Re: [HACKERS] Push down more UPDATEs/DELETEs in postgres_fdw

much cleaner the earlier version, and don't have any major concern as
such. Here are few comments:

1)

@@ -211,6 +211,12 @@ typedef struct PgFdwDirectModifyState
     PGresult   *result;            /* result for query */
     int            num_tuples;        /* # of result tuples */
     int            next_tuple;        /* index of next one to return */
+    Relation    resultRel;        /* relcache entry for the target table */


Why we need resultRel? Can't we directly use dmstate->rel ?


2) In the patch somewhere scanrelid condition being used as
fscan->scan.scanrelid == 0 where as some place its been used as
fsplan->scan.scanrelid > 0. Infact in the same function its been used
differently example postgresBeginDirectModify. Can make this consistent.

3)

+     * If UPDATE/DELETE on a join, create a RETURINING list used in the remote
+     * query.
+     */
+    if (fscan->scan.scanrelid == 0)
+        returningList = make_explicit_returning_list(resultRelation, rel,
+                                                     returningList);
+

Above block can be moved inside the if (plan->returningLists) condition above
the block. Like this:

    /*
     * Extract the relevant RETURNING list if any.
     */
    if (plan->returningLists)
    {
        returningList = (List *) list_nth(plan->returningLists, subplan_index);

        /*
         * If UPDATE/DELETE on a join, create a RETURINING list used in the remote
         * query.
         */
        if (fscan->scan.scanrelid == 0)
            returningList = make_explicit_returning_list(resultRelation,
                                                         rel,
                                                         returningList);
    }