Fix for bug#5138: hash indexes on heap tables support statistics.

 KEY::rec_per_key is updated every time 1/HEAP_STATS_UPDATE_THRESHOLD part of table records has been changed.


heap/_check.c:
  Hash indexes on heap tables now support statistics - number of hash buckets. The value is updated on any insert/delete operation.
heap/hp_clear.c:
  Hash indexes on heap tables now support statistics - number of hash buckets. The value is updated on any insert/delete operation.
heap/hp_create.c:
  Hash indexes on heap tables now support statistics - number of hash buckets. The value is updated on any insert/delete operation.
heap/hp_delete.c:
  Hash indexes on heap tables now support statistics - number of hash buckets. The value is updated on any insert/delete operation.
heap/hp_hash.c:
  Hash indexes on heap tables now support statistics - number of hash buckets. The value is updated on any insert/delete operation.
heap/hp_write.c:
  Hash indexes on heap tables now support statistics - number of hash buckets. The value is updated on any insert/delete operation.
include/heap.h:
  Hash indexes on heap tables now support statistics - number of hash buckets. The value is updated on any insert/delete operation.
mysql-test/r/heap.result:
  Fix for bug#5138: store/use statistics for hash indexes on heap tables
mysql-test/r/heap_hash.result:
  Fix for bug#5138: store/use statistics for hash indexes on heap tables
mysql-test/r/myisam.result:
  Fix for bug#5138: store/use statistics for hash indexes on heap tables
mysql-test/t/heap_hash.test:
  Fix for bug#5138: store/use statistics for hash indexes on heap tables
sql/structs.h:
  Added comments

heap/_check.c

@@ -102,9 +102,11 @@ static int check_one_key(HP_KEYDEF *keydef, uint keynr, ulong records,
   int error;
   uint i,found,max_links,seek,links;
   uint rec_link;				/* Only used with debugging */
+  uint hash_buckets_found;
   HASH_INFO *hash_info;
 
   error=0;
+  hash_buckets_found= 0;
   for (i=found=max_links=seek=0 ; i < records ; i++)
   {
     hash_info=hp_find_hash(&keydef->block,i);
@@ -128,21 +130,32 @@ static int check_one_key(HP_KEYDEF *keydef, uint keynr, ulong records,
 	  found++;
       }
       if (links > max_links) max_links=links;
+      hash_buckets_found++;
     }
   }
   if (found != records)
   {
-    DBUG_PRINT("error",("Found %ld of %ld records"));
+    DBUG_PRINT("error",("Found %ld of %ld records", found, records));
+    error=1;
+  }
+  if (keydef->hash_buckets != hash_buckets_found)
+  {
+    DBUG_PRINT("error",("Found %ld buckets, stats shows %ld buckets",
+                        hash_buckets_found, keydef->hash_buckets));
     error=1;
   }
   DBUG_PRINT("info",
-	     ("records: %ld   seeks: %d   max links: %d   hitrate: %.2f",
+	     ("records: %ld   seeks: %d   max links: %d   hitrate: %.2f   "
+              "buckets: %d",
 	      records,seek,max_links,
-	      (float) seek / (float) (records ? records : 1)));
+	      (float) seek / (float) (records ? records : 1), 
+              hash_buckets_found));
   if (print_status)
-    printf("Key: %d  records: %ld   seeks: %d   max links: %d   hitrate: %.2f\n",
+    printf("Key: %d  records: %ld   seeks: %d   max links: %d   "
+           "hitrate: %.2f   buckets: %d\n",
 	   keynr, records, seek, max_links,
-	   (float) seek / (float) (records ? records : 1));
+	   (float) seek / (float) (records ? records : 1), 
+           hash_buckets_found);
   return error;
 }
 

heap/hp_clear.c

@@ -97,6 +97,7 @@ void hp_clear_keys(HP_SHARE *info)
         VOID(hp_free_level(block,block->levels,block->root,(byte*) 0));
       block->levels=0;
       block->last_allocated=0;
+      keyinfo->hash_buckets= 0;
     }
   }
   info->index_length=0;

