I wonder if we need to set JOIN_TYPE_OUTER.
Before this patch it is used in a very odd way:

$ grep -rn JOIN_TYPE_OUTER sql
sql/sql_select.cc:31713:    /* JOIN_TYPE_OUTER is just a marker unrelated to real join */
sql/sql_derived.cc:803:        cursor->outer_join|= JOIN_TYPE_OUTER;
sql/table.h:2217:#define JOIN_TYPE_OUTER 4U     /* Marker that this is an outer join */

It is set only in sql_derived.cc: mysql_derived_prepare().
as far as I understand, the logic is: if we have a derived table dt that's LEFT JOIN'ed:

 ... LEFT JOIN  (SELECT .... FROM t1, t2  ) dt ON expr ```

then we set JOIN_TYPE_OUTER for all the tables in the derived table (t1, t2 in this example).

But then if a table has JOIN_TYPE_OUTER it does not mean that it will be joined with its peers using any outer join operation - here t1 and t2 are joined with inner join...

... and that's all the input.
Need to figure if we need to set JOIN_TYPE_OUTER here (I think, no) and then this patch is good.

... and that's all the input. Need to figure if we need to set JOIN_TYPE_OUTER here (I think, no) and then this patch is good.

It's logically sufficient to specify on JOIN_TYPE_FULL in combination with the LEFT/RIGHT bits as that will be unambiguous in the code. We don't strictly need to mark JOIN_TYPE_OUTER because FULL OUTER JOIN and FULL JOIN are equivalent. However, I wasn't sure if we wanted to keep the marking so as to preserve the user input when rendering back the query during EXPLAIN EXTENDED (or any other place where the query is shown, like during VIEW .frm file creation).