BUG#12232: Add a server option to treat NULL values as equal when calculating MyISAM index

statistics (like 4.0 did) (patch #3, with review #1 & #2 feedback addressed)

include/myisam.h

@@ -311,6 +311,20 @@ typedef struct st_sort_key_blocks		/* Used when sorting */
 } SORT_KEY_BLOCKS;
 
 
+/* 
+  MyISAM supports several statistics collection methods. Currently statistics 
+  collection method is not stored in MyISAM file and has to be specified for 
+  each table analyze/repair operation in  MI_CHECK::stats_method.
+*/
+
+typedef enum 
+{
+  /* Treat NULLs as inequal when collecting statistics (default for 4.1/5.0) */
+  MI_STATS_METHOD_NULLS_NOT_EQUAL,
+  /* Treat NULLs as equal when collecting statistics (like 4.0 did) */
+  MI_STATS_METHOD_NULLS_EQUAL
+} enum_mi_stats_method;
+
 typedef struct st_mi_check_param
 {
   ulonglong auto_increment_value;
@@ -341,6 +355,7 @@ typedef struct st_mi_check_param
   void *thd;
   char *db_name,*table_name;
   char *op_name;
+  enum_mi_stats_method stats_method;
 } MI_CHECK;
 
 typedef struct st_sort_ft_buf

myisam/mi_check.c

@@ -80,6 +80,7 @@ void myisamchk_init(MI_CHECK *param)
   param->start_check_pos=0;
   param->max_record_length= LONGLONG_MAX;
   param->key_cache_block_size= KEY_CACHE_BLOCK_SIZE;
+  param->stats_method= MI_STATS_METHOD_NULLS_NOT_EQUAL;
 }
 
 	/* Check the status flags for the table */
@@ -558,10 +559,11 @@ static int chk_index(MI_CHECK *param, MI_INFO *info, MI_KEYDEF *keyinfo,
 		     ha_checksum *key_checksum, uint level)
 {
   int flag;
-  uint used_length,comp_flag,nod_flag,key_length=0,not_used;
+  uint used_length,comp_flag,nod_flag,key_length=0;
   uchar key[MI_MAX_POSSIBLE_KEY_BUFF],*temp_buff,*keypos,*old_keypos,*endpos;
   my_off_t next_page,record;
   char llbuff[22];
+  uint diff_pos;
   DBUG_ENTER("chk_index");
   DBUG_DUMP("buff",(byte*) buff,mi_getint(buff));
 
@@ -619,7 +621,7 @@ static int chk_index(MI_CHECK *param, MI_INFO *info, MI_KEYDEF *keyinfo,
     }
     if ((*keys)++ &&
 	(flag=ha_key_cmp(keyinfo->seg,info->lastkey,key,key_length,
-			 comp_flag, &not_used)) >=0)
+			 comp_flag, &diff_pos)) >=0)
     {
       DBUG_DUMP("old",(byte*) info->lastkey, info->lastkey_length);
       DBUG_DUMP("new",(byte*) key, key_length);
@@ -635,11 +637,11 @@ static int chk_index(MI_CHECK *param, MI_INFO *info, MI_KEYDEF *keyinfo,
     {
       if (*keys != 1L)				/* not first_key */
       {
-	uint diff;
-	ha_key_cmp(keyinfo->seg,info->lastkey,key,USE_WHOLE_KEY,
-		   SEARCH_FIND | SEARCH_NULL_ARE_NOT_EQUAL,
-		   &diff);
-	param->unique_count[diff-1]++;
+        if (param->stats_method == MI_STATS_METHOD_NULLS_NOT_EQUAL)
+          ha_key_cmp(keyinfo->seg,info->lastkey,key,USE_WHOLE_KEY,
+                     SEARCH_FIND | SEARCH_NULL_ARE_NOT_EQUAL,
+                     &diff_pos);
+	param->unique_count[diff_pos-1]++;
       }
     }
     (*key_checksum)+= mi_byte_checksum((byte*) key,
@@ -2013,7 +2015,7 @@ int mi_repair_by_sort(MI_CHECK *param, register MI_INFO *info,
   sort_param.sort_info=&sort_info;
   sort_param.fix_datafile= (my_bool) (! rep_quick);
   sort_param.master =1;
-
+  
   del=info->state->del;
   param->glob_crc=0;
   if (param->testflag & T_CALC_CHECKSUM)
@@ -3249,9 +3251,10 @@ static int sort_key_write(MI_SORT_PARAM *sort_param, const void *a)
     cmp=ha_key_cmp(sort_param->seg,sort_info->key_block->lastkey,
 		   (uchar*) a, USE_WHOLE_KEY,SEARCH_FIND | SEARCH_UPDATE,
 		   &diff_pos);
-    ha_key_cmp(sort_param->seg,sort_info->key_block->lastkey,
-               (uchar*) a, USE_WHOLE_KEY,SEARCH_FIND | SEARCH_NULL_ARE_NOT_EQUAL,
-               &diff_pos);
+    if (param->stats_method == MI_STATS_METHOD_NULLS_NOT_EQUAL)
+      ha_key_cmp(sort_param->seg,sort_info->key_block->lastkey,
+                 (uchar*) a, USE_WHOLE_KEY, 
+                 SEARCH_FIND | SEARCH_NULL_ARE_NOT_EQUAL, &diff_pos);
     sort_param->unique[diff_pos-1]++;
   }
   else
@@ -3989,9 +3992,10 @@ void update_auto_increment_key(MI_CHECK *param, MI_INFO *info,
     unique[0]= (#different values of {keypart1}) - 1
     unique[1]= (#different values of {keypart2,keypart1} tuple) - unique[0] - 1
     ...
-    Here we assume that NULL != NULL (see SEARCH_NULL_ARE_NOT_EQUAL). The
-    'unique' array is collected in one sequential scan through the entire
+    The 'unique' array is collected in one sequential scan through the entire
     index. This is done in two places: in chk_index() and in sort_key_write().
+    Statistics collection may consider NULLs as either equal or inequal (see
+    SEARCH_NULL_ARE_NOT_EQUAL, MI_STATS_METHOD_*).
 
