Fixed bug mdev-4025.

The bug could lead to a wrong estimate of the number of expected rows
in the output of the EXPLAIN commands for queries with GROUP BY.
This could be observed in the test case for LP bug 934348.

mysql-test/r/subselect_sj_jcl6.result

@@ -2978,7 +2978,7 @@ EXPLAIN
 SELECT a FROM t1 t WHERE a IN (SELECT b FROM t1, t2 WHERE b = a)
 GROUP BY a HAVING a != 'z';
 id	select_type	table	type	possible_keys	key	key_len	ref	rows	Extra
-1	PRIMARY	t	index	idx_a	idx_a	4	NULL	1	Using index
+1	PRIMARY	t	index	idx_a	idx_a	4	NULL	3	Using index
 1	PRIMARY	<subquery2>	eq_ref	distinct_key	distinct_key	4	func	1	
 2	MATERIALIZED	t2	ALL	NULL	NULL	NULL	NULL	2	Using where
 2	MATERIALIZED	t1	ref	idx_a	idx_a	4	test.t2.b	2	Using index
@@ -2992,7 +2992,7 @@ EXPLAIN
 SELECT a FROM t1 t WHERE a IN (SELECT b FROM t1, t2 WHERE b = a)
 GROUP BY a HAVING a != 'z';
 id	select_type	table	type	possible_keys	key	key_len	ref	rows	Extra
-1	PRIMARY	t	index	idx_a	idx_a	4	NULL	1	Using index
+1	PRIMARY	t	index	idx_a	idx_a	4	NULL	3	Using index
 1	PRIMARY	<subquery2>	eq_ref	distinct_key	distinct_key	4	func	1	
 2	MATERIALIZED	t2	ALL	NULL	NULL	NULL	NULL	2	Using where
 2	MATERIALIZED	t1	ref	idx_a	idx_a	4	test.t2.b	2	Using index

sql/sql_select.cc

@@ -18371,8 +18371,9 @@ test_if_skip_sort_order(JOIN_TAB *tab,ORDER *order,ha_rows select_limit_arg,
             and as result we'll choose an index scan when using ref/range
             access + filesort will be cheaper.
 	  */
-          select_limit= (ha_rows) (select_limit < fanout ?
-                                   1 : select_limit/fanout);
+          if (select_limit_arg != HA_POS_ERROR)
+            select_limit= (ha_rows) (select_limit < fanout ?
+                                     1 : select_limit/fanout);
           /*
             We assume that each of the tested indexes is not correlated
             with ref_key. Thus, to select first N records we have to scan