test/analyze8.test

# 2011 August 13
#
# 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 is testing the capabilities of sqlite_stat3.
#

set testdir [file dirname $argv0]
source $testdir/tester.tcl

ifcapable !stat4&&!stat3 {
  finish_test
  return
}

set testprefix analyze8

proc eqp {sql {db db}} {
  uplevel execsql [list "EXPLAIN QUERY PLAN $sql"] $db
}

# Scenario:
#
#    Two indices.  One has mostly singleton entries, but for a few
#    values there are hundreds of entries.  The other has 10-20
#    entries per value.
#
# Verify that the query planner chooses the first index for the singleton
# entries and the second index for the others.
#
do_test 1.0 {
  db eval {
    CREATE TABLE t1(a,b,c,d);
    CREATE INDEX t1a ON t1(a);
    CREATE INDEX t1b ON t1(b);
    CREATE INDEX t1c ON t1(c);
  }
  for {set i 0} {$i<1000} {incr i} {
    if {$i%2==0} {set a $i} {set a [expr {($i%8)*100}]}
    set b [expr {$i/10}]
    set c [expr {$i/8}]
    set c [expr {$c*$c*$c}]
    db eval {INSERT INTO t1 VALUES($a,$b,$c,$i)}
  }
  db eval {ANALYZE}
} {}

# The a==100 comparison is expensive because there are many rows
# with a==100.  And so for those cases, choose the t1b index.
#
# Buf ro a==99 and a==101, there are far fewer rows so choose
# the t1a index.
#
do_test 1.1 {
  eqp {SELECT * FROM t1 WHERE a=100 AND b=55}
} {0 0 0 {SEARCH TABLE t1 USING INDEX t1b (b=?)}}
do_test 1.2 {
  eqp {SELECT * FROM t1 WHERE a=99 AND b=55}
} {0 0 0 {SEARCH TABLE t1 USING INDEX t1a (a=?)}}
do_test 1.3 {
  eqp {SELECT * FROM t1 WHERE a=101 AND b=55}
} {0 0 0 {SEARCH TABLE t1 USING INDEX t1a (a=?)}}
do_test 1.4 {
  eqp {SELECT * FROM t1 WHERE a=100 AND b=56}
} {0 0 0 {SEARCH TABLE t1 USING INDEX t1b (b=?)}}
do_test 1.5 {
  eqp {SELECT * FROM t1 WHERE a=99 AND b=56}
} {0 0 0 {SEARCH TABLE t1 USING INDEX t1a (a=?)}}
do_test 1.6 {
  eqp {SELECT * FROM t1 WHERE a=101 AND b=56}
} {0 0 0 {SEARCH TABLE t1 USING INDEX t1a (a=?)}}
do_test 2.1 {
  eqp {SELECT * FROM t1 WHERE a=100 AND b BETWEEN 50 AND 54}
} {0 0 0 {SEARCH TABLE t1 USING INDEX t1b (b>? AND b<?)}}

# There are many more values of c between 0 and 100000 than there are
# between 800000 and 900000.  So t1c is more selective for the latter
# range.
# 
# Test 3.2 is a little unstable. It depends on the planner estimating
# that (b BETWEEN 30 AND 34) will match more rows than (c BETWEEN
# 800000 AND 900000). Which is a pretty close call (50 vs. 32), so
# the planner could get it wrong with an unlucky set of samples. This
# case happens to work, but others ("b BETWEEN 40 AND 44" for example) 
# will fail.
#
do_execsql_test 3.0 {
  SELECT count(*) FROM t1 WHERE b BETWEEN 30 AND 34;
  SELECT count(*) FROM t1 WHERE c BETWEEN 0 AND 100000;
  SELECT count(*) FROM t1 WHERE c BETWEEN 800000 AND 900000;
} {50 376 32}
do_test 3.1 {
  eqp {SELECT * FROM t1 WHERE b BETWEEN 30 AND 34 AND c BETWEEN 0 AND 100000}
} {0 0 0 {SEARCH TABLE t1 USING INDEX t1b (b>? AND b<?)}}
do_test 3.2 {
  eqp {SELECT * FROM t1
       WHERE b BETWEEN 30 AND 34 AND c BETWEEN 800000 AND 900000}
} {0 0 0 {SEARCH TABLE t1 USING INDEX t1c (c>? AND c<?)}}
do_test 3.3 {
  eqp {SELECT * FROM t1 WHERE a=100 AND c BETWEEN 0 AND 100000}
} {0 0 0 {SEARCH TABLE t1 USING INDEX t1a (a=?)}}
do_test 3.4 {
  eqp {SELECT * FROM t1
       WHERE a=100 AND c BETWEEN 800000 AND 900000}
} {0 0 0 {SEARCH TABLE t1 USING INDEX t1c (c>? AND c<?)}}

finish_test