Like returns rows which do not macht if using NLS_SORT='BINARY_AI'

Question

Hi,
I have a table with a nvarchar column in Oracle Xe 11. Using a select with like rows are returned which do not match the like condition. E.g. 'CR_abc' LIKE 'CR,%' is true for NLS_SORT=BINARY_AI. For GERMAN_AI the condition is false!
Example to reproduce:
set LINESIZE 200
ALTER SESSION SET NLS_COMP='LINGUISTIC';
ALTER SESSION SET NLS_SORT='BINARY_AI';
CREATE TABLE "LIBFIELDS_CONFIGF"  ("id1" NVARCHAR2(254) NOT NULL ENABLE, "id2" VARCHAR2(254) NOT NULL ENABLE );
INSERT INTO "LIBFIELDS_CONFIGF" ("id1", "id2") VALUES ('CR_abc', 'CR_abc');
select "id1" from libfields_configf where "id1" like 'CR,%';
EXPLAIN PLAN FOR select "id1" from libfields_configf where "id1" like 'CR,%';
SELECT * FROM TABLE(dbms_xplan.display);
select "id2" from libfields_configf where "id2" like 'CR,%';
EXPLAIN PLAN FOR select "id2" from libfields_configf where "id2" like 'CR,%';
SELECT * FROM TABLE(dbms_xplan.display);
DROP TABLE libfields_configf;This does not work on the nvarchar column (plan output)
   1 - filter(NLSSORT("id1",'nls_sort=''BINARY_AI''')>=HEXTORAW('00630072002C00
              ')  AND NLSSORT("id1",'nls_sort=''BINARY_AI''')<HEXTORAW('63722D00') )This works correctly on the varchar column (plan output)
   1 - filter(NLSSORT("id2",'nls_sort=''BINARY_AI''')>=HEXTORAW('63722C00')
              AND NLSSORT("id2",'nls_sort=''BINARY_AI''')<HEXTORAW('63722D00') )Switching NLS_SORT to GERMAN_AI produces the correct result for both queries. Is there a fix for this available or do I missunderstand the behaviour of NLS_SORT for BINARY_AI and GERMAN_AI?
Thx Jürgen

Answer 1

I'm not totally clear on what you're trying to do, given the like 'CR,%' pattern match, given that the sample data has the wildcard '_' character in it.
And trying a few different NLS_COMP and SORT settings, I'm also seeing unexpected matching. You're probably more informed on the diacriticals and how they should sort and match for different locales, americanamerica.<charset> NLS settings don't distinguish much beyond an ASCII binary sort, i.e. does 'ü' precede or follow 'u', or maybe even precede 'ă' ... ?
A dump( <col>,16) can also help especially when nvarchar character values, and the unistr() to get a few more entries for comparison are helpful too:
insert into <table> values ( unistr('\00c7R_abc'),unistr('\00c7R_abc') ); -- capital C cedille
insert into <table> values ( unistr('\010dR_abc'),unistr('\010dR_abc') ); -- lower C caron
insert into <table> values ( unistr('\0106R_abc'),unistr('\0106R_abc') ); -- capital C acute
select "id1", "id2",dump( "id1", 16) as dump_id1, dump( "id2",16 ) as dump_id2 from <table> where "id1" like 'C%';
alter session set nls_sort = 'german_ci';
... <repeat select>
alter session set nls_sort = 'binary_ai';
... <repeat select>
alter session set nls_sort = 'binary';BTW, case sensitive column names are generally frowned upon, it makes for extra keystrokes as well as a possible source of confusion. Does one want the "Emp", "EMP", or "EmP" table, they should all be the same but won't be when double quoted.
Edited by: clcarter on Oct 20, 2011 6:52 PM
fix quoting