heap/hp_create.c

@@ -128,6 +128,7 @@ int heap_create(const char *name, uint keys, HP_KEYDEF *keydef,
 		   max_records);
 	keyinfo->delete_key= hp_delete_key;
 	keyinfo->write_key= hp_write_key;
+        keyinfo->hash_buckets= 0;
       }
     }
     share->min_records= min_records;

heap/hp_delete.c

@@ -151,6 +151,8 @@ int hp_delete_key(HP_INFO *info, register HP_KEYDEF *keyinfo,
     pos->ptr_to_rec=empty->ptr_to_rec;
     pos->next_key=empty->next_key;
   }
+  else
+    keyinfo->hash_buckets--;
 
   if (empty == lastpos)			/* deleted last hash key */
     DBUG_RETURN (0);
@@ -187,7 +189,11 @@ int hp_delete_key(HP_INFO *info, register HP_KEYDEF *keyinfo,
     }
     pos3= pos;				/* Link pos->next after lastpos */
   }
-  else pos3= 0;				/* Different positions merge */
+  else
+  {
+    pos3= 0;				/* Different positions merge */
+    keyinfo->hash_buckets--;
+  }
 
   empty[0]=lastpos[0];
   hp_movelink(pos3, empty, pos->next_key);

heap/hp_hash.c

@@ -196,7 +196,18 @@ byte *hp_search_next(HP_INFO *info, HP_KEYDEF *keyinfo, const byte *key,
 }
 
 
-	/* Calculate pos according to keys */
+/*
+  Calculate position number for hash value.
+  SYNOPSIS
+    hp_mask()
+      hashnr     Hash value
+      buffmax    Value such that
+                 2^(n-1) < maxlength <= 2^n = buffmax
+      maxlength  
+  
+  RETURN
+    Array index, in [0..maxlength)
+*/
 
 ulong hp_mask(ulong hashnr, ulong buffmax, ulong maxlength)
 {
@@ -205,7 +216,12 @@ ulong hp_mask(ulong hashnr, ulong buffmax, ulong maxlength)
 }
 
 
-	/* Change link from pos to new_link */
+/*
+  Change
+    next_link -> ... -> X -> pos
+  to
+    next_link -> ... -> X -> newlink
+*/
 
 void hp_movelink(HASH_INFO *pos, HASH_INFO *next_link, HASH_INFO *newlink)
 {

heap/hp_write.c

@@ -36,7 +36,7 @@ int heap_write(HP_INFO *info, const byte *record)
   byte *pos;
   HP_SHARE *share=info->s;
   DBUG_ENTER("heap_write");
-
+  printf("heap_write\n");
 #ifndef DBUG_OFF
   if (info->mode & O_RDONLY)
   {
@@ -180,15 +180,31 @@ int hp_write_key(HP_INFO *info, HP_KEYDEF *keyinfo,
     DBUG_RETURN(-1);				/* No more memory */
   halfbuff= (long) share->blength >> 1;
   pos= hp_find_hash(&keyinfo->block,(first_index=share->records-halfbuff));
-
+  
+  /*
+    We're about to add one more hash position, with hash_mask=#records. 
+    Entries that should be relocated to that position are currently members 
+    of the list that starts at #first_index position.
+    At #first_index position there may be either:
+    a) A list of items with hash_mask=first_index. The list contains 
+       1) entries that should be relocated to the list that starts at new 
+          position we're adding
+       2) entries that should be left in the list starting at #first_index 
+          position 
+    or
+    b) An entry with hashnr != first_index. We don't need to move it.
+  */
   if (pos != empty)				/* If some records */
   {
     do
     {
       hashnr = hp_rec_hashnr(keyinfo, pos->ptr_to_rec);
       if (flag == 0)				/* First loop; Check if ok */
+      {
+        /* Bail out if we're dealing with case b) from above comment */
 	if (hp_mask(hashnr, share->blength, share->records) != first_index)
 	  break;
+      }
       if (!(hashnr & halfbuff))
       {						/* Key will not move */
 	if (!(flag & LOWFIND))
@@ -245,6 +261,9 @@ int hp_write_key(HP_INFO *info, HP_KEYDEF *keyinfo,
       }
     }
     while ((pos=pos->next_key));
+    
+    if ((flag & (LOWFIND | HIGHFIND)) == (LOWFIND | HIGHFIND)) 
+      keyinfo->hash_buckets++;
 
