Do a better job of choosing the join table order when the tables having

very different numbers of rows.

FossilOrigin-Name: 952f5e8c69904c48f2decfabf8ea60a2e9f3e134

diff --git a/manifest b/manifest
index eb8af982a530e7d2506aecf43339c3c50688c2fb..fe1dc68240291c0104b80d84abdab2bfe4b8bd17 100644 (file)
--- a/manifest
+++ b/manifest
@@ -1,8 +1,8 @@
 -----BEGIN PGP SIGNED MESSAGE-----
 Hash: SHA1
 
-C Fix\squoting\sof\sthe\sresult\sin\srtreeB.test.
-D 2011-03-02T22:07:29.984
+C Do\sa\sbetter\sjob\sof\schoosing\sthe\sjoin\stable\sorder\swhen\sthe\stables\shaving\nvery\sdifferent\snumbers\sof\srows.
+D 2011-03-04T00:56:58.067
 F Makefile.arm-wince-mingw32ce-gcc d6df77f1f48d690bd73162294bbba7f59507c72f
 F Makefile.in 27701a1653595a1f2187dc61c8117e00a6c1d50f
 F Makefile.linux-gcc 91d710bdc4998cb015f39edf3cb314ec4f4d7e23
@@ -245,7 +245,7 @@ F src/vtab.c b297e8fa656ab5e66244ab15680d68db0adbec30
 F src/wal.c 5386fb5e13c2daa8ab9062597fdc17bd849da371
 F src/wal.h 7a5fbb00114b7f2cd40c7e1003d4c41ce9d26840
 F src/walker.c 3112bb3afe1d85dc52317cb1d752055e9a781f8f
-F src/where.c bf8c0f584a2c9becb9110c37e151f9d4f73fbf7e
+F src/where.c d7b073867e7bab0b851c659ad4899ce72cc9b81d
 F test/aggerror.test a867e273ef9e3d7919f03ef4f0e8c0d2767944f2
 F test/alias.test 4529fbc152f190268a15f9384a5651bbbabc9d87
 F test/all.test 51756962d522e474338e9b2ebb26e7364d4aa125
@@ -259,6 +259,7 @@ F test/analyze2.test 8f2b1534d43f5547ce9a6b736c021d4192c75be3
 F test/analyze3.test d61f55d8b472fc6e713160b1e577f7a68e63f38b
 F test/analyze4.test 757b37875cf9bb528d46f74497bc789c88365045
 F test/analyze5.test 0618d2fe8982a5dae1d4e92152acc8ecbaf52be2
+F test/analyze6.test 1ba1aea8fad25a77ffd71f24522d1bb9ecc949fc
 F test/async.test ad4ba51b77cd118911a3fe1356b0809da9c108c3
 F test/async2.test bf5e2ca2c96763b4cba3d016249ad7259a5603b6
 F test/async3.test 93edaa9122f498e56ea98c36c72abc407f4fb11e
@@ -913,14 +914,14 @@ F tool/speedtest2.tcl ee2149167303ba8e95af97873c575c3e0fab58ff
 F tool/speedtest8.c 2902c46588c40b55661e471d7a86e4dd71a18224
 F tool/speedtest8inst1.c 293327bc76823f473684d589a8160bde1f52c14e
 F tool/vdbe-compress.tcl d70ea6d8a19e3571d7ab8c9b75cba86d1173ff0f
-P ec55e8c6bb4f2419b3813aa2fd1a20d8f5016159
-R 4466b84201cf07d8b58b7f3df32e8610
+P c6532b35cc5a81cc753a22e0165bfdd5143941f1
+R a3d7e872cd52f3bb8cff4bf51476ab7a
 U drh
-Z e51cd0af4553d3bdc1671b46c3024fc0
+Z 87aba3dd3051578a2e8c32390f315773
 -----BEGIN PGP SIGNATURE-----
 Version: GnuPG v1.4.6 (GNU/Linux)
 
-iD8DBQFNbr+loxKgR168RlERAifFAJ4rxcZg2GHVUmhY14d6bC3+WoLAOQCfYvQK
-lkhmZIcwz4BBbhXLINTF0CM=
-=JloW
+iD8DBQFNcDjeoxKgR168RlERAn2TAJ9XKohvD0kx5t1CAQmiI1sUD+1b8QCdFh0P
+RilThkZz6kdLZmkRZY4ir5s=
+=inuw
 -----END PGP SIGNATURE-----

diff --git a/manifest.uuid b/manifest.uuid
index 94178786c021f6cc24e6ebf9fd6e4e9e0b4887d6..b5f246c15a0bcb5504edaa3d6aab9edb3064fd79 100644 (file)
--- a/manifest.uuid
+++ b/manifest.uuid
@@ -1 +1 @@
-c6532b35cc5a81cc753a22e0165bfdd5143941f1
\ No newline at end of file
+952f5e8c69904c48f2decfabf8ea60a2e9f3e134
\ No newline at end of file

