merge to mysql-5.1-bugteam

11 files changed, 163 insertions, 26 deletions

diff --git a/sql/ha_partition.cc b/sql/ha_partition.cc
index df5badccb1e..451631ff373 100644
--- a/sql/ha_partition.cc
+++ b/sql/ha_partition.cc
@@ -1280,10 +1280,10 @@ void ha_partition::cleanup_new_partition(uint part_count)
     m_file= m_added_file;
     m_added_file= NULL;
 
+    external_lock(ha_thd(), F_UNLCK);
     /* delete_table also needed, a bit more complex */
     close();
 
-    m_added_file= m_file;
     m_file= save_m_file;
   }
   DBUG_VOID_RETURN;
@@ -5011,8 +5011,9 @@ int ha_partition::info(uint flag)
       If the handler doesn't support statistics, it should set all of the
       above to 0.
 
-      We will allow the first handler to set the rec_per_key and use
-      this as an estimate on the total table.
+      We first scans through all partitions to get the one holding most rows.
+      We will then allow the handler with the most rows to set
+      the rec_per_key and use this as an estimate on the total table.
 
       max_data_file_length:     Maximum data file length
       We ignore it, is only used in
@@ -5024,14 +5025,33 @@ int ha_partition::info(uint flag)
       ref_length:               We set this to the value calculated
       and stored in local object
       create_time:              Creation time of table
-      Set by first handler
 
-      So we calculate these constants by using the variables on the first
-      handler.
+      So we calculate these constants by using the variables from the
+      handler with most rows.
     */
-    handler *file;
+    handler *file, **file_array;
+    ulonglong max_records= 0;
+    uint32 i= 0;
+    uint32 handler_instance= 0;
+
+    file_array= m_file;
+    do
+    {
+      file= *file_array;
+      /* Get variables if not already done */
+      if (!(flag & HA_STATUS_VARIABLE) ||
+          !bitmap_is_set(&(m_part_info->used_partitions),
+                         (file_array - m_file)))
+        file->info(HA_STATUS_VARIABLE);
+      if (file->stats.records > max_records)
+      {
+        max_records= file->stats.records;
+        handler_instance= i;
+      }
+      i++;
+    } while (*(++file_array));
 
-    file= m_file[0];
+    file= m_file[handler_instance];
     file->info(HA_STATUS_CONST);
     stats.create_time= file->stats.create_time;
     ref_length= m_ref_length;
diff --git a/sql/log_event.cc b/sql/log_event.cc
index d7921ad3c27..ae7c4335f59 100644
--- a/sql/log_event.cc
+++ b/sql/log_event.cc
@@ -1852,6 +1852,7 @@ Rows_log_event::print_verbose_one_row(IO_CACHE *file, table_def *td,
 {
   const uchar *value0= value;
   const uchar *null_bits= value;
+  uint null_bit_index= 0;
   char typestr[64]= "";
   
   value+= (m_width + 7) / 8;
@@ -1860,7 +1861,8 @@ Rows_log_event::print_verbose_one_row(IO_CACHE *file, table_def *td,
   
   for (size_t i= 0; i < td->size(); i ++)
   {
-    int is_null= (null_bits[i / 8] >> (i % 8))  & 0x01;
+    int is_null= (null_bits[null_bit_index / 8] 
+                  >> (null_bit_index % 8))  & 0x01;
 
     if (bitmap_is_set(cols_bitmap, i) == 0)
       continue;
@@ -1897,6 +1899,8 @@ Rows_log_event::print_verbose_one_row(IO_CACHE *file, table_def *td,
     }
     