     if ((flag & (LOWFIND | LOWUSED)) == LOWFIND)
     {
@@ -265,6 +284,7 @@ int hp_write_key(HP_INFO *info, HP_KEYDEF *keyinfo,
   {
     pos->ptr_to_rec=recpos;
     pos->next_key=0;
+    keyinfo->hash_buckets++;
   }
   else
   {
@@ -280,6 +300,7 @@ int hp_write_key(HP_INFO *info, HP_KEYDEF *keyinfo,
     }
     else
     {
+      keyinfo->hash_buckets++;
       pos->ptr_to_rec=recpos;
       pos->next_key=0;
       hp_movelink(pos, gpos, empty);

include/heap.h

@@ -87,6 +87,11 @@ typedef struct st_hp_keydef		/* Key definition with open */
   uint8 algorithm;			/* HASH / BTREE */
   HA_KEYSEG *seg;
   HP_BLOCK block;			/* Where keys are saved */
+  /*
+    Number of buckets used in hash table. Used only to provide
+    #records estimates for heap key scans.
+  */
+  ha_rows hash_buckets; 
   TREE rb_tree;
   int (*write_key)(struct st_heap_info *info, struct st_hp_keydef *keyinfo,
 		   const byte *record, byte *recpos);
@@ -102,7 +107,7 @@ typedef struct st_heap_share
   ulong min_records,max_records;	/* Params to open */
   ulong data_length,index_length;
   uint records;				/* records */
-  uint blength;
+  uint blength;				/* records rounded up to 2^n */
   uint deleted;				/* Deleted records in database */
   uint reclength;			/* Length of one record */
   uint changed;

mysql-test/r/heap.result

@@ -4,7 +4,7 @@ insert into t1 values(1,1),(2,2),(3,3),(4,4);
 delete from t1 where a=1 or a=0;
 show keys from t1;
 Table	Non_unique	Key_name	Seq_in_index	Column_name	Collation	Cardinality	Sub_part	Packed	Null	Index_type	Comment
-t1	0	PRIMARY	1	a	NULL	NULL	NULL	NULL		HASH	
+t1	0	PRIMARY	1	a	NULL	3	NULL	NULL		HASH	
 select * from t1;
 a	b
 2	2
@@ -169,7 +169,7 @@ id	select_type	table	type	possible_keys	key	key_len	ref	rows	Extra
 1	SIMPLE	t1	ALL	btn	NULL	NULL	NULL	11	Using where
 explain select * from t1 where btn="a" and new_col="a";
 id	select_type	table	type	possible_keys	key	key_len	ref	rows	Extra
-1	SIMPLE	t1	ref	btn	btn	11	const,const	10	Using where
+1	SIMPLE	t1	ref	btn	btn	11	const,const	2	Using where
 drop table t1;
 CREATE TABLE t1 (
 a int default NULL,
@@ -182,7 +182,7 @@ SELECT * FROM t1 WHERE a=NULL;
 a	b
 explain SELECT * FROM t1 WHERE a IS NULL;
 id	select_type	table	type	possible_keys	key	key_len	ref	rows	Extra
-1	SIMPLE	t1	ref	a	a	5	const	10	Using where
+1	SIMPLE	t1	ref	a	a	5	const	1	Using where
 SELECT * FROM t1 WHERE a<=>NULL;
 a	b
 NULL	99
@@ -204,7 +204,7 @@ key a (a)
 INSERT INTO t1 VALUES (10), (10), (10);
 EXPLAIN SELECT * FROM t1 WHERE a=10;
 id	select_type	table	type	possible_keys	key	key_len	ref	rows	Extra
-1	SIMPLE	t1	ref	a	a	5	const	10	Using where
+1	SIMPLE	t1	ref	a	a	5	const	3	Using where
 SELECT * FROM t1 WHERE a=10;
 a
 10

