Two-sweeps read index_merge plus several small index_merge fixes and improvements

mysql-test/r/index_merge_bdb.result

+drop table if exists t1;
+create table t1 (
+pk int primary key,
+key1 int,
+key2 int,  
+filler char(200),
+filler2 char(200),  
+index(key1),
+index(key2),  
+) type=bdb;
+select * from t1 where (key1 >= 2 and key1 <= 10) or (pk >= 4 and pk <=8 );
+pk	key1	key2	filler	filler2
+2	2	2	filler-data	filler-data-2
+3	3	3	filler-data	filler-data-2
+9	9	9	filler-data	filler-data-2
+10	10	10	filler-data	filler-data-2
+4	4	4	filler-data	filler-data-2
+5	5	5	filler-data	filler-data-2
+6	6	6	filler-data	filler-data-2
+7	7	7	filler-data	filler-data-2
+8	8	8	filler-data	filler-data-2
+set @maxv=1000;
+select * from t1 where 
+(pk < 5) or (pk > 10 and pk < 15) or (pk >= 50 and pk < 55 ) or (pk > @maxv-10) 
+or key1=18 or key1=60;
+pk	key1	key2	filler	filler2
+18	18	18	filler-data	filler-data-2
+60	60	60	filler-data	filler-data-2
+1	1	1	filler-data	filler-data-2
+2	2	2	filler-data	filler-data-2
+3	3	3	filler-data	filler-data-2
+4	4	4	filler-data	filler-data-2
+11	11	11	filler-data	filler-data-2
+12	12	12	filler-data	filler-data-2
+13	13	13	filler-data	filler-data-2
+14	14	14	filler-data	filler-data-2
+50	50	50	filler-data	filler-data-2
+51	51	51	filler-data	filler-data-2
+52	52	52	filler-data	filler-data-2
+53	53	53	filler-data	filler-data-2
+54	54	54	filler-data	filler-data-2
+991	991	991	filler-data	filler-data-2
+992	992	992	filler-data	filler-data-2
+993	993	993	filler-data	filler-data-2
+994	994	994	filler-data	filler-data-2
+995	995	995	filler-data	filler-data-2
+996	996	996	filler-data	filler-data-2
+997	997	997	filler-data	filler-data-2
+998	998	998	filler-data	filler-data-2
+999	999	999	filler-data	filler-data-2
+1000	1000	1000	filler-data	filler-data-2
+select * from t1 where 
+(pk < 5) or (pk > 10 and pk < 15) or (pk >= 50 and pk < 55 ) or (pk > @maxv-10) 
+or key1 < 3 or key1 > @maxv-11;
+pk	key1	key2	filler	filler2
+990	990	990	filler-data	filler-data-2
+1	1	1	filler-data	filler-data-2
+2	2	2	filler-data	filler-data-2
+3	3	3	filler-data	filler-data-2
+4	4	4	filler-data	filler-data-2
+11	11	11	filler-data	filler-data-2
+12	12	12	filler-data	filler-data-2
+13	13	13	filler-data	filler-data-2
+14	14	14	filler-data	filler-data-2
+50	50	50	filler-data	filler-data-2
+51	51	51	filler-data	filler-data-2
+52	52	52	filler-data	filler-data-2
+53	53	53	filler-data	filler-data-2
+54	54	54	filler-data	filler-data-2
+991	991	991	filler-data	filler-data-2
+992	992	992	filler-data	filler-data-2
+993	993	993	filler-data	filler-data-2
+994	994	994	filler-data	filler-data-2
+995	995	995	filler-data	filler-data-2
+996	996	996	filler-data	filler-data-2
+997	997	997	filler-data	filler-data-2
+998	998	998	filler-data	filler-data-2
+999	999	999	filler-data	filler-data-2
+1000	1000	1000	filler-data	filler-data-2
+select * from t1 where 
+(pk < 5) or (pk > 10 and pk < 15) or (pk >= 50 and pk < 55 ) or (pk > @maxv-10) 
+or
+(key1 < 5) or (key1 > 10 and key1 < 15) or (key1 >= 50 and key1 < 55 ) or (key1 > @maxv-10);
+pk	key1	key2	filler	filler2
+1	1	1	filler-data	filler-data-2
+2	2	2	filler-data	filler-data-2
+3	3	3	filler-data	filler-data-2
+4	4	4	filler-data	filler-data-2
+11	11	11	filler-data	filler-data-2
+12	12	12	filler-data	filler-data-2
+13	13	13	filler-data	filler-data-2
+14	14	14	filler-data	filler-data-2
+50	50	50	filler-data	filler-data-2
+51	51	51	filler-data	filler-data-2
+52	52	52	filler-data	filler-data-2
+53	53	53	filler-data	filler-data-2
+54	54	54	filler-data	filler-data-2
+991	991	991	filler-data	filler-data-2
+992	992	992	filler-data	filler-data-2
+993	993	993	filler-data	filler-data-2
+994	994	994	filler-data	filler-data-2
+995	995	995	filler-data	filler-data-2
+996	996	996	filler-data	filler-data-2
+997	997	997	filler-data	filler-data-2
+998	998	998	filler-data	filler-data-2
+999	999	999	filler-data	filler-data-2
+1000	1000	1000	filler-data	filler-data-2
+select * from t1 where 
+(pk > 10 and pk < 15) or (pk >= 50 and pk < 55 )
+or
+(key1 < 5) or (key1 > @maxv-10);
+pk	key1	key2	filler	filler2
+1	1	1	filler-data	filler-data-2
+2	2	2	filler-data	filler-data-2
+3	3	3	filler-data	filler-data-2
+4	4	4	filler-data	filler-data-2
+991	991	991	filler-data	filler-data-2
+992	992	992	filler-data	filler-data-2
+993	993	993	filler-data	filler-data-2
+994	994	994	filler-data	filler-data-2
+995	995	995	filler-data	filler-data-2
+996	996	996	filler-data	filler-data-2
+997	997	997	filler-data	filler-data-2
+998	998	998	filler-data	filler-data-2
+999	999	999	filler-data	filler-data-2
+1000	1000	1000	filler-data	filler-data-2
+11	11	11	filler-data	filler-data-2
+12	12	12	filler-data	filler-data-2
+13	13	13	filler-data	filler-data-2
+14	14	14	filler-data	filler-data-2
+50	50	50	filler-data	filler-data-2
+51	51	51	filler-data	filler-data-2
+52	52	52	filler-data	filler-data-2
+53	53	53	filler-data	filler-data-2
+54	54	54	filler-data	filler-data-2
+drop table t1;

