Implementation of SUM(DISTINCT), tests cases

sql/filesort.cc:
  Snippet of filesort() code moved to function reuse_freed_buff() - 
  change buffpek pointers to use buff from freed BUFFPEK
  Used in filesort() and merge_walk().
sql/item_sum.cc:
  Implementation of Item_sum_sum_distinct - SUM(DISTINCT) item, 
  which uses Unique to resolve duplicates
sql/item_sum.h:
  New sum Item added - Item_sum_sum_distinct - for SUM(DISTINCT) function
sql/sql_class.h:
  added walk() and reset() methods to Unique, used in Item_sum_sum_distinct.
sql/sql_sort.h:
  declaration for reuse_freed_buff() to be able to use it in uniques.cc
sql/sql_yacc.yy:
  parser extended to handle MIN(DISTICNT), MAX(DISTINCT), SUM(DISTINCT)
sql/uniques.cc:
  Implementation for Unique::reset(), Unique::walk() as well as for merge_walk() 
  algorithm.

1 files changed, 188 insertions, 0 deletions

diff --git a/mysql-test/t/sum_distinct.test b/mysql-test/t/sum_distinct.test
new file mode 100644
index 00000000000..efbb21a7b85
--- /dev/null
+++ b/mysql-test/t/sum_distinct.test
@@ -0,0 +1,188 @@
+#
+# Various tests for SUM(DISTINCT ...)
+#
+--disable_warnings
+DROP TABLE IF EXISTS t1;
+--enable_warnings
+
+CREATE TABLE t1 (
+  id INTEGER NOT NULL PRIMARY KEY AUTO_INCREMENT, 
+  gender CHAR(1),
+  name VARCHAR(20)
+);
+
+# According to ANSI SQL, SUM(DISTINCT ...) should return NULL for empty
+# record set
+
+SELECT SUM(DISTINCT LENGTH(name)) s1 FROM t1;
+
+# According to ANSI SQL, SUM(DISTINCT ...) should return NULL for records sets
+# entirely consisting of NULLs
+
+INSERT INTO t1 (gender, name) VALUES (NULL, NULL);
+INSERT INTO t1 (gender, name) VALUES (NULL, NULL);
+INSERT INTO t1 (gender, name) VALUES (NULL, NULL);
+
+SELECT SUM(DISTINCT LENGTH(name)) s1 FROM t1;
+
+
+# Filling table with t1
+
+INSERT INTO t1 (gender, name) VALUES ('F', 'Helen'), ('F', 'Anastasia'),
+('F', 'Katherine'), ('F', 'Margo'), ('F', 'Magdalene'), ('F', 'Mary');
+
+CREATE TABLE t2 SELECT name FROM t1;
+
+SELECT (SELECT SUM(DISTINCT LENGTH(name)) FROM t1) FROM t2;
+
+DROP TABLE t2;
+
+INSERT INTO t1 (gender, name) VALUES ('F', 'Eva'), ('F', 'Sofia'),
+('F', 'Sara'), ('F', 'Golda'), ('F', 'Toba'), ('F', 'Victory'),
+('F', 'Faina'), ('F', 'Miriam'), ('F', 'Beki'), ('F', 'America'),
+('F', 'Susan'), ('F', 'Glory'), ('F', 'Priscilla'), ('F', 'Rosmary'),
+('F', 'Rose'), ('F', 'Margareth'), ('F', 'Elizabeth'), ('F', 'Meredith'),
+('F', 'Julie'), ('F', 'Xenia'), ('F', 'Zena'), ('F', 'Olga'),
+('F', 'Brunhilda'), ('F', 'Nataly'), ('F', 'Lara'), ('F', 'Svetlana'),
+('F', 'Grethem'), ('F', 'Irene');
+
+SELECT
+  SUM(DISTINCT LENGTH(name)) s1,
+  SUM(DISTINCT SUBSTRING(NAME, 1, 3)) s2,
+  SUM(DISTINCT LENGTH(SUBSTRING(name, 1, 4))) s3
+FROM t1;
+
+SELECT
+  SUM(DISTINCT LENGTH(g1.name)) s1,
+  SUM(DISTINCT SUBSTRING(g2.name, 1, 3)) s2,
+  SUM(DISTINCT LENGTH(SUBSTRING(g3.name, 1, 4))) s3
+FROM t1 g1, t1 g2, t1 g3;
+
+SELECT
+  SUM(DISTINCT LENGTH(g1.name)) s1,
+  SUM(DISTINCT SUBSTRING(g2.name, 1, 3)) s2,
+  SUM(DISTINCT LENGTH(SUBSTRING(g3.name, 1, 4))) s3
+FROM t1 g1, t1 g2, t1 g3 GROUP BY LENGTH(SUBSTRING(g3.name, 5, 10));
+
+# here we explicitly request summing through temporary table (so
+# Item_sum_sum_distinct::copy_or_same() is called)
+
+SELECT SQL_BUFFER_RESULT
+  SUM(DISTINCT LENGTH(name)) s1,
+  SUM(DISTINCT SUBSTRING(NAME, 1, 3)) s2,
+  SUM(DISTINCT LENGTH(SUBSTRING(name, 1, 4))) s3
+FROM t1;
+
+SELECT SQL_BUFFER_RESULT
+  SUM(DISTINCT LENGTH(g1.name)) s1,
