Re: Virtual generated columns

On Sat, Feb 22, 2025 at 2:35 AM Dean Rasheed <dean(dot)a(dot)rasheed(at)gmail(dot)com> wrote:
> On Fri, 21 Feb 2025 at 06:16, Richard Guo <guofenglinux(at)gmail(dot)com> wrote:
> > * The expansion of virtual generated columns occurs after subquery
> > pullup, which can lead to issues. This was an oversight on my part.
> > Initially, I believed it wasn't possible for an RTE_RELATION RTE to
> > have 'lateral' set to true, so I assumed it would be safe to expand
> > virtual generated columns after subquery pullup. However, upon closer
> > look, this doesn't seem to be the case: if a subquery had a LATERAL
> > marker, that would be propagated to any of its child RTEs, even for
> > RTE_RELATION child RTE if this child rel has sampling info (see
> > pull_up_simple_subquery).
>
> Ah yes. That matches my initial instinct, which was to expand virtual
> generated columns early in the planning process, but I didn't properly
> understand why that was necessary.

After chewing on this point for a bit longer, I think the virtual
generated columns should be expanded after we have pulled up any
SubLinks within the query's quals; otherwise any virtual generated
column references within the SubLinks that should be transformed into
joins wouldn't get expanded. As an example, please consider:

create table t (a int, b int);
create table vt (a int, b int generated always as (a * 2));

insert into t values (1, 1);
insert into vt values (1);

# select 1 from t t1 where exists
(select 1 from vt where exists
(select t1.a from t t2 where vt.b = 2));
ERROR: unexpected virtual generated column reference

> LGTM aside from a comment in fireRIRrules() that needed updating and a
> minor issue in the callback function: when deciding whether to wrap
> newnode in a ReturningExpr, if newnode is a Var, it should now compare
> its varlevelsup with 0, not var->varlevelsup, since newnode hasn't had
> its varlevelsup adjusted at that point.

Nice catch.

Attached are the updated patches to fix all the mentioned issues. I
plan to push them early next week after staring at the code for a bit
longer, barring any objections.

Thanks
Richard