ("Adding scan on %s",
3249 3250 3251 3252
                        param->table->key_info[(*ror_scan_mark)->keynr].name));
    if (total_cost > read_time)
      DBUG_RETURN(NULL);
    /* F=F-covered by first(I) */
3253 3254
    bitmap_union(covered_fields, &(*ror_scan_mark)->covered_fields);
    all_covered= bitmap_is_subset(&param->needed_fields, covered_fields);
3255 3256 3257 3258
  } while ((++ror_scan_mark < ror_scans_end) && !all_covered);
  
  if (!all_covered || (ror_scan_mark - tree->ror_scans) == 1)
    DBUG_RETURN(NULL);
3259 3260 3261 3262 3263 3264 3265 3266 3267

  /*
    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););
3268

3269
  /* Add priority queue use cost. */
3270 3271
  total_cost += rows2double(records)*
                log((double)(ror_scan_mark - tree->ror_scans)) /
3272 3273 3274 3275 3276 3277 3278 3279 3280 3281 3282 3283 3284 3285
                (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);
3286
  memcpy(trp->first_scan, tree->ror_scans, best_num*sizeof(ROR_SCAN_INFO*));
3287
  trp->last_scan=  trp->first_scan + best_num;
monty@mysql.com's avatar
monty@mysql.com committed
3288
  trp->is_covering= TRUE;
3289 3290
  trp->read_cost= total_cost;
  trp->records= records;
3291
  trp->cpk_scan= NULL;
3292

3293 3294 3295
  DBUG_PRINT("info",
             ("Returning covering ROR-intersect plan: cost %g, records %lu",
              trp->read_cost, (ulong) trp->records));
3296
  DBUG_RETURN(trp);
3297 3298 3299
}


3300
/*
3301
  Get best "range" table read plan for given SEL_TREE.
3302
  Also update PARAM members and store ROR scans info in the SEL_TREE.
3303
  SYNOPSIS
3304
    get_key_scans_params
3305
      param        parameters from test_quick_select
3306
      tree         make range select for this SEL_TREE
monty@mysql.com's avatar
monty@mysql.com committed
3307
      index_read_must_be_used if TRUE, assume 'index only' option will be set
3308
                             (except for clustered PK indexes)
3309 3310
      read_time    don't create read plans with cost > read_time.
  RETURN
3311
    Best range read plan
3312
    NULL if no plan found or error occurred
3313 3314
*/

3315
static TRP_RANGE *get_key_scans_params(PARAM *param, SEL_TREE *tree,
3316
                                       bool index_read_must_be_used,
3317
                                       double read_time)
3318 3319
{
  int idx;
3320 3321 3322
  SEL_ARG **key,**end, **key_to_read= NULL;
  ha_rows best_records;
  TRP_RANGE* read_plan= NULL;
3323
  bool pk_is_clustered= param->table->file->primary_key_is_clustered();
3324 3325
  DBUG_ENTER("get_key_scans_params");
  LINT_INIT(best_records); /* protected by key_to_read */
3326
  /*
3327 3328
    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
3329
    is defined as "not null".
3330 3331
  */
  DBUG_EXECUTE("info", print_sel_tree(param, tree, &tree->keys_map,
3332 3333 3334 3335
                                      "tree scans"););
  tree->ror_scans_map.clear_all();
  tree->n_ror_scans= 0;
  for (idx= 0,key=tree->keys, end=key+param->keys;
3336 3337 3338 3339 3340 3341 3342
       key != end ;
       key++,idx++)
  {
    ha_rows found_records;
    double found_read_time;
    if (*key)
    {
3343
      uint keynr= param->real_keynr[idx];
3344 3345
      if ((*key)->type == SEL_ARG::MAYBE_KEY ||
          (*key)->maybe_flag)
3346
        param->needed_reg->set_bit(keynr);
3347

monty@mysql.com's avatar
monty@mysql.com committed
3348 3349
      bool read_index_only= index_read_must_be_used ? TRUE :
                            (bool) param->table->used_keys.is_set(keynr);
3350

3351 3352 3353 3354 3355 3356
      found_records= check_quick_select(param, idx, *key);
      if (param->is_ror_scan)
      {
        tree->n_ror_scans++;
        tree->ror_scans_map.set_bit(idx);
      }
3357
      double cpu_cost= (double) found_records / TIME_FOR_COMPARE;
3358
      if (found_records != HA_POS_ERROR && found_records > 2 &&
sergefp@mysql.com's avatar
sergefp@mysql.com committed
3359
          read_index_only &&
monty@mysql.com's avatar
monty@mysql.com committed
3360
          (param->table->file->index_flags(keynr, param->max_key_part,1) &
monty@mysql.com's avatar
monty@mysql.com committed
3361
           HA_KEYREAD_ONLY) &&
3362
          !(pk_is_clustered && keynr == param->table->s->primary_key))
3363 3364 3365 3366 3367
      {
        /*
          We can resolve this by only reading through this key. 
          0.01 is added to avoid races between range and 'index' scan.
        */
3368
        found_read_time= get_index_only_read_time(param,found_records,keynr) +
3369 3370
                         cpu_cost + 0.01;
      }
3371
      else
3372
      {
3373
        /*
3374 3375 3376
          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.
        */
3377 3378 3379
	found_read_time= param->table->file->read_time(keynr,
                                                       param->range_count,
                                                       found_records) +
3380 3381
			 cpu_cost + 0.01;
      }
3382 3383 3384
      DBUG_PRINT("info",("key %s: found_read_time: %g (cur. read_time: %g)",
                         param->table->key_info[keynr].name, found_read_time,
                         read_time));
3385

3386 3387
      if (read_time > found_read_time && found_records != HA_POS_ERROR
          /*|| read_time == DBL_MAX*/ )
3388
      {
3389
        read_time=    found_read_time;
3390
        best_records= found_records;
3391 3392 3393 3394 3395 3396 3397 3398 3399 3400 3401 3402 3403 3404 3405 3406
        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;
3407 3408 3409 3410
      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));
3411 3412 3413 3414 3415 3416 3417 3418 3419
    }
  }
  else
    DBUG_PRINT("info", ("No 'range' table read plan found"));

  DBUG_RETURN(read_plan);
}


3420
QUICK_SELECT_I *TRP_INDEX_MERGE::make_quick(PARAM *param,
3421 3422 3423 3424 3425 3426 3427 3428 3429 3430 3431
                                            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;
3432 3433
  for (TRP_RANGE **range_scan= range_scans; range_scan != range_scans_end;
       range_scan++)
3434 3435
  {
    if (!(quick= (QUICK_RANGE_SELECT*)
3436
          ((*range_scan)->make_quick(param, FALSE, &quick_imerge->alloc)))||
3437 3438 3439 3440 3441 3442 3443 3444 3445 3446
        quick_imerge->push_quick_back(quick))
    {
      delete quick;
      delete quick_imerge;
      return NULL;
    }
  }
  return quick_imerge;
}

3447
QUICK_SELECT_I *TRP_ROR_INTERSECT::make_quick(PARAM *param,
3448 3449 3450 3451 3452 3453 3454
                                              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;
3455 3456

  if ((quick_intrsect=
3457
         new QUICK_ROR_INTERSECT_SELECT(param->thd, param->table,
monty@mysql.com's avatar
monty@mysql.com committed
3458
                                        retrieve_full_rows? (!is_covering):FALSE,
3459 3460
                                        parent_alloc)))
  {
3461
    DBUG_EXECUTE("info", print_ror_scans_arr(param->table,
3462 3463 3464
                                             "creating ROR-intersect",
                                             first_scan, last_scan););
    alloc= parent_alloc? parent_alloc: &quick_intrsect->alloc;
3465
    for (; first_scan != last_scan;++first_scan)
3466 3467 3468 3469
    {
      if (!(quick= get_quick_select(param, (*first_scan)->idx,
                                    (*first_scan)->sel_arg, alloc)) ||
          quick_intrsect->push_quick_back(quick))
3470
      {
3471 3472
        delete quick_intrsect;
        DBUG_RETURN(NULL);
3473 3474
      }
    }
3475 3476 3477 3478
    if (cpk_scan)
    {
      if (!(quick= get_quick_select(param, cpk_scan->idx,
                                    cpk_scan->sel_arg, alloc)))
3479
      {
3480 3481
        delete quick_intrsect;
        DBUG_RETURN(NULL);
3482
      }
3483
      quick->file= NULL; 
3484
      quick_intrsect->cpk_quick= quick;
3485
    }
3486
    quick_intrsect->records= records;
3487
    quick_intrsect->read_time= read_cost;
3488
  }
3489 3490 3491
  DBUG_RETURN(quick_intrsect);
}

3492

3493
QUICK_SELECT_I *TRP_ROR_UNION::make_quick(PARAM *param,
3494 3495 3496 3497 3498 3499 3500
                                          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");
3501 3502
  /*
    It is impossible to construct a ROR-union that will not retrieve full
3503
    rows, ignore retrieve_full_rows parameter.
3504 3505 3506
  */
  if ((quick_roru= new QUICK_ROR_UNION_SELECT(param->thd, param->table)))
  {
3507
    for (scan= first_ror; scan != last_ror; scan++)
3508
    {
3509
      if (!(quick= (*scan)->make_quick(param, FALSE, &quick_roru->alloc)) ||
3510 3511 3512 3513 3514
          quick_roru->push_quick_back(quick))
        DBUG_RETURN(NULL);
    }
    quick_roru->records= records;
    quick_roru->read_time= read_cost;
3515
  }
3516
  DBUG_RETURN(quick_roru);
3517 3518
}

3519

3520
/*
monty@mysql.com's avatar
monty@mysql.com committed
3521
  Build a SEL_TREE for <> or NOT BETWEEN predicate
3522 3523 3524 3525 3526 3527
 
  SYNOPSIS
    get_ne_mm_tree()
      param       PARAM from SQL_SELECT::test_quick_select
      cond_func   item for the predicate
      field       field in the predicate
monty@mysql.com's avatar
monty@mysql.com committed
3528 3529
      lt_value    constant that field should be smaller
      gt_value    constant that field should be greaterr
3530 3531 3532
      cmp_type    compare type for the field

  RETURN 
monty@mysql.com's avatar
monty@mysql.com committed
3533 3534
    #  Pointer to tree built tree
    0  on error
3535 3536 3537
*/

static SEL_TREE *get_ne_mm_tree(PARAM *param, Item_func *cond_func, 
monty@mysql.com's avatar
monty@mysql.com committed
3538 3539
                                Field *field,
                                Item *lt_value, Item *gt_value,
3540 3541
                                Item_result cmp_type)
{
monty@mysql.com's avatar
monty@mysql.com committed
3542
  SEL_TREE *tree;
3543
  tree= get_mm_parts(param, cond_func, field, Item_func::LT_FUNC,
monty@mysql.com's avatar
monty@mysql.com committed
3544
                     lt_value, cmp_type);
3545 3546 3547 3548
  if (tree)
  {
    tree= tree_or(param, tree, get_mm_parts(param, cond_func, field,
					    Item_func::GT_FUNC,
monty@mysql.com's avatar
monty@mysql.com committed
3549
					    gt_value, cmp_type));
3550 3551 3552 3553 3554
  }
  return tree;
}
   

igor@rurik.mysql.com's avatar
igor@rurik.mysql.com committed
3555 3556 3557 3558 3559 3560 3561 3562 3563 3564
/*
  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
3565
      inv         TRUE <> NOT cond_func is considered
igor@rurik.mysql.com's avatar
igor@rurik.mysql.com committed
3566
                  (makes sense only when cond_func is BETWEEN or IN) 
igor@rurik.mysql.com's avatar
igor@rurik.mysql.com committed
3567 3568

  RETURN 
igor@rurik.mysql.com's avatar
igor@rurik.mysql.com committed
3569
    Pointer to the tree built tree
igor@rurik.mysql.com's avatar
igor@rurik.mysql.com committed
3570 3571
*/

3572 3573
static SEL_TREE *get_func_mm_tree(PARAM *param, Item_func *cond_func, 
                                  Field *field, Item *value,
3574
                                  Item_result cmp_type, bool inv)
3575 3576 3577 3578
{
  SEL_TREE *tree= 0;
  DBUG_ENTER("get_func_mm_tree");

igor@rurik.mysql.com's avatar
igor@rurik.mysql.com committed
3579
  switch (cond_func->functype()) {
3580

igor@rurik.mysql.com's avatar
igor@rurik.mysql.com committed
3581
  case Item_func::NE_FUNC:
monty@mysql.com's avatar
monty@mysql.com committed
3582
    tree= get_ne_mm_tree(param, cond_func, field, value, value, cmp_type);
igor@rurik.mysql.com's avatar
igor@rurik.mysql.com committed
3583
    break;
3584

igor@rurik.mysql.com's avatar
igor@rurik.mysql.com committed
3585
  case Item_func::BETWEEN:
igor@rurik.mysql.com's avatar
igor@rurik.mysql.com committed
3586
  {
evgen@moonbone.local's avatar
evgen@moonbone.local committed
3587
    if (!value)
3588
    {
igor@rurik.mysql.com's avatar
igor@rurik.mysql.com committed
3589 3590 3591 3592 3593 3594
      if (inv)
      {
        tree= get_ne_mm_tree(param, cond_func, field, cond_func->arguments()[1],
                             cond_func->arguments()[2], cmp_type);
      }
      else
3595
      {
igor@rurik.mysql.com's avatar
igor@rurik.mysql.com committed
3596 3597 3598 3599 3600 3601 3602 3603 3604
        tree= get_mm_parts(param, cond_func, field, Item_func::GE_FUNC,
		           cond_func->arguments()[1],cmp_type);
        if (tree)
        {
          tree= tree_and(param, tree, get_mm_parts(param, cond_func, field,
					           Item_func::LE_FUNC,
					           cond_func->arguments()[2],
                                                   cmp_type));
        }
3605
      }
3606
    }
igor@rurik.mysql.com's avatar
igor@rurik.mysql.com committed
3607 3608 3609
    else
      tree= get_mm_parts(param, cond_func, field,
                         (inv ?
evgen@sunlight.local's avatar
evgen@sunlight.local committed
3610 3611 3612 3613
                          (value == (Item*)1 ? Item_func::GT_FUNC :
                                               Item_func::LT_FUNC):
                          (value == (Item*)1 ? Item_func::LE_FUNC :
                                               Item_func::GE_FUNC)),
igor@rurik.mysql.com's avatar
igor@rurik.mysql.com committed
3614
                         cond_func->arguments()[0], cmp_type);
igor@rurik.mysql.com's avatar
igor@rurik.mysql.com committed
3615
    break;
igor@rurik.mysql.com's avatar
igor@rurik.mysql.com committed
3616
  }
igor@rurik.mysql.com's avatar
igor@rurik.mysql.com committed
3617
  case Item_func::IN_FUNC:
3618 3619
  {
    Item_func_in *func=(Item_func_in*) cond_func;
3620 3621

    if (inv)
3622
    {
3623
      if (func->array && func->cmp_type != ROW_RESULT)
3624
      {
3625
        /*
3626 3627 3628
          We get here for conditions in form "t.key NOT IN (c1, c2, ...)",
          where c{i} are constants. Our goal is to produce a SEL_TREE that 
          represents intervals:
3629 3630 3631 3632 3633
          
          ($MIN<t.key<c1) OR (c1<t.key<c2) OR (c2<t.key<c3) OR ...    (*)
          
          where $MIN is either "-inf" or NULL.
          
3634 3635 3636 3637 3638 3639 3640 3641 3642 3643 3644 3645 3646 3647 3648 3649 3650
          The most straightforward way to produce it is to convert NOT IN
          into "(t.key != c1) AND (t.key != c2) AND ... " and let the range
          analyzer to build SEL_TREE from that. The problem is that the
          range analyzer will use O(N^2) memory (which is probably a bug),
          and people do use big NOT IN lists (e.g. see BUG#15872, BUG#21282),
          will run out of memory.

          Another problem with big lists like (*) is that a big list is
          unlikely to produce a good "range" access, while considering that
          range access will require expensive CPU calculations (and for 
          MyISAM even index accesses). In short, big NOT IN lists are rarely
          worth analyzing.