mysql-test/r/index_merge_innodb2.result

+drop table if exists t1;
+create table t1 (
+pk int primary key,
+key1 int,
+key2 int,  
+filler char(200),
+filler2 char(200),  
+index(key1),
+index(key2),  
+) type=innodb;
+select * from t1 where (key1 >= 2 and key1 <= 10) or (pk >= 4 and pk <=8 );
+pk	key1	key2	filler	filler2
+2	2	2	filler-data	filler-data-2
+3	3	3	filler-data	filler-data-2
+9	9	9	filler-data	filler-data-2
+10	10	10	filler-data	filler-data-2
+4	4	4	filler-data	filler-data-2
+5	5	5	filler-data	filler-data-2
+6	6	6	filler-data	filler-data-2
+7	7	7	filler-data	filler-data-2
+8	8	8	filler-data	filler-data-2
+set @maxv=1000;
+select * from t1 where 
+(pk < 5) or (pk > 10 and pk < 15) or (pk >= 50 and pk < 55 ) or (pk > @maxv-10) 
+or key1=18 or key1=60;
+pk	key1	key2	filler	filler2
+18	18	18	filler-data	filler-data-2
+60	60	60	filler-data	filler-data-2
+1	1	1	filler-data	filler-data-2
+2	2	2	filler-data	filler-data-2
+3	3	3	filler-data	filler-data-2
+4	4	4	filler-data	filler-data-2
+11	11	11	filler-data	filler-data-2
+12	12	12	filler-data	filler-data-2
+13	13	13	filler-data	filler-data-2
+14	14	14	filler-data	filler-data-2
+50	50	50	filler-data	filler-data-2
+51	51	51	filler-data	filler-data-2
+52	52	52	filler-data	filler-data-2
+53	53	53	filler-data	filler-data-2
+54	54	54	filler-data	filler-data-2
+991	991	991	filler-data	filler-data-2
+992	992	992	filler-data	filler-data-2
+993	993	993	filler-data	filler-data-2
+994	994	994	filler-data	filler-data-2
+995	995	995	filler-data	filler-data-2
+996	996	996	filler-data	filler-data-2
+997	997	997	filler-data	filler-data-2
+998	998	998	filler-data	filler-data-2
+999	999	999	filler-data	filler-data-2
+1000	1000	1000	filler-data	filler-data-2
+select * from t1 where 
+(pk < 5) or (pk > 10 and pk < 15) or (pk >= 50 and pk < 55 ) or (pk > @maxv-10) 
+or key1 < 3 or key1 > @maxv-11;
+pk	key1	key2	filler	filler2
+990	990	990	filler-data	filler-data-2
+1	1	1	filler-data	filler-data-2
+2	2	2	filler-data	filler-data-2
+3	3	3	filler-data	filler-data-2
+4	4	4	filler-data	filler-data-2
+11	11	11	filler-data	filler-data-2
+12	12	12	filler-data	filler-data-2
+13	13	13	filler-data	filler-data-2
+14	14	14	filler-data	filler-data-2
+50	50	50	filler-data	filler-data-2
+51	51	51	filler-data	filler-data-2
+52	52	52	filler-data	filler-data-2
+53	53	53	filler-data	filler-data-2
+54	54	54	filler-data	filler-data-2
+991	991	991	filler-data	filler-data-2
+992	992	992	filler-data	filler-data-2
+993	993	993	filler-data	filler-data-2
+994	994	994	filler-data	filler-data-2
+995	995	995	filler-data	filler-data-2
+996	996	996	filler-data	filler-data-2
+997	997	997	filler-data	filler-data-2
+998	998	998	filler-data	filler-data-2
+999	999	999	filler-data	filler-data-2
+1000	1000	1000	filler-data	filler-data-2
+select * from t1 where 
+(pk < 5) or (pk > 10 and pk < 15) or (pk >= 50 and pk < 55 ) or (pk > @maxv-10) 
+or
+(key1 < 5) or (key1 > 10 and key1 < 15) or (key1 >= 50 and key1 < 55 ) or (key1 > @maxv-10);
+pk	key1	key2	filler	filler2
+1	1	1	filler-data	filler-data-2
+2	2	2	filler-data	filler-data-2
+3	3	3	filler-data	filler-data-2
+4	4	4	filler-data	filler-data-2
+11	11	11	filler-data	filler-data-2
+12	12	12	filler-data	filler-data-2
+13	13	13	filler-data	filler-data-2
+14	14	14	filler-data	filler-data-2
+50	50	50	filler-data	filler-data-2
+51	51	51	filler-data	filler-data-2
+52	52	52	filler-data	filler-data-2
+53	53	53	filler-data	filler-data-2
+54	54	54	filler-data	filler-data-2
+991	991	991	filler-data	filler-data-2
+992	992	992	filler-data	filler-data-2
+993	993	993	filler-data	filler-data-2
+994	994	994	filler-data	filler-data-2
+995	995	995	filler-data	filler-data-2
+996	996	996	filler-data	filler-data-2
+997	997	997	filler-data	filler-data-2
+998	998	998	filler-data	filler-data-2
+999	999	999	filler-data	filler-data-2
+1000	1000	1000	filler-data	filler-data-2
+select * from t1 where 
+(pk > 10 and pk < 15) or (pk >= 50 and pk < 55 )
+or
+(key1 < 5) or (key1 > @maxv-10);
+pk	key1	key2	filler	filler2
+1	1	1	filler-data	filler-data-2
+2	2	2	filler-data	filler-data-2
+3	3	3	filler-data	filler-data-2
+4	4	4	filler-data	filler-data-2
+991	991	991	filler-data	filler-data-2
+992	992	992	filler-data	filler-data-2
+993	993	993	filler-data	filler-data-2
+994	994	994	filler-data	filler-data-2
+995	995	995	filler-data	filler-data-2
+996	996	996	filler-data	filler-data-2
+997	997	997	filler-data	filler-data-2
+998	998	998	filler-data	filler-data-2
+999	999	999	filler-data	filler-data-2
+1000	1000	1000	filler-data	filler-data-2
+11	11	11	filler-data	filler-data-2
+12	12	12	filler-data	filler-data-2
+13	13	13	filler-data	filler-data-2
+14	14	14	filler-data	filler-data-2
+50	50	50	filler-data	filler-data-2
+51	51	51	filler-data	filler-data-2
+52	52	52	filler-data	filler-data-2
+53	53	53	filler-data	filler-data-2
+54	54	54	filler-data	filler-data-2
+drop table t1;