     my_b_printf(file, "\n");
+    
+    null_bit_index++;
   }
   return value - value0;
 }
diff --git a/sql/mysql_priv.h b/sql/mysql_priv.h
index 6fde16d3049..c04b1f5ae38 100644
--- a/sql/mysql_priv.h
+++ b/sql/mysql_priv.h
@@ -2433,6 +2433,7 @@ inline void setup_table_map(TABLE *table, TABLE_LIST *table_list, uint tablenr)
   table->tablenr= tablenr;
   table->map= (table_map) 1 << tablenr;
   table->force_index= table_list->force_index;
+  table->force_index_order= table->force_index_group= 0;
   table->covering_keys= table->s->keys_for_keyread;
   table->merge_keys.clear_all();
 }
diff --git a/sql/mysql_priv.h.pp b/sql/mysql_priv.h.pp
index 8bb31f64587..d874a2591d1 100644
--- a/sql/mysql_priv.h.pp
+++ b/sql/mysql_priv.h.pp
@@ -773,7 +773,6 @@ extern int wild_compare(const char *str,const char *wildstr,
 extern WF_PACK *wf_comp(char * str);
 extern int wf_test(struct wild_file_pack *wf_pack,const char *name);
 extern void wf_end(struct wild_file_pack *buffer);
-extern size_t strip_sp(char * str);
 extern my_bool array_append_string_unique(const char *str,
                                           const char **array, size_t size);
 extern void get_date(char * to,int timeflag,time_t use_time);
diff --git a/sql/opt_range.cc b/sql/opt_range.cc
index 05575e2744b..119f90bc97a 100644
--- a/sql/opt_range.cc
+++ b/sql/opt_range.cc
@@ -5876,6 +5876,7 @@ get_mm_leaf(RANGE_OPT_PARAM *param, COND *conf_func, Field *field,
       if (type == Item_func::LT_FUNC && (value->val_int() > 0))
         type = Item_func::LE_FUNC;
       else if (type == Item_func::GT_FUNC &&
+               (field->type() != FIELD_TYPE_BIT) &&
                !((Field_num*)field)->unsigned_flag &&
                !((Item_int*)value)->unsigned_flag &&
                (value->val_int() < 0))
@@ -5913,7 +5914,9 @@ get_mm_leaf(RANGE_OPT_PARAM *param, COND *conf_func, Field *field,
    */
   if (field->result_type() == INT_RESULT &&
       value->result_type() == INT_RESULT &&
-      ((Field_num*)field)->unsigned_flag && !((Item_int*)value)->unsigned_flag)
+      ((field->type() == FIELD_TYPE_BIT || 
+       ((Field_num *) field)->unsigned_flag) && 
+       !((Item_int*) value)->unsigned_flag))
   {
     longlong item_val= value->val_int();
     if (item_val < 0)
@@ -6509,6 +6512,63 @@ get_range(SEL_ARG **e1,SEL_ARG **e2,SEL_ARG *root1)
 }
 
 
+/**
+   Combine two range expression under a common OR. On a logical level, the
+   transformation is key_or( expr1, expr2 ) => expr1 OR expr2.
+
+   Both expressions are assumed to be in the SEL_ARG format. In a logic sense,
+   theformat is reminiscent of DNF, since an expression such as the following
+
+   ( 1 < kp1 < 10 AND p1 ) OR ( 10 <= kp2 < 20 AND p2 )
+
+   where there is a key consisting of keyparts ( kp1, kp2, ..., kpn ) and p1
+   and p2 are valid SEL_ARG expressions over keyparts kp2 ... kpn, is a valid
+   SEL_ARG condition. The disjuncts appear ordered by the minimum endpoint of
+   the first range and ranges must not overlap. It follows that they are also
+   ordered by maximum endpoints. Thus
+
+   ( 1 < kp1 <= 2 AND ( kp2 = 2 OR kp2 = 3 ) ) OR kp1 = 3
+
+   Is a a valid SER_ARG expression for a key of at least 2 keyparts.
+   
+   For simplicity, we will assume that expr2 is a single range predicate,
+   i.e. on the form ( a < x < b AND ... ). It is easy to generalize to a
+   disjunction of several predicates by subsequently call key_or for each
+   disjunct.
+
+   The algorithm iterates over each disjunct of expr1, and for each disjunct
+   where the first keypart's range overlaps with the first keypart's range in
+   expr2:
+   
+   If the predicates are equal for the rest of the keyparts, or if there are
+   no more, the range in expr2 has its endpoints copied in, and the SEL_ARG
+   node in expr2 is deallocated. If more ranges became connected in expr1, the
+   surplus is also dealocated. If they differ, two ranges are created.
+   
+   - The range leading up to the overlap. Empty if endpoints are equal.
+
+   - The overlapping sub-range. May be the entire range if they are equal.
+
+   Finally, there may be one more range if expr2's first keypart's range has a
+   greater maximum endpoint than the last range in expr1.
+
+   For the overlapping sub-range, we recursively call key_or. Thus in order to
+   compute key_or of
+
+     (1) ( 1 < kp1 < 10 AND 1 < kp2 < 10 ) 
+
+     (2) ( 2 < kp1 < 20 AND 4 < kp2 < 20 )
+
+   We create the ranges 1 < kp <= 2, 2 < kp1 < 10, 10 <= kp1 < 20. For the
+   first one, we simply hook on the condition for the second keypart from (1)
+   : 1 < kp2 < 10. For the second range 2 < kp1 < 10, key_or( 1 < kp2 < 10, 4
+   < kp2 < 20 ) is called, yielding 1 < kp2 < 20. For the last range, we reuse
+   the range 4 < kp2 < 20 from (2) for the second keypart. The result is thus
+   
+   ( 1  <  kp1 <= 2 AND 1 < kp2 < 10 ) OR
+   ( 2  <  kp1 < 10 AND 1 < kp2 < 20 ) OR
+   ( 10 <= kp1 < 20 AND 4 < kp2 < 20 )
+*/
 static SEL_ARG *
 key_or(RANGE_OPT_PARAM *param, SEL_ARG *key1,SEL_ARG *key2)
 {
@@ -6660,7 +6720,21 @@ key_or(RANGE_OPT_PARAM *param, SEL_ARG *key1,SEL_ARG *key2)
 	  key1=key1->tree_delete(save);
 	}
         last->copy_min(tmp);
-	if (last->copy_min(key2) || last->copy_max(key2))
+        bool full_range= last->copy_min(key2);
+        if (!full_range)
+        {
+          if (last->next && key2->cmp_max_to_min(last->next) >= 0)
+          {
+            last->max_value= last->next->min_value;
+            if (last->next->min_flag & NEAR_MIN)
+              last->max_flag&= ~NEAR_MAX;
+            else
+              last->max_flag|= NEAR_MAX;
+          }
+          else
+            full_range= last->copy_max(key2);
+        }
+	if (full_range)
 	{					// Full range
 	  key1->free_tree();
 	  for (; key2 ; key2=key2->next)
@@ -6670,8 +6744,6 @@ key_or(RANGE_OPT_PARAM *param, SEL_ARG *key1,SEL_ARG *key2)
 	  return 0;
 	}
       }
-      key2=key2->next;
-      continue;
     }
 
