Addressed all remaining issues from the review of the patch

that introduced engine independent persistent statistics.
In particular:
- added an enumeration type for possible values of the system
  variable use_stat_tables
- renamed KEY::real_rec_per_key to KEY::actual_rec_per_key
- optimized the collection of statistical data for any primary
  key defined only on one column.
  

16 files changed, 112 insertions, 45 deletions

diff --git a/sql/multi_range_read.cc b/sql/multi_range_read.cc
index 1361d5822c9..b9f49a83b4b 100644
--- a/sql/multi_range_read.cc
+++ b/sql/multi_range_read.cc
@@ -1201,7 +1201,7 @@ bool DsMrr_impl::setup_buffer_sharing(uint key_size_in_keybuf,
   uint parts= my_count_bits(key_tuple_map);
   ulong rpc;
   ulonglong rowids_size= rowid_buf_elem_size;
-  if ((rpc= key_info->real_rec_per_key(parts - 1)))
+  if ((rpc= key_info->actual_rec_per_key(parts - 1)))
     rowids_size= rowid_buf_elem_size * rpc;
 
   double fraction_for_rowids=
diff --git a/sql/opt_range.cc b/sql/opt_range.cc
index ab4e696c21c..6f13816d8ea 100644
--- a/sql/opt_range.cc
+++ b/sql/opt_range.cc
@@ -5502,8 +5502,8 @@ ha_rows records_in_index_intersect_extension(PARTIAL_INDEX_INTERSECT_INFO *curr,
     ha_rows ext_records= ext_index_scan->records;
     if (i < used_key_parts)
     {
-      ulong f1= key_info->real_rec_per_key(i-1);
-      ulong f2= key_info->real_rec_per_key(i);
+      ulong f1= key_info->actual_rec_per_key(i-1);
+      ulong f2= key_info->actual_rec_per_key(i);
       ext_records= (ha_rows) ((double) ext_records / f2 * f1);
     }
     if (ext_records < table_cardinality)
@@ -12642,7 +12642,7 @@ void cost_group_min_max(TABLE* table, KEY *index_info, uint used_key_parts,
   num_blocks= (uint)(table_records / keys_per_block) + 1;
 
   /* Compute the number of keys in a group. */
-  keys_per_group= index_info->real_rec_per_key(group_key_parts - 1);
+  keys_per_group= index_info->actual_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= (uint)(table_records / 10) + 1;
@@ -12662,7 +12662,7 @@ void cost_group_min_max(TABLE* table, KEY *index_info, uint used_key_parts,
       Compute the probability that two ends of a subgroup are inside
       different blocks.
     */
-    keys_per_subgroup= index_info->real_rec_per_key(used_key_parts - 1);
+    keys_per_subgroup= index_info->actual_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
diff --git a/sql/sql_admin.cc b/sql/sql_admin.cc
index f2124dd3bb8..e6bbef482a7 100644
--- a/sql/sql_admin.cc
+++ b/sql/sql_admin.cc
@@ -29,6 +29,7 @@
 #include "sql_parse.h"                       // check_table_access
 #include "strfunc.h"
 #include "sql_admin.h"
+#include "sql_statistics.h"
 
 /* Prepare, run and cleanup for mysql_recreate_table() */
 
@@ -718,7 +719,7 @@ static bool mysql_admin_table(THD* thd, TABLE_LIST* tables,
     if (compl_result_code == HA_ADMIN_OK &&
         operator_func == &handler::ha_analyze && 
         table->table->s->table_category == TABLE_CATEGORY_USER &&
-        (thd->variables.use_stat_tables > 0 ||
+        (get_use_stat_tables_mode(thd) > NEVER ||
          lex->with_persistent_for_clause)) 
     {
       if (!(compl_result_code=
diff --git a/sql/sql_base.cc b/sql/sql_base.cc
index 9db8ac1c732..a582dde4277 100644
--- a/sql/sql_base.cc
+++ b/sql/sql_base.cc
@@ -49,6 +49,7 @@
 #include "sql_trigger.h"
 #include "transaction.h"
 #include "sql_prepare.h"
+#include "sql_statistics.h"
 #include <m_ctype.h>
 #include <my_dir.h>
 #include <hash.h>
@@ -3142,7 +3143,7 @@ retry_share:
     while (table_cache_count > table_cache_size && unused_tables)
       free_cache_entry(unused_tables);
 
