Re: Source code question related to sql_optimizer.cc line 8583

[Øystein Grøvlen]

Hi,

IIUC, your are suggesting this change:

--- a/sql/sql_optimizer.cc
+++ b/sql/sql_optimizer.cc
@@ -8912,7 +8912,7 @@ bool JOIN::attach_join_conditions(plan_idx last_tab)
DBUG_RETURN(true);

// Add the guard turning the predicate off for the null-complemented row.
- cond= new Item_func_trig_cond(cond, NULL, this, first_inner,
+ cond= new Item_func_trig_cond(cond, NULL, this, i,
Item_func_trig_cond::IS_NOT_NULL_COMPL);
if (!cond)
DBUG_RETURN(true);

Running the test suite with this change leads to multiple seg faults; an indication that the change is not correct.

AFAIU, first_inner is used because you are supposed to pass the index for the first table of the join nest. This code in Item_func_trig_cond::get_table_range() indicates that:

JOIN_TAB *join_tab= m_join->best_ref[m_idx];
*first_table= join_tab->table_ref;
last_inner= join_tab->last_inner();
*last_table= m_join->best_ref[last_inner]->table_ref;

m_idx is here the value we are discussing.

Øystein Grøvlen,
Senior Principal Software Engineer,
MySQL Group, Oracle,
Trondheim, Norway