mysql-test/r/heap_hash.result

-drop table if exists t1;
+drop table if exists t1,t2;
 create table t1 (a int not null,b int not null, primary key using HASH (a)) engine=heap comment="testing heaps" avg_row_length=100 min_rows=1 max_rows=100;
 insert into t1 values(1,1),(2,2),(3,3),(4,4);
 delete from t1 where a=1 or a=0;
 show keys from t1;
 Table	Non_unique	Key_name	Seq_in_index	Column_name	Collation	Cardinality	Sub_part	Packed	Null	Index_type	Comment
-t1	0	PRIMARY	1	a	NULL	NULL	NULL	NULL		HASH	
+t1	0	PRIMARY	1	a	NULL	3	NULL	NULL		HASH	
 select * from t1;
 a	b
 2	2
@@ -169,7 +169,7 @@ id	select_type	table	type	possible_keys	key	key_len	ref	rows	Extra
 1	SIMPLE	t1	ALL	btn	NULL	NULL	NULL	11	Using where
 explain select * from t1 where btn="a" and new_col="a";
 id	select_type	table	type	possible_keys	key	key_len	ref	rows	Extra
-1	SIMPLE	t1	ref	btn	btn	11	const,const	10	Using where
+1	SIMPLE	t1	ref	btn	btn	11	const,const	2	Using where
 drop table t1;
 CREATE TABLE t1 (
 a int default NULL,
@@ -182,7 +182,7 @@ SELECT * FROM t1 WHERE a=NULL;
 a	b
 explain SELECT * FROM t1 WHERE a IS NULL;
 id	select_type	table	type	possible_keys	key	key_len	ref	rows	Extra
-1	SIMPLE	t1	ref	a	a	5	const	10	Using where
+1	SIMPLE	t1	ref	a	a	5	const	1	Using where
 SELECT * FROM t1 WHERE a<=>NULL;
 a	b
 NULL	99
@@ -203,3 +203,139 @@ DELETE from t1 where a < 100;
 SELECT * from t1;
 a
 DROP TABLE t1;
+create table t1
+(
+a char(8) not null,
+b char(20) not null,
+c int not null,
+key (a)
+) engine=heap;
+insert into t1 values ('aaaa', 'prefill-hash=5',0);
+insert into t1 values ('aaab', 'prefill-hash=0',0);
+insert into t1 values ('aaac', 'prefill-hash=7',0);
+insert into t1 values ('aaad', 'prefill-hash=2',0);
+insert into t1 values ('aaae', 'prefill-hash=1',0);
+insert into t1 values ('aaaf', 'prefill-hash=4',0);
+insert into t1 values ('aaag', 'prefill-hash=3',0);
+insert into t1 values ('aaah', 'prefill-hash=6',0);
+explain select * from t1 where a='aaaa';
+id	select_type	table	type	possible_keys	key	key_len	ref	rows	Extra
+1	SIMPLE	t1	ref	a	a	8	const	1	Using where
+explain select * from t1 where a='aaab';
+id	select_type	table	type	possible_keys	key	key_len	ref	rows	Extra
+1	SIMPLE	t1	ref	a	a	8	const	1	Using where
+explain select * from t1 where a='aaac';
+id	select_type	table	type	possible_keys	key	key_len	ref	rows	Extra
+1	SIMPLE	t1	ref	a	a	8	const	1	Using where
+explain select * from t1 where a='aaad';
+id	select_type	table	type	possible_keys	key	key_len	ref	rows	Extra
+1	SIMPLE	t1	ref	a	a	8	const	1	Using where
