Re: nls_sort on index definition causes index not to be used

  • From: Martin Berger <martin.a.berger@xxxxxxxxx>
  • To: Jonathan Lewis <jonathan@xxxxxxxxxxxxxxxxxx>
  • Date: Wed, 22 Feb 2017 20:16:17 +0100

I was sceptical about nls_sort=punctuation on numbers as well, but there
seems to be a difference:
https://livesql.oracle.com/apex/livesql/file/content_EL52ZTYR9GABVT909FLUJJZ8A.html

(even I can't imagine a business case for this ...)

Martin

2017-02-22 19:30 GMT+01:00 Jonathan Lewis <jonathan@xxxxxxxxxxxxxxxxxx>:


The problem is the nlssort() - you've got function-based indexes and
aren't applying the same function when you query the data.
Is employee_id a character type ? "_ID" commonly indicates numeric and it
looks really strange to apply nlssort() to a number:

Example of workaround - old scott.emp table:

SQL> create index e_i1 on emp(nlssort(sal,'nls_sort=''punctuation'''));

set autotrace traceonly explain:

SQL> select * from emp where sal = 1100;

Execution Plan
----------------------------------------------------------
Plan hash value: 3956160932

--------------------------------------------------------------------------
| Id  | Operation         | Name | Rows  | Bytes | Cost (%CPU)| Time     |
--------------------------------------------------------------------------
|   0 | SELECT STATEMENT  |      |     1 |    38 |    18   (0)| 00:00:01 |
|*  1 |  TABLE ACCESS FULL| EMP  |     1 |    38 |    18   (0)| 00:00:01 |
--------------------------------------------------------------------------

Predicate Information (identified by operation id):
---------------------------------------------------

   1 - filter("SAL"=1100)



SQL> select * from emp where nlssort(sal,'nls_sort=''punctuation''') =
nlssort(1100,'nls_sort=''punctuation''');

Execution Plan
----------------------------------------------------------
Plan hash value: 2872601085

------------------------------------------------------------
------------------------
| Id  | Operation                   | Name | Rows  | Bytes | Cost (%CPU)|
Time     |
------------------------------------------------------------
------------------------
|   0 | SELECT STATEMENT            |      |     1 |    38 |     2   (0)|
00:00:01 |
|   1 |  TABLE ACCESS BY INDEX ROWID| EMP  |     1 |    38 |     2   (0)|
00:00:01 |
|*  2 |   INDEX RANGE SCAN          | E_I1 |     1 |       |     1   (0)|
00:00:01 |
------------------------------------------------------------
------------------------

Predicate Information (identified by operation id):
---------------------------------------------------

   2 - access(NLSSORT(TO_CHAR("SAL"),'nls_sort=''PUNCTUATION''')=
HEXTORAW('A
              1A1A0A0000101010100') )


Regards
Jonathan Lewis


________________________________________
From: oracle-l-bounce@xxxxxxxxxxxxx <oracle-l-bounce@xxxxxxxxxxxxx> on
behalf of Beckstrom Jeffrey <JBECKSTROM@xxxxxxxxx>
Sent: 22 February 2017 17:47:25
To: oracle-l-freelists; oracle-db-l
Subject: nls_sort on index definition causes index not to be used

We have a 3rd party application for which indexes are built with
nls_sort=punctuation on some of the columns. From SQL Plus a simple select
on the column is using a full table scan. With a hint, it uses a full scan
of the index. I presume the problem is the nls_sort.  Example:
                                         U
N prefix
INDEX_NAME OWNER Q length COLUMN_POSITION COLUMN_NAME COLUMN_EXPRESSION
------------------------------ --------- - ---------- ---------------
------------------------------ --------------------
EMPLOYEE_IDX1 xxxxxx    Y 1 SYS_NC00042$ NLSSORT("EMPLOYEE_ID
",'nls_sort=''PUNCTU
ATION''')

Can someone tell me why?
--
//www.freelists.org/webpage/oracle-l





-- 
Martin Berger         +43 660 2978929 <+436602978929>
martin.a.berger@xxxxxxxxx @martinberx <https://twitter.com/martinberx>
^∆x      http://berxblog.blogspot.com

Other related posts: