gistfile1.txt

=== Old execution (like before the patch) ===

update series set val=1; SELECT t, next_seq_value() r FROM t1 IGNORE INDEX(t) GROUP BY t HAVING r = 1  ORDER BY t1.u;
+------+------+
| t    | r    |
+------+------+
|   10 |    1 |
|   12 |    1 |
|   14 |    1 |
|   16 |    1 |
+------+------+

Here I can see the code use end_write(), Group-by implementation based on a temporary table with a group key.
The SP-func is invoked for every record.
  +------+------+
  | t    | u    |
  +------+------+
* |   10 |   10 |  sp_func()=1 
  |   11 |   11 |  sp_func()=0
* |   12 |   12 |  sp_func()=1
  |   12 |   13 |  sp_func()=0
* |   14 |   15 |  sp_func()=1
  |   15 |   16 |  sp_func()=0
* |   16 |   17 |  sp_func()=1
  |   17 |   17 |  sp_func()=0
  +------+------+
Ok.

== New execution (after the patch) ==

update series set val=1; 
SELECT t, next_seq_value() r FROM t1 FORCE INDEX(t) GROUP BY t HAVING r = 1  ORDER BY t1.u;
+------+------+
| t    | r    |
+------+------+
|   10 |    1 |
|   12 |    1 |
|   15 |    1 |
|   17 |    1 |
+------+------+

The code uses end_write_group() - that is, it does group-by on the fly (one group follows other), and then writes the result into the temp.table

  +------+------+
  | t    | u    |
  +------+------+

* |   10 |   10 |  sp_func()=1

  |   11 |   11 |  sp_func()=0

* |   12 |   12 |  sp_func()=1
  |   12 |   13 |  Note -- there is NO call to sp_func() here. The code sees we're still in the same group and exits.

  |   14 |   15 |  sp_func()=0

* |   15 |   16 |  sp_func()=1

  |   16 |   17 |  sp_func()=0

* |   17 |   17 |  sp_func()=1
  +------+------+