Oracle regular expression having a hyphen doesn't give same result on Windows as on Unix

问题

I have the following query with a regular expression:

select REGEXP_REPLACE ('TEST 3304 V2', '[`~!@#$%^&*()_+-={}|;.:<>?,./]', ' ') as REG 
from dual;

When executed via SQL*Plus on a Windows machine it returns the following:

SQL>  select REGEXP_REPLACE ('TEST 3304 V2', '[`~!@#$%^&*()_+-={}|;.:<>?,./]', ' ') as REG from dual;

REG
------------
TEST 3304 V2

On a SunOS machine I get a different result:

SQL>  select REGEXP_REPLACE ('TEST 3304 V2', '[`~!@#$%^&*()_+-={}|;.:<>?,./]', ' ') as REG from dual;

REG
------------
TEST      V

These queries were run against the same Oracle server. Is there any reason for that difference in the output?

SQL*Plus version on Windows:

SQL*Plus: Release 11.2.0.1.0 Production on Mar. Oct. 14 15:36:35 2014

Copyright (c) 1982, 2010, Oracle.  All rights reserved.


Connected to:
Oracle Database 11g Enterprise Edition Release 11.2.0.2.0 - 64bit Production
With the Partitioning, OLAP, Data Mining and Real Application Testing options

SQL*Plus version on Unix:

SQL*Plus: Release 11.2.0.2.0 Production on Tue Oct 14 16:01:26 2014

Copyright (c) 1982, 2010, Oracle.  All rights reserved.


Connected to:
Oracle Database 11g Enterprise Edition Release 11.2.0.2.0 - 64bit Production
With the Partitioning, OLAP, Data Mining and Real Application Testing options

回答1:

As Avinash Raj said in comments, the hyphen in your regular expression pattern is being interpreted as a range. The behaviour seems to be dependent on the sorting algorithm being used by the two clients, based on the NLS_LANG environment variable, which influences the NLS_SORT value.

With NLS_LANG=ENGLISH_UNITED KINGDOM.WE8ISO8859P1:

SQL> select REGEXP_REPLACE ('TEST 3304 V2', '[`~!@#$%^&*()_+-={}|;.:<>?,./]', ' ') as REG from dual;

REG
------------
TEST      V

SQL> select value from nls_session_parameters where parameter = 'NLS_SORT';

VALUE
----------
BINARY

Going out on a limb as your profile says you're in Morocco, with NLS_LANG="ARABIC_MOROCCO.AR8MSWIN1256":

SQL> select REGEXP_REPLACE ('TEST 3304 V2', '[`~!@#$%^&*()_+-={}|;.:<>?,./]', ' ') as REG from dual;

REG
------------
TEST 3304 V2

SQL> select value from nls_session_parameters where parameter = 'NLS_SORT';

VALUE
----------
ARABIC

The reason is that the pattern segment +-= is treated as a range covering all characters from + to =. In the ISO8859-1 and Windows 1252 character set that is characters 43 to 61, and all the numeric digits fall within that range - zero is 48 for example - are within that range, so the regex replaces them. That is also true in the Windows 1256 character set. (And anything based on ASCII).

But your NLS_LANG is also implicitly changing the sort order, and it's switch from BINARY to ARABIC sorting that changes the behaviour. You can see that within a single session; with NLS_LANG=ENGLISH_UNITED KINGDOM.WE8ISO8859P1:

SQL> select REGEXP_REPLACE ('TEST 3304 V2', '[`~!@#$%^&*()_+-={}|;.:<>?,./]', ' ') as REG from dual;

REG
------------
TEST      V

SQL> alter session set NLS_SORT=ARABIC;

Session altered.

SQL> select REGEXP_REPLACE ('TEST 3304 V2', '[`~!@#$%^&*()_+-={}|;.:<>?,./]', ' ') as REG from dual;

REG
------------
TEST 3304 V2

You can also tell that it is a range issue by slightly modifying the range; changing +-= to +-3 so higher digits are not included, but leaving everything else the same:

SQL> alter session set NLS_SORT=BINARY;

Session altered.

SQL> select REGEXP_REPLACE ('TEST 3304 V2', '[`~!@#$%^&*()_+-3{}|;.:<>?,./]', ' ') as REG from dual;

REG
------------
TEST    4 V