-    if (thd->variables.use_stat_tables > 0)
+    if (get_use_stat_tables_mode(thd) > NEVER)
     {
       if (share->table_category != TABLE_CATEGORY_SYSTEM)
       {
@@ -4634,7 +4635,7 @@ open_and_process_table(THD *thd, LEX *lex, TABLE_LIST *tables,
     goto end;
   }
 
-  if (thd->variables.use_stat_tables > 0 && tables->table)
+  if (get_use_stat_tables_mode(thd) > NEVER && tables->table)
   {
     TABLE_SHARE *table_share= tables->table->s;
     if (table_share && table_share->table_category != TABLE_CATEGORY_SYSTEM)
diff --git a/sql/sql_base.h b/sql/sql_base.h
index c4cd7f467a0..aa2ba9e5680 100644
--- a/sql/sql_base.h
+++ b/sql/sql_base.h
@@ -310,22 +310,6 @@ int dynamic_column_error_message(enum_dyncol_func_result rc);
 /* open_and_lock_tables with optional derived handling */
 int open_and_lock_tables_derived(THD *thd, TABLE_LIST *tables, bool derived);
 
-int read_statistics_for_tables_if_needed(THD *thd, TABLE_LIST *tables);
-int collect_statistics_for_table(THD *thd, TABLE *table);
-int alloc_statistics_for_table_share(THD* thd, TABLE_SHARE *share,
-                                     bool is_safe);
-int alloc_statistics_for_table(THD *thd, TABLE *table);
-int update_statistics_for_table(THD *thd, TABLE *table);
-int delete_statistics_for_table(THD *thd, LEX_STRING *db, LEX_STRING *tab);
-int delete_statistics_for_column(THD *thd, TABLE *tab, Field *col);
-int delete_statistics_for_index(THD *thd, TABLE *tab, KEY *key_info,
-                                bool ext_prefixes_only);
-int rename_table_in_stat_tables(THD *thd, LEX_STRING *db, LEX_STRING *tab,
-                                LEX_STRING *new_db, LEX_STRING *new_tab);
-int rename_column_in_stat_tables(THD *thd, TABLE *tab, Field *col,
-                                  const char *new_name);
-void set_statistics_for_table(THD *thd, TABLE *table);
-
 extern "C" int simple_raw_key_cmp(void* arg, const void* key1,
                                   const void* key2);
 extern "C" int count_distinct_walk(void *elem, element_count count, void *arg);
diff --git a/sql/sql_db.cc b/sql/sql_db.cc
index 086445948bd..f4e9ccfc5e6 100644
--- a/sql/sql_db.cc
+++ b/sql/sql_db.cc
@@ -36,6 +36,7 @@
 #include "sp.h"
 #include "events.h"
 #include "sql_handler.h"
+#include "sql_statistics.h"
 #include <my_dir.h>
 #include <m_ctype.h>
 #include "log.h"
diff --git a/sql/sql_delete.cc b/sql/sql_delete.cc
index 2dd090578a7..4e35072d508 100644
--- a/sql/sql_delete.cc
+++ b/sql/sql_delete.cc
@@ -35,6 +35,7 @@
 #include "sql_select.h"
 #include "sp_head.h"
 #include "sql_trigger.h"
+#include "sql_statistics.h"
 #include "transaction.h"
 #include "records.h"                            // init_read_record,
 #include "sql_derived.h"                        // mysql_handle_list_of_derived
diff --git a/sql/sql_join_cache.cc b/sql/sql_join_cache.cc
index 225bb413195..ac96746c389 100644
--- a/sql/sql_join_cache.cc
+++ b/sql/sql_join_cache.cc
@@ -3812,7 +3812,8 @@ uint JOIN_TAB_SCAN_MRR::aux_buffer_incr(ulong recno)
   uint incr= 0;
   TABLE_REF *ref= &join_tab->ref;
   TABLE *tab= join_tab->table;
-  uint rec_per_key= tab->key_info[ref->key].real_rec_per_key(ref->key_parts-1);
+  uint rec_per_key=
+    tab->key_info[ref->key].actual_rec_per_key(ref->key_parts-1);
   set_if_bigger(rec_per_key, 1);
   if (recno == 1)
     incr=  ref->key_length + tab->file->ref_length;
diff --git a/sql/sql_rename.cc b/sql/sql_rename.cc
index d27989c3b01..c91623cee6e 100644
--- a/sql/sql_rename.cc
+++ b/sql/sql_rename.cc
@@ -28,6 +28,7 @@
 #include "lock.h"       // MYSQL_OPEN_SKIP_TEMPORARY
 #include "sql_base.h"   // tdc_remove_table, lock_table_names,
 #include "sql_handler.h"                        // mysql_ha_rm_tables
+#include "sql_statistics.h" 
 #include "datadict.h"
 
