opt_range.cc 340 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 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62
  This file contains:

  RangeAnalysisModule  
    A module that accepts a condition, index (or partitioning) description, 
    and builds lists of intervals (in index/partitioning space), such that 
    all possible records that match the condition are contained within the 
    intervals.
    The entry point for the range analysis module is get_mm_tree() function.
    
    The lists are returned in form of complicated structure of interlinked
    SEL_TREE/SEL_IMERGE/SEL_ARG objects.
    See check_quick_keys, find_used_partitions for examples of how to walk 
    this structure.
    All direct "users" of this module are located within this file, too.


  PartitionPruningModule
    A module that accepts a partitioned table, condition, and finds which
    partitions we will need to use in query execution. Search down for
    "PartitionPruningModule" for description.
    The module has single entry point - prune_partitions() function.


  Range/index_merge/groupby-minmax optimizer module  
    A module that accepts a table, condition, and returns 
     - a QUICK_*_SELECT object that can be used to retrieve rows that match
       the specified condition, or a "no records will match the condition" 
       statement.

    The module entry points are
      test_quick_select()
      get_quick_select_for_ref()


  Record retrieval code for range/index_merge/groupby-min-max.
    Implementations of QUICK_*_SELECT classes.
63 64
*/

65
#ifdef USE_PRAGMA_IMPLEMENTATION
bk@work.mysql.com's avatar
bk@work.mysql.com committed
66 67 68 69 70 71 72 73 74 75 76 77
#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

78
/*
79
  Convert double value to #rows. Currently this does floor(), and we
80 81
  might consider using round() instead.
*/
82
#define double2rows(x) ((ha_rows)(x))
83

bk@work.mysql.com's avatar
bk@work.mysql.com committed
84 85 86 87
static int sel_cmp(Field *f,char *a,char *b,uint8 a_flag,uint8 b_flag);

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

88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175

/*
  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
176 177 178 179 180 181
class SEL_ARG :public Sql_alloc
{
public:
  uint8 min_flag,max_flag,maybe_flag;
  uint8 part;					// Which key part
  uint8 maybe_null;
182 183 184 185 186 187 188 189 190 191 192 193
  /* 
    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
194 195 196
  Field *field;
  char *min_value,*max_value;			// Pointer to range

197 198 199 200
  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
201 202 203 204 205 206 207 208 209
  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)
210 211
    :min_flag(0),elements(1),use_count(1),left(0),next_key_part(0),
    color(BLACK), type(type_arg)
212
  {}
bk@work.mysql.com's avatar
bk@work.mysql.com committed
213 214
  inline bool is_same(SEL_ARG *arg)
  {
215
    if (type != arg->type || part != arg->part)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313
      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;
  }
314
  void store_min(uint length,char **min_key,uint min_key_flag)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
315
  {
316 317 318
    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
319 320 321 322
    {
      if (maybe_null && *min_value)
      {
	**min_key=1;
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
323
	bzero(*min_key+1,length-1);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
324 325
      }
      else
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
326 327
	memcpy(*min_key,min_value,length);
      (*min_key)+= length;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
328
    }
329
  }
bk@work.mysql.com's avatar
bk@work.mysql.com committed
330 331 332
  void store(uint length,char **min_key,uint min_key_flag,
	     char **max_key, uint max_key_flag)
  {
333
    store_min(length, min_key, min_key_flag);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
334 335 336 337 338 339
    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
340
	bzero(*max_key+1,length-1);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
341 342
      }
      else
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
343 344
	memcpy(*max_key,max_value,length);
      (*max_key)+= length;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
345 346 347 348 349 350
    }
  }

  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
351
    key_tree->store(key[key_tree->part].store_length,
bk@work.mysql.com's avatar
bk@work.mysql.com committed
352 353 354 355 356 357 358 359 360 361 362 363
		    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
364
    key_tree->store(key[key_tree->part].store_length,
bk@work.mysql.com's avatar
bk@work.mysql.com committed
365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415
		    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();
416

417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433

  /*
    Check if this SEL_ARG object represents a single-point interval

    SYNOPSIS
      is_singlepoint()
    
    DESCRIPTION
      Check if this SEL_ARG object (not tree) represents a single-point
      interval, i.e. if it represents a "keypart = const" or 
      "keypart IS NULL".

    RETURN
      TRUE   This SEL_ARG object represents a singlepoint interval
      FALSE  Otherwise
  */

434 435
  bool is_singlepoint()
  {
436 437 438 439 440 441
    /* 
      Check for NEAR_MIN ("strictly less") and NO_MIN_RANGE (-inf < field) 
      flags, and the same for right edge.
    */
    if (min_flag || max_flag)
      return FALSE;
442
    byte *min_val= (byte *)min_value;
443
    byte *max_val= (byte *)max_value;
444 445 446 447 448 449 450 451 452 453 454 455 456

    if (maybe_null)
    {
      /* First byte is a NULL value indicator */
      if (*min_val != *max_val)
        return FALSE;

      if (*min_val)
        return TRUE; /* This "x IS NULL" */
      min_val++;
      max_val++;
    }
    return !field->key_cmp(min_val, max_val);
457
  }
bk@work.mysql.com's avatar
bk@work.mysql.com committed
458 459
};

460
class SEL_IMERGE;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
461

462

bk@work.mysql.com's avatar
bk@work.mysql.com committed
463 464 465
class SEL_TREE :public Sql_alloc
{
public:
466 467 468 469 470
  /*
    Starting an effort to document this field:
    (for some i, keys[i]->type == SEL_ARG::IMPOSSIBLE) => 
       (type == SEL_TREE::IMPOSSIBLE)
  */
bk@work.mysql.com's avatar
bk@work.mysql.com committed
471 472
  enum Type { IMPOSSIBLE, ALWAYS, MAYBE, KEY, KEY_SMALLER } type;
  SEL_TREE(enum Type type_arg) :type(type_arg) {}
473
  SEL_TREE() :type(KEY)
474
  {
475
    keys_map.clear_all();
476 477
    bzero((char*) keys,sizeof(keys));
  }
478 479 480 481 482 483
  /*
    Note: there may exist SEL_TREE objects with sel_tree->type=KEY and
    keys[i]=0 for all i. (SergeyP: it is not clear whether there is any
    merit in range analyzer functions (e.g. get_mm_parts) returning a
    pointer to such SEL_TREE instead of NULL)
  */
bk@work.mysql.com's avatar
bk@work.mysql.com committed
484
  SEL_ARG *keys[MAX_KEY];
485 486
  key_map keys_map;        /* bitmask of non-NULL elements in keys */

487 488
  /*
    Possible ways to read rows using index_merge. The list is non-empty only
489 490 491
    if type==KEY. Currently can be non empty only if keys_map.is_clear_all().
  */
  List<SEL_IMERGE> merges;
492

493 494
  /* 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 */
495
  uint    n_ror_scans;     /* number of set bits in ror_scans_map */
496 497 498 499

  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
500 501
};

502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529
class RANGE_OPT_PARAM
{
public:
  THD	*thd;   /* Current thread handle */
  TABLE *table; /* Table being analyzed */
  COND *cond;   /* Used inside get_mm_tree(). */
  table_map prev_tables;
  table_map read_tables;
  table_map current_table; /* Bit of the table being analyzed */

  /* Array of parts of all keys for which range analysis is performed */
  KEY_PART *key_parts;
  KEY_PART *key_parts_end;
  MEM_ROOT *mem_root; /* Memory that will be freed when range analysis completes */
  MEM_ROOT *old_root; /* Memory that will last until the query end */
  /*
    Number of indexes used in range analysis (In SEL_TREE::keys only first
    #keys elements are not empty)
  */
  uint keys;
  
  /* 
    If true, the index descriptions describe real indexes (and it is ok to
    call field->optimize_range(real_keynr[...], ...).
    Otherwise index description describes fake indexes.
  */
  bool using_real_indexes;
  
530 531
  bool remove_jump_scans;
  
532 533 534 535 536 537
  /*
    used_key_no -> table_key_no translation table. Only makes sense if
    using_real_indexes==TRUE
  */
  uint real_keynr[MAX_KEY];
};
bk@work.mysql.com's avatar
bk@work.mysql.com committed
538

539 540 541
class PARAM : public RANGE_OPT_PARAM
{
public:
542
  KEY_PART *key[MAX_KEY]; /* First key parts of keys used in the query */
543
  uint baseflag, max_key_part, range_count;
544

545

bk@work.mysql.com's avatar
bk@work.mysql.com committed
546 547
  char min_key[MAX_KEY_LENGTH+MAX_FIELD_WIDTH],
    max_key[MAX_KEY_LENGTH+MAX_FIELD_WIDTH];
548
  bool quick;				// Don't calulate possible keys
549

550
  uint fields_bitmap_size;
551
  MY_BITMAP needed_fields;    /* bitmask of fields needed by the query */
552
  MY_BITMAP tmp_covered_fields;
553 554 555

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

556 557
  uint *imerge_cost_buff;     /* buffer for index_merge cost estimates */
  uint imerge_cost_buff_size; /* size of the buffer */
558

559
  /* TRUE if last checked tree->key can be used for ROR-scan */
560
  bool is_ror_scan;
561 562
  /* Number of ranges in the last checked tree->key */
  uint n_ranges;
563
};
bk@work.mysql.com's avatar
bk@work.mysql.com committed
564

565 566 567 568 569
class TABLE_READ_PLAN;
  class TRP_RANGE;
  class TRP_ROR_INTERSECT;
  class TRP_ROR_UNION;
  class TRP_ROR_INDEX_MERGE;
570
  class TRP_GROUP_MIN_MAX;
571 572 573

struct st_ror_scan_info;

574
static SEL_TREE * get_mm_parts(RANGE_OPT_PARAM *param,COND *cond_func,Field *field,
bk@work.mysql.com's avatar
bk@work.mysql.com committed
575 576
			       Item_func::Functype type,Item *value,
			       Item_result cmp_type);
577
static SEL_ARG *get_mm_leaf(RANGE_OPT_PARAM *param,COND *cond_func,Field *field,
578
			    KEY_PART *key_part,
bk@work.mysql.com's avatar
bk@work.mysql.com committed
579
			    Item_func::Functype type,Item *value);
580
static SEL_TREE *get_mm_tree(RANGE_OPT_PARAM *param,COND *cond);
581 582

static bool is_key_scan_ror(PARAM *param, uint keynr, uint8 nparts);
583 584
static ha_rows check_quick_select(PARAM *param,uint index,SEL_ARG *key_tree, 
                                  bool update_tbl_stats);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
585 586 587 588
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);

589
QUICK_RANGE_SELECT *get_quick_select(PARAM *param,uint index,
590
                                     SEL_ARG *key_tree,
591
                                     MEM_ROOT *alloc = NULL);
592
static TRP_RANGE *get_key_scans_params(PARAM *param, SEL_TREE *tree,
593
                                       bool index_read_must_be_used,
594
                                       bool update_tbl_stats,
595 596 597 598 599 600
                                       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
601 602
TRP_ROR_INTERSECT *get_best_covering_ror_intersect(PARAM *param,
                                                   SEL_TREE *tree,
603 604 605 606
                                                   double read_time);
static
TABLE_READ_PLAN *get_best_disjunct_quick(PARAM *param, SEL_IMERGE *imerge,
                                         double read_time);
607 608
static
TRP_GROUP_MIN_MAX *get_best_group_min_max(PARAM *param, SEL_TREE *tree);
609
static int get_index_merge_params(PARAM *param, key_map& needed_reg,
610
                           SEL_IMERGE *imerge, double *read_time,
611
                           ha_rows* imerge_rows);
612
static double get_index_only_read_time(const PARAM* param, ha_rows records,
613 614
                                       int keynr);

bk@work.mysql.com's avatar
bk@work.mysql.com committed
615
#ifndef DBUG_OFF
616 617
static void print_sel_tree(PARAM *param, SEL_TREE *tree, key_map *tree_map,
                           const char *msg);
618 619
static void print_ror_scans_arr(TABLE *table, const char *msg,
                                struct st_ror_scan_info **start,
620 621 622
                                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
623
#endif
624

625 626
static SEL_TREE *tree_and(RANGE_OPT_PARAM *param,SEL_TREE *tree1,SEL_TREE *tree2);
static SEL_TREE *tree_or(RANGE_OPT_PARAM *param,SEL_TREE *tree1,SEL_TREE *tree2);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
627 628 629 630
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);
631
bool get_quick_keys(PARAM *param,QUICK_RANGE_SELECT *quick,KEY_PART *key,
bk@work.mysql.com's avatar
bk@work.mysql.com committed
632 633 634 635 636
			   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);
637
static bool null_part_in_key(KEY_PART *key_part, const char *key,
638
                             uint length);
639
bool sel_trees_can_be_ored(SEL_TREE *tree1, SEL_TREE *tree2, RANGE_OPT_PARAM* param);
640 641 642


/*
643
  SEL_IMERGE is a list of possible ways to do index merge, i.e. it is
644
  a condition in the following form:
645
   (t_1||t_2||...||t_N) && (next)
646

647
  where all t_i are SEL_TREEs, next is another SEL_IMERGE and no pair
648 649 650 651 652 653 654 655 656 657 658
  (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:
659
  SEL_TREE *trees_prealloced[PREALLOCED_TREES];
660 661 662 663 664 665 666 667 668 669 670
  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)
  {}
671 672 673
  int or_sel_tree(RANGE_OPT_PARAM *param, SEL_TREE *tree);
  int or_sel_tree_with_checks(RANGE_OPT_PARAM *param, SEL_TREE *new_tree);
  int or_sel_imerge_with_checks(RANGE_OPT_PARAM *param, SEL_IMERGE* imerge);
674 675 676
};


677
/*
678 679
  Add SEL_TREE to this index_merge without any checks,

680 681
  NOTES
    This function implements the following:
682 683 684 685 686 687 688
      (x_1||...||x_N) || t = (x_1||...||x_N||t), where x_i, t are SEL_TREEs

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

689
int SEL_IMERGE::or_sel_tree(RANGE_OPT_PARAM *param, SEL_TREE *tree)
690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713
{
  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.
714

715 716 717 718 719
  SYNOPSIS
    or_sel_tree_with_checks()
      param    PARAM from SQL_SELECT::test_quick_select
      new_tree SEL_TREE with type KEY or KEY_SMALLER.

720
  NOTES
721
    This does the following:
722 723
    (t_1||...||t_k)||new_tree =
     either
724 725 726
       = (t_1||...||t_k||new_tree)
     or
       = (t_1||....||(t_j|| new_tree)||...||t_k),
727

728
     where t_i, y are SEL_TREEs.
729 730
    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
731 732
    read may depend on the order of conditions in WHERE part of the query.

733
  RETURN
734
    0  OK
735
    1  One of the trees was combined with new_tree to SEL_TREE::ALWAYS,
736 737 738 739
       and (*this) should be discarded.
   -1  An error occurred.
*/

740
int SEL_IMERGE::or_sel_tree_with_checks(RANGE_OPT_PARAM *param, SEL_TREE *new_tree)
741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758
{
  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;
    }
  }

759
  /* New tree cannot be combined with any of existing trees. */
760 761 762 763 764 765 766 767 768
  return or_sel_tree(param, new_tree);
}


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

  RETURN
    0 - OK
769
    1 - One of conditions in result is always TRUE and this SEL_IMERGE
770 771 772 773
        should be discarded.
   -1 - An error occurred
*/

774
int SEL_IMERGE::or_sel_imerge_with_checks(RANGE_OPT_PARAM *param, SEL_IMERGE* imerge)
775 776 777 778 779 780 781 782 783 784 785 786
{
  for (SEL_TREE** tree= imerge->trees;
       tree != imerge->trees_next;
       tree++)
  {
    if (or_sel_tree_with_checks(param, *tree))
      return 1;
  }
  return 0;
}


787
/*
788
  Perform AND operation on two index_merge lists and store result in *im1.
789 790 791 792 793 794 795 796 797 798 799
*/

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.

800
  NOTES
801 802 803
    The following conversion is implemented:
     (a_1 &&...&& a_N)||(b_1 &&...&& b_K) = AND_i,j(a_i || b_j) =>
      => (a_1||b_1).
804 805

    i.e. all conjuncts except the first one are currently dropped.
806 807
    This is done to avoid producing N*K ways to do index_merge.

monty@mysql.com's avatar
monty@mysql.com committed
808
    If (a_1||b_1) produce a condition that is always TRUE, NULL is returned
809
    and index_merge is discarded (while it is actually possible to try
810
    harder).
811

812 813
    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.
814 815

  RETURN
816
    0     OK, result is stored in *im1
817 818 819
    other Error, both passed lists are unusable
*/

820
int imerge_list_or_list(RANGE_OPT_PARAM *param,
821 822 823 824 825 826
                        List<SEL_IMERGE> *im1,
                        List<SEL_IMERGE> *im2)
{
  SEL_IMERGE *imerge= im1->head();
  im1->empty();
  im1->push_back(imerge);
827

828 829 830 831 832 833 834 835
  return imerge->or_sel_imerge_with_checks(param, im2->head());
}


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

  RETURN
836
    0     OK, result is stored in *im1.
837 838 839
    other Error
*/

840
int imerge_list_or_tree(RANGE_OPT_PARAM *param,
841 842 843 844 845
                        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
846
  while ((imerge= it++))
847 848 849 850 851 852
  {
    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
853 854

/***************************************************************************
855
** Basic functions for SQL_SELECT and QUICK_RANGE_SELECT
bk@work.mysql.com's avatar
bk@work.mysql.com committed
856 857 858 859 860 861 862 863 864
***************************************************************************/

	/* 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
865 866 867
			table_map read_tables, COND *conds,
                        bool allow_null_cond,
                        int *error)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
868 869 870 871 872
{
  SQL_SELECT *select;
  DBUG_ENTER("make_select");

  *error=0;
873 874

  if (!conds && !allow_null_cond)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
875 876 877
    DBUG_RETURN(0);
  if (!(select= new SQL_SELECT))
  {
878 879
    *error= 1;			// out of memory
    DBUG_RETURN(0);		/* purecov: inspected */
bk@work.mysql.com's avatar
bk@work.mysql.com committed
880 881 882 883 884 885
  }
  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
886
  if (head->sort.io_cache)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
887
  {
igor@hundin.mysql.fi's avatar
igor@hundin.mysql.fi committed
888
    select->file= *head->sort.io_cache;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
889 890
    select->records=(ha_rows) (select->file.end_of_file/
			       head->file->ref_length);
igor@hundin.mysql.fi's avatar
igor@hundin.mysql.fi committed
891 892
    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
893 894 895 896 897 898 899
  }
  DBUG_RETURN(select);
}


SQL_SELECT::SQL_SELECT() :quick(0),cond(0),free_cond(0)
{
serg@serg.mylan's avatar
serg@serg.mylan committed
900
  quick_keys.clear_all(); needed_reg.clear_all();
bk@work.mysql.com's avatar
bk@work.mysql.com committed
901 902 903 904
  my_b_clear(&file);
}


905
void SQL_SELECT::cleanup()
bk@work.mysql.com's avatar
bk@work.mysql.com committed
906 907
{
  delete quick;
908
  quick= 0;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
909
  if (free_cond)
910 911
  {
    free_cond=0;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
912
    delete cond;
913
    cond= 0;
914
  }
bk@work.mysql.com's avatar
bk@work.mysql.com committed
915 916 917
  close_cached_file(&file);
}

918 919 920 921 922 923

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

924
#undef index					// Fix for Unixware 7
bk@work.mysql.com's avatar
bk@work.mysql.com committed
925

sergefp@mysql.com's avatar
sergefp@mysql.com committed
926 927 928 929 930
QUICK_SELECT_I::QUICK_SELECT_I()
  :max_used_key_length(0),
   used_key_parts(0)
{}

931
QUICK_RANGE_SELECT::QUICK_RANGE_SELECT(THD *thd, TABLE *table, uint key_nr,
sergefp@mysql.com's avatar
sergefp@mysql.com committed
932
                                       bool no_alloc, MEM_ROOT *parent_alloc)
933
  :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
934
{
935 936 937 938
  my_bitmap_map *bitmap;
  DBUG_ENTER("QUICK_RANGE_SELECT::QUICK_RANGE_SELECT");

  in_ror_merged_scan= 0;
monty@mysql.com's avatar
monty@mysql.com committed
939
  sorted= 0;
sergefp@mysql.com's avatar
sergefp@mysql.com committed
940 941
  index= key_nr;
  head=  table;
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
942
  key_part_info= head->key_info[index].key_part;
943
  my_init_dynamic_array(&ranges, sizeof(QUICK_RANGE*), 16, 16);
sergefp@mysql.com's avatar
sergefp@mysql.com committed
944

sergefp@mysql.com's avatar
sergefp@mysql.com committed
945
  /* 'thd' is not accessible in QUICK_RANGE_SELECT::reset(). */
ingo@mysql.com's avatar
ingo@mysql.com committed
946 947 948 949 950 951
  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
952
  if (!no_alloc && !parent_alloc)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
953
  {
954 955
    // Allocates everything through the internal memroot
    init_sql_alloc(&alloc, thd->variables.range_alloc_block_size, 0);
956
    thd->mem_root= &alloc;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
957 958 959
  }
  else
    bzero((char*) &alloc,sizeof(alloc));
960 961
  file= head->file;
  record= head->record[0];
962 963 964 965 966 967 968 969 970 971 972 973 974
  save_read_set= head->read_set;
  save_write_set= head->write_set;

  /* Allocate a bitmap for used columns */
  if (!(bitmap= (my_bitmap_map*) my_malloc(head->s->column_bitmap_size,
                                           MYF(MY_WME))))
  {
    column_bitmap.bitmap= 0;
    error= 1;
  }
  else
    bitmap_init(&column_bitmap, bitmap, head->s->fields, FALSE);
  DBUG_VOID_RETURN;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
975 976
}

monty@mysql.com's avatar
monty@mysql.com committed
977

978 979
int QUICK_RANGE_SELECT::init()
{
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
980
  DBUG_ENTER("QUICK_RANGE_SELECT::init");
ingo@mysql.com's avatar
ingo@mysql.com committed
981

982 983
  if (file->inited != handler::NONE)
    file->ha_index_or_rnd_end();
tomas@poseidon.ndb.mysql.com's avatar
Merge  
tomas@poseidon.ndb.mysql.com committed
984
  DBUG_RETURN(error= file->ha_index_init(index, 1));
monty@mysql.com's avatar
monty@mysql.com committed
985 986 987 988 989 990
}


void QUICK_RANGE_SELECT::range_end()
{
  if (file->inited != handler::NONE)
991
    file->ha_index_or_rnd_end();
bk@work.mysql.com's avatar
bk@work.mysql.com committed
992 993
}

monty@mysql.com's avatar
monty@mysql.com committed
994

995
QUICK_RANGE_SELECT::~QUICK_RANGE_SELECT()
996
{
997
  DBUG_ENTER("QUICK_RANGE_SELECT::~QUICK_RANGE_SELECT");
998 999
  if (!dont_free)
  {
1000 1001
    /* file is NULL for CPK scan on covering ROR-intersection */
    if (file) 
1002
    {
1003 1004 1005 1006 1007
      range_end();
      if (free_file)
      {
        DBUG_PRINT("info", ("Freeing separate handler %p (free=%d)", file,
                            free_file));
1008
        file->ha_external_lock(current_thd, F_UNLCK);
1009
        file->close();
1010
        delete file;
1011
      }
1012 1013 1014 1015
      else
      {
        file->extra(HA_EXTRA_NO_KEYREAD);
      }
1016
    }
1017
    delete_dynamic(&ranges); /* ranges are allocated in alloc */
1018
    free_root(&alloc,MYF(0));
1019
    my_free((char*) column_bitmap.bitmap, MYF(MY_ALLOW_ZERO_PTR));
1020
  }
1021 1022 1023
  head->column_bitmaps_set(save_read_set, save_write_set);
  x_free(multi_range);
  x_free(multi_range_buff);
1024
  DBUG_VOID_RETURN;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1025 1026
}

1027

1028
QUICK_INDEX_MERGE_SELECT::QUICK_INDEX_MERGE_SELECT(THD *thd_param,
1029
                                                   TABLE *table)
sergefp@mysql.com's avatar
sergefp@mysql.com committed
1030
  :pk_quick_select(NULL), thd(thd_param)
1031
{
1032
  DBUG_ENTER("QUICK_INDEX_MERGE_SELECT::QUICK_INDEX_MERGE_SELECT");
1033 1034
  index= MAX_KEY;
  head= table;
1035
  bzero(&read_record, sizeof(read_record));
1036
  init_sql_alloc(&alloc, thd->variables.range_alloc_block_size, 0);
1037
  DBUG_VOID_RETURN;
1038 1039 1040 1041
}

int QUICK_INDEX_MERGE_SELECT::init()
{
sergefp@mysql.com's avatar
sergefp@mysql.com committed
1042 1043
  DBUG_ENTER("QUICK_INDEX_MERGE_SELECT::init");
  DBUG_RETURN(0);
1044 1045
}

1046
int QUICK_INDEX_MERGE_SELECT::reset()
1047
{
1048
  DBUG_ENTER("QUICK_INDEX_MERGE_SELECT::reset");
sergefp@mysql.com's avatar
sergefp@mysql.com committed
1049
  DBUG_RETURN(read_keys_and_merge());
1050 1051
}

1052
bool
1053 1054
QUICK_INDEX_MERGE_SELECT::push_quick_back(QUICK_RANGE_SELECT *quick_sel_range)
{
1055 1056
  /*
    Save quick_select that does scan on clustered primary key as it will be
1057
    processed separately.
1058
  */
1059
  if (head->file->primary_key_is_clustered() &&
1060
      quick_sel_range->index == head->s->primary_key)
1061 1062 1063 1064
    pk_quick_select= quick_sel_range;
  else
    return quick_selects.push_back(quick_sel_range);
  return 0;
1065 1066 1067 1068
}

QUICK_INDEX_MERGE_SELECT::~QUICK_INDEX_MERGE_SELECT()
{
sergefp@mysql.com's avatar
sergefp@mysql.com committed
1069 1070
  List_iterator_fast<QUICK_RANGE_SELECT> quick_it(quick_selects);
  QUICK_RANGE_SELECT* quick;
1071
  DBUG_ENTER("QUICK_INDEX_MERGE_SELECT::~QUICK_INDEX_MERGE_SELECT");
sergefp@mysql.com's avatar
sergefp@mysql.com committed
1072 1073 1074
  quick_it.rewind();
  while ((quick= quick_it++))
    quick->file= NULL;
1075
  quick_selects.delete_elements();
1076
  delete pk_quick_select;
1077
  free_root(&alloc,MYF(0));
1078
  DBUG_VOID_RETURN;
1079 1080
}

1081 1082 1083 1084 1085

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
1086
  : cpk_quick(NULL), thd(thd_param), need_to_fetch_row(retrieve_full_rows),
1087
    scans_inited(FALSE)
1088 1089
{
  index= MAX_KEY;
1090
  head= table;
1091 1092
  record= head->record[0];
  if (!parent_alloc)
1093
    init_sql_alloc(&alloc, thd->variables.range_alloc_block_size, 0);
1094 1095
  else
    bzero(&alloc, sizeof(MEM_ROOT));
1096
  last_rowid= (byte*)alloc_root(parent_alloc? parent_alloc : &alloc,
1097 1098 1099
                                head->file->ref_length);
}

1100

1101
/*
1102 1103 1104
  Do post-constructor initialization.
  SYNOPSIS
    QUICK_ROR_INTERSECT_SELECT::init()
1105

1106 1107 1108 1109 1110
  RETURN
    0      OK
    other  Error code
*/

1111 1112
int QUICK_ROR_INTERSECT_SELECT::init()
{
sergefp@mysql.com's avatar
sergefp@mysql.com committed
1113 1114 1115
  DBUG_ENTER("QUICK_ROR_INTERSECT_SELECT::init");
 /* Check if last_rowid was successfully allocated in ctor */
  DBUG_RETURN(!last_rowid);
1116 1117 1118 1119
}


/*
1120 1121 1122 1123
  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
1124
      reuse_handler If TRUE, use head->file, otherwise create a separate
1125 1126 1127 1128
                    handler object

  NOTES
    This function creates and prepares for subsequent use a separate handler
1129
    object if it can't reuse head->file. The reason for this is that during
1130 1131 1132
    ROR-merge several key scans are performed simultaneously, and a single
    handler is only capable of preserving context of a single key scan.

1133
    In ROR-merge the quick select doing merge does full records retrieval,
1134
    merged quick selects read only keys.
1135 1136

  RETURN
1137 1138 1139 1140
    0  ROR child scan initialized, ok to use.
    1  error
*/

1141
int QUICK_RANGE_SELECT::init_ror_merged_scan(bool reuse_handler)
1142
{
1143
  handler *save_file= file, *org_file;
1144
  THD *thd;
1145
  MY_BITMAP *bitmap;
1146
  DBUG_ENTER("QUICK_RANGE_SELECT::init_ror_merged_scan");
1147

1148
  in_ror_merged_scan= 1;
1149 1150
  if (reuse_handler)
  {
1151 1152
    DBUG_PRINT("info", ("Reusing handler 0x%lx", (long) file));
    if (init() || reset())
1153 1154 1155
    {
      DBUG_RETURN(1);
    }
1156 1157
    head->column_bitmaps_set(&column_bitmap, &column_bitmap);
    goto end;
1158 1159 1160 1161 1162 1163 1164 1165
  }

  /* Create a separate handler object for this quick select */
  if (free_file)
  {
    /* already have own 'handler' object. */
    DBUG_RETURN(0);
  }
1166

1167
  thd= head->in_use;
1168
  if (!(file= head->file->clone(thd->mem_root)))
1169
  {
1170
    /* Caller will free the memory */
1171 1172
    goto failure;
  }
1173 1174 1175

  head->column_bitmaps_set(&column_bitmap, &column_bitmap);

1176
  if (file->ha_external_lock(thd, F_RDLCK))
1177
    goto failure;
1178

1179
  if (init() || reset())
1180
  {
1181
    file->ha_external_lock(thd, F_UNLCK);
1182 1183 1184
    file->close();
    goto failure;
  }
monty@mysql.com's avatar
monty@mysql.com committed
1185
  free_file= TRUE;
1186
  last_rowid= file->ref;
1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203

end:
  /*
    We are only going to read key fields and call position() on 'file'
    The following sets head->tmp_set to only use this key and then updates
    head->read_set and head->write_set to use this bitmap.
    The now bitmap is stored in 'column_bitmap' which is used in ::get_next()
  */
  org_file= head->file;
  head->file= file;
  /* We don't have to set 'head->keyread' here as the 'file' is unique */
  head->mark_columns_used_by_index(index);
  head->prepare_for_position();
  head->file= org_file;
  bitmap_copy(&column_bitmap, head->read_set);
  head->column_bitmaps_set(&column_bitmap, &column_bitmap);

1204 1205 1206
  DBUG_RETURN(0);

failure:
1207 1208
  head->column_bitmaps_set(save_read_set, save_write_set);
  delete file;
1209 1210 1211 1212
  file= save_file;
  DBUG_RETURN(1);
}

1213 1214 1215 1216 1217

/*
  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
1218
      reuse_handler If TRUE, use head->file, otherwise create separate
1219
                    handler object.
1220
  RETURN
1221 1222 1223 1224
    0     OK
    other error code
*/
int QUICK_ROR_INTERSECT_SELECT::init_ror_merged_scan(bool reuse_handler)
1225 1226 1227
{
  List_iterator_fast<QUICK_RANGE_SELECT> quick_it(quick_selects);
  QUICK_RANGE_SELECT* quick;
1228
  DBUG_ENTER("QUICK_ROR_INTERSECT_SELECT::init_ror_merged_scan");
1229 1230

  /* Initialize all merged "children" quick selects */
sergefp@mysql.com's avatar
sergefp@mysql.com committed
1231
  DBUG_ASSERT(!need_to_fetch_row || reuse_handler);
1232 1233 1234
  if (!need_to_fetch_row && reuse_handler)
  {
    quick= quick_it++;
1235
    /*
1236
      There is no use of this->file. Use it for the first of merged range
1237 1238
      selects.
    */
monty@mysql.com's avatar
monty@mysql.com committed
1239
    if (quick->init_ror_merged_scan(TRUE))
1240 1241 1242
      DBUG_RETURN(1);
    quick->file->extra(HA_EXTRA_KEYREAD_PRESERVE_FIELDS);
  }
monty@mishka.local's avatar
monty@mishka.local committed
1243
  while ((quick= quick_it++))
1244
  {
monty@mysql.com's avatar
monty@mysql.com committed
1245
    if (quick->init_ror_merged_scan(FALSE))
1246 1247
      DBUG_RETURN(1);
    quick->file->extra(HA_EXTRA_KEYREAD_PRESERVE_FIELDS);
1248
    /* All merged scans share the same record buffer in intersection. */
1249 1250 1251
    quick->record= head->record[0];
  }

monty@mysql.com's avatar
monty@mysql.com committed
1252
  if (need_to_fetch_row && head->file->ha_rnd_init(1))
1253 1254 1255 1256 1257 1258 1259
  {
    DBUG_PRINT("error", ("ROR index_merge rnd_init call failed"));
    DBUG_RETURN(1);
  }
  DBUG_RETURN(0);
}

1260

1261
/*
1262 1263 1264 1265 1266 1267 1268 1269
  Initialize quick select for row retrieval.
  SYNOPSIS
    reset()
  RETURN
    0      OK
    other  Error code
*/

1270 1271 1272
int QUICK_ROR_INTERSECT_SELECT::reset()
{
  DBUG_ENTER("QUICK_ROR_INTERSECT_SELECT::reset");
sergefp@mysql.com's avatar
sergefp@mysql.com committed
1273 1274
  if (!scans_inited && init_ror_merged_scan(TRUE))
    DBUG_RETURN(1);
1275
  scans_inited= TRUE;
sergefp@mysql.com's avatar
sergefp@mysql.com committed
1276 1277 1278 1279 1280
  List_iterator_fast<QUICK_RANGE_SELECT> it(quick_selects);
  QUICK_RANGE_SELECT *quick;
  while ((quick= it++))
    quick->reset();
  DBUG_RETURN(0);
1281 1282
}

1283 1284 1285

/*
  Add a merged quick select to this ROR-intersection quick select.
1286

1287 1288 1289 1290 1291 1292
  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.
1293

1294
  RETURN
1295
    FALSE OK
monty@mysql.com's avatar
monty@mysql.com committed
1296
    TRUE  Out of memory.
1297 1298
*/

1299
bool
1300 1301
QUICK_ROR_INTERSECT_SELECT::push_quick_back(QUICK_RANGE_SELECT *quick)
{
1302
  return quick_selects.push_back(quick);
1303 1304 1305
}

QUICK_ROR_INTERSECT_SELECT::~QUICK_ROR_INTERSECT_SELECT()
1306
{
1307
  DBUG_ENTER("QUICK_ROR_INTERSECT_SELECT::~QUICK_ROR_INTERSECT_SELECT");
1308
  quick_selects.delete_elements();
1309 1310
  delete cpk_quick;
  free_root(&alloc,MYF(0));
monty@mysql.com's avatar
monty@mysql.com committed
1311 1312
  if (need_to_fetch_row && head->file->inited != handler::NONE)
    head->file->ha_rnd_end();
1313 1314 1315
  DBUG_VOID_RETURN;
}

monty@mysql.com's avatar
monty@mysql.com committed
1316

1317 1318
QUICK_ROR_UNION_SELECT::QUICK_ROR_UNION_SELECT(THD *thd_param,
                                               TABLE *table)
1319
  : thd(thd_param), scans_inited(FALSE)
1320 1321 1322 1323 1324 1325
{
  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
1326
  thd_param->mem_root= &alloc;
1327 1328
}

1329 1330 1331 1332 1333

/*
  Do post-constructor initialization.
  SYNOPSIS
    QUICK_ROR_UNION_SELECT::init()
1334

1335 1336 1337 1338 1339
  RETURN
    0      OK
    other  Error code
*/

1340 1341
int QUICK_ROR_UNION_SELECT::init()
{
sergefp@mysql.com's avatar
sergefp@mysql.com committed
1342
  DBUG_ENTER("QUICK_ROR_UNION_SELECT::init");
1343
  if (init_queue(&queue, quick_selects.elements, 0,
monty@mysql.com's avatar
monty@mysql.com committed
1344
                 FALSE , QUICK_ROR_UNION_SELECT::queue_cmp,
1345 1346 1347
                 (void*) this))
  {
    bzero(&queue, sizeof(QUEUE));
sergefp@mysql.com's avatar
sergefp@mysql.com committed
1348
    DBUG_RETURN(1);
1349
  }
1350

1351
  if (!(cur_rowid= (byte*)alloc_root(&alloc, 2*head->file->ref_length)))
sergefp@mysql.com's avatar
sergefp@mysql.com committed
1352
    DBUG_RETURN(1);
1353
  prev_rowid= cur_rowid + head->file->ref_length;
sergefp@mysql.com's avatar
sergefp@mysql.com committed
1354
  DBUG_RETURN(0);
1355 1356
}

1357

1358
/*
1359
  Comparison function to be used QUICK_ROR_UNION_SELECT::queue priority
1360 1361
  queue.

