sl@0: # 2007 November 29
sl@0: #
sl@0: # The author disclaims copyright to this source code.  In place of
sl@0: # a legal notice, here is a blessing:
sl@0: #
sl@0: #    May you do good and not evil.
sl@0: #    May you find forgiveness for yourself and forgive others.
sl@0: #    May you share freely, never taking more than you give.
sl@0: #
sl@0: #***********************************************************************
sl@0: # This file tests the optimisations made in November 2007 of expressions 
sl@0: # of the following form:
sl@0: #
sl@0: #     <value> IN (SELECT <column> FROM <table>)
sl@0: #
sl@0: # $Id: in3.test,v 1.5 2008/08/04 03:51:24 danielk1977 Exp $
sl@0: 
sl@0: set testdir [file dirname $argv0]
sl@0: source $testdir/tester.tcl
sl@0: 
sl@0: ifcapable !subquery {
sl@0:   finish_test
sl@0:   return
sl@0: }
sl@0: 
sl@0: # Return the number of OpenEphemeral instructions used in the
sl@0: # implementation of the sql statement passed as a an argument.
sl@0: #
sl@0: proc nEphemeral {sql} {
sl@0:   set nEph 0
sl@0:   foreach op [execsql "EXPLAIN $sql"] {
sl@0:     if {$op eq "OpenEphemeral"} {incr nEph}
sl@0:   }
sl@0:   set nEph
sl@0: }
sl@0: 
sl@0: # This proc works the same way as execsql, except that the number
sl@0: # of OpenEphemeral instructions used in the implementation of the
sl@0: # statement is inserted into the start of the returned list.
sl@0: #
sl@0: proc exec_neph {sql} {
sl@0:   return [concat [nEphemeral $sql] [execsql $sql]]
sl@0: }
sl@0: 
sl@0: do_test in3-1.1 {
sl@0:   execsql {
sl@0:     CREATE TABLE t1(a PRIMARY KEY, b);
sl@0:     INSERT INTO t1 VALUES(1, 2);
sl@0:     INSERT INTO t1 VALUES(3, 4);
sl@0:     INSERT INTO t1 VALUES(5, 6);
sl@0:   }
sl@0: } {}
sl@0: 
sl@0: # All of these queries should avoid using a temp-table:
sl@0: #
sl@0: do_test in3-1.2 {
sl@0:   exec_neph { SELECT rowid FROM t1 WHERE rowid IN (SELECT rowid FROM t1); }
sl@0: } {0 1 2 3}
sl@0: do_test in3-1.3 {
sl@0:   exec_neph { SELECT a FROM t1 WHERE a IN (SELECT a FROM t1); }
sl@0: } {0 1 3 5}
sl@0: do_test in3-1.4 {
sl@0:   exec_neph { SELECT rowid FROM t1 WHERE rowid+0 IN (SELECT rowid FROM t1); }
sl@0: } {0 1 2 3}
sl@0: do_test in3-1.5 {
sl@0:   exec_neph { SELECT a FROM t1 WHERE a+0 IN (SELECT a FROM t1); }
sl@0: } {0 1 3 5}
sl@0: 
sl@0: # Because none of the sub-select queries in the following statements
sl@0: # match the pattern ("SELECT <column> FROM <table>"), the following do 
sl@0: # require a temp table.
sl@0: #
sl@0: do_test in3-1.6 {
sl@0:   exec_neph { SELECT rowid FROM t1 WHERE rowid IN (SELECT rowid+0 FROM t1); }
sl@0: } {1 1 2 3}
sl@0: do_test in3-1.7 {
sl@0:   exec_neph { SELECT a FROM t1 WHERE a IN (SELECT a+0 FROM t1); }
sl@0: } {1 1 3 5}
sl@0: do_test in3-1.8 {
sl@0:   exec_neph { SELECT a FROM t1 WHERE a IN (SELECT a FROM t1 WHERE 1); }
sl@0: } {1 1 3 5}
sl@0: do_test in3-1.9 {
sl@0:   exec_neph { SELECT a FROM t1 WHERE a IN (SELECT a FROM t1 GROUP BY a); }
sl@0: } {1 1 3 5}
sl@0: 
sl@0: # This should not use a temp-table. Even though the sub-select does
sl@0: # not exactly match the pattern "SELECT <column> FROM <table>", in
sl@0: # this case the ORDER BY is a no-op and can be ignored.
sl@0: do_test in3-1.10 {
sl@0:   exec_neph { SELECT a FROM t1 WHERE a IN (SELECT a FROM t1 ORDER BY a); }
sl@0: } {0 1 3 5}
sl@0: 
sl@0: # These do use the temp-table. Adding the LIMIT clause means the 
sl@0: # ORDER BY cannot be ignored.
