Define variable a ROWTYPE based on a table name passed as a parameter

Question

I am trying to write a procedure that processes rows for tables of differing structures. Table_name is passed as an IN parameter to the procedure. I define a cursor within the function that is defined by a SELECT * FROM table_name. I also need to define two variables, l_this_row and l_prev_row. I need to define them as table_name%ROWTYPE. Of course, table_name is a VARCHAR2 that contains the name of a table, whereas Oracle is expecting an actual table name in the declaration. How can I define these variables to make something like the code shown below compile and work properly? I am using Oracle 10gR2.
DEFINE PROCEDURE foo(table_name IN VARCHAR2) IS
l_this_row table_name%ROWTYPE;
l_prev_row table_name%ROWTYPE;
l_cursor SYSREFCURSOR;
BEGIN
OPEN l_cursor FOR
'SELECT * FROM '
|| table_name
|| ' ORDER BY '
|| get_pk_col_name_for_table(table_name)
|| ', version_num'; --version_num is guaranteed to be a valid column name in every table, although the other column names vary in name, quantity, and type per table.
LOOP
FETCH l_cursor INTO l_this_record;
EXIT WHEN l_cursor%NOTFOUND;
-- dynamic comparison of l_this_record and l_prev_record and conditional processing based on the comparison occurs here
l_prev_record := l_this_record;
END LOOP;
END;

Answer 1

Billy  Verreynne  wrote:
Well, one method is to keep the SQL projection consistent across the number of dynamic SQLs being dealt with. With the projection being consistent, your code does not need to guess what the row structure the column returns.Unfortunately, what I need to compare is almost everything except the primary keys. So one table might contain addresses and another might contain invoice headers. There's no way to get them into the same structure.
Billy  Verreynne  wrote:
This approach can also be done in a generic fashion - dealing with an arbitrary number of columns per row. This dynamic structure is more flexible, but also increases the complexity of the code that has to deal with this. For example:
create or replace type TStrings is table of varchar2(4000);Now SQLs can be created to pass any number of columns to the code to process:
select TStrings( empid, ename, jobid, date_employed, date_of_birth ) from emp order by empidThe code then uses the Count method of the TStrings type to determine the number of columns passed and processes that.Then the problem becomes having to specify all the column names. I need to figure them out dynamically, as in "SELECT column_name FROM user_tab_cols WHERE column_name NOT IN ('ID', 'PREV_ID', 'NEXT_ID')". It would, of course, be best if I didn't have to populate each array element with an individual SELECT in its own EXECUTE IMMEDIATE statement. That would certainly slow things down. If I were trying to write this using ADO.NET, I could reference something like table_name(row_index).columns(ix) without having to know any of the column names, and I could exclude columns by name using a reference like table_name.columns(column_index).name. (I don't recall the exact syntax, but it is something reasonably close to that.) But PL/SQL doesn't seem to have anything close to this.
Ideally, I want to get it to where I can do something like this in a private function:
FOR ix IN 1..TStrings.Count LOOP
   IF this_row(ix) <> last_row(ix) THEN
      RETURN FALSE;
   END IF;
END LOOP;
RETURN TRUE;
Billy  Verreynne  wrote:Last comment. I have to echo what the others said. This is not optimal. Dealing with variant/dynamic structures and processing rows using PL/SQL (in a slow-by-slow fashion) does not scale well. Scalability and performance comes from processing data sets using SQL.I completely understand. I'm a big fan of replacing a lot of code with one SQL statement, but that just won't work for this problem.