     if (cmp >= 0 && tmp->cmp_min_to_min(key2) < 0)
diff --git a/sql/sql_base.cc b/sql/sql_base.cc
index d1e96fcdbb3..e706bd04ea6 100644
--- a/sql/sql_base.cc
+++ b/sql/sql_base.cc
@@ -2305,7 +2305,8 @@ bool reopen_name_locked_table(THD* thd, TABLE_LIST* table_list, bool link_in)
   table->tablenr=thd->current_tablenr++;
   table->used_fields=0;
   table->const_table=0;
-  table->null_row= table->maybe_null= table->force_index= 0;
+  table->null_row= table->maybe_null= 0;
+  table->force_index= table->force_index_order= table->force_index_group= 0;
   table->status=STATUS_NO_RECORD;
   DBUG_RETURN(FALSE);
 }
@@ -2963,7 +2964,8 @@ TABLE *open_table(THD *thd, TABLE_LIST *table_list, MEM_ROOT *mem_root,
   table->tablenr=thd->current_tablenr++;
   table->used_fields=0;
   table->const_table=0;
-  table->null_row= table->maybe_null= table->force_index= 0;
+  table->null_row= table->maybe_null= 0;
+  table->force_index= table->force_index_order= table->force_index_group= 0;
   table->status=STATUS_NO_RECORD;
   table->insert_values= 0;
   table->fulltext_searched= 0;
diff --git a/sql/sql_parse.cc b/sql/sql_parse.cc
index 83ef525e3eb..27688e41d4f 100644
--- a/sql/sql_parse.cc
+++ b/sql/sql_parse.cc
@@ -6258,6 +6258,7 @@ TABLE_LIST *st_select_lex::add_table_to_list(THD *thd,
   ptr->table_name_length=table->table.length;
   ptr->lock_type=   lock_type;
   ptr->updating=    test(table_options & TL_OPTION_UPDATING);
+  /* TODO: remove TL_OPTION_FORCE_INDEX as it looks like it's not used */
   ptr->force_index= test(table_options & TL_OPTION_FORCE_INDEX);
   ptr->ignore_leaves= test(table_options & TL_OPTION_IGNORE_LEAVES);
   ptr->derived=	    table->sel;
diff --git a/sql/sql_select.cc b/sql/sql_select.cc
index 3f1432914a0..9dacb2c2ce4 100644
--- a/sql/sql_select.cc
+++ b/sql/sql_select.cc
@@ -1231,13 +1231,22 @@ JOIN::optimize()
       (!group_list && tmp_table_param.sum_func_count))
     order=0;
 
