Re: misleading error message in ProcessUtilitySlow T_CreateStatsStmt

On Thu, Nov 13, 2025 at 5:41 PM Álvaro Herrera <alvherre@kurilemu.de> wrote:
>

>
> > @@ -15658,10 +15658,19 @@ ATPostAlterTypeParse(Oid oldId, Oid oldRelId, Oid refRelId, char *cmd,
> >                       querytree_list = list_concat(querytree_list, afterStmts);
> >               }
> >               else if (IsA(stmt, CreateStatsStmt))
> > +             {
> > +                     RangeTblEntry   *rte;
> > +                     CreateStatsStmt *ss = castNode(CreateStatsStmt, stmt);
> > +
> > +                     rte = makeNode(RangeTblEntry);
> > +                     rte->rtekind = RTE_RELATION;
> > +                     rte->relid = oldRelId;
> > +                     rte->rellockmode = ShareUpdateExclusiveLock;
> > +                     ss->rtable = list_make1(rte);
> > +
> >                       querytree_list = lappend(querytree_list,
> > -                                                                      transformStatsStmt(oldRelId,
> > -                                                                                                             (CreateStatsStmt *) stmt,
> > -                                                                                                             cmd));
> > +                                                                      transformStatsStmt(ss, cmd));
> > +             }
> >               else
> >                       querytree_list = lappend(querytree_list, stmt);
> >       }
>
> Hmm, how would this part here work in the hypothetical world where a
> stats object references multiple relations?
>
hi.

For extended statistics that span multiple relations (hypothetically),
we should have the OIDs of
all involved relations within the pg_statistic_ext catalog. Without this , it's
hard to do statistic's expression deparse, changing column data types
or generation
expressions. Catalog pg_depend do have statistics associated relation OIDs
information, relying on it for statistical expression deparsing would be more
complicated.

Once pg_statistic_ext have all relation oid information,
RememberStatisticsForRebuilding will not only collecting
AlteredTableInfo->changedStatisticsOids, it will aslo collect the OID of other
relations associated with this statistic.

For the above quoted part, we should construct a RangeTblEntry for
each associated
relation (OID).
Obviously transformStatsStmt itself needs to figure out how to
deal with multi-relation expressions.

rebased and minor polishing.



--
jian
https://www.enterprisedb.com