Re: Virtual generated columns - Mailing list pgsql-hackers

On Fri, May 16, 2025 at 3:26 PM Richard Guo <guofenglinux@gmail.com> wrote:
>
> On Fri, May 16, 2025 at 1:00 PM Alexander Lakhin <exclusion@gmail.com> wrote:
> > I've discovered yet another way to trigger that error:
> > create table vt (a int, b int generated always as (a * 2), c int);
> > insert into vt values(1);
> > alter table vt alter column c type bigint using b + c;
> >
> > ERROR:  XX000: unexpected virtual generated column reference
> > LOCATION:  CheckVarSlotCompatibility, execExprInterp.c:2410
>
> Thank you for the report.  It seems that we fail to expand references
> to virtual generated columns in the NewColumnValues expression when
> altering tables.  We might be able to fix it by:
>
> @@ -6203,7 +6203,7 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap)
>         NewColumnValue *ex = lfirst(l);
>
>         /* expr already planned */
> -       ex->exprstate = ExecInitExpr((Expr *) ex->expr, NULL);
> +       ex->exprstate = ExecInitExpr((Expr *)
> expand_generated_columns_in_expr((Node *) ex->expr, oldrel, 1), NULL);
>

we have used the USING expression in ATPrepAlterColumnType,
ATColumnChangeRequiresRewrite.
expanding it on ATPrepAlterColumnType seems to make more sense?

@@ -14467,7 +14467,7 @@ ATPrepAlterColumnType(List **wqueue,
                 */
                newval = (NewColumnValue *) palloc0(sizeof(NewColumnValue));
                newval->attnum = attnum;
-               newval->expr = (Expr *) transform;
+               newval->expr = (Expr *)
expand_generated_columns_in_expr(transform, rel, 1);
                newval->is_generated = false;