Tom Lane wrote:
>
> > I'm not sure. Seems that I made assumption here that
> > varlevelsup is _absolute_ level number and seems that
> > _replace_var() and _new_param() replace parser' varlevelsup
> > with absolute level value.
>
> After looking through all the references to varlevelsup, it's clear
> that all pieces of the system *except* subselect.c treat varlevelsup
> as a relative level number, so-many-levels-out-from-current-subplan.
> subselect.c has a couple of places that think nonzero varlevelsup
> is an absolute level number, with 1 as the top plan. This is certainly
> a source of bugs --- it happens to work for two-level plans, but will
> fail for anything more deeply nested. I will work on fixing subselect.c
> to bring it in line with the rest of the world...
subselect.c uses varlevelsup as absolute level number only
for correlation vars <--> params mapping, so why should it be
source of bugs? SS_replace_correlation_vars replaces all
correlation vars with parameters. Vars with absolute varlevelsup
are in PlannerParamVar only. To identify correlation vars and
to know is parameter already assigned to a var we obviously
need in absolute level number.
Vadim