1362 1363 1364 1365 1366 1367
  SYNPOSIS
    QUICK_ROR_UNION_SELECT::queue_cmp()
      arg   Pointer to QUICK_ROR_UNION_SELECT
      val1  First merged select
      val2  Second merged select
*/
1368

1369 1370
int QUICK_ROR_UNION_SELECT::queue_cmp(void *arg, byte *val1, byte *val2)
{
1371
  QUICK_ROR_UNION_SELECT *self= (QUICK_ROR_UNION_SELECT*)arg;
1372 1373 1374 1375
  return self->head->file->cmp_ref(((QUICK_SELECT_I*)val1)->last_rowid,
                                   ((QUICK_SELECT_I*)val2)->last_rowid);
}

1376

1377
/*
1378 1379 1380
  Initialize quick select for row retrieval.
  SYNOPSIS
    reset()
1381

1382 1383 1384 1385 1386
  RETURN
    0      OK
    other  Error code
*/

1387 1388 1389 1390 1391
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
1392
  have_prev_rowid= FALSE;
sergefp@mysql.com's avatar
sergefp@mysql.com committed
1393 1394 1395 1396 1397 1398 1399 1400 1401
  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);
    }
1402
    scans_inited= TRUE;
sergefp@mysql.com's avatar
sergefp@mysql.com committed
1403 1404
  }
  queue_remove_all(&queue);
1405 1406
  /*
    Initialize scans for merged quick selects and put all merged quick
1407 1408 1409 1410 1411
    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
1412
    if (quick->reset())
1413
      DBUG_RETURN(1);
1414 1415 1416 1417
    if ((error= quick->get_next()))
    {
      if (error == HA_ERR_END_OF_FILE)
        continue;
monty@mysql.com's avatar
monty@mysql.com committed
1418
      DBUG_RETURN(error);
1419 1420 1421 1422 1423
    }
    quick->save_last_pos();
    queue_insert(&queue, (byte*)quick);
  }

monty@mysql.com's avatar
monty@mysql.com committed
1424
  if (head->file->ha_rnd_init(1))
1425 1426 1427 1428 1429 1430 1431 1432 1433
  {
    DBUG_PRINT("error", ("ROR index_merge rnd_init call failed"));
    DBUG_RETURN(1);
  }

  DBUG_RETURN(0);
}


1434
bool
1435 1436 1437 1438 1439 1440 1441 1442 1443
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);
1444
  quick_selects.delete_elements();
1445 1446
  if (head->file->inited != handler::NONE)
    head->file->ha_rnd_end();
1447 1448
  free_root(&alloc,MYF(0));
  DBUG_VOID_RETURN;
1449 1450
}

1451

bk@work.mysql.com's avatar
bk@work.mysql.com committed
1452 1453 1454 1455 1456 1457 1458 1459 1460 1461 1462 1463 1464 1465 1466 1467 1468 1469 1470 1471 1472 1473 1474 1475 1476 1477 1478 1479 1480 1481 1482 1483 1484 1485 1486 1487 1488 1489 1490 1491 1492 1493 1494 1495 1496 1497 1498 1499 1500 1501 1502 1503 1504
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)
  {
1505 1506
    if (!(tmp= new SEL_ARG(type)))
      return 0;					// out of memory
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1507 1508 1509 1510 1511 1512
    tmp->prev= *next_arg;			// Link into next/prev chain
    (*next_arg)->next=tmp;
    (*next_arg)= tmp;
  }
  else
  {
1513 1514 1515
    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
1516 1517 1518 1519 1520 1521 1522 1523 1524 1525
    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)
1526 1527
      if (!(tmp->right= right->clone(tmp,next_arg)))
	return 0;				// OOM
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1528 1529
  }
  increment_use_count(1);
1530
  tmp->color= color;
1531
  tmp->elements= this->elements;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1532 1533 1534 1535 1536 1537 1538 1539 1540 1541 1542 1543 1544 1545 1546 1547 1548 1549 1550 1551 1552 1553 1554
  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;
}

1555

bk@work.mysql.com's avatar
bk@work.mysql.com committed
1556 1557 1558
/*
  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
1559
*/
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1560 1561 1562 1563 1564 1565 1566 1567 1568 1569 1570 1571 1572 1573 1574 1575 1576 1577 1578 1579 1580 1581 1582

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
1583
    a++; b++;					// Skip NULL marker
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1584 1585 1586 1587 1588 1589 1590 1591 1592 1593 1594 1595 1596 1597 1598 1599 1600 1601 1602 1603 1604 1605 1606 1607
  }
  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;
1608
  root= clone((SEL_ARG *) 0, &next_arg);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1609 1610
  next_arg->next=0;				// Fix last link
  tmp_link.next->prev=0;			// Fix first link
1611 1612
  if (root)					// If not OOM
    root->use_count= 0;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1613 1614 1615
  return root;
}

1616

1617
/*
1618
  Find the best index to retrieve first N records in given order
1619 1620 1621 1622 1623 1624 1625 1626

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

  DESCRIPTION
1627 1628 1629 1630
    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
1631
    Run through all table indexes and find the shortest index that allows
1632 1633
    records to be retrieved in given order. We look for the shortest index
    as we will have fewer index pages to read with it.
1634 1635 1636 1637 1638 1639 1640 1641 1642 1643 1644 1645 1646 1647 1648 1649 1650 1651 1652 1653 1654 1655

    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
1656
  for (idx= 0; idx < table->s->keys; idx++)
1657 1658 1659 1660 1661 1662 1663 1664 1665 1666 1667 1668 1669 1670 1671 1672 1673 1674 1675 1676 1677 1678 1679 1680 1681 1682 1683 1684 1685 1686 1687 1688 1689 1690 1691 1692 1693 1694 1695 1696 1697 1698 1699 1700 1701 1702 1703 1704 1705
  {
    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
1706
/*
1707
  Table rows retrieval plan. Range optimizer creates QUICK_SELECT_I-derived
1708 1709 1710 1711 1712
  objects from table read plans.
*/
class TABLE_READ_PLAN
{
public:
1713 1714
  /*
    Plan read cost, with or without cost of full row retrieval, depending
1715 1716
    on plan creation parameters.
  */
1717
  double read_cost;
1718
  ha_rows records; /* estimate of #rows to be examined */
serg@serg.mylan's avatar
serg@serg.mylan committed
1719

1720 1721
  /*
    If TRUE, the scan returns rows in rowid order. This is used only for
1722 1723
    scans that can be both ROR and non-ROR.
  */
1724
  bool is_ror;
1725

1726 1727 1728 1729 1730
  /*
    Create quick select for this plan.
    SYNOPSIS
     make_quick()
       param               Parameter from test_quick_select
monty@mysql.com's avatar
monty@mysql.com committed
1731
       retrieve_full_rows  If TRUE, created quick select will do full record
1732 1733
                           retrieval.
       parent_alloc        Memory pool to use, if any.
1734

1735 1736
    NOTES
      retrieve_full_rows is ignored by some implementations.
1737 1738

    RETURN
1739 1740 1741
      created quick select
      NULL on any error.
  */
1742 1743 1744 1745
  virtual QUICK_SELECT_I *make_quick(PARAM *param,
                                     bool retrieve_full_rows,
                                     MEM_ROOT *parent_alloc=NULL) = 0;

1746
  /* Table read plans are allocated on MEM_ROOT and are never deleted */
1747 1748
  static void *operator new(size_t size, MEM_ROOT *mem_root)
  { return (void*) alloc_root(mem_root, (uint) size); }
1749
  static void operator delete(void *ptr,size_t size) { TRASH(ptr, size); }
1750
  static void operator delete(void *ptr, MEM_ROOT *mem_root) { /* Never called */ }
1751 1752
  virtual ~TABLE_READ_PLAN() {}               /* Remove gcc warning */

1753 1754 1755 1756 1757 1758 1759
};

class TRP_ROR_INTERSECT;
class TRP_ROR_UNION;
class TRP_INDEX_MERGE;


1760
/*
1761
  Plan for a QUICK_RANGE_SELECT scan.
1762 1763 1764
  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
1765
*/
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1766

1767
class TRP_RANGE : public TABLE_READ_PLAN
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1768
{
1769
public:
1770 1771
  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
1772

1773
  TRP_RANGE(SEL_ARG *key_arg, uint idx_arg)
1774 1775
   : key(key_arg), key_idx(idx_arg)
  {}
1776
  virtual ~TRP_RANGE() {}                     /* Remove gcc warning */
1777

1778 1779 1780 1781 1782 1783 1784 1785 1786 1787 1788 1789 1790
  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
1791 1792


1793 1794
/* Plan for QUICK_ROR_INTERSECT_SELECT scan. */

1795 1796 1797
class TRP_ROR_INTERSECT : public TABLE_READ_PLAN
{
public:
1798 1799
  TRP_ROR_INTERSECT() {}                      /* Remove gcc warning */
  virtual ~TRP_ROR_INTERSECT() {}             /* Remove gcc warning */
1800
  QUICK_SELECT_I *make_quick(PARAM *param, bool retrieve_full_rows,
1801
                             MEM_ROOT *parent_alloc);
1802

1803
  /* Array of pointers to ROR range scans used in this intersection */
1804
  struct st_ror_scan_info **first_scan;
1805 1806
  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
1807
  bool is_covering; /* TRUE if no row retrieval phase is necessary */
1808
  double index_scan_costs; /* SUM(cost(index_scan)) */
1809 1810
};

1811

1812
/*
1813 1814
  Plan for QUICK_ROR_UNION_SELECT scan.
  QUICK_ROR_UNION_SELECT always retrieves full rows, so retrieve_full_rows
1815
  is ignored by make_quick.
1816
*/
1817

1818 1819 1820
class TRP_ROR_UNION : public TABLE_READ_PLAN
{
public:
1821 1822
  TRP_ROR_UNION() {}                          /* Remove gcc warning */
  virtual ~TRP_ROR_UNION() {}                 /* Remove gcc warning */
1823
  QUICK_SELECT_I *make_quick(PARAM *param, bool retrieve_full_rows,
1824
                             MEM_ROOT *parent_alloc);
1825 1826
  TABLE_READ_PLAN **first_ror; /* array of ptrs to plans for merged scans */
  TABLE_READ_PLAN **last_ror;  /* end of the above array */
1827 1828
};

1829 1830 1831 1832

/*
  Plan for QUICK_INDEX_MERGE_SELECT scan.
  QUICK_ROR_INTERSECT_SELECT always retrieves full rows, so retrieve_full_rows
1833
  is ignored by make_quick.
1834 1835
*/

1836 1837 1838
class TRP_INDEX_MERGE : public TABLE_READ_PLAN
{
public:
1839 1840
  TRP_INDEX_MERGE() {}                        /* Remove gcc warning */
  virtual ~TRP_INDEX_MERGE() {}               /* Remove gcc warning */
1841
  QUICK_SELECT_I *make_quick(PARAM *param, bool retrieve_full_rows,
1842
                             MEM_ROOT *parent_alloc);
1843 1844
  TRP_RANGE **range_scans; /* array of ptrs to plans of merged scans */
  TRP_RANGE **range_scans_end; /* end of the array */
1845 1846 1847
};


1848 1849 1850 1851 1852 1853 1854 1855 1856 1857 1858 1859 1860 1861 1862 1863 1864 1865 1866 1867 1868 1869 1870
/*
  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:
1871 1872 1873 1874
  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,
1875 1876
                    uint index_arg, uint key_infix_len_arg,
                    byte *key_infix_arg,
1877 1878 1879 1880 1881 1882 1883 1884 1885
                    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)
1886 1887 1888 1889
    {
      if (key_infix_len)
        memcpy(this->key_infix, key_infix_arg, key_infix_len);
    }
1890
  virtual ~TRP_GROUP_MIN_MAX() {}             /* Remove gcc warning */
1891 1892 1893 1894 1895 1896

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


1897
/*
1898
  Fill param->needed_fields with bitmap of fields used in the query.
1899
  SYNOPSIS
1900 1901
    fill_used_fields_bitmap()
      param Parameter from test_quick_select function.
1902

1903 1904 1905
  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).
1906 1907 1908
  RETURN
    0  Ok
    1  Out of memory.
1909 1910 1911 1912 1913
*/

static int fill_used_fields_bitmap(PARAM *param)
{
  TABLE *table= param->table;
1914
  my_bitmap_map *tmp;
1915
  uint pk;
1916
  param->tmp_covered_fields.bitmap= 0;
1917 1918 1919 1920
  param->fields_bitmap_size= table->s->column_bitmap_size;
  if (!(tmp= (my_bitmap_map*) alloc_root(param->mem_root,
                                  param->fields_bitmap_size)) ||
      bitmap_init(&param->needed_fields, tmp, table->s->fields, FALSE))
1921
    return 1;
1922

1923 1924
  bitmap_copy(&param->needed_fields, table->read_set);
  bitmap_union(&param->needed_fields, table->write_set);
1925

1926
  pk= param->table->s->primary_key;
1927
  if (pk != MAX_KEY && param->table->file->primary_key_is_clustered())
1928
  {
1929
    /* The table uses clustered PK and it is not internally generated */
1930
    KEY_PART_INFO *key_part= param->table->key_info[pk].key_part;
1931
    KEY_PART_INFO *key_part_end= key_part +
1932
                                 param->table->key_info[pk].key_parts;
1933
    for (;key_part != key_part_end; ++key_part)
1934
      bitmap_clear_bit(&param->needed_fields, key_part->fieldnr-1);
1935 1936 1937 1938 1939
  }
  return 0;
}


serg@serg.mylan's avatar
serg@serg.mylan committed
1940
/*
1941
  Test if a key can be used in different ranges
serg@serg.mylan's avatar
serg@serg.mylan committed
1942 1943

  SYNOPSIS
1944 1945 1946 1947 1948
    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)
1949 1950 1951
      limit             Query limit
      force_quick_range Prefer to use range (instead of full table scan) even
                        if it is more expensive.
1952 1953 1954 1955 1956

  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.
1957

1958
    In the table struct the following information is updated:
1959 1960 1961 1962 1963 1964 1965 1966 1967 1968 1969 1970 1971 1972 1973 1974
      quick_keys           - Which keys can be used
      quick_rows           - How many rows the key matches
      quick_condition_rows - E(# rows that will satisfy the table condition)

  IMPLEMENTATION
    quick_condition_rows value is obtained as follows:
      
      It is a minimum of E(#output rows) for all considered table access
      methods (range and index_merge accesses over various indexes).
    
    The obtained value is not a true E(#rows that satisfy table condition)
    but rather a pessimistic estimate. To obtain a true E(#...) one would
    need to combine estimates of various access methods, taking into account
    correlations between sets of rows they will return.
    
    For example, if values of tbl.key1 and tbl.key2 are independent (a right
1975
    assumption if we have no information about their correlation) then the
1976 1977 1978
    correct estimate will be:
    
      E(#rows("tbl.key1 < c1 AND tbl.key2 < c2")) = 
1979
      = E(#rows(tbl.key1 < c1)) / total_rows(tbl) * E(#rows(tbl.key2 < c2)
1980

1981 1982 1983 1984 1985
    which is smaller than 
      
       MIN(E(#rows(tbl.key1 < c1), E(#rows(tbl.key2 < c2)))

    which is currently produced.
serg@serg.mylan's avatar
serg@serg.mylan committed
1986

1987
  TODO
1988 1989 1990 1991 1992 1993 1994
   * Change the value returned in quick_condition_rows from a pessimistic
     estimate to true E(#rows that satisfy table condition). 
     (we can re-use some of E(#rows) calcuation code from index_merge/intersection 
      for this)
   
   * Check if this function really needs to modify keys_to_use, and change the
     code to pass it by reference if it doesn't.
1995

1996 1997 1998
   * In addition to force_quick_range other means can be (an usually are) used
     to make this function prefer range over full table scan. Figure out if
     force_quick_range is really needed.
1999

2000 2001 2002 2003
  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
2004
*/
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2005

2006 2007
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
2008 2009 2010 2011
				  ha_rows limit, bool force_quick_range)
{
  uint idx;
  double scan_time;
2012
  DBUG_ENTER("SQL_SELECT::test_quick_select");
serg@serg.mylan's avatar
serg@serg.mylan committed
2013 2014 2015
  DBUG_PRINT("enter",("keys_to_use: %lu  prev_tables: %lu  const_tables: %lu",
		      keys_to_use.to_ulonglong(), (ulong) prev_tables,
		      (ulong) const_tables));
2016
  DBUG_PRINT("info", ("records: %lu", (ulong) head->file->stats.records));
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2017 2018
  delete quick;
  quick=0;
2019 2020 2021
  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
2022 2023
      !limit)
    DBUG_RETURN(0); /* purecov: inspected */
2024 2025
  if (keys_to_use.is_clear_all())
    DBUG_RETURN(0);
2026
  records= head->file->stats.records;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2027 2028
  if (!records)
    records++;					/* purecov: inspected */
2029 2030
  scan_time= (double) records / TIME_FOR_COMPARE + 1;
  read_time= (double) head->file->scan_time() + scan_time + 1.1;
2031 2032
  if (head->force_index)
    scan_time= read_time= DBL_MAX;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2033
  if (limit < records)
2034
    read_time= (double) records + scan_time + 1; // Force to use index
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2035
  else if (read_time <= 2.0 && !force_quick_range)
2036
    DBUG_RETURN(0);				/* No need for quick select */
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2037

2038
  DBUG_PRINT("info",("Time to scan table: %g", read_time));
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2039

2040 2041
  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
2042
  {
2043
    MEM_ROOT alloc;
2044
    SEL_TREE *tree= NULL;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2045
    KEY_PART *key_parts;
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
2046
    KEY *key_info;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2047
    PARAM param;
serg@serg.mylan's avatar
serg@serg.mylan committed
2048

bk@work.mysql.com's avatar
bk@work.mysql.com committed
2049
    /* set up parameter that is passed to all functions */
2050
    param.thd= thd;
2051
    param.baseflag=head->file->ha_table_flags();
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2052 2053 2054 2055 2056
    param.prev_tables=prev_tables | const_tables;
    param.read_tables=read_tables;
    param.current_table= head->map;
    param.table=head;
    param.keys=0;
2057
    param.mem_root= &alloc;
2058
    param.old_root= thd->mem_root;
2059
    param.needed_reg= &needed_reg;
2060
    param.imerge_cost_buff_size= 0;
2061
    param.using_real_indexes= TRUE;
2062
    param.remove_jump_scans= TRUE;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2063

monty@mashka.mysql.fi's avatar
monty@mashka.mysql.fi committed
2064
    thd->no_errors=1;				// Don't warn about NULL
2065
    init_sql_alloc(&alloc, thd->variables.range_alloc_block_size, 0);
2066 2067 2068 2069
    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
2070
    {
monty@mashka.mysql.fi's avatar
monty@mashka.mysql.fi committed
2071
      thd->no_errors=0;
2072
      free_root(&alloc,MYF(0));			// Return memory & allocator
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2073 2074 2075
      DBUG_RETURN(0);				// Can't use range
    }
    key_parts= param.key_parts;
2076
    thd->mem_root= &alloc;
2077 2078 2079 2080

    /*
      Make an array with description of all key parts of all table keys.
      This is used in get_mm_parts function.
2081
    */
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
2082
    key_info= head->key_info;
2083
    for (idx=0 ; idx < head->s->keys ; idx++, key_info++)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2084
    {
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
2085
      KEY_PART_INFO *key_part_info;
2086
      if (!keys_to_use.is_set(idx))
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2087 2088 2089 2090 2091
	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
2092 2093 2094
      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
2095
      {
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
2096 2097 2098 2099 2100 2101
	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;
2102
        key_parts->image_type =
2103
          (key_info->flags & HA_SPATIAL) ? Field::itMBR : Field::itRAW;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2104 2105 2106 2107 2108
      }
      param.real_keynr[param.keys++]=idx;
    }
    param.key_parts_end=key_parts;

sergefp@mysql.com's avatar
sergefp@mysql.com committed
2109 2110 2111 2112
    /* 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);
2113 2114 2115
      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
2116 2117 2118 2119 2120
      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;
    }
2121

2122 2123 2124 2125 2126
    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
2127
    {
2128 2129 2130 2131 2132 2133 2134 2135
      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;
        }
2136 2137 2138 2139 2140 2141
        /*
          If the tree can't be used for range scans, proceed anyway, as we
          can construct a group-min-max quick select
        */
        if (tree->type != SEL_TREE::KEY && tree->type != SEL_TREE::KEY_SMALLER)
          tree= NULL;
2142
      }
2143 2144 2145 2146 2147 2148 2149
    }

    /*
      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);
2150
    if (group_trp)
2151
    {
2152 2153 2154 2155 2156 2157 2158
      param.table->quick_condition_rows= min(group_trp->records,
                                             head->file->stats.records);
      if (group_trp->read_cost < best_read_time)
      {
        best_trp= group_trp;
        best_read_time= best_trp->read_cost;
      }
2159 2160 2161
    }

    if (tree)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2162
    {
monty@mysql.com's avatar
monty@mysql.com committed
2163 2164 2165
      /*
        It is possible to use a range-based quick select (but it might be
        slower than 'all' table scan).
2166 2167
      */
      if (tree->merges.is_empty())
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2168
      {
2169 2170 2171 2172 2173
        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 */
2174
        if ((range_trp= get_key_scans_params(&param, tree, FALSE, TRUE,
2175 2176 2177 2178 2179 2180
                                             best_read_time)))
        {
          best_trp= range_trp;
          best_read_time= best_trp->read_cost;
        }

2181
        /*
2182 2183 2184
          Simultaneous key scans and row deletes on several handler
          objects are not allowed so don't use ROR-intersection for
          table deletes.
2185
        */
2186 2187 2188 2189
        if ((thd->lex->sql_command != SQLCOM_DELETE))
#ifdef NOT_USED
          if ((thd->lex->sql_command != SQLCOM_UPDATE))
#endif
2190
        {
2191
          /*
2192 2193
            Get best non-covering ROR-intersection plan and prepare data for
            building covering ROR-intersection.
2194
          */
2195 2196
          if ((rori_trp= get_best_ror_intersect(&param, tree, best_read_time,
                                                &can_build_covering)))
2197
          {
2198 2199
            best_trp= rori_trp;
            best_read_time= best_trp->read_cost;
2200 2201
            /*
              Try constructing covering ROR-intersect only if it looks possible
2202 2203
              and worth doing.
            */
2204 2205 2206 2207
            if (!rori_trp->is_covering && can_build_covering &&
                (rori_trp= get_best_covering_ror_intersect(&param, tree,
                                                           best_read_time)))
              best_trp= rori_trp;
2208 2209
          }
        }
2210 2211 2212 2213 2214 2215 2216 2217 2218 2219 2220
      }
      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++))
2221
        {
2222
          new_conj_trp= get_best_disjunct_quick(&param, imerge, best_read_time);
2223 2224 2225
          if (new_conj_trp)
            set_if_smaller(param.table->quick_condition_rows, 
                           new_conj_trp->records);
2226 2227 2228
          if (!best_conj_trp || (new_conj_trp && new_conj_trp->read_cost <
                                 best_conj_trp->read_cost))
            best_conj_trp= new_conj_trp;
2229
        }
2230 2231 2232 2233
        if (best_conj_trp)
          best_trp= best_conj_trp;
      }
    }
2234

2235
    thd->mem_root= param.old_root;
2236 2237 2238 2239 2240 2241 2242 2243 2244

    /* 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
2245 2246
      }
    }
2247 2248

  free_mem:
2249
    free_root(&alloc,MYF(0));			// Return memory & allocator
2250
    thd->mem_root= param.old_root;
monty@mashka.mysql.fi's avatar
monty@mashka.mysql.fi committed
2251
    thd->no_errors=0;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2252
  }
2253

2254
  DBUG_EXECUTE("info", print_quick(quick, &needed_reg););
2255

bk@work.mysql.com's avatar
bk@work.mysql.com committed
2256 2257 2258 2259 2260 2261 2262
  /*
    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);
}

2263
/****************************************************************************
2264
 * Partition pruning module
2265 2266 2267
 ****************************************************************************/
#ifdef WITH_PARTITION_STORAGE_ENGINE

2268 2269 2270 2271 2272 2273 2274 2275 2276 2277 2278 2279 2280 2281 2282 2283 2284 2285 2286 2287 2288 2289 2290 2291 2292 2293 2294 2295 2296 2297 2298 2299 2300 2301 2302 2303 2304 2305 2306 2307 2308
/*
  PartitionPruningModule

  This part of the code does partition pruning. Partition pruning solves the
  following problem: given a query over partitioned tables, find partitions
  that we will not need to access (i.e. partitions that we can assume to be
  empty) when executing the query.
  The set of partitions to prune doesn't depend on which query execution
  plan will be used to execute the query.
  
  HOW IT WORKS
  
  Partition pruning module makes use of RangeAnalysisModule. The following
  examples show how the problem of partition pruning can be reduced to the 
  range analysis problem:
  
  EXAMPLE 1
    Consider a query:
    
      SELECT * FROM t1 WHERE (t1.a < 5 OR t1.a = 10) AND t1.a > 3 AND t1.b='z'
    
    where table t1 is partitioned using PARTITION BY RANGE(t1.a).  An apparent
    way to find the used (i.e. not pruned away) partitions is as follows:
    
    1. analyze the WHERE clause and extract the list of intervals over t1.a
       for the above query we will get this list: {(3 < t1.a < 5), (t1.a=10)}

    2. for each interval I
       {
         find partitions that have non-empty intersection with I;
         mark them as used;
       }
       
  EXAMPLE 2
    Suppose the table is partitioned by HASH(part_func(t1.a, t1.b)). Then
    we need to:

    1. Analyze the WHERE clause and get a list of intervals over (t1.a, t1.b).
       The list of intervals we'll obtain will look like this:
       ((t1.a, t1.b) = (1,'foo')),
       ((t1.a, t1.b) = (2,'bar')), 
2309
       ((t1,a, t1.b) > (10,'zz'))
2310 2311 2312 2313 2314 2315 2316 2317 2318 2319 2320 2321 2322 2323 2324 2325 2326 2327 2328 2329 2330 2331 2332 2333 2334 2335 2336 2337 2338
       
    2. for each interval I 
       {
         if (the interval has form "(t1.a, t1.b) = (const1, const2)" )
         {
           calculate HASH(part_func(t1.a, t1.b));
           find which partition has records with this hash value and mark
             it as used;
         }
         else
         {
           mark all partitions as used; 
           break;
         }
       }

   For both examples the step #1 is exactly what RangeAnalysisModule could
   be used to do, if it was provided with appropriate index description
   (array of KEY_PART structures). 
   In example #1, we need to provide it with description of index(t1.a), 
   in example #2, we need to provide it with description of index(t1.a, t1.b).
   
   These index descriptions are further called "partitioning index
   descriptions". Note that it doesn't matter if such indexes really exist,
   as range analysis module only uses the description.
   
   Putting it all together, partitioning module works as follows:
   
   prune_partitions() {
2339
     call create_partition_index_description();
2340 2341 2342 2343 2344 2345 2346 2347 2348

     call get_mm_tree(); // invoke the RangeAnalysisModule
     
     // analyze the obtained interval list and get used partitions 
     call find_used_partitions();
  }

*/

2349 2350 2351 2352 2353 2354 2355 2356 2357 2358
struct st_part_prune_param;
struct st_part_opt_info;

typedef void (*mark_full_part_func)(partition_info*, uint32);

/*
  Partition pruning operation context
*/
typedef struct st_part_prune_param
{
2359
  RANGE_OPT_PARAM range_param; /* Range analyzer parameters */
2360

2361 2362 2363 2364 2365 2366 2367 2368 2369 2370 2371 2372 2373 2374 2375 2376 2377 2378 2379 2380 2381 2382 2383 2384 2385 2386 2387 2388 2389 2390 2391 2392 2393 2394 2395 2396 2397 2398 2399 2400 2401 2402 2403 2404 2405 2406
  /***************************************************************
   Following fields are filled in based solely on partitioning 
   definition and not modified after that:
   **************************************************************/
  partition_info *part_info; /* Copy of table->part_info */
  /* Function to get partition id from partitioning fields only */
  get_part_id_func get_top_partition_id_func;
  /* Function to mark a partition as used (w/all subpartitions if they exist)*/
  mark_full_part_func mark_full_partition_used;
 
  /* Partitioning 'index' description, array of key parts */
  KEY_PART *key;
  
  /*
    Number of fields in partitioning 'index' definition created for
    partitioning (0 if partitioning 'index' doesn't include partitioning
    fields)
  */
  uint part_fields;
  uint subpart_fields; /* Same as above for subpartitioning */
  
  /* 
    Number of the last partitioning field keypart in the index, or -1 if
    partitioning index definition doesn't include partitioning fields.
  */
  int last_part_partno;
  int last_subpart_partno; /* Same as above for supartitioning */

  /*
    is_part_keypart[i] == test(keypart #i in partitioning index is a member
                               used in partitioning)
    Used to maintain current values of cur_part_fields and cur_subpart_fields
  */
  my_bool *is_part_keypart;
  /* Same as above for subpartitioning */
  my_bool *is_subpart_keypart;

  /***************************************************************
   Following fields form find_used_partitions() recursion context:
   **************************************************************/
  SEL_ARG **arg_stack;     /* "Stack" of SEL_ARGs */
  SEL_ARG **arg_stack_end; /* Top of the stack    */
  /* Number of partitioning fields for which we have a SEL_ARG* in arg_stack */
  uint cur_part_fields;
  /* Same as cur_part_fields, but for subpartitioning */
  uint cur_subpart_fields;
2407

2408 2409 2410 2411 2412
  /* Iterator to be used to obtain the "current" set of used partitions */
  PARTITION_ITERATOR part_iter;

  /* Initialized bitmap of no_subparts size */
  MY_BITMAP subparts_bitmap;
2413 2414
} PART_PRUNE_PARAM;

2415
static bool create_partition_index_description(PART_PRUNE_PARAM *prune_par);
2416
static int find_used_partitions(PART_PRUNE_PARAM *ppar, SEL_ARG *key_tree);
2417 2418 2419 2420
static int find_used_partitions_imerge(PART_PRUNE_PARAM *ppar,
                                       SEL_IMERGE *imerge);
static int find_used_partitions_imerge_list(PART_PRUNE_PARAM *ppar,
                                            List<SEL_IMERGE> &merges);
2421 2422 2423 2424 2425 2426 2427 2428
static void mark_all_partitions_as_used(partition_info *part_info);
static uint32 part_num_to_part_id_range(PART_PRUNE_PARAM* prune_par, 
                                        uint32 num);

#ifndef DBUG_OFF
static void print_partitioning_index(KEY_PART *parts, KEY_PART *parts_end);
static void dbug_print_field(Field *field);
static void dbug_print_segment_range(SEL_ARG *arg, KEY_PART *part);
2429
static void dbug_print_singlepoint_range(SEL_ARG **start, uint num);
2430 2431 2432 2433 2434 2435 2436 2437 2438 2439 2440 2441 2442 2443 2444 2445 2446 2447 2448 2449 2450 2451 2452 2453 2454 2455 2456 2457 2458 2459 2460 2461 2462 2463 2464 2465 2466 2467 2468 2469 2470 2471 2472 2473 2474 2475
#endif


/*
  Perform partition pruning for a given table and condition.

  SYNOPSIS
    prune_partitions()
      thd           Thread handle
      table         Table to perform partition pruning for
      pprune_cond   Condition to use for partition pruning
  
  DESCRIPTION
    This function assumes that all partitions are marked as unused when it
    is invoked. The function analyzes the condition, finds partitions that
    need to be used to retrieve the records that match the condition, and 
    marks them as used by setting appropriate bit in part_info->used_partitions
    In the worst case all partitions are marked as used.

  NOTE
    This function returns promptly if called for non-partitioned table.