 static TABLE_LIST *rename_tables(THD *thd, TABLE_LIST *table_list,
diff --git a/sql/sql_select.cc b/sql/sql_select.cc
index fcb650b4757..85908623324 100644
--- a/sql/sql_select.cc
+++ b/sql/sql_select.cc
@@ -5416,7 +5416,7 @@ best_access_path(JOIN      *join,
             else
             {
               uint key_parts= table->actual_n_key_parts(keyinfo);
-              if (!(records= keyinfo->real_rec_per_key(key_parts-1)))
+              if (!(records= keyinfo->actual_rec_per_key(key_parts-1)))
               {                                   /* Prefer longer keys */
                 records=
                   ((double) s->records / (double) rec *
@@ -5516,7 +5516,7 @@ best_access_path(JOIN      *join,
             else
             {
               /* Check if we have statistic about the distribution */
-              if ((records= keyinfo->real_rec_per_key(max_key_part-1)))
+              if ((records= keyinfo->actual_rec_per_key(max_key_part-1)))
               {
                 /* 
                   Fix for the case where the index statistics is too
@@ -22974,7 +22974,7 @@ test_if_cheaper_ordering(const JOIN_TAB *tab, ORDER *order, TABLE *table,
           if (used_key_parts > used_index_parts)
             used_pk_parts= used_key_parts-used_index_parts;
           rec_per_key= used_key_parts ?
-	               keyinfo->real_rec_per_key(used_key_parts-1) : 1;
+	               keyinfo->actual_rec_per_key(used_key_parts-1) : 1;
           /* Take into account the selectivity of the used pk prefix */
           if (used_pk_parts)
 	  {
@@ -22989,8 +22989,8 @@ test_if_cheaper_ordering(const JOIN_TAB *tab, ORDER *order, TABLE *table,
               rec_per_key= 1;                 
             if (rec_per_key > 1)
 	    {
-              rec_per_key*= pkinfo->real_rec_per_key(used_pk_parts-1);
-              rec_per_key/= pkinfo->real_rec_per_key(0);
+              rec_per_key*= pkinfo->actual_rec_per_key(used_pk_parts-1);
+              rec_per_key/= pkinfo->actual_rec_per_key(0);
               /* 
                 The value of rec_per_key for the extended key has
                 to be adjusted accordingly if some components of
@@ -23004,9 +23004,9 @@ test_if_cheaper_ordering(const JOIN_TAB *tab, ORDER *order, TABLE *table,
                     We presume here that for any index rec_per_key[i] != 0
                     if rec_per_key[0] != 0.
 	          */
-                  DBUG_ASSERT(pkinfo->real_rec_per_key(i));
-                  rec_per_key*= pkinfo->real_rec_per_key(i-1);
-                  rec_per_key/= pkinfo->real_rec_per_key(i);
+                  DBUG_ASSERT(pkinfo->actual_rec_per_key(i));
+                  rec_per_key*= pkinfo->actual_rec_per_key(i-1);
+                  rec_per_key/= pkinfo->actual_rec_per_key(i);
                 }
 	      }
             }    
@@ -23051,7 +23051,7 @@ test_if_cheaper_ordering(const JOIN_TAB *tab, ORDER *order, TABLE *table,
           select_limit= (ha_rows) (select_limit *
                                    (double) table_records /
                                     table->quick_condition_rows);
-        rec_per_key= keyinfo->real_rec_per_key(keyinfo->key_parts-1);
+        rec_per_key= keyinfo->actual_rec_per_key(keyinfo->key_parts-1);
         set_if_bigger(rec_per_key, 1);
         /*
           Here we take into account the fact that rows are
diff --git a/sql/sql_show.cc b/sql/sql_show.cc
index 19a25c28942..861509b30de 100644
--- a/sql/sql_show.cc
+++ b/sql/sql_show.cc
@@ -45,6 +45,7 @@
 #include "set_var.h"
 #include "sql_trigger.h"
 #include "sql_derived.h"
+#include "sql_statistics.h"
 #include "sql_connect.h"
 #include "authors.h"
 #include "contributors.h"
@@ -5765,7 +5766,7 @@ static int get_schema_stat_record(THD *thd, TABLE_LIST *tables,
           if (key->rec_per_key[j])
           {
             ha_rows records=((double) show_table->stat_records() /
-                             key->real_rec_per_key(j));
+                             key->actual_rec_per_key(j));
             table->field[9]->store((longlong) records, TRUE);
             table->field[9]->set_notnull();
           }
diff --git a/sql/sql_statistics.cc b/sql/sql_statistics.cc
index 93d3a1c581f..1cb4398caa2 100644
--- a/sql/sql_statistics.cc
+++ b/sql/sql_statistics.cc
@@ -206,6 +206,8 @@ private:
   Count_distinct_field *count_distinct; /* The container for distinct 
                                            column values */
 