          Considering the above, we'll handle NOT IN as follows:
          * if the number of entries in the NOT IN list is less than
            NOT_IN_IGNORE_THRESHOLD, construct the SEL_TREE (*) manually.
          * Otherwise, don't produce a SEL_TREE.
3651
        */
3652
#define NOT_IN_IGNORE_THRESHOLD 1000
3653 3654
        MEM_ROOT *tmp_root= param->mem_root;
        param->thd->mem_root= param->old_root;
3655 3656 3657 3658 3659 3660 3661 3662 3663 3664 3665
        /* 
          Create one Item_type constant object. We'll need it as
          get_mm_parts only accepts constant values wrapped in Item_Type
          objects.
          We create the Item on param->mem_root which points to
          per-statement mem_root (while thd->mem_root is currently pointing
          to mem_root local to range optimizer).
        */
        Item *value_item= func->array->create_item();
        param->thd->mem_root= tmp_root;

3666
        if (func->array->count > NOT_IN_IGNORE_THRESHOLD || !value_item)
3667
          break;
3668

3669
        /* Get a SEL_TREE for "(-inf|NULL) < X < c_0" interval.  */
3670 3671 3672 3673 3674 3675 3676 3677 3678 3679 3680 3681 3682 3683 3684
        uint i=0;
        do 
        {
          func->array->value_to_item(i, value_item);
          tree= get_mm_parts(param, cond_func, field, Item_func::LT_FUNC,
                             value_item, cmp_type);
          if (!tree)
            break;
          i++;
        } while (i < func->array->count && tree->type == SEL_TREE::IMPOSSIBLE);

        if (!tree || tree->type == SEL_TREE::IMPOSSIBLE)
        {
          /* We get here in cases like "t.unsigned NOT IN (-1,-2,-3) */
          tree= NULL;
3685
          break;
3686
        }
3687
        SEL_TREE *tree2;
3688
        for (; i < func->array->count; i++)
3689
        {
3690
          if (func->array->compare_elems(i, i-1))
3691
          {
3692 3693 3694 3695 3696
            /* Get a SEL_TREE for "-inf < X < c_i" interval */
            func->array->value_to_item(i, value_item);
            tree2= get_mm_parts(param, cond_func, field, Item_func::LT_FUNC,
                                value_item, cmp_type);
            if (!tree2)
3697
            {
3698 3699 3700
              tree= NULL;
              break;
            }
3701

3702 3703 3704 3705
            /* Change all intervals to be "c_{i-1} < X < c_i" */
            for (uint idx= 0; idx < param->keys; idx++)
            {
              SEL_ARG *new_interval, *last_val;
3706 3707
              if (((new_interval= tree2->keys[idx])) &&
                  (tree->keys[idx]) &&
3708
                  ((last_val= tree->keys[idx]->last())))
3709
              {
3710 3711
                new_interval->min_value= last_val->max_value;
                new_interval->min_flag= NEAR_MIN;
3712 3713
              }
            }
3714 3715 3716 3717 3718
            /* 
              The following doesn't try to allocate memory so no need to
              check for NULL.
            */
            tree= tree_or(param, tree, tree2);
3719 3720
          }
        }
3721 3722 3723 3724 3725 3726 3727 3728 3729 3730 3731
        
        if (tree && tree->type != SEL_TREE::IMPOSSIBLE)
        {
          /* 
            Get the SEL_TREE for the last "c_last < X < +inf" interval 
            (value_item cotains c_last already)
          */
          tree2= get_mm_parts(param, cond_func, field, Item_func::GT_FUNC,
                              value_item, cmp_type);
          tree= tree_or(param, tree, tree2);
        }
3732 3733 3734 3735 3736 3737 3738
      }
      else
      {
        tree= get_ne_mm_tree(param, cond_func, field,
                             func->arguments()[1], func->arguments()[1],
                             cmp_type);
        if (tree)
3739
        {
3740 3741 3742 3743 3744 3745 3746
          Item **arg, **end;
          for (arg= func->arguments()+2, end= arg+func->argument_count()-2;
               arg < end ; arg++)
          {
            tree=  tree_and(param, tree, get_ne_mm_tree(param, cond_func, field, 
                                                        *arg, *arg, cmp_type));
          }
3747 3748 3749 3750 3751 3752 3753 3754 3755 3756 3757 3758 3759 3760 3761 3762 3763
        }
      }
    }
    else
    {    
      tree= get_mm_parts(param, cond_func, field, Item_func::EQ_FUNC,
                         func->arguments()[1], cmp_type);
      if (tree)
      {
        Item **arg, **end;
        for (arg= func->arguments()+2, end= arg+func->argument_count()-2;
             arg < end ; arg++)
        {
          tree= tree_or(param, tree, get_mm_parts(param, cond_func, field, 
                                                  Item_func::EQ_FUNC,
                                                  *arg, cmp_type));
        }
3764 3765
      }
    }
igor@rurik.mysql.com's avatar
igor@rurik.mysql.com committed
3766
    break;
3767
  }
igor@rurik.mysql.com's avatar
igor@rurik.mysql.com committed
3768
  default: 
3769
  {
igor@rurik.mysql.com's avatar
igor@rurik.mysql.com committed
3770 3771 3772 3773 3774 3775 3776
    /* 
       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.
    */
3777 3778 3779
    Item_func::Functype func_type=
      (value != cond_func->arguments()[0]) ? cond_func->functype() :
        ((Item_bool_func2*) cond_func)->rev_functype();
3780
    tree= get_mm_parts(param, cond_func, field, func_type, value, cmp_type);
3781
  }
igor@rurik.mysql.com's avatar
igor@rurik.mysql.com committed
3782 3783
  }

3784 3785 3786
  DBUG_RETURN(tree);
}

igor@rurik.mysql.com's avatar
igor@rurik.mysql.com committed
3787 3788 3789 3790 3791 3792 3793 3794 3795 3796 3797 3798 3799 3800 3801 3802 3803 3804 3805 3806 3807 3808 3809 3810 3811 3812 3813 3814 3815 3816 3817 3818 3819 3820 3821 3822 3823 3824 3825 3826 3827 3828 3829 3830 3831 3832 3833 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 3895 3896 3897 3898

/*
  Build conjunction of all SEL_TREEs for a simple predicate applying equalities
 
  SYNOPSIS
    get_full_func_mm_tree()
      param       PARAM from SQL_SELECT::test_quick_select
      cond_func   item for the predicate
      field_item  field in the predicate
      value       constant in the predicate
                  (for BETWEEN it contains the number of the field argument,
                   for IN it's always 0) 
      inv         TRUE <> NOT cond_func is considered
                  (makes sense only when cond_func is BETWEEN or IN)

  DESCRIPTION
    For a simple SARGable predicate of the form (f op c), where f is a field and
    c is a constant, the function builds a conjunction of all SEL_TREES that can
    be obtained by the substitution of f for all different fields equal to f.

  NOTES  
    If the WHERE condition contains a predicate (fi op c),
    then not only SELL_TREE for this predicate is built, but
    the trees for the results of substitution of fi for
    each fj belonging to the same multiple equality as fi
    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.

    A BETWEEN predicate of the form (fi [NOT] BETWEEN c1 AND c2) is treated
    in a similar way: we build a conjuction of trees for the results
    of all substitutions of fi for equal fj.
    Yet a predicate of the form (c BETWEEN f1i AND f2i) is processed
    differently. It is considered as a conjuction of two SARGable
    predicates (f1i <= c) and (f2i <=c) and the function get_full_func_mm_tree
    is called for each of them separately producing trees for 
       AND j (f1j <=c ) and AND j (f2j <= c) 
    After this these two trees are united in one conjunctive tree.
    It's easy to see that the same tree is obtained for
       AND j,k (f1j <=c AND f2k<=c)
    which is equivalent to 
       AND j,k (c BETWEEN f1j AND f2k).
    The validity of the processing of the predicate (c NOT BETWEEN f1i AND f2i)
    which equivalent to (f1i > c OR f2i < c) is not so obvious. Here the
    function get_full_func_mm_tree is called for (f1i > c) and (f2i < c)
    producing trees for AND j (f1j > c) and AND j (f2j < c). Then this two
    trees are united in one OR-tree. The expression 
      (AND j (f1j > c) OR AND j (f2j < c)
    is equivalent to the expression
      AND j,k (f1j > c OR f2k < c) 
    which is just a translation of 
      AND j,k (c NOT BETWEEN f1j AND f2k)

    In the cases when one of the items f1, f2 is a constant c1 we do not create
    a tree for it at all. It works for BETWEEN predicates but does not
    work for NOT BETWEEN predicates as we have to evaluate the expression
    with it. If it is TRUE then the other tree can be completely ignored.
    We do not do it now and no trees are built in these cases for
    NOT BETWEEN predicates.

    As to IN predicates only ones of the form (f IN (c1,...,cn)),
    where f1 is a field and c1,...,cn are constant, are considered as
    SARGable. We never try to narrow the index scan using predicates of
    the form (c IN (c1,...,f,...,cn)). 
      
  RETURN 
    Pointer to the tree representing the built conjunction of SEL_TREEs
*/

static SEL_TREE *get_full_func_mm_tree(PARAM *param, Item_func *cond_func, 
                                       Item_field *field_item, Item *value, 
                                       bool inv)
{
  SEL_TREE *tree= 0;
  SEL_TREE *ftree= 0;
  table_map ref_tables= 0;
  table_map param_comp= ~(param->prev_tables | param->read_tables |
		          param->current_table);
  DBUG_ENTER("get_full_func_mm_tree");

  for (uint i= 0; i < cond_func->arg_count; i++)
  {
    Item *arg= cond_func->arguments()[i]->real_item();
    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, inv);
  Item_equal *item_equal= field_item->item_equal;
  if (item_equal)
  {
    Item_equal_iterator it(*item_equal);
    Item_field *item;
    while ((item= it++))
    {
      Field *f= item->field;
      if (field->eq(f))
        continue;
      if (!((ref_tables | f->table->map) & param_comp))
      {
        tree= get_func_mm_tree(param, cond_func, f, value, cmp_type, inv);
        ftree= !ftree ? tree : tree_and(param, ftree, tree);
      }
    }
  }
  DBUG_RETURN(ftree);
}

bk@work.mysql.com's avatar
bk@work.mysql.com committed
3899 3900 3901 3902 3903
	/* make a select tree of all keys in condition */

static SEL_TREE *get_mm_tree(PARAM *param,COND *cond)
{
  SEL_TREE *tree=0;
3904 3905
  SEL_TREE *ftree= 0;
  Item_field *field_item= 0;
3906
  bool inv= FALSE;
igor@rurik.mysql.com's avatar
igor@rurik.mysql.com committed
3907
  Item *value= 0;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3908 3909 3910 3911 3912 3913 3914 3915 3916 3917 3918 3919 3920
  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);
3921
	if (param->thd->is_fatal_error)
3922
	  DBUG_RETURN(0);	// out of memory
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3923 3924 3925 3926 3927 3928 3929 3930 3931 3932 3933 3934 3935 3936 3937
	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)
3938
	    DBUG_RETURN(0);	// out of memory
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3939 3940 3941 3942 3943 3944 3945 3946 3947 3948 3949
	  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())
  {
3950 3951 3952 3953 3954 3955 3956 3957 3958 3959 3960 3961
    /*
      During the cond->val_int() evaluation we can come across a subselect 
      item which may allocate memory on the thd->mem_root and assumes 
      all the memory allocated has the same life span as the subselect 
      item itself. So we have to restore the thread's mem_root here.
    */
    MEM_ROOT *tmp_root= param->mem_root;
    param->thd->mem_root= param->old_root;
    tree= cond->val_int() ? new(tmp_root) SEL_TREE(SEL_TREE::ALWAYS) :
                            new(tmp_root) SEL_TREE(SEL_TREE::IMPOSSIBLE);
    param->thd->mem_root= tmp_root;
    DBUG_RETURN(tree);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3962
  }
3963

3964 3965 3966
  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
3967 3968
  if (cond->type() != Item::FUNC_ITEM)
  {						// Should be a field
3969
    ref_tables= cond->used_tables();
3970 3971
    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
3972 3973 3974
      DBUG_RETURN(0);
    DBUG_RETURN(new SEL_TREE(SEL_TREE::MAYBE));
  }
3975

bk@work.mysql.com's avatar
bk@work.mysql.com committed
3976
  Item_func *cond_func= (Item_func*) cond;
3977 3978 3979
  if (cond_func->functype() == Item_func::BETWEEN ||
      cond_func->functype() == Item_func::IN_FUNC)
    inv= ((Item_func_opt_neg *) cond_func)->negated;
3980
  else if (cond_func->select_optimize() == Item_func::OPTIMIZE_NONE)
igor@rurik.mysql.com's avatar
igor@rurik.mysql.com committed
3981
    DBUG_RETURN(0);			       
3982

igor@rurik.mysql.com's avatar
igor@rurik.mysql.com committed
3983 3984
  param->cond= cond;

igor@rurik.mysql.com's avatar
igor@rurik.mysql.com committed
3985 3986
  switch (cond_func->functype()) {
  case Item_func::BETWEEN:
igor@rurik.mysql.com's avatar
igor@rurik.mysql.com committed
3987 3988 3989 3990 3991 3992 3993 3994 3995 3996 3997 3998 3999 4000 4001 4002 4003 4004 4005 4006 4007 4008 4009 4010 4011 4012 4013 4014 4015 4016 4017
    if (cond_func->arguments()[0]->real_item()->type() == Item::FIELD_ITEM)
    {
      field_item= (Item_field*) (cond_func->arguments()[0]->real_item());
      ftree= get_full_func_mm_tree(param, cond_func, field_item, NULL, inv);
    }

    /*
      Concerning the code below see the NOTES section in
      the comments for the function get_full_func_mm_tree()
    */
    for (uint i= 1 ; i < cond_func->arg_count ; i++)
    {
     
      if (cond_func->arguments()[i]->real_item()->type() == Item::FIELD_ITEM)
      {
        field_item= (Item_field*) (cond_func->arguments()[i]->real_item());
        SEL_TREE *tmp= get_full_func_mm_tree(param, cond_func, 
                                    field_item, (Item*) i, inv);
        if (inv)
          tree= !tree ? tmp : tree_or(param, tree, tmp);
        else 
          tree= tree_and(param, tree, tmp);
      }
      else if (inv)
      { 
        tree= 0;
        break;
      }
    }

    ftree = tree_and(param, ftree, tree);
igor@rurik.mysql.com's avatar
igor@rurik.mysql.com committed
4018 4019
    break;
  case Item_func::IN_FUNC:
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4020 4021
  {
    Item_func_in *func=(Item_func_in*) cond_func;
4022
    if (func->key_item()->real_item()->type() != Item::FIELD_ITEM)
4023
      DBUG_RETURN(0);
4024
    field_item= (Item_field*) (func->key_item()->real_item());
igor@rurik.mysql.com's avatar
igor@rurik.mysql.com committed
4025
    ftree= get_full_func_mm_tree(param, cond_func, field_item, NULL, inv);
igor@rurik.mysql.com's avatar
igor@rurik.mysql.com committed
4026
    break;
4027
  }
igor@rurik.mysql.com's avatar
igor@rurik.mysql.com committed
4028
  case Item_func::MULT_EQUAL_FUNC:
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4029
  {
4030 4031
    Item_equal *item_equal= (Item_equal *) cond;    
    if (!(value= item_equal->get_const()))
igor@rurik.mysql.com's avatar
igor@rurik.mysql.com committed
4032 4033 4034 4035
      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
4036
    {
igor@rurik.mysql.com's avatar
igor@rurik.mysql.com committed
4037 4038 4039
      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
4040
      {
4041
        tree= get_mm_parts(param, cond, field, Item_func::EQ_FUNC,
igor@rurik.mysql.com's avatar
igor@rurik.mysql.com committed
4042 4043
		           value,cmp_type);
        ftree= !ftree ? tree : tree_and(param, ftree, tree);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4044 4045
      }
    }
igor@rurik.mysql.com's avatar
igor@rurik.mysql.com committed
4046
    
4047
    DBUG_RETURN(ftree);
4048 4049
  }
  default:
4050
    if (cond_func->arguments()[0]->real_item()->type() == Item::FIELD_ITEM)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4051
    {
4052
      field_item= (Item_field*) (cond_func->arguments()[0]->real_item());
4053
      value= cond_func->arg_count > 1 ? cond_func->arguments()[1] : 0;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4054
    }
4055
    else if (cond_func->have_rev_func() &&
4056 4057
             cond_func->arguments()[1]->real_item()->type() ==
                                                            Item::FIELD_ITEM)
4058
    {
4059
      field_item= (Item_field*) (cond_func->arguments()[1]->real_item());
4060 4061 4062 4063
      value= cond_func->arguments()[0];
    }
    else
      DBUG_RETURN(0);
igor@rurik.mysql.com's avatar
igor@rurik.mysql.com committed
4064
    ftree= get_full_func_mm_tree(param, cond_func, field_item, value, inv);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4065
  }
4066 4067

  DBUG_RETURN(ftree);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4068 4069 4070 4071
}


static SEL_TREE *
4072
get_mm_parts(PARAM *param, COND *cond_func, Field *field,
4073
	     Item_func::Functype type,
4074
	     Item *value, Item_result cmp_type)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4075 4076 4077 4078 4079
{
  DBUG_ENTER("get_mm_parts");
  if (field->table != param->table)
    DBUG_RETURN(0);

4080 4081
  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
4082 4083 4084 4085
  SEL_TREE *tree=0;
  if (value &&
      value->used_tables() & ~(param->prev_tables | param->read_tables))
    DBUG_RETURN(0);
4086
  for (; key_part != end ; key_part++)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4087 4088 4089 4090
  {
    if (field->eq(key_part->field))
    {
      SEL_ARG *sel_arg=0;
4091
      if (!tree && !(tree=new SEL_TREE()))
4092
	DBUG_RETURN(0);				// OOM
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4093 4094
      if (!value || !(value->used_tables() & ~param->read_tables))
      {
4095 4096
	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
4097 4098 4099 4100 4101 4102 4103 4104
	if (!sel_arg)
	  continue;
	if (sel_arg->type == SEL_ARG::IMPOSSIBLE)
	{
	  tree->type=SEL_TREE::IMPOSSIBLE;
	  DBUG_RETURN(tree);
	}
      }
4105 4106
      else
      {
4107
	// This key may be used later
4108
	if (!(sel_arg= new SEL_ARG(SEL_ARG::MAYBE_KEY)))
4109
	  DBUG_RETURN(0);			// OOM
4110
      }
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4111 4112
      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
4113
      tree->keys_map.set_bit(key_part->key);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4114 4115
    }
  }
4116

bk@work.mysql.com's avatar
bk@work.mysql.com committed
4117 4118 4119 4120 4121
  DBUG_RETURN(tree);
}


static SEL_ARG *
4122
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
4123 4124
	    Item_func::Functype type,Item *value)
{
4125
  uint maybe_null=(uint) field->real_maybe_null();
4126
  bool optimize_range;
4127 4128
  SEL_ARG *tree= 0;
  MEM_ROOT *alloc= param->mem_root;
4129
  char *str;
evgen@moonbone.local's avatar
evgen@moonbone.local committed
4130
  ulong orig_sql_mode;
4131
  int err;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4132 4133
  DBUG_ENTER("get_mm_leaf");

4134 4135
  /*
    We need to restore the runtime mem_root of the thread in this
konstantin@mysql.com's avatar
konstantin@mysql.com committed
4136
    function because it evaluates the value of its argument, while
4137 4138 4139 4140 4141 4142
    the argument can be any, e.g. a subselect. The subselect
    items, in turn, assume that all the memory allocated during
    the evaluation has the same life span as the item itself.
    TODO: opt_range.cc should not reset thd->mem_root at all.
  */
  param->thd->mem_root= param->old_root;
4143 4144
  if (!value)					// IS NULL or IS NOT NULL
  {
4145
    if (field->table->maybe_null)		// Can't use a key on this
4146
      goto end;
4147
    if (!maybe_null)				// Not null field
4148 4149 4150 4151 4152 4153 4154
    {
      if (type == Item_func::ISNULL_FUNC)
        tree= &null_element;
      goto end;
    }
    if (!(tree= new (alloc) SEL_ARG(field,is_null_string,is_null_string)))
      goto end;                                 // out of memory
4155 4156 4157 4158 4159
    if (type == Item_func::ISNOTNULL_FUNC)
    {
      tree->min_flag=NEAR_MIN;		    /* IS NOT NULL ->  X > NULL */
      tree->max_flag=NO_MAX_RANGE;
    }
4160
    goto end;
4161 4162 4163
  }

  /*
4164 4165 4166 4167 4168 4169 4170 4171 4172 4173
    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 '

4174 4175 4176 4177
  */
  if (field->result_type() == STRING_RESULT &&
      value->result_type() == STRING_RESULT &&
      key_part->image_type == Field::itRAW &&
4178 4179
      ((Field_str*)field)->charset() != conf_func->compare_collation() &&
      !(conf_func->compare_collation()->state & MY_CS_BINSORT))
4180
    goto end;
4181