mysql-test/t/index_merge_bdb.test

+#
+# 2-sweeps read Index_merge test
+#
+-- source include/have_bdb.inc
+
+--disable_warnings
+drop table if exists t1;
+--enable_warnings
+
+create table t1 (
+  pk int primary key,
+  key1 int,
+  key2 int,  
+  filler char(200),
+  filler2 char(200),  
+  index(key1),
+  index(key2),  
+) type=bdb;
+
+
+--disable_query_log
+let $1=1000;
+while ($1)
+{
+  eval insert into t1 values($1, $1, $1, 'filler-data','filler-data-2');
+  dec $1;
+}
+--enable_query_log
+
+select * from t1 where (key1 >= 2 and key1 <= 10) or (pk >= 4 and pk <=8 );
+
+set @maxv=1000;
+
+select * from t1 where 
+  (pk < 5) or (pk > 10 and pk < 15) or (pk >= 50 and pk < 55 ) or (pk > @maxv-10) 
+  or key1=18 or key1=60;
+
+select * from t1 where 
+  (pk < 5) or (pk > 10 and pk < 15) or (pk >= 50 and pk < 55 ) or (pk > @maxv-10) 
+  or key1 < 3 or key1 > @maxv-11;
+
+select * from t1 where 
+  (pk < 5) or (pk > 10 and pk < 15) or (pk >= 50 and pk < 55 ) or (pk > @maxv-10) 
+  or
+  (key1 < 5) or (key1 > 10 and key1 < 15) or (key1 >= 50 and key1 < 55 ) or (key1 > @maxv-10);
+
+select * from t1 where 
+  (pk > 10 and pk < 15) or (pk >= 50 and pk < 55 )
+  or
+  (key1 < 5) or (key1 > @maxv-10);
+
+drop table t1;

mysql-test/t/index_merge_innodb2.test

+#
+# 2-sweeps read Index_merge test
+#
+-- source include/have_innodb.inc
+
+--disable_warnings
+drop table if exists t1;
+--enable_warnings
+
+create table t1 (
+  pk int primary key,
+  key1 int,
+  key2 int,  
+  filler char(200),
+  filler2 char(200),  
+  index(key1),
+  index(key2),  
+) type=innodb;
+
+
+--disable_query_log
+let $1=1000;
+while ($1)
+{
+  eval insert into t1 values($1, $1, $1, 'filler-data','filler-data-2');
+  dec $1;
+}
+--enable_query_log
+
+select * from t1 where (key1 >= 2 and key1 <= 10) or (pk >= 4 and pk <=8 );
+
+set @maxv=1000;
+
+select * from t1 where 
+  (pk < 5) or (pk > 10 and pk < 15) or (pk >= 50 and pk < 55 ) or (pk > @maxv-10) 
+  or key1=18 or key1=60;
+
+select * from t1 where 
+  (pk < 5) or (pk > 10 and pk < 15) or (pk >= 50 and pk < 55 ) or (pk > @maxv-10) 
+  or key1 < 3 or key1 > @maxv-11;
+
+select * from t1 where 
+  (pk < 5) or (pk > 10 and pk < 15) or (pk >= 50 and pk < 55 ) or (pk > @maxv-10) 
+  or
+  (key1 < 5) or (key1 > 10 and key1 < 15) or (key1 >= 50 and key1 < 55 ) or (key1 > @maxv-10);
+
+select * from t1 where 
+  (pk > 10 and pk < 15) or (pk >= 50 and pk < 55 )
+  or
+  (key1 < 5) or (key1 > @maxv-10);
+
+drop table t1;