     Output is an array:
     rec_per_key_part[k] = 

myisam/myisamchk.c

@@ -67,6 +67,7 @@ static const char *field_pack[]=
  "no zeros", "blob", "constant", "table-lockup",
  "always zero","varchar","unique-hash","?","?"};
 
+static const char *myisam_stats_method_str="nulls_inequal";
 
 static void get_options(int *argc,char * * *argv);
 static void print_version(void);
@@ -155,7 +156,7 @@ enum options_mc {
   OPT_READ_BUFFER_SIZE, OPT_WRITE_BUFFER_SIZE, OPT_SORT_BUFFER_SIZE,
   OPT_SORT_KEY_BLOCKS, OPT_DECODE_BITS, OPT_FT_MIN_WORD_LEN,
   OPT_FT_MAX_WORD_LEN, OPT_FT_STOPWORD_FILE,
-  OPT_MAX_RECORD_LENGTH, OPT_AUTO_CLOSE
+  OPT_MAX_RECORD_LENGTH, OPT_AUTO_CLOSE, OPT_STATS_METHOD
 };
 
 static struct my_option my_long_options[] =
@@ -336,6 +337,11 @@ static struct my_option my_long_options[] =
     "Use stopwords from this file instead of built-in list.",
     (gptr*) &ft_stopword_file, (gptr*) &ft_stopword_file, 0, GET_STR,
     REQUIRED_ARG, 0, 0, 0, 0, 0, 0},
+  {"stats_method", OPT_STATS_METHOD,
+   "Specifies how index statistics collection code should threat NULLs. "
+   "Possible values of name are \"nulls_inequal\" (default behavior for 4.1/5.0), and \"nulls_equal\" (emulate 4.0 behavior).",
+   (gptr*) &myisam_stats_method_str, (gptr*) &myisam_stats_method_str, 0,
+    GET_STR, REQUIRED_ARG, 0, 0, 0, 0, 0, 0},
   { 0, 0, 0, 0, 0, 0, GET_NO_ARG, NO_ARG, 0, 0, 0, 0, 0, 0}
 };
 
@@ -465,6 +471,12 @@ static void usage(void)
 