  RETURN
    TRUE   We've inferred that no partitions need to be used (i.e. no table
           records will satisfy pprune_cond)
    FALSE  Otherwise
*/

bool prune_partitions(THD *thd, TABLE *table, Item *pprune_cond)
{
  bool retval= FALSE;
  partition_info *part_info = table->part_info;
  DBUG_ENTER("prune_partitions");

  if (!part_info)
    DBUG_RETURN(FALSE); /* not a partitioned table */
  
  if (!pprune_cond)
  {
    mark_all_partitions_as_used(part_info);
    DBUG_RETURN(FALSE);
  }
  
  PART_PRUNE_PARAM prune_param;
  MEM_ROOT alloc;
  RANGE_OPT_PARAM  *range_par= &prune_param.range_param;
2476
  my_bitmap_map *old_read_set, *old_write_set;
2477 2478 2479 2480 2481 2482

  prune_param.part_info= part_info;
  init_sql_alloc(&alloc, thd->variables.range_alloc_block_size, 0);
  range_par->mem_root= &alloc;
  range_par->old_root= thd->mem_root;

2483
  if (create_partition_index_description(&prune_param))
2484 2485 2486 2487 2488 2489
  {
    mark_all_partitions_as_used(part_info);
    free_root(&alloc,MYF(0));		// Return memory & allocator
    DBUG_RETURN(FALSE);
  }
  
2490 2491
  old_write_set= dbug_tmp_use_all_columns(table, table->write_set);
  old_read_set=  dbug_tmp_use_all_columns(table, table->read_set);
2492 2493 2494 2495 2496 2497 2498 2499
  range_par->thd= thd;
  range_par->table= table;
  /* range_par->cond doesn't need initialization */
  range_par->prev_tables= range_par->read_tables= 0;
  range_par->current_table= table->map;

  range_par->keys= 1; // one index
  range_par->using_real_indexes= FALSE;
2500
  range_par->remove_jump_scans= FALSE;
2501 2502 2503 2504
  range_par->real_keynr[0]= 0;

  thd->no_errors=1;				// Don't warn about NULL
  thd->mem_root=&alloc;
patg@govinda.patg.net's avatar
patg@govinda.patg.net committed
2505 2506 2507

  bitmap_clear_all(&part_info->used_partitions);

2508 2509 2510 2511 2512 2513
  prune_param.key= prune_param.range_param.key_parts;
  SEL_TREE *tree;
  SEL_ARG *arg;
  int res;

  tree= get_mm_tree(range_par, pprune_cond);
2514
  if (!tree)
2515 2516 2517 2518 2519 2520 2521
    goto all_used;

  if (tree->type == SEL_TREE::IMPOSSIBLE)
  {
    retval= TRUE;
    goto end;
  }
2522 2523 2524

  if (tree->type != SEL_TREE::KEY && tree->type != SEL_TREE::KEY_SMALLER)
    goto all_used;
2525

2526 2527
  if (tree->merges.is_empty())
  {
2528
    /* Range analysis has produced a single list of intervals. */
2529 2530 2531
    prune_param.arg_stack_end= prune_param.arg_stack;
    prune_param.cur_part_fields= 0;
    prune_param.cur_subpart_fields= 0;
2532
    init_all_partitions_iterator(part_info, &prune_param.part_iter);
2533 2534 2535 2536 2537 2538
    if (!tree->keys[0] || (-1 == (res= find_used_partitions(&prune_param,
                                                            tree->keys[0]))))
      goto all_used;
  }
  else
  {
2539 2540
    if (tree->merges.elements == 1)
    {
2541 2542 2543 2544 2545 2546 2547 2548 2549 2550 2551
      /* 
        Range analysis has produced a "merge" of several intervals lists, a 
        SEL_TREE that represents an expression in form         
          sel_imerge = (tree1 OR tree2 OR ... OR treeN)
        that cannot be reduced to one tree. This can only happen when 
        partitioning index has several keyparts and the condition is OR of
        conditions that refer to different key parts. For example, we'll get
        here for "partitioning_field=const1 OR subpartitioning_field=const2"
      */
      if (-1 == (res= find_used_partitions_imerge(&prune_param,
                                                  tree->merges.head())))
2552 2553 2554
        goto all_used;
    }
    else
2555
    {
2556 2557 2558 2559 2560 2561 2562 2563 2564
      /* 
        Range analysis has produced a list of several imerges, i.e. a
        structure that represents a condition in form 
        imerge_list= (sel_imerge1 AND sel_imerge2 AND ... AND sel_imergeN)
        This is produced for complicated WHERE clauses that range analyzer
        can't really analyze properly.
      */
      if (-1 == (res= find_used_partitions_imerge_list(&prune_param,
                                                       tree->merges)))
2565 2566 2567 2568
        goto all_used;
    }
  }
  
2569 2570 2571 2572 2573 2574
  /*
    res == 0 => no used partitions => retval=TRUE
    res == 1 => some used partitions => retval=FALSE
    res == -1 - we jump over this line to all_used:
  */
  retval= test(!res);
2575 2576 2577
  goto end;

all_used:
2578
  retval= FALSE; // some partitions are used
2579 2580
  mark_all_partitions_as_used(prune_param.part_info);
end:
2581 2582
  dbug_tmp_restore_column_map(table->write_set, old_write_set);
  dbug_tmp_restore_column_map(table->read_set,  old_read_set);
2583 2584 2585 2586 2587 2588 2589 2590
  thd->no_errors=0;
  thd->mem_root= range_par->old_root;
  free_root(&alloc,MYF(0));			// Return memory & allocator
  DBUG_RETURN(retval);
}


/*
2591
  Store field key image to table record
2592 2593

  SYNOPSIS
2594 2595 2596 2597 2598 2599 2600 2601 2602 2603
    store_key_image_to_rec()
      field  Field which key image should be stored
      ptr    Field value in key format
      len    Length of the value, in bytes

  DESCRIPTION
    Copy the field value from its key image to the table record. The source
    is the value in key image format, occupying len bytes in buffer pointed
    by ptr. The destination is table record, in "field value in table record"
    format.
2604 2605
*/

2606
void store_key_image_to_rec(Field *field, char *ptr, uint len)
2607 2608
{
  /* Do the same as print_key() does */ 
2609 2610
  my_bitmap_map *old_map;

2611 2612 2613 2614 2615 2616 2617
  if (field->real_maybe_null())
  {
    if (*ptr)
    {
      field->set_null();
      return;
    }
2618
    field->set_notnull();
2619 2620
    ptr++;
  }    
2621 2622
  old_map= dbug_tmp_use_all_columns(field->table,
                                    field->table->write_set);
2623
  field->set_key_image(ptr, len); 
2624
  dbug_tmp_restore_column_map(field->table->write_set, old_map);
2625 2626 2627 2628 2629 2630 2631 2632 2633
}


/*
  For SEL_ARG* array, store sel_arg->min values into table record buffer

  SYNOPSIS
    store_selargs_to_rec()
      ppar   Partition pruning context
2634
      start  Array of SEL_ARG* for which the minimum values should be stored
2635
      num    Number of elements in the array
2636 2637 2638 2639

  DESCRIPTION
    For each SEL_ARG* interval in the specified array, store the left edge
    field value (sel_arg->min, key image format) into the table record.
2640 2641 2642 2643 2644 2645 2646 2647 2648 2649 2650 2651 2652 2653 2654 2655 2656 2657 2658
*/

static void store_selargs_to_rec(PART_PRUNE_PARAM *ppar, SEL_ARG **start,
                                 int num)
{
  KEY_PART *parts= ppar->range_param.key_parts;
  for (SEL_ARG **end= start + num; start != end; start++)
  {
    SEL_ARG *sel_arg= (*start);
    store_key_image_to_rec(sel_arg->field, sel_arg->min_value,
                           parts[sel_arg->part].length);
  }
}


/* Mark a partition as used in the case when there are no subpartitions */
static void mark_full_partition_used_no_parts(partition_info* part_info,
                                              uint32 part_id)
{
patg@govinda.patg.net's avatar
patg@govinda.patg.net committed
2659 2660
  DBUG_ENTER("mark_full_partition_used_no_parts");
  DBUG_PRINT("enter", ("Mark partition %u as used", part_id));
2661
  bitmap_set_bit(&part_info->used_partitions, part_id);
patg@govinda.patg.net's avatar
patg@govinda.patg.net committed
2662
  DBUG_VOID_RETURN;
2663 2664 2665 2666 2667 2668 2669 2670 2671
}


/* Mark a partition as used in the case when there are subpartitions */
static void mark_full_partition_used_with_parts(partition_info *part_info,
                                                uint32 part_id)
{
  uint32 start= part_id * part_info->no_subparts;
  uint32 end=   start + part_info->no_subparts; 
patg@govinda.patg.net's avatar
patg@govinda.patg.net committed
2672 2673
  DBUG_ENTER("mark_full_partition_used_with_parts");

2674
  for (; start != end; start++)
patg@govinda.patg.net's avatar
patg@govinda.patg.net committed
2675 2676
  {
    DBUG_PRINT("info", ("1:Mark subpartition %u as used", start));
2677
    bitmap_set_bit(&part_info->used_partitions, start);
patg@govinda.patg.net's avatar
patg@govinda.patg.net committed
2678 2679
  }
  DBUG_VOID_RETURN;
2680 2681
}

2682 2683 2684 2685 2686 2687 2688 2689 2690 2691 2692
/*
  Find the set of used partitions for List<SEL_IMERGE>
  SYNOPSIS
    find_used_partitions_imerge_list
      ppar      Partition pruning context.
      key_tree  Intervals tree to perform pruning for.
      
  DESCRIPTION
    List<SEL_IMERGE> represents "imerge1 AND imerge2 AND ...". 
    The set of used partitions is an intersection of used partitions sets
    for imerge_{i}.
2693
    We accumulate this intersection in a separate bitmap.
2694 2695 2696 2697 2698 2699 2700 2701 2702 2703
 
  RETURN 
    See find_used_partitions()
*/

static int find_used_partitions_imerge_list(PART_PRUNE_PARAM *ppar,
                                            List<SEL_IMERGE> &merges)
{
  MY_BITMAP all_merges;
  uint bitmap_bytes;
2704
  my_bitmap_map *bitmap_buf;
2705 2706
  uint n_bits= ppar->part_info->used_partitions.n_bits;
  bitmap_bytes= bitmap_buffer_size(n_bits);
2707 2708
  if (!(bitmap_buf= (my_bitmap_map*) alloc_root(ppar->range_param.mem_root,
                                                bitmap_bytes)))
2709
  {
patg@govinda.patg.net's avatar
patg@govinda.patg.net committed
2710
    /*
2711
      Fallback, process just the first SEL_IMERGE. This can leave us with more
2712 2713 2714 2715 2716 2717
      partitions marked as used then actually needed.
    */
    return find_used_partitions_imerge(ppar, merges.head());
  }
  bitmap_init(&all_merges, bitmap_buf, n_bits, FALSE);
  bitmap_set_prefix(&all_merges, n_bits);
patg@govinda.patg.net's avatar
patg@govinda.patg.net committed
2718

2719 2720 2721 2722 2723 2724 2725 2726 2727 2728
  List_iterator<SEL_IMERGE> it(merges);
  SEL_IMERGE *imerge;
  while ((imerge=it++))
  {
    int res= find_used_partitions_imerge(ppar, imerge);
    if (!res)
    {
      /* no used partitions on one ANDed imerge => no used partitions at all */
      return 0;
    }
patg@govinda.patg.net's avatar
patg@govinda.patg.net committed
2729

2730 2731 2732 2733 2734 2735 2736 2737 2738 2739 2740 2741 2742 2743 2744 2745 2746 2747 2748 2749 2750 2751 2752 2753 2754 2755 2756 2757 2758 2759
    if (res != -1)
      bitmap_intersect(&all_merges, &ppar->part_info->used_partitions);

    if (bitmap_is_clear_all(&all_merges))
      return 0;

    bitmap_clear_all(&ppar->part_info->used_partitions);
  }
  memcpy(ppar->part_info->used_partitions.bitmap, all_merges.bitmap,
         bitmap_bytes);
  return 1;
}


/*
  Find the set of used partitions for SEL_IMERGE structure
  SYNOPSIS
    find_used_partitions_imerge()
      ppar      Partition pruning context.
      key_tree  Intervals tree to perform pruning for.
      
  DESCRIPTION
    SEL_IMERGE represents "tree1 OR tree2 OR ...". The implementation is
    trivial - just use mark used partitions for each tree and bail out early
    if for some tree_{i} all partitions are used.
 
  RETURN 
    See find_used_partitions().
*/

2760
static
2761
int find_used_partitions_imerge(PART_PRUNE_PARAM *ppar, SEL_IMERGE *imerge)
2762
{
2763 2764 2765 2766 2767 2768
  int res= 0;
  for (SEL_TREE **ptree= imerge->trees; ptree < imerge->trees_next; ptree++)
  {
    ppar->arg_stack_end= ppar->arg_stack;
    ppar->cur_part_fields= 0;
    ppar->cur_subpart_fields= 0;
2769
    init_all_partitions_iterator(ppar->part_info, &ppar->part_iter);
2770 2771
    SEL_ARG *key_tree= (*ptree)->keys[0];
    if (!key_tree || (-1 == (res |= find_used_partitions(ppar, key_tree))))
2772 2773 2774
      return -1;
  }
  return res;
2775 2776 2777 2778
}


/*
2779
  Collect partitioning ranges for the SEL_ARG tree and mark partitions as used
2780 2781 2782 2783

  SYNOPSIS
    find_used_partitions()
      ppar      Partition pruning context.
2784
      key_tree  SEL_ARG range tree to perform pruning for
2785 2786 2787

  DESCRIPTION
    This function 
2788 2789
      * recursively walks the SEL_ARG* tree collecting partitioning "intervals"
      * finds the partitions one needs to use to get rows in these intervals
2790
      * marks these partitions as used.
2791 2792 2793 2794 2795 2796 2797 2798 2799 2800 2801 2802 2803 2804 2805 2806 2807
    The next session desribes the process in greater detail.
 
  IMPLEMENTATION
    TYPES OF RESTRICTIONS THAT WE CAN OBTAIN PARTITIONS FOR    
    We can find out which [sub]partitions to use if we obtain restrictions on 
    [sub]partitioning fields in the following form:
    1.  "partition_field1=const1 AND ... AND partition_fieldN=constN"
    1.1  Same as (1) but for subpartition fields

    If partitioning supports interval analysis (i.e. partitioning is a
    function of a single table field, and partition_info::
    get_part_iter_for_interval != NULL), then we can also use condition in
    this form:
    2.  "const1 <=? partition_field <=? const2"
    2.1  Same as (2) but for subpartition_field

    INFERRING THE RESTRICTIONS FROM SEL_ARG TREE
2808
    
2809
    The below is an example of what SEL_ARG tree may represent:
2810
    
2811 2812 2813 2814 2815 2816 2817 2818 2819 2820 2821 2822 2823 2824 2825 2826 2827 2828 2829 2830 2831 2832 2833 2834 2835 2836 2837 2838
    (start)
     |                           $
     |   Partitioning keyparts   $  subpartitioning keyparts
     |                           $
     |     ...          ...      $
     |      |            |       $
     | +---------+  +---------+  $  +-----------+  +-----------+
     \-| par1=c1 |--| par2=c2 |-----| subpar1=c3|--| subpar2=c5|
       +---------+  +---------+  $  +-----------+  +-----------+
            |                    $        |             |
            |                    $        |        +-----------+ 
            |                    $        |        | subpar2=c6|
            |                    $        |        +-----------+ 
            |                    $        |
            |                    $  +-----------+  +-----------+
            |                    $  | subpar1=c4|--| subpar2=c8|
            |                    $  +-----------+  +-----------+
            |                    $         
            |                    $
       +---------+               $  +------------+  +------------+
       | par1=c2 |------------------| subpar1=c10|--| subpar2=c12|
       +---------+               $  +------------+  +------------+
            |                    $
           ...                   $

    The up-down connections are connections via SEL_ARG::left and
    SEL_ARG::right. A horizontal connection to the right is the
    SEL_ARG::next_key_part connection.
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 2872 2873 2874 2875
    find_used_partitions() traverses the entire tree via recursion on
     * SEL_ARG::next_key_part (from left to right on the picture)
     * SEL_ARG::left|right (up/down on the pic). Left-right recursion is
       performed for each depth level.
    
    Recursion descent on SEL_ARG::next_key_part is used to accumulate (in
    ppar->arg_stack) constraints on partitioning and subpartitioning fields.
    For the example in the above picture, one of stack states is:
      in find_used_partitions(key_tree = "subpar2=c5") (***)
      in find_used_partitions(key_tree = "subpar1=c3")
      in find_used_partitions(key_tree = "par2=c2")   (**)
      in find_used_partitions(key_tree = "par1=c1")
      in prune_partitions(...)
    We apply partitioning limits as soon as possible, e.g. when we reach the
    depth (**), we find which partition(s) correspond to "par1=c1 AND par2=c2",
    and save them in ppar->part_iter.
    When we reach the depth (***), we find which subpartition(s) correspond to
    "subpar1=c3 AND subpar2=c5", and then mark appropriate subpartitions in
    appropriate subpartitions as used.
    
    It is possible that constraints on some partitioning fields are missing.
    For the above example, consider this stack state:
      in find_used_partitions(key_tree = "subpar2=c12") (***)
      in find_used_partitions(key_tree = "subpar1=c10")
      in find_used_partitions(key_tree = "par1=c2")
      in prune_partitions(...)
    Here we don't have constraints for all partitioning fields. Since we've
    never set the ppar->part_iter to contain used set of partitions, we use
    its default "all partitions" value.  We get  subpartition id for 
    "subpar1=c3 AND subpar2=c5", and mark that subpartition as used in every
    partition.

    The inverse is also possible: we may get constraints on partitioning
    fields, but not constraints on subpartitioning fields. In that case,
    calls to find_used_partitions() with depth below (**) will return -1,
    and we will mark entire partition as used.
2876

2877 2878
  TODO
    Replace recursion on SEL_ARG::left and SEL_ARG::right with a loop
2879 2880 2881 2882 2883

  RETURN
    1   OK, one or more [sub]partitions are marked as used.
    0   The passed condition doesn't match any partitions
   -1   Couldn't infer any partition pruning "intervals" from the passed 
2884 2885
        SEL_ARG* tree (which means that all partitions should be marked as
        used) Marking partitions as used is the responsibility of the caller.
2886 2887 2888 2889 2890 2891 2892 2893 2894 2895 2896 2897 2898 2899 2900 2901 2902 2903
*/

static 
int find_used_partitions(PART_PRUNE_PARAM *ppar, SEL_ARG *key_tree)
{
  int res, left_res=0, right_res=0;
  int partno= (int)key_tree->part;
  bool pushed= FALSE;
  bool set_full_part_if_bad_ret= FALSE;

  if (key_tree->left != &null_element)
  {
    if (-1 == (left_res= find_used_partitions(ppar,key_tree->left)))
      return -1;
  }

  if (key_tree->type == SEL_ARG::KEY_RANGE)
  {
2904
    if (partno == 0 && (NULL != ppar->part_info->get_part_iter_for_interval))
2905 2906 2907
    {
      /* 
        Partitioning is done by RANGE|INTERVAL(monotonic_expr(fieldX)), and
2908
        we got "const1 CMP fieldX CMP const2" interval <-- psergey-todo: change
2909 2910 2911 2912
      */
      DBUG_EXECUTE("info", dbug_print_segment_range(key_tree,
                                                    ppar->range_param.
                                                    key_parts););
2913 2914 2915 2916 2917 2918 2919 2920
      res= ppar->part_info->
           get_part_iter_for_interval(ppar->part_info,
                                      FALSE,
                                      key_tree->min_value, 
                                      key_tree->max_value,
                                      key_tree->min_flag | key_tree->max_flag,
                                      &ppar->part_iter);
      if (!res)
2921
        goto go_right; /* res==0 --> no satisfying partitions */
2922
      if (res == -1)
2923
      {
2924 2925
        //get a full range iterator
        init_all_partitions_iterator(ppar->part_info, &ppar->part_iter);
2926 2927
      }
      /* 
2928
        Save our intent to mark full partition as used if we will not be able
2929 2930 2931 2932 2933 2934
        to obtain further limits on subpartitions
      */
      set_full_part_if_bad_ret= TRUE;
      goto process_next_key_part;
    }

2935 2936 2937 2938 2939 2940 2941 2942 2943 2944 2945 2946 2947 2948 2949 2950 2951 2952 2953 2954
    if (partno == ppar->last_subpart_partno && 
        (NULL != ppar->part_info->get_subpart_iter_for_interval))
    {
      PARTITION_ITERATOR subpart_iter;
      DBUG_EXECUTE("info", dbug_print_segment_range(key_tree,
                                                    ppar->range_param.
                                                    key_parts););
      res= ppar->part_info->
           get_subpart_iter_for_interval(ppar->part_info,
                                         TRUE,
                                         key_tree->min_value, 
                                         key_tree->max_value,
                                         key_tree->min_flag | key_tree->max_flag,
                                         &subpart_iter);
      DBUG_ASSERT(res); /* We can't get "no satisfying subpartitions" */
      if (res == -1)
        return -1; /* all subpartitions satisfy */
        
      uint32 subpart_id;
      bitmap_clear_all(&ppar->subparts_bitmap);
2955 2956
      while ((subpart_id= subpart_iter.get_next(&subpart_iter)) !=
             NOT_A_PARTITION_ID)
2957 2958 2959 2960 2961 2962 2963 2964 2965 2966 2967 2968 2969 2970 2971
        bitmap_set_bit(&ppar->subparts_bitmap, subpart_id);

      /* Mark each partition as used in each subpartition.  */
      uint32 part_id;
      while ((part_id= ppar->part_iter.get_next(&ppar->part_iter)) !=
              NOT_A_PARTITION_ID)
      {
        for (uint i= 0; i < ppar->part_info->no_subparts; i++)
          if (bitmap_is_set(&ppar->subparts_bitmap, i))
            bitmap_set_bit(&ppar->part_info->used_partitions,
                           part_id * ppar->part_info->no_subparts + i);
      }
      goto go_right;
    }

2972 2973 2974 2975 2976 2977 2978 2979 2980 2981 2982 2983 2984 2985 2986
    if (key_tree->is_singlepoint())
    {
      pushed= TRUE;
      ppar->cur_part_fields+=    ppar->is_part_keypart[partno];
      ppar->cur_subpart_fields+= ppar->is_subpart_keypart[partno];
      *(ppar->arg_stack_end++) = key_tree;

      if (partno == ppar->last_part_partno &&
          ppar->cur_part_fields == ppar->part_fields)
      {
        /* 
          Ok, we've got "fieldN<=>constN"-type SEL_ARGs for all partitioning
          fields. Save all constN constants into table record buffer.
        */
        store_selargs_to_rec(ppar, ppar->arg_stack, ppar->part_fields);
2987
        DBUG_EXECUTE("info", dbug_print_singlepoint_range(ppar->arg_stack,
2988 2989
                                                       ppar->part_fields););
        uint32 part_id;
2990
        longlong func_value;
2991
        /* Find in which partition the {const1, ...,constN} tuple goes */
2992 2993
        if (ppar->get_top_partition_id_func(ppar->part_info, &part_id,
                                            &func_value))
2994 2995 2996 2997 2998
        {
          res= 0; /* No satisfying partitions */
          goto pop_and_go_right;
        }
        /* Rembember the limit we got - single partition #part_id */
2999
        init_single_partition_iterator(part_id, &ppar->part_iter);
3000 3001 3002 3003 3004 3005 3006 3007 3008
        
        /*
          If there are no subpartitions/we fail to get any limit for them, 
          then we'll mark full partition as used. 
        */
        set_full_part_if_bad_ret= TRUE;
        goto process_next_key_part;
      }

3009 3010
      if (partno == ppar->last_subpart_partno &&
          ppar->cur_subpart_fields == ppar->subpart_fields)
3011 3012 3013 3014 3015 3016 3017
      {
        /* 
          Ok, we've got "fieldN<=>constN"-type SEL_ARGs for all subpartitioning
          fields. Save all constN constants into table record buffer.
        */
        store_selargs_to_rec(ppar, ppar->arg_stack_end - ppar->subpart_fields,
                             ppar->subpart_fields);
3018
        DBUG_EXECUTE("info", dbug_print_singlepoint_range(ppar->arg_stack_end- 
3019 3020 3021 3022 3023 3024 3025
                                                       ppar->subpart_fields,
                                                       ppar->subpart_fields););
        /* Find the subpartition (it's HASH/KEY so we always have one) */
        partition_info *part_info= ppar->part_info;
        uint32 subpart_id= part_info->get_subpartition_id(part_info);
        
        /* Mark this partition as used in each subpartition. */
3026 3027 3028
        uint32 part_id;
        while ((part_id= ppar->part_iter.get_next(&ppar->part_iter)) !=
                NOT_A_PARTITION_ID)
3029 3030
        {
          bitmap_set_bit(&part_info->used_partitions,
3031
                         part_id * part_info->no_subparts + subpart_id);
3032 3033 3034 3035 3036 3037 3038 3039 3040 3041 3042 3043 3044 3045 3046 3047 3048 3049 3050 3051
        }
        res= 1; /* Some partitions were marked as used */
        goto pop_and_go_right;
      }
    }
    else
    {
      /* 
        Can't handle condition on current key part. If we're that deep that 
        we're processing subpartititoning's key parts, this means we'll not be
        able to infer any suitable condition, so bail out.
      */
      if (partno >= ppar->last_part_partno)
        return -1;
    }
  }

process_next_key_part:
  if (key_tree->next_key_part)
    res= find_used_partitions(ppar, key_tree->next_key_part);
3052
  else
3053
    res= -1;
3054 3055
 
  if (set_full_part_if_bad_ret)
3056
  {
3057
    if (res == -1)
3058
    {
3059 3060 3061
      /* Got "full range" for subpartitioning fields */
      uint32 part_id;
      bool found= FALSE;
3062 3063
      while ((part_id= ppar->part_iter.get_next(&ppar->part_iter)) !=
             NOT_A_PARTITION_ID)
3064
      {
3065 3066
        ppar->mark_full_partition_used(ppar->part_info, part_id);
        found= TRUE;
3067
      }
3068
      res= test(found);
3069
    }
3070 3071 3072 3073
    /*
      Restore the "used partitions iterator" to the default setting that
      specifies iteration over all partitions.
    */
3074
    init_all_partitions_iterator(ppar->part_info, &ppar->part_iter);
3075 3076 3077 3078 3079 3080 3081 3082 3083 3084
  }

  if (pushed)
  {
pop_and_go_right:
    /* Pop this key part info off the "stack" */
    ppar->arg_stack_end--;
    ppar->cur_part_fields-=    ppar->is_part_keypart[partno];
    ppar->cur_subpart_fields-= ppar->is_subpart_keypart[partno];
  }
3085 3086 3087 3088

  if (res == -1)
    return -1;
go_right:
3089 3090 3091 3092 3093 3094 3095 3096 3097 3098 3099 3100 3101 3102 3103 3104 3105 3106 3107 3108 3109 3110 3111 3112 3113 3114 3115 3116 3117 3118 3119 3120 3121 3122 3123 3124 3125 3126 3127 3128 3129 3130 3131 3132 3133 3134 3135 3136 3137 3138 3139 3140 3141 3142 3143 3144 3145
  if (key_tree->right != &null_element)
  {
    if (-1 == (right_res= find_used_partitions(ppar,key_tree->right)))
      return -1;
  }
  return (left_res || right_res || res);
}
 

static void mark_all_partitions_as_used(partition_info *part_info)
{
  bitmap_set_all(&part_info->used_partitions);
}


/*
  Check if field types allow to construct partitioning index description
 
  SYNOPSIS
    fields_ok_for_partition_index()
      pfield  NULL-terminated array of pointers to fields.

  DESCRIPTION
    For an array of fields, check if we can use all of the fields to create
    partitioning index description.
    
    We can't process GEOMETRY fields - for these fields singlepoint intervals
    cant be generated, and non-singlepoint are "special" kinds of intervals
    to which our processing logic can't be applied.

    It is not known if we could process ENUM fields, so they are disabled to be
    on the safe side.

  RETURN 
    TRUE   Yes, fields can be used in partitioning index
    FALSE  Otherwise
*/

static bool fields_ok_for_partition_index(Field **pfield)
{
  if (!pfield)
    return FALSE;
  for (; (*pfield); pfield++)
  {
    enum_field_types ftype= (*pfield)->real_type();
    if (ftype == FIELD_TYPE_ENUM || ftype == FIELD_TYPE_GEOMETRY)
      return FALSE;
  }
  return TRUE;
}


/*
  Create partition index description and fill related info in the context
  struct

  SYNOPSIS
3146
    create_partition_index_description()
3147 3148 3149 3150 3151 3152 3153 3154 3155 3156 3157 3158 3159 3160 3161 3162
      prune_par  INOUT Partition pruning context

  DESCRIPTION
    Create partition index description. Partition index description is:

      part_index(used_fields_list(part_expr), used_fields_list(subpart_expr))

    If partitioning/sub-partitioning uses BLOB or Geometry fields, then
    corresponding fields_list(...) is not included into index description
    and we don't perform partition pruning for partitions/subpartitions.

  RETURN
    TRUE   Out of memory or can't do partition pruning at all
    FALSE  OK
*/

3163
static bool create_partition_index_description(PART_PRUNE_PARAM *ppar)
3164 3165 3166 3167 3168 3169 3170 3171 3172 3173 3174 3175 3176 3177 3178 3179 3180 3181 3182 3183
{
  RANGE_OPT_PARAM *range_par= &(ppar->range_param);
  partition_info *part_info= ppar->part_info;
  uint used_part_fields, used_subpart_fields;

  used_part_fields= fields_ok_for_partition_index(part_info->part_field_array) ?
                      part_info->no_part_fields : 0;
  used_subpart_fields= 
    fields_ok_for_partition_index(part_info->subpart_field_array)? 
      part_info->no_subpart_fields : 0;
  
  uint total_parts= used_part_fields + used_subpart_fields;

  ppar->part_fields=      used_part_fields;
  ppar->last_part_partno= (int)used_part_fields - 1;

  ppar->subpart_fields= used_subpart_fields;
  ppar->last_subpart_partno= 
    used_subpart_fields?(int)(used_part_fields + used_subpart_fields - 1): -1;

3184
  if (part_info->is_sub_partitioned())
3185 3186 3187 3188 3189 3190 3191 3192 3193 3194 3195 3196 3197 3198 3199 3200 3201 3202 3203 3204 3205 3206
  {
    ppar->mark_full_partition_used=  mark_full_partition_used_with_parts;
    ppar->get_top_partition_id_func= part_info->get_part_partition_id;
  }
  else
  {
    ppar->mark_full_partition_used=  mark_full_partition_used_no_parts;
    ppar->get_top_partition_id_func= part_info->get_partition_id;
  }

  KEY_PART *key_part;
  MEM_ROOT *alloc= range_par->mem_root;
  if (!total_parts || 
      !(key_part= (KEY_PART*)alloc_root(alloc, sizeof(KEY_PART)*
                                               total_parts)) ||
      !(ppar->arg_stack= (SEL_ARG**)alloc_root(alloc, sizeof(SEL_ARG*)* 
                                                      total_parts)) ||
      !(ppar->is_part_keypart= (my_bool*)alloc_root(alloc, sizeof(my_bool)*
                                                           total_parts)) ||
      !(ppar->is_subpart_keypart= (my_bool*)alloc_root(alloc, sizeof(my_bool)*
                                                           total_parts)))
    return TRUE;
3207 3208 3209
 
  if (ppar->subpart_fields)
  {
3210
    my_bitmap_map *buf;
3211
    uint32 bufsize= bitmap_buffer_size(ppar->part_info->no_subparts);
3212
    if (!(buf= (my_bitmap_map*) alloc_root(alloc, bufsize)))
3213
      return TRUE;
3214 3215
    bitmap_init(&ppar->subparts_bitmap, buf, ppar->part_info->no_subparts,
                FALSE);
3216
  }
3217 3218 3219
  range_par->key_parts= key_part;
  Field **field= (ppar->part_fields)? part_info->part_field_array :
                                           part_info->subpart_field_array;
3220
  bool in_subpart_fields= FALSE;
3221 3222 3223 3224 3225 3226 3227
  for (uint part= 0; part < total_parts; part++, key_part++)
  {
    key_part->key=          0;
    key_part->part=	    part;
    key_part->length=       (*field)->pack_length_in_rec();
    /* 
      psergey-todo: check yet again if this is correct for tricky field types,
3228
      e.g. see "Fix a fatal error in decimal key handling" in open_binary_frm()
3229 3230 3231 3232 3233 3234 3235 3236 3237 3238 3239 3240
    */
    key_part->store_length= (*field)->pack_length();
    if ((*field)->real_maybe_null())
      key_part->store_length+= HA_KEY_NULL_LENGTH;
    if ((*field)->type() == FIELD_TYPE_BLOB || 
        (*field)->real_type() == MYSQL_TYPE_VARCHAR)
      key_part->store_length+= HA_KEY_BLOB_LENGTH;

    key_part->field=        (*field);
    key_part->image_type =  Field::itRAW;
    /* We don't set key_parts->null_bit as it will not be used */

3241 3242
    ppar->is_part_keypart[part]= !in_subpart_fields;
    ppar->is_subpart_keypart[part]= in_subpart_fields;
3243 3244 3245 3246 3247 3248

    /*
      Check if this was last field in this array, in this case we
      switch to subpartitioning fields. (This will only happens if
      there are subpartitioning fields to cater for).
    */
3249 3250 3251
    if (!*(++field))
    {
      field= part_info->subpart_field_array;
3252
      in_subpart_fields= TRUE;
3253 3254 3255 3256 3257 3258 3259 3260 3261 3262 3263 3264 3265 3266 3267 3268 3269 3270 3271 3272 3273
    }
  }
  range_par->key_parts_end= key_part;

  DBUG_EXECUTE("info", print_partitioning_index(range_par->key_parts,
                                                range_par->key_parts_end););
  return FALSE;
}


#ifndef DBUG_OFF

static void print_partitioning_index(KEY_PART *parts, KEY_PART *parts_end)
{
  DBUG_ENTER("print_partitioning_index");
  DBUG_LOCK_FILE;
  fprintf(DBUG_FILE, "partitioning INDEX(");
  for (KEY_PART *p=parts; p != parts_end; p++)
  {
    fprintf(DBUG_FILE, "%s%s", p==parts?"":" ,", p->field->field_name);
  }
3274
  fputs(");\n", DBUG_FILE);
3275 3276 3277 3278 3279 3280 3281 3282 3283 3284 3285
  DBUG_UNLOCK_FILE;
  DBUG_VOID_RETURN;
}

/* Print field value into debug trace, in NULL-aware way. */
static void dbug_print_field(Field *field)
{
  if (field->is_real_null())
    fprintf(DBUG_FILE, "NULL");
  else
  {
3286 3287 3288
    char buf[256];
    String str(buf, sizeof(buf), &my_charset_bin);
    str.length(0);
3289 3290 3291 3292 3293 3294 3295 3296 3297 3298 3299 3300 3301 3302
    String *pstr;
    pstr= field->val_str(&str);
    fprintf(DBUG_FILE, "'%s'", pstr->c_ptr_safe());
  }
}


/* Print a "c1 < keypartX < c2" - type interval into debug trace. */
static void dbug_print_segment_range(SEL_ARG *arg, KEY_PART *part)
{
  DBUG_ENTER("dbug_print_segment_range");
  DBUG_LOCK_FILE;
  if (!(arg->min_flag & NO_MIN_RANGE))
  {
3303
    store_key_image_to_rec(part->field, (char*)(arg->min_value), part->length);
3304 3305 3306 3307 3308 3309 3310 3311 3312 3313 3314 3315 3316 3317 3318
    dbug_print_field(part->field);
    if (arg->min_flag & NEAR_MIN)
      fputs(" < ", DBUG_FILE);
    else
      fputs(" <= ", DBUG_FILE);
  }

  fprintf(DBUG_FILE, "%s", part->field->field_name);

  if (!(arg->max_flag & NO_MAX_RANGE))
  {
    if (arg->max_flag & NEAR_MAX)
      fputs(" < ", DBUG_FILE);
    else
      fputs(" <= ", DBUG_FILE);
3319
    store_key_image_to_rec(part->field, (char*)(arg->max_value), part->length);
3320 3321
    dbug_print_field(part->field);
  }
3322
  fputs("\n", DBUG_FILE);
3323 3324 3325 3326 3327 3328 3329 3330 3331
  DBUG_UNLOCK_FILE;
  DBUG_VOID_RETURN;
}


/*
  Print a singlepoint multi-keypart range interval to debug trace
 
  SYNOPSIS
3332
    dbug_print_singlepoint_range()
3333 3334 3335 3336 3337 3338 3339 3340
      start  Array of SEL_ARG* ptrs representing conditions on key parts
      num    Number of elements in the array.