sql/filesort.cc

@@ -87,9 +87,15 @@ ha_rows filesort(THD *thd, TABLE *table, SORT_FIELD *sortorder, uint s_length,
   DBUG_PUSH("");		/* No DBUG here */
 #endif
   FILESORT_INFO table_sort;
-  bzero(&table_sort, sizeof(FILESORT_INFO));
+  /* 
+    don't use table->sort in filesort as it is also used by 
+    QUICK_INDEX_MERGE_SELECT. work with a copy of it and put it back at the 
+    end when index_merge select has finished with it.
+  */
+  memcpy(&table_sort, &table->sort, sizeof(FILESORT_INFO));
+  table->sort.io_cache= NULL;
   
-  outfile= table->sort.io_cache;
+  outfile= table_sort.io_cache;
   my_b_clear(&tempfile);
   my_b_clear(&buffpek_pointers);
   buffpek=0;
@@ -261,7 +267,6 @@ ha_rows filesort(THD *thd, TABLE *table, SORT_FIELD *sortorder, uint s_length,
   DBUG_POP();			/* Ok to DBUG */
 #endif
   memcpy(&table->sort, &table_sort, sizeof(FILESORT_INFO));
-  table->sort.io_cache= outfile;
   DBUG_PRINT("exit",("records: %ld",records));
   DBUG_RETURN(error ? HA_POS_ERROR : records);
 } /* filesort */
@@ -445,7 +450,13 @@ static ha_rows find_all_keys(SORTPARAM *param, SQL_SELECT *select,
       file->unlock_row();
   }
   if (quick_select)
+  {
+    /*
+      index_merge quick select uses table->sort when retrieving rows, so free 
+      resoures it has allocated.
+    */
     end_read_record(&read_record_info);
+  }
   else
   {
     (void) file->extra(HA_EXTRA_NO_CACHE);	/* End cacheing of records */

sql/ha_berkeley.h

@@ -167,6 +167,7 @@ class ha_berkeley: public handler
   longlong get_auto_increment();
   void print_error(int error, myf errflag);
   uint8 table_cache_type() { return HA_CACHE_TBL_TRANSACT; }
+  bool primary_key_is_clustered_covering() { return true; }
 };
 
 extern bool berkeley_skip, berkeley_shared_data;

sql/ha_innodb.cc

@@ -2001,10 +2001,12 @@ build_template(
 		update field->query_id so that the formula
 		thd->query_id == field->query_id did not work. */
 
-		if (templ_type == ROW_MYSQL_REC_FIELDS
-		    && !(fetch_all_in_key
-			 && dict_index_contains_col_or_prefix(index, i))
-		    && thd->query_id != field->query_id) {
+                ibool index_contains_field = dict_index_contains_col_or_prefix(index, i);
+
+		if (templ_type == ROW_MYSQL_REC_FIELDS && 
+                    ((prebuilt->read_just_key && !index_contains_field) ||
+		    (!(fetch_all_in_key && index_contains_field)
+		     && thd->query_id != field->query_id))) {
 
 			/* This field is not needed in the query, skip it */
 

sql/ha_innodb.h

@@ -187,6 +187,7 @@ class ha_innobase: public handler
 	void init_table_handle_for_HANDLER(); 
 	longlong get_auto_increment();
         uint8 table_cache_type() { return HA_CACHE_TBL_ASKTRANSACT; }
+        bool primary_key_is_clustered_covering() { return true; }
 };
 
 extern bool innodb_skip;

sql/handler.h

@@ -368,6 +368,13 @@ class handler :public Sql_alloc
   */
   static bool caching_allowed(THD* thd, char* table_key,
 			      uint key_length, uint8 cahe_type);
+
+  /*
+   RETURN
+     true  primary key (if there is one) is clustered key covering all fields
+     false otherwise
+  */
+  virtual bool primary_key_is_clustered_covering() { return false; }
 };
 
 	/* Some extern variables used with handlers */

sql/opt_range.cc

@@ -482,7 +482,7 @@ int SEL_IMERGE::or_sel_imerge_with_checks(PARAM *param, SEL_IMERGE* imerge)
 
 
 /* 
-  Perform AND operation on two index_merge lists, storing result in *im1.
+  Perform AND operation on two index_merge lists and store result in *im1.
 
 */
 
@@ -617,10 +617,11 @@ QUICK_SELECT_I::QUICK_SELECT_I()
 
 QUICK_RANGE_SELECT::QUICK_RANGE_SELECT(THD *thd, TABLE *table, uint key_nr, 
                                        bool no_alloc, MEM_ROOT *parent_alloc)
-  :dont_free(0),error(0),it(ranges),range(0)
+  :dont_free(0),error(0),range(0),cur_range(NULL)
 {
   index= key_nr;
   head=  table;
+  my_init_dynamic_array(&ranges, sizeof(QUICK_RANGE*), 16, 16);
 
   if (!no_alloc && !parent_alloc)
   {
@@ -644,17 +645,20 @@ QUICK_RANGE_SELECT::~QUICK_RANGE_SELECT()
   if (!dont_free)
   {
     file->index_end();
+    delete_dynamic(&ranges); /* ranges are allocated in alloc */
     free_root(&alloc,MYF(0));
   }
 }
 
