PL/SQL to create a temp table that will be dropped after session ends

Question

Is it possible in PL/SQL to create a temp table that will be dropped after the session ends? Please provide example if possible. I can create a global temp table in PL/SQL but I am not sure how (if possible) to have it 'drop' once the session ends.
DB: 10g
OS: Wiindoze 2003 Server
:-)

Answer 1

As others have mentioned (but probably not clearly explained), Oracle treats temporary tables differently to SQL Server.
In SQL Server you create a temporary table and it gets dropped (automatically I assume, I dont do SQL Server) after the session finishes. This will obviously allow each session to "request" a temporary table to use, then use it, and not have to worry about cleaning up the database after the session has finished.
Oracle takes a different approach...
On the assumption that each session is likely to be creating a temporary table for the same purposes, with the same structure, Oracle let's you create a Global Temporary Table a.k.a. GTT (which you've already come across). You only have to create this table once and you leave it on the database. This then means that any code written to use that table doesn't have to be dynamic code and can be verified and checked at compile time, just like code written for any other table. The difference of a GTT from a regular table is that any data you put into that table can only be seen by that session and will not interfere with any data of other sessions and, when you either commit, or end the session (depending on the "on commit delete rows" or "on commit preserve rows" option used when creating the GTT), that data from your own session will automatically be removed and hence the table is cleaned up that way, whilst the actual table itself remains.
Some people from SQL Server backgrounds try and create and drop tables dynamically in their PL/SQL code, but this leads to problems...
SQL> ed
Wrote file afiedt.buf
  1  begin
  2    execute immediate 'create table my_temp (x number)';
  3    insert into my_temp values (1);
  4    execute immediate 'drop table my_temp';
  5* end;
SQL> /
  insert into my_temp values (1);
ERROR at line 3:
ORA-06550: line 3, column 15:
PL/SQL: ORA-00942: table or view does not exist
ORA-06550: line 3, column 3:
PL/SQL: SQL Statement ignoredi.e. the code will not compile for direct DML statements trying to use that table.
They then try and get around this issue by making their DML statements dynamic too...
SQL> ed
Wrote file afiedt.buf
  1  create or replace procedure my_proc is
  2  begin
  3    execute immediate 'create table my_temp (x number)';
  4    execute immediate 'insert into my_temp values (''A'')';
  5    execute immediate 'drop table my_temp';
  6* end;
SQL> /
Procedure created.... which looks great and it compiles ok... but... when they try and run it...
SQL> exec my_proc;
BEGIN my_proc; END;
ERROR at line 1:
ORA-01722: invalid number
ORA-06512: at "SCOTT.MY_PROC", line 4
ORA-06512: at line 1... oops the code has a bug in it. Our DML statement was invalid.
This is really something that would have been caught at compile time, if the statement had been a direct DML statement rather than dynamic. And thus we see the problem with people trying to write all their code as dynamic SQL... it's more likely to contain bugs that won't be detected at compile time and only come to light at run time... sometimes only under certain conditions and sometimes once it's got into a production environment. Bad Idea!!!! ;)
Far better to never create tables (or most other database objects) at run time. Just create them once as part of the database design/implementation and use them as required, allowing you to catch the most common coding errors up front before they get anywhere near a test environment or worse still, a production environment.