Fix Bug#11333 "Stored Procedure: Memory blow up on repeated SELECT ... INTO query"

mysql-test/r/sp.result:
  update result
mysql-test/t/sp.test:
  Add test for Bug #11333 "Stored Procedure: Memory blow up on repeated SELECT ... INTO query"
sql/item.cc:
  we should call destructors for Items before reuse
sql/item.h:
  Add new method and constructor for Item_string.
sql/sp_head.cc:
  String allocation should be done on the system heap for now.

1 files changed, 19 insertions, 4 deletions

diff --git a/sql/sp_head.cc b/sql/sp_head.cc
index f119ef1ec22..0a3521e8855 100644
--- a/sql/sp_head.cc
+++ b/sql/sp_head.cc
@@ -249,10 +249,25 @@ sp_eval_func_item(THD *thd, Item **it_addr, enum enum_field_types type,
         DBUG_PRINT("info",("default result: %*s",
                            s->length(), s->c_ptr_quick()));
         CREATE_ON_CALLERS_ARENA(it= new(reuse, &rsize)
-                              Item_string(thd->strmake(s->ptr(),
-                                          s->length()), s->length(),
-                                          it->collation.collation),
-                          use_callers_arena, &backup_current_arena);
+                                Item_string(it->collation.collation),
+                                use_callers_arena, &backup_current_arena);
+        /*
+          We have to use special constructor and allocate string
+          on system heap here. This is because usual Item_string
+          constructor would allocate memory in the callers arena.
+          This would lead to the memory leak in SP loops.
+          See Bug #11333 "Stored Procedure: Memory blow up on
+          repeated SELECT ... INTO query" for sample of such SP.
+          TODO: Usage of the system heap gives significant overhead,
+          however usual "reuse" mechanism does not work here, as
+          Item_string has no max size. That is, if we have a loop, which
+          has string variable with constantly increasing size, we would have
+          to allocate new pieces of memory again and again on each iteration.
+          In future we should probably reserve some area of memory for
+          not-very-large strings and reuse it. But for large strings
+          we would have to use system heap anyway.
+        */
+        ((Item_string*) it)->set_str_with_copy(s->ptr(), s->length());
       }
       break;
     }