 
 QUICK_INDEX_MERGE_SELECT::QUICK_INDEX_MERGE_SELECT(THD *thd_param, TABLE *table)
-  :cur_quick_it(quick_selects), thd(thd_param), unique(NULL)
+  :cur_quick_it(quick_selects), thd(thd_param), unique(NULL), 
+   pk_quick_select(NULL)
 {
   index= MAX_KEY;
   head= table;
   reset_called= false;
+  bzero(&read_record, sizeof(read_record));
   init_sql_alloc(&alloc,1024,0);
 }
 
@@ -662,7 +666,7 @@ int QUICK_INDEX_MERGE_SELECT::init()
 {
   cur_quick_it.rewind();
   cur_quick_select= cur_quick_it++;
-  return cur_quick_select->init();
+  return 0;
 }
 
 int QUICK_INDEX_MERGE_SELECT::reset()
@@ -673,19 +677,30 @@ int QUICK_INDEX_MERGE_SELECT::reset()
     DBUG_RETURN(0);
 
   reset_called= true;
-  result = cur_quick_select->reset() && prepare_unique();  
+  result = cur_quick_select->reset() || prepare_unique();
   DBUG_RETURN(result);
 }
 
 bool 
 QUICK_INDEX_MERGE_SELECT::push_quick_back(QUICK_RANGE_SELECT *quick_sel_range)
 {
+  /* 
+    Save quick_select that does scan on clustered covering primary key as 
+    it will be processed separately
+  */
+  if (head->file->primary_key_is_clustered_covering() && 
+      quick_sel_range->index == head->primary_key)
+    pk_quick_select= quick_sel_range;
+  else
     return quick_selects.push_back(quick_sel_range);
+  return 0;
 }
 
 QUICK_INDEX_MERGE_SELECT::~QUICK_INDEX_MERGE_SELECT()
 {
+  delete unique;
   quick_selects.delete_elements();
+  delete pk_quick_select;
   free_root(&alloc,MYF(0));
 }
 
@@ -1021,6 +1036,7 @@ int SQL_SELECT::test_quick_select(THD *thd, key_map keys_to_use,
           ha_rows min_imerge_records;
           
           List_iterator_fast<SEL_IMERGE> it(tree->merges);
+          /* find index_merge with minimal cost */
           while ((imerge= it++))
           {
             double  imerge_cost= 0;
@@ -1031,6 +1047,13 @@ int SQL_SELECT::test_quick_select(THD *thd, key_map keys_to_use,
               (SEL_ARG***)alloc_root(&alloc,
                                      (imerge->trees_next - imerge->trees)*
                                      sizeof(void*));
+            /*
+              It may be possible to use different keys for index_merge, e.g for
+              queries like 
+              ...WHERE (key1 < c2 AND key2 < c2) OR (key3 < c3 AND key4 < c4)
+              We assume we get the best index_merge if we choose the best key
+              read inside each of the conjuncts.
+            */
             for (SEL_TREE **ptree= imerge->trees;
                  ptree != imerge->trees_next;
                  ptree++)
@@ -1062,7 +1085,11 @@ imerge_fail:;
             goto end_free;
           
           records= min_imerge_records;
-          /* ok, got minimal imerge, *min_imerge, with cost min_imerge_cost */
+          /* 
+            Ok, got minimal index merge, *min_imerge, with cost min_imerge_cost
+            Compare its cost with "all" scan cost (or "all+using index" if 
+            it is possible) and choose the best.
+          */
           
           if (!head->used_keys.is_clear_all())
           {
@@ -1184,8 +1211,8 @@ imerge_fail:;
 
 
 /*
-  Calculate quick select read time, # of records, and best key to use 
-  without constructing QUICK_SELECT
+  Calculate quick range select read time, # of records, and best key to use
+  without constructing QUICK_RANGE_SELECT object.
 */
 
 static int get_quick_select_params(SEL_TREE *tree, PARAM& param, 
@@ -1448,6 +1475,7 @@ get_mm_parts(PARAM *param, Field *field, Item_func::Functype type,
     if (tree2)
       tree= tree_or(param,tree,tree2);
   }
+
   DBUG_RETURN(tree);
 }
 
@@ -1751,14 +1779,16 @@ tree_and(PARAM *param,SEL_TREE *tree1,SEL_TREE *tree2)
 
 
 /*
-  Check if two SEL_TREES can be combined into one without using index_merge
+  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 
+  using index_merge.
 */
 
 bool sel_trees_can_be_ored(SEL_TREE *tree1, SEL_TREE *tree2, PARAM* param)
 {
   key_map common_keys= tree1->keys_map;
-  common_keys.intersect(tree2->keys_map);
   DBUG_ENTER("sel_trees_can_be_ored");
+  common_keys.intersect(tree2->keys_map);
 
   if (common_keys.is_clear_all())
     DBUG_RETURN(false);
@@ -2973,7 +3003,9 @@ get_quick_keys(PARAM *param,QUICK_RANGE_SELECT *quick,KEY_PART *key,
   set_if_bigger(quick->max_used_key_length,range->min_length);
   set_if_bigger(quick->max_used_key_length,range->max_length);
   set_if_bigger(quick->used_key_parts, (uint) key_tree->part+1);
-  quick->ranges.push_back(range);
+  if (insert_dynamic(&quick->ranges, (gptr)&range))
+    return 1;
+
 
