Re: Removing Functionally Dependent GROUP BY Columns

On 14/02/2016 5:11 pm, "Tom Lane" <tgl(at)sss(dot)pgh(dot)pa(dot)us> wrote:
>
> David Rowley <david(dot)rowley(at)2ndquadrant(dot)com> writes:
> > On 12/02/2016 12:01 am, "Tom Lane" <tgl(at)sss(dot)pgh(dot)pa(dot)us> wrote:
> > I can't quite understand what you're seeing here.
>
> The second loop is iterating through the original GROUP BY list, so it
> will see again any outer Vars that were excluded by the first loop.
> It needs to exclude them exactly the same, because they are outside
> the scope of its data structures. Consider something like (admittedly
> pretty silly, but legal SQL)
>
> create table up (u1 int, u2 int, u3 int);
> create table down (f1 int primary key, f2 int);
>
> select * from othertable, up
> where u1 in (select f2 from down group by f1, f2, up.u3);
>
> up.u3 would have varlevelsup = 1, varno = 2, varattno = 3.
> If you don't skip it then the surplusvars[var->varno] access
> will be trying to fetch off the end of the surplusvars[] array,
> because there is only one RTE in the subquery's rangetable
> though there are two in the outer query's rangetable.

Thanks for explaining this. Clearly I missed the case of the varno pointing
off the end of the array. Thanks for noticing and fixing.