How to output a CLOB to file using UTF-8

Question

I am working on an 9.2.04 database. I'm generating xml using SQLX functions and inserting the result into a table with a column of XMLTYPE datatype.
The database characterset is WE8ISO8859P15. I need to write the contents of my XMLTYPE column to a file on the server using UTF-8 encoding. I've tried The following :
declare
temp_xml xmltype;
cl clob;
begin
select myxml into temp_xml from xml_table;
cl := temp_xml.getClobVal;
clob_to_xmlfile('XMLDIR','myfilename.xml',cl);
end;
The code for clob_to_xmlfile is :
PROCEDURE clob_to_xmlfile( p_dir IN VARCHAR2,
p_file IN VARCHAR2,
p_clob IN CLOB )
AS
l_output utl_file.file_type;
l_amt number default 32000;
l_offset number default 1;
l_length number default nvl(dbms_lob.getlength(p_clob),0);
BEGIN
-- notice use of fopen_nchar and put_nchar to use unicode
l_output := utl_file.fopen_nchar(p_dir, p_file, 'w', 32760);
while ( l_offset < l_length )
loop
utl_file.put_nchar(l_output,
dbms_lob.substr(p_clob,l_amt,l_offset) );
utl_file.fflush(l_output);
l_offset := l_offset + l_amt;
end loop;
utl_file.new_line(l_output);
utl_file.fclose(l_output);
END clob_to_xmlfile;
However, the resultant file has somehow 'lost' characters, thereby making the xml malformed. I know by inspection that the XMLTYPE column in my table contains valid xml, so the 'corruption' is occuring when I try writing it to file.
Is this something to do with characterset s ?? Can anyone help me with this ??

Answer 1

In 10g, DBMS_XMLDOM has writeToFile() writing XML into files without the temporary lobs. It lets you specify a character set parameter. However, 9.2.0.4 doesn't not support character set conversion so this may not work. The options would be upgrading to 10g or building the function like what you have.
The presented code seems to have the followings issues:
(1) the length expansion is not taken into accout; utf-8 encoding needs equal or grater number of bytes to represent the same string in iso-8859-15,
potentially up to 3x. If your buffer was 32k, you should not read more than 10k at a time in order to make sure all characters read from CLOB can fit into the buffer. This problem apparently can cause unexpected behavior: a
symptom may be lost characters.
(2) while ( offset < length ) has a off-by-one problem, as we start counting LOB offset with 1, not zero.
(3) I think utl_file.fflush needs the new line character at the end of buffer so it can tell how much to write. It would be difficult to write a bufferful of data because you have no idea where a line begins and ends
while reading a CLOB.
Probably a more practical approach to make this work would be using the PUT_RAW function with autoflush turned on, where data can be written blindly regardless of when the buffer should be flushed. That is:
-read 10k from the clob (DBMS_LOB.READ)
-manually convert it into UTF-8 to a 30k buffer (UTL_RAW.CONVERT)
-then write it with UTL_FILE.PUT_RAW (UTL_RAW.CAST_TO_RAW)