4182 4183 4184
  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
4185 4186 4187 4188
  if (type == Item_func::LIKE_FUNC)
  {
    bool like_error;
    char buff1[MAX_FIELD_WIDTH],*min_str,*max_str;
4189
    String tmp(buff1,sizeof(buff1),value->collation.collation),*res;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4190
    uint length,offset,min_length,max_length;
4191
    uint field_length= field->pack_length()+maybe_null;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4192

4193
    if (!optimize_range)
4194
      goto end;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4195
    if (!(res= value->val_str(&tmp)))
4196 4197 4198 4199
    {
      tree= &null_element;
      goto end;
    }
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4200

4201 4202 4203 4204 4205
    /*
      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
4206 4207 4208 4209 4210 4211
    if (res != &tmp)
    {
      tmp.copy(*res);				// Get own copy
      res= &tmp;
    }
    if (field->cmp_type() != STRING_RESULT)
4212
      goto end;                                 // Can only optimize strings
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4213 4214

    offset=maybe_null;
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
4215 4216 4217
    length=key_part->store_length;

    if (length != key_part->length  + maybe_null)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4218
    {
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
4219 4220 4221
      /* 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
4222 4223 4224
    }
    else
    {
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
4225 4226 4227 4228 4229 4230 4231 4232
      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
4233
      else
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
4234
	field_length= length;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4235 4236
    }
    length+=offset;
4237 4238
    if (!(min_str= (char*) alloc_root(alloc, length*2)))
      goto end;
4239

bk@work.mysql.com's avatar
bk@work.mysql.com committed
4240 4241 4242
    max_str=min_str+length;
    if (maybe_null)
      max_str[0]= min_str[0]=0;
4243

4244
    field_length-= maybe_null;
4245
    like_error= my_like_range(field->charset(),
monty@mysql.com's avatar
monty@mysql.com committed
4246
			      res->ptr(), res->length(),
monty@mysql.com's avatar
monty@mysql.com committed
4247 4248
			      ((Item_func_like*)(param->cond))->escape,
			      wild_one, wild_many,
4249
			      field_length,
monty@mysql.com's avatar
monty@mysql.com committed
4250 4251
			      min_str+offset, max_str+offset,
			      &min_length, &max_length);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4252
    if (like_error)				// Can't optimize with LIKE
4253
      goto end;
monty@mysql.com's avatar
monty@mysql.com committed
4254

4255
    if (offset != maybe_null)			// BLOB or VARCHAR
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4256 4257 4258 4259
    {
      int2store(min_str+maybe_null,min_length);
      int2store(max_str+maybe_null,max_length);
    }
4260 4261
    tree= new (alloc) SEL_ARG(field, min_str, max_str);
    goto end;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4262 4263
  }

4264
  if (!optimize_range &&
4265
      type != Item_func::EQ_FUNC &&
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4266
      type != Item_func::EQUAL_FUNC)
4267
    goto end;                                   // Can't optimize this
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4268

4269 4270 4271 4272
  /*
    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
4273 4274 4275
  if (field->result_type() == STRING_RESULT &&
      value->result_type() != STRING_RESULT &&
      field->cmp_type() != value->result_type())
4276
    goto end;
4277
  /* For comparison purposes allow invalid dates like 2000-01-32 */
evgen@moonbone.local's avatar
evgen@moonbone.local committed
4278
  orig_sql_mode= field->table->in_use->variables.sql_mode;
4279 4280 4281 4282
  if (value->real_item()->type() == Item::STRING_ITEM &&
      (field->type() == FIELD_TYPE_DATE ||
       field->type() == FIELD_TYPE_DATETIME))
    field->table->in_use->variables.sql_mode|= MODE_INVALID_DATES;
4283 4284 4285 4286 4287 4288 4289
  err= value->save_in_field_no_warnings(field, 1);
  if (err > 0 && field->cmp_type() != value->result_type())
  {
    tree= 0;
    goto end;
  } 
  if (err < 0)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4290
  {
4291
    field->table->in_use->variables.sql_mode= orig_sql_mode;
4292
    /* This happens when we try to insert a NULL field in a not null column */
4293 4294
    tree= &null_element;                        // cmp with NULL is never TRUE
    goto end;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4295
  }
4296
  field->table->in_use->variables.sql_mode= orig_sql_mode;
4297
  str= (char*) alloc_root(alloc, key_part->store_length+1);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4298
  if (!str)
4299
    goto end;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4300
  if (maybe_null)
4301
    *str= (char) field->is_real_null();		// Set to 1 if null
4302
  field->get_key_image(str+maybe_null, key_part->length, key_part->image_type);
4303 4304
  if (!(tree= new (alloc) SEL_ARG(field, str, str)))
    goto end;                                   // out of memory
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4305

timour@mysql.com's avatar
timour@mysql.com committed
4306 4307 4308 4309 4310 4311 4312 4313 4314 4315 4316
  /*
    Check if we are comparing an UNSIGNED integer with a negative constant.
    In this case we know that:
    (a) (unsigned_int [< | <=] negative_constant) == FALSE
    (b) (unsigned_int [> | >=] negative_constant) == TRUE
    In case (a) the condition is false for all values, and in case (b) it
    is true for all values, so we can avoid unnecessary retrieval and condition
    testing, and we also get correct comparison of unsinged integers with
    negative integers (which otherwise fails because at query execution time
    negative integers are cast to unsigned if compared with unsigned).
   */
igor@rurik.mysql.com's avatar
igor@rurik.mysql.com committed
4317 4318
  if (field->result_type() == INT_RESULT &&
      value->result_type() == INT_RESULT &&
timour@mysql.com's avatar
timour@mysql.com committed
4319 4320 4321 4322 4323 4324 4325 4326
      ((Field_num*)field)->unsigned_flag && !((Item_int*)value)->unsigned_flag)
  {
    longlong item_val= value->val_int();
    if (item_val < 0)
    {
      if (type == Item_func::LT_FUNC || type == Item_func::LE_FUNC)
      {
        tree->type= SEL_ARG::IMPOSSIBLE;
4327
        goto end;
timour@mysql.com's avatar
timour@mysql.com committed
4328 4329
      }
      if (type == Item_func::GT_FUNC || type == Item_func::GE_FUNC)
4330 4331 4332 4333
      {
        tree= 0;
        goto end;
      }
timour@mysql.com's avatar
timour@mysql.com committed
4334 4335 4336
    }
  }

bk@work.mysql.com's avatar
bk@work.mysql.com committed
4337 4338 4339 4340 4341 4342 4343 4344 4345 4346 4347 4348 4349 4350 4351
  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:
4352 4353 4354
    /* Don't use open ranges for partial key_segments */
    if (field_is_equal_to_item(field,value) &&
        !(key_part->flag & HA_PART_KEY_SEG))
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4355 4356 4357 4358 4359
      tree->min_flag=NEAR_MIN;
    /* fall through */
  case Item_func::GE_FUNC:
    tree->max_flag=NO_MAX_RANGE;
    break;
4360
  case Item_func::SP_EQUALS_FUNC:
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
4361 4362 4363
    tree->min_flag=GEOM_FLAG | HA_READ_MBR_EQUAL;// NEAR_MIN;//512;
    tree->max_flag=NO_MAX_RANGE;
    break;
4364
  case Item_func::SP_DISJOINT_FUNC:
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
4365 4366 4367
    tree->min_flag=GEOM_FLAG | HA_READ_MBR_DISJOINT;// NEAR_MIN;//512;
    tree->max_flag=NO_MAX_RANGE;
    break;
4368
  case Item_func::SP_INTERSECTS_FUNC:
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
4369 4370 4371
    tree->min_flag=GEOM_FLAG | HA_READ_MBR_INTERSECT;// NEAR_MIN;//512;
    tree->max_flag=NO_MAX_RANGE;
    break;
4372
  case Item_func::SP_TOUCHES_FUNC:
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
4373 4374 4375
    tree->min_flag=GEOM_FLAG | HA_READ_MBR_INTERSECT;// NEAR_MIN;//512;
    tree->max_flag=NO_MAX_RANGE;
    break;
4376 4377

  case Item_func::SP_CROSSES_FUNC:
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
4378 4379 4380
    tree->min_flag=GEOM_FLAG | HA_READ_MBR_INTERSECT;// NEAR_MIN;//512;
    tree->max_flag=NO_MAX_RANGE;
    break;
4381
  case Item_func::SP_WITHIN_FUNC:
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
4382 4383 4384
    tree->min_flag=GEOM_FLAG | HA_READ_MBR_WITHIN;// NEAR_MIN;//512;
    tree->max_flag=NO_MAX_RANGE;
    break;
4385 4386

  case Item_func::SP_CONTAINS_FUNC:
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
4387 4388 4389
    tree->min_flag=GEOM_FLAG | HA_READ_MBR_CONTAIN;// NEAR_MIN;//512;
    tree->max_flag=NO_MAX_RANGE;
    break;
4390
  case Item_func::SP_OVERLAPS_FUNC:
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
4391 4392 4393
    tree->min_flag=GEOM_FLAG | HA_READ_MBR_INTERSECT;// NEAR_MIN;//512;
    tree->max_flag=NO_MAX_RANGE;
    break;
4394

bk@work.mysql.com's avatar
bk@work.mysql.com committed
4395 4396 4397
  default:
    break;
  }
4398 4399 4400

end:
  param->thd->mem_root= alloc;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4401 4402 4403 4404 4405 4406 4407 4408 4409
  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
4410 4411
** IMPOSSIBLE:	 Condition is never TRUE
** ALWAYS:	 Condition is always TRUE
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4412 4413 4414 4415 4416 4417
** 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
******************************************************************************/

/*
4418 4419
  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
4420 4421 4422 4423 4424 4425 4426 4427 4428 4429 4430 4431 4432 4433 4434 4435 4436 4437 4438 4439 4440 4441 4442 4443 4444 4445 4446 4447 4448 4449 4450 4451 4452 4453 4454 4455 4456 4457 4458 4459 4460 4461 4462 4463 4464 4465 4466 4467 4468 4469 4470 4471 4472 4473 4474 4475 4476 4477 4478 4479 4480
*/

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
4481 4482
  key_map  result_keys;
  result_keys.clear_all();
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4483 4484 4485 4486 4487 4488 4489 4490 4491 4492 4493 4494 4495
  /* 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);
4496
      if (*key1 && (*key1)->type == SEL_ARG::IMPOSSIBLE)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4497 4498
      {
	tree1->type= SEL_TREE::IMPOSSIBLE;
4499
        DBUG_RETURN(tree1);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4500
      }
sergefp@mysql.com's avatar
sergefp@mysql.com committed
4501
      result_keys.set_bit(key1 - tree1->keys);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4502
#ifdef EXTRA_DEBUG
4503 4504
      if (*key1)
        (*key1)->test_use_count(*key1);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4505 4506 4507
#endif
    }
  }
4508 4509
  tree1->keys_map= result_keys;
  /* dispose index_merge if there is a "range" option */
sergefp@mysql.com's avatar
sergefp@mysql.com committed
4510
  if (!result_keys.is_clear_all())
4511 4512 4513 4514 4515 4516 4517
  {
    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
4518 4519 4520 4521
  DBUG_RETURN(tree1);
}


4522
/*
4523 4524
  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
4525
  using index_merge.
4526 4527 4528 4529
*/

bool sel_trees_can_be_ored(SEL_TREE *tree1, SEL_TREE *tree2, PARAM* param)
{
sergefp@mysql.com's avatar
sergefp@mysql.com committed
4530
  key_map common_keys= tree1->keys_map;
4531
  DBUG_ENTER("sel_trees_can_be_ored");
4532
  common_keys.intersect(tree2->keys_map);
4533

sergefp@mysql.com's avatar
sergefp@mysql.com committed
4534
  if (common_keys.is_clear_all())
monty@mysql.com's avatar
monty@mysql.com committed
4535
    DBUG_RETURN(FALSE);
4536 4537

  /* trees have a common key, check if they refer to same key part */
4538
  SEL_ARG **key1,**key2;
sergefp@mysql.com's avatar
sergefp@mysql.com committed
4539
  for (uint key_no=0; key_no < param->keys; key_no++)
4540
  {
sergefp@mysql.com's avatar
sergefp@mysql.com committed
4541
    if (common_keys.is_set(key_no))
4542 4543 4544 4545 4546
    {
      key1= tree1->keys + key_no;
      key2= tree2->keys + key_no;
      if ((*key1)->part == (*key2)->part)
      {
monty@mysql.com's avatar
monty@mysql.com committed
4547
        DBUG_RETURN(TRUE);
4548 4549 4550
      }
    }
  }
monty@mysql.com's avatar
monty@mysql.com committed
4551
  DBUG_RETURN(FALSE);
4552
}
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4553 4554 4555 4556 4557 4558 4559 4560 4561 4562 4563 4564 4565 4566 4567 4568

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);

4569
  SEL_TREE *result= 0;
sergefp@mysql.com's avatar
sergefp@mysql.com committed
4570 4571
  key_map  result_keys;
  result_keys.clear_all();
4572
  if (sel_trees_can_be_ored(tree1, tree2, param))
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4573
  {
4574 4575 4576 4577
    /* 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
4578
    {
4579 4580 4581 4582
      *key1=key_or(*key1,*key2);
      if (*key1)
      {
        result=tree1;				// Added to tree1
sergefp@mysql.com's avatar
sergefp@mysql.com committed
4583
        result_keys.set_bit(key1 - tree1->keys);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4584
#ifdef EXTRA_DEBUG
4585
        (*key1)->test_use_count(*key1);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4586
#endif
4587 4588 4589 4590 4591 4592 4593 4594 4595 4596 4597 4598 4599 4600 4601 4602 4603 4604 4605 4606 4607 4608 4609 4610 4611 4612 4613 4614 4615 4616 4617
      }
    }
    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
4618
        swap_variables(SEL_TREE*, tree1, tree2);
4619 4620 4621 4622 4623 4624

      /* 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
4625 4626 4627 4628 4629 4630 4631 4632 4633 4634 4635 4636 4637 4638 4639 4640 4641 4642 4643 4644 4645
    }
  }
  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)
  {
4646 4647
    key1->right= key1->left= &null_element;
    key1->next= key1->prev= 0;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4648 4649 4650 4651 4652 4653 4654 4655 4656 4657 4658 4659 4660 4661 4662 4663 4664 4665 4666 4667 4668 4669 4670 4671 4672
  }
  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;
}


4673 4674 4675 4676 4677 4678 4679 4680 4681 4682 4683 4684 4685
/*
  Produce a SEL_ARG graph that represents "key1 AND key2"

  SYNOPSIS
    key_and()
      key1   First argument, root of its RB-tree
      key2   Second argument, root of its RB-tree

  RETURN
    RB-tree root of the resulting SEL_ARG graph.
    NULL if the result of AND operation is an empty interval {0}.
*/

bk@work.mysql.com's avatar
bk@work.mysql.com committed
4686
static SEL_ARG *
4687
key_and(SEL_ARG *key1, SEL_ARG *key2, uint clone_flag)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4688 4689 4690 4691 4692 4693 4694 4695 4696
{
  if (!key1)
    return key2;
  if (!key2)
    return key1;
  if (key1->part != key2->part)
  {
    if (key1->part > key2->part)
    {
4697
      swap_variables(SEL_ARG *, key1, key2);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4698 4699 4700 4701 4702
      clone_flag=swap_clone_flag(clone_flag);
    }
    // key1->part < key2->part
    key1->use_count--;
    if (key1->use_count > 0)
4703 4704
      if (!(key1= key1->clone_tree()))
	return 0;				// OOM
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4705 4706 4707 4708
    return and_all_keys(key1,key2,clone_flag);
  }

  if (((clone_flag & CLONE_KEY2_MAYBE) &&
4709 4710
       !(clone_flag & CLONE_KEY1_MAYBE) &&
       key2->type != SEL_ARG::MAYBE_KEY) ||
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4711 4712
      key1->type == SEL_ARG::MAYBE_KEY)
  {						// Put simple key in key2
4713
    swap_variables(SEL_ARG *, key1, key2);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4714 4715 4716
    clone_flag=swap_clone_flag(clone_flag);
  }

4717
  /* If one of the key is MAYBE_KEY then the found region may be smaller */
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4718 4719 4720 4721 4722
  if (key2->type == SEL_ARG::MAYBE_KEY)
  {
    if (key1->use_count > 1)
    {
      key1->use_count--;
4723 4724
      if (!(key1=key1->clone_tree()))
	return 0;				// OOM
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4725 4726 4727 4728 4729 4730 4731 4732 4733 4734 4735 4736 4737 4738
      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)
4739 4740
      {
	key1->use_count--;			// Incremented in and_all_keys
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4741
	return and_all_keys(key1,key2,clone_flag);
4742
      }
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4743 4744 4745 4746 4747
      key2->use_count--;			// Key2 doesn't have a tree
    }
    return key1;
  }

4748 4749
  if ((key1->min_flag | key2->min_flag) & GEOM_FLAG)
  {
4750
    /* TODO: why not leave one of the trees? */
4751 4752 4753 4754 4755
    key1->free_tree();
    key2->free_tree();
    return 0;					// Can't optimize this
  }

4756 4757 4758
  if ((key1->min_flag | key2->min_flag) & GEOM_FLAG)
  {
    key1->free_tree();
4759 4760 4761 4762
    key2->free_tree();
    return 0;					// Can't optimize this
  }

bk@work.mysql.com's avatar
bk@work.mysql.com committed
4763 4764 4765 4766 4767 4768 4769 4770 4771 4772 4773 4774 4775 4776 4777 4778 4779 4780 4781 4782
  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);
4783 4784
      if (!new_arg)
	return &null_element;			// End of memory
bk@work.mysql.com's avatar
bk@work.mysql.com committed
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 4834 4835
      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;
  }
4836
  if (!key2)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4837 4838 4839 4840 4841 4842 4843 4844
  {
    key1->use_count--;
    key1->free_tree();
    return 0;
  }
  key1->use_count--;
  key2->use_count--;

4845 4846
  if (key1->part != key2->part || 
      (key1->min_flag | key2->min_flag) & GEOM_FLAG)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4847 4848 4849 4850 4851 4852 4853 4854 4855 4856 4857 4858 4859 4860 4861 4862 4863 4864 4865 4866 4867 4868 4869 4870
  {
    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)
    {
4871
      swap_variables(SEL_ARG *,key1,key2);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4872
    }
4873
    if (key1->use_count > 0 || !(key1=key1->clone_tree()))
4874
      return 0;					// OOM
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4875 4876 4877 4878 4879 4880 4881 4882 4883 4884 4885 4886 4887 4888 4889 4890 4891 4892 4893 4894 4895 4896 4897 4898 4899
  }