  end:
   if (key_tree->right != &null_element)
@@ -2991,8 +3023,8 @@ bool QUICK_RANGE_SELECT::unique_key_range()
 {
   if (ranges.elements == 1)
   {
-    QUICK_RANGE *tmp;
-    if (((tmp=ranges.head())->flag & (EQ_RANGE | NULL_RANGE)) == EQ_RANGE)
+    QUICK_RANGE *tmp= *((QUICK_RANGE**)ranges.buffer);
+    if ((tmp->flag & (EQ_RANGE | NULL_RANGE)) == EQ_RANGE)
     {
       KEY *key=head->key_info+index;
       return ((key->flags & HA_NOSAME) &&
@@ -3045,7 +3077,6 @@ QUICK_RANGE_SELECT *get_quick_select_for_ref(THD *thd, TABLE *table,
   {
     if (thd->is_fatal_error)
       return 0;					// out of memory
-    return quick;				// empty range
   }
 
   QUICK_RANGE *range= new QUICK_RANGE();
@@ -3070,7 +3101,7 @@ QUICK_RANGE_SELECT *get_quick_select_for_ref(THD *thd, TABLE *table,
       key_part->part_length+=HA_KEY_BLOB_LENGTH;
     key_part->null_bit=     key_info->key_part[part].null_bit;
   }
-  if (!quick->ranges.push_back(range))
+  if (!insert_dynamic(&quick->ranges,(gptr)&range))
     return quick;
 
 err:
@@ -3079,18 +3110,41 @@ QUICK_RANGE_SELECT *get_quick_select_for_ref(THD *thd, TABLE *table,
 }
 
 
-#define MEM_STRIP_BUF_SIZE current_thd->variables.sortbuff_size
+#define MEM_STRIP_BUF_SIZE thd->variables.sortbuff_size
+
 /*
   Fetch all row ids into unique.
+
+  If table has a clustered primary key(PK) that contains all rows (bdb and 
+     innodb currently) and one of the index_merge scans is a scan on primary key,
+  then 
+    primary key scan rowids are not put into Unique and also 
+    rows that will be retrieved by PK scan are not put into Unique
+  
+  RETURN
+    0     OK
+    other error
 */
+
 int QUICK_INDEX_MERGE_SELECT::prepare_unique()
 {
   int result;
   DBUG_ENTER("QUICK_INDEX_MERGE_SELECT::prepare_unique");
   
-  /* we're going to just read rowids */
+  /* we're going to just read rowids. */
   head->file->extra(HA_EXTRA_KEYREAD);
 
+  /* 
+    Make innodb retrieve all PK member fields, so 
+     * ha_innobase::position (which uses them) call works.
+     * we filter out rows retrieved by CCPK.
+    (This also creates a deficiency - it is possible that we will retrieve
+     parts of key that are not used by current query at all)
+  */
+  head->file->extra(HA_EXTRA_RETRIEVE_ALL_COLS);
+
+  cur_quick_select->init();
+
   unique= new Unique(refposcmp2, (void *) &head->file->ref_length,
                      head->file->ref_length,
                      MEM_STRIP_BUF_SIZE);
@@ -3103,9 +3157,12 @@ int QUICK_INDEX_MERGE_SELECT::prepare_unique()
       cur_quick_select= cur_quick_it++;
       if (!cur_quick_select)
         break;
-      cur_quick_select->init();
-      if (cur_quick_select->reset())
+
+      if (cur_quick_select->init())
         DBUG_RETURN(1);
+
+      /* QUICK_RANGE_SELECT::reset never fails */
+      cur_quick_select->reset();
     }
 
     if (result)
@@ -3125,28 +3182,61 @@ int QUICK_INDEX_MERGE_SELECT::prepare_unique()
     if (thd->killed)
       DBUG_RETURN(1);
     
+    /* skip row if it will be retrieved by clustered covering PK scan */
+    if (pk_quick_select && pk_quick_select->row_in_ranges())
+      continue;
+
     cur_quick_select->file->position(cur_quick_select->record);
-    if (unique->unique_add((char*)cur_quick_select->file->ref))
+    result= unique->unique_add((char*)cur_quick_select->file->ref);
+    
+    if (result)
       DBUG_RETURN(1);
 
   }while(true);  
 
   /* ok, all row ids are in Unique */
   result= unique->get(head);
-
+  doing_pk_scan= false;
+  init_read_record(&read_record, thd, head, NULL, 1, 1);
   /* index_merge currently doesn't support "using index" at all */
   head->file->extra(HA_EXTRA_NO_KEYREAD);
+
   DBUG_RETURN(result);
 }
 
+/*
+  Get next row for index_merge.
+  NOTES
+  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.
+*/
 int QUICK_INDEX_MERGE_SELECT::get_next()
 {
+  int result;
   DBUG_ENTER("QUICK_INDEX_MERGE_SELECT::get_next");
   
-  DBUG_PRINT("QUICK_INDEX_MERGE_SELECT", 
-             ("ERROR: index merge error: get_next should not be called "));
-  DBUG_ASSERT(0);
-  DBUG_RETURN(HA_ERR_END_OF_FILE);
+  if (doing_pk_scan)
+    DBUG_RETURN(pk_quick_select->get_next());
+
+  result= read_record.read_record(&read_record);
+
+  if (result == -1)
+  {
+    result= HA_ERR_END_OF_FILE;
+    /* All rows from Unique have been retrieved, do a CCPK scan */
+    end_read_record(&read_record);
+    if(pk_quick_select)
+    {
+      doing_pk_scan= true;
+      if ((result= pk_quick_select->init()))
+        DBUG_RETURN(result);
+      DBUG_RETURN(pk_quick_select->get_next());
+    }
+  }
+
+  DBUG_RETURN(result);
 }
 
