rt-thread/components/external/SQLite-3.8.1/test/select3.test

265 lines
6.8 KiB
Plaintext

# 2001 September 15
#
# 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 regression tests for SQLite library. The
# focus of this file is testing aggregate functions and the
# GROUP BY and HAVING clauses of SELECT statements.
#
# $Id: select3.test,v 1.23 2008/01/16 18:20:42 danielk1977 Exp $
set testdir [file dirname $argv0]
source $testdir/tester.tcl
# Build some test data
#
do_test select3-1.0 {
execsql {
CREATE TABLE t1(n int, log int);
BEGIN;
}
for {set i 1} {$i<32} {incr i} {
for {set j 0} {(1<<$j)<$i} {incr j} {}
execsql "INSERT INTO t1 VALUES($i,$j)"
}
execsql {
COMMIT
}
execsql {SELECT DISTINCT log FROM t1 ORDER BY log}
} {0 1 2 3 4 5}
# Basic aggregate functions.
#
do_test select3-1.1 {
execsql {SELECT count(*) FROM t1}
} {31}
do_test select3-1.2 {
execsql {
SELECT min(n),min(log),max(n),max(log),sum(n),sum(log),avg(n),avg(log)
FROM t1
}
} {1 0 31 5 496 124 16.0 4.0}
do_test select3-1.3 {
execsql {SELECT max(n)/avg(n), max(log)/avg(log) FROM t1}
} {1.9375 1.25}
# Try some basic GROUP BY clauses
#
do_test select3-2.1 {
execsql {SELECT log, count(*) FROM t1 GROUP BY log ORDER BY log}
} {0 1 1 1 2 2 3 4 4 8 5 15}
do_test select3-2.2 {
execsql {SELECT log, min(n) FROM t1 GROUP BY log ORDER BY log}
} {0 1 1 2 2 3 3 5 4 9 5 17}
do_test select3-2.3.1 {
execsql {SELECT log, avg(n) FROM t1 GROUP BY log ORDER BY log}
} {0 1.0 1 2.0 2 3.5 3 6.5 4 12.5 5 24.0}
do_test select3-2.3.2 {
execsql {SELECT log, avg(n)+1 FROM t1 GROUP BY log ORDER BY log}
} {0 2.0 1 3.0 2 4.5 3 7.5 4 13.5 5 25.0}
do_test select3-2.4 {
execsql {SELECT log, avg(n)-min(n) FROM t1 GROUP BY log ORDER BY log}
} {0 0.0 1 0.0 2 0.5 3 1.5 4 3.5 5 7.0}
do_test select3-2.5 {
execsql {SELECT log*2+1, avg(n)-min(n) FROM t1 GROUP BY log ORDER BY log}
} {1 0.0 3 0.0 5 0.5 7 1.5 9 3.5 11 7.0}
do_test select3-2.6 {
execsql {
SELECT log*2+1 as x, count(*) FROM t1 GROUP BY x ORDER BY x
}
} {1 1 3 1 5 2 7 4 9 8 11 15}
do_test select3-2.7 {
execsql {
SELECT log*2+1 AS x, count(*) AS y FROM t1 GROUP BY x ORDER BY y, x
}
} {1 1 3 1 5 2 7 4 9 8 11 15}
do_test select3-2.8 {
execsql {
SELECT log*2+1 AS x, count(*) AS y FROM t1 GROUP BY x ORDER BY 10-(x+y)
}
} {11 15 9 8 7 4 5 2 3 1 1 1}
#do_test select3-2.9 {
# catchsql {
# SELECT log, count(*) FROM t1 GROUP BY 'x' ORDER BY log;
# }
#} {1 {GROUP BY terms must not be non-integer constants}}
do_test select3-2.10 {
catchsql {
SELECT log, count(*) FROM t1 GROUP BY 0 ORDER BY log;
}
} {1 {1st GROUP BY term out of range - should be between 1 and 2}}
do_test select3-2.11 {
catchsql {
SELECT log, count(*) FROM t1 GROUP BY 3 ORDER BY log;
}
} {1 {1st GROUP BY term out of range - should be between 1 and 2}}
do_test select3-2.12 {
catchsql {
SELECT log, count(*) FROM t1 GROUP BY 1 ORDER BY log;
}
} {0 {0 1 1 1 2 2 3 4 4 8 5 15}}
# Cannot have an empty GROUP BY
do_test select3-2.13 {
catchsql {
SELECT log, count(*) FROM t1 GROUP BY ORDER BY log;
}
} {1 {near "ORDER": syntax error}}
do_test select3-2.14 {
catchsql {
SELECT log, count(*) FROM t1 GROUP BY;
}
} {1 {near ";": syntax error}}
# Cannot have a HAVING without a GROUP BY
#
do_test select3-3.1 {
