jian he <jian.universality@gmail.com> writes:
> get_attnum says:
> Returns InvalidAttrNumber if the attr doesn't exist (or is dropped).
> So I conclude that "attnum == 0" is not related to the idea of a system column.
attnum = 0 is also used for whole-row Vars. This is a pretty
unfortunate choice given the alternative meaning of "invalid",
but cleaning it up would be a daunting task (with not a whole
lot of payoff in the end, AFAICS). It's deeply embedded.
That being the case, you have to tread *very* carefully when
considering making changes like this.
> for example, ATExecColumnDefault, following code snippet,
> the second ereport should be "if (attnum < 0)"
> /* Prevent them from altering a system attribute */
> if (attnum <= 0)
I think that's just fine as-is. Sure, the == case is unreachable,
but it is very very common to consider whole-row Vars as being more
like system attributes than user attributes. In this particular
case, for sure we don't want to permit attaching a default to a
whole-row Var. So I'm content to allow the duplicative rejection.
regards, tom lane