Oracle 11g Linguistic NLS SORT Throws Exception

Question

Hi All,
I am using oracle 11g and when i try NLS_Sort with Generic_m option, throws ORA-00910 exception. See below for more details
Select * From v$version
1 Oracle Database 11g Release 11.1.0.6.0 - Production
2 PL/SQL Release 11.1.0.6.0 - Production
3 CORE 11.1.0.6.0 Production
4 TNS for 32-bit Windows: Version 11.1.0.6.0 - Production
5 NLSRTL Version 11.1.0.6.0 - Production
Select * From v$NLS_PARAMETERS;
1 NLS_LANGUAGE AMERICAN
2 NLS_TERRITORY AMERICA
3 NLS_CURRENCY $
4 NLS_ISO_CURRENCY AMERICA
5 NLS_NUMERIC_CHARACTERS .,
6 NLS_CALENDAR GREGORIAN
7 NLS_DATE_FORMAT DD-MON-RR
8 NLS_DATE_LANGUAGE AMERICAN
9 NLS_CHARACTERSET WE8MSWIN1252
10 NLS_SORT    BINARY
11 NLS_TIME_FORMAT HH.MI.SSXFF AM
12 NLS_TIMESTAMP_FORMAT DD-MON-RR HH.MI.SSXFF AM
13 NLS_TIME_TZ_FORMAT HH.MI.SSXFF AM TZR
14 NLS_TIMESTAMP_TZ_FORMAT DD-MON-RR HH.MI.SSXFF AM TZR
15 NLS_DUAL_CURRENCY $
16 NLS_NCHAR_CHARACTERSET AL16UTF16
17 NLS_COMP    BINARY
18 NLS_LENGTH_SEMANTICS BYTE
19 NLS_NCHAR_CONV_EXCP FALSE
I have one table 'Test' in that column TestDesc varchar2(4000). Now if i retrieve the records by
select sts, it is successfully executed. The same if i do after exec of the below sts, it is
throwing ORA-00910:specified length too long for its datatype.
1. ALTER SESSION SET NLS_SORT=GENERIC_m;
2. Select * From Test order by TestDesc; (It is throwing ORA-00910 exception)
But the same query/setup is executed successfully in oracle 10g.
My question:
Why it is throwing this exception especially in 11g?
Do i need to excute any alter sts like changing of charactor set..
Kindly help me to sove this issue.

Answer 1

Does that mean that 11.2.0.1 (without pse), NLSSORT does not work as intended (documented)?
Correct. At least, if the related information in bug reports is correct.
Is there in 11.2.0.2 still silent truncation instead of "... calculates the collation key for a maximum prefix", if there's a difference?
There is no difference. The term "silent truncation" is not really precise, so the documentation describes the process in a more elaborate way.
Shouldn't docs mention that a "calculated result" could mean that sorts may sometimes be out of sequence (in-exact collation)?
Yes, it should. It is sometimes too easy to assume that things obvious to oneself are also obvious to anybody else ;-) I have asked for the following explanation to be added after the relevant doc paragraph:
"The above behavior implies that two character values whose collation keys (NLSSORT results) are compared to find the linguistic ordering are considered equal if they do not differ in the prefix described above even though they may differ at some further character position. As the NLSSORT function is used implicitly to find linguistic ordering for comparison conditions, BETWEEN condition, IN condition, ORDER BY, GROUP BY, and COUNT(DISTINCT), those operations may return results that are only approximate for long character values. This is a restriction of the current comparison architecture. Currently, the only way to guarantee precise linguistic comparison results is to not compare character values that are longer than 499 characters for monolingual collations and 124 characters for multilingual collations."
-- Sergiusz
Edited by: S. Wolicki, Oracle on May 5, 2011 1:17 PM
Lowered the pessimistic length for multilingual sorts from 249=floor((2000-3)/8) to 124=floor((2000-3)/16)