set v [catch {execsql {SELECT log, count(*) FROM t1 HAVING log>=4}} msg]
lappend v $msg
} {1 {a GROUP BY clause is required before HAVING}}
# Toss in some HAVING clauses
#
do_test select3-4.1 {
execsql {SELECT log, count(*) FROM t1 GROUP BY log HAVING log>=4 ORDER BY log}
} {4 8 5 15}
do_test select3-4.2 {
execsql {
SELECT log, count(*) FROM t1
GROUP BY log
HAVING count(*)>=4
ORDER BY log
}
} {3 4 4 8 5 15}
do_test select3-4.3 {
execsql {
SELECT log, count(*) FROM t1
GROUP BY log
HAVING count(*)>=4
ORDER BY max(n)+0
}
} {3 4 4 8 5 15}
do_test select3-4.4 {
execsql {
SELECT log AS x, count(*) AS y FROM t1
GROUP BY x
HAVING y>=4
ORDER BY max(n)+0
}
} {3 4 4 8 5 15}
do_test select3-4.5 {
execsql {
SELECT log AS x FROM t1
GROUP BY x
HAVING count(*)>=4
ORDER BY max(n)+0
}
} {3 4 5}
do_test select3-5.1 {
execsql {
SELECT log, count(*), avg(n), max(n+log*2) FROM t1
GROUP BY log
ORDER BY max(n+log*2)+0, avg(n)+0
}
} {0 1 1.0 1 1 1 2.0 4 2 2 3.5 8 3 4 6.5 14 4 8 12.5 24 5 15 24.0 41}
do_test select3-5.2 {
execsql {
SELECT log, count(*), avg(n), max(n+log*2) FROM t1
GROUP BY log
ORDER BY max(n+log*2)+0, min(log,avg(n))+0
}
} {0 1 1.0 1 1 1 2.0 4 2 2 3.5 8 3 4 6.5 14 4 8 12.5 24 5 15 24.0 41}
# Test sorting of GROUP BY results in the presence of an index
# on the GROUP BY column.
#
do_test select3-6.1 {
execsql {
SELECT log, min(n) FROM t1 GROUP BY log ORDER BY log;
}
} {0 1 1 2 2 3 3 5 4 9 5 17}
do_test select3-6.2 {
execsql {
SELECT log, min(n) FROM t1 GROUP BY log ORDER BY log DESC;
}
} {5 17 4 9 3 5 2 3 1 2 0 1}
do_test select3-6.3 {
execsql {
SELECT log, min(n) FROM t1 GROUP BY log ORDER BY 1;
}
} {0 1 1 2 2 3 3 5 4 9 5 17}
do_test select3-6.4 {
execsql {
SELECT log, min(n) FROM t1 GROUP BY log ORDER BY 1 DESC;
}
} {5 17 4 9 3 5 2 3 1 2 0 1}
do_test select3-6.5 {
execsql {
CREATE INDEX i1 ON t1(log);
SELECT log, min(n) FROM t1 GROUP BY log ORDER BY log;
}
} {0 1 1 2 2 3 3 5 4 9 5 17}
do_test select3-6.6 {
execsql {
SELECT log, min(n) FROM t1 GROUP BY log ORDER BY log DESC;
}
} {5 17 4 9 3 5 2 3 1 2 0 1}
do_test select3-6.7 {
execsql {
SELECT log, min(n) FROM t1 GROUP BY log ORDER BY 1;
}
} {0 1 1 2 2 3 3 5 4 9 5 17}
do_test select3-6.8 {
execsql {
SELECT log, min(n) FROM t1 GROUP BY log ORDER BY 1 DESC;
}
} {5 17 4 9 3 5 2 3 1 2 0 1}
# Sometimes an aggregate query can return no rows at all.
#
do_test select3-7.1 {
execsql {
CREATE TABLE t2(a,b);
INSERT INTO t2 VALUES(1,2);
SELECT a, sum(b) FROM t2 WHERE b=5 GROUP BY a;
}
} {}
do_test select3-7.2 {
execsql {
SELECT a, sum(b) FROM t2 WHERE b=5;
}
} {{} {}}
# If a table column is of type REAL but we are storing integer values
# in it, the values are stored as integers to take up less space. The
# values are converted by to REAL as they are read out of the table.
# Make sure the GROUP BY clause does this conversion correctly.
# Ticket #2251.
#
do_test select3-8.1 {
execsql {
CREATE TABLE A (
A1 DOUBLE,
A2 VARCHAR COLLATE NOCASE,
A3 DOUBLE
);
INSERT INTO A VALUES(39136,'ABC',1201900000);
INSERT INTO A VALUES(39136,'ABC',1207000000);
SELECT typeof(sum(a3)) FROM a;
}
} {real}
do_test select3-8.2 {
execsql {
SELECT typeof(sum(a3)) FROM a GROUP BY a1;
}
} {real}
finish_test