use oracle cursor within a sys_refcursor

Lloyd picture Lloyd · Jul 8, 2009 · Viewed 10.8k times · Source

I've got a PL/SQL package that returns a sys_refcursor based on the id that you pass it. I'd like to iterate through some ids and create a new ref cursor with one column from the original result set repeated for each id. (Sort of a cross tab.) A very simplified version of the PL/SQL block looks like:

create or replace package body dashboard_package is

   procedure visits(RC in out sys_refcursor, IdNumber varchar2) as 
   BEGIN

      OPEN RC FOR 


      select *
  from (
        select cat, cat_order, subcat, label_text
               , trim(to_char(sum(v.current_month),'9,999,999,999')) current_month
               , trim(to_char(sum(v.ly_month),'9,999,999,999')) ly_month
               , trim(to_char(sum(v.ytd_tot),'9,999,999,999')) ytd_tot
               , trim(to_char(sum(v.lytd_tot),'9,999,999,999')) lytd_tot
               , trim(to_char(sum(v.ly_tot),'9,999,999,999')) ly_tot
          from dashboard v
         where v.id_number = IdNumber
         group by cat_order, subcat, cat, label_text

            union all
            ...
             ) a

     order by cat_order, subcat;

       END; 
END;

I think if I had something like this

create or replace procedure test_refcur is
   refCursorValue SYS_REFCURSOR;   
begin
   dashboard_package.visits(refCursorValue,12345);
   for cursrow in refCursorValue loop
      dbms_output.put_line(cursrow.ytd_tot);
   end loop;
end test_refcur;

working, I could take it from there... any thoughts? Or perhaps clarification on the question that I should be asking.

Answer

Jeffrey Kemp picture Jeffrey Kemp · Jul 8, 2009

If you're coming in with a number of IDs, then first prize would be to run only one SQL query to fetch the lot in one go, using a bulk in-bind for the IDs. This would probably require a modification to dashboard_package.visits, or writing a new version of the visits procedure to accept a PL/SQL table of IDs instead of a single ID.

If your hands are tied WRT modifying dashboard_package, then you could write a pipelined function to return the rows for a set of IDs:

-- create some helper types for the pipelined function
create type visitobj as object
(id             number
,cat            dashboard.cat%type
,cat_order      dashboard.cat_order%type
,subcat         dashboard.subcat%type
,label_text     dashboard.label_text%type
,current_month  varchar2(13)
,ly_month       varchar2(13)
,ytd_tot        varchar2(13)
,lytd_tot       varchar2(13)
,ly_tot         varchar2(13));
create type visittable as table of visitobj;

create or replace function test_refcur
   return visittable deterministic pipelined is
   refCursorValue SYS_REFCURSOR;
   cat            dashboard.cat%type;
   cat_order      dashboard.cat_order%type;
   subcat         dashboard.subcat%type;
   label_text     dashboard.label_text%type;
   current_month  varchar2(13);
   ly_month       varchar2(13);
   ytd_tot        varchar2(13);
   lytd_tot       varchar2(13);
   ly_tot         varchar2(13);
begin
  for id in (/*iterate through the IDs*/) loop
   dashboard_package.visits(refCursorValue, id);
   loop
      fetch refCursorValue into cat, cat_order, subcat, label_text,
                                current_month, ly_month, ytd_tot,
                                lytd_tot, ly_tot;
      exit when refCursorValue%NOTFOUND;
      pipe row (visitobj (id, cat, cat_order, subcat, label_text,
                          current_month, ly_month, ytd_tot,
                          lytd_tot, ly_tot));
   end loop;
  end loop;
  return;
end test_refcur;

-- now you can simply do this:
SELECT * FROM TABLE(test_refcur);

(Of course, "/*iterate through the IDs*/" would be whatever method you want to use to gather the IDs for which the function should be called - e.g. could be a PL/SQL table of IDs, or perhaps another query).

Again I'd stress that "first prize" is to not do any of this extra work at all - just have a dashboard_package.visits that does it all in one SQL.

On a side note, trim(to_char(sum(v.ly_tot),'9,999,999,999')) can be simplified to to_char(sum(v.ly_tot),'FM9,999,999,999'). Also, if you use the format 'FM9G999G999G999' instead, it will be non-locale-specific.