-  // Can't use sort on head table if using row cache
+  // Can't use sort on head table if using join buffering
   if (full_join)
   {
-    if (group_list)
-      simple_group=0;
-    if (order)
-      simple_order=0;
+    TABLE *stable= (sort_by_table == (TABLE *) 1 ? 
+      join_tab[const_tables].table : sort_by_table);
+    /* 
+      FORCE INDEX FOR ORDER BY can be used to prevent join buffering when
+      sorting on the first table.
+    */
+    if (!stable || !stable->force_index_order)
+    {
+      if (group_list)
+        simple_group= 0;
+      if (order)
+        simple_order= 0;
+    }
   }
 
   /*
diff --git a/sql/sql_select.h b/sql/sql_select.h
index a0366d47149..3f06b402638 100644
--- a/sql/sql_select.h
+++ b/sql/sql_select.h
@@ -357,6 +357,8 @@ public:
     simple_xxxxx is set if ORDER/GROUP BY doesn't include any references
     to other tables than the first non-constant table in the JOIN.
     It's also set if ORDER/GROUP BY is empty.
+    Used for deciding for or against using a temporary table to compute 
+    GROUP/ORDER BY.
   */
   bool simple_order, simple_group;
   /**
diff --git a/sql/table.cc b/sql/table.cc
index 04f2a3fbcf8..d2538eb4d59 100644
--- a/sql/table.cc
+++ b/sql/table.cc
@@ -4637,7 +4637,8 @@ Item_subselect *TABLE_LIST::containing_subselect()
     (TABLE_LIST::index_hints). Using the information in this tagged list
     this function sets the members st_table::keys_in_use_for_query, 
     st_table::keys_in_use_for_group_by, st_table::keys_in_use_for_order_by,
-    st_table::force_index and st_table::covering_keys.
+    st_table::force_index, st_table::force_index_order, 
+    st_table::force_index_group and st_table::covering_keys.
 
     Current implementation of the runtime does not allow mixing FORCE INDEX
     and USE INDEX, so this is checked here. Then the FORCE INDEX list 
@@ -4765,14 +4766,28 @@ bool TABLE_LIST::process_index_hints(TABLE *tbl)
     }
 
     /* process FORCE INDEX as USE INDEX with a flag */
+    if (!index_order[INDEX_HINT_FORCE].is_clear_all())
+    {
+      tbl->force_index_order= TRUE;
+      index_order[INDEX_HINT_USE].merge(index_order[INDEX_HINT_FORCE]);
+    }
+
+    if (!index_group[INDEX_HINT_FORCE].is_clear_all())
+    {
+      tbl->force_index_group= TRUE;
+      index_group[INDEX_HINT_USE].merge(index_group[INDEX_HINT_FORCE]);
+    }
+
+    /*
+      TODO: get rid of tbl->force_index (on if any FORCE INDEX is specified) and
+      create tbl->force_index_join instead.
+      Then use the correct force_index_XX instead of the global one.
+    */
     if (!index_join[INDEX_HINT_FORCE].is_clear_all() ||
-        !index_order[INDEX_HINT_FORCE].is_clear_all() ||
-        !index_group[INDEX_HINT_FORCE].is_clear_all())
+        tbl->force_index_group || tbl->force_index_order)
     {
       tbl->force_index= TRUE;
       index_join[INDEX_HINT_USE].merge(index_join[INDEX_HINT_FORCE]);
-      index_order[INDEX_HINT_USE].merge(index_order[INDEX_HINT_FORCE]);
-      index_group[INDEX_HINT_USE].merge(index_group[INDEX_HINT_FORCE]);
     }
 
     /* apply USE INDEX */
diff --git a/sql/table.h b/sql/table.h
index 40372fa91cf..e4a382c799f 100644
--- a/sql/table.h
+++ b/sql/table.h
@@ -752,6 +752,18 @@ struct st_table {
     bytes, it would take up 4.
   */
   my_bool force_index;
+
+  /**
+    Flag set when the statement contains FORCE INDEX FOR ORDER BY
+    See TABLE_LIST::process_index_hints().
+  */
+  my_bool force_index_order;
+
+  /**
+    Flag set when the statement contains FORCE INDEX FOR GROUP BY
+    See TABLE_LIST::process_index_hints().
+  */
+  my_bool force_index_group;
   my_bool distinct,const_table,no_rows;
 
   /**