+  SUM(DISTINCT SUBSTRING(g2.name, 1, 3)) s2,
+  SUM(DISTINCT LENGTH(SUBSTRING(g3.name, 1, 4))) s3
+FROM t1 g1, t1 g2, t1 g3 GROUP BY LENGTH(SUBSTRING(g3.name, 5, 10));
+
+# this test demonstrates that strings are automatically converted to numbers
+# before summing
+
+SET @l=1;
+UPDATE t1 SET name=CONCAT(name, @l:=@l+1);
+
+SELECT SUM(DISTINCT RIGHT(name, 1)) FROM t1;
+
+# this is a test case for ordinary t1 
+
+SELECT SUM(DISTINCT id) FROM t1;
+SELECT SUM(DISTINCT id % 11) FROM t1;
+
+DROP TABLE t1;
+
+#
+# Test the case when distinct values doesn't fit in memory and 
+# filesort is used (see uniques.cc:merge_walk)
+#
+
+CREATE TABLE t1 (id INTEGER);
+CREATE TABLE t2 (id INTEGER);
+
+INSERT INTO t1 (id) VALUES (1), (1), (1),(1);
+INSERT INTO t2 (id) SELECT id FROM t1;
+INSERT INTO t1 (id) SELECT id FROM t2; /* 8 */
+INSERT INTO t1 (id) SELECT id FROM t2; /* 12 */
+INSERT INTO t1 (id) SELECT id FROM t2; /* 16 */
+INSERT INTO t1 (id) SELECT id FROM t2; /* 20 */
+INSERT INTO t1 (id) SELECT id FROM t2; /* 24 */
+DELETE FROM t2;
+INSERT INTO t2 (id) SELECT id+1 FROM t1;
+INSERT INTO t1 SELECT id FROM t2;
+DELETE FROM t2;
+INSERT INTO t2 (id) SELECT id+2 FROM t1;
+INSERT INTO t1 SELECT id FROM t2;
+DELETE FROM t2;
+INSERT INTO t2 (id) SELECT id+4 FROM t1;
+INSERT INTO t1 SELECT id FROM t2;
+DELETE FROM t2;
+INSERT INTO t2 (id) SELECT id+8 FROM t1;
+INSERT INTO t1 SELECT id FROM t2;
+DELETE FROM t2;
+INSERT INTO t2 (id) SELECT id+16 FROM t1;
+INSERT INTO t1 SELECT id FROM t2;
+DELETE FROM t2;
+INSERT INTO t2 (id) SELECT id+32 FROM t1;
+INSERT INTO t1 SELECT id FROM t2;
+DELETE FROM t2;
+INSERT INTO t2 (id) SELECT id+64 FROM t1;
+INSERT INTO t1 SELECT id FROM t2;
+DELETE FROM t2;
+INSERT INTO t2 (id) SELECT id+128 FROM t1;
+INSERT INTO t1 SELECT id FROM t2;
+DELETE FROM t2;
+INSERT INTO t2 (id) SELECT id+256 FROM t1;
+INSERT INTO t1 SELECT id FROM t2;
+DELETE FROM t2;
+INSERT INTO t2 (id) SELECT id+512 FROM t1;
+INSERT INTO t1 SELECT id FROM t2;
+DELETE FROM t2;
+INSERT INTO t2 (id) SELECT id+1024 FROM t1;
+INSERT INTO t1 SELECT id FROM t2;
+DELETE FROM t2;
+INSERT INTO t2 (id) SELECT id+2048 FROM t1;
+INSERT INTO t1 SELECT id FROM t2;
+DELETE FROM t2;
+INSERT INTO t2 (id) SELECT id+4096 FROM t1;
+INSERT INTO t1 SELECT id FROM t2;
+DELETE FROM t2;
+INSERT INTO t2 (id) SELECT id+8192 FROM t1;
+INSERT INTO t1 SELECT id FROM t2;
+DELETE FROM t2;
+#INSERT INTO t2 (id) SELECT id+16384 FROM t1;
+#INSERT INTO t1 SELECT id FROM t2;
+#DELETE FROM t2;
+#INSERT INTO t2 (id) SELECT id+32768 FROM t1;
+#INSERT INTO t1 SELECT id FROM t2;
+#DELETE FROM t2;
+#INSERT INTO t2 (id) SELECT id+65536 FROM t1;
+#INSERT INTO t1 SELECT id FROM t2;
+#DELETE FROM t2;
+INSERT INTO t2 SELECT id FROM t1 ORDER BY id*rand();
+
+# SELECT '++++++++++++++++++++++++++++++++++++++++++++++++++';
+
+SELECT SUM(DISTINCT id) sm FROM t1;
+SELECT SUM(DISTINCT id) sm FROM t2;
+SELECT SUM(DISTINCT id) sm FROM t1 group by id % 13;
+
+# this limit for max_heap_table_size is set to force testing the case, when
+# all distinct sum values can not fit in memory and must be stored in a
+# temporary table
+
+SET max_heap_table_size=16384;
+
+# to check that max_heap_table_size was actually set (hard limit for minimum
+# max_heap_table_size is set in mysqld.cc):
+
+SHOW variables LIKE 'max_heap_table_size';
+
+SELECT SUM(DISTINCT id) sm FROM t1;
+SELECT SUM(DISTINCT id) sm FROM t2;
+SELECT SUM(DISTINCT id) sm FROM t1 GROUP BY id % 13;
+
+DROP TABLE t1;
+DROP TABLE t2;