  DESCRIPTION
    This function prints a "keypartN=constN AND ... AND keypartK=constK"-type 
    interval to debug trace.
*/

3341
static void dbug_print_singlepoint_range(SEL_ARG **start, uint num)
3342
{
3343
  DBUG_ENTER("dbug_print_singlepoint_range");
3344 3345
  DBUG_LOCK_FILE;
  SEL_ARG **end= start + num;
3346

3347 3348 3349 3350 3351 3352
  for (SEL_ARG **arg= start; arg != end; arg++)
  {
    Field *field= (*arg)->field;
    fprintf(DBUG_FILE, "%s%s=", (arg==start)?"":", ", field->field_name);
    dbug_print_field(field);
  }
3353
  fputs("\n", DBUG_FILE);
3354 3355 3356 3357 3358 3359 3360 3361 3362 3363
  DBUG_UNLOCK_FILE;
  DBUG_VOID_RETURN;
}
#endif

/****************************************************************************
 * Partition pruning code ends
 ****************************************************************************/
#endif

3364

3365
/*
3366 3367 3368 3369
  Get cost of 'sweep' full records retrieval.
  SYNOPSIS
    get_sweep_read_cost()
      param            Parameter from test_quick_select
3370
      records          # of records to be retrieved
3371
  RETURN
3372
    cost of sweep
3373
*/
3374

3375
double get_sweep_read_cost(const PARAM *param, ha_rows records)
3376
{
3377
  double result;
3378
  DBUG_ENTER("get_sweep_read_cost");
3379 3380
  if (param->table->file->primary_key_is_clustered())
  {
3381
    result= param->table->file->read_time(param->table->s->primary_key,
3382
                                          records, records);
3383 3384
  }
  else
3385
  {
3386
    double n_blocks=
3387 3388
      ceil(ulonglong2double(param->table->file->stats.data_file_length) /
           IO_SIZE);
3389 3390 3391 3392
    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;
3393
    DBUG_PRINT("info",("sweep: nblocks=%g, busy_blocks=%g", n_blocks,
3394
                       busy_blocks));
3395
    /*
3396
      Disabled: Bail out if # of blocks to read is bigger than # of blocks in
3397 3398 3399 3400 3401 3402 3403 3404
      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' */
3405
      result= busy_blocks*(DISK_SEEK_BASE_COST +
3406 3407 3408 3409
                          DISK_SEEK_PROP_COST*n_blocks/busy_blocks);
    }
    else
    {
3410
      /*
3411 3412 3413
        Possibly this is a join with source table being non-last table, so
        assume that disk seeks are random here.
      */
3414
      result= busy_blocks;
3415 3416
    }
  }
3417
  DBUG_PRINT("info",("returning cost=%g", result));
3418
  DBUG_RETURN(result);
3419
}
3420 3421


3422 3423 3424 3425
/*
  Get best plan for a SEL_IMERGE disjunctive expression.
  SYNOPSIS
    get_best_disjunct_quick()
3426 3427
      param     Parameter from check_quick_select function
      imerge    Expression to use
3428
      read_time Don't create scans with cost > read_time
3429

3430
  NOTES
3431
    index_merge cost is calculated as follows:
3432
    index_merge_cost =
3433 3434 3435 3436 3437
      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))
3438 3439
       For non-CPK scans,
         cost(index_read_i) = {cost of ordinary 'index only' scan}
3440 3441 3442 3443 3444
       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
3445
        cost(rowid_to_row_scan) =
3446
          {cost of ordinary clustered PK scan with n_ranges=n_rows}
3447 3448

      Otherwise, we use the following model to calculate costs:
3449
      We need to retrieve n_rows rows from file that occupies n_blocks blocks.
3450
      We assume that offsets of rows we need are independent variates with
3451
      uniform distribution in [0..max_file_offset] range.
3452

3453 3454
      We'll denote block as "busy" if it contains row(s) we need to retrieve
      and "empty" if doesn't contain rows we need.
3455

3456
      Probability that a block is empty is (1 - 1/n_blocks)^n_rows (this
3457
      applies to any block in file). Let x_i be a variate taking value 1 if
3458
      block #i is empty and 0 otherwise.
3459

3460 3461
      Then E(x_i) = (1 - 1/n_blocks)^n_rows;

3462 3463
      E(n_empty_blocks) = E(sum(x_i)) = sum(E(x_i)) =
        = n_blocks * ((1 - 1/n_blocks)^n_rows) =
3464 3465 3466 3467
       ~= 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)).
3468

3469 3470
      Average size of "hole" between neighbor non-empty blocks is
           E(hole_size) = n_blocks/E(n_busy_blocks).
3471

3472 3473 3474 3475 3476 3477
      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.
3478 3479 3480 3481 3482

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

  RETURN
3483 3484
    Created read plan
    NULL - Out of memory or no read scan could be built.
3485
*/
3486

3487 3488
static
TABLE_READ_PLAN *get_best_disjunct_quick(PARAM *param, SEL_IMERGE *imerge,
3489
                                         double read_time)
3490 3491 3492 3493 3494 3495 3496
{
  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
3497
  bool imerge_too_expensive= FALSE;
3498 3499 3500 3501
  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
3502 3503
  bool all_scans_ror_able= TRUE;
  bool all_scans_rors= TRUE;
3504 3505 3506 3507 3508 3509 3510 3511 3512
  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));

3513
  if (!(range_scans= (TRP_RANGE**)alloc_root(param->mem_root,
3514 3515 3516
                                             sizeof(TRP_RANGE*)*
                                             n_child_scans)))
    DBUG_RETURN(NULL);
3517
  /*
3518 3519 3520
    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.
3521
  */
3522
  for (ptree= imerge->trees, cur_child= range_scans;
3523
       ptree != imerge->trees_next;
3524
       ptree++, cur_child++)
3525
  {
3526 3527
    DBUG_EXECUTE("info", print_sel_tree(param, *ptree, &(*ptree)->keys_map,
                                        "tree in SEL_IMERGE"););
3528
    if (!(*cur_child= get_key_scans_params(param, *ptree, TRUE, FALSE, read_time)))
3529 3530
    {
      /*
3531
        One of index scans in this index_merge is more expensive than entire
3532 3533 3534
        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.
3535
      */
monty@mysql.com's avatar
monty@mysql.com committed
3536
      imerge_too_expensive= TRUE;
3537 3538 3539
    }
    if (imerge_too_expensive)
      continue;
3540

3541 3542 3543
    imerge_cost += (*cur_child)->read_cost;
    all_scans_ror_able &= ((*ptree)->n_ror_scans > 0);
    all_scans_rors &= (*cur_child)->is_ror;
3544
    if (pk_is_clustered &&
3545 3546
        param->real_keynr[(*cur_child)->key_idx] ==
        param->table->s->primary_key)
3547
    {
3548 3549
      cpk_scan= cur_child;
      cpk_scan_records= (*cur_child)->records;
3550 3551
    }
    else
3552
      non_cpk_scan_records += (*cur_child)->records;
3553
  }
3554

3555
  DBUG_PRINT("info", ("index_merge scans cost=%g", imerge_cost));
3556
  if (imerge_too_expensive || (imerge_cost > read_time) ||
3557
      (non_cpk_scan_records+cpk_scan_records >= param->table->file->stats.records) &&
3558
      read_time != DBL_MAX)
3559
  {
3560 3561
    /*
      Bail out if it is obvious that both index_merge and ROR-union will be
3562
      more expensive
3563
    */
3564 3565
    DBUG_PRINT("info", ("Sum of index_merge scans is more expensive than "
                        "full table scan, bailing out"));
3566
    DBUG_RETURN(NULL);
3567
  }
3568
  if (all_scans_rors)
3569
  {
3570 3571
    roru_read_plans= (TABLE_READ_PLAN**)range_scans;
    goto skip_to_ror_scan;
3572
  }
3573 3574
  if (cpk_scan)
  {
3575 3576
    /*
      Add one ROWID comparison for each row retrieved on non-CPK scan.  (it
3577 3578 3579
      is done in QUICK_RANGE_SELECT::row_in_ranges)
     */
    imerge_cost += non_cpk_scan_records / TIME_FOR_COMPARE_ROWID;
3580 3581 3582
  }

  /* Calculate cost(rowid_to_row_scan) */
3583
  imerge_cost += get_sweep_read_cost(param, non_cpk_scan_records);
3584
  DBUG_PRINT("info",("index_merge cost with rowid-to-row scan: %g",
3585
                     imerge_cost));
3586 3587
  if (imerge_cost > read_time)
    goto build_ror_index_merge;
3588 3589

  /* Add Unique operations cost */
3590 3591
  unique_calc_buff_size=
    Unique::get_cost_calc_buff_size(non_cpk_scan_records,
3592 3593 3594 3595 3596 3597
                                    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)))
3598
      DBUG_RETURN(NULL);
3599 3600 3601
    param->imerge_cost_buff_size= unique_calc_buff_size;
  }

3602
  imerge_cost +=
3603
    Unique::get_use_cost(param->imerge_cost_buff, non_cpk_scan_records,
3604 3605
                         param->table->file->ref_length,
                         param->thd->variables.sortbuff_size);
3606
  DBUG_PRINT("info",("index_merge total cost: %g (wanted: less then %g)",
3607 3608 3609 3610 3611 3612 3613
                     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;
3614
      imerge_trp->records= min(imerge_trp->records,
3615
                               param->table->file->stats.records);
3616 3617 3618 3619 3620
      imerge_trp->range_scans= range_scans;
      imerge_trp->range_scans_end= range_scans + n_child_scans;
      read_time= imerge_cost;
    }
  }
3621

3622
build_ror_index_merge:
3623 3624
  if (!all_scans_ror_able || param->thd->lex->sql_command == SQLCOM_DELETE)
    DBUG_RETURN(imerge_trp);
3625

3626 3627
  /* Ok, it is possible to build a ROR-union, try it. */
  bool dummy;
3628
  if (!(roru_read_plans=
3629 3630 3631 3632 3633 3634 3635 3636 3637 3638 3639 3640 3641
          (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++)
3642
  {
3643 3644
    /*
      Assume the best ROR scan is the one that has cheapest full-row-retrieval
3645 3646
      scan cost.
      Also accumulate index_only scan costs as we'll need them to calculate
3647 3648 3649 3650 3651 3652 3653
      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->
3654
              read_time(param->real_keynr[(*cur_child)->key_idx], 1,
3655 3656 3657 3658 3659 3660 3661
                        (*cur_child)->records) +
              rows2double((*cur_child)->records) / TIME_FOR_COMPARE;
    }
    else
      cost= read_time;

    TABLE_READ_PLAN *prev_plan= *cur_child;
3662
    if (!(*cur_roru_plan= get_best_ror_intersect(param, *ptree, cost,
3663 3664 3665 3666 3667 3668 3669 3670 3671
                                                 &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
3672 3673
      roru_index_costs +=
        ((TRP_ROR_INTERSECT*)(*cur_roru_plan))->index_scan_costs;
3674
    roru_total_records += (*cur_roru_plan)->records;
3675
    roru_intersect_part *= (*cur_roru_plan)->records /
3676
                           param->table->file->stats.records;
3677
  }
3678

3679 3680
  /*
    rows to retrieve=
3681
      SUM(rows_in_scan_i) - table_rows * PROD(rows_in_scan_i / table_rows).
3682
    This is valid because index_merge construction guarantees that conditions
3683 3684 3685
    in disjunction do not share key parts.
  */
  roru_total_records -= (ha_rows)(roru_intersect_part*
3686
                                  param->table->file->stats.records);
3687 3688
  /* ok, got a ROR read plan for each of the disjuncts
    Calculate cost:
3689 3690 3691 3692 3693 3694
    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.
  */
3695

3696
  double roru_total_cost;
3697 3698 3699
  roru_total_cost= roru_index_costs +
                   rows2double(roru_total_records)*log((double)n_child_scans) /
                   (TIME_FOR_COMPARE_ROWID * M_LN2) +
3700 3701
                   get_sweep_read_cost(param, roru_total_records);

3702
  DBUG_PRINT("info", ("ROR-union: cost %g, %d members", roru_total_cost,
3703 3704 3705 3706 3707 3708 3709 3710 3711 3712 3713 3714 3715 3716
                      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);
3717 3718 3719 3720 3721 3722 3723
}


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

  SYNOPSIS
3724
    get_index_only_read_time()
3725 3726 3727 3728 3729
      param    parameters structure
      records  #of records to read
      keynr    key to read

  NOTES
3730
    It is assumed that we will read trough the whole key range and that all
3731 3732 3733 3734
    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.
3735 3736 3737 3738 3739 3740

  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)
3741 3742
*/

3743
static double get_index_only_read_time(const PARAM* param, ha_rows records,
3744
                                       int keynr)
3745 3746
{
  double read_time;
3747
  uint keys_per_block= (param->table->file->stats.block_size/2/
3748 3749 3750 3751
			(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);
3752
  return read_time;
3753 3754
}

3755

3756 3757
typedef struct st_ror_scan_info
{
3758 3759 3760 3761 3762
  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. */
3763
  SEL_ARG   *sel_arg;
3764 3765

  /* Fields used in the query and covered by this ROR scan. */
3766 3767
  MY_BITMAP covered_fields;
  uint      used_fields_covered; /* # of set bits in covered_fields */
3768
  int       key_rec_length; /* length of key record (including rowid) */
3769 3770

  /*
3771 3772
    Cost of reading all index records with values in sel_arg intervals set
    (assuming there is no need to access full table records)
3773 3774
  */
  double    index_read_cost;
3775 3776 3777
  uint      first_uncovered_field; /* first unused bit in covered_fields */
  uint      key_components; /* # of parts in the key */
} ROR_SCAN_INFO;
3778 3779 3780


/*
3781
  Create ROR_SCAN_INFO* structure with a single ROR scan on index idx using
3782
  sel_arg set of intervals.
3783

3784 3785
  SYNOPSIS
    make_ror_scan()
3786 3787 3788
      param    Parameter from test_quick_select function
      idx      Index of key in param->keys
      sel_arg  Set of intervals for a given key
3789

3790
  RETURN
3791
    NULL - out of memory
3792
    ROR scan structure containing a scan for {idx, sel_arg}
3793 3794 3795 3796 3797 3798
*/

static
ROR_SCAN_INFO *make_ror_scan(const PARAM *param, int idx, SEL_ARG *sel_arg)
{
  ROR_SCAN_INFO *ror_scan;
3799
  my_bitmap_map *bitmap_buf;
3800 3801
  uint keynr;
  DBUG_ENTER("make_ror_scan");
3802

3803 3804 3805 3806 3807 3808
  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];
3809 3810
  ror_scan->key_rec_length= (param->table->key_info[keynr].key_length +
                             param->table->file->ref_length);
3811 3812
  ror_scan->sel_arg= sel_arg;
  ror_scan->records= param->table->quick_rows[keynr];
3813

3814 3815
  if (!(bitmap_buf= (my_bitmap_map*) alloc_root(param->mem_root,
                                                param->fields_bitmap_size)))
3816
    DBUG_RETURN(NULL);
3817

3818
  if (bitmap_init(&ror_scan->covered_fields, bitmap_buf,
3819
                  param->table->s->fields, FALSE))
3820 3821
    DBUG_RETURN(NULL);
  bitmap_clear_all(&ror_scan->covered_fields);
3822

3823
  KEY_PART_INFO *key_part= param->table->key_info[keynr].key_part;
3824
  KEY_PART_INFO *key_part_end= key_part +
3825 3826 3827
                               param->table->key_info[keynr].key_parts;
  for (;key_part != key_part_end; ++key_part)
  {
3828 3829
    if (bitmap_is_set(&param->needed_fields, key_part->fieldnr-1))
      bitmap_set_bit(&ror_scan->covered_fields, key_part->fieldnr-1);
3830
  }
3831
  ror_scan->index_read_cost=
3832 3833 3834 3835 3836 3837
    get_index_only_read_time(param, param->table->quick_rows[ror_scan->keynr],
                             ror_scan->keynr);
  DBUG_RETURN(ror_scan);
}


3838
/*
3839 3840 3841 3842 3843 3844 3845
  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
3846
   -1 a < b
3847 3848
    0 a = b
    1 a > b
3849
*/
3850

3851
static int cmp_ror_scan_info(ROR_SCAN_INFO** a, ROR_SCAN_INFO** b)
3852 3853 3854 3855 3856 3857 3858
{
  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;
}

/*
3859 3860 3861
  Compare two ROR_SCAN_INFO** by
   (#covered fields in F desc,
    #components asc,
3862
    number of first not covered component asc)
3863 3864 3865 3866 3867 3868 3869

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

  RETURN
3870
   -1 a < b
3871 3872
    0 a = b
    1 a > b
3873
*/
3874

3875
static int cmp_ror_scan_info_covering(ROR_SCAN_INFO** a, ROR_SCAN_INFO** b)
3876 3877 3878 3879 3880 3881 3882 3883 3884 3885 3886 3887 3888 3889 3890 3891
{
  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;
}

3892

3893
/* Auxiliary structure for incremental ROR-intersection creation */
3894
typedef struct
3895 3896 3897
{
  const PARAM *param;
  MY_BITMAP covered_fields; /* union of fields covered by all scans */
3898
  /*
3899
    Fraction of table records that satisfies conditions of all scans.
3900
    This is the number of full records that will be retrieved if a
3901 3902
    non-index_only index intersection will be employed.
  */
3903 3904 3905 3906
  double out_rows;
  /* TRUE if covered_fields is a superset of needed_fields */
  bool is_covering;

3907
  ha_rows index_records; /* sum(#records to look in indexes) */
3908 3909
  double index_scan_costs; /* SUM(cost of 'index-only' scans) */
  double total_cost;
3910
} ROR_INTERSECT_INFO;
3911 3912


3913 3914 3915 3916
/*
  Allocate a ROR_INTERSECT_INFO and initialize it to contain zero scans.

  SYNOPSIS
3917 3918 3919
    ror_intersect_init()
      param         Parameter from test_quick_select

3920 3921 3922 3923 3924 3925
  RETURN
    allocated structure
    NULL on error
*/

static
3926
ROR_INTERSECT_INFO* ror_intersect_init(const PARAM *param)
3927 3928
{
  ROR_INTERSECT_INFO *info;
3929
  my_bitmap_map* buf;
3930
  if (!(info= (ROR_INTERSECT_INFO*)alloc_root(param->mem_root,
3931 3932 3933
                                              sizeof(ROR_INTERSECT_INFO))))
    return NULL;
  info->param= param;
3934 3935
  if (!(buf= (my_bitmap_map*) alloc_root(param->mem_root,
                                         param->fields_bitmap_size)))
3936
    return NULL;
3937
  if (bitmap_init(&info->covered_fields, buf, param->table->s->fields,
monty@mysql.com's avatar
monty@mysql.com committed
3938
                  FALSE))
3939
    return NULL;
3940
  info->is_covering= FALSE;
3941
  info->index_scan_costs= 0.0;
3942
  info->index_records= 0;
3943
  info->out_rows= param->table->file->stats.records;
3944
  bitmap_clear_all(&info->covered_fields);
3945 3946 3947
  return info;
}

3948 3949 3950 3951
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, 
3952
         no_bytes_in_map(&src->covered_fields));
3953 3954 3955 3956 3957 3958
  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;
}
3959 3960


3961
/*
3962
  Get selectivity of a ROR scan wrt ROR-intersection.
3963

3964
  SYNOPSIS
3965 3966 3967 3968
    ror_scan_selectivity()
      info  ROR-interection 
      scan  ROR scan
      
3969
  NOTES
3970
    Suppose we have a condition on several keys
3971 3972
    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
3973
          ...
3974
         k_n1=c_n1 AND k_n3=c_n3 AND ...  (1) //parts of the key used by *scan
3975

3976 3977
    where k_ij may be the same as any k_pq (i.e. keys may have common parts).

3978
    A full row is retrieved if entire condition holds.
3979 3980

    The recursive procedure for finding P(cond) is as follows:
3981

3982
    First step:
3983
    Pick 1st part of 1st key and break conjunction (1) into two parts:
3984 3985
      cond= (k_11=c_11 AND R)

3986
    Here R may still contain condition(s) equivalent to k_11=c_11.
3987 3988
    Nevertheless, the following holds:

3989
      P(k_11=c_11 AND R) = P(k_11=c_11) * P(R | k_11=c_11).
3990 3991 3992 3993 3994

    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:
3995
    We have a set of fixed fields/satisfied conditions) F, probability P(F),
3996 3997 3998
    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).
3999
    Lets denote k_ij as t,  R = t AND R1, where R1 may still contain t. Then
4000

4001
     P((t AND R1)|F) = P(t|F) * P(R1|t|F) = P(t|F) * P(R1|(t AND F)) (2)
4002 4003 4004 4005 4006 4007 4008

    (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

4009 4010
    b) F doesn't contain condition on field used in t. Then F and t are
     considered independent.
4011

4012
     P(t|F) = P(t|(fields_before_t_in_key AND other_fields)) =
4013 4014
          = P(t|fields_before_t_in_key).

4015 4016
     P(t|fields_before_t_in_key) = #records(fields_before_t_in_key) /
                                   #records(fields_before_t_in_key, t)
4017 4018

    The second multiplier is calculated by applying this step recursively.
4019

4020 4021 4022 4023 4024
  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.

4025
    The calculation is conducted as follows:
4026
    Lets denote #records(keypart1, ... keypartK) as n_k. We need to calculate
4027

4028 4029
     n_{k1}      n_{k_2}
    --------- * ---------  * .... (3)
4030
     n_{k1-1}    n_{k2_1}
4031

4032 4033 4034 4035
    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
4036
    as fixed, we calculate (3) as
4037 4038 4039

                                  n_{i1}      n_{i_2}
    (3) = n_{max_key_part}  / (   --------- * ---------  * ....  )
4040 4041 4042 4043
                                  n_{i1-1}    n_{i2_1}

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

4044 4045
    In order to minimize number of expensive records_in_range calls we group
    and reduce adjacent fractions.
4046

4047
  RETURN
4048 4049
    Selectivity of given ROR scan.
    
4050 4051
*/

4052 4053
static double ror_scan_selectivity(const ROR_INTERSECT_INFO *info, 
                                   const ROR_SCAN_INFO *scan)
4054 4055
{
  double selectivity_mult= 1.0;
4056
  KEY_PART_INFO *key_part= info->param->table->key_info[scan->keynr].key_part;
sergefp@mysql.com's avatar
sergefp@mysql.com committed
4057
  byte key_val[MAX_KEY_LENGTH+MAX_FIELD_WIDTH]; /* key values tuple */
4058
  char *key_ptr= (char*) key_val;
4059 4060
  SEL_ARG *sel_arg, *tuple_arg= NULL;
  bool cur_covered;
4061
  bool prev_covered= test(bitmap_is_set(&info->covered_fields,
4062
                                        key_part->fieldnr-1));
sergefp@mysql.com's avatar
sergefp@mysql.com committed
4063 4064 4065 4066 4067 4068
  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;
4069
  ha_rows prev_records= info->param->table->file->stats.records;
4070
  DBUG_ENTER("ror_intersect_selectivity");
4071 4072 4073

  for (sel_arg= scan->sel_arg; sel_arg;
       sel_arg= sel_arg->next_key_part)
4074
  {
4075
    DBUG_PRINT("info",("sel_arg step"));
4076
    cur_covered= test(bitmap_is_set(&info->covered_fields,
4077
                                    key_part[sel_arg->part].fieldnr-1));
4078
    if (cur_covered != prev_covered)
4079
    {
4080
      /* create (part1val, ..., part{n-1}val) tuple. */
4081 4082
      ha_rows records;
      if (!tuple_arg)
4083
      {
4084 4085
        tuple_arg= scan->sel_arg;
        /* Here we use the length of the first key part */
4086
        tuple_arg->store_min(key_part->store_length, &key_ptr, 0);
4087 4088 4089 4090
      }
      while (tuple_arg->next_key_part != sel_arg)
      {
        tuple_arg= tuple_arg->next_key_part;
4091
        tuple_arg->store_min(key_part[tuple_arg->part].store_length, &key_ptr, 0);
4092
      }
4093
      min_range.length= max_range.length= ((char*) key_ptr - (char*) key_val);
4094 4095
      records= (info->param->table->file->
                records_in_range(scan->keynr, &min_range, &max_range));
4096 4097 4098 4099 4100 4101 4102 4103 4104 4105 4106
      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 */
4107
        prev_records= records;
4108
      }
4109
    }
4110 4111 4112 4113
    prev_covered= cur_covered;
  }
  if (!prev_covered)
  {
4114
    double tmp= rows2double(info->param->table->quick_rows[scan->keynr]) /
4115 4116
                rows2double(prev_records);
    DBUG_PRINT("info", ("Selectivity multiplier: %g", tmp));
4117
    selectivity_mult *= tmp;
4118
  }
4119 4120 4121
  DBUG_PRINT("info", ("Returning multiplier: %g", selectivity_mult));
  DBUG_RETURN(selectivity_mult);
}
4122

4123

4124 4125 4126 4127 4128 4129 4130 4131 4132 4133 4134 4135 4136 4137 4138 4139 4140 4141 4142 4143 4144 4145 4146 4147 4148 4149 4150 4151 4152 4153 4154 4155 4156 4157 4158 4159 4160
/*
  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,
4161
                              ROR_SCAN_INFO* ror_scan, bool is_cpk_scan)
4162 4163 4164 4165 4166 4167 4168 4169 4170 4171
{
  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);
4172 4173 4174
  if (selectivity_mult == 1.0)
  {
    /* Don't add this scan if it doesn't improve selectivity. */
4175
    DBUG_PRINT("info", ("The scan doesn't improve selectivity."));
4176
    DBUG_RETURN(FALSE);
4177
  }
4178 4179 4180 4181
  
  info->out_rows *= selectivity_mult;
  DBUG_PRINT("info", ("info->total_cost= %g", info->total_cost));
  
4182
  if (is_cpk_scan)
4183
  {
4184 4185 4186 4187 4188 4189
    /*
      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) / 
4190 4191 4192 4193
                              TIME_FOR_COMPARE_ROWID;
  }
  else
  {
4194
    info->index_records += info->param->table->quick_rows[ror_scan->keynr];
4195 4196
    info->index_scan_costs += ror_scan->index_read_cost;
    bitmap_union(&info->covered_fields, &ror_scan->covered_fields);
4197 4198 4199 4200 4201 4202
    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;
    }
4203
  }
4204

4205
  info->total_cost= info->index_scan_costs;
4206
  DBUG_PRINT("info", ("info->total_cost: %g", info->total_cost));
4207 4208
  if (!info->is_covering)
  {
4209 4210 4211
    info->total_cost += 
      get_sweep_read_cost(info->param, double2rows(info->out_rows));
    DBUG_PRINT("info", ("info->total_cost= %g", info->total_cost));
4212
  }
4213 4214
  DBUG_PRINT("info", ("New out_rows: %g", info->out_rows));
  DBUG_PRINT("info", ("New cost: %g, %scovering", info->total_cost,
4215
                      info->is_covering?"" : "non-"));
4216
  DBUG_RETURN(TRUE);
4217 4218
}

4219

4220 4221
/*
  Get best ROR-intersection plan using non-covering ROR-intersection search
4222 4223 4224 4225
  algorithm. The returned plan may be covering.

  SYNOPSIS
    get_best_ror_intersect()
4226 4227 4228
      param            Parameter from test_quick_select function.
      tree             Transformed restriction condition to be used to look
                       for ROR scans.
4229
      read_time        Do not return read plans with cost > read_time.
4230
      are_all_covering [out] set to TRUE if union of all scans covers all
4231 4232
                       fields needed by the query (and it is possible to build
                       a covering ROR-intersection)
4233

4234
  NOTES
4235 4236 4237 4238 4239
    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.
4240

4241
  IMPLEMENTATION
4242
    The approximate best non-covering plan search algorithm is as follows:
4243

4244 4245 4246 4247
    find_min_ror_intersection_scan()
    {
      R= select all ROR scans;
      order R by (E(#records_matched) * key_record_length).
4248

4249 4250 4251 4252 4253 4254
      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)
      {
4255 4256
        firstR= R - first(R);
        if (!selectivity(S + firstR < selectivity(S)))
4257
          continue;
4258
          
4259 4260 4261 4262 4263 4264 4265 4266 4267
        S= S + first(R);
        if (cost(S) < min_cost)
        {
          min_cost= cost(S);
          min_scan= make_scan(S);
        }
      }
      return min_scan;
    }
4268

4269
    See ror_intersect_add function for ROR intersection costs.
4270

4271
    Special handling for Clustered PK scans
4272 4273
    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
4274 4275
    expensive in this case.
    Clustered PK scan has special handling in ROR-intersection: it is not used
4276
    to retrieve rows, instead its condition is used to filter row references
4277
    we get from scans on other keys.
4278 4279

  RETURN
4280
    ROR-intersection table read plan
4281
    NULL if out of memory or no suitable plan found.
4282 4283
*/

4284 4285 4286 4287 4288 4289
static
TRP_ROR_INTERSECT *get_best_ror_intersect(const PARAM *param, SEL_TREE *tree,
                                          double read_time,
                                          bool *are_all_covering)
{
  uint idx;
4290
  double min_cost= DBL_MAX;
4291
  DBUG_ENTER("get_best_ror_intersect");
4292

4293
  if ((tree->n_ror_scans < 2) || !param->table->file->stats.records)
4294
    DBUG_RETURN(NULL);
4295 4296

  /*
4297 4298
    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.
4299
  */
4300
  ROR_SCAN_INFO **cur_ror_scan;
4301
  ROR_SCAN_INFO *cpk_scan= NULL;
4302
  uint cpk_no;
monty@mysql.com's avatar
monty@mysql.com committed
4303
  bool cpk_scan_used= FALSE;
4304

4305 4306 4307 4308
  if (!(tree->ror_scans= (ROR_SCAN_INFO**)alloc_root(param->mem_root,
                                                     sizeof(ROR_SCAN_INFO*)*
                                                     param->keys)))
    return NULL;
4309 4310
  cpk_no= ((param->table->file->primary_key_is_clustered()) ?
           param->table->s->primary_key : MAX_KEY);
4311

4312
  for (idx= 0, cur_ror_scan= tree->ror_scans; idx < param->keys; idx++)
4313
  {
4314
    ROR_SCAN_INFO *scan;
4315
    if (!tree->ror_scans_map.is_set(idx))
4316
      continue;
4317
    if (!(scan= make_ror_scan(param, idx, tree->keys[idx])))
4318
      return NULL;
4319
    if (param->real_keynr[idx] == cpk_no)
4320
    {
4321 4322
      cpk_scan= scan;
      tree->n_ror_scans--;
4323 4324
    }
    else
4325
      *(cur_ror_scan++)= scan;
4326
  }
4327

4328
  tree->ror_scans_end= cur_ror_scan;
4329 4330
  DBUG_EXECUTE("info",print_ror_scans_arr(param->table, "original",
                                          tree->ror_scans,
4331 4332
                                          tree->ror_scans_end););
  /*
4333
    Ok, [ror_scans, ror_scans_end) is array of ptrs to initialized
4334 4335
    ROR_SCAN_INFO's.
    Step 2: Get best ROR-intersection using an approximate algorithm.
4336 4337 4338
  */
  qsort(tree->ror_scans, tree->n_ror_scans, sizeof(ROR_SCAN_INFO*),
        (qsort_cmp)cmp_ror_scan_info);
4339 4340
  DBUG_EXECUTE("info",print_ror_scans_arr(param->table, "ordered",
                                          tree->ror_scans,
4341
                                          tree->ror_scans_end););
4342

4343 4344 4345 4346 4347 4348 4349 4350 4351
  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. */
4352 4353 4354
  ROR_INTERSECT_INFO *intersect, *intersect_best;
  if (!(intersect= ror_intersect_init(param)) || 
      !(intersect_best= ror_intersect_init(param)))
4355
    return NULL;
4356

4357
  /* [intersect_scans,intersect_scans_best) will hold the best intersection */
4358
  ROR_SCAN_INFO **intersect_scans_best;
4359
  cur_ror_scan= tree->ror_scans;
4360
  intersect_scans_best= intersect_scans;
4361
  while (cur_ror_scan != tree->ror_scans_end && !intersect->is_covering)
4362
  {
4363
    /* S= S + first(R);  R= R - first(R); */
4364
    if (!ror_intersect_add(intersect, *cur_ror_scan, FALSE))
4365 4366 4367 4368 4369 4370
    {
      cur_ror_scan++;
      continue;
    }
    
    *(intersect_scans_end++)= *(cur_ror_scan++);
4371

4372
    if (intersect->total_cost < min_cost)
4373
    {
4374
      /* Local minimum found, save it */
4375
      ror_intersect_cpy(intersect_best, intersect);
4376
      intersect_scans_best= intersect_scans_end;
4377
      min_cost = intersect->total_cost;
4378 4379
    }
  }
4380

4381 4382 4383 4384 4385 4386
  if (intersect_scans_best == intersect_scans)
  {
    DBUG_PRINT("info", ("None of scans increase selectivity"));
    DBUG_RETURN(NULL);
  }
    
4387 4388 4389 4390
  DBUG_EXECUTE("info",print_ror_scans_arr(param->table,
                                          "best ROR-intersection",
                                          intersect_scans,
                                          intersect_scans_best););
4391

4392
  *are_all_covering= intersect->is_covering;
4393
  uint best_num= intersect_scans_best - intersect_scans;
4394 4395
  ror_intersect_cpy(intersect, intersect_best);

4396 4397
  /*
    Ok, found the best ROR-intersection of non-CPK key scans.
4398 4399
    Check if we should add a CPK scan. If the obtained ROR-intersection is 
    covering, it doesn't make sense to add CPK scan.
4400 4401
  */
  if (cpk_scan && !intersect->is_covering)
4402
  {
4403
    if (ror_intersect_add(intersect, cpk_scan, TRUE) && 
4404
        (intersect->total_cost < min_cost))
4405
    {
monty@mysql.com's avatar
monty@mysql.com committed
4406
      cpk_scan_used= TRUE;
4407
      intersect_best= intersect; //just set pointer here
4408 4409
    }
  }
4410

4411
  /* Ok, return ROR-intersect plan if we have found one */
4412
  TRP_ROR_INTERSECT *trp= NULL;
4413
  if (min_cost < read_time && (cpk_scan_used || best_num > 1))
4414
  {
4415 4416
    if (!(trp= new (param->mem_root) TRP_ROR_INTERSECT))
      DBUG_RETURN(trp);
4417 4418
    if (!(trp->first_scan=
           (ROR_SCAN_INFO**)alloc_root(param->mem_root,
4419 4420 4421 4422
                                       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;
4423 4424 4425 4426 4427 4428
    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;
4429
    set_if_smaller(param->table->quick_condition_rows, best_rows);
4430
    trp->records= best_rows;
4431 4432 4433 4434 4435
    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));
4436
  }
4437
  DBUG_RETURN(trp);
4438 4439 4440 4441
}


/*
4442
  Get best covering ROR-intersection.
4443
  SYNOPSIS
4444
    get_best_covering_ror_intersect()
4445 4446 4447
      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.
4448

4449 4450
  RETURN
    Best covering ROR-intersection plan
4451
    NULL if no plan found.
4452 4453

  NOTES
4454
    get_best_ror_intersect must be called for a tree before calling this
4455
    function for it.
4456
    This function invalidates tree->ror_scans member values.
4457

4458 4459 4460 4461 4462
  The following approximate algorithm is used:
    I=set of all covering indexes
    F=set of all fields to cover
    S={}

4463 4464
    do
    {
4465 4466 4467 4468 4469 4470 4471
      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.
4472 4473
*/

4474
static
4475 4476
TRP_ROR_INTERSECT *get_best_covering_ror_intersect(PARAM *param,
                                                   SEL_TREE *tree,
4477
                                                   double read_time)
4478
{
4479
  ROR_SCAN_INFO **ror_scan_mark;
4480
  ROR_SCAN_INFO **ror_scans_end= tree->ror_scans_end;
4481 4482 4483
  DBUG_ENTER("get_best_covering_ror_intersect");

  for (ROR_SCAN_INFO **scan= tree->ror_scans; scan != ror_scans_end; ++scan)
4484
    (*scan)->key_components=
4485
      param->table->key_info[(*scan)->keynr].key_parts;
4486

4487 4488
  /*
    Run covering-ROR-search algorithm.
4489
    Assume set I is [ror_scan .. ror_scans_end)
4490
  */
4491

4492 4493
  /*I=set of all covering indexes */
  ror_scan_mark= tree->ror_scans;
4494

4495 4496
  MY_BITMAP *covered_fields= &param->tmp_covered_fields;
  if (!covered_fields->bitmap) 
4497
    covered_fields->bitmap= (my_bitmap_map*)alloc_root(param->mem_root,
4498 4499
                                               param->fields_bitmap_size);
  if (!covered_fields->bitmap ||
4500 4501
      bitmap_init(covered_fields, covered_fields->bitmap,
                  param->table->s->fields, FALSE))
4502
    DBUG_RETURN(0);
4503
  bitmap_clear_all(covered_fields);
4504 4505 4506

  double total_cost= 0.0f;
  ha_rows records=0;
4507 4508
  bool all_covered;

4509 4510 4511 4512
  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););
4513 4514
  do
  {
4515
    /*
4516
      Update changed sorting info:
4517
        #covered fields,
4518
	number of first not covered component
4519 4520 4521 4522
      Calculate and save these values for each of remaining scans.
    */
    for (ROR_SCAN_INFO **scan= ror_scan_mark; scan != ror_scans_end; ++scan)
    {
4523
      bitmap_subtract(&(*scan)->covered_fields, covered_fields);
4524
      (*scan)->used_fields_covered=
4525
        bitmap_bits_set(&(*scan)->covered_fields);
4526
      (*scan)->first_uncovered_field=
4527 4528 4529 4530 4531 4532 4533 4534 4535
        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););
4536

4537 4538 4539
    /* I=I-first(I) */
    total_cost += (*ror_scan_mark)->index_read_cost;
    records += (*ror_scan_mark)->records;
4540
    DBUG_PRINT("info", ("Adding scan on %s",
4541 4542 4543 4544
                        param->table->key_info[(*ror_scan_mark)->keynr].name));
    if (total_cost > read_time)
      DBUG_RETURN(NULL);
    /* F=F-covered by first(I) */
4545 4546
    bitmap_union(covered_fields, &(*ror_scan_mark)->covered_fields);
    all_covered= bitmap_is_subset(&param->needed_fields, covered_fields);
4547 4548 4549 4550
  } while ((++ror_scan_mark < ror_scans_end) && !all_covered);
  
  if (!all_covered || (ror_scan_mark - tree->ror_scans) == 1)
    DBUG_RETURN(NULL);
4551 4552 4553 4554 4555 4556 4557 4558 4559

