MDEV-21932 Another attempt to fix the bug .

The first patch for the bug was erroneous: it did not take into account
the fact that the modified function get_key_scans_params() was called in
different contexts. As a result the patch caused a regression bug MDEV-22191.
The patch for this bug introduced an extra parameter. Actually we can
do without this parameter and use the fourth parameter for the same
puropose - to differentiate between the calls of the function for range
access and for index merge access.

Also removed the call of get_key_scans_params() in the code of the function
merge_same_index_scans() as not needed.

1 files changed, 16 insertions, 18 deletions

diff --git a/sql/opt_range.cc b/sql/opt_range.cc
index 9e96e94ac91..5658bb065b7 100644
--- a/sql/opt_range.cc
+++ b/sql/opt_range.cc
@@ -948,9 +948,8 @@ QUICK_RANGE_SELECT *get_quick_select(PARAM *param,uint index,
                                      uint mrr_buf_size, MEM_ROOT *alloc);
 static TRP_RANGE *get_key_scans_params(PARAM *param, SEL_TREE *tree,
                                        bool index_read_must_be_used,
-                                       bool update_tbl_stats,
-                                       double read_time,
-                                       bool ror_scans_required);
+                                       bool for_range_access,
+                                       double read_time);
 static
 TRP_INDEX_INTERSECT *get_best_index_intersect(PARAM *param, SEL_TREE *tree,
                                               double read_time);
@@ -3147,7 +3146,7 @@ int SQL_SELECT::test_quick_select(THD *thd, key_map keys_to_use,
 
       /* Get best 'range' plan and prepare data for making other plans */
       if ((range_trp= get_key_scans_params(&param, tree, FALSE, TRUE,
-                                           best_read_time, FALSE)))
+                                           best_read_time)))
       {
         best_trp= range_trp;
         best_read_time= best_trp->read_cost;
@@ -4673,7 +4672,6 @@ TABLE_READ_PLAN *get_best_disjunct_quick(PARAM *param, SEL_IMERGE *imerge,
   double roru_index_costs;
   ha_rows roru_total_records;
   double roru_intersect_part= 1.0;
-  bool only_ror_scans_required= FALSE;
   DBUG_ENTER("get_best_disjunct_quick");
   DBUG_PRINT("info", ("Full table scan cost: %g", read_time));
 
@@ -4700,8 +4698,6 @@ TABLE_READ_PLAN *get_best_disjunct_quick(PARAM *param, SEL_IMERGE *imerge,
                                              n_child_scans)))
     DBUG_RETURN(NULL);
 
-  only_ror_scans_required= !optimizer_flag(param->thd,
-                                      OPTIMIZER_SWITCH_INDEX_MERGE_SORT_UNION);
   /*
     Collect best 'range' scan for each of disjuncts, and, while doing so,
     analyze possibility of ROR scans. Also calculate some values needed by
@@ -4714,8 +4710,7 @@ TABLE_READ_PLAN *get_best_disjunct_quick(PARAM *param, SEL_IMERGE *imerge,
     DBUG_EXECUTE("info", print_sel_tree(param, *ptree, &(*ptree)->keys_map,
                                         "tree in SEL_IMERGE"););
     if (!(*cur_child= get_key_scans_params(param, *ptree, TRUE, FALSE,
-                                           read_time,
-                                           only_ror_scans_required)))
+                                           read_time)))
     {
       /*
         One of index scans in this index_merge is more expensive than entire
@@ -5035,9 +5030,12 @@ TABLE_READ_PLAN *merge_same_index_scans(PARAM *param, SEL_IMERGE *imerge,
          a random order
       2. the functions that estimate the cost of a range scan and an
          index merge retrievals are not well calibrated
+
+      As the best range access has been already chosen it does not
+      make sense to evaluate the one obtained from a degenerated
+      index merge.
     */
-    trp= get_key_scans_params(param, *imerge->trees, FALSE, TRUE,
-                              read_time, FALSE);
+    trp= 0;
   }
 
   DBUG_RETURN(trp); 
@@ -6753,9 +6751,9 @@ TRP_ROR_INTERSECT *get_best_covering_ror_intersect(PARAM *param,
       tree         make range select for this SEL_TREE
       index_read_must_be_used if TRUE, assume 'index only' option will be set
                              (except for clustered PK indexes)
+      for_range_access     if TRUE the function is called to get the best range
+                           plan for range access, not for index merge access
       read_time    don't create read plans with cost > read_time.
-      only_ror_scans_required         set to TRUE when we are only interested
-                                      in ROR scan
   RETURN
     Best range read plan
     NULL if no plan found or error occurred
@@ -6763,9 +6761,8 @@ TRP_ROR_INTERSECT *get_best_covering_ror_intersect(PARAM *param,
 
 static TRP_RANGE *get_key_scans_params(PARAM *param, SEL_TREE *tree,
                                        bool index_read_must_be_used, 
-                                       bool update_tbl_stats,
-                                       double read_time,
-                                       bool only_ror_scans_required)
+                                       bool for_range_access,
+                                       double read_time)
 {
   uint idx;
   SEL_ARG **key,**end, **key_to_read= NULL;
@@ -6809,10 +6806,11 @@ static TRP_RANGE *get_key_scans_params(PARAM *param, SEL_TREE *tree,
                             (bool) param->table->covering_keys.is_set(keynr);
 
       found_records= check_quick_select(param, idx, read_index_only, *key,
-                                        update_tbl_stats, &mrr_flags,
+                                        for_range_access, &mrr_flags,
                                         &buf_size, &cost);
 
-      if (only_ror_scans_required && !param->is_ror_scan)
+      if (!for_range_access && !param->is_ror_scan &&
+          !optimizer_flag(param->thd,OPTIMIZER_SWITCH_INDEX_MERGE_SORT_UNION))
       {
         /* The scan is not a ROR-scan, just skip it */
         continue;