+  bool is_single_pk_col; /* TRUE <-> the only column of the primary key */ 
+
 public:
 
   inline void init(THD *thd, Field * table_field);
@@ -1399,6 +1401,8 @@ private:
     
 public:
 
+  bool is_single_comp_pk;
+
   Index_prefix_calc(TABLE *table, KEY *key_info)
     : index_table(table), index_info(key_info)
   {
@@ -1407,6 +1411,16 @@ public:
     uint key_parts= table->actual_n_key_parts(key_info);
     empty= TRUE;
     prefixes= 0;
+
+    is_single_comp_pk= FALSE;
+    uint pk= table->s->primary_key;
+    if (table->key_info - key_info == pk && table->key_info[pk].key_parts == 1)
+    {
+      prefixes= 1;
+      is_single_comp_pk= TRUE;
+      return;
+    }
+        
     if ((calc_state=
          (Prefix_calc_state *) sql_alloc(sizeof(Prefix_calc_state)*key_parts)))
     {
@@ -1430,6 +1444,7 @@ public:
     }
   }
 
+
   /** 
     @breif
     Change the elements of calc_state after reading the next index entry
@@ -1487,6 +1502,13 @@ public:
   {
     uint i;
     Prefix_calc_state *state;
+
+    if (is_single_comp_pk)
+    {
+      index_info->collected_stats->set_avg_frequency(0, 1.0);
+      return;
+    }
+
     for (i= 0, state= calc_state; i < prefixes; i++, state++)
     {
       if (i < prefixes)
@@ -1658,7 +1680,7 @@ void create_min_max_stistical_fields_for_table_share(THD *thd,
 int alloc_statistics_for_table(THD* thd, TABLE *table)
 { 
   Field **field_ptr;
-  uint cnt= 0;
+  uint fields;
 
   DBUG_ENTER("alloc_statistics_for_table");
 