  /*
    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););
4560

4561
  /* Add priority queue use cost. */
4562 4563
  total_cost += rows2double(records)*
                log((double)(ror_scan_mark - tree->ror_scans)) /
4564 4565 4566 4567 4568 4569 4570 4571 4572 4573 4574 4575 4576 4577
                (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);
4578
  memcpy(trp->first_scan, tree->ror_scans, best_num*sizeof(ROR_SCAN_INFO*));
4579
  trp->last_scan=  trp->first_scan + best_num;
monty@mysql.com's avatar
monty@mysql.com committed
4580
  trp->is_covering= TRUE;
4581 4582
  trp->read_cost= total_cost;
  trp->records= records;
4583
  trp->cpk_scan= NULL;
4584
  set_if_smaller(param->table->quick_condition_rows, records); 
4585

4586 4587 4588
  DBUG_PRINT("info",
             ("Returning covering ROR-intersect plan: cost %g, records %lu",
              trp->read_cost, (ulong) trp->records));
4589
  DBUG_RETURN(trp);
4590 4591 4592
}


4593
/*
4594
  Get best "range" table read plan for given SEL_TREE.
4595
  Also update PARAM members and store ROR scans info in the SEL_TREE.
4596
  SYNOPSIS
4597
    get_key_scans_params
4598
      param        parameters from test_quick_select
4599
      tree         make range select for this SEL_TREE
monty@mysql.com's avatar
monty@mysql.com committed
4600
      index_read_must_be_used if TRUE, assume 'index only' option will be set
4601
                             (except for clustered PK indexes)
4602 4603
      read_time    don't create read plans with cost > read_time.
  RETURN
4604
    Best range read plan
4605
    NULL if no plan found or error occurred
4606 4607
*/

4608
static TRP_RANGE *get_key_scans_params(PARAM *param, SEL_TREE *tree,
4609 4610
                                       bool index_read_must_be_used, 
                                       bool update_tbl_stats,
4611
                                       double read_time)
4612 4613
{
  int idx;
4614 4615 4616
  SEL_ARG **key,**end, **key_to_read= NULL;
  ha_rows best_records;
  TRP_RANGE* read_plan= NULL;
4617
  bool pk_is_clustered= param->table->file->primary_key_is_clustered();
4618 4619
  DBUG_ENTER("get_key_scans_params");
  LINT_INIT(best_records); /* protected by key_to_read */
4620
  /*
4621 4622
    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
4623
    is defined as "not null".
4624 4625
  */
  DBUG_EXECUTE("info", print_sel_tree(param, tree, &tree->keys_map,
4626 4627 4628 4629
                                      "tree scans"););
  tree->ror_scans_map.clear_all();
  tree->n_ror_scans= 0;
  for (idx= 0,key=tree->keys, end=key+param->keys;
4630 4631 4632 4633 4634 4635 4636
       key != end ;
       key++,idx++)
  {
    ha_rows found_records;
    double found_read_time;
    if (*key)
    {
4637
      uint keynr= param->real_keynr[idx];
4638 4639
      if ((*key)->type == SEL_ARG::MAYBE_KEY ||
          (*key)->maybe_flag)
4640
        param->needed_reg->set_bit(keynr);
4641

monty@mysql.com's avatar
monty@mysql.com committed
4642 4643
      bool read_index_only= index_read_must_be_used ? TRUE :
                            (bool) param->table->used_keys.is_set(keynr);
4644

4645
      found_records= check_quick_select(param, idx, *key, update_tbl_stats);
4646 4647 4648 4649 4650
      if (param->is_ror_scan)
      {
        tree->n_ror_scans++;
        tree->ror_scans_map.set_bit(idx);
      }
4651
      double cpu_cost= (double) found_records / TIME_FOR_COMPARE;
4652
      if (found_records != HA_POS_ERROR && found_records > 2 &&
sergefp@mysql.com's avatar
sergefp@mysql.com committed
4653
          read_index_only &&
monty@mysql.com's avatar
monty@mysql.com committed
4654
          (param->table->file->index_flags(keynr, param->max_key_part,1) &
monty@mysql.com's avatar
monty@mysql.com committed
4655
           HA_KEYREAD_ONLY) &&
4656
          !(pk_is_clustered && keynr == param->table->s->primary_key))
4657 4658 4659 4660 4661
      {
        /*
          We can resolve this by only reading through this key. 
          0.01 is added to avoid races between range and 'index' scan.
        */
4662
        found_read_time= get_index_only_read_time(param,found_records,keynr) +
4663 4664
                         cpu_cost + 0.01;
      }
4665
      else
4666
      {
4667
        /*
4668 4669 4670
          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.
        */
4671 4672 4673
	found_read_time= param->table->file->read_time(keynr,
                                                       param->range_count,
                                                       found_records) +
4674 4675
			 cpu_cost + 0.01;
      }
4676 4677 4678
      DBUG_PRINT("info",("key %s: found_read_time: %g (cur. read_time: %g)",
                         param->table->key_info[keynr].name, found_read_time,
                         read_time));
4679

4680 4681
      if (read_time > found_read_time && found_records != HA_POS_ERROR
          /*|| read_time == DBL_MAX*/ )
4682
      {
4683
        read_time=    found_read_time;
4684
        best_records= found_records;
4685 4686 4687 4688 4689 4690 4691 4692 4693 4694 4695 4696 4697 4698 4699 4700
        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;
4701 4702 4703 4704
      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));
4705 4706 4707 4708 4709 4710 4711 4712 4713
    }
  }
  else
    DBUG_PRINT("info", ("No 'range' table read plan found"));

  DBUG_RETURN(read_plan);
}


4714
QUICK_SELECT_I *TRP_INDEX_MERGE::make_quick(PARAM *param,
4715 4716 4717 4718 4719 4720 4721 4722 4723 4724 4725
                                            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;
4726 4727
  for (TRP_RANGE **range_scan= range_scans; range_scan != range_scans_end;
       range_scan++)
4728 4729
  {
    if (!(quick= (QUICK_RANGE_SELECT*)
4730
          ((*range_scan)->make_quick(param, FALSE, &quick_imerge->alloc)))||
4731 4732 4733 4734 4735 4736 4737 4738 4739 4740
        quick_imerge->push_quick_back(quick))
    {
      delete quick;
      delete quick_imerge;
      return NULL;
    }
  }
  return quick_imerge;
}

4741
QUICK_SELECT_I *TRP_ROR_INTERSECT::make_quick(PARAM *param,
4742 4743 4744 4745 4746 4747 4748
                                              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;
4749 4750

  if ((quick_intrsect=
4751
         new QUICK_ROR_INTERSECT_SELECT(param->thd, param->table,
4752 4753
                                        (retrieve_full_rows? (!is_covering) :
                                         FALSE),
4754 4755
                                        parent_alloc)))
  {
4756
    DBUG_EXECUTE("info", print_ror_scans_arr(param->table,
4757 4758 4759
                                             "creating ROR-intersect",
                                             first_scan, last_scan););
    alloc= parent_alloc? parent_alloc: &quick_intrsect->alloc;
4760
    for (; first_scan != last_scan;++first_scan)
4761 4762 4763 4764
    {
      if (!(quick= get_quick_select(param, (*first_scan)->idx,
                                    (*first_scan)->sel_arg, alloc)) ||
          quick_intrsect->push_quick_back(quick))
4765
      {
4766 4767
        delete quick_intrsect;
        DBUG_RETURN(NULL);
4768 4769
      }
    }
4770 4771 4772 4773
    if (cpk_scan)
    {
      if (!(quick= get_quick_select(param, cpk_scan->idx,
                                    cpk_scan->sel_arg, alloc)))
4774
      {
4775 4776
        delete quick_intrsect;
        DBUG_RETURN(NULL);
4777
      }
4778
      quick->file= NULL; 
4779
      quick_intrsect->cpk_quick= quick;
4780
    }
4781
    quick_intrsect->records= records;
4782
    quick_intrsect->read_time= read_cost;
4783
  }
4784 4785 4786
  DBUG_RETURN(quick_intrsect);
}

4787

4788
QUICK_SELECT_I *TRP_ROR_UNION::make_quick(PARAM *param,
4789 4790 4791 4792 4793 4794 4795
                                          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");
4796 4797
  /*
    It is impossible to construct a ROR-union that will not retrieve full
4798
    rows, ignore retrieve_full_rows parameter.
4799 4800 4801
  */
  if ((quick_roru= new QUICK_ROR_UNION_SELECT(param->thd, param->table)))
  {
4802
    for (scan= first_ror; scan != last_ror; scan++)
4803
    {
4804
      if (!(quick= (*scan)->make_quick(param, FALSE, &quick_roru->alloc)) ||
4805 4806 4807 4808 4809
          quick_roru->push_quick_back(quick))
        DBUG_RETURN(NULL);
    }
    quick_roru->records= records;
    quick_roru->read_time= read_cost;
4810
  }
4811
  DBUG_RETURN(quick_roru);
4812 4813
}

4814

4815
/*
monty@mysql.com's avatar
monty@mysql.com committed
4816
  Build a SEL_TREE for <> or NOT BETWEEN predicate
4817 4818 4819 4820 4821 4822
 
  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
4823 4824
      lt_value    constant that field should be smaller
      gt_value    constant that field should be greaterr
4825 4826 4827
      cmp_type    compare type for the field

  RETURN 
monty@mysql.com's avatar
monty@mysql.com committed
4828 4829
    #  Pointer to tree built tree
    0  on error
4830 4831
*/

4832
static SEL_TREE *get_ne_mm_tree(RANGE_OPT_PARAM *param, Item_func *cond_func, 
monty@mysql.com's avatar
monty@mysql.com committed
4833 4834
                                Field *field,
                                Item *lt_value, Item *gt_value,
4835 4836
                                Item_result cmp_type)
{
monty@mysql.com's avatar
monty@mysql.com committed
4837
  SEL_TREE *tree;
4838
  tree= get_mm_parts(param, cond_func, field, Item_func::LT_FUNC,
monty@mysql.com's avatar
monty@mysql.com committed
4839
                     lt_value, cmp_type);
4840 4841 4842 4843
  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
4844
					    gt_value, cmp_type));
4845 4846 4847 4848 4849
  }
  return tree;
}
   

igor@rurik.mysql.com's avatar
igor@rurik.mysql.com committed
4850 4851 4852 4853 4854 4855 4856 4857 4858 4859
/*
  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
4860
      inv         TRUE <> NOT cond_func is considered
igor@rurik.mysql.com's avatar
igor@rurik.mysql.com committed
4861
                  (makes sense only when cond_func is BETWEEN or IN) 
igor@rurik.mysql.com's avatar
igor@rurik.mysql.com committed
4862 4863

  RETURN 
igor@rurik.mysql.com's avatar
igor@rurik.mysql.com committed
4864
    Pointer to the tree built tree
igor@rurik.mysql.com's avatar
igor@rurik.mysql.com committed
4865 4866
*/

4867
static SEL_TREE *get_func_mm_tree(RANGE_OPT_PARAM *param, Item_func *cond_func, 
4868
                                  Field *field, Item *value,
4869
                                  Item_result cmp_type, bool inv)
4870 4871 4872 4873
{
  SEL_TREE *tree= 0;
  DBUG_ENTER("get_func_mm_tree");

igor@rurik.mysql.com's avatar
igor@rurik.mysql.com committed
4874
  switch (cond_func->functype()) {
4875

igor@rurik.mysql.com's avatar
igor@rurik.mysql.com committed
4876
  case Item_func::NE_FUNC:
monty@mysql.com's avatar
monty@mysql.com committed
4877
    tree= get_ne_mm_tree(param, cond_func, field, value, value, cmp_type);
igor@rurik.mysql.com's avatar
igor@rurik.mysql.com committed
4878
    break;
4879

igor@rurik.mysql.com's avatar
igor@rurik.mysql.com committed
4880
  case Item_func::BETWEEN:
igor@rurik.mysql.com's avatar
igor@rurik.mysql.com committed
4881
  {
evgen@moonbone.local's avatar
evgen@moonbone.local committed
4882
    if (!value)
4883
    {
igor@rurik.mysql.com's avatar
igor@rurik.mysql.com committed
4884
      if (inv)
4885
      {
igor@rurik.mysql.com's avatar
igor@rurik.mysql.com committed
4886 4887 4888 4889
        tree= get_ne_mm_tree(param, cond_func, field, cond_func->arguments()[1],
                             cond_func->arguments()[2], cmp_type);
      }
      else
4890
      {
igor@rurik.mysql.com's avatar
igor@rurik.mysql.com committed
4891 4892 4893 4894 4895 4896 4897 4898 4899
        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));
        }
4900
      }
4901
    }
igor@rurik.mysql.com's avatar
igor@rurik.mysql.com committed
4902 4903 4904
    else
      tree= get_mm_parts(param, cond_func, field,
                         (inv ?
evgen@sunlight.local's avatar
evgen@sunlight.local committed
4905 4906 4907 4908
                          (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
4909
                         cond_func->arguments()[0], cmp_type);
igor@rurik.mysql.com's avatar
igor@rurik.mysql.com committed
4910
    break;
igor@rurik.mysql.com's avatar
igor@rurik.mysql.com committed
4911
  }
igor@rurik.mysql.com's avatar
igor@rurik.mysql.com committed
4912
  case Item_func::IN_FUNC:
4913 4914
  {
    Item_func_in *func=(Item_func_in*) cond_func;
4915

4916 4917 4918 4919 4920 4921 4922 4923
    /*
      Array for IN() is constructed when all values have the same result
      type. Tree won't be built for values with different result types,
      so we check it here to avoid unnecessary work.
    */
    if (!func->array)
      break;

4924
    if (inv)
4925
    {
4926
      if (func->array->result_type() != ROW_RESULT)
4927
      {
4928
        /*
4929 4930 4931
          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:
4932 4933 4934 4935 4936
          
          ($MIN<t.key<c1) OR (c1<t.key<c2) OR (c2<t.key<c3) OR ...    (*)
          
          where $MIN is either "-inf" or NULL.
          
4937 4938 4939 4940 4941 4942 4943 4944 4945 4946 4947 4948 4949 4950 4951 4952 4953
          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.
4954
        */
4955
#define NOT_IN_IGNORE_THRESHOLD 1000
4956 4957
        MEM_ROOT *tmp_root= param->mem_root;
        param->thd->mem_root= param->old_root;
4958 4959 4960 4961 4962 4963 4964 4965 4966 4967 4968
        /* 
          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;

4969
        if (func->array->count > NOT_IN_IGNORE_THRESHOLD || !value_item)
4970
          break;
4971

4972
        /* Get a SEL_TREE for "(-inf|NULL) < X < c_0" interval.  */
4973 4974 4975 4976 4977 4978 4979 4980 4981 4982 4983 4984 4985 4986 4987
        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;
4988
          break;
4989
        }
4990
        SEL_TREE *tree2;
4991
        for (; i < func->array->count; i++)
4992
        {
4993
          if (func->array->compare_elems(i, i-1))
4994
          {
4995 4996 4997 4998 4999
            /* 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)
5000
            {
5001 5002 5003
              tree= NULL;
              break;
            }
5004

5005 5006 5007 5008 5009 5010
            /* 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;
              if (((new_interval= tree2->keys[idx])) && 
                  ((last_val= tree->keys[idx]->last())))
5011
              {
5012 5013
                new_interval->min_value= last_val->max_value;
                new_interval->min_flag= NEAR_MIN;
5014 5015
              }
            }
5016 5017 5018 5019 5020
            /* 
              The following doesn't try to allocate memory so no need to
              check for NULL.
            */
            tree= tree_or(param, tree, tree2);
5021 5022
          }
        }
5023 5024 5025 5026 5027 5028 5029 5030 5031 5032 5033
        
        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);
        }
5034 5035
      }
      else
5036
      {
5037 5038 5039 5040
        tree= get_ne_mm_tree(param, cond_func, field,
                             func->arguments()[1], func->arguments()[1],
                             cmp_type);
        if (tree)
5041
        {
5042 5043 5044 5045 5046 5047 5048
          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));
          }
5049 5050 5051 5052 5053 5054 5055 5056 5057 5058 5059 5060 5061 5062 5063 5064 5065
        }
      }
    }
    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));
        }
5066 5067
      }
    }
igor@rurik.mysql.com's avatar
igor@rurik.mysql.com committed
5068
    break;
5069
  }
igor@rurik.mysql.com's avatar
igor@rurik.mysql.com committed
5070
  default: 
5071
  {
igor@rurik.mysql.com's avatar
igor@rurik.mysql.com committed
5072 5073 5074 5075 5076 5077 5078
    /* 
       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.
    */
5079 5080 5081
    Item_func::Functype func_type=
      (value != cond_func->arguments()[0]) ? cond_func->functype() :
        ((Item_bool_func2*) cond_func)->rev_functype();
5082
    tree= get_mm_parts(param, cond_func, field, func_type, value, cmp_type);
5083
  }
igor@rurik.mysql.com's avatar
igor@rurik.mysql.com committed
5084 5085
  }

5086 5087
  DBUG_RETURN(tree);
}
5088

igor@rurik.mysql.com's avatar
igor@rurik.mysql.com committed
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 5114 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

/*
  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
*/

5160 5161
static SEL_TREE *get_full_func_mm_tree(RANGE_OPT_PARAM *param,
                                       Item_func *cond_func,
igor@rurik.mysql.com's avatar
igor@rurik.mysql.com committed
5162 5163 5164 5165 5166 5167 5168 5169 5170 5171 5172 5173 5174 5175 5176 5177 5178 5179 5180 5181 5182 5183 5184 5185 5186 5187 5188 5189 5190 5191 5192 5193 5194 5195 5196 5197 5198 5199
                                       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);
5200 5201
}

bk@work.mysql.com's avatar
bk@work.mysql.com committed
5202 5203
	/* make a select tree of all keys in condition */

5204
static SEL_TREE *get_mm_tree(RANGE_OPT_PARAM *param,COND *cond)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
5205 5206
{
  SEL_TREE *tree=0;
5207 5208
  SEL_TREE *ftree= 0;
  Item_field *field_item= 0;
5209
  bool inv= FALSE;
igor@rurik.mysql.com's avatar
igor@rurik.mysql.com committed
5210
  Item *value= 0;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
5211 5212 5213 5214 5215 5216 5217 5218 5219 5220 5221 5222 5223
  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);
5224
	if (param->thd->is_fatal_error)
5225
	  DBUG_RETURN(0);	// out of memory
bk@work.mysql.com's avatar
bk@work.mysql.com committed
5226 5227 5228 5229 5230 5231 5232 5233 5234 5235 5236 5237 5238 5239 5240
	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)
5241
	    DBUG_RETURN(0);	// out of memory
bk@work.mysql.com's avatar
bk@work.mysql.com committed
5242 5243 5244 5245 5246 5247 5248 5249 5250 5251 5252
	  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())
  {
5253 5254 5255 5256 5257 5258 5259 5260 5261 5262 5263 5264
    /*
      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
5265
  }
5266

5267 5268 5269
  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
5270 5271
  if (cond->type() != Item::FUNC_ITEM)
  {						// Should be a field
5272
    ref_tables= cond->used_tables();
5273 5274
    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
5275 5276 5277
      DBUG_RETURN(0);
    DBUG_RETURN(new SEL_TREE(SEL_TREE::MAYBE));
  }
5278

bk@work.mysql.com's avatar
bk@work.mysql.com committed
5279
  Item_func *cond_func= (Item_func*) cond;
5280 5281 5282
  if (cond_func->functype() == Item_func::BETWEEN ||
      cond_func->functype() == Item_func::IN_FUNC)
    inv= ((Item_func_opt_neg *) cond_func)->negated;
5283
  else if (cond_func->select_optimize() == Item_func::OPTIMIZE_NONE)
igor@rurik.mysql.com's avatar
igor@rurik.mysql.com committed
5284
    DBUG_RETURN(0);			       
bk@work.mysql.com's avatar
bk@work.mysql.com committed
5285

igor@rurik.mysql.com's avatar
igor@rurik.mysql.com committed
5286 5287
  param->cond= cond;

igor@rurik.mysql.com's avatar
igor@rurik.mysql.com committed
5288 5289
  switch (cond_func->functype()) {
  case Item_func::BETWEEN:
igor@rurik.mysql.com's avatar
igor@rurik.mysql.com committed
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
    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
5321 5322
    break;
  case Item_func::IN_FUNC:
bk@work.mysql.com's avatar
bk@work.mysql.com committed
5323 5324
  {
    Item_func_in *func=(Item_func_in*) cond_func;
5325
    if (func->key_item()->real_item()->type() != Item::FIELD_ITEM)
5326
      DBUG_RETURN(0);
5327
    field_item= (Item_field*) (func->key_item()->real_item());
igor@rurik.mysql.com's avatar
igor@rurik.mysql.com committed
5328
    ftree= get_full_func_mm_tree(param, cond_func, field_item, NULL, inv);
igor@rurik.mysql.com's avatar
igor@rurik.mysql.com committed
5329
    break;
5330
  }
igor@rurik.mysql.com's avatar
igor@rurik.mysql.com committed
5331
  case Item_func::MULT_EQUAL_FUNC:
bk@work.mysql.com's avatar
bk@work.mysql.com committed
5332
  {
5333 5334
    Item_equal *item_equal= (Item_equal *) cond;    
    if (!(value= item_equal->get_const()))
igor@rurik.mysql.com's avatar
igor@rurik.mysql.com committed
5335 5336 5337 5338
      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
5339
    {
igor@rurik.mysql.com's avatar
igor@rurik.mysql.com committed
5340 5341 5342
      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
5343
      {
5344
        tree= get_mm_parts(param, cond, field, Item_func::EQ_FUNC,
igor@rurik.mysql.com's avatar
igor@rurik.mysql.com committed
5345 5346
		           value,cmp_type);
        ftree= !ftree ? tree : tree_and(param, ftree, tree);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
5347 5348
      }
    }
igor@rurik.mysql.com's avatar
igor@rurik.mysql.com committed
5349
    
5350
    DBUG_RETURN(ftree);
5351 5352
  }
  default:
5353
    if (cond_func->arguments()[0]->real_item()->type() == Item::FIELD_ITEM)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
5354
    {
5355
      field_item= (Item_field*) (cond_func->arguments()[0]->real_item());
5356
      value= cond_func->arg_count > 1 ? cond_func->arguments()[1] : 0;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
5357
    }
5358
    else if (cond_func->have_rev_func() &&
5359 5360
             cond_func->arguments()[1]->real_item()->type() ==
                                                            Item::FIELD_ITEM)
5361
    {
5362
      field_item= (Item_field*) (cond_func->arguments()[1]->real_item());
5363 5364 5365 5366
      value= cond_func->arguments()[0];
    }
    else
      DBUG_RETURN(0);
igor@rurik.mysql.com's avatar
igor@rurik.mysql.com committed
5367
    ftree= get_full_func_mm_tree(param, cond_func, field_item, value, inv);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
5368
  }
5369 5370

  DBUG_RETURN(ftree);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
5371 5372 5373 5374
}


static SEL_TREE *
5375
get_mm_parts(RANGE_OPT_PARAM *param, COND *cond_func, Field *field,
5376
	     Item_func::Functype type,
5377
	     Item *value, Item_result cmp_type)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
5378 5379 5380 5381 5382
{
  DBUG_ENTER("get_mm_parts");
  if (field->table != param->table)
    DBUG_RETURN(0);

5383 5384
  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
5385 5386 5387 5388
  SEL_TREE *tree=0;
  if (value &&
      value->used_tables() & ~(param->prev_tables | param->read_tables))
    DBUG_RETURN(0);
5389
  for (; key_part != end ; key_part++)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
5390 5391 5392 5393
  {
    if (field->eq(key_part->field))
    {
      SEL_ARG *sel_arg=0;
5394
      if (!tree && !(tree=new SEL_TREE()))
5395
	DBUG_RETURN(0);				// OOM
bk@work.mysql.com's avatar
bk@work.mysql.com committed
5396 5397
      if (!value || !(value->used_tables() & ~param->read_tables))
      {
5398 5399
	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
5400 5401 5402 5403 5404 5405 5406 5407
	if (!sel_arg)
	  continue;
	if (sel_arg->type == SEL_ARG::IMPOSSIBLE)
	{
	  tree->type=SEL_TREE::IMPOSSIBLE;
	  DBUG_RETURN(tree);
	}
      }
5408 5409
      else
      {
5410
	// This key may be used later
5411
	if (!(sel_arg= new SEL_ARG(SEL_ARG::MAYBE_KEY)))
5412
	  DBUG_RETURN(0);			// OOM
5413
      }
bk@work.mysql.com's avatar
bk@work.mysql.com committed
5414 5415
      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
5416
      tree->keys_map.set_bit(key_part->key);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
5417 5418
    }
  }
5419
  
bk@work.mysql.com's avatar
bk@work.mysql.com committed
5420 5421 5422 5423 5424
  DBUG_RETURN(tree);
}


static SEL_ARG *
monty@mysql.com's avatar
monty@mysql.com committed
5425 5426
get_mm_leaf(RANGE_OPT_PARAM *param, COND *conf_func, Field *field,
            KEY_PART *key_part, Item_func::Functype type,Item *value)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
5427
{
5428
  uint maybe_null=(uint) field->real_maybe_null();
5429
  bool optimize_range;
5430 5431
  SEL_ARG *tree= 0;
  MEM_ROOT *alloc= param->mem_root;
5432
  char *str;
evgen@moonbone.local's avatar
evgen@moonbone.local committed
5433
  ulong orig_sql_mode;
5434
  int err;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
5435 5436
  DBUG_ENTER("get_mm_leaf");

5437 5438
  /*
    We need to restore the runtime mem_root of the thread in this
konstantin@mysql.com's avatar
konstantin@mysql.com committed
5439
    function because it evaluates the value of its argument, while
5440 5441 5442 5443 5444 5445
    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;
5446 5447
  if (!value)					// IS NULL or IS NOT NULL
  {
5448
    if (field->table->maybe_null)		// Can't use a key on this
5449
      goto end;
5450
    if (!maybe_null)				// Not null field
5451 5452 5453 5454 5455 5456 5457
    {
      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
5458 5459 5460 5461 5462
    if (type == Item_func::ISNOTNULL_FUNC)
    {
      tree->min_flag=NEAR_MIN;		    /* IS NOT NULL ->  X > NULL */
      tree->max_flag=NO_MAX_RANGE;
    }
5463
    goto end;
5464 5465 5466
  }

  /*
5467 5468 5469 5470 5471 5472 5473 5474 5475 5476
    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 '

5477 5478 5479 5480
  */
  if (field->result_type() == STRING_RESULT &&
      value->result_type() == STRING_RESULT &&
      key_part->image_type == Field::itRAW &&
5481 5482
      ((Field_str*)field)->charset() != conf_func->compare_collation() &&
      !(conf_func->compare_collation()->state & MY_CS_BINSORT))
5483
    goto end;
5484

5485 5486 5487 5488 5489
  if (param->using_real_indexes)
    optimize_range= field->optimize_range(param->real_keynr[key_part->key],
                                          key_part->part);
  else
    optimize_range= TRUE;
5490

bk@work.mysql.com's avatar
bk@work.mysql.com committed
5491 5492 5493 5494
  if (type == Item_func::LIKE_FUNC)
  {
    bool like_error;
    char buff1[MAX_FIELD_WIDTH],*min_str,*max_str;
5495
    String tmp(buff1,sizeof(buff1),value->collation.collation),*res;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
5496
    uint length,offset,min_length,max_length;
5497
    uint field_length= field->pack_length()+maybe_null;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
5498

5499
    if (!optimize_range)
5500
      goto end;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
5501
    if (!(res= value->val_str(&tmp)))
5502 5503 5504 5505
    {
      tree= &null_element;
      goto end;
    }
bk@work.mysql.com's avatar
bk@work.mysql.com committed
5506

5507 5508 5509 5510 5511
    /*
      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
5512 5513 5514 5515 5516 5517
    if (res != &tmp)
    {
      tmp.copy(*res);				// Get own copy
      res= &tmp;
    }
    if (field->cmp_type() != STRING_RESULT)
5518
      goto end;                                 // Can only optimize strings
bk@work.mysql.com's avatar
bk@work.mysql.com committed
5519 5520

    offset=maybe_null;
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
5521 5522 5523
    length=key_part->store_length;

    if (length != key_part->length  + maybe_null)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
5524
    {
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
5525 5526 5527
      /* 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
5528 5529 5530
    }
    else
    {
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
5531 5532 5533 5534 5535 5536 5537 5538
      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
5539
      else
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
5540
	field_length= length;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
5541 5542
    }
    length+=offset;
5543 5544
    if (!(min_str= (char*) alloc_root(alloc, length*2)))
      goto end;
5545

bk@work.mysql.com's avatar
bk@work.mysql.com committed
5546 5547 5548
    max_str=min_str+length;
    if (maybe_null)
      max_str[0]= min_str[0]=0;
5549

5550
    field_length-= maybe_null;
5551
    like_error= my_like_range(field->charset(),
monty@mysql.com's avatar
monty@mysql.com committed
5552
			      res->ptr(), res->length(),
monty@mysql.com's avatar
monty@mysql.com committed
5553 5554
			      ((Item_func_like*)(param->cond))->escape,
			      wild_one, wild_many,
5555
			      field_length,
monty@mysql.com's avatar
monty@mysql.com committed
5556 5557
			      min_str+offset, max_str+offset,
			      &min_length, &max_length);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
5558
    if (like_error)				// Can't optimize with LIKE
5559
      goto end;
monty@mysql.com's avatar
monty@mysql.com committed
5560

5561
    if (offset != maybe_null)			// BLOB or VARCHAR
bk@work.mysql.com's avatar
bk@work.mysql.com committed
5562 5563 5564 5565
    {
      int2store(min_str+maybe_null,min_length);
      int2store(max_str+maybe_null,max_length);
    }
5566 5567
    tree= new (alloc) SEL_ARG(field, min_str, max_str);
    goto end;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
5568 5569
  }

5570
  if (!optimize_range &&
5571
      type != Item_func::EQ_FUNC &&
bk@work.mysql.com's avatar
bk@work.mysql.com committed
5572
      type != Item_func::EQUAL_FUNC)
5573
    goto end;                                   // Can't optimize this
bk@work.mysql.com's avatar
bk@work.mysql.com committed
5574

5575 5576 5577 5578
  /*
    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
5579 5580 5581
  if (field->result_type() == STRING_RESULT &&
      value->result_type() != STRING_RESULT &&
      field->cmp_type() != value->result_type())
5582
    goto end;
5583
  /* For comparison purposes allow invalid dates like 2000-01-32 */
evgen@moonbone.local's avatar
evgen@moonbone.local committed
5584
  orig_sql_mode= field->table->in_use->variables.sql_mode;
5585 5586 5587 5588
  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;
5589 5590 5591 5592 5593 5594 5595
  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
5596
  {
5597
    field->table->in_use->variables.sql_mode= orig_sql_mode;
5598
    /* This happens when we try to insert a NULL field in a not null column */
5599 5600
    tree= &null_element;                        // cmp with NULL is never TRUE
    goto end;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
5601
  }
5602
  field->table->in_use->variables.sql_mode= orig_sql_mode;
5603
  str= (char*) alloc_root(alloc, key_part->store_length+1);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
5604
  if (!str)
5605
    goto end;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
5606
  if (maybe_null)
5607
    *str= (char) field->is_real_null();		// Set to 1 if null
5608
  field->get_key_image(str+maybe_null, key_part->length, key_part->image_type);
5609 5610
  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
5611

timour@mysql.com's avatar
timour@mysql.com committed
5612 5613 5614 5615 5616 5617 5618 5619 5620 5621 5622
  /*
    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
5623 5624
  if (field->result_type() == INT_RESULT &&
      value->result_type() == INT_RESULT &&
timour@mysql.com's avatar
timour@mysql.com committed
5625 5626 5627 5628 5629 5630 5631 5632
      ((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;
5633
        goto end;
timour@mysql.com's avatar
timour@mysql.com committed
5634 5635
      }
      if (type == Item_func::GT_FUNC || type == Item_func::GE_FUNC)
5636 5637 5638 5639
      {
        tree= 0;
        goto end;
      }
timour@mysql.com's avatar
timour@mysql.com committed
5640 5641
    }
  }
bk@work.mysql.com's avatar
bk@work.mysql.com committed
5642 5643 5644 5645 5646 5647 5648 5649 5650 5651 5652 5653 5654 5655 5656 5657 5658 5659 5660 5661 5662 5663

  switch (type) {
  case Item_func::LT_FUNC:
    if (field_is_equal_to_item(field,value))
      tree->max_flag=NEAR_MAX;
    /* fall through */
  case Item_func::LE_FUNC:
    if (!maybe_null)
      tree->min_flag=NO_MIN_RANGE;		/* From start */
    else
    {						// > NULL
      tree->min_value=is_null_string;
      tree->min_flag=NEAR_MIN;
    }
    break;
  case Item_func::GT_FUNC:
    if (field_is_equal_to_item(field,value))
      tree->min_flag=NEAR_MIN;
    /* fall through */
  case Item_func::GE_FUNC:
    tree->max_flag=NO_MAX_RANGE;
    break;
5664
  case Item_func::SP_EQUALS_FUNC:
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
5665 5666 5667
    tree->min_flag=GEOM_FLAG | HA_READ_MBR_EQUAL;// NEAR_MIN;//512;
    tree->max_flag=NO_MAX_RANGE;
    break;
5668
  case Item_func::SP_DISJOINT_FUNC:
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
5669 5670 5671
    tree->min_flag=GEOM_FLAG | HA_READ_MBR_DISJOINT;// NEAR_MIN;//512;
    tree->max_flag=NO_MAX_RANGE;
    break;
5672
  case Item_func::SP_INTERSECTS_FUNC:
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
5673 5674 5675
    tree->min_flag=GEOM_FLAG | HA_READ_MBR_INTERSECT;// NEAR_MIN;//512;
    tree->max_flag=NO_MAX_RANGE;
    break;
5676
  case Item_func::SP_TOUCHES_FUNC:
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
5677 5678 5679
    tree->min_flag=GEOM_FLAG | HA_READ_MBR_INTERSECT;// NEAR_MIN;//512;
    tree->max_flag=NO_MAX_RANGE;
    break;
5680 5681

  case Item_func::SP_CROSSES_FUNC:
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
5682 5683 5684
    tree->min_flag=GEOM_FLAG | HA_READ_MBR_INTERSECT;// NEAR_MIN;//512;
    tree->max_flag=NO_MAX_RANGE;
    break;
5685
  case Item_func::SP_WITHIN_FUNC:
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
5686 5687 5688
    tree->min_flag=GEOM_FLAG | HA_READ_MBR_WITHIN;// NEAR_MIN;//512;
    tree->max_flag=NO_MAX_RANGE;
    break;
5689 5690

  case Item_func::SP_CONTAINS_FUNC:
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
5691 5692 5693
    tree->min_flag=GEOM_FLAG | HA_READ_MBR_CONTAIN;// NEAR_MIN;//512;
    tree->max_flag=NO_MAX_RANGE;
    break;
5694
  case Item_func::SP_OVERLAPS_FUNC:
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
5695 5696 5697
    tree->min_flag=GEOM_FLAG | HA_READ_MBR_INTERSECT;// NEAR_MIN;//512;
    tree->max_flag=NO_MAX_RANGE;
    break;
5698

bk@work.mysql.com's avatar
bk@work.mysql.com committed
5699 5700 5701
  default:
    break;
  }
5702 5703 5704

end:
  param->thd->mem_root= alloc;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
5705 5706 5707 5708 5709 5710 5711 5712 5713
  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
5714 5715
** IMPOSSIBLE:	 Condition is never TRUE
** ALWAYS:	 Condition is always TRUE
bk@work.mysql.com's avatar
bk@work.mysql.com committed
5716 5717 5718 5719 5720 5721
** 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
******************************************************************************/

/*
5722 5723
  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
5724 5725 5726 5727 5728 5729 5730 5731 5732 5733 5734 5735 5736 5737 5738 5739 5740 5741 5742 5743 5744 5745 5746 5747 5748 5749 5750 5751 5752 5753 5754 5755 5756 5757 5758 5759 5760 5761
*/

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 *
5762
tree_and(RANGE_OPT_PARAM *param,SEL_TREE *tree1,SEL_TREE *tree2)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
5763 5764 5765 5766 5767 5768 5769 5770 5771 5772 5773 5774 5775 5776 5777 5778 5779 5780 5781 5782 5783 5784
{
  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
5785 5786
  key_map  result_keys;
  result_keys.clear_all();
bk@work.mysql.com's avatar
bk@work.mysql.com committed
5787 5788 5789 5790 5791 5792 5793 5794 5795 5796 5797 5798 5799
  /* 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);
5800
      if (*key1 && (*key1)->type == SEL_ARG::IMPOSSIBLE)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
5801 5802
      {
	tree1->type= SEL_TREE::IMPOSSIBLE;
5803
        DBUG_RETURN(tree1);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
5804
      }
sergefp@mysql.com's avatar
sergefp@mysql.com committed
5805
      result_keys.set_bit(key1 - tree1->keys);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
5806
#ifdef EXTRA_DEBUG
5807 5808
      if (*key1)
        (*key1)->test_use_count(*key1);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
5809 5810 5811
#endif
    }
  }
5812 5813
  tree1->keys_map= result_keys;
  /* dispose index_merge if there is a "range" option */
sergefp@mysql.com's avatar
sergefp@mysql.com committed
5814
  if (!result_keys.is_clear_all())
5815 5816 5817 5818 5819 5820 5821
  {
    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
5822 5823 5824 5825
  DBUG_RETURN(tree1);
}


5826
/*
5827 5828
  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
5829
  using index_merge.
5830 5831
*/

5832 5833
bool sel_trees_can_be_ored(SEL_TREE *tree1, SEL_TREE *tree2, 
                           RANGE_OPT_PARAM* param)
5834
{
sergefp@mysql.com's avatar
sergefp@mysql.com committed
5835
  key_map common_keys= tree1->keys_map;
5836
  DBUG_ENTER("sel_trees_can_be_ored");
5837
  common_keys.intersect(tree2->keys_map);
5838

sergefp@mysql.com's avatar
sergefp@mysql.com committed
5839
  if (common_keys.is_clear_all())
monty@mysql.com's avatar
monty@mysql.com committed
5840
    DBUG_RETURN(FALSE);
5841 5842

  /* trees have a common key, check if they refer to same key part */
5843
  SEL_ARG **key1,**key2;
sergefp@mysql.com's avatar
sergefp@mysql.com committed
5844
  for (uint key_no=0; key_no < param->keys; key_no++)
5845
  {
sergefp@mysql.com's avatar
sergefp@mysql.com committed
5846
    if (common_keys.is_set(key_no))
5847 5848 5849 5850 5851
    {
      key1= tree1->keys + key_no;
      key2= tree2->keys + key_no;
      if ((*key1)->part == (*key2)->part)
      {
monty@mysql.com's avatar
monty@mysql.com committed
5852
        DBUG_RETURN(TRUE);
5853 5854 5855
      }
    }
  }
monty@mysql.com's avatar
monty@mysql.com committed
5856
  DBUG_RETURN(FALSE);
5857
}
bk@work.mysql.com's avatar
bk@work.mysql.com committed
5858

5859 5860 5861 5862 5863 5864 5865 5866 5867 5868 5869 5870 5871 5872 5873 5874 5875 5876 5877 5878 5879 5880 5881 5882 5883 5884 5885 5886 5887 5888 5889 5890 5891 5892 5893 5894 5895 5896 5897 5898 5899 5900 5901 5902 5903 5904 5905 5906 5907 5908 5909 5910 5911 5912 5913 5914 5915 5916 5917 5918 5919 5920 5921 5922

/*
  Remove the trees that are not suitable for record retrieval.
  SYNOPSIS
    param  Range analysis parameter
    tree   Tree to be processed, tree->type is KEY or KEY_SMALLER
 
  DESCRIPTION
    This function walks through tree->keys[] and removes the SEL_ARG* trees
    that are not "maybe" trees (*) and cannot be used to construct quick range
    selects.
    (*) - have type MAYBE or MAYBE_KEY. Perhaps we should remove trees of
          these types here as well.