sl@0: do_test in3-1.11 {
sl@0:   exec_neph {SELECT a FROM t1 WHERE a IN (SELECT a FROM t1 ORDER BY a LIMIT 1)}
sl@0: } {1 1}
sl@0: do_test in3-1.12 {
sl@0:   exec_neph {
sl@0:     SELECT a FROM t1 WHERE a IN (SELECT a FROM t1 ORDER BY a LIMIT 1 OFFSET 1)
sl@0:   }
sl@0: } {1 3}
sl@0: 
sl@0: # Has to use a temp-table because of the compound sub-select.
sl@0: #
sl@0: ifcapable compound {
sl@0:   do_test in3-1.13 {
sl@0:     exec_neph {
sl@0:       SELECT a FROM t1 WHERE a IN (
sl@0:         SELECT a FROM t1 UNION ALL SELECT a FROM t1
sl@0:       )
sl@0:     }
sl@0:   } {1 1 3 5}
sl@0: }
sl@0: 
sl@0: # The first of these queries has to use the temp-table, because the 
sl@0: # collation sequence used for the index on "t1.a" does not match the
sl@0: # collation sequence used by the "IN" comparison. The second does not
sl@0: # require a temp-table, because the collation sequences match.
sl@0: #
sl@0: do_test in3-1.14 {
sl@0:   exec_neph { SELECT a FROM t1 WHERE a COLLATE nocase IN (SELECT a FROM t1) }
sl@0: } {1 1 3 5}
sl@0: do_test in3-1.15 {
sl@0:   exec_neph { SELECT a FROM t1 WHERE a COLLATE binary IN (SELECT a FROM t1) }
sl@0: } {0 1 3 5}
sl@0: 
sl@0: # Neither of these queries require a temp-table. The collation sequence
sl@0: # makes no difference when using a rowid.
sl@0: #
sl@0: do_test in3-1.16 {
sl@0:   exec_neph {SELECT a FROM t1 WHERE a COLLATE nocase IN (SELECT rowid FROM t1)}
sl@0: } {0 1 3}
sl@0: do_test in3-1.17 {
sl@0:   exec_neph {SELECT a FROM t1 WHERE a COLLATE binary IN (SELECT rowid FROM t1)}
sl@0: } {0 1 3}
sl@0: 
sl@0: # The following tests - in3.2.* - test a bug that was difficult to track
sl@0: # down during development. They are not particularly well focused.
sl@0: #
sl@0: do_test in3-2.1 {
sl@0:   execsql {
sl@0:     DROP TABLE IF EXISTS t1;
sl@0:     CREATE TABLE t1(w int, x int, y int);
sl@0:     CREATE TABLE t2(p int, q int, r int, s int);
sl@0:   }
sl@0:   for {set i 1} {$i<=100} {incr i} {
sl@0:     set w $i
sl@0:     set x [expr {int(log($i)/log(2))}]
sl@0:     set y [expr {$i*$i + 2*$i + 1}]
sl@0:     execsql "INSERT INTO t1 VALUES($w,$x,$y)"
sl@0:   }
sl@0:   set maxy [execsql {select max(y) from t1}]
sl@0:   db eval { INSERT INTO t2 SELECT 101-w, x, $maxy+1-y, y FROM t1 }
sl@0: } {}
sl@0: do_test in3-2.2 {
sl@0:   execsql {
sl@0:     SELECT rowid 
sl@0:     FROM t1 
sl@0:     WHERE rowid IN (SELECT rowid FROM t1 WHERE rowid IN (1, 2));
sl@0:   }
sl@0: } {1 2}
sl@0: do_test in3-2.3 {
sl@0:   execsql {
sl@0:     select rowid from t1 where rowid IN (-1,2,4)
sl@0:   }
sl@0: } {2 4}
sl@0: do_test in3-2.4 {
sl@0:   execsql {
sl@0:     SELECT rowid FROM t1 WHERE rowid IN 
sl@0:        (select rowid from t1 where rowid IN (-1,2,4))
sl@0:   }
sl@0: } {2 4}
sl@0: 
sl@0: #-------------------------------------------------------------------------
sl@0: # This next block of tests - in3-3.* - verify that column affinity is
sl@0: # correctly handled in cases where an index might be used to optimise
sl@0: # an IN (SELECT) expression.
sl@0: #
sl@0: do_test in3-3.1 {
sl@0:   catch {execsql {
sl@0:     DROP TABLE t1;
sl@0:     DROP TABLE t2;
sl@0:   }}
sl@0: 
sl@0:   execsql {
sl@0: 
sl@0:     CREATE TABLE t1(a BLOB, b NUMBER ,c TEXT);
sl@0:     CREATE UNIQUE INDEX t1_i1 ON t1(a);        /* no affinity */
sl@0:     CREATE UNIQUE INDEX t1_i2 ON t1(b);        /* numeric affinity */
sl@0:     CREATE UNIQUE INDEX t1_i3 ON t1(c);        /* text affinity */
sl@0: 
sl@0:     CREATE TABLE t2(x BLOB, y NUMBER, z TEXT);
sl@0:     CREATE UNIQUE INDEX t2_i1 ON t2(x);        /* no affinity */
sl@0:     CREATE UNIQUE INDEX t2_i2 ON t2(y);        /* numeric affinity */
sl@0:     CREATE UNIQUE INDEX t2_i3 ON t2(z);        /* text affinity */
sl@0: 
sl@0:     INSERT INTO t1 VALUES(1, 1, 1);
sl@0:     INSERT INTO t2 VALUES('1', '1', '1');
sl@0:   }
sl@0: } {}
sl@0: 
sl@0: do_test in3-3.2 {
sl@0:   # No affinity is applied before comparing "x" and "a". Therefore
sl@0:   # the index can be used (the comparison is false, text!=number).