+insert into t1 select * from t1;
+explain select * from t1 where a='aaaa';
+id	select_type	table	type	possible_keys	key	key_len	ref	rows	Extra
+1	SIMPLE	t1	ref	a	a	8	const	2	Using where
+explain select * from t1 where a='aaab';
+id	select_type	table	type	possible_keys	key	key_len	ref	rows	Extra
+1	SIMPLE	t1	ref	a	a	8	const	2	Using where
+explain select * from t1 where a='aaac';
+id	select_type	table	type	possible_keys	key	key_len	ref	rows	Extra
+1	SIMPLE	t1	ref	a	a	8	const	2	Using where
+explain select * from t1 where a='aaad';
+id	select_type	table	type	possible_keys	key	key_len	ref	rows	Extra
+1	SIMPLE	t1	ref	a	a	8	const	2	Using where
+create table t2 as select * from t1;
+delete from t1;
+insert into t1 select * from t2;
+explain select * from t1 where a='aaaa';
+id	select_type	table	type	possible_keys	key	key_len	ref	rows	Extra
+1	SIMPLE	t1	ref	a	a	8	const	2	Using where
+explain select * from t1 where a='aaab';
+id	select_type	table	type	possible_keys	key	key_len	ref	rows	Extra
+1	SIMPLE	t1	ref	a	a	8	const	2	Using where
+explain select * from t1 where a='aaac';
+id	select_type	table	type	possible_keys	key	key_len	ref	rows	Extra
+1	SIMPLE	t1	ref	a	a	8	const	2	Using where
+explain select * from t1 where a='aaad';
+id	select_type	table	type	possible_keys	key	key_len	ref	rows	Extra
+1	SIMPLE	t1	ref	a	a	8	const	2	Using where
+drop table t1, t2;
+create table t1 (
+id int unsigned not null primary key auto_increment, 
+name varchar(20) not null,
+index heap_idx(name),
+index btree_idx using btree(name)
+) engine=heap;
+create table t2 (
+id int unsigned not null primary key auto_increment, 
+name varchar(20) not null,
+index btree_idx using btree(name),
+index heap_idx(name)
+) engine=heap;
+insert into t1 (name) values ('Matt'), ('Lilu'), ('Corbin'), ('Carly'), 
+('Suzy'), ('Hoppy'), ('Burrito'), ('Mimi'), ('Sherry'), ('Ben'), ('Phil'), 
+('Emily'), ('Mike');
+insert into t2 select * from t1;
+explain select * from t1 where name='matt';
+id	select_type	table	type	possible_keys	key	key_len	ref	rows	Extra
+1	SIMPLE	t1	ref	heap_idx,btree_idx	heap_idx	20	const	1	Using where
+explain select * from t2 where name='matt';
+id	select_type	table	type	possible_keys	key	key_len	ref	rows	Extra
+1	SIMPLE	t2	ref	btree_idx,heap_idx	btree_idx	20	const	1	Using where
+explain select * from t1 where name='Lilu';
+id	select_type	table	type	possible_keys	key	key_len	ref	rows	Extra
+1	SIMPLE	t1	ref	heap_idx,btree_idx	heap_idx	20	const	1	Using where
+explain select * from t2 where name='Lilu';
