fix for bug #958 and tests for it

aligned with Sergey notes.


sql/records.cc:
  fix for bug #958 - big select with 
  count distinct and group by
  aligned with Sergey notes.

3 files changed, 69 insertions, 1 deletions

diff --git a/mysql-test/r/count_distinct3.result b/mysql-test/r/count_distinct3.result
new file mode 100644
index 00000000000..633bb2fa252
--- /dev/null
+++ b/mysql-test/r/count_distinct3.result
@@ -0,0 +1,7 @@
+DROP TABLE IF EXISTS t1, t2;
+CREATE TABLE t1 (id INTEGER, grp TINYINT, id_rev INTEGER);
+SELECT COUNT(*) FROM t1;
+COUNT(*)
+4181000
+SELECT COUNT(DISTINCT id) FROM t1 GROUP BY grp;
+DROP TABLE t1;
diff --git a/mysql-test/t/count_distinct3.test b/mysql-test/t/count_distinct3.test
new file mode 100644
index 00000000000..e6cc98d47df
--- /dev/null
+++ b/mysql-test/t/count_distinct3.test
@@ -0,0 +1,55 @@
+#
+# this is a test for error 1032 in count(distinct) + group by, introduced in 
+# mysql-4.1
+#
+
+--disable_warnings
+DROP TABLE IF EXISTS t1, t2;
+--enable_warnings
+
+CREATE TABLE t1 (id INTEGER, grp TINYINT, id_rev INTEGER);
+
+--disable_query_log
+SET @rnd_max= 2147483647;
+let $1 = 1000;
+while ($1)
+{
+  SET @rnd= RAND();
+  SET @id = CAST(@rnd * @rnd_max AS UNSIGNED);
+  SET @id_rev= @rnd_max - @id;
+  SET @grp= CAST(128.0 * @rnd AS UNSIGNED); 
+  INSERT INTO t1 (id, grp, id_rev) VALUES (@id, @grp, @id_rev); 
+  dec $1;
+}
+CREATE TABLE t2 SELECT * FROM t1;
+INSERT INTO t1 (id, grp, id_rev) SELECT id, grp, id_rev FROM t2;
+INSERT INTO t2 (id, grp, id_rev) SELECT id, grp, id_rev FROM t1;
+INSERT INTO t1 (id, grp, id_rev) SELECT id, grp, id_rev FROM t2;
+INSERT INTO t2 (id, grp, id_rev) SELECT id, grp, id_rev FROM t1;
+INSERT INTO t1 (id, grp, id_rev) SELECT id, grp, id_rev FROM t2;
+INSERT INTO t2 (id, grp, id_rev) SELECT id, grp, id_rev FROM t1;
+INSERT INTO t1 (id, grp, id_rev) SELECT id, grp, id_rev FROM t2;
+INSERT INTO t2 (id, grp, id_rev) SELECT id, grp, id_rev FROM t1;
+INSERT INTO t1 (id, grp, id_rev) SELECT id, grp, id_rev FROM t2;
+INSERT INTO t2 (id, grp, id_rev) SELECT id, grp, id_rev FROM t1;
+INSERT INTO t1 (id, grp, id_rev) SELECT id, grp, id_rev FROM t2;
+INSERT INTO t2 (id, grp, id_rev) SELECT id, grp, id_rev FROM t1;
+INSERT INTO t1 (id, grp, id_rev) SELECT id, grp, id_rev FROM t2;
+INSERT INTO t2 (id, grp, id_rev) SELECT id, grp, id_rev FROM t1;
+INSERT INTO t1 (id, grp, id_rev) SELECT id, grp, id_rev FROM t2;
+INSERT INTO t2 (id, grp, id_rev) SELECT id, grp, id_rev FROM t1;
+INSERT INTO t1 (id, grp, id_rev) SELECT id, grp, id_rev FROM t2;
+INSERT INTO t2 (id, grp, id_rev) SELECT id, grp, id_rev FROM t1;
+DROP TABLE t2;
+--enable_query_log
+
+SELECT COUNT(*) FROM t1;
+
+# As t1 contains random numbers, results are different from test to test. 
+# That's okay, because we test only that select doesn't yield an
+# error. Note, that --disable_result_log doesn't suppress error output.
+
+--disable_result_log
+SELECT COUNT(DISTINCT id) FROM t1 GROUP BY grp;
+--enable_result_log
+DROP TABLE t1;
diff --git a/sql/records.cc b/sql/records.cc
index 783ca663dfe..72a6d480356 100644
--- a/sql/records.cc
+++ b/sql/records.cc
@@ -73,7 +73,13 @@ void init_read_record(READ_RECORD *info,THD *thd, TABLE *table,
     info->ref_pos=table->file->ref;
     table->file->rnd_init(0);
 
-    if (! (specialflag & SPECIAL_SAFE_MODE) &&
+    /*
+      table->sort.addon_field is checked because if we use addon fields,
+      it doesn't make sense to use cache - we don't read from the table
+      and table->sort.io_cache is read sequentially
+    */
+    if (!table->sort.addon_field &&
+        ! (specialflag & SPECIAL_SAFE_MODE) &&
 	thd->variables.read_rnd_buff_size &&
 	!table->file->fast_key_read() &&
 	(table->db_stat & HA_READ_ONLY ||