  // 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
4900
	  if (!(key2=new SEL_ARG(*key2)))
4901
	    return 0;		// out of memory
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4902 4903 4904 4905 4906 4907 4908 4909 4910 4911 4912 4913 4914 4915 4916 4917 4918 4919 4920 4921 4922 4923 4924 4925 4926 4927 4928 4929 4930 4931 4932 4933 4934 4935 4936 4937 4938 4939 4940 4941
	  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)
	  {
4942 4943
	    SEL_ARG *cpy= new SEL_ARG(*key2);	// Must make copy
	    if (!cpy)
4944
	      return 0;				// OOM
4945
	    key1=key1->insert(cpy);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4946 4947 4948 4949 4950 4951 4952 4953 4954 4955 4956 4957 4958 4959 4960 4961 4962 4963 4964 4965 4966 4967 4968 4969 4970 4971 4972 4973
	    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
4974
        last->copy_min(tmp);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4975 4976 4977 4978 4979 4980 4981 4982 4983 4984 4985 4986 4987 4988 4989 4990 4991
	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);
4992 4993
      if (!new_arg)
	return 0;				// OOM
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4994 4995 4996 4997 4998 4999 5000 5001 5002 5003 5004 5005 5006
      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);
5007 5008
	if (!new_arg)
	  return 0;				// OOM
bk@work.mysql.com's avatar
bk@work.mysql.com committed
5009 5010 5011 5012 5013 5014 5015 5016 5017 5018 5019 5020 5021 5022
	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))
	{
5023 5024 5025 5026
	  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
5027 5028 5029 5030 5031
	  key2=key2->next;
	  goto end;
	}
	if (tmp->cmp_min_to_max(&key) > 0)
	{
5032 5033 5034 5035
	  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
5036 5037 5038 5039 5040 5041
	  break;
	}
      }
      else
      {
	SEL_ARG *new_arg=tmp->clone_last(&key); // tmp.min <= x <= key.max
5042 5043
	if (!new_arg)
	  return 0;				// OOM
bk@work.mysql.com's avatar
bk@work.mysql.com committed
5044 5045
	tmp->copy_max_to_min(&key);
	tmp->increment_use_count(key1->use_count+1);
5046 5047
	/* 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
5048 5049 5050 5051 5052 5053 5054 5055 5056 5057 5058 5059 5060 5061
	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)
    {
5062 5063 5064
      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
5065
      key2->increment_use_count(key1->use_count+1);
5066
      key1=key1->insert(tmp);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
5067 5068 5069 5070 5071 5072 5073 5074 5075 5076 5077 5078 5079 5080 5081 5082 5083 5084 5085 5086 5087 5088 5089 5090 5091 5092 5093 5094 5095 5096 5097 5098 5099 5100 5101 5102 5103 5104 5105 5106 5107 5108 5109 5110 5111 5112 5113
    }
    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);
5114

bk@work.mysql.com's avatar
bk@work.mysql.com committed
5115 5116 5117 5118 5119 5120 5121 5122 5123 5124 5125 5126 5127 5128 5129 5130 5131 5132 5133 5134 5135 5136 5137 5138 5139 5140 5141 5142 5143 5144 5145 5146 5147 5148 5149 5150 5151 5152 5153 5154 5155 5156 5157 5158 5159 5160 5161 5162 5163 5164 5165 5166 5167 5168 5169 5170 5171 5172 5173 5174 5175 5176 5177 5178 5179 5180 5181
  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;
  }
}


/*
5182 5183 5184 5185 5186
  Remove a element from the tree

  SYNOPSIS
    tree_delete()
    key		Key that is to be deleted from tree (this)
5187

5188 5189 5190 5191 5192
  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
5193 5194 5195 5196 5197 5198 5199
*/

SEL_ARG *
SEL_ARG::tree_delete(SEL_ARG *key)
{
  enum leaf_color remove_color;
  SEL_ARG *root,*nod,**par,*fix_par;
5200 5201 5202 5203
  DBUG_ENTER("tree_delete");

  root=this;
  this->parent= 0;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
5204 5205 5206 5207 5208 5209 5210 5211 5212 5213 5214 5215 5216 5217 5218 5219 5220 5221 5222 5223 5224 5225 5226 5227 5228 5229 5230 5231 5232 5233 5234 5235 5236 5237 5238 5239 5240 5241 5242 5243 5244 5245 5246 5247 5248 5249

  /* 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)
5250
    DBUG_RETURN(0);				// Maybe root later
bk@work.mysql.com's avatar
bk@work.mysql.com committed
5251 5252 5253 5254 5255 5256 5257
  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;
5258
  DBUG_RETURN(root);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
5259 5260 5261 5262 5263 5264 5265 5266 5267 5268 5269 5270 5271 5272 5273 5274 5275 5276 5277 5278 5279 5280 5281 5282 5283 5284 5285 5286 5287 5288 5289 5290 5291 5292 5293 5294 5295 5296 5297 5298 5299 5300 5301 5302 5303 5304 5305 5306 5307 5308 5309 5310 5311 5312 5313 5314 5315 5316 5317 5318 5319 5320 5321 5322 5323 5324 5325 5326 5327 5328 5329 5330 5331 5332 5333 5334 5335 5336 5337 5338 5339 5340 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 5369 5370 5371 5372 5373 5374 5375 5376 5377 5378 5379 5380 5381 5382 5383 5384 5385 5386 5387 5388 5389 5390 5391 5392 5393 5394 5395 5396 5397 5398 5399 5400 5401 5402 5403 5404 5405 5406 5407 5408 5409 5410 5411 5412 5413 5414 5415 5416 5417 5418 5419 5420 5421 5422 5423 5424 5425 5426 5427 5428 5429 5430 5431 5432 5433
}


	/* 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;
}


5434
	/* Test that the properties for a red-black tree hold */
bk@work.mysql.com's avatar
bk@work.mysql.com committed
5435 5436 5437 5438 5439 5440 5441 5442 5443 5444 5445 5446 5447 5448 5449 5450 5451 5452 5453 5454 5455 5456 5457 5458 5459 5460 5461 5462 5463 5464 5465 5466 5467 5468 5469 5470 5471

#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
}

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 5509 5510 5511 5512 5513 5514 5515 5516

/*
  Count how many times SEL_ARG graph "root" refers to its part "key"
  
  SYNOPSIS
    count_key_part_usage()
      root  An RB-Root node in a SEL_ARG graph.
      key   Another RB-Root node in that SEL_ARG graph.

  DESCRIPTION
    The passed "root" node may refer to "key" node via root->next_key_part,
    root->next->n

    This function counts how many times the node "key" is referred (via
    SEL_ARG::next_key_part) by 
     - intervals of RB-tree pointed by "root", 
     - intervals of RB-trees that are pointed by SEL_ARG::next_key_part from 
       intervals of RB-tree pointed by "root",
     - and so on.
    
    Here is an example (horizontal links represent next_key_part pointers, 
    vertical links - next/prev prev pointers):  
    
         +----+               $
         |root|-----------------+
         +----+               $ |
           |                  $ |
           |                  $ |
         +----+       +---+   $ |     +---+    Here the return value
         |    |- ... -|   |---$-+--+->|key|    will be 4.
         +----+       +---+   $ |  |  +---+
           |                  $ |  |
          ...                 $ |  |
           |                  $ |  |
         +----+   +---+       $ |  |
         |    |---|   |---------+  |
         +----+   +---+       $    |
           |        |         $    |
          ...     +---+       $    |
                  |   |------------+
                  +---+       $
  RETURN 
    Number of links to "key" from nodes reachable from "root".
*/

bk@work.mysql.com's avatar
bk@work.mysql.com committed
5517 5518 5519 5520 5521 5522 5523 5524 5525 5526 5527 5528 5529 5530 5531 5532 5533
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;
}


5534 5535 5536 5537 5538 5539 5540 5541 5542 5543 5544 5545 5546 5547
/*
  Check if SEL_ARG::use_count value is correct

  SYNOPSIS
    SEL_ARG::test_use_count()
      root  The root node of the SEL_ARG graph (an RB-tree root node that
            has the least value of sel_arg->part in the entire graph, and
            thus is the "origin" of the graph)

  DESCRIPTION
    Check if SEL_ARG::use_count value is correct. See the definition of
    use_count for what is "correct".
*/

bk@work.mysql.com's avatar
bk@work.mysql.com committed
5548 5549
void SEL_ARG::test_use_count(SEL_ARG *root)
{
5550
  uint e_count=0;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
5551 5552
  if (this == root && use_count != 1)
  {
monty@mysql.com's avatar
monty@mysql.com committed
5553
    sql_print_information("Use_count: Wrong count %lu for root",use_count);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
5554 5555 5556 5557 5558 5559 5560 5561 5562 5563 5564 5565
    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)
      {
5566 5567 5568
        sql_print_information("Use_count: Wrong count for key at 0x%lx, %lu "
                              "should be %lu", (long unsigned int)pos,
                              pos->next_key_part->use_count, count);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
5569 5570 5571 5572 5573 5574
	return;
      }
      pos->next_key_part->test_use_count(root);
    }
  }
  if (e_count != elements)
monty@mysql.com's avatar
monty@mysql.com committed
5575
    sql_print_warning("Wrong use count: %u (should be %u) for tree at 0x%lx",
5576
                      e_count, elements, (long unsigned int) this);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
5577 5578 5579 5580 5581
}

#endif


5582 5583 5584 5585 5586 5587 5588 5589 5590 5591
/*
  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
5592
    param->is_ror_scan is set to reflect if the key scan is a ROR (see
5593
    is_key_scan_ror function for more info)
5594
    param->table->quick_*, param->range_count (and maybe others) are
5595
    updated with data of given key scan, see check_quick_keys for details.
5596 5597

  RETURN
5598
    Estimate # of records to be retrieved.
5599
    HA_POS_ERROR if estimate calculation failed due to table handler problems.
bk@work.mysql.com's avatar
bk@work.mysql.com committed
5600

5601
*/
bk@work.mysql.com's avatar
bk@work.mysql.com committed
5602 5603 5604 5605 5606

static ha_rows
check_quick_select(PARAM *param,uint idx,SEL_ARG *tree)
{
  ha_rows records;
5607 5608
  bool    cpk_scan;
  uint key;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
5609
  DBUG_ENTER("check_quick_select");
5610

monty@mysql.com's avatar
monty@mysql.com committed
5611
  param->is_ror_scan= FALSE;
5612

bk@work.mysql.com's avatar
bk@work.mysql.com committed
5613 5614
  if (!tree)
    DBUG_RETURN(HA_POS_ERROR);			// Can't use it
monty@narttu.mysql.fi's avatar
monty@narttu.mysql.fi committed
5615 5616
  param->max_key_part=0;
  param->range_count=0;
5617 5618
  key= param->real_keynr[idx];

bk@work.mysql.com's avatar
bk@work.mysql.com committed
5619 5620 5621 5622
  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
5623 5624 5625 5626 5627

  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
5628
    cpk_scan= FALSE;
5629 5630 5631 5632 5633 5634 5635
  }
  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).
    */
5636 5637
    cpk_scan= ((param->table->s->primary_key == param->real_keynr[idx]) &&
               param->table->file->primary_key_is_clustered());
5638
    param->is_ror_scan= !cpk_scan;
5639
  }
5640
  param->n_ranges= 0;
5641

bk@work.mysql.com's avatar
bk@work.mysql.com committed
5642 5643
  records=check_quick_keys(param,idx,tree,param->min_key,0,param->max_key,0);
  if (records != HA_POS_ERROR)
5644
  {
5645
    param->table->quick_keys.set_bit(key);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
5646 5647
    param->table->quick_rows[key]=records;
    param->table->quick_key_parts[key]=param->max_key_part+1;
5648
    param->table->quick_n_ranges[key]= param->n_ranges;
5649
    if (cpk_scan)
monty@mysql.com's avatar
monty@mysql.com committed
5650
      param->is_ror_scan= TRUE;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
5651
  }
5652 5653
  if (param->table->file->index_flags(key, 0, TRUE) & HA_KEY_SCAN_NOT_ROR)
    param->is_ror_scan= FALSE;
5654
  DBUG_PRINT("exit", ("Records: %lu", (ulong) records));
bk@work.mysql.com's avatar
bk@work.mysql.com committed
5655 5656 5657 5658
  DBUG_RETURN(records);
}


5659
/*
5660 5661
  Recursively calculate estimate of # rows that will be retrieved by
  key scan on key idx.
5662 5663
  SYNOPSIS
    check_quick_keys()
5664
      param         Parameter from test_quick select function.
5665
      idx           Number of key to use in PARAM::keys in list of used keys
5666 5667 5668
                    (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
5669
      min_key_flag
5670
      max_key       Buffer with partial max key value tuple
5671 5672
      max_key_flag

5673
  NOTES
5674 5675
    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
5676 5677
    are calculated using records_in_range calls at the leaf nodes and then
    summed.
5678

5679 5680
    param->min_key and param->max_key are used to hold prefixes of key value
    tuples.
5681 5682

    The side effects are:
5683

5684 5685
    param->max_key_part is updated to hold the maximum number of key parts used
      in scan minus 1.
5686 5687

    param->range_count is incremented if the function finds a range that
5688
      wasn't counted by the caller.
5689

5690 5691 5692
    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)
5693 5694
*/

bk@work.mysql.com's avatar
bk@work.mysql.com committed
5695 5696 5697 5698 5699
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
5700 5701 5702
  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
5703 5704 5705 5706

  param->max_key_part=max(param->max_key_part,key_tree->part);
  if (key_tree->left != &null_element)
  {
5707 5708 5709 5710 5711 5712
    /*
      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
5713
    param->is_ror_scan= FALSE;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
5714 5715 5716 5717 5718 5719
    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
5720 5721
  tmp_min_key= min_key;
  tmp_max_key= max_key;
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
5722
  key_tree->store(param->key[idx][key_tree->part].store_length,
bk@work.mysql.com's avatar
bk@work.mysql.com committed
5723
		  &tmp_min_key,min_key_flag,&tmp_max_key,max_key_flag);
monty@mysql.com's avatar
monty@mysql.com committed
5724 5725
  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
5726

5727 5728
  if (param->is_ror_scan)
  {
5729
    /*
5730
      If the index doesn't cover entire key, mark the scan as non-ROR scan.
5731
      Actually we're cutting off some ROR scans here.
5732 5733 5734
    */
    uint16 fieldnr= param->table->key_info[param->real_keynr[idx]].
                    key_part[key_tree->part].fieldnr - 1;
5735
    if (param->table->field[fieldnr]->key_length() !=
5736
        param->key[idx][key_tree->part].length)
monty@mysql.com's avatar
monty@mysql.com committed
5737
      param->is_ror_scan= FALSE;
5738 5739
  }

bk@work.mysql.com's avatar
bk@work.mysql.com committed
5740 5741 5742 5743 5744 5745 5746 5747 5748 5749 5750 5751 5752
  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
    }
5753
    else
5754 5755
    {
      /* The interval for current key part is not c1 <= keyXpartY <= c1 */
monty@mysql.com's avatar
monty@mysql.com committed
5756
      param->is_ror_scan= FALSE;
5757
    }
5758

bk@work.mysql.com's avatar
bk@work.mysql.com committed
5759 5760 5761 5762 5763 5764 5765 5766 5767 5768 5769 5770 5771 5772 5773 5774 5775 5776
    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
5777
  param->range_count++;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
5778 5779
  if (!tmp_min_flag && ! tmp_max_flag &&
      (uint) key_tree->part+1 == param->table->key_info[keynr].key_parts &&
5780 5781
      (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
5782 5783
      min_key_length == max_key_length &&
      !memcmp(param->min_key,param->max_key,min_key_length))
5784
  {
bk@work.mysql.com's avatar
bk@work.mysql.com committed
5785
    tmp=1;					// Max one record
5786 5787
    param->n_ranges++;
  }
bk@work.mysql.com's avatar
bk@work.mysql.com committed
5788
  else
5789
  {
5790 5791
    if (param->is_ror_scan)
    {
5792 5793 5794 5795 5796 5797 5798 5799 5800
      /*
        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.
      */
5801 5802
      if (!(min_key_length == max_key_length &&
            !memcmp(min_key,max_key, (uint) (tmp_max_key - max_key)) &&
5803
            !key_tree->min_flag && !key_tree->max_flag &&
5804
            is_key_scan_ror(param, keynr, key_tree->part + 1)))
monty@mysql.com's avatar
monty@mysql.com committed
5805
        param->is_ror_scan= FALSE;
5806
    }
5807
    param->n_ranges++;
5808

5809
    if (tmp_min_flag & GEOM_FLAG)
5810
    {
5811 5812 5813 5814 5815 5816 5817 5818
      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);
5819 5820 5821
    }
    else
    {
5822 5823 5824 5825 5826 5827
      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
5828
      max_range.key=    (byte*) param->max_key;
5829 5830 5831 5832 5833 5834 5835 5836
      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));
5837 5838
    }
  }
bk@work.mysql.com's avatar
bk@work.mysql.com committed
5839 5840 5841 5842 5843 5844
 end:
  if (tmp == HA_POS_ERROR)			// Impossible range
    return tmp;
  records+=tmp;
  if (key_tree->right != &null_element)
  {
5845 5846 5847 5848 5849 5850
    /*
      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
5851
    param->is_ror_scan= FALSE;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
5852 5853 5854 5855 5856 5857 5858 5859 5860
    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;
}

5861

5862
/*
5863
  Check if key scan on given index with equality conditions on first n key
5864 5865 5866 5867
  parts is a ROR scan.

  SYNOPSIS
    is_key_scan_ror()
5868
      param  Parameter from test_quick_select
5869 5870 5871 5872
      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.
5873

5874 5875 5876
  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)
5877

5878 5879 5880
    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)
5881

5882 5883
    where the index is defined on (key1_1, ..., key1_N [,a_1, ..., a_n])

5884
    and the table has a clustered Primary Key
5885

5886
    PRIMARY KEY(a_1, ..., a_n, b1, ..., b_k) with first key parts being
5887
    identical to uncovered parts ot the key being scanned (2)
5888 5889

    Scans on HASH indexes are not ROR scans,
5890 5891 5892 5893 5894 5895
    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.