+id	select_type	table	type	possible_keys	key	key_len	ref	rows	Extra
+1	SIMPLE	t2	ref	btree_idx,heap_idx	btree_idx	20	const	1	Using where
+explain select * from t1 where name='Phil';
+id	select_type	table	type	possible_keys	key	key_len	ref	rows	Extra
+1	SIMPLE	t1	ref	heap_idx,btree_idx	heap_idx	20	const	1	Using where
+explain select * from t2 where name='Phil';
+id	select_type	table	type	possible_keys	key	key_len	ref	rows	Extra
+1	SIMPLE	t2	ref	btree_idx,heap_idx	btree_idx	20	const	1	Using where
+explain select * from t1 where name='Lilu';
+id	select_type	table	type	possible_keys	key	key_len	ref	rows	Extra
+1	SIMPLE	t1	ref	heap_idx,btree_idx	heap_idx	20	const	1	Using where
+explain select * from t2 where name='Lilu';
+id	select_type	table	type	possible_keys	key	key_len	ref	rows	Extra
+1	SIMPLE	t2	ref	btree_idx,heap_idx	btree_idx	20	const	1	Using where
+insert into t1 (name) select name from t2;
+insert into t1 (name) select name from t2;
+insert into t1 (name) select name from t2;
+insert into t1 (name) select name from t2;
+insert into t1 (name) select name from t2;
+insert into t1 (name) select name from t2;
+select count(*) from t1 where name='Matt';
+count(*)
+7
+explain select * from t1 ignore index (btree_idx) where name='matt';
+id	select_type	table	type	possible_keys	key	key_len	ref	rows	Extra
+1	SIMPLE	t1	ref	heap_idx	heap_idx	20	const	7	Using where
+show index from t1;
+Table	Non_unique	Key_name	Seq_in_index	Column_name	Collation	Cardinality	Sub_part	Packed	Null	Index_type	Comment
+t1	0	PRIMARY	1	id	NULL	91	NULL	NULL		HASH	
+t1	1	heap_idx	1	name	NULL	15	NULL	NULL		HASH	
+t1	1	btree_idx	1	name	A	NULL	NULL	NULL		BTREE	
+flush tables;
+show index from t1;
+Table	Non_unique	Key_name	Seq_in_index	Column_name	Collation	Cardinality	Sub_part	Packed	Null	Index_type	Comment
+t1	0	PRIMARY	1	id	NULL	91	NULL	NULL		HASH	
+t1	1	heap_idx	1	name	NULL	13	NULL	NULL		HASH	
+t1	1	btree_idx	1	name	A	NULL	NULL	NULL		BTREE	
+create table t3
+(
+a varchar(20) not null,
+b varchar(20) not null,
+key (a,b)
+) engine=heap;
+insert into t3 select name, name from t1;
+show index from t3;
+Table	Non_unique	Key_name	Seq_in_index	Column_name	Collation	Cardinality	Sub_part	Packed	Null	Index_type	Comment
+t3	1	a	1	a	NULL	NULL	NULL	NULL		HASH	
+t3	1	a	2	b	NULL	15	NULL	NULL		HASH	
+flush tables;
+show index from t3;
+Table	Non_unique	Key_name	Seq_in_index	Column_name	Collation	Cardinality	Sub_part	Packed	Null	Index_type	Comment
+t3	1	a	1	a	NULL	NULL	NULL	NULL		HASH	
+t3	1	a	2	b	NULL	13	NULL	NULL		HASH	
+drop table t1,t2;

