Unicode characters in file name

Question

Hi,
I try to open a file (using UTL_FILE functionalities) whose name contains polish characters (e.g. 'test-ś.txt').
In return, I get error message:
ORA-29283: invalid file operation
ORA-06512: at "SYS.UTL_FILE", line 633
ORA-29283: invalid file operation
Error is not due to missing rights on file/directory because when I replace the polish character by a latin one, file is opened successfuly.
I also tried to rename a file (using UTL_FILE.FRENAME) from latin to polish characters (e.g 'test-s.txt' -> 'test-ś.txt').
File is renamed but polish characters are lost (final result is something like 'test-Å›.txt').
What's wrong with my environment or code?
Thanks in advance for your help,
Arnaud
Here's my environment description, PL/SQL code and results.
Environment:
OS Windows in Polish for client box
* code page ACP=1250
* NLS_LANG=POLISH_POLAND.EE8MSWIN1250
OS Windows in US/English for database server
* code page ACP=1252
* NLS_LANG=AMERICAN_AMERICA.AL32UTF8
Oracle 10.2.0.5
* NLS_CHARACTERSET=AL32UTF8
* NLS_NCHAR_CHARACTERSET=AL16UTF16
Tests are executed from SQL Developer on client box.
The file I'm trying to open is located on database server.
So, Oracle Directory path used in FOPEN procedure is something like '\\server\directory'.
PL/SQL code:
SET SERVEROUTPUT ON;
declare
Message varchar2(1000);
Filename varchar2(1000); -- nvarchar2(1000);
FileHandler UTL_FILE.FILE_TYPE;
OraDir varchar2(30) := 'SGINSURANCE_DIR_SOURCE';
begin
dbms_output.enable(10000);
--Filename := 'test-s.txt';
Filename := 'test-ś.txt';
Message := 'Opening file ['||Filename||']';
dbms_output.put_line(Message);
--FileHandler := UTL_FILE.FOPEN_NCHAR(OraDir, Filename, 'r');
FileHandler := UTL_FILE.FOPEN(OraDir, Filename, 'r');
Message := 'Closing file';
dbms_output.put_line(Message);
UTL_FILE.FCLOSE(FileHandler);
exception
when others then
Message := 'Error: '||SQLERRM;
dbms_output.put_line(Message);
if UTL_FILE.IS_OPEN(FileHandler) then
Message := 'Closing file ['||Filename||']';
dbms_output.put_line(Message);
UTL_FILE.FCLOSE(FileHandler);
end if;
end;
Results:
Test with polish characters -> error ORA-29283: invalid file operation
anonymous block completed
Opening file [test-ś.txt]
Error: ORA-29283: invalid file operation
ORA-06512: at "SYS.UTL_FILE", line 536
ORA-29283: invalid file operation
Test without polish characters -> no error
anonymous block completed
Opening file [test-s.txt]
Closing file
-----------------------------------------------------------

Answer 1

Hello,
I tested this issue on Oracle-10-XE on Windows-XP with different Language settings.
It seems to me that UTL_FILE doesn't use wide character Windows API functions like _wfopen,
but simply old fopen based on 8-bit character strings.
Looks like UTL_FILE.FOPEN do not any character conversion
on the filename, but pass this filename "as is" directly to the operating system,
for example for a string "teść" with polish characters the following char codes are passed:
SELECT dump( 'teść', 16 ) from dual;
DUMP('TEŚĆ',16)                
Typ=96 Len=6: 74,65,c5,9b,c4,87ś - is : c5, 9b
ć - is : c4 87
In windows API functions based on on 8-bit char * strings are interpreted as being in the system code page
- look at this thread -> [ http://stackoverflow.com/questions/480849/windows-codepage-interactions-with-standard-c-c-filenames]
So if your code page is a Windows ANSI 1252, these characters are treated as:
ś -> c5 is "Å" , 9b is "›" --> Å›
ć -> c4 -> Ä, 87 -> ‡ --> ć
so instead of a 'teść', Windows converts it to 'teŘć' ;)
Here is a table of codes of CP-1252 -> [http://en.wikipedia.org/wiki/Windows-1252]
CP 1525 doesn't support polish characters, the only Windows ANSII code page that supports them is CP 1250
I've changed the system code page to 1250 on the server side, and this have worked fine:
declare
  fh UTL_FILE.FILE_TYPE;  
  strbuffer NVARCHAR2(1000);
begin
  fh := UTL_FILE.FOPEN_NCHAR( 'DIR_USER_FILES', CONVERT('teść.txt', 'EE8MSWIN1250' ), 'w' );
  utl_file.put_line_nchar( fh, 'chrząszcz brzmi w trzcinie');
  utl_file.put_line_nchar( fh, 'teść żócał mięśńęm');
  utl_file.fclose( fh );
  fh := UTL_FILE.FOPEN_NCHAR( 'DIR_USER_FILES', CONVERT('teść.txt', 'EE8MSWIN1250' ), 'r' );
  LOOP
    BEGIN
      utl_file.get_line_nchar( fh, strbuffer );
      dbms_output.put_line( strbuffer );
    EXCEPTION
      WHEN OTHERS THEN
        EXIT;
    END;
  END LOOP;
  utl_file.fclose( fh );
END;
/I've leaved untouched the users locale as "English (United States), only *the system locale* has been changed to "Polish"
- there are two different locales, look at this thread for details [http://mihai-nita.net/2005/06/11/setting-the-user-and-system-locales/]
If you change the server's system locale, this will affect all other non-unicode programs running on this server,
so something other may stop running properly.