5896
  RETURN
monty@mysql.com's avatar
monty@mysql.com committed
5897 5898
    TRUE  If the scan is ROR-scan
    FALSE otherwise
5899
*/
5900

5901 5902 5903 5904
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;
5905 5906 5907
  KEY_PART_INFO *key_part_end= (table_key->key_part +
                                table_key->key_parts);
  uint pk_number;
5908

5909
  if (key_part == key_part_end)
monty@mysql.com's avatar
monty@mysql.com committed
5910
    return TRUE;
5911
  pk_number= param->table->s->primary_key;
5912
  if (!param->table->file->primary_key_is_clustered() || pk_number == MAX_KEY)
monty@mysql.com's avatar
monty@mysql.com committed
5913
    return FALSE;
5914 5915

  KEY_PART_INFO *pk_part= param->table->key_info[pk_number].key_part;
5916
  KEY_PART_INFO *pk_part_end= pk_part +
5917
                              param->table->key_info[pk_number].key_parts;
5918 5919
  for (;(key_part!=key_part_end) && (pk_part != pk_part_end);
       ++key_part, ++pk_part)
5920
  {
5921
    if ((key_part->field != pk_part->field) ||
5922
        (key_part->length != pk_part->length))
5923
      return FALSE;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
5924
  }
5925
  return (key_part == key_part_end);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
5926 5927 5928
}


5929 5930
/*
  Create a QUICK_RANGE_SELECT from given key and SEL_ARG tree for that key.
5931

5932 5933
  SYNOPSIS
    get_quick_select()
5934
      param
5935
      idx          Index of used key in param->key.
5936 5937
      key_tree     SEL_ARG tree for the used key
      parent_alloc If not NULL, use it to allocate memory for
5938
                   quick select data. Otherwise use quick->alloc.
5939
  NOTES
5940
    The caller must call QUICK_SELECT::init for returned quick select
5941

5942
    CAUTION! This function may change thd->mem_root to a MEM_ROOT which will be
5943
    deallocated when the returned quick select is deleted.
5944 5945 5946 5947

  RETURN
    NULL on error
    otherwise created quick select
5948
*/
5949

5950 5951 5952
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
5953
{
5954
  QUICK_RANGE_SELECT *quick;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
5955
  DBUG_ENTER("get_quick_select");
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
5956 5957 5958 5959 5960 5961 5962 5963 5964

  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
5965
                                 test(parent_alloc));
5966

pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
5967
  if (quick)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
5968 5969 5970 5971 5972 5973 5974 5975 5976 5977 5978
  {
    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*)
5979 5980 5981 5982
        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
5983
    }
5984
  }
bk@work.mysql.com's avatar
bk@work.mysql.com committed
5985 5986 5987 5988 5989 5990 5991
  DBUG_RETURN(quick);
}


/*
** Fix this to get all possible sub_ranges
*/
5992 5993
bool
get_quick_keys(PARAM *param,QUICK_RANGE_SELECT *quick,KEY_PART *key,
bk@work.mysql.com's avatar
bk@work.mysql.com committed
5994 5995 5996 5997 5998 5999 6000 6001 6002 6003 6004 6005 6006
	       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
6007
  key_tree->store(key[key_tree->part].store_length,
bk@work.mysql.com's avatar
bk@work.mysql.com committed
6008 6009 6010 6011 6012 6013 6014 6015 6016 6017 6018 6019 6020 6021 6022 6023 6024 6025 6026 6027 6028 6029 6030 6031 6032 6033 6034 6035
		  &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
6036 6037 6038 6039
  {
    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
6040

6041 6042 6043 6044 6045
  /*
    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)
6046 6047 6048 6049 6050 6051 6052 6053 6054 6055
  {
    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
6056 6057 6058 6059 6060 6061 6062 6063
  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;
6064 6065
      if ((table_key->flags & (HA_NOSAME | HA_END_SPACE_KEY)) == HA_NOSAME &&
	  key->part == table_key->key_parts-1)
6066 6067 6068 6069 6070 6071 6072 6073 6074
      {
	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
6075 6076 6077 6078
    }
  }

  /* Get range for retrieving rows in QUICK_SELECT::get_next */
6079
  if (!(range= new QUICK_RANGE((const char *) param->min_key,
6080
			       (uint) (tmp_min_key - param->min_key),
6081
			       (const char *) param->max_key,
6082 6083
			       (uint) (tmp_max_key - param->max_key),
			       flag)))
6084 6085
    return 1;			// out of memory

bk@work.mysql.com's avatar
bk@work.mysql.com committed
6086 6087
  set_if_bigger(quick->max_used_key_length,range->min_length);
  set_if_bigger(quick->max_used_key_length,range->max_length);
6088
  set_if_bigger(quick->used_key_parts, (uint) key_tree->part+1);
6089 6090 6091
  if (insert_dynamic(&quick->ranges, (gptr)&range))
    return 1;

bk@work.mysql.com's avatar
bk@work.mysql.com committed
6092 6093 6094 6095 6096 6097 6098 6099 6100 6101 6102 6103
 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
*/

6104
bool QUICK_RANGE_SELECT::unique_key_range()
bk@work.mysql.com's avatar
bk@work.mysql.com committed
6105 6106 6107
{
  if (ranges.elements == 1)
  {
6108 6109
    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
6110 6111
    {
      KEY *key=head->key_info+index;
6112
      return ((key->flags & (HA_NOSAME | HA_END_SPACE_KEY)) == HA_NOSAME &&
bk@work.mysql.com's avatar
bk@work.mysql.com committed
6113 6114 6115 6116 6117 6118
	      key->key_length == tmp->min_length);
    }
  }
  return 0;
}

6119

monty@mysql.com's avatar
monty@mysql.com committed
6120
/* Returns TRUE if any part of the key is NULL */
6121 6122 6123

static bool null_part_in_key(KEY_PART *key_part, const char *key, uint length)
{
6124
  for (const char *end=key+length ;
6125
       key < end;
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
6126
       key+= key_part++->store_length)
6127
  {
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
6128 6129
    if (key_part->null_bit && *key)
      return 1;
6130 6131 6132 6133
  }
  return 0;
}

pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
6134

6135
bool QUICK_SELECT_I::is_keys_used(List<Item> *fields)
6136
{
6137
  return is_key_used(head, index, *fields);
6138 6139
}

6140
bool QUICK_INDEX_MERGE_SELECT::is_keys_used(List<Item> *fields)
6141 6142 6143 6144 6145
{
  QUICK_RANGE_SELECT *quick;
  List_iterator_fast<QUICK_RANGE_SELECT> it(quick_selects);
  while ((quick= it++))
  {
6146
    if (is_key_used(head, quick->index, *fields))
6147 6148 6149 6150 6151
      return 1;
  }
  return 0;
}

6152
bool QUICK_ROR_INTERSECT_SELECT::is_keys_used(List<Item> *fields)
6153 6154 6155 6156 6157
{
  QUICK_RANGE_SELECT *quick;
  List_iterator_fast<QUICK_RANGE_SELECT> it(quick_selects);
  while ((quick= it++))
  {
6158
    if (is_key_used(head, quick->index, *fields))
6159 6160 6161 6162 6163
      return 1;
  }
  return 0;
}

6164
bool QUICK_ROR_UNION_SELECT::is_keys_used(List<Item> *fields)
6165 6166 6167 6168 6169
{
  QUICK_SELECT_I *quick;
  List_iterator_fast<QUICK_SELECT_I> it(quick_selects);
  while ((quick= it++))
  {
6170
    if (quick->is_keys_used(fields))
6171 6172 6173 6174 6175
      return 1;
  }
  return 0;
}

monty@mysql.com's avatar
monty@mysql.com committed
6176

sergefp@mysql.com's avatar
sergefp@mysql.com committed
6177 6178
/*
  Create quick select from ref/ref_or_null scan.
6179

sergefp@mysql.com's avatar
sergefp@mysql.com committed
6180 6181 6182 6183 6184 6185 6186 6187 6188 6189 6190 6191 6192 6193 6194
  SYNOPSIS
    get_quick_select_for_ref()
      thd      Thread handle
      table    Table to access
      ref      ref[_or_null] scan parameters
      records  Estimate of number of records (needed only to construct 
               quick select)
  NOTES
    This allocates things in a new memory root, as this may be called many
    times during a query.
  
  RETURN 
    Quick select that retrieves the same rows as passed ref scan
    NULL on error.
*/
bk@work.mysql.com's avatar
bk@work.mysql.com committed
6195

6196
QUICK_RANGE_SELECT *get_quick_select_for_ref(THD *thd, TABLE *table,
sergefp@mysql.com's avatar
sergefp@mysql.com committed
6197
                                             TABLE_REF *ref, ha_rows records)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
6198
{
6199 6200
  MEM_ROOT *old_root, *alloc;
  QUICK_RANGE_SELECT *quick;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
6201 6202
  KEY *key_info = &table->key_info[ref->key];
  KEY_PART *key_part;
serg@serg.mylan's avatar
serg@serg.mylan committed
6203
  QUICK_RANGE *range;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
6204
  uint part;
6205 6206 6207 6208 6209 6210

  old_root= thd->mem_root;
  /* The following call may change thd->mem_root */
  quick= new QUICK_RANGE_SELECT(thd, table, ref->key, 0);
  /* save mem_root set by QUICK_RANGE_SELECT constructor */
  alloc= thd->mem_root;
6211 6212 6213 6214 6215
  /*
    return back default mem_root (thd->mem_root) changed by
    QUICK_RANGE_SELECT constructor
  */
  thd->mem_root= old_root;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
6216 6217

  if (!quick)
6218
    return 0;			/* no ranges found */
sergefp@mysql.com's avatar
sergefp@mysql.com committed
6219
  if (quick->init())
monty@mysql.com's avatar
monty@mysql.com committed
6220
    goto err;
sergefp@mysql.com's avatar
sergefp@mysql.com committed
6221
  quick->records= records;
6222

sergefp@mysql.com's avatar
sergefp@mysql.com committed
6223
  if (cp_buffer_from_ref(thd,ref) && thd->is_fatal_error ||
6224
      !(range= new(alloc) QUICK_RANGE()))
monty@mysql.com's avatar
monty@mysql.com committed
6225
    goto err;                                   // out of memory
6226

bk@work.mysql.com's avatar
bk@work.mysql.com committed
6227 6228 6229
  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 &&
6230 6231
		 (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
6232 6233

  if (!(quick->key_parts=key_part=(KEY_PART *)
6234
	alloc_root(&quick->alloc,sizeof(KEY_PART)*ref->key_parts)))
bk@work.mysql.com's avatar
bk@work.mysql.com committed
6235 6236 6237 6238 6239 6240
    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
6241 6242
    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
6243
    key_part->null_bit=     key_info->key_part[part].null_bit;
6244
    key_part->flag=         key_info->key_part[part].key_part_flag;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
6245
  }
pem@mysql.com's avatar
pem@mysql.com committed
6246
  if (insert_dynamic(&quick->ranges,(gptr)&range))
6247 6248
    goto err;

6249
  /*
6250 6251 6252 6253 6254
     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.
  */
6255 6256 6257 6258 6259
  if (ref->null_ref_key)
  {
    QUICK_RANGE *null_range;

    *ref->null_ref_key= 1;		// Set null byte then create a range
6260 6261 6262 6263 6264
    if (!(null_range= new (alloc) QUICK_RANGE((char*)ref->key_buff,
                                              ref->key_length,
                                              (char*)ref->key_buff,
                                              ref->key_length,
                                              EQ_RANGE)))
6265 6266
      goto err;
    *ref->null_ref_key= 0;		// Clear null byte
pem@mysql.com's avatar
pem@mysql.com committed
6267
    if (insert_dynamic(&quick->ranges,(gptr)&null_range))
6268 6269 6270 6271
      goto err;
  }

  return quick;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
6272 6273 6274 6275 6276 6277

err:
  delete quick;
  return 0;
}

6278 6279

/*
sergefp@mysql.com's avatar
sergefp@mysql.com committed
6280 6281 6282 6283 6284 6285
  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.
  
6286
  If table has a clustered primary key that covers all rows (TRUE for bdb
6287
     and innodb currently) and one of the index_merge scans is a scan on PK,
6288
  then
sergefp@mysql.com's avatar
sergefp@mysql.com committed
6289 6290
    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.
6291

6292 6293 6294
  RETURN
    0     OK
    other error
6295
*/
6296

sergefp@mysql.com's avatar
sergefp@mysql.com committed
6297
int QUICK_INDEX_MERGE_SELECT::read_keys_and_merge()
6298
{
sergefp@mysql.com's avatar
sergefp@mysql.com committed
6299 6300
  List_iterator_fast<QUICK_RANGE_SELECT> cur_quick_it(quick_selects);
  QUICK_RANGE_SELECT* cur_quick;
6301
  int result;
sergefp@mysql.com's avatar
sergefp@mysql.com committed
6302
  Unique *unique;
6303
  DBUG_ENTER("QUICK_INDEX_MERGE_SELECT::prepare_unique");
6304

6305
  /* We're going to just read rowids. */
6306 6307
  if (head->file->extra(HA_EXTRA_KEYREAD))
    DBUG_RETURN(1);
6308

6309 6310
  /*
    Make innodb retrieve all PK member fields, so
6311
     * ha_innobase::position (which uses them) call works.
6312
     * We can filter out rows that will be retrieved by clustered PK.
6313
    (This also creates a deficiency - it is possible that we will retrieve
6314
     parts of key that are not used by current query at all.)
6315
  */
6316 6317
  if (head->file->extra(HA_EXTRA_RETRIEVE_PRIMARY_KEY))
    DBUG_RETURN(1);
6318

sergefp@mysql.com's avatar
sergefp@mysql.com committed
6319 6320
  cur_quick_it.rewind();
  cur_quick= cur_quick_it++;
6321
  DBUG_ASSERT(cur_quick != 0);
sergefp@mysql.com's avatar
sergefp@mysql.com committed
6322 6323 6324 6325 6326
  
  /*
    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
6327
  if (cur_quick->init() || cur_quick->reset())
sergefp@mysql.com's avatar
sergefp@mysql.com committed
6328
    DBUG_RETURN(1);
6329

6330
  unique= new Unique(refpos_order_cmp, (void *)head->file,
6331
                     head->file->ref_length,
6332
                     thd->variables.sortbuff_size);
6333 6334
  if (!unique)
    DBUG_RETURN(1);
monty@mysql.com's avatar
monty@mysql.com committed
6335
  for (;;)
6336
  {
sergefp@mysql.com's avatar
sergefp@mysql.com committed
6337
    while ((result= cur_quick->get_next()) == HA_ERR_END_OF_FILE)
6338
    {
sergefp@mysql.com's avatar
sergefp@mysql.com committed
6339 6340 6341
      cur_quick->range_end();
      cur_quick= cur_quick_it++;
      if (!cur_quick)
6342
        break;
6343

sergefp@mysql.com's avatar
sergefp@mysql.com committed
6344 6345
      if (cur_quick->file->inited != handler::NONE) 
        cur_quick->file->ha_index_end();
sergefp@mysql.com's avatar
sergefp@mysql.com committed
6346
      if (cur_quick->init() || cur_quick->reset())
6347
        DBUG_RETURN(1);
6348 6349 6350
    }

    if (result)
6351
    {
6352
      if (result != HA_ERR_END_OF_FILE)
sergefp@mysql.com's avatar
sergefp@mysql.com committed
6353 6354
      {
        cur_quick->range_end();
6355
        DBUG_RETURN(result);
sergefp@mysql.com's avatar
sergefp@mysql.com committed
6356
      }
6357
      break;
6358
    }
6359

6360 6361
    if (thd->killed)
      DBUG_RETURN(1);
6362

6363
    /* skip row if it will be retrieved by clustered PK scan */
6364 6365
    if (pk_quick_select && pk_quick_select->row_in_ranges())
      continue;
6366

sergefp@mysql.com's avatar
sergefp@mysql.com committed
6367 6368
    cur_quick->file->position(cur_quick->record);
    result= unique->unique_add((char*)cur_quick->file->ref);
6369
    if (result)
6370 6371
      DBUG_RETURN(1);

monty@mysql.com's avatar
monty@mysql.com committed
6372
  }
6373

6374 6375
  /* ok, all row ids are in Unique */
  result= unique->get(head);
sergefp@mysql.com's avatar
sergefp@mysql.com committed
6376
  delete unique;
monty@mysql.com's avatar
monty@mysql.com committed
6377
  doing_pk_scan= FALSE;
monty@mysql.com's avatar
monty@mysql.com committed
6378 6379
  /* start table scan */
  init_read_record(&read_record, thd, head, (SQL_SELECT*) 0, 1, 1);
6380 6381
  /* index_merge currently doesn't support "using index" at all */
  head->file->extra(HA_EXTRA_NO_KEYREAD);
6382

6383 6384 6385
  DBUG_RETURN(result);
}

6386

6387 6388 6389
/*
  Get next row for index_merge.
  NOTES
6390 6391 6392 6393
    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.
6394
*/
6395

6396 6397
int QUICK_INDEX_MERGE_SELECT::get_next()
{
6398
  int result;
6399
  DBUG_ENTER("QUICK_INDEX_MERGE_SELECT::get_next");
6400

6401 6402 6403 6404 6405 6406 6407 6408 6409
  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);
6410
    /* All rows from Unique have been retrieved, do a clustered PK scan */
monty@mysql.com's avatar
monty@mysql.com committed
6411
    if (pk_quick_select)
6412
    {
monty@mysql.com's avatar
monty@mysql.com committed
6413
      doing_pk_scan= TRUE;
sergefp@mysql.com's avatar
sergefp@mysql.com committed
6414
      if ((result= pk_quick_select->init()) || (result= pk_quick_select->reset()))
6415 6416 6417 6418 6419 6420
        DBUG_RETURN(result);
      DBUG_RETURN(pk_quick_select->get_next());
    }
  }

  DBUG_RETURN(result);
6421 6422
}

6423 6424

/*
6425
  Retrieve next record.
6426
  SYNOPSIS
6427 6428
     QUICK_ROR_INTERSECT_SELECT::get_next()

6429
  NOTES
6430 6431
    Invariant on enter/exit: all intersected selects have retrieved all index
    records with rowid <= some_rowid_val and no intersected select has
6432 6433 6434 6435
    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.

6436
    If a Clustered PK scan is present, it is used only to check if row
6437 6438 6439 6440 6441
    satisfies its condition (and never used for row retrieval).

