> This would be logical to me because we allow function creation.
> Consider, if someone creates an SQL function on a table being updated
> that returns a count of all rows who's index matches an int. Which
> would mean that an insert using that function to select rows could
> need to be evaluated for each row.
> Well, I guess what I'm trying to say is that the current behavior
> could be desired/being used by someone.
As you point out, the current behavior allows functions with
side-effects to work, where if we change it then functions must be
side-effect-free. Most functions do not have side effects, though one
using the SPI interface for example might be prone to them.
We _could_ add a column attribute for each function which declares a
function as being "safe" or not.
- Tom