    A SEL_ARG* tree cannot be used to construct quick select if it has
    tree->part != 0. (e.g. it could represent "keypart2 < const").

    WHY THIS FUNCTION IS NEEDED
    
    Normally we allow construction of SEL_TREE objects that have SEL_ARG
    trees that do not allow quick range select construction. For example for
    " keypart1=1 AND keypart2=2 " the execution will proceed as follows:
    tree1= SEL_TREE { SEL_ARG{keypart1=1} }
    tree2= SEL_TREE { SEL_ARG{keypart2=2} } -- can't make quick range select
                                               from this
    call tree_and(tree1, tree2) -- this joins SEL_ARGs into a usable SEL_ARG
                                   tree.
    
    There is an exception though: when we construct index_merge SEL_TREE,
    any SEL_ARG* tree that cannot be used to construct quick range select can
    be removed, because current range analysis code doesn't provide any way
    that tree could be later combined with another tree.
    Consider an example: we should not construct
    st1 = SEL_TREE { 
      merges = SEL_IMERGE { 
                            SEL_TREE(t.key1part1 = 1), 
                            SEL_TREE(t.key2part2 = 2)   -- (*)
                          } 
                   };
    because 
     - (*) cannot be used to construct quick range select, 
     - There is no execution path that would cause (*) to be converted to 
       a tree that could be used.

    The latter is easy to verify: first, notice that the only way to convert
    (*) into a usable tree is to call tree_and(something, (*)).

    Second look at what tree_and/tree_or function would do when passed a
    SEL_TREE that has the structure like st1 tree has, and conlcude that 
    tree_and(something, (*)) will not be called.

  RETURN
    0  Ok, some suitable trees left
    1  No tree->keys[] left.
*/

static bool remove_nonrange_trees(RANGE_OPT_PARAM *param, SEL_TREE *tree)
{
  bool res= FALSE;
  for (uint i=0; i < param->keys; i++)
  {
    if (tree->keys[i])
    {
      if (tree->keys[i]->part)
5923
      {
5924
        tree->keys[i]= NULL;
5925 5926
        tree->keys_map.clear_bit(i);
      }
5927 5928 5929 5930
      else
        res= TRUE;
    }
  }
5931
  return !res;
5932 5933 5934
}


bk@work.mysql.com's avatar
bk@work.mysql.com committed
5935
static SEL_TREE *
5936
tree_or(RANGE_OPT_PARAM *param,SEL_TREE *tree1,SEL_TREE *tree2)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
5937 5938 5939 5940 5941 5942 5943 5944 5945 5946 5947 5948 5949
{
  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);

5950
  SEL_TREE *result= 0;
sergefp@mysql.com's avatar
sergefp@mysql.com committed
5951 5952
  key_map  result_keys;
  result_keys.clear_all();
5953
  if (sel_trees_can_be_ored(tree1, tree2, param))
bk@work.mysql.com's avatar
bk@work.mysql.com committed
5954
  {
5955 5956 5957 5958
    /* 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
5959
    {
5960 5961 5962 5963
      *key1=key_or(*key1,*key2);
      if (*key1)
      {
        result=tree1;				// Added to tree1
sergefp@mysql.com's avatar
sergefp@mysql.com committed
5964
        result_keys.set_bit(key1 - tree1->keys);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
5965
#ifdef EXTRA_DEBUG
5966
        (*key1)->test_use_count(*key1);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
5967
#endif
5968 5969 5970 5971 5972 5973 5974 5975 5976 5977
      }
    }
    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())
    {
5978 5979 5980 5981 5982 5983 5984
      if (param->remove_jump_scans)
      {
        bool no_trees= remove_nonrange_trees(param, tree1);
        no_trees= no_trees || remove_nonrange_trees(param, tree2);
        if (no_trees)
          DBUG_RETURN(new SEL_TREE(SEL_TREE::ALWAYS));
      }
5985 5986 5987 5988 5989 5990 5991 5992 5993 5994 5995 5996 5997 5998 5999 6000 6001 6002 6003 6004 6005
      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
6006
        swap_variables(SEL_TREE*, tree1, tree2);
6007 6008 6009
      
      if (param->remove_jump_scans && remove_nonrange_trees(param, tree2))
         DBUG_RETURN(new SEL_TREE(SEL_TREE::ALWAYS));
6010 6011 6012 6013 6014
      /* 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
6015 6016 6017 6018 6019 6020 6021 6022 6023 6024 6025 6026 6027 6028 6029 6030 6031 6032 6033 6034 6035
    }
  }
  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)
  {
6036 6037
    key1->right= key1->left= &null_element;
    key1->next= key1->prev= 0;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
6038 6039 6040 6041 6042 6043 6044 6045 6046 6047 6048 6049 6050 6051 6052 6053 6054 6055 6056 6057 6058 6059 6060 6061 6062
  }
  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;
}


6063 6064 6065 6066 6067 6068 6069 6070 6071 6072 6073 6074 6075
/*
  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
6076
static SEL_ARG *
6077
key_and(SEL_ARG *key1, SEL_ARG *key2, uint clone_flag)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
6078 6079 6080 6081 6082 6083 6084 6085 6086
{
  if (!key1)
    return key2;
  if (!key2)
    return key1;
  if (key1->part != key2->part)
  {
    if (key1->part > key2->part)
    {
6087
      swap_variables(SEL_ARG *, key1, key2);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
6088 6089 6090 6091 6092
      clone_flag=swap_clone_flag(clone_flag);
    }
    // key1->part < key2->part
    key1->use_count--;
    if (key1->use_count > 0)
6093 6094
      if (!(key1= key1->clone_tree()))
	return 0;				// OOM
bk@work.mysql.com's avatar
bk@work.mysql.com committed
6095 6096 6097 6098
    return and_all_keys(key1,key2,clone_flag);
  }

  if (((clone_flag & CLONE_KEY2_MAYBE) &&
6099 6100
       !(clone_flag & CLONE_KEY1_MAYBE) &&
       key2->type != SEL_ARG::MAYBE_KEY) ||
bk@work.mysql.com's avatar
bk@work.mysql.com committed
6101 6102
      key1->type == SEL_ARG::MAYBE_KEY)
  {						// Put simple key in key2
6103
    swap_variables(SEL_ARG *, key1, key2);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
6104 6105 6106
    clone_flag=swap_clone_flag(clone_flag);
  }

6107
  /* 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
6108 6109 6110 6111 6112
  if (key2->type == SEL_ARG::MAYBE_KEY)
  {
    if (key1->use_count > 1)
    {
      key1->use_count--;
6113 6114
      if (!(key1=key1->clone_tree()))
	return 0;				// OOM
bk@work.mysql.com's avatar
bk@work.mysql.com committed
6115 6116 6117 6118 6119 6120 6121 6122 6123 6124 6125 6126 6127 6128
      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)
6129 6130
      {
	key1->use_count--;			// Incremented in and_all_keys
bk@work.mysql.com's avatar
bk@work.mysql.com committed
6131
	return and_all_keys(key1,key2,clone_flag);
6132
      }
bk@work.mysql.com's avatar
bk@work.mysql.com committed
6133 6134 6135 6136 6137
      key2->use_count--;			// Key2 doesn't have a tree
    }
    return key1;
  }

6138 6139
  if ((key1->min_flag | key2->min_flag) & GEOM_FLAG)
  {
6140
    /* TODO: why not leave one of the trees? */
6141 6142 6143 6144 6145
    key1->free_tree();
    key2->free_tree();
    return 0;					// Can't optimize this
  }

6146 6147 6148
  if ((key1->min_flag | key2->min_flag) & GEOM_FLAG)
  {
    key1->free_tree();
6149 6150 6151 6152
    key2->free_tree();
    return 0;					// Can't optimize this
  }

bk@work.mysql.com's avatar
bk@work.mysql.com committed
6153 6154 6155 6156 6157 6158 6159 6160 6161 6162 6163 6164 6165 6166 6167 6168 6169 6170 6171 6172
  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);
6173 6174
      if (!new_arg)
	return &null_element;			// End of memory
bk@work.mysql.com's avatar
bk@work.mysql.com committed
6175 6176 6177 6178 6179 6180 6181 6182 6183 6184 6185 6186 6187 6188 6189 6190 6191 6192 6193 6194 6195 6196 6197 6198 6199 6200 6201 6202 6203 6204 6205 6206 6207 6208 6209 6210 6211 6212 6213 6214 6215 6216 6217 6218 6219 6220 6221 6222 6223 6224 6225
      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;
  }
6226
  if (!key2)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
6227 6228 6229 6230 6231 6232 6233 6234
  {
    key1->use_count--;
    key1->free_tree();
    return 0;
  }
  key1->use_count--;
  key2->use_count--;

6235 6236
  if (key1->part != key2->part || 
      (key1->min_flag | key2->min_flag) & GEOM_FLAG)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
6237 6238 6239 6240 6241 6242 6243 6244 6245 6246 6247 6248 6249 6250 6251 6252 6253 6254 6255 6256 6257 6258 6259 6260
  {
    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)
    {
6261
      swap_variables(SEL_ARG *,key1,key2);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
6262
    }
6263
    if (key1->use_count > 0 || !(key1=key1->clone_tree()))
6264
      return 0;					// OOM
bk@work.mysql.com's avatar
bk@work.mysql.com committed
6265 6266 6267 6268 6269 6270 6271 6272 6273 6274 6275 6276 6277 6278 6279 6280 6281 6282 6283 6284 6285 6286 6287 6288 6289
  }

  // 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
6290
	  if (!(key2=new SEL_ARG(*key2)))
6291
	    return 0;		// out of memory
bk@work.mysql.com's avatar
bk@work.mysql.com committed
6292 6293 6294 6295 6296 6297 6298 6299 6300 6301 6302 6303 6304 6305 6306 6307 6308 6309 6310 6311 6312 6313 6314 6315 6316 6317 6318 6319 6320 6321 6322 6323 6324 6325 6326 6327 6328 6329 6330 6331
	  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)
	  {
6332 6333
	    SEL_ARG *cpy= new SEL_ARG(*key2);	// Must make copy
	    if (!cpy)
6334
	      return 0;				// OOM
6335
	    key1=key1->insert(cpy);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
6336 6337 6338 6339 6340 6341 6342 6343 6344 6345 6346 6347 6348 6349 6350 6351 6352 6353 6354 6355 6356 6357 6358 6359 6360 6361 6362 6363
	    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
6364
        last->copy_min(tmp);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
6365 6366 6367 6368 6369 6370 6371 6372 6373 6374 6375 6376 6377 6378 6379 6380 6381
	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);
6382 6383
      if (!new_arg)
	return 0;				// OOM
bk@work.mysql.com's avatar
bk@work.mysql.com committed
6384 6385 6386 6387 6388 6389 6390 6391 6392 6393 6394 6395 6396
      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);
6397 6398
	if (!new_arg)
	  return 0;				// OOM
bk@work.mysql.com's avatar
bk@work.mysql.com committed
6399 6400 6401 6402 6403 6404 6405 6406 6407 6408 6409 6410 6411 6412
	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))
	{
6413 6414 6415 6416
	  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
6417 6418 6419 6420 6421
	  key2=key2->next;
	  goto end;
	}
	if (tmp->cmp_min_to_max(&key) > 0)
	{
6422 6423 6424 6425
	  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
6426 6427 6428 6429 6430 6431
	  break;
	}
      }
      else
      {
	SEL_ARG *new_arg=tmp->clone_last(&key); // tmp.min <= x <= key.max
6432 6433
	if (!new_arg)
	  return 0;				// OOM
bk@work.mysql.com's avatar
bk@work.mysql.com committed
6434 6435
	tmp->copy_max_to_min(&key);
	tmp->increment_use_count(key1->use_count+1);
6436 6437
	/* 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
6438 6439 6440 6441 6442 6443 6444 6445 6446 6447 6448 6449 6450 6451
	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)
    {
6452 6453 6454
      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
6455
      key2->increment_use_count(key1->use_count+1);
6456
      key1=key1->insert(tmp);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
6457 6458 6459 6460 6461 6462 6463 6464 6465 6466 6467 6468 6469 6470 6471 6472 6473 6474 6475 6476 6477 6478 6479 6480 6481 6482 6483 6484 6485 6486 6487 6488 6489 6490 6491 6492 6493 6494 6495 6496 6497 6498 6499 6500 6501 6502
    }
    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;
6503 6504
  LINT_INIT(par);
  LINT_INIT(last_element);
6505

bk@work.mysql.com's avatar
bk@work.mysql.com committed
6506 6507 6508 6509 6510 6511 6512 6513 6514 6515 6516 6517 6518 6519 6520 6521 6522 6523 6524 6525 6526 6527 6528 6529 6530 6531 6532 6533 6534 6535 6536 6537 6538 6539 6540 6541 6542 6543 6544 6545 6546 6547 6548 6549 6550 6551 6552 6553 6554 6555 6556 6557 6558 6559 6560 6561 6562 6563 6564 6565 6566 6567 6568 6569 6570 6571 6572
  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;
  }
}


/*
6573 6574 6575 6576 6577
  Remove a element from the tree

  SYNOPSIS
    tree_delete()
    key		Key that is to be deleted from tree (this)
6578

6579 6580 6581 6582 6583
  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
6584 6585 6586 6587 6588 6589 6590
*/

SEL_ARG *
SEL_ARG::tree_delete(SEL_ARG *key)
{
  enum leaf_color remove_color;
  SEL_ARG *root,*nod,**par,*fix_par;
6591 6592 6593 6594
  DBUG_ENTER("tree_delete");

  root=this;
  this->parent= 0;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
6595 6596 6597 6598 6599 6600 6601 6602 6603 6604 6605 6606 6607 6608 6609 6610 6611 6612 6613 6614 6615 6616 6617 6618 6619 6620 6621 6622 6623 6624 6625 6626 6627 6628 6629 6630 6631 6632 6633 6634 6635 6636 6637 6638 6639 6640

  /* 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)
6641
    DBUG_RETURN(0);				// Maybe root later
bk@work.mysql.com's avatar
bk@work.mysql.com committed
6642 6643 6644 6645 6646 6647 6648
  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;
6649
  DBUG_RETURN(root);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
6650 6651 6652 6653 6654 6655 6656 6657 6658 6659 6660 6661 6662 6663 6664 6665 6666 6667 6668 6669 6670 6671 6672 6673 6674 6675 6676 6677 6678 6679 6680 6681 6682 6683 6684 6685 6686 6687 6688 6689 6690 6691 6692 6693 6694 6695 6696 6697 6698 6699 6700 6701 6702 6703 6704 6705 6706 6707 6708 6709 6710 6711 6712 6713 6714 6715 6716 6717 6718 6719 6720 6721 6722 6723 6724 6725 6726 6727 6728 6729 6730 6731 6732 6733 6734 6735 6736 6737 6738 6739 6740 6741 6742 6743 6744 6745 6746 6747 6748 6749 6750 6751 6752 6753 6754 6755 6756 6757 6758 6759 6760 6761 6762 6763 6764 6765 6766 6767 6768 6769 6770 6771 6772 6773 6774 6775 6776 6777 6778 6779 6780 6781 6782 6783 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 6813 6814 6815 6816 6817 6818 6819 6820 6821 6822 6823 6824
}


	/* 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;
}


6825
	/* Test that the properties for a red-black tree hold */
bk@work.mysql.com's avatar
bk@work.mysql.com committed
6826 6827 6828 6829 6830 6831 6832 6833 6834 6835 6836 6837 6838 6839 6840 6841 6842 6843 6844 6845 6846 6847 6848 6849 6850 6851 6852 6853 6854 6855 6856 6857 6858 6859 6860 6861 6862

#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
}

6863 6864 6865 6866 6867 6868 6869 6870 6871 6872 6873 6874 6875 6876 6877 6878 6879 6880 6881 6882 6883 6884 6885 6886 6887 6888 6889 6890 6891 6892 6893 6894 6895 6896 6897 6898 6899 6900 6901 6902 6903 6904 6905 6906 6907

/*
  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
6908 6909 6910 6911 6912 6913 6914 6915 6916 6917 6918 6919 6920 6921 6922 6923 6924
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;
}


6925 6926 6927 6928 6929 6930 6931 6932 6933 6934 6935 6936 6937 6938
/*
  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
6939 6940
void SEL_ARG::test_use_count(SEL_ARG *root)
{
6941
  uint e_count=0;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
6942 6943
  if (this == root && use_count != 1)
  {
monty@mysql.com's avatar
monty@mysql.com committed
6944
    sql_print_information("Use_count: Wrong count %lu for root",use_count);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
6945 6946 6947 6948 6949 6950 6951 6952 6953 6954 6955 6956
    return;
  }
  if (this->type != SEL_ARG::KEY_RANGE)
    return;
  for (SEL_ARG *pos=first(); pos ; pos=pos->next)
  {
    e_count++;
    if (pos->next_key_part)
    {
      ulong count=count_key_part_usage(root,pos->next_key_part);
      if (count > pos->next_key_part->use_count)
      {
monty@mysql.com's avatar
monty@mysql.com committed
6957
	sql_print_information("Use_count: Wrong count for key at 0x%lx, %lu should be %lu",
bk@work.mysql.com's avatar
bk@work.mysql.com committed
6958 6959 6960 6961 6962 6963 6964
			pos,pos->next_key_part->use_count,count);
	return;
      }
      pos->next_key_part->test_use_count(root);
    }
  }
  if (e_count != elements)
monty@mysql.com's avatar
monty@mysql.com committed
6965
    sql_print_warning("Wrong use count: %u (should be %u) for tree at 0x%lx",
6966
		    e_count, elements, (gptr) this);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
6967 6968 6969 6970 6971
}

#endif


6972 6973 6974 6975 6976 6977
/*
  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
6978 6979 6980 6981 6982 6983
      idx               Number of index to use in tree->keys
      tree              Transformed selection condition, tree->keys[idx]
                        holds the range tree to be used for scanning.
      update_tbl_stats  If true, update table->quick_keys with information
                        about range scan we've evaluated.

6984
  NOTES
6985
    param->is_ror_scan is set to reflect if the key scan is a ROR (see
6986
    is_key_scan_ror function for more info)
6987
    param->table->quick_*, param->range_count (and maybe others) are
6988
    updated with data of given key scan, see check_quick_keys for details.
6989 6990

  RETURN
6991
    Estimate # of records to be retrieved.
6992
    HA_POS_ERROR if estimate calculation failed due to table handler problems.
bk@work.mysql.com's avatar
bk@work.mysql.com committed
6993

6994
*/
bk@work.mysql.com's avatar
bk@work.mysql.com committed
6995 6996

static ha_rows
6997
check_quick_select(PARAM *param,uint idx,SEL_ARG *tree, bool update_tbl_stats)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
6998 6999
{
  ha_rows records;
7000 7001
  bool    cpk_scan;
  uint key;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
7002
  DBUG_ENTER("check_quick_select");
7003

monty@mysql.com's avatar
monty@mysql.com committed
7004
  param->is_ror_scan= FALSE;
7005

bk@work.mysql.com's avatar
bk@work.mysql.com committed
7006 7007
  if (!tree)
    DBUG_RETURN(HA_POS_ERROR);			// Can't use it
monty@narttu.mysql.fi's avatar
monty@narttu.mysql.fi committed
7008 7009
  param->max_key_part=0;
  param->range_count=0;
7010 7011
  key= param->real_keynr[idx];

bk@work.mysql.com's avatar
bk@work.mysql.com committed
7012 7013 7014 7015
  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
7016 7017 7018 7019 7020

  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
7021
    cpk_scan= FALSE;
7022 7023 7024 7025 7026 7027 7028
  }
  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).
    */
7029 7030
    cpk_scan= ((param->table->s->primary_key == param->real_keynr[idx]) &&
               param->table->file->primary_key_is_clustered());
7031
    param->is_ror_scan= !cpk_scan;
7032
  }
7033
  param->n_ranges= 0;
7034

bk@work.mysql.com's avatar
bk@work.mysql.com committed
7035 7036
  records=check_quick_keys(param,idx,tree,param->min_key,0,param->max_key,0);
  if (records != HA_POS_ERROR)
7037
  {
7038 7039 7040 7041 7042 7043 7044 7045 7046 7047 7048 7049
    if (update_tbl_stats)
    {
      param->table->quick_keys.set_bit(key);
      param->table->quick_key_parts[key]=param->max_key_part+1;
      param->table->quick_n_ranges[key]= param->n_ranges;
      param->table->quick_condition_rows=
        min(param->table->quick_condition_rows, records);
    }
    /*
      Need to save quick_rows in any case as it is used when calculating
      cost of ROR intersection:
    */
bk@work.mysql.com's avatar
bk@work.mysql.com committed
7050
    param->table->quick_rows[key]=records;
7051
    if (cpk_scan)
monty@mysql.com's avatar
monty@mysql.com committed
7052
      param->is_ror_scan= TRUE;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
7053
  }
7054 7055
  if (param->table->file->index_flags(key, 0, TRUE) & HA_KEY_SCAN_NOT_ROR)
    param->is_ror_scan= FALSE;
7056
  DBUG_PRINT("exit", ("Records: %lu", (ulong) records));
bk@work.mysql.com's avatar
bk@work.mysql.com committed
7057 7058 7059 7060
  DBUG_RETURN(records);
}


7061
/*
7062 7063
  Recursively calculate estimate of # rows that will be retrieved by
  key scan on key idx.
7064 7065
  SYNOPSIS
    check_quick_keys()
7066
      param         Parameter from test_quick select function.
7067
      idx           Number of key to use in PARAM::keys in list of used keys
7068 7069 7070
                    (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
7071
      min_key_flag
7072
      max_key       Buffer with partial max key value tuple
7073 7074
      max_key_flag

7075
  NOTES
7076 7077
    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
7078 7079
    are calculated using records_in_range calls at the leaf nodes and then
    summed.
7080

7081 7082
    param->min_key and param->max_key are used to hold prefixes of key value
    tuples.
7083 7084

    The side effects are:
7085

7086 7087
    param->max_key_part is updated to hold the maximum number of key parts used
      in scan minus 1.
7088 7089

    param->range_count is incremented if the function finds a range that
7090
      wasn't counted by the caller.
7091

7092 7093 7094
    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)
7095 7096
*/

bk@work.mysql.com's avatar
bk@work.mysql.com committed
7097 7098 7099 7100 7101
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
7102 7103 7104
  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
7105 7106 7107 7108

  param->max_key_part=max(param->max_key_part,key_tree->part);
  if (key_tree->left != &null_element)
  {
7109 7110 7111 7112 7113 7114
    /*
      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
7115
    param->is_ror_scan= FALSE;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
7116 7117 7118 7119 7120 7121
    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
7122 7123
  tmp_min_key= min_key;
  tmp_max_key= max_key;
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
7124
  key_tree->store(param->key[idx][key_tree->part].store_length,
bk@work.mysql.com's avatar
bk@work.mysql.com committed
7125
		  &tmp_min_key,min_key_flag,&tmp_max_key,max_key_flag);
monty@mysql.com's avatar
monty@mysql.com committed
7126 7127
  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
7128

7129 7130
  if (param->is_ror_scan)
  {
7131
    /*
7132
      If the index doesn't cover entire key, mark the scan as non-ROR scan.
7133
      Actually we're cutting off some ROR scans here.
7134 7135 7136
    */
    uint16 fieldnr= param->table->key_info[param->real_keynr[idx]].
                    key_part[key_tree->part].fieldnr - 1;
7137
    if (param->table->field[fieldnr]->key_length() !=
7138
        param->key[idx][key_tree->part].length)
monty@mysql.com's avatar
monty@mysql.com committed
7139
      param->is_ror_scan= FALSE;
7140 7141
  }

bk@work.mysql.com's avatar
bk@work.mysql.com committed
7142 7143 7144 7145 7146 7147 7148 7149 7150 7151 7152 7153 7154
  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
    }
7155
    else
7156 7157
    {
      /* The interval for current key part is not c1 <= keyXpartY <= c1 */
monty@mysql.com's avatar
monty@mysql.com committed
7158
      param->is_ror_scan= FALSE;
7159
    }
7160

bk@work.mysql.com's avatar
bk@work.mysql.com committed
7161 7162 7163 7164 7165 7166 7167 7168 7169 7170 7171 7172 7173 7174 7175 7176 7177 7178
    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
7179
  param->range_count++;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
7180 7181
  if (!tmp_min_flag && ! tmp_max_flag &&
      (uint) key_tree->part+1 == param->table->key_info[keynr].key_parts &&
7182 7183
      (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
7184 7185
      min_key_length == max_key_length &&
      !memcmp(param->min_key,param->max_key,min_key_length))
7186
  {
bk@work.mysql.com's avatar
bk@work.mysql.com committed
7187
    tmp=1;					// Max one record
7188 7189
    param->n_ranges++;
  }
bk@work.mysql.com's avatar
bk@work.mysql.com committed
7190
  else
7191
  {
7192 7193
    if (param->is_ror_scan)
    {
7194 7195 7196 7197 7198 7199 7200 7201 7202
      /*
        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.
      */
7203 7204
      if (!(min_key_length == max_key_length &&
            !memcmp(min_key,max_key, (uint) (tmp_max_key - max_key)) &&
7205
            !key_tree->min_flag && !key_tree->max_flag &&
7206
            is_key_scan_ror(param, keynr, key_tree->part + 1)))
monty@mysql.com's avatar
monty@mysql.com committed
7207
        param->is_ror_scan= FALSE;
7208
    }
7209
    param->n_ranges++;
7210

7211
    if (tmp_min_flag & GEOM_FLAG)
7212
    {
7213 7214 7215 7216 7217 7218 7219 7220
      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);
7221 7222 7223
    }
    else
    {
7224 7225 7226 7227 7228 7229
      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
7230
      max_range.key=    (byte*) param->max_key;
7231 7232 7233 7234 7235 7236 7237 7238
      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));
7239 7240
    }
  }
bk@work.mysql.com's avatar
bk@work.mysql.com committed
7241 7242 7243 7244 7245 7246
 end:
  if (tmp == HA_POS_ERROR)			// Impossible range
    return tmp;
  records+=tmp;
  if (key_tree->right != &null_element)
  {
7247 7248 7249 7250 7251 7252
    /*
      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
7253
    param->is_ror_scan= FALSE;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
7254 7255 7256 7257 7258 7259 7260 7261 7262
    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;
}

7263

7264
/*
7265
  Check if key scan on given index with equality conditions on first n key
7266 7267 7268 7269
  parts is a ROR scan.

  SYNOPSIS
    is_key_scan_ror()
7270
      param  Parameter from test_quick_select
7271 7272 7273 7274
      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.
7275

7276 7277 7278
  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)
7279

7280 7281 7282
    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)
7283

7284 7285
    where the index is defined on (key1_1, ..., key1_N [,a_1, ..., a_n])

7286
    and the table has a clustered Primary Key
7287

7288
    PRIMARY KEY(a_1, ..., a_n, b1, ..., b_k) with first key parts being
7289
    identical to uncovered parts ot the key being scanned (2)
7290 7291

    Scans on HASH indexes are not ROR scans,
7292 7293 7294 7295 7296 7297
    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.

