Optimize a data parser that converts CLOB into table

Question

I am importing complex array data from Cold Fusion into ORACLE using a stored procedure. What I've done is convert the data into a CLOB, pass it into ORACLE, then parse it into a temporary table using a pipelined function. Unfortunately the array is large (16k rows) and it's taking a while to run.
What I'm looking is a better optimization for the data conversion. I've tried XML and the dbms_xmlstore but it was too slow. I am on 10g and don't have access to the log files or even the ability to trace anything. All I can do is look for better ways to do things and then try them.
Any suggestions or help to improve things would be welcome.
Warren Koch
The function converts the array of object into a custom delimited CLOB ( ; betwen values, | between rows) that looks like this:
prop1;prop2;prop3|prop1;prop2;prop3|prop1;prop2;prop3|prop1;prop2;prop3|
Here is the code I created.
CREATE OR REPLACE TYPE THE_IMAGE_TYPE AS OBJECT
IMAGE_NAME VARCHAR2(100),
IMAGE_SIZE INTEGER,
IMAGE_LOC INTEGER
CREATE OR REPLACE TYPE THE_IMAGE_TABLE AS TABLE OF THE_IMAGE_TYPE
CREATE GLOBAL TEMPORARY TABLE IMAGE_IMPORT
image_name VARCHAR2(25 BYTE),
loc NUMBER(11),
image_size NUMBER(11)
ON COMMIT PRESERVE ROWS
NOCACHE;
FUNCTION convert_image_data(
     in_image_clob IN CLOB
     RETURN the_image_table PIPELINED
AS
     image_clob CLOB := in_image_clob;
     image_delim_index PLS_INTEGER;
     image_index PLS_INTEGER := 1;
     row_string VARCHAR2(1000);
     out_rec THE_image_type;
BEGIN
     out_rec := the_image_type(NULL, NULL, NULL);
     IF SUBSTR(in_image_clob, -1, 1) != '|'
     THEN
          image_clob := image_clob || '|';
     END IF;
     image_delim_index := INSTR(image_clob, '|', image_index);
     WHILE image_delim_index > 0
     LOOP
          row_string := SUBSTR(image_clob, image_index, image_delim_index - image_index);
          row_string := REPLACE(row_string || '::::', ':', ' :');
          out_rec.image_name := SUBSTR(TRIM(REGEXP_SUBSTR(row_string, '[^:]+', 1, 1)), 1, 25);
          out_rec.image_loc := make_number(REGEXP_SUBSTR(row_string, '[^:]+', 1, 2));
          out_rec.image_size := make_number(REGEXP_SUBSTR(row_string, '[^:]+', 1, 3));
          PIPE ROW(out_rec);
          image_index := image_delim_index + 1;
          image_delim_index := INSTR(image_clob, '|', image_index);
     END LOOP;
     RETURN;
EXCEPTION
     WHEN OTHERS
     THEN
          RAISE;
END;
Used in my code like this:
EXECUTE IMMEDIATE 'TRUNCATE TABLE image_import';
FOR x IN (SELECT * FROM TABLE(convert_image_data(in_imagelist))
WHERE image_name IS NOT NULL)
LOOP
INSERT INTO image_import (image_name, loc, image_size)
               VALUES (x.image_name, x.image_loc, x.image_size);
END LOOP;

Answer 1

Is there any chance you can just use 1 SQL statement here? I gave it a shot based on the string you posted, not by trying to reverse engineer your code so it's possible i got it entirely wrong :)
select
    regexp_substr(image_row, '[^;]+', 1, 1) as image_name,
    regexp_substr(image_row, '[^;]+', 1, 2) as image_size,
    regexp_substr(image_row, '[^;]+', 1, 3) as image_loc   
from
   select
       regexp_substr(the_data, '[^|]+', 1, level) as image_row
   from
      select
         'prop1;prop2;prop3|prop4;prop5;prop6|propa;propb;propc|propd;prope;propf|' as the_data
      from dual
   connect by level <= length (regexp_replace(the_data, '[^|]+'))  + 1
17  where image_row is not null;
IMAGE_NAME                     IMAGE_SIZE                     IMAGE_LOC
prop1                          prop2                          prop3
prop4                          prop5                          prop6
propa                          propb                          propc
propd                          prope                          propf
4 rows selected.
Elapsed: 00:00:00.00