Re: Crash when calling a pl/pgsql function with no row to pass as an argument

Chris Campbell <chris(at)bignerdranch(dot)com> writes:
> postmaster crashes if it tries to call a pl/plgsql function that
> requires a table row as an argument, and there is no row produced in
> the query that can be passed in. There is currently an assertion in the
> code to guard against this case, but it's not an error case, so it
> needs to be handled more gracefully than crashing. :)

I'm actually a bit surprised that there's no crash further upstream,
as the executor's code for handling whole-tuple variables is pretty
messy. But yes, this needs to be fixed.

> Change the assertion protecting against this case in
> src/pl/plpgsql/src/pl_exec.c to an if statement, so that the row
> argument is only copied into the function's arguments if the row
> actually exists.

I think you need to do something to explicitly set up the row value
with NULLs. Fortunately, exec_move_row will do that if you just pass
it a NULL tuple, so the fix as applied looks like

*** src/pl/plpgsql/src/pl_exec.c.orig Tue Feb 3 12:34:04 2004
--- src/pl/plpgsql/src/pl_exec.c Mon Feb 23 20:36:06 2004
***************
*** 262,271 ****
HeapTuple tup;
TupleDesc tupdesc;

! Assert(slot != NULL && !fcinfo->argnull[i]);
! tup = slot->val;
! tupdesc = slot->ttc_tupleDescriptor;
! exec_move_row(&estate, NULL, row, tup, tupdesc);
}
break;

--- 262,279 ----
HeapTuple tup;
TupleDesc tupdesc;

! if (!fcinfo->argnull[i])
! {
! Assert(slot != NULL);
! tup = slot->val;
! tupdesc = slot->ttc_tupleDescriptor;
! exec_move_row(&estate, NULL, row, tup, tupdesc);
! }
! else
! {
! /* If arg is null, treat it as an empty row */
! exec_move_row(&estate, NULL, row, NULL, NULL);
! }
}
break;

Thanks for the report!

regards, tom lane