  RETURN
   0     - Ok
   other - Error code if any error occurred.
6442 6443 6444 6445 6446 6447 6448 6449 6450
*/

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");
6451

6452 6453 6454 6455 6456 6457 6458 6459 6460 6461
  /* 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();
6462

6463 6464 6465 6466 6467 6468
  if (error)
    DBUG_RETURN(error);

  quick->file->position(quick->record);
  memcpy(last_rowid, quick->file->ref, head->file->ref_length);
  last_rowid_count= 1;
6469

6470 6471 6472 6473 6474 6475 6476
  while (last_rowid_count < quick_selects.elements)
  {
    if (!(quick= quick_it++))
    {
      quick_it.rewind();
      quick= quick_it++;
    }
6477

6478 6479 6480 6481
    do {
      if ((error= quick->get_next()))
        DBUG_RETURN(error);
      quick->file->position(quick->record);
6482
      cmp= head->file->cmp_ref(quick->file->ref, last_rowid);
6483 6484 6485 6486 6487 6488 6489 6490 6491 6492 6493 6494 6495 6496 6497
    } 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);
6498
      last_rowid_count= 1;
6499 6500 6501 6502 6503 6504 6505 6506 6507 6508 6509 6510 6511 6512 6513
    }
    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);
}


6514 6515
/*
  Retrieve next record.
6516 6517
  SYNOPSIS
    QUICK_ROR_UNION_SELECT::get_next()
6518

6519
  NOTES
6520 6521
    Enter/exit invariant:
    For each quick select in the queue a {key,rowid} tuple has been
6522
    retrieved but the corresponding row hasn't been passed to output.
6523

6524
  RETURN
6525 6526
   0     - Ok
   other - Error code if any error occurred.
6527 6528 6529 6530 6531 6532 6533 6534
*/

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");
6535

6536 6537 6538 6539
  do
  {
    if (!queue.elements)
      DBUG_RETURN(HA_ERR_END_OF_FILE);
6540
    /* Ok, we have a queue with >= 1 scans */
6541 6542 6543 6544 6545 6546 6547 6548 6549 6550 6551 6552 6553 6554 6555 6556

    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);
    }
6557

6558 6559 6560
    if (!have_prev_rowid)
    {
      /* No rows have been returned yet */
monty@mysql.com's avatar
monty@mysql.com committed
6561 6562
      dup_row= FALSE;
      have_prev_rowid= TRUE;
6563 6564 6565 6566
    }
    else
      dup_row= !head->file->cmp_ref(cur_rowid, prev_rowid);
  }while (dup_row);
6567

6568 6569 6570 6571 6572 6573 6574 6575
  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
6576
int QUICK_RANGE_SELECT::reset()
ingo@mysql.com's avatar
ingo@mysql.com committed
6577 6578 6579
{
  uint  mrange_bufsiz;
  byte  *mrange_buff;
sergefp@mysql.com's avatar
sergefp@mysql.com committed
6580 6581 6582
  DBUG_ENTER("QUICK_RANGE_SELECT::reset");
  next=0;
  range= NULL;
6583
  in_range= FALSE;
sergefp@mysql.com's avatar
sergefp@mysql.com committed
6584
  cur_range= (QUICK_RANGE**) ranges.buffer;
igor@rurik.mysql.com's avatar
igor@rurik.mysql.com committed
6585 6586

  if (file->inited == handler::NONE && (error= file->ha_index_init(index)))
igor@rurik.mysql.com's avatar
igor@rurik.mysql.com committed
6587
    DBUG_RETURN(error);
igor@rurik.mysql.com's avatar
igor@rurik.mysql.com committed
6588
 
ingo@mysql.com's avatar
ingo@mysql.com committed
6589 6590 6591 6592 6593 6594 6595
  /* 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
6596 6597
  /* Allocate the ranges array. */
  DBUG_ASSERT(ranges.elements);
ingo@mysql.com's avatar
ingo@mysql.com committed
6598 6599 6600 6601 6602 6603 6604 6605 6606 6607 6608 6609 6610 6611 6612 6613
  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
6614
  /* Allocate the handler buffer if necessary.  */
ingo@mysql.com's avatar
ingo@mysql.com committed
6615 6616 6617
  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
6618
                       (QUICK_SELECT_I::records + 1)* head->s->reclength);
ingo@mysql.com's avatar
ingo@mysql.com committed
6619 6620 6621 6622 6623 6624 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 6651 6652 6653 6654 6655 6656 6657 6658 6659

    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
6660

6661
int QUICK_RANGE_SELECT::get_next()
bk@work.mysql.com's avatar
bk@work.mysql.com committed
6662
{
ingo@mysql.com's avatar
ingo@mysql.com committed
6663 6664 6665 6666
  int             result;
  KEY_MULTI_RANGE *mrange;
  key_range       *start_key;
  key_range       *end_key;
6667
  DBUG_ENTER("QUICK_RANGE_SELECT::get_next");
ingo@mysql.com's avatar
ingo@mysql.com committed
6668 6669 6670
  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
6671 6672 6673

  for (;;)
  {
ingo@mysql.com's avatar
ingo@mysql.com committed
6674
    if (in_range)
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
6675
    {
ingo@mysql.com's avatar
ingo@mysql.com committed
6676 6677
      /* 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
6678
      if (result != HA_ERR_END_OF_FILE)
ingo@mysql.com's avatar
ingo@mysql.com committed
6679 6680
      {
        in_range= ! result;
6681
	DBUG_RETURN(result);
ingo@mysql.com's avatar
ingo@mysql.com committed
6682
      }
bk@work.mysql.com's avatar
bk@work.mysql.com committed
6683
    }
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
6684

ingo@mysql.com's avatar
ingo@mysql.com committed
6685 6686 6687 6688 6689 6690 6691 6692 6693 6694 6695 6696 6697 6698 6699 6700 6701 6702 6703 6704 6705 6706 6707 6708 6709 6710 6711 6712 6713 6714
    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
6715

ingo@mysql.com's avatar
ingo@mysql.com committed
6716 6717
      mrange_slot->range_flag= range->flag;
    }
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
6718

ingo@mysql.com's avatar
ingo@mysql.com committed
6719 6720
    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
6721
    if (result != HA_ERR_END_OF_FILE)
ingo@mysql.com's avatar
ingo@mysql.com committed
6722 6723
    {
      in_range= ! result;
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
6724
      DBUG_RETURN(result);
ingo@mysql.com's avatar
ingo@mysql.com committed
6725 6726
    }
    in_range= FALSE; /* No matching rows; go to next set of ranges. */
bk@work.mysql.com's avatar
bk@work.mysql.com committed
6727 6728 6729
  }
}

6730 6731 6732 6733 6734 6735
/*
  Get the next record with a different prefix.

  SYNOPSIS
    QUICK_RANGE_SELECT::get_next_prefix()
    prefix_length  length of cur_prefix
6736
    cur_prefix     prefix of a key to be searched for
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

  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. */
6769
      DBUG_ASSERT(cur_prefix != 0);
6770 6771 6772 6773 6774 6775
      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
6776 6777 6778 6779 6780 6781 6782 6783
    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++);
6784 6785 6786 6787 6788 6789 6790 6791 6792 6793 6794 6795 6796 6797 6798 6799 6800 6801 6802 6803 6804 6805 6806 6807 6808 6809 6810 6811 6812

    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
6813
/* Get next for geometrical indexes */
bk@work.mysql.com's avatar
bk@work.mysql.com committed
6814

pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
6815
int QUICK_RANGE_SELECT_GEOM::get_next()
bk@work.mysql.com's avatar
bk@work.mysql.com committed
6816
{
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
6817
  DBUG_ENTER("QUICK_RANGE_SELECT_GEOM::get_next");
bk@work.mysql.com's avatar
bk@work.mysql.com committed
6818

pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
6819
  for (;;)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
6820
  {
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
6821 6822
    int result;
    if (range)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
6823
    {
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
6824 6825 6826 6827 6828
      // 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
6829
    }
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
6830

ingo@mysql.com's avatar
ingo@mysql.com committed
6831 6832 6833 6834 6835 6836 6837 6838
    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
6839 6840 6841 6842 6843 6844 6845 6846

    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
6847 6848 6849
  }
}

6850

6851 6852 6853 6854
/*
  Check if current row will be retrieved by this QUICK_RANGE_SELECT

  NOTES
6855 6856
    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
6857
    quick select.
6858
    The implementation does a binary search on sorted array of disjoint
6859 6860
    ranges, without taking size of range into account.

6861
    This function is used to filter out clustered PK scan rows in
6862 6863
    index_merge quick select.

6864
  RETURN
monty@mysql.com's avatar
monty@mysql.com committed
6865 6866
    TRUE  if current row will be retrieved by this quick select
    FALSE if not
6867 6868 6869 6870 6871 6872 6873 6874 6875 6876
*/

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)
6877
  {
6878 6879 6880 6881 6882 6883 6884 6885 6886 6887 6888 6889 6890
    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));
}

6891
/*
6892 6893 6894 6895 6896 6897 6898
  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.
6899
 */
6900

6901
QUICK_SELECT_DESC::QUICK_SELECT_DESC(QUICK_RANGE_SELECT *q,
6902 6903
                                     uint used_key_parts)
 : QUICK_RANGE_SELECT(*q), rev_it(rev_ranges)
6904
{
monty@hundin.mysql.fi's avatar
monty@hundin.mysql.fi committed
6905
  QUICK_RANGE *r;
6906

6907 6908
  QUICK_RANGE **pr= (QUICK_RANGE**)ranges.buffer;
  QUICK_RANGE **last_range= pr + ranges.elements;
monty@mysql.com's avatar
monty@mysql.com committed
6909 6910
  for (; pr!=last_range; pr++)
    rev_ranges.push_front(*pr);
monty@hundin.mysql.fi's avatar
monty@hundin.mysql.fi committed
6911

6912
  /* Remove EQ_RANGE flag for keys that are not using the full key */
monty@hundin.mysql.fi's avatar
monty@hundin.mysql.fi committed
6913
  for (r = rev_it++; r; r = rev_it++)
6914 6915 6916 6917 6918 6919 6920 6921
  {
    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;
6922 6923
}

6924

6925 6926 6927 6928 6929 6930
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
6931 6932
   *   - 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
6933 6934 6935 6936 6937 6938 6939 6940 6941 6942 6943 6944
   *     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
6945 6946 6947
      result = ((range->flag & EQ_RANGE)
		? file->index_next_same(record, (byte*) range->min_key,
					range->min_length) :
6948 6949 6950 6951 6952 6953 6954 6955 6956 6957 6958 6959 6960 6961 6962
		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
    {
6963 6964 6965
      int local_error;
      if ((local_error=file->index_last(record)))
	DBUG_RETURN(local_error);		// Empty table
6966 6967 6968 6969 6970 6971
      if (cmp_prev(range) == 0)
	DBUG_RETURN(0);
      range=0;			// No matching records; go to next range
      continue;
    }

6972
    if (range->flag & EQ_RANGE)
6973 6974 6975 6976 6977 6978
    {
      result = file->index_read(record, (byte*) range->max_key,
				range->max_length, HA_READ_KEY_EXACT);
    }
    else
    {
6979 6980 6981 6982 6983
      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));
6984 6985 6986 6987 6988 6989 6990 6991 6992 6993 6994 6995 6996 6997 6998 6999 7000 7001
    }
    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
  }
}

7002

pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
7003 7004 7005 7006 7007 7008 7009 7010 7011 7012 7013 7014 7015 7016 7017 7018 7019 7020 7021 7022 7023 7024 7025 7026 7027 7028 7029 7030 7031 7032 7033 7034 7035 7036 7037 7038 7039 7040 7041 7042 7043
/*
  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
}


7044
/*
7045 7046 7047
  Returns 0 if found key is inside range (found key >= range->min_key).
*/

7048
int QUICK_RANGE_SELECT::cmp_prev(QUICK_RANGE *range_arg)
7049
{
7050
  int cmp;
7051
  if (range_arg->flag & NO_MIN_RANGE)
7052
    return 0;					/* key can't be to small */
7053

monty@mysql.com's avatar
monty@mysql.com committed
7054 7055
  cmp= key_cmp(key_part_info, (byte*) range_arg->min_key,
               range_arg->min_length);
7056 7057 7058
  if (cmp > 0 || cmp == 0 && !(range_arg->flag & NEAR_MIN))
    return 0;
  return 1;                                     // outside of range
7059 7060
}

7061

7062
/*
monty@mysql.com's avatar
monty@mysql.com committed
7063
 * TRUE if this range will require using HA_READ_AFTER_KEY
7064
   See comment in get_next() about this
7065
 */
7066

7067
bool QUICK_SELECT_DESC::range_reads_after_key(QUICK_RANGE *range_arg)
7068
{
7069
  return ((range_arg->flag & (NO_MAX_RANGE | NEAR_MAX)) ||
7070
	  !(range_arg->flag & EQ_RANGE) ||
7071
	  head->key_info[index].key_length != range_arg->max_length) ? 1 : 0;
7072 7073
}

7074

monty@mysql.com's avatar
monty@mysql.com committed
7075
/* TRUE if we are reading over a key that may have a NULL value */
7076

7077
#ifdef NOT_USED
7078
bool QUICK_SELECT_DESC::test_if_null_range(QUICK_RANGE *range_arg,
7079 7080
					   uint used_key_parts)
{
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
7081
  uint offset, end;
7082 7083 7084
  KEY_PART *key_part = key_parts,
           *key_part_end= key_part+used_key_parts;

7085
  for (offset= 0,  end = min(range_arg->min_length, range_arg->max_length) ;
7086
       offset < end && key_part != key_part_end ;
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
7087
       offset+= key_part++->store_length)
7088
  {
7089 7090
    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
7091
		key_part->store_length))
7092
      continue;
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
7093 7094

    if (key_part->null_bit && range_arg->min_key[offset])
7095 7096 7097 7098 7099 7100 7101 7102 7103 7104 7105 7106
      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)
  {
7107
    if (offset >= range_arg->min_length || range_arg->min_key[offset])
7108 7109 7110 7111 7112 7113 7114 7115 7116 7117 7118 7119
      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;
}
7120
#endif
7121 7122


7123 7124 7125 7126 7127 7128 7129 7130 7131
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
7132
  bool first= TRUE;
7133
  List_iterator_fast<QUICK_RANGE_SELECT> it(quick_selects);
7134
  str->append(STRING_WITH_LEN("sort_union("));
7135 7136 7137 7138 7139
  while ((quick= it++))
  {
    if (!first)
      str->append(',');
    else
monty@mysql.com's avatar
monty@mysql.com committed
7140
      first= FALSE;
7141 7142 7143 7144 7145 7146 7147 7148 7149 7150 7151 7152
    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)
{
7153
  bool first= TRUE;
7154 7155
  QUICK_RANGE_SELECT *quick;
  List_iterator_fast<QUICK_RANGE_SELECT> it(quick_selects);
7156
  str->append(STRING_WITH_LEN("intersect("));
7157 7158 7159 7160 7161
  while ((quick= it++))
  {
    KEY *key_info= head->key_info + quick->index;
    if (!first)
      str->append(',');
7162
    else
monty@mysql.com's avatar
monty@mysql.com committed
7163
      first= FALSE;
7164 7165 7166 7167 7168 7169 7170 7171 7172 7173 7174 7175 7176
    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)
{
7177
  bool first= TRUE;
7178 7179
  QUICK_SELECT_I *quick;
  List_iterator_fast<QUICK_SELECT_I> it(quick_selects);
7180
  str->append(STRING_WITH_LEN("union("));
7181 7182 7183 7184 7185
  while ((quick= it++))
  {
    if (!first)
      str->append(',');
    else
monty@mysql.com's avatar
monty@mysql.com committed
7186
      first= FALSE;
7187 7188 7189 7190 7191 7192
    quick->add_info_string(str);
  }
  str->append(')');
}


7193
void QUICK_RANGE_SELECT::add_keys_and_lengths(String *key_names,
7194
                                              String *used_lengths)
7195 7196 7197 7198 7199 7200 7201 7202 7203
{
  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);
}

7204 7205
void QUICK_INDEX_MERGE_SELECT::add_keys_and_lengths(String *key_names,
                                                    String *used_lengths)
7206 7207 7208
{
  char buf[64];
  uint length;
monty@mysql.com's avatar
monty@mysql.com committed
7209
  bool first= TRUE;
7210
  QUICK_RANGE_SELECT *quick;
7211

7212 7213 7214
  List_iterator_fast<QUICK_RANGE_SELECT> it(quick_selects);
  while ((quick= it++))
  {
7215
    if (first)
monty@mysql.com's avatar
monty@mysql.com committed
7216
      first= FALSE;
7217 7218
    else
    {
7219 7220
      key_names->append(',');
      used_lengths->append(',');
7221
    }
7222

7223 7224
    KEY *key_info= head->key_info + quick->index;
    key_names->append(key_info->name);
7225 7226 7227 7228 7229 7230 7231 7232 7233 7234 7235 7236 7237 7238
    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);
  }
}

7239 7240
void QUICK_ROR_INTERSECT_SELECT::add_keys_and_lengths(String *key_names,
                                                      String *used_lengths)
7241 7242 7243
{
  char buf[64];
  uint length;
7244
  bool first= TRUE;
7245 7246 7247 7248 7249 7250
  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
7251
      first= FALSE;
7252
    else
7253 7254
    {
      key_names->append(',');
7255
      used_lengths->append(',');
7256 7257
    }
    key_names->append(key_info->name);
7258 7259 7260
    length= longlong2str(quick->max_used_key_length, buf, 10) - buf;
    used_lengths->append(buf, length);
  }
7261

7262 7263 7264 7265 7266 7267 7268 7269 7270 7271 7272
  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);
  }
}

7273 7274
void QUICK_ROR_UNION_SELECT::add_keys_and_lengths(String *key_names,
                                                  String *used_lengths)
7275
{
7276
  bool first= TRUE;
7277 7278 7279 7280 7281
  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
7282
      first= FALSE;
7283
    else
7284
    {
7285 7286 7287
      used_lengths->append(',');
      key_names->append(',');
    }
7288
    quick->add_keys_and_lengths(key_names, used_lengths);
7289 7290 7291
  }
}

7292 7293 7294 7295 7296 7297 7298 7299 7300

/*******************************************************************************
* 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
7301
get_constant_key_infix(KEY *index_info, SEL_ARG *index_range_tree,
7302
                       KEY_PART_INFO *first_non_group_part,
7303 7304 7305 7306
                       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);
7307
static bool
7308 7309
check_group_min_max_predicates(COND *cond, Item_field *min_max_arg_item,
                               Field::imagetype image_type);
7310

7311 7312 7313 7314 7315 7316
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);
7317

timour@mysql.com's avatar
timour@mysql.com committed
7318

7319 7320 7321 7322 7323 7324 7325 7326 7327 7328 7329 7330 7331 7332 7333 7334 7335 7336 7337 7338 7339 7340 7341 7342 7343
/*
  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
7344 7345
        - NGA = QA - (GA union C) = {NG_1, ..., NG_m} - the ones not in
          GROUP BY and not referenced by MIN/MAX functions.
7346
        with the following properties specified below.
7347 7348
    B3. If Q has a GROUP BY WITH ROLLUP clause the access method is not 
        applicable.
7349 7350 7351 7352 7353 7354 7355 7356 7357 7358