sl@0:   exec_neph { SELECT x IN (SELECT a FROM t1) FROM t2 }
sl@0: } {0 0}
sl@0: do_test in3-3.3 {
sl@0:   # Logically, numeric affinity is applied to both sides before 
sl@0:   # the comparison.  Therefore it is possible to use index t1_i2.
sl@0:   exec_neph { SELECT x IN (SELECT b FROM t1) FROM t2 }
sl@0: } {0 1}
sl@0: do_test in3-3.4 {
sl@0:   # No affinity is applied before the comparison takes place. Making
sl@0:   # it possible to use index t1_i3.
sl@0:   exec_neph { SELECT x IN (SELECT c FROM t1) FROM t2 }
sl@0: } {0 1}
sl@0: 
sl@0: do_test in3-3.5 {
sl@0:   # Numeric affinity should be applied to each side before the comparison
sl@0:   # takes place. Therefore we cannot use index t1_i1, which has no affinity.
sl@0:   exec_neph { SELECT y IN (SELECT a FROM t1) FROM t2 }
sl@0: } {1 1}
sl@0: do_test in3-3.6 {
sl@0:   # Numeric affinity is applied to both sides before 
sl@0:   # the comparison.  Therefore it is possible to use index t1_i2.
sl@0:   exec_neph { SELECT y IN (SELECT b FROM t1) FROM t2 }
sl@0: } {0 1}
sl@0: do_test in3-3.7 {
sl@0:   # Numeric affinity is applied before the comparison takes place. 
sl@0:   # Making it impossible to use index t1_i3.
sl@0:   exec_neph { SELECT y IN (SELECT c FROM t1) FROM t2 }
sl@0: } {1 1}
sl@0: 
sl@0: #---------------------------------------------------------------------
sl@0: #
sl@0: # Test using a multi-column index.
sl@0: #
sl@0: do_test in3-4.1 {
sl@0:   execsql {
sl@0:     CREATE TABLE t3(a, b, c);
sl@0:     CREATE UNIQUE INDEX t3_i ON t3(b, a);
sl@0:   }
sl@0: 
sl@0:   execsql {
sl@0:     INSERT INTO t3 VALUES(1, 'numeric', 2);
sl@0:     INSERT INTO t3 VALUES(2, 'text', 2);
sl@0:     INSERT INTO t3 VALUES(3, 'real', 2);
sl@0:     INSERT INTO t3 VALUES(4, 'none', 2);
sl@0:   }
sl@0: } {}
sl@0: do_test in3-4.2 {
sl@0:   exec_neph { SELECT 'text' IN (SELECT b FROM t3) }
sl@0: } {0 1}
sl@0: do_test in3-4.3 {
sl@0:   exec_neph { SELECT 'TEXT' COLLATE nocase IN (SELECT b FROM t3) }
sl@0: } {1 1}
sl@0: do_test in3-4.4 {
sl@0:   # A temp table must be used because t3_i.b is not guaranteed to be unique.
sl@0:   exec_neph { SELECT b FROM t3 WHERE b IN (SELECT b FROM t3) }
sl@0: } {1 none numeric real text}
sl@0: do_test in3-4.5 {
sl@0:   execsql { CREATE UNIQUE INDEX t3_i2 ON t3(b) }
sl@0:   exec_neph { SELECT b FROM t3 WHERE b IN (SELECT b FROM t3) }
sl@0: } {0 none numeric real text}
sl@0: do_test in3-4.6 {
sl@0:   execsql { DROP INDEX t3_i2 }
sl@0: } {}
sl@0: 
sl@0: # The following two test cases verify that ticket #2991 has been fixed.
sl@0: #
sl@0: do_test in3-5.1 {
sl@0:   execsql {
sl@0:     CREATE TABLE Folders(
sl@0:       folderid INTEGER PRIMARY KEY, 
sl@0:       parentid INTEGER, 
sl@0:       rootid INTEGER, 
sl@0:       path VARCHAR(255)
sl@0:     );
sl@0:   }
sl@0: } {}
sl@0: do_test in3-5.2 {
sl@0:   catchsql {
sl@0:     DELETE FROM Folders WHERE folderid IN
sl@0:     (SELECT folderid FROM Folder WHERE path LIKE 'C:\MP3\Albums\' || '%');
sl@0:   }
sl@0: } {1 {no such table: Folder}}
sl@0: 
sl@0: finish_test