mysql-test/r/myisam.result

@@ -543,7 +543,7 @@ Warnings:
 Note	1031	Table storage engine for 't1' doesn't have this option
 show keys from t1;
 Table	Non_unique	Key_name	Seq_in_index	Column_name	Collation	Cardinality	Sub_part	Packed	Null	Index_type	Comment
-t1	1	a	1	a	NULL	NULL	NULL	NULL	YES	HASH	
+t1	1	a	1	a	NULL	1000	NULL	NULL	YES	HASH	
 drop table t1,t2;
 create table t1 ( a tinytext, b char(1), index idx (a(1),b) );
 insert into t1 values (null,''), (null,'');

mysql-test/t/heap_hash.test

@@ -3,7 +3,7 @@
 #
 
 --disable_warnings
-drop table if exists t1;
+drop table if exists t1,t2;
 --enable_warnings
 
 create table t1 (a int not null,b int not null, primary key using HASH (a)) engine=heap comment="testing heaps" avg_row_length=100 min_rows=1 max_rows=100;
@@ -141,3 +141,105 @@ INSERT into t1 values (1),(2),(3),(4),(5),(6),(7),(8),(9),(10),(11);
 DELETE from t1 where a < 100;
 SELECT * from t1;
 DROP TABLE t1;
+
+
+#
+# Hash index # records estimate test
+#
+create table t1
+(
+  a char(8) not null,
+  b char(20) not null,
+  c int not null,
+  key (a)
+) engine=heap;
+
+insert into t1 values ('aaaa', 'prefill-hash=5',0);
+insert into t1 values ('aaab', 'prefill-hash=0',0);
+insert into t1 values ('aaac', 'prefill-hash=7',0);
+insert into t1 values ('aaad', 'prefill-hash=2',0);
+insert into t1 values ('aaae', 'prefill-hash=1',0);
+insert into t1 values ('aaaf', 'prefill-hash=4',0);
+insert into t1 values ('aaag', 'prefill-hash=3',0);
+insert into t1 values ('aaah', 'prefill-hash=6',0);
+
+explain select * from t1 where a='aaaa';
+explain select * from t1 where a='aaab';
+explain select * from t1 where a='aaac';
+explain select * from t1 where a='aaad';
+insert into t1 select * from t1;
+explain select * from t1 where a='aaaa';
+explain select * from t1 where a='aaab';
+explain select * from t1 where a='aaac';
+explain select * from t1 where a='aaad';
+
+# Check if delete_all_rows() updates #hash_buckets
+create table t2 as select * from t1;
+delete from t1;
+insert into t1 select * from t2;
+explain select * from t1 where a='aaaa';
+explain select * from t1 where a='aaab';
+explain select * from t1 where a='aaac';
+explain select * from t1 where a='aaad';
+drop table t1, t2;
+
+
+# Btree and hash index use costs. 
+create table t1 (
+  id int unsigned not null primary key auto_increment, 
+  name varchar(20) not null,
+  index heap_idx(name),
+  index btree_idx using btree(name)
+) engine=heap;
+
+create table t2 (
+  id int unsigned not null primary key auto_increment, 
+  name varchar(20) not null,
+  index btree_idx using btree(name),
+  index heap_idx(name)
+) engine=heap;
+
+insert into t1 (name) values ('Matt'), ('Lilu'), ('Corbin'), ('Carly'), 
+  ('Suzy'), ('Hoppy'), ('Burrito'), ('Mimi'), ('Sherry'), ('Ben'), ('Phil'), 
+  ('Emily'), ('Mike');
+insert into t2 select * from t1;
+
+explain select * from t1 where name='matt';
+explain select * from t2 where name='matt';
+
+explain select * from t1 where name='Lilu';
+explain select * from t2 where name='Lilu';
+
+explain select * from t1 where name='Phil';
+explain select * from t2 where name='Phil';
+
+explain select * from t1 where name='Lilu';
+explain select * from t2 where name='Lilu';
+
+insert into t1 (name) select name from t2;
+insert into t1 (name) select name from t2;
+insert into t1 (name) select name from t2;
+insert into t1 (name) select name from t2;
+insert into t1 (name) select name from t2;
+insert into t1 (name) select name from t2;
+
+select count(*) from t1 where name='Matt';
+explain select * from t1 ignore index (btree_idx) where name='matt';
+show index from t1;
+
+flush tables;
+show index from t1;
+
+create table t3
+(
+  a varchar(20) not null,
+  b varchar(20) not null,
+  key (a,b)
+) engine=heap;
+insert into t3 select name, name from t1;
+show index from t3;
+flush tables;
+show index from t3;
+
+drop table t1,t2;
+

sql/ha_heap.cc

@@ -30,6 +30,12 @@
 const char **ha_heap::bas_ext() const
 { static const char *ext[1]= { NullS }; return ext; }
 
+/*
+  Hash index statistics is updated (copied from HP_KEYDEF::hash_buckets to 
+  rec_per_key) after 1/HEAP_STATS_UPDATE_THRESHOLD records have been inserted/
+  updated/deleted. delete_all_rows() and table flush cause immediate update.
+*/
+#define HEAP_STATS_UPDATE_THRESHOLD 10
 
 int ha_heap::open(const char *name, int mode, uint test_if_locked)
 {
@@ -48,6 +54,8 @@ int ha_heap::open(const char *name, int mode, uint test_if_locked)
   {
     /* Initialize variables for the opened table */
     set_keys_for_scanning();
+    if (table->tmp_table == NO_TMP_TABLE)
+      update_key_stats();
   }
   return (file ? 0 : 1);
 }
@@ -84,28 +92,59 @@ void ha_heap::set_keys_for_scanning(void)
   }
 }
 