    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
7359 7360
         - C IS NOT NULL
         - C != const
7361 7362 7363
    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.
7364
    SA5. The select list in DISTINCT queries should not contain expressions.
7365 7366 7367 7368 7369 7370 7371 7372 7373 7374 7375 7376 7377 7378 7379 7380 7381 7382 7383 7384 7385 7386 7387 7388 7389 7390 7391 7392 7393
    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:
7394 7395 7396 7397
       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)]
7398 7399
         [AND PC(C)]
         [AND PA(A_i1,...,A_iq)]
7400 7401 7402 7403
       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:
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
       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?
7433 7434
  - Lift the limitation in condition (B3), that is, make this access method 
    applicable to ROLLUP queries.
7435 7436 7437 7438 7439 7440 7441 7442 7443 7444 7445 7446 7447 7448 7449 7450 7451 7452 7453 7454 7455 7456 7457 7458 7459 7460 7461 7462 7463 7464 7465 7466 7467 7468 7469 7470 7471 7472 7473

  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. */
7474 7475
       (!join->select_distinct)) ||
      (thd->lex->select_lex.olap == ROLLUP_TYPE)) /* Check (B3) for ROLLUP */
7476
    DBUG_RETURN(NULL);
7477
  if (table->s->keys == 0)        /* There are no indexes to use. */
7478 7479 7480
    DBUG_RETURN(NULL);

  /* Analyze the query in more detail. */
7481
  List_iterator<Item> select_items_it(join->fields_list);
7482

7483
  /* Check (SA1,SA4) and store the only MIN/MAX argument - the C attribute.*/
monty@mishka.local's avatar
monty@mishka.local committed
7484
  if (join->make_sum_func_list(join->all_fields, join->fields_list, 1))
7485 7486
    DBUG_RETURN(NULL);
  if (join->sum_funcs[0])
7487
  {
7488 7489 7490
    Item_sum *min_max_item;
    Item_sum **func_ptr= join->sum_funcs;
    while ((min_max_item= *(func_ptr++)))
7491
    {
7492 7493 7494 7495 7496
      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
7497 7498
        DBUG_RETURN(NULL);

7499 7500
      Item *expr= min_max_item->args[0];    /* The argument of MIN/MAX. */
      if (expr->type() == Item::FIELD_ITEM) /* Is it an attribute? */
7501
      {
7502 7503 7504 7505
        if (! min_max_arg_item)
          min_max_arg_item= (Item_field*) expr;
        else if (! min_max_arg_item->eq(expr, 1))
          DBUG_RETURN(NULL);
7506
      }
7507 7508
      else
        DBUG_RETURN(NULL);
7509
    }
7510
  }
7511

7512 7513 7514 7515
  /* Check (SA5). */
  if (join->select_distinct)
  {
    while ((item= select_items_it++))
7516
    {
7517 7518
      if (item->type() != Item::FIELD_ITEM)
        DBUG_RETURN(NULL);
7519 7520 7521 7522 7523 7524 7525 7526 7527 7528 7529 7530 7531 7532 7533 7534
    }
  }

  /* 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;
7535
  KEY *cur_index_info_end= cur_index_info + table->s->keys;
7536
  KEY_PART_INFO *cur_part= NULL;
7537 7538 7539 7540 7541 7542 7543 7544 7545 7546 7547 7548 7549 7550 7551 7552 7553 7554
  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;
7555
  uint cur_param_idx=MAX_KEY;
timour@mysql.com's avatar
timour@mysql.com committed
7556
  key_map cur_used_key_parts;
timour@mysql.com's avatar
timour@mysql.com committed
7557
  uint pk= param->table->s->primary_key;
7558 7559 7560 7561 7562 7563 7564

  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;
7565

timour@mysql.com's avatar
timour@mysql.com committed
7566 7567 7568 7569 7570 7571 7572 7573 7574 7575 7576 7577 7578 7579 7580 7581 7582 7583 7584 7585 7586 7587 7588 7589
    /*
      If the current storage manager is such that it appends the primary key to
      each index, then the above condition is insufficient to check if the
      index is covering. In such cases it may happen that some fields are
      covered by the PK index, but not by the current index. Since we can't
      use the concatenation of both indexes for index lookup, such an index
      does not qualify as covering in our case. If this is the case, below
      we check that all query fields are indeed covered by 'cur_index'.
    */
    if (pk < MAX_KEY && cur_index != pk &&
        (table->file->table_flags() & HA_PRIMARY_KEY_IN_READ_INDEX))
    {
      /* For each table field */
      for (uint i= 0; i < table->s->fields; i++)
      {
        Field *cur_field= table->field[i];
        /*
          If the field is used in the current query, check that the
          field is covered by some keypart of the current index.
        */
        if (thd->query_id == cur_field->query_id)
        {
          KEY_PART_INFO *key_part= cur_index_info->key_part;
          KEY_PART_INFO *key_part_end= key_part + cur_index_info->key_parts;
7590
          for (;;)
timour@mysql.com's avatar
timour@mysql.com committed
7591 7592 7593
          {
            if (key_part->field == cur_field)
              break;
7594 7595
            if (++key_part == key_part_end)
              goto next_index;                  // Field was not part of key
timour@mysql.com's avatar
timour@mysql.com committed
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
    /*
      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))
        {
7622 7623
          cur_group_prefix_len+= cur_part->store_length;
          ++cur_group_key_parts;
7624 7625 7626 7627 7628 7629 7630 7631 7632 7633 7634 7635 7636 7637 7638
        }
        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)
    {
7639
      select_items_it.rewind();
timour@mysql.com's avatar
timour@mysql.com committed
7640
      cur_used_key_parts.clear_all();
7641
      uint max_key_part= 0;
7642
      while ((item= select_items_it++))
7643
      {
7644
        item_field= (Item_field*) item; /* (SA5) already checked above. */
7645 7646
        /* 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
7647 7648 7649 7650 7651 7652
        /*
          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;
7653
        if (key_part_nr < 1 || key_part_nr > join->fields_list.elements)
7654 7655
          goto next_index;
        cur_part= cur_index_info->key_part + key_part_nr - 1;
7656
        cur_group_prefix_len+= cur_part->store_length;
timour@mysql.com's avatar
timour@mysql.com committed
7657 7658
        cur_used_key_parts.set_bit(key_part_nr);
        ++cur_group_key_parts;
7659
        max_key_part= max(max_key_part,key_part_nr);
7660
      }
7661 7662 7663 7664 7665 7666 7667 7668 7669 7670 7671
      /*
        Check that used key parts forms a prefix of the index.
        To check this we compare bits in all_parts and cur_parts.
        all_parts have all bits set from 0 to (max_key_part-1).
        cur_parts have bits set for only used keyparts.
      */
      ulonglong all_parts, cur_parts;
      all_parts= (1<<max_key_part) - 1;
      cur_parts= cur_used_key_parts.to_ulonglong() >> 1;
      if (all_parts != cur_parts)
        goto next_index;
7672 7673 7674 7675 7676 7677 7678 7679
    }
    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);
7680
      if (key_part_nr <= cur_group_key_parts)
7681 7682 7683 7684 7685 7686 7687 7688
        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.
    */
7689 7690 7691 7692 7693 7694 7695 7696 7697 7698 7699 7700 7701 7702 7703 7704 7705 7706 7707 7708 7709 7710

    /*
      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)))
7711
    {
7712 7713 7714 7715 7716 7717 7718 7719 7720 7721 7722 7723 7724
      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))
timour@mysql.com's avatar
timour@mysql.com committed
7725
      {
7726 7727 7728 7729
        /*
          There is a gap but no range tree, thus no predicates at all for the
          non-group keyparts.
        */
7730
        goto next_index;
timour@mysql.com's avatar
timour@mysql.com committed
7731 7732 7733 7734 7735 7736 7737 7738 7739 7740 7741 7742 7743 7744 7745 7746 7747 7748 7749 7750 7751 7752 7753 7754
      }
      else if (first_non_group_part && join->conds)
      {
        /*
          If there is no MIN/MAX function in the query, but some index
          key part is referenced in the WHERE clause, then this index
          cannot be used because the WHERE condition over the keypart's
          field cannot be 'pushed' to the index (because there is no
          range 'tree'), and the WHERE clause must be evaluated before
          GROUP BY/DISTINCT.
        */
        /*
          Store the first and last keyparts that need to be analyzed
          into one array that can be passed as parameter.
        */
        KEY_PART_INFO *key_part_range[2];
        key_part_range[0]= first_non_group_part;
        key_part_range[1]= last_part;

        /* Check if cur_part is referenced in the WHERE clause. */
        if (join->conds->walk(&Item::find_item_in_field_list_processor,
                              (byte*) key_part_range))
          goto next_index;
      }
7755 7756
    }

7757 7758 7759 7760 7761 7762 7763 7764 7765 7766 7767 7768 7769
    /*
      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;
      }
    }

7770
    /* If we got to this point, cur_index_info passes the test. */
7771 7772 7773
    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;
7774

7775 7776 7777 7778 7779 7780 7781 7782 7783 7784 7785 7786 7787 7788
    /* 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
7789 7790 7791 7792 7793 7794
    /*
      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))
7795
    {
7796
      DBUG_ASSERT(tree != 0 || cur_param_idx == MAX_KEY);
7797 7798 7799 7800 7801 7802 7803 7804 7805 7806
      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;
    }
7807 7808

  next_index:
7809 7810
    cur_group_key_parts= 0;
    cur_group_prefix_len= 0;
7811 7812 7813 7814
  }
  if (!index_info) /* No usable index found. */
    DBUG_RETURN(NULL);

7815 7816 7817
  /* Check (SA3) for the where clause. */
  if (join->conds && min_max_arg_item &&
      !check_group_min_max_predicates(join->conds, min_max_arg_item,
7818 7819
                                      (index_info->flags & HA_SPATIAL) ?
                                      Field::itMBR : Field::itRAW))
7820 7821 7822 7823
    DBUG_RETURN(NULL);

  /* The query passes all tests, so construct a new TRP object. */
  read_plan= new (param->mem_root)
7824 7825 7826 7827
                 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,
7828
                                   (key_infix_len > 0) ? key_infix : NULL,
7829
                                   tree, best_index_tree, best_param_idx,
7830
                                   best_quick_prefix_records);
7831 7832 7833 7834 7835
  if (read_plan)
  {
    if (tree && read_plan->quick_prefix_records == 0)
      DBUG_RETURN(NULL);

7836 7837 7838
    read_plan->read_cost= best_read_cost;
    read_plan->records=   best_records;

7839 7840 7841 7842 7843 7844 7845 7846 7847 7848
    DBUG_PRINT("info",
               ("Returning group min/max plan: cost: %g, records: %lu",
                read_plan->read_cost, (ulong) read_plan->records));
  }

  DBUG_RETURN(read_plan);
}


/*
7849 7850
  Check that the MIN/MAX attribute participates only in range predicates
  with constants.
7851 7852 7853 7854 7855 7856

  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)
7857
    min_max_arg_part  the keypart of the MIN/MAX argument if any
7858 7859 7860

  DESCRIPTION
    The function walks recursively over the cond tree representing a WHERE
7861
    clause, and checks condition (SA3) - if a field is referenced by a MIN/MAX
7862 7863
    aggregate function, it is referenced only by one of the following
    predicates: {=, !=, <, <=, >, >=, between, is null, is not null}.
7864 7865 7866 7867 7868 7869 7870

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

static bool
7871 7872
check_group_min_max_predicates(COND *cond, Item_field *min_max_arg_item,
                               Field::imagetype image_type)
7873 7874
{
  DBUG_ENTER("check_group_min_max_predicates");
7875
  DBUG_ASSERT(cond && min_max_arg_item);
7876 7877 7878 7879 7880 7881 7882 7883 7884

  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++))
    {
monty@mishka.local's avatar
monty@mishka.local committed
7885
      if (!check_group_min_max_predicates(and_or_arg, min_max_arg_item,
7886
                                         image_type))
7887 7888 7889 7890 7891
        DBUG_RETURN(FALSE);
    }
    DBUG_RETURN(TRUE);
  }

7892 7893 7894 7895 7896 7897 7898 7899 7900 7901 7902 7903 7904
  /*
    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. */
7905 7906 7907 7908 7909 7910 7911 7912 7913 7914 7915 7916 7917
  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)
    {
7918
      if (min_max_arg_item->eq(cur_arg, 1)) 
7919 7920 7921
      {
       /*
         If pred references the MIN/MAX argument, check whether pred is a range
7922
         condition that compares the MIN/MAX argument with a constant.
7923 7924
       */
        Item_func::Functype pred_type= pred->functype();
7925 7926 7927 7928 7929 7930 7931 7932 7933 7934
        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)
7935 7936 7937 7938
          DBUG_RETURN(FALSE);

        /* Check that pred compares min_max_arg_item with a constant. */
        Item *args[3];
7939
        bzero(args, 3 * sizeof(Item*));
7940 7941 7942 7943
        bool inv;
        /* Test if this is a comparison of a field and a constant. */
        if (!simple_pred(pred, args, &inv))
          DBUG_RETURN(FALSE);
7944 7945 7946 7947 7948 7949 7950 7951 7952 7953 7954 7955 7956 7957 7958 7959 7960 7961 7962

        /* 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);
7963 7964 7965 7966
      }
    }
    else if (cur_arg->type() == Item::FUNC_ITEM)
    {
monty@mishka.local's avatar
monty@mishka.local committed
7967
      if (!check_group_min_max_predicates(cur_arg, min_max_arg_item,
7968
                                         image_type))
7969 7970 7971 7972 7973 7974 7975 7976 7977 7978 7979 7980 7981 7982 7983 7984 7985 7986 7987
        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()
7988 7989 7990 7991 7992 7993 7994 7995 7996
    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)
7997 7998 7999
    
  DESCRIPTION
    Test conditions (NGA1, NGA2) from get_best_group_min_max(). Namely,
8000 8001
    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
8002 8003
    (const_ci = NGF_i).
    Thus all the NGF_i attributes must fill the 'gap' between the last group-by
8004 8005 8006 8007 8008 8009
    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
8010
    TRUE  if the index passes the test
8011 8012 8013 8014
    FALSE o/w
*/

static bool
8015
get_constant_key_infix(KEY *index_info, SEL_ARG *index_range_tree,
8016
                       KEY_PART_INFO *first_non_group_part,
8017 8018 8019 8020
                       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)
8021 8022 8023
{
  SEL_ARG       *cur_range;
  KEY_PART_INFO *cur_part;
8024 8025
  /* End part for the first loop below. */
  KEY_PART_INFO *end_part= min_max_arg_part ? min_max_arg_part : last_part;
8026 8027 8028 8029 8030 8031 8032 8033 8034 8035 8036 8037 8038 8039 8040 8041 8042

  *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)
    {
8043 8044 8045 8046 8047 8048 8049
      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;
      }
8050 8051 8052 8053 8054 8055 8056 8057 8058 8059 8060 8061 8062 8063 8064 8065 8066 8067 8068 8069 8070 8071 8072 8073
    }

    /* 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;
  }

8074 8075 8076
  if (!min_max_arg_part && (cur_part == last_part))
    *first_non_infix_part= last_part;

8077 8078 8079 8080
  return TRUE;
}


8081 8082 8083 8084 8085 8086 8087 8088 8089 8090 8091 8092 8093 8094 8095 8096 8097 8098 8099 8100
/*
  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)
{
8101
  KEY_PART_INFO *part, *end;
8102

8103
  for (part= index->key_part, end= part + index->key_parts; part < end; part++)
8104 8105
  {
    if (field->eq(part->field))
ram@gw.mysql.r18.ru's avatar
ram@gw.mysql.r18.ru committed
8106
      return part - index->key_part + 1;
8107
  }
8108
  return 0;
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
}


/*
  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]);
}


8150
/*
8151
  Compute the cost of a quick_group_min_max_select for a particular index.
8152 8153

  SYNOPSIS
8154 8155 8156 8157 8158 8159 8160
    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
8161 8162
    quick_prefix_records [in] Number of records retrieved by the internally
			      used quick range select if any
8163 8164 8165 8166
    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
8167 8168

  DESCRIPTION
monty@mysql.com's avatar
monty@mysql.com committed
8169 8170
    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.
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

  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
*/

8210 8211 8212 8213 8214
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)
8215 8216 8217 8218 8219 8220 8221 8222 8223 8224 8225 8226
{
  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
8227
  DBUG_ENTER("cost_group_min_max");
monty@mysql.com's avatar
monty@mysql.com committed
8228

8229 8230 8231 8232 8233 8234 8235 8236 8237 8238 8239 8240 8241 8242 8243 8244 8245 8246
  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
8247
    num_groups= (uint) rint(num_groups * quick_prefix_selectivity);
igor@rurik.mysql.com's avatar
igor@rurik.mysql.com committed
8248
    set_if_bigger(num_groups, 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 8273 8274 8275 8276 8277 8278 8279
  }

  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;

8280
  *read_cost= io_cost + cpu_cost;
8281
  *records= num_groups;
8282 8283

  DBUG_PRINT("info",
8284 8285
             ("table rows=%u, keys/block=%u, keys/group=%u, result rows=%u, blocks=%u",
              table_records, keys_per_block, keys_per_group, *records,
8286 8287 8288 8289 8290 8291 8292 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
              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");

8319 8320 8321 8322 8323
  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
8324 8325
                                        key_infix_len, key_infix,
                                        parent_alloc);
8326 8327 8328 8329 8330 8331 8332 8333 8334 8335 8336 8337 8338 8339 8340 8341
  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. */
8342 8343
      quick->quick_prefix_select= get_quick_select(param, param_idx,
                                                   index_tree,
8344 8345 8346 8347 8348 8349 8350 8351 8352 8353 8354 8355 8356 8357 8358 8359 8360 8361 8362 8363 8364 8365
                                                   &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)
      {
8366
        if (quick->add_range(min_max_range))
8367 8368 8369 8370 8371 8372 8373 8374 8375 8376 8377 8378 8379
        {
          delete quick;
          quick= NULL;
          DBUG_RETURN(NULL);
        }
        min_max_range= min_max_range->next;
      }
    }
  }
  else
    quick->quick_prefix_select= NULL;

  quick->update_key_stat();
8380
  quick->adjust_prefix_ranges();
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

  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
8410 8411 8412 8413 8414 8415 8416 8417 8418 8419 8420 8421 8422
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),
8423 8424
   key_infix_len(key_infix_len_arg), min_functions_it(NULL),
   max_functions_it(NULL)
8425 8426 8427 8428 8429 8430
{
  head=       table;
  file=       head->file;
  index=      use_index;
  record=     head->record[0];
  tmp_record= head->record[1];
8431 8432 8433
  read_time= read_cost_arg;
  records= records_arg;
  used_key_parts= used_key_parts_arg;
8434 8435 8436
  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
8437 8438 8439 8440 8441 8442

  /*
    We can't have parent_alloc set as the init function can't handle this case
    yet.
  */
  DBUG_ASSERT(!parent_alloc);
8443 8444 8445
  if (!parent_alloc)
  {
    init_sql_alloc(&alloc, join->thd->variables.range_alloc_block_size, 0);
monty@mysql.com's avatar
monty@mysql.com committed
8446
    join->thd->mem_root= &alloc;
8447 8448
  }
  else
8449
    bzero(&alloc, sizeof(MEM_ROOT));            // ensure that it's not used
8450 8451 8452 8453 8454 8455 8456 8457 8458
}


/*
  Do post-constructor initialization.