 	/* get next possible record using quick-struct */
@@ -3172,16 +3262,21 @@ int QUICK_RANGE_SELECT::get_next()
 
       if (!result)
       {
-	if ((range->flag & GEOM_FLAG) || !cmp_next(*it.ref()))
+	if ((range->flag & GEOM_FLAG) || !cmp_next(*cur_range))
 	  DBUG_RETURN(0);
       }
       else if (result != HA_ERR_END_OF_FILE)
 	DBUG_RETURN(result);
     }
     
-    if (!(range=it++))
-      DBUG_RETURN(HA_ERR_END_OF_FILE);		// All ranges used
+    if (!cur_range)
+      range= *(cur_range= (QUICK_RANGE**)ranges.buffer);
+    else 
+      range= (cur_range == ((QUICK_RANGE**)ranges.buffer + ranges.elements - 1))? 
+            NULL: *(++cur_range);
 
+    if (!range)
+      DBUG_RETURN(HA_ERR_END_OF_FILE);		// All ranges used
     if (range->flag & GEOM_FLAG)
     {
       if ((result = file->index_read(record,
@@ -3271,6 +3366,45 @@ int QUICK_RANGE_SELECT::cmp_next(QUICK_RANGE *range_arg)
 }
 
 
+/*
+  Check if current row will be retrieved by this QUICK_RANGE_SELECT
+  (this is used to filter out CCPK scan rows in index_merge).
+
+  NOTES
+    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 
+    quick select.
+
+    The implementation does a binary search on sorted array of disjoint 
+    ranges, without taking size of range into account.
+
+  RETURN
+    true  if current row will be retrieved by this quick select
+    false if not
+*/
+
+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)
+  {    
+    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));
+}
+
 /*
   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
@@ -3288,13 +3422,14 @@ QUICK_SELECT_DESC::QUICK_SELECT_DESC(QUICK_RANGE_SELECT *q,
   bool not_read_after_key = file->table_flags() & HA_NOT_READ_AFTER_KEY;
   QUICK_RANGE *r;
   
-  it.rewind();
-  for (r = it++; r; r = it++)
+  QUICK_RANGE **pr= (QUICK_RANGE**)ranges.buffer;
+  QUICK_RANGE **last_range= pr + ranges.elements;
+  for (; pr!=last_range; ++pr)
   {
+    r= *pr;
     rev_ranges.push_front(r);
     if (not_read_after_key && range_reads_after_key(r))
     {
-      it.rewind();				// Reset range
       error = HA_ERR_UNSUPPORTED;
       dont_free=1;				// Don't free memory from 'q'
       return;
@@ -3412,7 +3547,7 @@ int QUICK_SELECT_DESC::get_next()
   Returns 0 if found key is inside range (found key >= range->min_key).
 */
 
-int QUICK_SELECT_DESC::cmp_prev(QUICK_RANGE *range_arg)
+int QUICK_RANGE_SELECT::cmp_prev(QUICK_RANGE *range_arg)
 {
   if (range_arg->flag & NO_MIN_RANGE)
     return 0;					/* key can't be to small */
@@ -3563,6 +3698,9 @@ static void print_quick_sel_imerge(QUICK_INDEX_MERGE_SELECT *quick,
   {
     print_quick_sel_range(quick_range_sel, needed_reg);
   }
+  if (quick->pk_quick_select)
+    print_quick_sel_range(quick->pk_quick_select, needed_reg);
+
   DBUG_VOID_RETURN;
 }
 
@@ -3574,12 +3712,15 @@ void print_quick_sel_range(QUICK_RANGE_SELECT *quick,const key_map *needed_reg)
   if (! _db_on_ || !quick)
     DBUG_VOID_RETURN;
 
-  List_iterator<QUICK_RANGE> li(quick->ranges);
   DBUG_LOCK_FILE;
   fprintf(DBUG_FILE,"Used quick_range on key: %d (other_keys: 0x%s):\n",
 	  quick->index, needed_reg->print(buf));
-  while ((range=li++))
+
+  QUICK_RANGE **pr= (QUICK_RANGE**)quick->ranges.buffer;
+  QUICK_RANGE **last_range= pr + quick->ranges.elements;
+  for (; pr!=last_range; ++pr)
   {
+    range= *pr;
     if (!(range->flag & NO_MIN_RANGE))
     {
       print_key(quick->key_parts,range->min_key,range->min_length);

sql/opt_range.h

@@ -68,7 +68,7 @@ class QUICK_RANGE :public Sql_alloc {
 
 /*
   Quick select interface. 
-  This class is parent for all QUICK_*_SELECT and FT_SELECT classes.
+  This class is a parent for all QUICK_*_SELECT and FT_SELECT classes.
 */
 
 class QUICK_SELECT_I
@@ -128,19 +128,29 @@ class QUICK_RANGE_SELECT : public QUICK_SELECT_I
                                               SEL_ARG *key_tree,
                                               MEM_ROOT *alloc);
   friend class QUICK_SELECT_DESC;
+  friend class QUICK_INDEX_MERGE_SELECT;
+
+  DYNAMIC_ARRAY ranges;     /* ordered array of range ptrs */
+  QUICK_RANGE **cur_range;  /* current element in ranges  */
 