+void ha_heap::update_key_stats()
+{
+  printf("update_key_stats\n");
+  for (uint i= 0; i < table->keys; i++)
+  {
+    KEY *key=table->key_info+i;
+    if (key->algorithm != HA_KEY_ALG_BTREE)
+    {
+      ha_rows hash_buckets= file->s->keydef[i].hash_buckets;
+      key->rec_per_key[key->key_parts-1]= 
+        hash_buckets ? file->s->records/hash_buckets : 0;
+    }
+  }
+  records_changed= 0;
+}
+
 int ha_heap::write_row(byte * buf)
 {
+  int res;
   statistic_increment(ha_write_count,&LOCK_status);
   if (table->timestamp_default_now)
     update_timestamp(buf+table->timestamp_default_now-1);
   if (table->next_number_field && buf == table->record[0])
     update_auto_increment();
-  return heap_write(file,buf);
+  res= heap_write(file,buf);
+  if (!res && table->tmp_table == NO_TMP_TABLE && 
+      ++records_changed*HEAP_STATS_UPDATE_THRESHOLD > file->s->records)
+    update_key_stats();
+  return res;
 }
 
 int ha_heap::update_row(const byte * old_data, byte * new_data)
 {
+  int res;
   statistic_increment(ha_update_count,&LOCK_status);
   if (table->timestamp_on_update_now)
     update_timestamp(new_data+table->timestamp_on_update_now-1);
-  return heap_update(file,old_data,new_data);
+  res= heap_update(file,old_data,new_data);
+  if (!res && table->tmp_table == NO_TMP_TABLE && 
+      ++records_changed*HEAP_STATS_UPDATE_THRESHOLD > file->s->records)
+    update_key_stats();
+  return res;
 }
 
 int ha_heap::delete_row(const byte * buf)
 {
+  int res;
   statistic_increment(ha_delete_count,&LOCK_status);
-  return heap_delete(file,buf);
+  res= heap_delete(file,buf);
+  if (!res && table->tmp_table == NO_TMP_TABLE && 
+      ++records_changed*HEAP_STATS_UPDATE_THRESHOLD > file->s->records)
+    update_key_stats();
+  return res;
 }
 
 int ha_heap::index_read(byte * buf, const byte * key, uint key_len,
@@ -227,6 +266,8 @@ int ha_heap::extra(enum ha_extra_function operation)
 int ha_heap::delete_all_rows()
 {
   heap_clear(file);
+  if (table->tmp_table == NO_TMP_TABLE)
+    update_key_stats();
   return 0;
 }
 
@@ -383,7 +424,14 @@ ha_rows ha_heap::records_in_range(uint inx, key_range *min_key,
       min_key->flag != HA_READ_KEY_EXACT ||
       max_key->flag != HA_READ_AFTER_KEY)
     return HA_POS_ERROR;			// Can only use exact keys
-  return 10;					// Good guess
+  else
+  {
+    ha_rows records= file->s->records;
+    if (!records)
+      return 0;
+    ha_rows res= records / file->s->keydef[inx].hash_buckets;
+    return res ? res : 1;
+  }
 }
 
 

sql/ha_heap.h

@@ -27,9 +27,10 @@ class ha_heap: public handler
 {
   HP_INFO *file;
   key_map btree_keys;
-
- public:
-  ha_heap(TABLE *table): handler(table), file(0) {}
+  /* number of records changed since last statistics update */
+  uint    records_changed;
+public:
+  ha_heap(TABLE *table): handler(table), file(0), records_changed(0) {}
   ~ha_heap() {}
   const char *table_type() const { return "HEAP"; }
   const char *index_type(uint inx)
@@ -91,5 +92,6 @@ class ha_heap: public handler
 
   THR_LOCK_DATA **store_lock(THD *thd, THR_LOCK_DATA **to,
 			     enum thr_lock_type lock_type);
-
+private:
+  void update_key_stats();
 };

sql/structs.h

@@ -89,7 +89,12 @@ typedef struct st_key {
   enum  ha_key_alg algorithm;
   KEY_PART_INFO *key_part;
   char	*name;				/* Name of key */
-  ulong *rec_per_key;			/* Key part distribution */
+  /*
+    Array of AVG(#records with the same field value) for 1st ... Nth key part.
+    0 means 'not known'.
+    For temporary heap tables this member is NULL.
+  */
+  ulong *rec_per_key;
   union {
     int  bdb_return_if_eq;
   } handler;