FOR loop messes up the DBMS_UTILITY.FORMAT_ERROR_BACKTRACE

Question

Hello!
I've encountered the following problem which could be resolved for now
by eliminating FOR loop, but that's not a good solution.
The problem is that if some exception is raised inside the `for' loop
with cursor, and the outer block contains exception handler with
`format_error_backtrace' call, the last one shows not the exact line
where the exception was raised, but the loop beginning. Let me show
you.
The next block loops over some table and tries to assign one of the
field's value to local variable with intentionally low
precision. Please look at line where exception is raised:
========================================
SQL> ed
Wrote file afiedt.buf
  1  declare
  2    x number(1);
  3  begin
  4    for x_r in (select xlong from xtable)
  5    loop
  6      dbms_output.put_line('Processing... '|| x_r.xlong);
  7      x := x_r.xlong;
  8    end loop;
  9* end;
SQL> /
Processing... 611911
declare
ERROR at line 1:
ORA-06502: PL/SQL: numeric or value error: number precision too large
ORA-06512: at line 7========================================
Next, I'll put exception handler with `format_error_message' call in
it:
========================================
SQL> ed
Wrote file afiedt.buf
  1  declare
  2    x number(1);
  3  begin
  4    for x_r in (select xlong from xtable)
  5    loop
  6      dbms_output.put_line('Processing... '|| x_r.xlong);
  7      x := x_r.xlong;
  8    end loop;
  9  exception
10    when others then
11      dbms_output.put_line(dbms_utility.format_error_backtrace);
12* end;
SQL> /
Processing... 611911
ORA-06512: at line 4========================================
Now look how line number changed; instead of 7'th line it is 4'th
now -- at the beginning of the loop.
Now, let's expand this loop into explicit open/fetch/close sequence:
========================================
SQL> ed
Wrote file afiedt.buf
  1  declare
  2    x number(1);
  3    cursor x_c is select xlong from xtable;
  4    x_r x_c%rowtype;
  5  begin
  6    open x_c;
  7    loop
  8      fetch x_c into x_r;
  9      exit when x_c%notfound;
10      dbms_output.put_line('Processing... '|| x_r.xlong);
11      x := x_r.xlong;
12    end loop;
13    close x_c;
14  exception
15    when others then
16      dbms_output.put_line(dbms_utility.format_error_backtrace);
17* end;
SQL> /
Processing... 611911
ORA-06512: at line 11========================================
Here we see that line number where exception was raised is now
correct!
I suppose the problem is in some hidden exception handling behind the
`for' loop execution. Any help will be appreciated, thanks in advance!
Message was edited by:
uj2

Answer 1

uj2,
I think the reason is probably PL/SQL optimisation. Because you are using backtrace, it means you are on 10g. This means you have an automatic PLSQL_OPTIMIZE_LEVEL of 2. At this level, the compiler will re-write your cursor-for-loop to array fetches "under the covers", so my feeling is that the internal code re-organisation might have something to do with it.
Can you try your cursor-for-loop example with these settings:
ALTER SESSION SET PLSQL_OPTIMIZE_LEVEL = 0;
ALTER SESSION SET PLSQL_OPTIMIZE_LEVEL = 1;
Neither of these levels optimise cursor-for-loops.
The reason your explicit fetch doesn't display the same behaviour is that the compiler doesn't optimise those.
Of course I could be wrong, but this is where I'd start testing.
Regards