@@ -1666,10 +1688,11 @@ int alloc_statistics_for_table(THD* thd, TABLE *table)
     (Table_statistics *) alloc_root(&table->mem_root,
                                     sizeof(Table_statistics));
 
-  for (field_ptr= table->field; *field_ptr; field_ptr++, cnt++) ; 
+  fields= table->s->fields ; 
   Column_statistics_collected *column_stats=
     (Column_statistics_collected *) alloc_root(&table->mem_root,
-                                    sizeof(Column_statistics_collected) * cnt);
+                                    sizeof(Column_statistics_collected) *
+                                    fields);
 
   uint keys= table->s->keys;
   Index_statistics *index_stats=
@@ -1688,7 +1711,7 @@ int alloc_statistics_for_table(THD* thd, TABLE *table)
   table_stats->index_stats= index_stats;
   table_stats->idx_avg_frequency= idx_avg_frequency;
   
-  memset(column_stats, 0, sizeof(Column_statistics) * cnt);
+  memset(column_stats, 0, sizeof(Column_statistics) * fields);
 
   for (field_ptr= table->field; *field_ptr; field_ptr++, column_stats++)
     (*field_ptr)->collected_stats= column_stats;
@@ -1838,13 +1861,23 @@ inline
 void Column_statistics_collected::init(THD *thd, Field *table_field)
 {
   uint max_heap_table_size= thd->variables.max_heap_table_size;
+  TABLE *table= table_field->table;
+  uint pk= table->s->primary_key;
+  
+  is_single_pk_col= FALSE;
 
+  if (pk != MAX_KEY && table->key_info[pk].key_parts == 1 &&
+      table->key_info[pk].key_part[0].fieldnr == table_field->field_index + 1)
+    is_single_pk_col= TRUE;  
+  
   column= table_field;
 
   set_all_nulls();
 
   nulls= 0;
   column_total_length= 0;
+  if (is_single_pk_col)
+    count_distinct= NULL;
   if (table_field->flags & BLOB_FLAG)
     count_distinct= NULL;
   else
@@ -1923,6 +1956,12 @@ void Column_statistics_collected::finish(ha_rows rows)
     delete count_distinct;
     count_distinct= NULL;
   }
+  else if (is_single_pk_col)
+  {
+    val= 1.0;
+    set_avg_frequency(val); 
+    set_not_null(COLUMN_STAT_AVG_FREQUENCY);
+  } 
 }
 
 
@@ -1986,6 +2025,12 @@ int collect_statistics_for_index(THD *thd, TABLE *table, uint index)
   DEBUG_SYNC(table->in_use, "statistics_collection_start1");
   DEBUG_SYNC(table->in_use, "statistics_collection_start2");
 
+  if (index_prefix_calc.is_single_comp_pk)
+  {
+    index_prefix_calc.get_avg_frequency();
+    DBUG_RETURN(rc);
+  }
+
   table->key_read= 1;
   table->file->extra(HA_EXTRA_KEYREAD);
 
@@ -2078,7 +2123,7 @@ int collect_statistics_for_table(THD *thd, TABLE *table)
 
   table->collected_stats->cardinality_is_null= TRUE;
   table->collected_stats->cardinality= 0;
- 
+
   for (field_ptr= table->field; *field_ptr; field_ptr++)
   {
     table_field= *field_ptr;   
@@ -2949,9 +2994,9 @@ int rename_column_in_stat_tables(THD *thd, TABLE *tab, Field *col,
 