diff --git a/src/where.c b/src/where.c
index 9ad2ad320253afc97bd0c4f2da5b8a7b29bff874..727abdf3f6d23e45ecde5d018b0c06d1df88cfa3 100644 (file)
--- a/src/where.c
+++ b/src/where.c
@@ -4610,8 +4610,8 @@ WhereInfo *sqlite3WhereBegin(
         **   (1) The table must not depend on other tables that have not
         **       yet run.
         **
-        **   (2) A full-table-scan plan cannot supercede another plan unless
-        **       it is an "optimal" plan as defined above.
+        **   (2) A full-table-scan plan cannot supercede indexed plan unless
+        **       the full-table-scan is an "optimal" plan as defined above.
         **
         **   (3) All tables have an INDEXED BY clause or this table lacks an
         **       INDEXED BY clause or this table uses the specific
@@ -4627,6 +4627,7 @@ WhereInfo *sqlite3WhereBegin(
         */
         if( (sCost.used&notReady)==0                       /* (1) */
             && (bestJ<0 || (notIndexed&m)!=0               /* (2) */
+                || (bestPlan.plan.wsFlags & WHERE_NOT_FULLSCAN)==0
                 || (sCost.plan.wsFlags & WHERE_NOT_FULLSCAN)!=0)
             && (nUnconstrained==0 || pTabItem->pIndex==0   /* (3) */
                 || NEVER((sCost.plan.wsFlags & WHERE_NOT_FULLSCAN)!=0))

diff --git a/test/analyze6.test b/test/analyze6.test
new file mode 100644 (file)
index 0000000..515747c
--- /dev/null
+++ b/test/analyze6.test
@@ -0,0 +1,74 @@
+# 2011 March 3
+#
+# The author disclaims copyright to this source code.  In place of
+# a legal notice, here is a blessing:
+#
+#    May you do good and not evil.
+#    May you find forgiveness for yourself and forgive others.
+#    May you share freely, never taking more than you give.
+#
+#***********************************************************************
+#
+# This file implements tests for SQLite library.  The focus of the tests
+# in this file a corner-case query planner optimization involving the
+# join order of two tables of different sizes.
+#
+
+set testdir [file dirname $argv0]
+source $testdir/tester.tcl
+
+ifcapable !stat2 {
+  finish_test
+  return
+}
+
+set testprefix analyze6
+
+proc eqp {sql {db db}} {
+  uplevel execsql [list "EXPLAIN QUERY PLAN $sql"] $db
+}
+
+do_test analyze6-1.0 {
+  db eval {
+    CREATE TABLE cat(x INT);
+    CREATE UNIQUE INDEX catx ON cat(x);
+    /* Give cat 16 unique integers */
+    INSERT INTO cat VALUES(1);
+    INSERT INTO cat VALUES(2);
+    INSERT INTO cat SELECT x+2 FROM cat;
+    INSERT INTO cat SELECT x+4 FROM cat;
+    INSERT INTO cat SELECT x+8 FROM cat;
+
+    CREATE TABLE ev(y INT);
+    CREATE INDEX evy ON ev(y);
+    /* ev will hold 32 copies of 16 integers found in cat */
+    INSERT INTO ev SELECT x FROM cat;
+    INSERT INTO ev SELECT x FROM cat;
+    INSERT INTO ev SELECT y FROM ev;
+    INSERT INTO ev SELECT y FROM ev;
+    INSERT INTO ev SELECT y FROM ev;
+    INSERT INTO ev SELECT y FROM ev;
+    ANALYZE;
+    SELECT count(*) FROM cat;
+    SELECT count(*) FROM ev;
+  }
+} {16 512}
+
+# The lowest cost plan is to scan CAT and for each integer there, do a single
+# lookup of the first corresponding entry in EV then read off the equal values
+# in EV.  (Prior to the 2011-03-04 enhancement to where.c, this query would
+# have used EV for the outer loop instead of CAT - which was about 3x slower.)
+#
+do_test analyze6-1.1 {
+  eqp {SELECT count(*) FROM ev, cat WHERE x=y}
+} {0 0 1 {SCAN TABLE cat (~16 rows)} 0 1 0 {SEARCH TABLE ev USING COVERING INDEX evy (y=?) (~32 rows)}}
+
+# The same plan is chosen regardless of the order of the tables in the
+# FROM clause.
+#
+do_test analyze6-1.2 {
+  eqp {SELECT count(*) FROM cat, ev WHERE x=y}
+} {0 0 0 {SCAN TABLE cat (~16 rows)} 0 1 1 {SEARCH TABLE ev USING COVERING INDEX evy (y=?) (~32 rows)}}
+
+
+finish_test