"Jonathan Hull" <[EMAIL PROTECTED]> writes: > The key feature for the error is that when a result structure (eg : pg_foo) > is defined with a domain type that is not null, only PG 8.2 errors if the > result is an empty set.
The problem is explained well enough by this comment in plpgsql's code for FOR-over-query: /* * If the query didn't return any rows, set the target to NULL and return * with FOUND = false. */ At the time this code was written, there weren't any potential negative side-effects of trying to set a row value to all NULLs, but now it's possible that that fails because of domain constraints. I think the idea was to ensure that a record variable would have the correct structure (matching the query output) post-loop, even if the query produced zero rows. But it's not clear that that is really useful for anything, given plpgsql's dearth of introspection facilities. So we could make Jonathan's problem go away if we just take out the assignment of nulls, and say that FOR over no rows leaves the record variable unchanged. The documentation doesn't specify the current behavior. Looking through the code, I see another place that does the same thing: FETCH from a cursor, when the cursor has no more rows to return. It's a bit harder to argue that it's sane to leave the variable unchanged in this case. However, the documentation doesn't actually promise that the target gets set to null in this case either. Thoughts? regards, tom lane ---------------------------(end of broadcast)--------------------------- TIP 9: In versions below 8.0, the planner will ignore your desire to choose an index scan if your joining column's datatypes do not match