 void set_statistics_for_table(THD *thd, TABLE *table)
 {
-  uint use_stat_table_mode= thd->variables.use_stat_tables;
+  Use_stat_tables_mode use_stat_table_mode= get_use_stat_tables_mode(thd);
   table->used_stat_records= 
-    (use_stat_table_mode <= 1 ||
+    (use_stat_table_mode <= COMPLEMENTARY ||
      !table->s->stats_is_read || !table->s->read_stats ||
      table->s->read_stats->cardinality_is_null) ?
     table->file->stats.records : table->s->read_stats->cardinality;
@@ -2960,7 +3005,8 @@ void set_statistics_for_table(THD *thd, TABLE *table)
        key_info < key_info_end; key_info++)
   {
     key_info->is_statistics_from_stat_tables=
-      (use_stat_table_mode > 1  && table->s->stats_is_read &&
+      (use_stat_table_mode > COMPLEMENTARY &&
+       table->s->stats_is_read &&
        key_info->read_stats &&
        key_info->read_stats->avg_frequency_is_inited() &&
        key_info->read_stats->get_avg_frequency(0) > 0.5);
diff --git a/sql/sql_statistics.h b/sql/sql_statistics.h
index 14a16170c3b..e24e4beae74 100644
--- a/sql/sql_statistics.h
+++ b/sql/sql_statistics.h
@@ -25,6 +25,14 @@
   generated automatically by the table definitions.   
 */
 
+typedef
+enum enum_use_stat_tables_mode
+{
+  NEVER,
+  COMPLEMENTARY,
+  PEFERABLY,
+} Use_stat_tables_mode;
+
 enum enum_stat_tables
 {
   TABLE_STAT,
@@ -60,6 +68,27 @@ enum enum_index_stat_col
   INDEX_STAT_AVG_FREQUENCY
 };
 
+inline
+Use_stat_tables_mode get_use_stat_tables_mode(THD *thd)
+{ 
+  return (Use_stat_tables_mode) (thd->variables.use_stat_tables);
+}
+
+int read_statistics_for_tables_if_needed(THD *thd, TABLE_LIST *tables);
+int collect_statistics_for_table(THD *thd, TABLE *table);
+int alloc_statistics_for_table_share(THD* thd, TABLE_SHARE *share,
+                                     bool is_safe);
+int alloc_statistics_for_table(THD *thd, TABLE *table);
+int update_statistics_for_table(THD *thd, TABLE *table);
+int delete_statistics_for_table(THD *thd, LEX_STRING *db, LEX_STRING *tab);
+int delete_statistics_for_column(THD *thd, TABLE *tab, Field *col);
+int delete_statistics_for_index(THD *thd, TABLE *tab, KEY *key_info,
+                                bool ext_prefixes_only);
+int rename_table_in_stat_tables(THD *thd, LEX_STRING *db, LEX_STRING *tab,
+                                LEX_STRING *new_db, LEX_STRING *new_tab);
+int rename_column_in_stat_tables(THD *thd, TABLE *tab, Field *col,
+                                  const char *new_name);
+void set_statistics_for_table(THD *thd, TABLE *table);
 
 class Columns_statistics;
 class Index_statistics;
diff --git a/sql/sql_table.cc b/sql/sql_table.cc
index b63c1859582..5bd2b105a92 100644
--- a/sql/sql_table.cc
+++ b/sql/sql_table.cc
@@ -43,6 +43,7 @@
 #include "discover.h"                  // readfrm
 #include "my_pthread.h"                // pthread_mutex_t
 #include "log_event.h"                 // Query_log_event
+#include "sql_statistics.h"
 #include <hash.h>
 #include <myisam.h>
 #include <my_dir.h>
diff --git a/sql/structs.h b/sql/structs.h
index 13bb0574b24..a3a54c524e6 100644
--- a/sql/structs.h
+++ b/sql/structs.h
@@ -142,7 +142,7 @@ typedef struct st_key {
   engine_option_value *option_list;
   ha_index_option_struct *option_struct;                  /* structure with parsed options */
 
-  double real_rec_per_key(uint i);
+  double actual_rec_per_key(uint i);
 
 } KEY;
 
diff --git a/sql/table.cc b/sql/table.cc
index 11a8c553566..3ce0074a86f 100644
--- a/sql/table.cc
+++ b/sql/table.cc
@@ -6783,7 +6783,7 @@ uint TABLE_SHARE::actual_n_key_parts(THD *thd)
 }  
 
 
-double KEY::real_rec_per_key(uint i)
+double KEY::actual_rec_per_key(uint i)
 { 
   if (rec_per_key == 0)
     return 0;