7298
  RETURN
monty@mysql.com's avatar
monty@mysql.com committed
7299 7300
    TRUE  If the scan is ROR-scan
    FALSE otherwise
7301
*/
7302

7303 7304 7305 7306
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;
7307 7308 7309
  KEY_PART_INFO *key_part_end= (table_key->key_part +
                                table_key->key_parts);
  uint pk_number;
7310

7311
  if (key_part == key_part_end)
monty@mysql.com's avatar
monty@mysql.com committed
7312
    return TRUE;
7313
  pk_number= param->table->s->primary_key;
7314
  if (!param->table->file->primary_key_is_clustered() || pk_number == MAX_KEY)
monty@mysql.com's avatar
monty@mysql.com committed
7315
    return FALSE;
7316 7317

  KEY_PART_INFO *pk_part= param->table->key_info[pk_number].key_part;
7318
  KEY_PART_INFO *pk_part_end= pk_part +
7319
                              param->table->key_info[pk_number].key_parts;
7320 7321
  for (;(key_part!=key_part_end) && (pk_part != pk_part_end);
       ++key_part, ++pk_part)
7322
  {
7323
    if ((key_part->field != pk_part->field) ||
7324
        (key_part->length != pk_part->length))
7325
      return FALSE;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
7326
  }
7327
  return (key_part == key_part_end);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
7328 7329 7330
}


7331 7332
/*
  Create a QUICK_RANGE_SELECT from given key and SEL_ARG tree for that key.
7333

7334 7335
  SYNOPSIS
    get_quick_select()
7336
      param
7337
      idx          Index of used key in param->key.
7338 7339
      key_tree     SEL_ARG tree for the used key
      parent_alloc If not NULL, use it to allocate memory for
7340
                   quick select data. Otherwise use quick->alloc.
7341
  NOTES
7342
    The caller must call QUICK_SELECT::init for returned quick select
7343

7344
    CAUTION! This function may change thd->mem_root to a MEM_ROOT which will be
7345
    deallocated when the returned quick select is deleted.
7346 7347 7348 7349

  RETURN
    NULL on error
    otherwise created quick select
7350
*/
7351

7352 7353 7354
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
7355
{
7356
  QUICK_RANGE_SELECT *quick;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
7357
  DBUG_ENTER("get_quick_select");
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
7358 7359 7360 7361 7362 7363 7364 7365 7366

  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
7367
                                 test(parent_alloc));
7368

pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
7369
  if (quick)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
7370 7371 7372 7373 7374 7375 7376 7377 7378 7379 7380
  {
    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*)
7381 7382 7383 7384
        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
7385
    }
7386
  }
bk@work.mysql.com's avatar
bk@work.mysql.com committed
7387 7388 7389 7390 7391 7392 7393
  DBUG_RETURN(quick);
}


/*
** Fix this to get all possible sub_ranges
*/
7394 7395
bool
get_quick_keys(PARAM *param,QUICK_RANGE_SELECT *quick,KEY_PART *key,
bk@work.mysql.com's avatar
bk@work.mysql.com committed
7396 7397 7398 7399 7400 7401 7402 7403 7404 7405 7406 7407 7408
	       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
7409
  key_tree->store(key[key_tree->part].store_length,
bk@work.mysql.com's avatar
bk@work.mysql.com committed
7410 7411 7412 7413 7414 7415 7416 7417 7418 7419 7420 7421 7422 7423 7424 7425 7426 7427 7428 7429 7430 7431 7432 7433 7434 7435 7436 7437
		  &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
7438 7439 7440 7441
  {
    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
7442

7443 7444 7445 7446 7447
  /*
    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)
7448 7449 7450 7451 7452 7453 7454 7455 7456 7457
  {
    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
7458 7459 7460 7461 7462 7463 7464 7465
  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;
7466 7467
      if ((table_key->flags & (HA_NOSAME | HA_END_SPACE_KEY)) == HA_NOSAME &&
	  key->part == table_key->key_parts-1)
7468 7469 7470 7471 7472 7473 7474 7475 7476
      {
	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
7477 7478 7479 7480
    }
  }

  /* Get range for retrieving rows in QUICK_SELECT::get_next */
7481
  if (!(range= new QUICK_RANGE((const char *) param->min_key,
7482
			       (uint) (tmp_min_key - param->min_key),
7483
			       (const char *) param->max_key,
7484 7485
			       (uint) (tmp_max_key - param->max_key),
			       flag)))
7486 7487
    return 1;			// out of memory

bk@work.mysql.com's avatar
bk@work.mysql.com committed
7488 7489
  set_if_bigger(quick->max_used_key_length,range->min_length);
  set_if_bigger(quick->max_used_key_length,range->max_length);
7490
  set_if_bigger(quick->used_key_parts, (uint) key_tree->part+1);
7491 7492 7493
  if (insert_dynamic(&quick->ranges, (gptr)&range))
    return 1;

bk@work.mysql.com's avatar
bk@work.mysql.com committed
7494 7495 7496 7497 7498 7499 7500 7501 7502 7503 7504 7505
 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
*/

7506
bool QUICK_RANGE_SELECT::unique_key_range()
bk@work.mysql.com's avatar
bk@work.mysql.com committed
7507 7508 7509
{
  if (ranges.elements == 1)
  {
7510 7511
    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
7512 7513
    {
      KEY *key=head->key_info+index;
7514
      return ((key->flags & (HA_NOSAME | HA_END_SPACE_KEY)) == HA_NOSAME &&
bk@work.mysql.com's avatar
bk@work.mysql.com committed
7515 7516 7517 7518 7519 7520
	      key->key_length == tmp->min_length);
    }
  }
  return 0;
}

7521

monty@mysql.com's avatar
monty@mysql.com committed
7522
/* Returns TRUE if any part of the key is NULL */
7523 7524 7525

static bool null_part_in_key(KEY_PART *key_part, const char *key, uint length)
{
7526
  for (const char *end=key+length ;
7527
       key < end;
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
7528
       key+= key_part++->store_length)
7529
  {
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
7530 7531
    if (key_part->null_bit && *key)
      return 1;
7532 7533 7534 7535
  }
  return 0;
}

pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
7536

7537 7538
bool QUICK_SELECT_I::check_if_keys_used(List<Item> *fields)
{
7539
  return check_if_key_used(head, index, *fields);
7540 7541 7542 7543 7544 7545 7546 7547 7548 7549 7550 7551 7552 7553 7554 7555 7556 7557 7558 7559 7560 7561 7562 7563 7564 7565 7566 7567 7568 7569 7570 7571 7572 7573 7574 7575 7576 7577
}

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

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

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

monty@mysql.com's avatar
monty@mysql.com committed
7578

sergefp@mysql.com's avatar
sergefp@mysql.com committed
7579 7580
/*
  Create quick select from ref/ref_or_null scan.
7581

sergefp@mysql.com's avatar
sergefp@mysql.com committed
7582 7583 7584 7585 7586 7587 7588 7589 7590 7591 7592 7593 7594 7595 7596
  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
7597

7598
QUICK_RANGE_SELECT *get_quick_select_for_ref(THD *thd, TABLE *table,
sergefp@mysql.com's avatar
sergefp@mysql.com committed
7599
                                             TABLE_REF *ref, ha_rows records)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
7600
{
7601 7602
  MEM_ROOT *old_root, *alloc;
  QUICK_RANGE_SELECT *quick;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
7603 7604
  KEY *key_info = &table->key_info[ref->key];
  KEY_PART *key_part;
serg@serg.mylan's avatar
serg@serg.mylan committed
7605
  QUICK_RANGE *range;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
7606
  uint part;
7607 7608 7609 7610 7611 7612

  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;
7613 7614 7615 7616 7617
  /*
    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
7618 7619

  if (!quick)
7620
    return 0;			/* no ranges found */
sergefp@mysql.com's avatar
sergefp@mysql.com committed
7621
  if (quick->init())
monty@mysql.com's avatar
monty@mysql.com committed
7622
    goto err;
sergefp@mysql.com's avatar
sergefp@mysql.com committed
7623
  quick->records= records;
7624

7625
  if (cp_buffer_from_ref(thd, table, ref) && thd->is_fatal_error ||
7626
      !(range= new(alloc) QUICK_RANGE()))
monty@mysql.com's avatar
monty@mysql.com committed
7627
    goto err;                                   // out of memory
7628

bk@work.mysql.com's avatar
bk@work.mysql.com committed
7629 7630 7631
  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 &&
7632 7633
		 (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
7634 7635

  if (!(quick->key_parts=key_part=(KEY_PART *)
7636
	alloc_root(&quick->alloc,sizeof(KEY_PART)*ref->key_parts)))
bk@work.mysql.com's avatar
bk@work.mysql.com committed
7637 7638 7639 7640 7641 7642
    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
7643 7644
    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
7645 7646
    key_part->null_bit=     key_info->key_part[part].null_bit;
  }
pem@mysql.com's avatar
pem@mysql.com committed
7647
  if (insert_dynamic(&quick->ranges,(gptr)&range))
7648 7649
    goto err;

7650
  /*
7651 7652 7653 7654 7655
     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.
  */
7656 7657 7658 7659 7660
  if (ref->null_ref_key)
  {
    QUICK_RANGE *null_range;

    *ref->null_ref_key= 1;		// Set null byte then create a range
7661 7662 7663 7664 7665
    if (!(null_range= new (alloc) QUICK_RANGE((char*)ref->key_buff,
                                              ref->key_length,
                                              (char*)ref->key_buff,
                                              ref->key_length,
                                              EQ_RANGE)))
7666 7667
      goto err;
    *ref->null_ref_key= 0;		// Clear null byte
pem@mysql.com's avatar
pem@mysql.com committed
7668
    if (insert_dynamic(&quick->ranges,(gptr)&null_range))
7669 7670 7671 7672
      goto err;
  }

  return quick;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
7673 7674 7675 7676 7677 7678

err:
  delete quick;
  return 0;
}

7679 7680

/*
sergefp@mysql.com's avatar
sergefp@mysql.com committed
7681 7682 7683 7684 7685 7686
  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.
  
7687
  If table has a clustered primary key that covers all rows (TRUE for bdb
7688 7689 7690
  and innodb currently) and one of the index_merge scans is a scan on PK,
  then 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.
7691

7692 7693 7694
  RETURN
    0     OK
    other error
7695
*/
7696

sergefp@mysql.com's avatar
sergefp@mysql.com committed
7697
int QUICK_INDEX_MERGE_SELECT::read_keys_and_merge()
7698
{
sergefp@mysql.com's avatar
sergefp@mysql.com committed
7699 7700
  List_iterator_fast<QUICK_RANGE_SELECT> cur_quick_it(quick_selects);
  QUICK_RANGE_SELECT* cur_quick;
7701
  int result;
sergefp@mysql.com's avatar
sergefp@mysql.com committed
7702
  Unique *unique;
7703 7704
  handler *file= head->file;
  DBUG_ENTER("QUICK_INDEX_MERGE_SELECT::read_keys_and_merge");
7705

7706 7707
  file->extra(HA_EXTRA_KEYREAD);
  head->prepare_for_position();
7708

sergefp@mysql.com's avatar
sergefp@mysql.com committed
7709 7710
  cur_quick_it.rewind();
  cur_quick= cur_quick_it++;
7711
  DBUG_ASSERT(cur_quick != 0);
sergefp@mysql.com's avatar
sergefp@mysql.com committed
7712 7713 7714 7715 7716
  
  /*
    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
7717
  if (cur_quick->init() || cur_quick->reset())
sergefp@mysql.com's avatar
sergefp@mysql.com committed
7718
    DBUG_RETURN(1);
7719

7720 7721
  unique= new Unique(refpos_order_cmp, (void *)file,
                     file->ref_length,
7722
                     thd->variables.sortbuff_size);
7723 7724
  if (!unique)
    DBUG_RETURN(1);
monty@mysql.com's avatar
monty@mysql.com committed
7725
  for (;;)
7726
  {
sergefp@mysql.com's avatar
sergefp@mysql.com committed
7727
    while ((result= cur_quick->get_next()) == HA_ERR_END_OF_FILE)
7728
    {
sergefp@mysql.com's avatar
sergefp@mysql.com committed
7729 7730 7731
      cur_quick->range_end();
      cur_quick= cur_quick_it++;
      if (!cur_quick)
7732
        break;
7733

sergefp@mysql.com's avatar
sergefp@mysql.com committed
7734 7735
      if (cur_quick->file->inited != handler::NONE) 
        cur_quick->file->ha_index_end();
sergefp@mysql.com's avatar
sergefp@mysql.com committed
7736
      if (cur_quick->init() || cur_quick->reset())
7737
        DBUG_RETURN(1);
7738 7739 7740
    }

    if (result)
7741
    {
7742
      if (result != HA_ERR_END_OF_FILE)
sergefp@mysql.com's avatar
sergefp@mysql.com committed
7743 7744
      {
        cur_quick->range_end();
7745
        DBUG_RETURN(result);
sergefp@mysql.com's avatar
sergefp@mysql.com committed
7746
      }
7747
      break;
7748
    }
7749

7750 7751
    if (thd->killed)
      DBUG_RETURN(1);
7752

7753
    /* skip row if it will be retrieved by clustered PK scan */
7754 7755
    if (pk_quick_select && pk_quick_select->row_in_ranges())
      continue;
7756

sergefp@mysql.com's avatar
sergefp@mysql.com committed
7757 7758
    cur_quick->file->position(cur_quick->record);
    result= unique->unique_add((char*)cur_quick->file->ref);
7759
    if (result)
7760 7761
      DBUG_RETURN(1);

monty@mysql.com's avatar
monty@mysql.com committed
7762
  }
7763

7764
  DBUG_PRINT("info", ("ok"));
7765 7766
  /* ok, all row ids are in Unique */
  result= unique->get(head);
sergefp@mysql.com's avatar
sergefp@mysql.com committed
7767
  delete unique;
monty@mysql.com's avatar
monty@mysql.com committed
7768
  doing_pk_scan= FALSE;
7769 7770
  /* index_merge currently doesn't support "using index" at all */
  file->extra(HA_EXTRA_NO_KEYREAD);
monty@mysql.com's avatar
monty@mysql.com committed
7771 7772
  /* start table scan */
  init_read_record(&read_record, thd, head, (SQL_SELECT*) 0, 1, 1);
7773 7774 7775
  DBUG_RETURN(result);
}

7776

7777 7778 7779
/*
  Get next row for index_merge.
  NOTES
7780 7781 7782 7783
    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.
7784
*/
7785

7786 7787
int QUICK_INDEX_MERGE_SELECT::get_next()
{
7788
  int result;
7789
  DBUG_ENTER("QUICK_INDEX_MERGE_SELECT::get_next");
7790

7791 7792 7793
  if (doing_pk_scan)
    DBUG_RETURN(pk_quick_select->get_next());

7794
  if ((result= read_record.read_record(&read_record)) == -1)
7795 7796 7797
  {
    result= HA_ERR_END_OF_FILE;
    end_read_record(&read_record);
7798
    /* All rows from Unique have been retrieved, do a clustered PK scan */
monty@mysql.com's avatar
monty@mysql.com committed
7799
    if (pk_quick_select)
7800
    {
monty@mysql.com's avatar
monty@mysql.com committed
7801
      doing_pk_scan= TRUE;
7802 7803
      if ((result= pk_quick_select->init()) ||
          (result= pk_quick_select->reset()))
7804 7805 7806 7807 7808 7809
        DBUG_RETURN(result);
      DBUG_RETURN(pk_quick_select->get_next());
    }
  }

  DBUG_RETURN(result);
7810 7811
}

7812 7813

/*
7814
  Retrieve next record.
7815
  SYNOPSIS
7816 7817
     QUICK_ROR_INTERSECT_SELECT::get_next()

7818
  NOTES
7819 7820
    Invariant on enter/exit: all intersected selects have retrieved all index
    records with rowid <= some_rowid_val and no intersected select has
7821 7822 7823 7824
    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.

7825
    If a Clustered PK scan is present, it is used only to check if row
7826 7827 7828 7829 7830
    satisfies its condition (and never used for row retrieval).

  RETURN
   0     - Ok
   other - Error code if any error occurred.
7831 7832 7833 7834 7835 7836 7837 7838 7839
*/

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");
7840

7841
  do
7842
  {
7843 7844
    /* Get a rowid for first quick and save it as a 'candidate' */
    quick= quick_it++;
7845
    error= quick->get_next();
7846 7847
    if (cpk_quick)
    {
7848
      while (!error && !cpk_quick->row_in_ranges())
7849 7850 7851 7852
        error= quick->get_next();
    }
    if (error)
      DBUG_RETURN(error);
7853

7854 7855 7856
    quick->file->position(quick->record);
    memcpy(last_rowid, quick->file->ref, head->file->ref_length);
    last_rowid_count= 1;
7857

7858
    while (last_rowid_count < quick_selects.elements)
7859
    {
7860 7861 7862 7863 7864
      if (!(quick= quick_it++))
      {
        quick_it.rewind();
        quick= quick_it++;
      }
7865

7866 7867 7868 7869 7870 7871 7872 7873 7874 7875
      do
      {
        if ((error= quick->get_next()))
          DBUG_RETURN(error);
        quick->file->position(quick->record);
        cmp= head->file->cmp_ref(quick->file->ref, last_rowid);
      } while (cmp < 0);

      /* Ok, current select 'caught up' and returned ref >= cur_ref */
      if (cmp > 0)
7876
      {
7877 7878
        /* Found a row with ref > cur_ref. Make it a new 'candidate' */
        if (cpk_quick)
7879
        {
7880 7881 7882 7883 7884
          while (!cpk_quick->row_in_ranges())
          {
            if ((error= quick->get_next()))
              DBUG_RETURN(error);
          }
7885
        }
7886 7887 7888 7889 7890 7891 7892
        memcpy(last_rowid, quick->file->ref, head->file->ref_length);
        last_rowid_count= 1;
      }
      else
      {
        /* current 'candidate' row confirmed by this select */
        last_rowid_count++;
7893 7894 7895
      }
    }

7896
    /* We get here if we got the same row ref in all scans. */
7897 7898 7899
    if (need_to_fetch_row)
      error= head->file->rnd_pos(head->record[0], last_rowid);
  } while (error == HA_ERR_RECORD_DELETED);
7900 7901 7902 7903
  DBUG_RETURN(error);
}


7904 7905
/*
  Retrieve next record.
7906 7907
  SYNOPSIS
    QUICK_ROR_UNION_SELECT::get_next()
7908

7909
  NOTES
7910 7911
    Enter/exit invariant:
    For each quick select in the queue a {key,rowid} tuple has been
7912
    retrieved but the corresponding row hasn't been passed to output.
7913

7914
  RETURN
7915 7916
   0     - Ok
   other - Error code if any error occurred.
7917 7918 7919 7920 7921 7922 7923 7924
*/

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");
7925

7926 7927
  do
  {
7928 7929 7930 7931 7932
    do
    {
      if (!queue.elements)
        DBUG_RETURN(HA_ERR_END_OF_FILE);
      /* Ok, we have a queue with >= 1 scans */
7933

7934 7935
      quick= (QUICK_SELECT_I*)queue_top(&queue);
      memcpy(cur_rowid, quick->last_rowid, rowid_length);
7936

7937 7938 7939 7940 7941 7942 7943 7944 7945 7946 7947 7948
      /* 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);
      }
7949

7950 7951 7952 7953 7954 7955 7956 7957 7958
      if (!have_prev_rowid)
      {
        /* No rows have been returned yet */
        dup_row= FALSE;
        have_prev_rowid= TRUE;
      }
      else
        dup_row= !head->file->cmp_ref(cur_rowid, prev_rowid);
    } while (dup_row);
7959

7960 7961 7962
    tmp= cur_rowid;
    cur_rowid= prev_rowid;
    prev_rowid= tmp;
7963

7964 7965
    error= head->file->rnd_pos(quick->record, prev_rowid);
  } while (error == HA_ERR_RECORD_DELETED);
7966 7967 7968
  DBUG_RETURN(error);
}

7969

sergefp@mysql.com's avatar
sergefp@mysql.com committed
7970
int QUICK_RANGE_SELECT::reset()
ingo@mysql.com's avatar
ingo@mysql.com committed
7971 7972 7973
{
  uint  mrange_bufsiz;
  byte  *mrange_buff;
sergefp@mysql.com's avatar
sergefp@mysql.com committed
7974 7975 7976
  DBUG_ENTER("QUICK_RANGE_SELECT::reset");
  next=0;
  range= NULL;
7977
  in_range= FALSE;
sergefp@mysql.com's avatar
sergefp@mysql.com committed
7978
  cur_range= (QUICK_RANGE**) ranges.buffer;
igor@rurik.mysql.com's avatar
igor@rurik.mysql.com committed
7979

7980
  if (file->inited == handler::NONE && (error= file->ha_index_init(index,1)))
igor@rurik.mysql.com's avatar
igor@rurik.mysql.com committed
7981
    DBUG_RETURN(error);
igor@rurik.mysql.com's avatar
igor@rurik.mysql.com committed
7982
 
ingo@mysql.com's avatar
ingo@mysql.com committed
7983 7984 7985 7986 7987 7988 7989
  /* 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
7990 7991
  /* Allocate the ranges array. */
  DBUG_ASSERT(ranges.elements);
ingo@mysql.com's avatar
ingo@mysql.com committed
7992 7993 7994 7995 7996 7997 7998 7999 8000 8001 8002 8003 8004 8005 8006 8007
  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
8008
  /* Allocate the handler buffer if necessary.  */
8009
  if (file->ha_table_flags() & HA_NEED_READ_RANGE_BUFFER)
ingo@mysql.com's avatar
ingo@mysql.com committed
8010 8011
  {
    mrange_bufsiz= min(multi_range_bufsiz,
joreland@mysql.com's avatar
merge  
joreland@mysql.com committed
8012
                       (QUICK_SELECT_I::records + 1)* head->s->reclength);
ingo@mysql.com's avatar
ingo@mysql.com committed
8013 8014 8015 8016 8017 8018 8019 8020 8021 8022 8023 8024 8025 8026 8027 8028 8029 8030 8031 8032 8033 8034

    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;
8035 8036 8037 8038 8039 8040 8041 8042
#ifdef HAVE_purify
    /*
      We need this until ndb will use the buffer efficiently
      (Now ndb stores  complete row in here, instead of only the used fields
      which gives us valgrind warnings in compare_record[])
    */
    bzero((char*) mrange_buff, mrange_bufsiz);
#endif
ingo@mysql.com's avatar
ingo@mysql.com committed
8043 8044 8045 8046 8047 8048 8049 8050 8051 8052 8053 8054 8055 8056 8057 8058 8059 8060 8061
  }
  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
8062

8063
int QUICK_RANGE_SELECT::get_next()
bk@work.mysql.com's avatar
bk@work.mysql.com committed
8064
{
ingo@mysql.com's avatar
ingo@mysql.com committed
8065 8066 8067 8068
  int             result;
  KEY_MULTI_RANGE *mrange;
  key_range       *start_key;
  key_range       *end_key;
8069
  DBUG_ENTER("QUICK_RANGE_SELECT::get_next");
ingo@mysql.com's avatar
ingo@mysql.com committed
8070 8071 8072
  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
8073

8074 8075 8076 8077 8078 8079 8080 8081 8082
  if (in_ror_merged_scan)
  {
    /*
      We don't need to signal the bitmap change as the bitmap is always the
      same for this head->file
    */
    head->column_bitmaps_set_no_signal(&column_bitmap, &column_bitmap);
  }

bk@work.mysql.com's avatar
bk@work.mysql.com committed
8083 8084
  for (;;)
  {
ingo@mysql.com's avatar
ingo@mysql.com committed
8085
    if (in_range)
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
8086
    {
ingo@mysql.com's avatar
ingo@mysql.com committed
8087 8088
      /* 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
8089
      if (result != HA_ERR_END_OF_FILE)
8090
        goto end;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
8091
    }
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
8092

ingo@mysql.com's avatar
ingo@mysql.com committed
8093 8094 8095 8096 8097 8098
    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;
8099 8100
      if (in_ror_merged_scan)
        head->column_bitmaps_set_no_signal(save_read_set, save_write_set);
ingo@mysql.com's avatar
ingo@mysql.com committed
8101 8102 8103 8104 8105 8106 8107 8108 8109 8110 8111 8112 8113 8114 8115 8116 8117 8118 8119 8120 8121 8122 8123 8124
      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
8125

ingo@mysql.com's avatar
ingo@mysql.com committed
8126 8127
      mrange_slot->range_flag= range->flag;
    }
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
8128

ingo@mysql.com's avatar
ingo@mysql.com committed
8129 8130
    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
8131
    if (result != HA_ERR_END_OF_FILE)
8132
      goto end;
ingo@mysql.com's avatar
ingo@mysql.com committed
8133
    in_range= FALSE; /* No matching rows; go to next set of ranges. */
bk@work.mysql.com's avatar
bk@work.mysql.com committed
8134
  }
8135 8136 8137 8138 8139 8140 8141 8142 8143

end:
  in_range= ! result;
  if (in_ror_merged_scan)
  {
    /* Restore bitmaps set on entry */
    head->column_bitmaps_set_no_signal(save_read_set, save_write_set);
  }
  DBUG_RETURN(result);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
8144 8145
}

8146 8147 8148 8149 8150 8151
/*
  Get the next record with a different prefix.

  SYNOPSIS
    QUICK_RANGE_SELECT::get_next_prefix()
    prefix_length  length of cur_prefix
8152
    cur_prefix     prefix of a key to be searched for
8153 8154 8155 8156 8157 8158 8159 8160 8161 8162 8163 8164 8165 8166 8167 8168 8169 8170 8171 8172 8173 8174 8175 8176 8177 8178 8179 8180 8181 8182 8183 8184

  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. */
8185
      DBUG_ASSERT(cur_prefix != 0);
8186 8187 8188 8189 8190 8191
      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
8192 8193 8194 8195 8196 8197 8198 8199
    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++);
8200 8201 8202 8203 8204 8205 8206 8207 8208 8209 8210 8211 8212 8213 8214 8215 8216 8217 8218 8219 8220 8221 8222 8223 8224 8225 8226 8227 8228

    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
8229
/* Get next for geometrical indexes */
bk@work.mysql.com's avatar
bk@work.mysql.com committed
8230

pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
8231
int QUICK_RANGE_SELECT_GEOM::get_next()
bk@work.mysql.com's avatar
bk@work.mysql.com committed
8232
{
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
8233
  DBUG_ENTER("QUICK_RANGE_SELECT_GEOM::get_next");
bk@work.mysql.com's avatar
bk@work.mysql.com committed
8234

pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
8235
  for (;;)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
8236
  {
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
8237 8238
    int result;
    if (range)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
8239
    {
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
8240 8241 8242 8243 8244
      // 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
8245
    }
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
8246

ingo@mysql.com's avatar
ingo@mysql.com committed
8247 8248 8249 8250 8251 8252 8253 8254
    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
8255 8256 8257 8258 8259

    result= file->index_read(record,
			     (byte*) range->min_key,
			     range->min_length,
			     (ha_rkey_function)(range->flag ^ GEOM_FLAG));
8260
    if (result != HA_ERR_KEY_NOT_FOUND && result != HA_ERR_END_OF_FILE)
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
8261 8262
      DBUG_RETURN(result);
    range=0;				// Not found, to next range
bk@work.mysql.com's avatar
bk@work.mysql.com committed
8263 8264 8265
  }
}

8266

8267 8268 8269 8270
/*
  Check if current row will be retrieved by this QUICK_RANGE_SELECT

  NOTES
8271 8272
    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
8273
    quick select.
8274
    The implementation does a binary search on sorted array of disjoint
8275 8276
    ranges, without taking size of range into account.

8277
    This function is used to filter out clustered PK scan rows in
8278 8279
    index_merge quick select.

8280
  RETURN
monty@mysql.com's avatar
monty@mysql.com committed
8281 8282
    TRUE  if current row will be retrieved by this quick select
    FALSE if not
8283 8284 8285 8286 8287 8288 8289 8290 8291 8292
*/

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)
8293
  {
8294 8295 8296 8297 8298 8299 8300 8301 8302 8303 8304 8305 8306
    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));
}

8307
/*
8308 8309 8310 8311 8312 8313 8314
  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.
8315
 */
8316

8317
QUICK_SELECT_DESC::QUICK_SELECT_DESC(QUICK_RANGE_SELECT *q,
8318
                                     uint used_key_parts)
8319
 :QUICK_RANGE_SELECT(*q), rev_it(rev_ranges)
8320
{
monty@hundin.mysql.fi's avatar
monty@hundin.mysql.fi committed
8321
  QUICK_RANGE *r;
8322

8323 8324
  QUICK_RANGE **pr= (QUICK_RANGE**)ranges.buffer;
  QUICK_RANGE **last_range= pr + ranges.elements;
monty@mysql.com's avatar
monty@mysql.com committed
8325 8326
  for (; pr!=last_range; pr++)
    rev_ranges.push_front(*pr);
monty@hundin.mysql.fi's avatar
monty@hundin.mysql.fi committed
8327

8328
  /* Remove EQ_RANGE flag for keys that are not using the full key */
monty@hundin.mysql.fi's avatar
monty@hundin.mysql.fi committed
8329
  for (r = rev_it++; r; r = rev_it++)
8330 8331 8332 8333 8334 8335 8336 8337
  {
    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;
8338 8339
}

8340

8341 8342 8343 8344 8345 8346
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
8347 8348
   *   - 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
8349 8350 8351 8352 8353 8354 8355 8356 8357 8358 8359 8360
   *     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
8361 8362 8363
      result = ((range->flag & EQ_RANGE)
		? file->index_next_same(record, (byte*) range->min_key,
					range->min_length) :
8364 8365 8366 8367 8368 8369 8370 8371 8372 8373 8374 8375 8376 8377 8378
		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
    {
8379 8380 8381
      int local_error;
      if ((local_error=file->index_last(record)))
	DBUG_RETURN(local_error);		// Empty table
8382 8383 8384 8385 8386 8387
      if (cmp_prev(range) == 0)
	DBUG_RETURN(0);
      range=0;			// No matching records; go to next range
      continue;
    }

8388
    if (range->flag & EQ_RANGE)
8389 8390 8391 8392 8393 8394
    {
      result = file->index_read(record, (byte*) range->max_key,
				range->max_length, HA_READ_KEY_EXACT);
    }
    else
    {
8395 8396 8397 8398 8399
      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));
8400 8401 8402
    }
    if (result)
    {
8403
      if (result != HA_ERR_KEY_NOT_FOUND && result != HA_ERR_END_OF_FILE)
8404 8405 8406 8407 8408 8409 8410 8411 8412 8413 8414 8415 8416 8417
	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
  }
}

8418

pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
8419 8420 8421 8422 8423 8424 8425 8426 8427 8428 8429 8430 8431 8432 8433 8434 8435 8436 8437 8438 8439 8440 8441 8442 8443 8444 8445 8446 8447 8448 8449 8450 8451 8452 8453 8454 8455 8456 8457 8458 8459
/*
  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
}


8460
/*
8461 8462 8463
  Returns 0 if found key is inside range (found key >= range->min_key).
*/

8464
int QUICK_RANGE_SELECT::cmp_prev(QUICK_RANGE *range_arg)
8465
{
8466
  int cmp;
8467
  if (range_arg->flag & NO_MIN_RANGE)
8468
    return 0;					/* key can't be to small */
8469

monty@mysql.com's avatar
monty@mysql.com committed
8470 8471
  cmp= key_cmp(key_part_info, (byte*) range_arg->min_key,
               range_arg->min_length);
8472 8473 8474
  if (cmp > 0 || cmp == 0 && !(range_arg->flag & NEAR_MIN))
    return 0;
  return 1;                                     // outside of range
8475 8476
}

8477

8478
/*
monty@mysql.com's avatar
monty@mysql.com committed
8479
 * TRUE if this range will require using HA_READ_AFTER_KEY
8480
   See comment in get_next() about this
8481
 */
8482

8483
bool QUICK_SELECT_DESC::range_reads_after_key(QUICK_RANGE *range_arg)
8484
{
8485
  return ((range_arg->flag & (NO_MAX_RANGE | NEAR_MAX)) ||
8486
	  !(range_arg->flag & EQ_RANGE) ||
8487
	  head->key_info[index].key_length != range_arg->max_length) ? 1 : 0;
8488 8489
}

8490

monty@mysql.com's avatar
monty@mysql.com committed
8491
/* TRUE if we are reading over a key that may have a NULL value */
8492

8493
#ifdef NOT_USED
8494
bool QUICK_SELECT_DESC::test_if_null_range(QUICK_RANGE *range_arg,
8495 8496
					   uint used_key_parts)
{
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
8497
  uint offset, end;
8498 8499 8500
  KEY_PART *key_part = key_parts,
           *key_part_end= key_part+used_key_parts;

8501
  for (offset= 0,  end = min(range_arg->min_length, range_arg->max_length) ;
8502
       offset < end && key_part != key_part_end ;
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
8503
       offset+= key_part++->store_length)
8504
  {
8505 8506
    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
8507
		key_part->store_length))
8508
      continue;
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
8509 8510

    if (key_part->null_bit && range_arg->min_key[offset])
8511 8512 8513 8514 8515 8516 8517 8518 8519 8520 8521 8522
      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)
  {
8523
    if (offset >= range_arg->min_length || range_arg->min_key[offset])
8524 8525 8526 8527 8528 8529 8530 8531 8532 8533 8534 8535
      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;
}
8536
#endif
8537 8538


8539 8540 8541 8542 8543 8544 8545 8546 8547
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
8548
  bool first= TRUE;
8549
  List_iterator_fast<QUICK_RANGE_SELECT> it(quick_selects);
8550
  str->append(STRING_WITH_LEN("sort_union("));
8551 8552 8553 8554 8555
  while ((quick= it++))
  {
    if (!first)
      str->append(',');
    else
monty@mysql.com's avatar
monty@mysql.com committed
8556
      first= FALSE;
8557 8558 8559 8560 8561 8562 8563 8564 8565 8566 8567 8568
    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)
{
8569
  bool first= TRUE;
8570 8571
  QUICK_RANGE_SELECT *quick;
  List_iterator_fast<QUICK_RANGE_SELECT> it(quick_selects);
8572
  str->append(STRING_WITH_LEN("intersect("));
8573 8574 8575 8576 8577
  while ((quick= it++))
  {
    KEY *key_info= head->key_info + quick->index;
    if (!first)
      str->append(',');
8578
    else
monty@mysql.com's avatar
monty@mysql.com committed
8579
      first= FALSE;
8580 8581 8582 8583 8584 8585 8586 8587 8588 8589 8590 8591 8592
    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)
{
8593
  bool first= TRUE;
8594 8595
  QUICK_SELECT_I *quick;
  List_iterator_fast<QUICK_SELECT_I> it(quick_selects);
8596
  str->append(STRING_WITH_LEN("union("));
8597 8598 8599 8600 8601
  while ((quick= it++))
  {
    if (!first)
      str->append(',');
    else
monty@mysql.com's avatar
monty@mysql.com committed
8602
      first= FALSE;
8603 8604 8605 8606 8607 8608
    quick->add_info_string(str);
  }
  str->append(')');
}


8609
void QUICK_RANGE_SELECT::add_keys_and_lengths(String *key_names,
8610
                                              String *used_lengths)
8611 8612 8613 8614 8615 8616 8617 8618 8619
{
  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);
}

8620 8621
void QUICK_INDEX_MERGE_SELECT::add_keys_and_lengths(String *key_names,
                                                    String *used_lengths)
8622 8623 8624
{
  char buf[64];
  uint length;
monty@mysql.com's avatar
monty@mysql.com committed
8625
  bool first= TRUE;
8626
  QUICK_RANGE_SELECT *quick;
8627

8628 8629 8630
  List_iterator_fast<QUICK_RANGE_SELECT> it(quick_selects);
  while ((quick= it++))
  {
8631
    if (first)
monty@mysql.com's avatar
monty@mysql.com committed
8632
      first= FALSE;
8633 8634
    else
    {
8635 8636
      key_names->append(',');
      used_lengths->append(',');
8637
    }
8638

8639 8640
    KEY *key_info= head->key_info + quick->index;
    key_names->append(key_info->name);
8641 8642 8643 8644 8645 8646 8647 8648 8649 8650 8651 8652 8653 8654
    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);
  }
}

8655 8656
void QUICK_ROR_INTERSECT_SELECT::add_keys_and_lengths(String *key_names,
                                                      String *used_lengths)
8657 8658 8659
{
  char buf[64];
  uint length;
8660
  bool first= TRUE;
8661 8662 8663 8664 8665 8666
  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
8667
      first= FALSE;
8668
    else
8669 8670
    {
      key_names->append(',');
8671
      used_lengths->append(',');
8672 8673
    }
    key_names->append(key_info->name);
8674 8675 8676
    length= longlong2str(quick->max_used_key_length, buf, 10) - buf;
    used_lengths->append(buf, length);
  }
8677

8678 8679 8680 8681 8682 8683 8684 8685 8686 8687 8688
  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);
  }
}

8689 8690
void QUICK_ROR_UNION_SELECT::add_keys_and_lengths(String *key_names,
                                                  String *used_lengths)
8691
{
8692
  bool first= TRUE;
8693 8694 8695 8696 8697
  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
8698
      first= FALSE;
8699
    else
8700
    {
8701 8702 8703
      used_lengths->append(',');
      key_names->append(',');
    }
8704
    quick->add_keys_and_lengths(key_names, used_lengths);
8705 8706 8707
  }
}

8708 8709 8710 8711 8712 8713 8714 8715 8716

/*******************************************************************************
* 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
8717
get_constant_key_infix(KEY *index_info, SEL_ARG *index_range_tree,
8718
                       KEY_PART_INFO *first_non_group_part,
8719 8720 8721 8722
                       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);
8723
static bool
8724 8725
check_group_min_max_predicates(COND *cond, Item_field *min_max_arg_item,
                               Field::imagetype image_type);
8726

8727 8728 8729 8730 8731 8732
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);
8733

timour@mysql.com's avatar
timour@mysql.com committed
8734

8735 8736 8737 8738 8739 8740 8741 8742 8743 8744 8745 8746 8747 8748 8749 8750 8751 8752 8753 8754 8755 8756 8757 8758 8759
/*
  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
8760 8761
        - NGA = QA - (GA union C) = {NG_1, ..., NG_m} - the ones not in
          GROUP BY and not referenced by MIN/MAX functions.
8762
        with the following properties specified below.
8763 8764
    B3. If Q has a GROUP BY WITH ROLLUP clause the access method is not 
        applicable.
8765 8766 8767 8768 8769 8770 8771 8772 8773 8774

    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
8775 8776
         - C IS NOT NULL
         - C != const
8777 8778 8779
    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.
8780
    SA5. The select list in DISTINCT queries should not contain expressions.
8781 8782 8783 8784 8785 8786 8787 8788 8789 8790 8791 8792 8793 8794
    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,
8795 8796 8797 8798
         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>.
8799 8800 8801 8802 8803 8804 8805
    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
8806 8807 8808 8809
         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).
8810 8811

    C) Overall query form:
8812 8813 8814 8815
       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)]
8816 8817
         [AND PC(C)]
         [AND PA(A_i1,...,A_iq)]
8818 8819 8820 8821
       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:
8822 8823 8824 8825 8826 8827 8828 8829 8830 8831 8832 8833 8834 8835 8836 8837 8838 8839 8840 8841 8842 8843 8844 8845 8846 8847 8848 8849 8850
       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?
8851 8852
  - Lift the limitation in condition (B3), that is, make this access method 
    applicable to ROLLUP queries.
8853 8854 8855 8856 8857 8858 8859 8860 8861 8862 8863 8864 8865 8866 8867 8868 8869 8870

  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. */
8871
  Item_field *min_max_arg_item= NULL; // The argument of all MIN/MAX functions
8872 8873 8874 8875
  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. */
8876
  uint group_key_parts= 0;  // Number of index key parts in the group prefix.
8877 8878 8879 8880 8881 8882 8883 8884 8885 8886 8887 8888 8889 8890 8891
  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. */
8892 8893
       (!join->select_distinct)) ||
      (thd->lex->select_lex.olap == ROLLUP_TYPE)) /* Check (B3) for ROLLUP */
8894
    DBUG_RETURN(NULL);
8895
  if (table->s->keys == 0)        /* There are no indexes to use. */
8896 8897 8898
    DBUG_RETURN(NULL);

  /* Analyze the query in more detail. */
8899
  List_iterator<Item> select_items_it(join->fields_list);
8900

8901
  /* Check (SA1,SA4) and store the only MIN/MAX argument - the C attribute.*/
monty@mishka.local's avatar
monty@mishka.local committed
8902
  if (join->make_sum_func_list(join->all_fields, join->fields_list, 1))
8903 8904
    DBUG_RETURN(NULL);
  if (join->sum_funcs[0])
8905
  {
8906 8907 8908
    Item_sum *min_max_item;
    Item_sum **func_ptr= join->sum_funcs;
    while ((min_max_item= *(func_ptr++)))
8909
    {
8910 8911 8912 8913 8914
      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
8915 8916
        DBUG_RETURN(NULL);

