Strange expdp errors

Question

I've exported schema from database with CL8MSWIN1251 encoding and now I try to import it into database with AL32UTF8 character encoding. DBMS version is 10.2.0.4.0.
At first, I've set nls_length_semantics to char in database.
After that I've restarted instance and imported metadata from schema.
But on importing data I'm getting a lot of the following errors :
ORA-02374: conversion error loading table "SFEDU_TEST"."ELEMS"
ORA-12899: value too large for column ES_COMMENT (actual: 2456, maximum: 4000)
ORA-02372: data for row: ES_COMMENT : 0X'200D0A09CAE0F4E5E4F0E020F1EEE7E4E0EDE020E220313939'
It's interesting that Oracle says: "value too large for column ES_COMMENT (actual: 2456, maximum: 4000)", so maximum size is more than actual...
How can I deal with this issue?
Edited by: alp on Apr 29, 2009 4:48 AM

Answer 1

There may be two errors here:
1. The actual data expansion issue. Setting NLS_LENGTH_SEMANTICS to CHAR does not change the semantics of objects recreated by import. Import uses explicit semantics specifications per column and, therefore, does not use the default value set by NLS_LENGTH_SEMANTICS.
What does DESCRIBE "SFEDU_TEST"."ELEMS" show in the target database?To deal with this error I've run thr following code:
begin
for cur_col in ( select table_name,column_name, DATA_TYPE,CHAR_COL_DECL_LENGTH,DATA_DEFAULT from user_tab_cols T, user_objects O where DATA_TYPE='VARCHAR2' and O.object_name=T.table_name and O.OBJECT_TYPE='TABLE' and T.column_name not like 'SYS_%' and column_name not in ('EDP_KURS','EDP_SNSP_SHIFR','EDP_SNP_SHIFR'))
loop
begin
     if cur_col.DATA_DEFAULT IS NULL then
     begin
          dbms_output.put_line('alter table ' || cur_col.table_name||' modify ' || cur_col.column_name || ' VARCHAR2(' || cur_col.CHAR_COL_DECL_LENGTH ||' CHAR )');
          execute immediate 'alter table ' || cur_col.table_name||' modify ' || cur_col.column_name || ' VARCHAR2(' || cur_col.CHAR_COL_DECL_LENGTH ||' CHAR )';
     end;
     else
     begin
          dbms_output.put_line('alter table ' || cur_col.table_name||' modify ' || cur_col.column_name || ' VARCHAR2(' || cur_col.CHAR_COL_DECL_LENGTH ||' CHAR ) default '||cur_col.data_default);
          execute immediate 'alter table ' || cur_col.table_name||' modify ' || cur_col.column_name || ' VARCHAR2(' || cur_col.CHAR_COL_DECL_LENGTH ||' CHAR ) default '||cur_col.data_default;          
     end;
     end if ;
end;
end loop;
end;
now descr says the following:
descr sfedu_test.elems
Name                         Null? Type
ES_ID                              NUMBER(16)
ES_PHONE                         VARCHAR2(4000 CHAR)
ES_URL                          VARCHAR2(4000 CHAR)
ES_EMAIL                         VARCHAR2(4000 CHAR)
ES_COMMENT                         VARCHAR2(4000 CHAR)
EST_EST_ID                         NUMBER(10)
ES_ES_ID                         NUMBER(16)
ES_EDU                          NUMBER(1)
ES_ADR                          VARCHAR2(4000 CHAR)
ES_PHOTO                         VARCHAR2(4000 CHAR)
but I still have :
KUP-11007: conversion error loading table "SFEDU_TEST"."ELEMS"
ORA-12899: value too large for column ES_COMMENT (actual: 2446, maximum: 4000)
KUP-11009: data for row: ES_COMMENT : 0X'CAE0F4E5E4F0E020FDEAEEEBEEE3E8E820E820EFF0E8F0EEE4'
>
2. There is a bug in the specific place where the error message ORA-12899 is generated in your case (this message comes from many places in code).
As the message is most probably generated in the server code, you could try to issue:
ALTER SYSTEM SET EVENTS '12899 trace name errorstack level 1, forever';before starting the import. This should generate a trace (*.trc) file in the udump or bdump directory with ORA-12899 and the associated call stack. If you can get the stack, paste it here.There is no new trace files in udump,bdump after enabling tracing...