-  List<QUICK_RANGE> ranges;
-  List_iterator<QUICK_RANGE> it;
   QUICK_RANGE *range;
   MEM_ROOT alloc;
   KEY_PART *key_parts;  
   int cmp_next(QUICK_RANGE *range);
+  int cmp_prev(QUICK_RANGE *range);
+  bool row_in_ranges();
 public:
   QUICK_RANGE_SELECT(THD *thd, TABLE *table,uint index_arg,bool no_alloc=0,
                      MEM_ROOT *parent_alloc=NULL);
   ~QUICK_RANGE_SELECT();
   
-  int reset(void) { next=0; it.rewind(); return 0; }
+  int reset(void)
+  {
+    next=0; 
+    range= NULL; 
+    cur_range= NULL;
+    return 0;
+  }
   int init();
   int get_next();
   bool reverse_sorted() { return 0; }
@@ -148,9 +158,60 @@ class QUICK_RANGE_SELECT : public QUICK_SELECT_I
   int get_type() { return QS_TYPE_RANGE; }
 };
 
+
 /*
-  Index merge quick select. 
-  It is implemented as a container for several QUICK_RANGE_SELECTs.
+QUICK_INDEX_MERGE_SELECT - index_merge acces method quick select.
+
+  QUICK_INDEX_MERGE_SELECT uses 
+   * QUICK_RANGE_SELECTs to get rows
+   * Unique class to remove duplicate rows
+
+INDEX MERGE OPTIMIZER
+  Current implementation doesn't detect all cases where index_merge could be 
+  used, in particular:
+   * index_merge will never be used if range scan is possible (even if range 
+     scan is more expensive)
+
+   * index_merge+'using index' is not supported (this the consequence of the 
+     above restriction)
+   
+   * If WHERE part contains complex nested AND and OR conditions, some ways to
+     retrieve rows using index_merge will not be considered. The choice of 
+     read plan may depend on the order of conjuncts/disjuncts in WHERE part of
+     the query, see comments near SEL_IMERGE::or_sel_tree_with_checks and 
+     imerge_list_or_list function for details.
+
+   * there is no "index_merge_ref" method (but index_merge on non-first table 
+     in join is possible with 'range checked for each record').
+
+   See comments around SEL_IMERGE class and test_quick_select for more details.
+
+ROW RETRIEVAL ALGORITHM
+
+  index_merge uses Unique class for duplicates removal.  Index merge takes 
+  advantage of clustered covering primary key (CCPK) if the table has one.
+  The algorithm is as follows:
+
+  prepare() //implemented in QUICK_INDEX_MERGE_SELECT::prepare_unique
+  {
+    activate 'index only';
+    while(retrieve next row for non-CCPK scan)
+    {
+      if (there is a CCPK scan and row will be retrieved by it)
+        skip this row;
+      else
+        put rowid into Unique;
+    }
+    deactivate 'index only';
+  }
+
+  fetch() //implemented as sequence of QUICK_INDEX_MERGE_SELECT::get_next calls
+  {
+    retrieve all rows from row pointers stored in Unique;
+    free Unique;
+    retrieve all rows for CCPK scan;
+  }
+
 */
 
 class QUICK_INDEX_MERGE_SELECT : public QUICK_SELECT_I 
@@ -175,15 +236,24 @@ class QUICK_INDEX_MERGE_SELECT : public QUICK_SELECT_I
   List_iterator_fast<QUICK_RANGE_SELECT> cur_quick_it;
   QUICK_RANGE_SELECT* cur_quick_select;
   
-  /* last element in quick_selects list. */
+  /* last element in quick_selects list */
   QUICK_RANGE_SELECT* last_quick_select;
 
+  /* quick select that uses Covering Clustered Primary Key (NULL if none) */
+  QUICK_RANGE_SELECT* pk_quick_select;
+  
+  /* true if this select is currently doing a CCPK scan */
+  bool  doing_pk_scan;
+  
   Unique  *unique;
   MEM_ROOT alloc;
 
   THD *thd;
   int prepare_unique();
   bool reset_called;
+  
+  /* used to get rows collected in Unique */
+  READ_RECORD read_record;
 };
 
 class QUICK_SELECT_DESC: public QUICK_RANGE_SELECT
@@ -194,7 +264,6 @@ class QUICK_SELECT_DESC: public QUICK_RANGE_SELECT
   bool reverse_sorted() { return 1; }
   int get_type() { return QS_TYPE_RANGE_DESC; }
 private:
-  int cmp_prev(QUICK_RANGE *range);
   bool range_reads_after_key(QUICK_RANGE *range);
 #ifdef NOT_USED
   bool test_if_null_range(QUICK_RANGE *range, uint used_key_parts);

sql/records.cc

@@ -97,8 +97,8 @@ void init_read_record(READ_RECORD *info,THD *thd, TABLE *table,
       }
     }
   }
-  else if (select && select->quick && 
-           (select->quick->get_type() != QUICK_SELECT_I::QS_TYPE_INDEX_MERGE))
+  else if (select && select->quick)
+           //&& (select->quick->get_type() != QUICK_SELECT_I::QS_TYPE_INDEX_MERGE))
   {
     DBUG_PRINT("info",("using rr_quick"));
     info->read_record=rr_quick;