8917 8918
      Item *expr= min_max_item->args[0];    /* The argument of MIN/MAX. */
      if (expr->type() == Item::FIELD_ITEM) /* Is it an attribute? */
8919
      {
8920 8921 8922 8923
        if (! min_max_arg_item)
          min_max_arg_item= (Item_field*) expr;
        else if (! min_max_arg_item->eq(expr, 1))
          DBUG_RETURN(NULL);
8924
      }
8925 8926
      else
        DBUG_RETURN(NULL);
8927
    }
8928
  }
8929

8930 8931 8932 8933
  /* Check (SA5). */
  if (join->select_distinct)
  {
    while ((item= select_items_it++))
8934
    {
8935 8936
      if (item->type() != Item::FIELD_ITEM)
        DBUG_RETURN(NULL);
8937 8938 8939 8940 8941 8942 8943 8944 8945 8946 8947 8948 8949 8950 8951 8952
    }
  }

  /* 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;
8953
  KEY *cur_index_info_end= cur_index_info + table->s->keys;
8954
  KEY_PART_INFO *cur_part= NULL;
8955 8956 8957 8958 8959 8960 8961 8962 8963 8964 8965 8966 8967 8968 8969 8970 8971 8972
  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;
8973
  uint cur_param_idx=MAX_KEY;
timour@mysql.com's avatar
timour@mysql.com committed
8974
  key_map cur_used_key_parts;
timour@mysql.com's avatar
timour@mysql.com committed
8975
  uint pk= param->table->s->primary_key;
8976 8977 8978 8979 8980 8981 8982

  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;
8983

timour@mysql.com's avatar
timour@mysql.com committed
8984 8985 8986 8987 8988 8989 8990 8991 8992 8993
    /*
      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 &&
8994
        (table->file->ha_table_flags() & HA_PRIMARY_KEY_IN_READ_INDEX))
timour@mysql.com's avatar
timour@mysql.com committed
8995 8996 8997 8998 8999 9000
    {
      /* For each table field */
      for (uint i= 0; i < table->s->fields; i++)
      {
        Field *cur_field= table->field[i];
        /*
9001 9002
          If the field is used in the current query ensure that it's
          part of 'cur_index'
timour@mysql.com's avatar
timour@mysql.com committed
9003
        */
9004 9005 9006
        if (bitmap_is_set(table->read_set, cur_field->field_index) &&
            !cur_field->part_of_key_not_clustered.is_set(cur_index))
          goto next_index;                  // Field was not part of key
timour@mysql.com's avatar
timour@mysql.com committed
9007 9008 9009
      }
    }

9010 9011 9012 9013 9014 9015 9016 9017 9018 9019 9020 9021 9022 9023 9024 9025 9026 9027 9028 9029 9030
    /*
      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))
        {
9031 9032
          cur_group_prefix_len+= cur_part->store_length;
          ++cur_group_key_parts;
9033 9034 9035 9036 9037 9038 9039 9040 9041 9042 9043 9044 9045 9046 9047
        }
        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)
    {
9048
      select_items_it.rewind();
timour@mysql.com's avatar
timour@mysql.com committed
9049
      cur_used_key_parts.clear_all();
9050
      uint max_key_part= 0;
9051
      while ((item= select_items_it++))
9052
      {
9053
        item_field= (Item_field*) item; /* (SA5) already checked above. */
9054 9055
        /* 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
9056 9057 9058 9059 9060 9061
        /*
          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;
9062
        if (key_part_nr < 1 || key_part_nr > join->fields_list.elements)
9063 9064
          goto next_index;
        cur_part= cur_index_info->key_part + key_part_nr - 1;
9065
        cur_group_prefix_len+= cur_part->store_length;
timour@mysql.com's avatar
timour@mysql.com committed
9066 9067
        cur_used_key_parts.set_bit(key_part_nr);
        ++cur_group_key_parts;
9068
        max_key_part= max(max_key_part,key_part_nr);
9069
      }
9070 9071 9072 9073 9074 9075 9076 9077 9078 9079 9080
      /*
        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;
9081 9082 9083 9084 9085 9086 9087 9088
    }
    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);
9089
      if (key_part_nr <= cur_group_key_parts)
9090 9091 9092 9093 9094 9095 9096 9097
        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.
    */
9098 9099 9100 9101 9102 9103 9104 9105 9106 9107 9108 9109 9110 9111 9112 9113 9114 9115 9116 9117 9118 9119

    /*
      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)))
9120
    {
9121 9122 9123 9124 9125 9126 9127 9128 9129 9130 9131 9132 9133
      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
9134
      {
9135 9136 9137 9138
        /*
          There is a gap but no range tree, thus no predicates at all for the
          non-group keyparts.
        */
9139
        goto next_index;
timour@mysql.com's avatar
timour@mysql.com committed
9140 9141 9142 9143 9144 9145 9146 9147 9148 9149 9150 9151 9152 9153 9154 9155 9156 9157 9158 9159
      }
      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. */
9160
        if (join->conds->walk(&Item::find_item_in_field_list_processor, 0,
timour@mysql.com's avatar
timour@mysql.com committed
9161 9162 9163
                              (byte*) key_part_range))
          goto next_index;
      }
9164 9165
    }

9166 9167 9168 9169 9170 9171 9172 9173
    /*
      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++)
      {
9174
        if (bitmap_is_set(table->read_set, cur_part->field->field_index))
9175 9176 9177 9178
          goto next_index;
      }
    }

9179
    /* If we got to this point, cur_index_info passes the test. */
9180 9181 9182
    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;
9183

9184 9185 9186 9187 9188 9189 9190 9191
    /* 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,
9192
                                                    cur_index_tree, TRUE);
9193 9194 9195 9196 9197
    }
    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
9198 9199 9200 9201 9202 9203
    /*
      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))
9204
    {
9205
      DBUG_ASSERT(tree != 0 || cur_param_idx == MAX_KEY);
9206 9207 9208 9209 9210 9211 9212 9213 9214 9215
      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;
    }
9216 9217

  next_index:
9218 9219
    cur_group_key_parts= 0;
    cur_group_prefix_len= 0;
9220 9221 9222 9223
  }
  if (!index_info) /* No usable index found. */
    DBUG_RETURN(NULL);

9224 9225 9226
  /* Check (SA3) for the where clause. */
  if (join->conds && min_max_arg_item &&
      !check_group_min_max_predicates(join->conds, min_max_arg_item,
9227 9228
                                      (index_info->flags & HA_SPATIAL) ?
                                      Field::itMBR : Field::itRAW))
9229 9230 9231 9232
    DBUG_RETURN(NULL);

  /* The query passes all tests, so construct a new TRP object. */
  read_plan= new (param->mem_root)
9233 9234 9235 9236
                 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,
9237
                                   (key_infix_len > 0) ? key_infix : NULL,
9238
                                   tree, best_index_tree, best_param_idx,
9239
                                   best_quick_prefix_records);
9240 9241 9242 9243 9244
  if (read_plan)
  {
    if (tree && read_plan->quick_prefix_records == 0)
      DBUG_RETURN(NULL);

9245 9246 9247
    read_plan->read_cost= best_read_cost;
    read_plan->records=   best_records;

9248 9249 9250 9251 9252 9253 9254 9255 9256 9257
    DBUG_PRINT("info",
               ("Returning group min/max plan: cost: %g, records: %lu",
                read_plan->read_cost, (ulong) read_plan->records));
  }

  DBUG_RETURN(read_plan);
}


/*
9258 9259
  Check that the MIN/MAX attribute participates only in range predicates
  with constants.
9260 9261 9262 9263 9264 9265

  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)
9266
    min_max_arg_part  the keypart of the MIN/MAX argument if any
9267 9268 9269

  DESCRIPTION
    The function walks recursively over the cond tree representing a WHERE
9270
    clause, and checks condition (SA3) - if a field is referenced by a MIN/MAX
9271 9272
    aggregate function, it is referenced only by one of the following
    predicates: {=, !=, <, <=, >, >=, between, is null, is not null}.
9273 9274 9275 9276 9277 9278 9279

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

static bool
9280 9281
check_group_min_max_predicates(COND *cond, Item_field *min_max_arg_item,
                               Field::imagetype image_type)
9282 9283
{
  DBUG_ENTER("check_group_min_max_predicates");
9284
  DBUG_ASSERT(cond && min_max_arg_item);
9285 9286 9287 9288 9289 9290 9291 9292 9293

  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
9294
      if (!check_group_min_max_predicates(and_or_arg, min_max_arg_item,
9295
                                         image_type))
9296 9297 9298 9299 9300
        DBUG_RETURN(FALSE);
    }
    DBUG_RETURN(TRUE);
  }

9301 9302 9303 9304 9305 9306 9307 9308 9309 9310 9311 9312 9313
  /*
    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. */
9314 9315 9316 9317 9318 9319 9320 9321 9322 9323 9324 9325 9326
  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)
    {
9327
      if (min_max_arg_item->eq(cur_arg, 1)) 
9328 9329 9330
      {
       /*
         If pred references the MIN/MAX argument, check whether pred is a range
9331
         condition that compares the MIN/MAX argument with a constant.
9332 9333
       */
        Item_func::Functype pred_type= pred->functype();
9334 9335 9336 9337 9338 9339 9340 9341 9342 9343
        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)
9344 9345 9346 9347
          DBUG_RETURN(FALSE);

        /* Check that pred compares min_max_arg_item with a constant. */
        Item *args[3];
9348
        bzero(args, 3 * sizeof(Item*));
9349 9350 9351 9352
        bool inv;
        /* Test if this is a comparison of a field and a constant. */
        if (!simple_pred(pred, args, &inv))
          DBUG_RETURN(FALSE);
9353 9354 9355 9356 9357 9358 9359 9360 9361 9362 9363 9364 9365 9366 9367 9368 9369 9370 9371

        /* 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);
9372 9373 9374 9375
      }
    }
    else if (cur_arg->type() == Item::FUNC_ITEM)
    {
monty@mishka.local's avatar
monty@mishka.local committed
9376
      if (!check_group_min_max_predicates(cur_arg, min_max_arg_item,
9377
                                         image_type))
9378 9379 9380 9381 9382 9383 9384 9385 9386 9387 9388 9389 9390 9391 9392 9393 9394 9395 9396
        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()
9397 9398 9399 9400 9401 9402 9403 9404 9405
    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)
9406 9407 9408
    
  DESCRIPTION
    Test conditions (NGA1, NGA2) from get_best_group_min_max(). Namely,
9409 9410
    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
9411 9412
    (const_ci = NGF_i).
    Thus all the NGF_i attributes must fill the 'gap' between the last group-by
9413 9414 9415 9416 9417 9418
    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
9419
    TRUE  if the index passes the test
9420 9421 9422 9423
    FALSE o/w
*/

static bool
9424
get_constant_key_infix(KEY *index_info, SEL_ARG *index_range_tree,
9425
                       KEY_PART_INFO *first_non_group_part,
9426 9427 9428 9429
                       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)
9430 9431 9432
{
  SEL_ARG       *cur_range;
  KEY_PART_INFO *cur_part;
9433 9434
  /* End part for the first loop below. */
  KEY_PART_INFO *end_part= min_max_arg_part ? min_max_arg_part : last_part;
9435 9436 9437 9438 9439 9440 9441 9442 9443 9444 9445 9446 9447 9448 9449 9450 9451

  *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)
    {
9452 9453 9454 9455 9456 9457 9458
      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;
      }
9459 9460 9461 9462 9463 9464 9465 9466 9467 9468 9469 9470
    }

    /* 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 &&
monty@mysql.com's avatar
monty@mysql.com committed
9471 9472 9473 9474
         cur_range->min_value[0] && cur_range->max_value[0]) ||
        !memcmp(cur_range->min_value, cur_range->max_value, field_length))
    {
      /* cur_range specifies 'IS NULL' or an equality condition. */
9475 9476 9477 9478 9479 9480 9481 9482
      memcpy(key_ptr, cur_range->min_value, field_length);
      key_ptr+= field_length;
      *key_infix_len+= field_length;
    }
    else
      return FALSE;
  }

9483 9484 9485
  if (!min_max_arg_part && (cur_part == last_part))
    *first_non_infix_part= last_part;

9486 9487 9488 9489
  return TRUE;
}


9490 9491 9492 9493 9494 9495 9496 9497 9498 9499 9500 9501 9502 9503 9504 9505 9506 9507 9508 9509
/*
  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)
{
9510
  KEY_PART_INFO *part, *end;
9511

9512
  for (part= index->key_part, end= part + index->key_parts; part < end; part++)
9513 9514
  {
    if (field->eq(part->field))
ram@gw.mysql.r18.ru's avatar
ram@gw.mysql.r18.ru committed
9515
      return part - index->key_part + 1;
9516
  }
9517
  return 0;
9518 9519 9520 9521 9522 9523 9524 9525 9526 9527 9528 9529 9530 9531 9532 9533 9534 9535 9536 9537 9538 9539 9540 9541 9542 9543 9544 9545 9546 9547 9548 9549 9550 9551 9552 9553 9554 9555 9556 9557 9558
}


/*
  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]);
}


9559
/*
9560
  Compute the cost of a quick_group_min_max_select for a particular index.
9561 9562

  SYNOPSIS
9563 9564 9565 9566 9567 9568 9569
    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
9570 9571
    quick_prefix_records [in] Number of records retrieved by the internally
			      used quick range select if any
9572 9573 9574 9575
    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
9576 9577

  DESCRIPTION
monty@mysql.com's avatar
monty@mysql.com committed
9578 9579
    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.
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 9618

  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
*/

9619 9620 9621 9622 9623
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)
9624 9625 9626 9627 9628 9629 9630 9631 9632 9633 9634 9635
{
  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
9636
  DBUG_ENTER("cost_group_min_max");
monty@mysql.com's avatar
monty@mysql.com committed
9637

9638 9639
  table_records= table->file->stats.records;
  keys_per_block= (table->file->stats.block_size / 2 /
9640 9641 9642 9643 9644 9645 9646 9647 9648 9649 9650 9651 9652 9653 9654 9655
                   (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
9656
    num_groups= (uint) rint(num_groups * quick_prefix_selectivity);
igor@rurik.mysql.com's avatar
igor@rurik.mysql.com committed
9657
    set_if_bigger(num_groups, 1);
9658 9659 9660 9661 9662 9663 9664 9665 9666 9667 9668 9669 9670 9671 9672 9673 9674 9675 9676 9677 9678 9679 9680 9681 9682 9683 9684 9685 9686 9687 9688
  }

  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;

9689
  *read_cost= io_cost + cpu_cost;
9690
  *records= num_groups;
9691 9692

  DBUG_PRINT("info",
9693 9694
             ("table rows=%u, keys/block=%u, keys/group=%u, result rows=%u, blocks=%u",
              table_records, keys_per_block, keys_per_group, *records,
9695 9696 9697 9698 9699 9700 9701 9702 9703 9704 9705 9706 9707 9708 9709 9710 9711 9712 9713 9714 9715 9716 9717 9718 9719 9720 9721 9722 9723 9724 9725 9726 9727
              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");

9728 9729 9730 9731 9732
  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
9733 9734
                                        key_infix_len, key_infix,
                                        parent_alloc);
9735 9736 9737 9738 9739 9740 9741 9742 9743 9744 9745 9746 9747 9748 9749 9750
  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. */
9751 9752
      quick->quick_prefix_select= get_quick_select(param, param_idx,
                                                   index_tree,
9753 9754 9755 9756 9757 9758 9759 9760 9761 9762 9763 9764 9765 9766 9767 9768 9769 9770 9771 9772 9773 9774
                                                   &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)
      {
9775
        if (quick->add_range(min_max_range))
9776 9777 9778 9779 9780 9781 9782 9783 9784 9785 9786 9787 9788 9789 9790 9791 9792 9793 9794 9795 9796 9797 9798 9799 9800 9801 9802 9803 9804 9805 9806 9807 9808 9809 9810 9811 9812 9813 9814 9815 9816 9817
        {
          delete quick;
          quick= NULL;
          DBUG_RETURN(NULL);
        }
        min_max_range= min_max_range->next;
      }
    }
  }
  else
    quick->quick_prefix_select= NULL;

  quick->update_key_stat();

  DBUG_RETURN(quick);
}


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

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

  RETURN
    None
*/

monty@mysql.com's avatar
monty@mysql.com committed
9818 9819 9820 9821 9822 9823 9824 9825 9826 9827 9828 9829 9830
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),
9831 9832
   key_infix_len(key_infix_len_arg), min_functions_it(NULL),
   max_functions_it(NULL)
9833 9834 9835 9836 9837 9838
{
  head=       table;
  file=       head->file;
  index=      use_index;
  record=     head->record[0];
  tmp_record= head->record[1];
9839 9840 9841
  read_time= read_cost_arg;
  records= records_arg;
  used_key_parts= used_key_parts_arg;
9842 9843 9844
  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
9845 9846 9847 9848 9849 9850

  /*
    We can't have parent_alloc set as the init function can't handle this case
    yet.
  */
  DBUG_ASSERT(!parent_alloc);
9851 9852 9853
  if (!parent_alloc)
  {
    init_sql_alloc(&alloc, join->thd->variables.range_alloc_block_size, 0);
monty@mysql.com's avatar
monty@mysql.com committed
9854
    join->thd->mem_root= &alloc;
9855 9856
  }
  else
9857
    bzero(&alloc, sizeof(MEM_ROOT));            // ensure that it's not used
9858 9859 9860 9861 9862 9863 9864 9865 9866
}


/*
  Do post-constructor initialization.

  SYNOPSIS
    QUICK_GROUP_MIN_MAX_SELECT::init()
  
9867 9868 9869 9870 9871 9872
  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.

9873 9874 9875 9876 9877 9878 9879 9880 9881 9882 9883 9884 9885 9886 9887 9888 9889 9890 9891 9892 9893 9894 9895 9896 9897 9898 9899 9900 9901 9902 9903 9904 9905 9906 9907
  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
9908
    if (my_init_dynamic_array(&min_max_ranges, sizeof(QUICK_RANGE*), 16, 16))
9909 9910
      return 1;

9911 9912
    if (have_min)
    {
monty@mishka.local's avatar
monty@mishka.local committed
9913
      if (!(min_functions= new List<Item_sum>))
9914 9915 9916 9917 9918 9919
        return 1;
    }
    else
      min_functions= NULL;
    if (have_max)
    {
monty@mishka.local's avatar
monty@mishka.local committed
9920
      if (!(max_functions= new List<Item_sum>))
9921 9922 9923 9924
        return 1;
    }
    else
      max_functions= NULL;
9925

9926 9927 9928
    Item_sum *min_max_item;
    Item_sum **func_ptr= join->sum_funcs;
    while ((min_max_item= *(func_ptr++)))
9929
    {
9930 9931 9932 9933
      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);
9934 9935
    }

9936 9937 9938 9939 9940 9941 9942 9943 9944 9945 9946
    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;
    }
9947
  }
igor@rurik.mysql.com's avatar
igor@rurik.mysql.com committed
9948 9949
  else
    min_max_ranges.elements= 0;
9950 9951 9952 9953 9954 9955 9956 9957 9958 9959 9960 9961 9962

  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));
9963 9964
  delete min_functions_it;
  delete max_functions_it;
9965 9966 9967 9968 9969 9970 9971 9972 9973 9974 9975 9976 9977 9978 9979 9980 9981 9982 9983
  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
9984 9985
    FALSE on success
    TRUE  otherwise
9986 9987 9988 9989 9990 9991 9992 9993
*/

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
9994
  if ((range_flag & NO_MIN_RANGE) && (range_flag & NO_MAX_RANGE))
9995
    return FALSE;
9996 9997 9998 9999 10000 10001 10002 10003 10004 10005 10006 10007 10008 10009 10010

  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)
10011
    return TRUE;
10012
  if (insert_dynamic(&min_max_ranges, (gptr)&range))
10013 10014
    return TRUE;
  return FALSE;
10015 10016 10017 10018 10019 10020 10021 10022 10023 10024 10025 10026 10027 10028 10029 10030 10031 10032 10033 10034 10035 10036 10037 10038 10039 10040 10041 10042 10043
}


/*
  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)
  {
10044
    QUICK_RANGE *cur_range;
10045 10046 10047 10048 10049 10050 10051
    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;
10052
        used_key_parts++;
10053 10054 10055 10056 10057 10058 10059 10060 10061
        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;
10062
        used_key_parts++;
10063 10064 10065 10066
        return;
      }
    }
  }
10067 10068
  else if (have_min && min_max_arg_part &&
           min_max_arg_part->field->real_maybe_null())
10069
  {
10070 10071 10072 10073 10074 10075 10076 10077
    /*
      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.
    */
10078
    max_used_key_length+= min_max_arg_len;
10079
    used_key_parts++;
10080 10081 10082 10083 10084 10085 10086 10087 10088 10089
  }
}


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

  SYNOPSIS
    QUICK_GROUP_MIN_MAX_SELECT::reset()

10090 10091 10092 10093
  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.

10094 10095 10096 10097 10098 10099 10100 10101 10102 10103 10104
  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 */
10105
  result= file->ha_index_init(index, 1);
10106
  result= file->index_last(record);
sergefp@mysql.com's avatar
sergefp@mysql.com committed
10107 10108
  if (result == HA_ERR_END_OF_FILE)
    DBUG_RETURN(0);
10109 10110
  if (result)
    DBUG_RETURN(result);
sergefp@mysql.com's avatar
sergefp@mysql.com committed
10111 10112
  if (quick_prefix_select && quick_prefix_select->reset())
    DBUG_RETURN(1);
10113 10114 10115 10116 10117 10118 10119 10120 10121 10122 10123 10124 10125 10126 10127 10128 10129 10130 10131 10132 10133 10134 10135 10136 10137 10138 10139 10140 10141 10142 10143 10144 10145 10146 10147 10148 10149 10150 10151
  /* 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
10152 10153 10154 10155 10156 10157 10158
#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
10159
  int result;
timour@mysql.com's avatar
timour@mysql.com committed
10160
#endif
10161 10162 10163 10164 10165 10166 10167 10168 10169 10170 10171 10172 10173 10174 10175 10176 10177 10178 10179 10180
  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;
10181
    if (result)
10182 10183 10184 10185 10186 10187 10188 10189 10190 10191 10192 10193 10194 10195 10196 10197 10198 10199 10200
      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)));
    }
10201
    /*
10202
      If this is just a GROUP BY or DISTINCT without MIN or MAX and there
10203 10204 10205 10206 10207 10208 10209
      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);

10210
    result= have_min ? min_res : have_max ? max_res : result;
10211 10212
  } while ((result == HA_ERR_KEY_NOT_FOUND || result == HA_ERR_END_OF_FILE) &&
           is_last_prefix != 0);
10213 10214

  if (result == 0)
10215
  {
10216 10217 10218 10219 10220 10221 10222
    /*
      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);
10223
  }
10224 10225 10226 10227 10228 10229 10230 10231 10232 10233 10234 10235 10236 10237
  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
10238 10239
    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.
10240 10241 10242 10243 10244 10245 10246 10247 10248 10249

  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.
10250
    HA_ERR_END_OF_FILE   - "" -
10251 10252 10253 10254 10255 10256 10257 10258 10259 10260 10261 10262 10263 10264 10265 10266
    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
  {
10267
    /* Apply the constant equality conditions to the non-group select fields */
10268 10269 10270 10271 10272 10273 10274 10275 10276 10277 10278 10279 10280 10281 10282 10283 10284 10285 10286 10287 10288 10289 10290 10291 10292 10293 10294 10295 10296 10297 10298 10299 10300
    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)
      {
10301
        if (key_cmp(index_info->key_part, group_prefix, real_prefix_len))
10302
          key_restore(record, tmp_record, index_info, 0);
10303
      }
10304
      else if (result == HA_ERR_KEY_NOT_FOUND || result == HA_ERR_END_OF_FILE)
10305 10306 10307 10308 10309 10310 10311 10312 10313 10314 10315 10316 10317 10318 10319 10320 10321 10322 10323
        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
10324
    Lookup the maximal key of the group, and store it into this->record.
10325 10326 10327 10328

  RETURN
    0                    on success
    HA_ERR_KEY_NOT_FOUND if no MAX key was found that fulfills all conditions.
10329
    HA_ERR_END_OF_FILE	 - "" -
10330 10331 10332 10333 10334 10335 10336 10337 10338 10339 10340 10341 10342 10343 10344 10345 10346 10347 10348 10349 10350 10351 10352 10353 10354 10355 10356 10357 10358 10359 10360 10361 10362 10363 10364 10365 10366 10367 10368 10369 10370 10371 10372 10373 10374 10375 10376 10377 10378 10379 10380 10381 10382 10383 10384 10385 10386 10387 10388 10389 10390 10391 10392 10393 10394 10395 10396 10397 10398 10399 10400 10401 10402 10403 10404 10405 10406 10407 10408 10409 10410 10411 10412 10413 10414 10415 10416 10417 10418 10419 10420 10421 10422 10423 10424 10425 10426 10427 10428 10429
    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
10430
    HA_ERR_END_OF_FILE   - "" -
10431 10432 10433 10434 10435 10436 10437 10438 10439 10440 10441 10442 10443 10444 10445 10446 10447 10448 10449 10450 10451 10452
    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) &&
10453
        (key_cmp(min_max_arg_part, (const byte*) cur_range->max_key,
10454
                 min_max_arg_len) == 1))
10455 10456 10457 10458 10459 10460 10461 10462 10463 10464 10465 10466 10467 10468 10469 10470 10471 10472 10473 10474
      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);
10475
    if (result)
10476
    {
10477 10478 10479 10480
      if ((result == HA_ERR_KEY_NOT_FOUND || result == HA_ERR_END_OF_FILE) &&
          (cur_range->flag & (EQ_RANGE | NULL_RANGE)))
        continue; /* Check the next range. */

10481 10482 10483 10484 10485
      /*
        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.
      */
10486
      break;
10487
    }
10488 10489 10490 10491 10492 10493

    /* 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)
10494 10495 10496 10497 10498 10499
    {
      /*
        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);
10500 10501 10502 10503 10504 10505 10506
      found_null= TRUE;
      continue;
    }

    /* Check if record belongs to the current group. */
    if (key_cmp(index_info->key_part, group_prefix, real_prefix_len))
    {
10507
      result= HA_ERR_KEY_NOT_FOUND;
10508 10509 10510 10511 10512 10513 10514 10515 10516 10517 10518 10519 10520 10521 10522 10523 10524
      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)))
      {
10525
        result= HA_ERR_KEY_NOT_FOUND;
10526 10527 10528 10529 10530 10531 10532 10533 10534 10535 10536 10537 10538 10539
        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)
  {
10540
    memcpy(record, tmp_record, head->s->rec_buff_length);
10541 10542 10543 10544 10545 10546 10547 10548 10549 10550 10551 10552 10553 10554 10555 10556 10557 10558 10559 10560 10561 10562 10563
    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
10564
    HA_ERR_END_OF_FILE   - "" -
10565 10566 10567 10568 10569 10570 10571 10572 10573 10574 10575 10576 10577 10578 10579 10580 10581 10582 10583 10584 10585 10586
    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) &&
10587
        (key_cmp(min_max_arg_part, (const byte*) cur_range->min_key,
10588
                 min_max_arg_len) == -1))
10589 10590 10591 10592 10593 10594 10595 10596 10597 10598 10599 10600 10601 10602 10603 10604 10605 10606 10607 10608 10609
      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);

monty@mysql.com's avatar
monty@mysql.com committed
10610 10611
    if (result)
    {
10612 10613 10614 10615
      if ((result == HA_ERR_KEY_NOT_FOUND || result == HA_ERR_END_OF_FILE) &&
          (cur_range->flag & EQ_RANGE))
        continue; /* Check the next range. */

10616 10617 10618 10619 10620
      /*
        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
10621
    }
10622 10623
    /* A key was found. */
    if (cur_range->flag & EQ_RANGE)
monty@mysql.com's avatar
monty@mysql.com committed
10624
      return 0; /* No need to perform the checks below for equal keys. */
10625 10626 10627

    /* 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
10628
      continue;                                 // Row not found
10629 10630 10631 10632 10633 10634 10635 10636 10637 10638 10639 10640 10641 10642 10643 10644 10645 10646 10647 10648 10649 10650 10651 10652 10653 10654 10655 10656 10657 10658 10659 10660 10661 10662 10663 10664 10665 10666 10667 10668 10669 10670 10671 10672 10673 10674 10675 10676 10677 10678 10679 10680 10681 10682 10683 10684 10685 10686 10687 10688 10689 10690 10691 10692 10693 10694 10695 10696 10697 10698 10699 10700 10701 10702 10703 10704 10705 10706 10707 10708 10709 10710 10711 10712 10713 10714 10715 10716

    /* 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();
}


10717 10718 10719 10720 10721 10722 10723 10724 10725 10726 10727 10728 10729 10730 10731
/*
  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.

*/

10732 10733 10734 10735 10736 10737 10738 10739 10740 10741 10742
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);
}


10743
#ifndef DBUG_OFF
10744

10745 10746 10747 10748 10749 10750 10751
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");
10752

10753 10754 10755 10756 10757 10758 10759 10760 10761 10762 10763 10764 10765 10766 10767
  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())
10768
    tmp.append(STRING_WITH_LEN("(empty)"));
10769

10770
  DBUG_PRINT("info", ("SEL_TREE %p (%s) scans:%s", tree, msg, tmp.ptr()));
10771

10772 10773
  DBUG_VOID_RETURN;
}
10774

10775 10776 10777 10778

static void print_ror_scans_arr(TABLE *table, const char *msg,
                                struct st_ror_scan_info **start,
                                struct st_ror_scan_info **end)
10779
{
serg@serg.mylan's avatar
serg@serg.mylan committed
10780
  DBUG_ENTER("print_ror_scans_arr");
10781 10782 10783 10784

  char buff[1024];
  String tmp(buff,sizeof(buff),&my_charset_bin);
  tmp.length(0);
10785
  for (;start != end; start++)
10786
  {
10787 10788 10789
    if (tmp.length())
      tmp.append(',');
    tmp.append(table->key_info[(*start)->keynr].name);
10790
  }
10791
  if (!tmp.length())
10792
    tmp.append(STRING_WITH_LEN("(empty)"));
10793 10794
  DBUG_PRINT("info", ("ROR key scans (%s): %s", msg, tmp.ptr()));
  DBUG_VOID_RETURN;
10795 10796
}

bk@work.mysql.com's avatar
bk@work.mysql.com committed
10797 10798 10799 10800 10801 10802 10803 10804 10805 10806 10807
/*****************************************************************************
** 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
10808
  const char *key_end= key+used_length;
10809
  String tmp(buff,sizeof(buff),&my_charset_bin);
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
10810
  uint store_length;
10811 10812 10813 10814
  TABLE *table= key_part->field->table;
  my_bitmap_map *old_write_set, *old_read_set;
  old_write_set= dbug_tmp_use_all_columns(table, table->write_set);
  old_read_set=  dbug_tmp_use_all_columns(table, table->read_set);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
10815

pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
10816
  for (; key < key_end; key+=store_length, key_part++)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
10817
  {
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
10818 10819 10820
    Field *field=      key_part->field;
    store_length= key_part->store_length;

bk@work.mysql.com's avatar
bk@work.mysql.com committed
10821 10822
    if (field->real_maybe_null())
    {
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
10823
      if (*key)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
10824 10825 10826 10827
      {
	fwrite("NULL",sizeof(char),4,DBUG_FILE);
	continue;
      }
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
10828 10829
      key++;					// Skip null byte
      store_length--;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
10830
    }
10831
    field->set_key_image((char*) key, key_part->length);
monty@mysql.com's avatar
monty@mysql.com committed
10832 10833 10834 10835
    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
10836
    fwrite(tmp.ptr(),sizeof(char),tmp.length(),DBUG_FILE);
pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
10837 10838
    if (key+store_length < key_end)
      fputc('/',DBUG_FILE);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
10839
  }
10840 10841
  dbug_tmp_restore_column_map(table->write_set, old_write_set);
  dbug_tmp_restore_column_map(table->read_set, old_read_set);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
10842 10843
}

pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
10844

10845
static void print_quick(QUICK_SELECT_I *quick, const key_map *needed_reg)
10846
{
10847
  char buf[MAX_KEY/8+1];
10848 10849
  TABLE *table;
  my_bitmap_map *old_read_map, *old_write_map;
10850
  DBUG_ENTER("print_quick");
serg@serg.mylan's avatar
serg@serg.mylan committed
10851
  if (!quick)
10852
    DBUG_VOID_RETURN;
10853
  DBUG_LOCK_FILE;
10854

10855 10856 10857
  table= quick->head;
  old_read_map=  dbug_tmp_use_all_columns(table, table->read_set);
  old_write_map= dbug_tmp_use_all_columns(table, table->write_set);
monty@mysql.com's avatar
monty@mysql.com committed
10858
  quick->dbug_dump(0, TRUE);
10859 10860 10861
  dbug_tmp_restore_column_map(table->read_set, old_read_map);
  dbug_tmp_restore_column_map(table->write_set, old_write_map);

10862
  fprintf(DBUG_FILE,"other_keys: 0x%s:\n", needed_reg->print(buf));
10863

10864
  DBUG_UNLOCK_FILE;
10865 10866 10867
  DBUG_VOID_RETURN;
}

pem@mysql.comhem.se's avatar
pem@mysql.comhem.se committed
10868

10869
static void print_rowid(byte* val, int len)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
10870
{
10871
  byte *pb;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
10872
  DBUG_LOCK_FILE;
10873 10874 10875 10876 10877 10878 10879 10880 10881 10882
  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;
}
10883

10884 10885 10886 10887
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);
10888

10889
  if (verbose)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
10890
  {
10891 10892
    QUICK_RANGE *range;
    QUICK_RANGE **pr= (QUICK_RANGE**)ranges.buffer;
10893
    QUICK_RANGE **last_range= pr + ranges.elements;
10894
    for (; pr!=last_range; ++pr)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
10895
    {
10896 10897 10898 10899 10900 10901 10902 10903 10904 10905 10906
      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
10907

10908 10909 10910 10911 10912 10913 10914 10915 10916
      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
10917 10918
    }
  }
10919 10920 10921 10922 10923 10924 10925 10926 10927 10928 10929 10930
}

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)
  {
10931
    fprintf(DBUG_FILE, "%*sclustered PK quick:\n", indent, "");
10932 10933 10934 10935 10936 10937 10938 10939 10940
    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;
10941
  fprintf(DBUG_FILE, "%*squick ROR-intersect select, %scovering\n",
10942 10943 10944
          indent, "", need_to_fetch_row? "":"non-");
  fprintf(DBUG_FILE, "%*smerged scans {\n", indent, "");
  while ((quick= it++))
10945
    quick->dbug_dump(indent+2, verbose);
10946 10947
  if (cpk_quick)
  {
10948
    fprintf(DBUG_FILE, "%*sclustered PK quick:\n", indent, "");
10949 10950 10951 10952 10953 10954 10955 10956 10957 10958 10959 10960 10961 10962
    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
10963 10964
}

10965 10966 10967 10968 10969 10970 10971 10972 10973 10974 10975 10976 10977 10978 10979 10980 10981 10982 10983 10984 10985 10986 10987 10988 10989 10990 10991 10992 10993 10994 10995 10996 10997 10998 10999 11000 11001 11002 11003 11004 11005 11006 11007 11008

/*
  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
11009
#endif /* NOT_USED */
bk@work.mysql.com's avatar
bk@work.mysql.com committed
11010 11011

/*****************************************************************************
11012
** Instantiate templates
bk@work.mysql.com's avatar
bk@work.mysql.com committed
11013 11014
*****************************************************************************/

11015
#ifdef HAVE_EXPLICIT_TEMPLATE_INSTANTIATION
bk@work.mysql.com's avatar
bk@work.mysql.com committed
11016 11017 11018
template class List<QUICK_RANGE>;
template class List_iterator<QUICK_RANGE>;
#endif