  SYNOPSIS
    QUICK_GROUP_MIN_MAX_SELECT::init()
  
8459 8460 8461 8462 8463 8464
  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.

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
  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)
  {
monty@mishka.local's avatar
monty@mishka.local committed
8500
    if (my_init_dynamic_array(&min_max_ranges, sizeof(QUICK_RANGE*), 16, 16))
8501 8502
      return 1;

8503 8504
    if (have_min)
    {
monty@mishka.local's avatar
monty@mishka.local committed
8505
      if (!(min_functions= new List<Item_sum>))
8506 8507 8508 8509 8510 8511
        return 1;
    }
    else
      min_functions= NULL;
    if (have_max)
    {
monty@mishka.local's avatar
monty@mishka.local committed
8512
      if (!(max_functions= new List<Item_sum>))
8513 8514 8515 8516
        return 1;
    }
    else
      max_functions= NULL;
8517

8518 8519 8520
    Item_sum *min_max_item;
    Item_sum **func_ptr= join->sum_funcs;
    while ((min_max_item= *(func_ptr++)))
8521
    {
8522 8523 8524 8525
      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);
8526 8527
    }

8528 8529 8530 8531 8532 8533 8534 8535 8536 8537 8538
    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;
    }
8539
  }
igor@rurik.mysql.com's avatar
igor@rurik.mysql.com committed
8540 8541
  else
    min_max_ranges.elements= 0;
8542 8543 8544 8545 8546 8547 8548 8549 8550 8551 8552 8553 8554

  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));
8555 8556
  delete min_functions_it;
  delete max_functions_it;
8557 8558 8559 8560 8561 8562 8563 8564 8565 8566 8567 8568 8569 8570 8571 8572 8573 8574 8575
  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
8576 8577
    FALSE on success
    TRUE  otherwise
8578 8579 8580 8581 8582 8583 8584 8585
*/

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). */
monty@mishka.local's avatar
monty@mishka.local committed
8586
  if ((range_flag & NO_MIN_RANGE) && (range_flag & NO_MAX_RANGE))
8587
    return FALSE;
8588 8589 8590 8591 8592 8593 8594 8595 8596 8597 8598 8599 8600 8601 8602

  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)
8603
    return TRUE;
8604
  if (insert_dynamic(&min_max_ranges, (gptr)&range))
8605 8606
    return TRUE;
  return FALSE;
8607 8608 8609
}


8610 8611 8612 8613 8614 8615 8616 8617 8618 8619 8620 8621 8622 8623 8624 8625 8626 8627 8628 8629 8630 8631 8632 8633 8634 8635 8636 8637 8638 8639 8640 8641 8642 8643 8644 8645
/*
  Opens the ranges if there are more conditions in quick_prefix_select than
  the ones used for jumping through the prefixes.

  SYNOPSIS
    QUICK_GROUP_MIN_MAX_SELECT::adjust_prefix_ranges()

  NOTES
    quick_prefix_select is made over the conditions on the whole key.
    It defines a number of ranges of length x. 
    However when jumping through the prefixes we use only the the first 
    few most significant keyparts in the range key. However if there
    are more keyparts to follow the ones we are using we must make the 
    condition on the key inclusive (because x < "ab" means 
    x[0] < 'a' OR (x[0] == 'a' AND x[1] < 'b').
    To achive the above we must turn off the NEAR_MIN/NEAR_MAX
*/
void QUICK_GROUP_MIN_MAX_SELECT::adjust_prefix_ranges ()
{
  if (quick_prefix_select &&
      group_prefix_len < quick_prefix_select->max_used_key_length)
  {
    DYNAMIC_ARRAY *arr;
    uint inx;

    for (inx= 0, arr= &quick_prefix_select->ranges; inx < arr->elements; inx++)
    {
      QUICK_RANGE *range;

      get_dynamic(arr, (gptr)&range, inx);
      range->flag &= ~(NEAR_MIN | NEAR_MAX);
    }
  }
}


8646 8647 8648 8649 8650 8651 8652 8653 8654 8655 8656 8657 8658 8659 8660 8661 8662 8663 8664 8665 8666 8667 8668 8669 8670 8671
/*
  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)
  {
8672
    QUICK_RANGE *cur_range;
8673 8674 8675 8676 8677 8678 8679
    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;
8680
        used_key_parts++;
8681 8682 8683 8684 8685 8686 8687 8688 8689
        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;
8690
        used_key_parts++;
8691 8692 8693 8694
        return;
      }
    }
  }
8695 8696
  else if (have_min && min_max_arg_part &&
           min_max_arg_part->field->real_maybe_null())
8697
  {
8698 8699 8700 8701 8702 8703 8704 8705
    /*
      If a MIN/MAX argument value is NULL, we can quickly determine
      that we're in the beginning of the next group, because NULLs
      are always < any other value. This allows us to quickly
      determine the end of the current group and jump to the next
      group (see next_min()) and thus effectively increases the
      usable key length.
    */
8706
    max_used_key_length+= min_max_arg_len;
8707
    used_key_parts++;
8708 8709 8710 8711 8712 8713 8714 8715 8716 8717
  }
}


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

  SYNOPSIS
    QUICK_GROUP_MIN_MAX_SELECT::reset()

8718 8719 8720 8721
  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.

8722 8723 8724 8725 8726 8727 8728 8729 8730 8731 8732 8733 8734
  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);
sergefp@mysql.com's avatar
sergefp@mysql.com committed
8735 8736
  if (result == HA_ERR_END_OF_FILE)
    DBUG_RETURN(0);
8737 8738
  if (result)
    DBUG_RETURN(result);
sergefp@mysql.com's avatar
sergefp@mysql.com committed
8739 8740
  if (quick_prefix_select && quick_prefix_select->reset())
    DBUG_RETURN(1);
8741 8742 8743 8744 8745 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
  /* 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
8780 8781 8782 8783 8784 8785 8786
#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
8787
  int result;
timour@mysql.com's avatar
timour@mysql.com committed
8788
#endif
8789 8790 8791 8792 8793 8794 8795 8796 8797 8798 8799 8800 8801 8802 8803 8804 8805 8806 8807 8808 8809 8810 8811 8812 8813 8814 8815 8816 8817 8818 8819 8820 8821 8822 8823 8824 8825 8826 8827 8828
  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)));
    }
8829
    /*
8830
      If this is just a GROUP BY or DISTINCT without MIN or MAX and there
8831 8832 8833 8834 8835 8836 8837
      are equality predicates for the key parts after the group, find the
      first sub-group with the extended prefix.
    */
    if (!have_min && !have_max && key_infix_len > 0)
      result= file->index_read(record, group_prefix, real_prefix_len,
                               HA_READ_KEY_EXACT);

8838 8839 8840 8841 8842 8843 8844 8845 8846 8847 8848 8849 8850 8851 8852 8853 8854 8855 8856 8857 8858 8859 8860 8861 8862 8863
    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
8864 8865
    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.
8866 8867 8868 8869 8870 8871 8872 8873 8874 8875 8876 8877 8878 8879 8880 8881 8882 8883 8884 8885 8886 8887 8888 8889 8890 8891

  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
  {
8892
    /* Apply the constant equality conditions to the non-group select fields */
8893 8894 8895 8896 8897 8898 8899 8900 8901 8902 8903 8904 8905 8906 8907 8908 8909 8910 8911 8912 8913 8914 8915 8916 8917 8918 8919 8920 8921 8922 8923 8924 8925
    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)
      {
8926
        if (key_cmp(index_info->key_part, group_prefix, real_prefix_len))
8927
          key_restore(record, tmp_record, index_info, 0);
8928 8929
      }
      else if (result == HA_ERR_KEY_NOT_FOUND) 
8930 8931 8932 8933 8934 8935 8936 8937 8938 8939 8940 8941 8942 8943 8944 8945 8946 8947 8948
        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
8949
    Lookup the maximal key of the group, and store it into this->record.
8950 8951 8952 8953 8954 8955 8956 8957 8958 8959 8960 8961 8962 8963 8964 8965 8966 8967 8968 8969 8970 8971 8972 8973 8974 8975 8976 8977 8978 8979 8980 8981 8982 8983 8984 8985 8986 8987 8988 8989 8990 8991 8992 8993 8994 8995 8996 8997 8998 8999 9000 9001 9002 9003 9004 9005 9006 9007 9008 9009 9010 9011 9012 9013 9014 9015 9016 9017 9018 9019 9020 9021 9022 9023 9024 9025 9026 9027 9028 9029 9030 9031 9032 9033 9034 9035 9036 9037 9038 9039 9040 9041 9042 9043 9044 9045 9046 9047 9048 9049 9050 9051 9052 9053 9054 9055 9056 9057 9058 9059 9060 9061 9062 9063 9064 9065 9066 9067 9068 9069 9070 9071 9072 9073 9074 9075

  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) &&
9076
        (key_cmp(min_max_arg_part, (const byte*) cur_range->max_key,
9077
                 min_max_arg_len) == 1))
9078 9079 9080 9081 9082 9083 9084 9085 9086 9087 9088 9089 9090 9091 9092 9093 9094 9095 9096 9097 9098 9099 9100 9101
      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)
9102 9103 9104 9105 9106 9107
    {
      /*
        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.
      */
9108
      break;
9109
    }
9110 9111 9112 9113 9114 9115

    /* 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)
9116 9117 9118 9119 9120 9121
    {
      /*
        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);
9122 9123 9124 9125 9126 9127 9128 9129 9130 9131 9132 9133 9134 9135 9136 9137 9138 9139 9140 9141 9142 9143 9144 9145 9146 9147 9148 9149 9150 9151 9152 9153 9154 9155 9156 9157 9158 9159 9160 9161
      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)
  {
9162
    memcpy(record, tmp_record, head->s->rec_buff_length);
9163 9164 9165 9166 9167 9168 9169 9170 9171 9172 9173 9174 9175 9176 9177 9178 9179 9180 9181 9182 9183 9184 9185 9186 9187 9188 9189 9190 9191 9192 9193 9194 9195 9196 9197 9198 9199 9200 9201 9202 9203 9204 9205 9206 9207
    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) &&
9208
        (key_cmp(min_max_arg_part, (const byte*) cur_range->min_key,
9209
                 min_max_arg_len) == -1))
9210 9211 9212 9213 9214 9215 9216 9217 9218 9219 9220 9221 9222 9223 9224 9225 9226 9227 9228 9229 9230 9231 9232
      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. */
monty@mysql.com's avatar
monty@mysql.com committed
9233 9234
    if (result)
    {
9235 9236 9237 9238 9239
      /*
        In no key was found with this upper bound, there certainly are no keys
        in the ranges to the left.
      */
      return result;
monty@mysql.com's avatar
monty@mysql.com committed
9240
    }
9241 9242
    /* A key was found. */
    if (cur_range->flag & EQ_RANGE)
monty@mysql.com's avatar
monty@mysql.com committed
9243
      return 0; /* No need to perform the checks below for equal keys. */
9244 9245 9246

    /* Check if record belongs to the current group. */
    if (key_cmp(index_info->key_part, group_prefix, real_prefix_len))
monty@mysql.com's avatar
monty@mysql.com committed
9247
      continue;                                 // Row not found
9248 9249 9250 9251 9252 9253 9254 9255 9256 9257 9258 9259 9260 9261 9262 9263 9264 9265 9266 9267 9268 9269 9270 9271 9272 9273 9274 9275 9276 9277 9278 9279 9280 9281 9282 9283 9284 9285 9286 9287 9288 9289 9290 9291 9292 9293 9294 9295 9296 9297 9298 9299 9300 9301 9302 9303 9304 9305 9306 9307 9308 9309 9310 9311 9312 9313 9314 9315 9316 9317 9318 9319 9320 9321 9322 9323 9324 9325 9326 9327 9328 9329 9330 9331 9332 9333 9334 9335

    /* 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();
}


9336 9337 9338 9339 9340 9341 9342 9343 9344 9345 9346 9347 9348 9349 9350
/*
  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.

*/

9351 9352 9353 9354 9355 9356 9357 9358 9359 9360 9361
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);
}


9362
#ifndef DBUG_OFF
9363

9364 9365 9366 9367 9368 9369 9370 9371 9372
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;
9373

9374 9375 9376 9377 9378 9379 9380 9381 9382 9383 9384 9385 9386 9387 9388
  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())
9389
    tmp.append(STRING_WITH_LEN("(empty)"));
9390

9391
  DBUG_PRINT("info", ("SEL_TREE %p (%s) scans:%s", tree, msg, tmp.ptr()));
9392

9393 9394
  DBUG_VOID_RETURN;
}
9395

9396 9397 9398 9399

static void print_ror_scans_arr(TABLE *table, const char *msg,
                                struct st_ror_scan_info **start,
                                struct st_ror_scan_info **end)
9400
{
9401 9402 9403 9404 9405 9406 9407
  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);
9408
  for (;start != end; start++)
9409
  {
9410 9411 9412
    if (tmp.length())
      tmp.append(',');
    tmp.append(table->key_info[(*start)->keynr].name);
9413
  }
9414
  if (!tmp.length())
9415
    tmp.append(STRING_WITH_LEN("(empty)"));
9416 9417
  DBUG_PRINT("info", ("ROR key scans (%s): %s", msg, tmp.ptr()));
  DBUG_VOID_RETURN;
9418 9419
}

bk@work.mysql.com's avatar
bk@work.mysql.com committed
9420 9421 9422 9423 9424 9425 9426 9427 9428 9429 9430
/*****************************************************************************
** 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
9431
  const char *key_end= key+used_length;
9432
  String tmp(buff,sizeof(buff),&my_charset_bin);
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
9433
  uint store_length;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
9434

pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
9435
  for (; key < key_end; key+=store_length, key_part++)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
9436
  {
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
9437 9438 9439
    Field *field=      key_part->field;
    store_length= key_part->store_length;

bk@work.mysql.com's avatar
bk@work.mysql.com committed
9440 9441
    if (field->real_maybe_null())
    {
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
9442
      if (*key)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
9443 9444 9445 9446
      {
	fwrite("NULL",sizeof(char),4,DBUG_FILE);
	continue;
      }
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
9447 9448
      key++;					// Skip null byte
      store_length--;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
9449
    }
9450
    field->set_key_image((char*) key, key_part->length);
monty@mysql.com's avatar
monty@mysql.com committed
9451 9452 9453 9454
    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
9455
    fwrite(tmp.ptr(),sizeof(char),tmp.length(),DBUG_FILE);
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
9456 9457
    if (key+store_length < key_end)
      fputc('/',DBUG_FILE);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
9458 9459 9460
  }
}

pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
9461

9462
static void print_quick(QUICK_SELECT_I *quick, const key_map *needed_reg)
9463
{
9464
  char buf[MAX_KEY/8+1];
9465
  DBUG_ENTER("print_quick");
9466 9467
  if (! _db_on_ || !quick)
    DBUG_VOID_RETURN;
9468
  DBUG_LOCK_FILE;
9469

monty@mysql.com's avatar
monty@mysql.com committed
9470
  quick->dbug_dump(0, TRUE);
9471
  fprintf(DBUG_FILE,"other_keys: 0x%s:\n", needed_reg->print(buf));
9472

9473
  DBUG_UNLOCK_FILE;
9474 9475 9476
  DBUG_VOID_RETURN;
}

pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
9477

9478
static void print_rowid(byte* val, int len)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
9479
{
9480
  byte *pb;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
9481
  DBUG_LOCK_FILE;
9482 9483 9484 9485 9486 9487 9488 9489 9490 9491
  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;
}
9492

9493 9494 9495 9496
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);
9497

9498
  if (verbose)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
9499
  {
9500 9501
    QUICK_RANGE *range;
    QUICK_RANGE **pr= (QUICK_RANGE**)ranges.buffer;
9502
    QUICK_RANGE **last_range= pr + ranges.elements;
9503
    for (; pr!=last_range; ++pr)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
9504
    {
9505 9506 9507 9508 9509 9510 9511 9512 9513 9514 9515
      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
9516

9517 9518 9519 9520 9521 9522 9523 9524 9525
      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
9526 9527
    }
  }
9528 9529 9530 9531 9532 9533 9534 9535 9536 9537 9538 9539
}

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)
  {
9540
    fprintf(DBUG_FILE, "%*sclustered PK quick:\n", indent, "");
9541 9542 9543 9544 9545 9546 9547 9548 9549
    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;
9550
  fprintf(DBUG_FILE, "%*squick ROR-intersect select, %scovering\n",
9551 9552 9553
          indent, "", need_to_fetch_row? "":"non-");
  fprintf(DBUG_FILE, "%*smerged scans {\n", indent, "");
  while ((quick= it++))
9554
    quick->dbug_dump(indent+2, verbose);
9555 9556
  if (cpk_quick)
  {
9557
    fprintf(DBUG_FILE, "%*sclustered PK quick:\n", indent, "");
9558 9559 9560 9561 9562 9563 9564 9565 9566 9567 9568 9569 9570 9571
    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
9572 9573
}

9574 9575 9576 9577 9578 9579 9580 9581 9582 9583 9584 9585 9586 9587 9588 9589 9590 9591 9592 9593 9594 9595 9596 9597 9598 9599 9600 9601 9602 9603 9604 9605 9606 9607 9608 9609 9610 9611 9612 9613 9614 9615 9616 9617

/*
  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
9618
#endif /* NOT_USED */
bk@work.mysql.com's avatar
bk@work.mysql.com committed
9619 9620

/*****************************************************************************
9621
** Instantiate templates
bk@work.mysql.com's avatar
bk@work.mysql.com committed
9622 9623
*****************************************************************************/

9624
#ifdef HAVE_EXPLICIT_TEMPLATE_INSTANTIATION
bk@work.mysql.com's avatar
bk@work.mysql.com committed
9625 9626 9627
template class List<QUICK_RANGE>;
template class List_iterator<QUICK_RANGE>;
#endif