用于EagleEye3.0 规则集漏报和误报测试的示例项目,项目收集于github和gitee
You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
 
 
 
 
 
 

202 lines
4.2 KiB

--source include/no_valgrind_without_big.inc
--echo #
--echo # WL#7755 mysqlpump: Extend mysqldump functionalities
--echo #
#Not with ps-protocol
--source include/no_ps_protocol.inc
#
# Check all charsets/collation combinations
#
let $check_std_csets= 1;
if ($check_std_csets)
{
# single byte 8 bit encoding charset
# cp1251
let $cset= cp1251;
let $coll= cp1251_bin;
--source include/check_charset.inc
# variable width multi byte encoding charset
# euckr
let $cset= euckr;
let $coll= euckr_bin;
--source include/check_charset.inc
# eucjpms
let $cset= eucjpms;
let $coll= eucjpms_bin;
--source include/check_charset.inc
# gb18030
let $cset= gb18030;
let $coll= gb18030_bin;
--source include/check_charset.inc
let $cset= cp932;
let $coll= cp932_bin;
--source include/check_charset.inc
}
let $check_std_csets= 0;
let $check_ucs2_csets=1;
if ($check_ucs2_csets)
{
# ucs2
let $cset= ucs2;
let $coll= ucs2_bin;
--source include/check_charset.inc
let $cset= ucs2;
let $coll= ucs2_unicode_ci;
--source include/check_charset.inc
}
let $check_std_csets= 0;
let $check_ucs2_csets= 0;
let $check_utf8_csets= 1;
if ($check_utf8_csets)
{
# utf8
let $cset= utf8;
let $coll= utf8_bin;
--source include/check_charset.inc
let $cset= utf8;
let $coll= utf8_unicode_ci;
--source include/check_charset.inc
}
#
# Test for non ascii characters for object names
#
let $check_std_csets= 0;
let $check_ucs2_csets= 0;
let $check_utf8_csets= 0;
let $check_for_object_names=1;
if ($check_for_object_names)
{
# cp1251
let $cset= cp1251;
let $coll= cp1251_bulgarian_ci;
--source include/check_charset.inc
# euckr
let $cset= euckr;
let $coll= euckr_korean_ci;
--source include/check_charset.inc
# gb18030
let $cset= gb18030;
let $coll= gb18030_bin;
--source include/check_charset.inc
# ujis
let $cset= ujis;
let $coll= ujis_japanese_ci;
--source include/check_charset.inc
# sjis
let $cset= sjis;
let $coll= sjis_japanese_ci;
--source include/check_charset.inc
}
# Misc tests
CREATE DATABASE db1_charset;
USE db1_charset;
CREATE TABLE t_latin1(c CHAR(40)) ENGINE=InnoDB DEFAULT CHARSET=latin1;
INSERT INTO t_latin1 VALUES ('aaa'), ('ÁÂÃÄ');
INSERT INTO t_latin1 VALUES (_latin1'ςσ');
CREATE TABLE t_utf8(c CHAR(40)) ENGINE=InnoDB DEFAULT CHARSET=utf8;
INSERT INTO t_utf8 VALUES ('aaa'), ('ÁÂÃÄ');
CREATE TABLE t_koi8r (c1 VARBINARY(255), c2 VARBINARY(255));
SET CHARACTER_SET_CLIENT=koi8r,
CHARACTER_SET_CONNECTION=cp1251,
CHARACTER_SET_RESULTS=koi8r;
INSERT INTO t_koi8r (c1, c2) VALUES ('îÕ, ÚÁ ÒÙÂÁÌËÕ','îÕ, ÚÁ ÒÙÂÁÌËÕ');
SET NAMES UTF8;
SELECT * FROM t_latin1 ORDER BY 1;
SELECT * FROM t_utf8 ORDER BY 1;
SELECT hex(c1), hex(c2) from t_koi8r;
--exec $MYSQL_PUMP --databases db1_charset > $MYSQLTEST_VARDIR/tmp/db1_charset.sql
DROP DATABASE db1_charset;
CREATE DATABASE db1_charset;
--exec $MYSQL db1_charset < $MYSQLTEST_VARDIR/tmp/db1_charset.sql
--remove_file $MYSQLTEST_VARDIR/tmp/db1_charset.sql
USE db1_charset;
SET NAMES UTF8;
SELECT * FROM t_latin1 ORDER BY 1;
SELECT * FROM t_utf8 ORDER BY 1;
SELECT hex(c1), hex(c2) FROM t_koi8r ORDER BY 1;
DROP DATABASE db1_charset;
CREATE DATABASE db1_charset;
USE db1_charset;
CREATE TABLE t_allcharsets
(ucs2 CHAR(40) character set ucs2,
utf8 CHAR(40) character set utf8,
big5 CHAR(40) character set big5,
cp932 CHAR(40) character set cp932,
eucjpms CHAR(40) character set eucjpms,
euckr CHAR(40) character set euckr,
gb2312 CHAR(40) character set gb2312,
gbk CHAR(40) character set gbk,
sjis CHAR(40) character set sjis,
ujis CHAR(40) character set ujis);
INSERT INTO t_allcharsets (ucs2) VALUES (0x30da);
UPDATE t_allcharsets
SET utf8=ucs2,
big5=ucs2,
cp932=ucs2,
eucjpms=ucs2,
euckr=ucs2,
gb2312=ucs2,
gbk=ucs2,
sjis=ucs2,
ujis=ucs2;
SELECT * FROM t_allcharsets;
--exec $MYSQL_PUMP --databases db1_charset > $MYSQLTEST_VARDIR/tmp/db1_charset.sql
DROP DATABASE db1_charset;
--exec $MYSQL < $MYSQLTEST_VARDIR/tmp/db1_charset.sql
--remove_file $MYSQLTEST_VARDIR/tmp/db1_charset.sql
USE db1_charset;
SELECT * FROM t_allcharsets;
DROP DATABASE db1_charset;