 #include <help_end.h>
 
+const char *myisam_stats_method_names[] = {"nulls_inequal", "nulls_equal",
+                                           NullS};
+TYPELIB myisam_stats_method_typelib= {
+  array_elements(myisam_stats_method_names) - 1, "",
+  myisam_stats_method_names, NULL};
+
 	 /* Read options */
 
 static my_bool
@@ -684,6 +696,19 @@ get_one_option(int optid,
     else
       check_param.testflag|= T_CALC_CHECKSUM;
     break;
+  case OPT_STATS_METHOD:
+  {
+    myisam_stats_method_str= argument;
+    int method;
+    if ((method=find_type(argument, &myisam_stats_method_typelib, 2)) <= 0)
+    {
+      fprintf(stderr, "Invalid value of stats_method: %s.\n", argument);
+      exit(1);
+    }
+    check_param.stats_method= test(method-1)? MI_STATS_METHOD_NULLS_EQUAL :
+                                              MI_STATS_METHOD_NULLS_NOT_EQUAL;
+    break;
+  }
 #ifdef DEBUG					/* Only useful if debugging */
   case OPT_START_CHECK_POS:
     check_param.start_check_pos= strtoull(argument, NULL, 0);

mysql-test/r/myisam.result

@@ -609,3 +609,64 @@ checksum table t2;
 Table	Checksum
 test.t2	984116287
 drop table t1, t2;
+show variables like 'myisam_stats_method';
+Variable_name	Value
+myisam_stats_method	nulls_inequal
+create table t1 (a int, key(a));
+insert into t1 values (0),(1),(2),(3),(4);
+insert into t1 select NULL from t1;
+analyze table t1;
+Table	Op	Msg_type	Msg_text
+test.t1	analyze	status	OK
+show index 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	A	10	NULL	NULL	YES	BTREE	
+insert into t1 values (11);
+delete from t1 where a=11;
+check table t1;
+Table	Op	Msg_type	Msg_text
+test.t1	check	status	OK
+show index 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	A	10	NULL	NULL	YES	BTREE	
+set myisam_stats_method=nulls_equal;
+show variables like 'myisam_stats_method';
+Variable_name	Value
+myisam_stats_method	nulls_equal
+insert into t1 values (11);
+delete from t1 where a=11;
+analyze table t1;
+Table	Op	Msg_type	Msg_text
+test.t1	analyze	status	OK
+show index 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	A	5	NULL	NULL	YES	BTREE	
+insert into t1 values (11);
+delete from t1 where a=11;
+check table t1;
+Table	Op	Msg_type	Msg_text
+test.t1	check	status	OK
+show index 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	A	5	NULL	NULL	YES	BTREE	
+set myisam_stats_method=DEFAULT;
+show variables like 'myisam_stats_method';
+Variable_name	Value
+myisam_stats_method	nulls_inequal
+insert into t1 values (11);
+delete from t1 where a=11;
+analyze table t1;
+Table	Op	Msg_type	Msg_text
+test.t1	analyze	status	OK
+show index 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	A	10	NULL	NULL	YES	BTREE	
+insert into t1 values (11);
+delete from t1 where a=11;
+check table t1;
+Table	Op	Msg_type	Msg_text
+test.t1	check	status	OK
+show index 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	A	10	NULL	NULL	YES	BTREE	
+drop table t1;

mysql-test/t/myisam.test

@@ -590,4 +590,51 @@ checksum table t1;
 checksum table t2;
 drop table t1, t2;
 
+# BUG#12232: New myisam_stats_method variable.
+show variables like 'myisam_stats_method';
+
+create table t1 (a int, key(a));
+insert into t1 values (0),(1),(2),(3),(4);
+insert into t1 select NULL from t1;
+
+# default: NULLs considered inequal
+analyze table t1; 
+show index from t1;
+insert into t1 values (11);
+delete from t1 where a=11;
+check table t1;
+show index from t1;
+
+# Set nulls to be equal:
+set myisam_stats_method=nulls_equal;
+show variables like 'myisam_stats_method';
+insert into t1 values (11);
+delete from t1 where a=11;
+
+analyze table t1; 
+show index from t1;
+
+insert into t1 values (11);
+delete from t1 where a=11;
+
+check table t1;
+show index from t1;
+
+# Set nulls back to be equal 
+set myisam_stats_method=DEFAULT;
+show variables like 'myisam_stats_method';
+insert into t1 values (11);
+delete from t1 where a=11;
+
+analyze table t1; 
+show index from t1;
+
+insert into t1 values (11);
+delete from t1 where a=11;
+
+check table t1;
+show index from t1;
+
+drop table t1;
+
 # End of 4.1 tests

sql/ha_myisam.cc

@@ -39,6 +39,12 @@ const char *myisam_recover_names[] =
 TYPELIB myisam_recover_typelib= {array_elements(myisam_recover_names)-1,"",
 				 myisam_recover_names, NULL};
 
+const char *myisam_stats_method_names[] = {"nulls_inequal", "nulls_equal",
+                                           NullS};
+TYPELIB myisam_stats_method_typelib= {
+  array_elements(myisam_stats_method_names) - 1, "",
+  myisam_stats_method_names, NULL};
+
 
 /*****************************************************************************
 ** MyISAM tables
@@ -278,6 +284,7 @@ int ha_myisam::check(THD* thd, HA_CHECK_OPT* check_opt)
   param.db_name    = table->table_cache_key;
   param.table_name = table->table_name;
   param.testflag = check_opt->flags | T_CHECK | T_SILENT;
+  param.stats_method= (enum_mi_stats_method)thd->variables.myisam_stats_method;
 
   if (!(table->db_stat & HA_READ_ONLY))
     param.testflag|= T_STATISTICS;
@@ -367,6 +374,7 @@ int ha_myisam::analyze(THD *thd, HA_CHECK_OPT* check_opt)
   param.testflag=(T_FAST | T_CHECK | T_SILENT | T_STATISTICS |
 		  T_DONT_CHECK_CHECKSUM);
   param.using_global_keycache = 1;
+  param.stats_method= (enum_mi_stats_method)thd->variables.myisam_stats_method;
 
   if (!(share->state.changed & STATE_NOT_ANALYZED))
     return HA_ADMIN_ALREADY_DONE;
@@ -920,6 +928,7 @@ int ha_myisam::enable_indexes(uint mode)
                       T_CREATE_MISSING_KEYS);
     param.myf_rw&= ~MY_WAIT_IF_FULL;
     param.sort_buffer_length=  thd->variables.myisam_sort_buff_size;
+    param.stats_method= (enum_mi_stats_method)thd->variables.myisam_stats_method;
     param.tmpdir=&mysql_tmpdir_list;
     if ((error= (repair(thd,param,0) != HA_ADMIN_OK)) && param.retry_repair)
     {

sql/handler.h

@@ -522,6 +522,7 @@ class handler :public Sql_alloc
 extern struct show_table_type_st sys_table_types[];
 extern const char *ha_row_type[];
 extern TYPELIB tx_isolation_typelib;
+extern TYPELIB myisam_stats_method_typelib;
 
 	/* Wrapper functions */
 #define ha_commit_stmt(thd) (ha_commit_trans((thd), &((thd)->transaction.stmt)))

sql/mysqld.cc

@@ -369,6 +369,7 @@ char *mysqld_unix_port, *opt_mysql_tmpdir;
 char *my_bind_addr_str;
 const char **errmesg;			/* Error messages */
 const char *myisam_recover_options_str="OFF";
+const char *myisam_stats_method_str="nulls_inequal";
 const char *sql_mode_str="OFF";
 /* name of reference on left espression in rewritten IN subquery */
 const char *in_left_expr_name= "<left expr>";
@@ -4169,6 +4170,7 @@ enum options_mysqld
   OPT_MAX_ERROR_COUNT, OPT_MYISAM_DATA_POINTER_SIZE,
   OPT_MYISAM_BLOCK_SIZE, OPT_MYISAM_MAX_EXTRA_SORT_FILE_SIZE,
   OPT_MYISAM_MAX_SORT_FILE_SIZE, OPT_MYISAM_SORT_BUFFER_SIZE,
+  OPT_MYISAM_STATS_METHOD,
   OPT_NET_BUFFER_LENGTH, OPT_NET_RETRY_COUNT,
   OPT_NET_READ_TIMEOUT, OPT_NET_WRITE_TIMEOUT,
   OPT_OPEN_FILES_LIMIT,
@@ -5208,6 +5210,11 @@ The minimum value for this variable is 4096.",
    (gptr*) &global_system_variables.myisam_sort_buff_size,
    (gptr*) &max_system_variables.myisam_sort_buff_size, 0,
    GET_ULONG, REQUIRED_ARG, 8192*1024, 4, ~0L, 0, 1, 0},
+  {"myisam_stats_method", OPT_MYISAM_STATS_METHOD,
+   "Specifies how MyISAM index statistics collection code should threat NULLs. "
+   "Possible values of name are \"nulls_inequal\" (default behavior for 4.1/5.0), and \"nulls_equal\" (emulate 4.0 behavior).",
+   (gptr*) &myisam_stats_method_str, (gptr*) &myisam_stats_method_str, 0,
+    GET_STR, REQUIRED_ARG, 0, 0, 0, 0, 0, 0},
   {"net_buffer_length", OPT_NET_BUFFER_LENGTH,
    "Buffer length for TCP/IP and socket communication.",
    (gptr*) &global_system_variables.net_buffer_length,
@@ -5759,6 +5766,7 @@ static void mysql_init_variables(void)
   query_id= thread_id= 1L;
   strmov(server_version, MYSQL_SERVER_VERSION);
   myisam_recover_options_str= sql_mode_str= "OFF";
+  myisam_stats_method_str= "nulls_inequal";
   my_bind_addr = htonl(INADDR_ANY);
   threads.empty();
   thread_cache.empty();
@@ -5807,6 +5815,12 @@ static void mysql_init_variables(void)
   global_system_variables.max_join_size= (ulonglong) HA_POS_ERROR;
   max_system_variables.max_join_size=   (ulonglong) HA_POS_ERROR;
   global_system_variables.old_passwords= 0;
+  
+  /*
+    Default behavior for 4.1 and 5.0 is to treat NULL values as inequal
+    when collecting index statistics for MyISAM tables.
+  */
+  global_system_variables.myisam_stats_method= MI_STATS_METHOD_NULLS_NOT_EQUAL;
 
   /* Variables that depends on compile options */
 #ifndef DBUG_OFF
@@ -6388,6 +6402,20 @@ get_one_option(int optid, const struct my_option *opt __attribute__((unused)),
     ha_open_options|=HA_OPEN_ABORT_IF_CRASHED;
     break;
   }
+  case OPT_MYISAM_STATS_METHOD:
+  {
+    myisam_stats_method_str= argument;
+    int method;
+    if ((method=find_type(argument, &myisam_stats_method_typelib, 2)) <= 0)
+    {
+      fprintf(stderr, "Invalid value of myisam_stats_method: %s.\n", argument);
+      exit(1);
+    }
+    global_system_variables.myisam_stats_method= 
+      test(method-1)? MI_STATS_METHOD_NULLS_EQUAL : 
+                      MI_STATS_METHOD_NULLS_NOT_EQUAL;
+    break;
+  }
   case OPT_SQL_MODE:
   {
     sql_mode_str= argument;

sql/set_var.cc

@@ -256,6 +256,12 @@ sys_var_thd_ulonglong	sys_myisam_max_extra_sort_file_size("myisam_max_extra_sort
 sys_var_thd_ulonglong	sys_myisam_max_sort_file_size("myisam_max_sort_file_size", &SV::myisam_max_sort_file_size, fix_myisam_max_sort_file_size, 1);
 sys_var_thd_ulong       sys_myisam_repair_threads("myisam_repair_threads", &SV::myisam_repair_threads);
 sys_var_thd_ulong	sys_myisam_sort_buffer_size("myisam_sort_buffer_size", &SV::myisam_sort_buff_size);
+
+sys_var_thd_enum        sys_myisam_stats_method("myisam_stats_method",
+                                                &SV::myisam_stats_method,
+                                                &myisam_stats_method_typelib,
+                                                NULL);
+
 sys_var_thd_ulong	sys_net_buffer_length("net_buffer_length",
 					      &SV::net_buffer_length);
 sys_var_thd_ulong	sys_net_read_timeout("net_read_timeout",
@@ -574,6 +580,7 @@ sys_var *sys_variables[]=
   &sys_myisam_max_sort_file_size,
   &sys_myisam_repair_threads,
   &sys_myisam_sort_buffer_size,
+  &sys_myisam_stats_method,
   &sys_net_buffer_length,
   &sys_net_read_timeout,
   &sys_net_retry_count,
@@ -810,6 +817,9 @@ struct show_var_st init_vars[]= {
   {sys_myisam_repair_threads.name, (char*) &sys_myisam_repair_threads,
    SHOW_SYS},
   {sys_myisam_sort_buffer_size.name, (char*) &sys_myisam_sort_buffer_size, SHOW_SYS},
+  
+  {sys_myisam_stats_method.name, (char*) &sys_myisam_stats_method, SHOW_SYS},
+  
 #ifdef __NT__
   {"named_pipe",	      (char*) &opt_enable_named_pipe,       SHOW_MY_BOOL},
 #endif

sql/sql_class.h

@@ -363,6 +363,7 @@ struct system_variables
   ulong max_insert_delayed_threads;
   ulong myisam_repair_threads;
   ulong myisam_sort_buff_size;
+  ulong myisam_stats_method;
   ulong net_buffer_length;
   ulong net_interactive_timeout;
   ulong net_read_timeout;