[pghintplan/pg_hint_plan.git] / core.c

/*-------------------------------------------------------------------------
 *
 * core.c
 *        Routines copied from PostgreSQL core distribution.
 *
 * The main purpose of this files is having access to static functions in core.
 * Another purpose is tweaking functions behavior by replacing part of them by
 * macro definitions. See at the end of pg_hint_plan.c for details. Anyway,
 * this file *must* contain required functions without making any change.
 *
 * This file contains the following functions from corresponding files.
 *
 * src/backend/optimizer/path/allpaths.c
 *
 *      static functions:
 *     set_plain_rel_pathlist()
 *     create_plain_partial_paths()
 *     set_append_rel_pathlist()
 *     generate_mergeappend_paths()
 *     get_cheapest_parameterized_child_path()
 *     accumulate_append_subpath()
 *
 *  public functions:
 *     standard_join_search(): This funcion is not static. The reason for
 *        including this function is make_rels_by_clause_joins. In order to
 *        avoid generating apparently unwanted join combination, we decided to
 *        change the behavior of make_join_rel, which is called under this
 *        function.
 *
 * src/backend/optimizer/path/joinrels.c
 *
 *      public functions:
 *     join_search_one_level(): We have to modify this to call my definition of
 *                  make_rels_by_clause_joins.
 *
 *      static functions:
 *     make_rels_by_clause_joins()
 *     make_rels_by_clauseless_joins()
 *     join_is_legal()
 *     has_join_restriction()
 *     is_dummy_rel()
 *     mark_dummy_rel()
 *     restriction_is_constant_false()
 *
 *
 * Portions Copyright (c) 1996-2019, PostgreSQL Global Development Group
 * Portions Copyright (c) 1994, Regents of the University of California
 *
 *-------------------------------------------------------------------------
 */


/*
 * set_plain_rel_pathlist
 *        Build access paths for a plain relation (no subquery, no inheritance)
 */
static void
set_plain_rel_pathlist(PlannerInfo *root, RelOptInfo *rel, RangeTblEntry *rte)
{
        Relids          required_outer;

        /*
         * We don't support pushing join clauses into the quals of a seqscan, but
         * it could still have required parameterization due to LATERAL refs in
         * its tlist.
         */
        required_outer = rel->lateral_relids;

        /* Consider sequential scan */
        add_path(rel, create_seqscan_path(root, rel, required_outer, 0));

        /* If appropriate, consider parallel sequential scan */
        if (rel->consider_parallel && required_outer == NULL)
                create_plain_partial_paths(root, rel);

        /* Consider index scans */
        create_index_paths(root, rel);

        /* Consider TID scans */
        create_tidscan_paths(root, rel);
}

/*
 * create_plain_partial_paths
 *        Build partial access paths for parallel scan of a plain relation
 */
static void
create_plain_partial_paths(PlannerInfo *root, RelOptInfo *rel)
{
        int                     parallel_workers;

        /*
         * If the user has set the parallel_workers reloption, use that; otherwise
         * select a default number of workers.
         */
        if (rel->rel_parallel_workers != -1)
                parallel_workers = rel->rel_parallel_workers;
        else
        {
                int                     parallel_threshold;

                /*
                 * If this relation is too small to be worth a parallel scan, just
                 * return without doing anything ... unless it's an inheritance child.
                 * In that case, we want to generate a parallel path here anyway.  It
                 * might not be worthwhile just for this relation, but when combined
                 * with all of its inheritance siblings it may well pay off.
                 */
                if (rel->pages < (BlockNumber) min_parallel_relation_size &&
                        rel->reloptkind == RELOPT_BASEREL)
                        return;

                /*
                 * Select the number of workers based on the log of the size of the
                 * relation.  This probably needs to be a good deal more
                 * sophisticated, but we need something here for now.  Note that the
                 * upper limit of the min_parallel_relation_size GUC is chosen to
                 * prevent overflow here.
                 */
                parallel_workers = 1;
                parallel_threshold = Max(min_parallel_relation_size, 1);
                while (rel->pages >= (BlockNumber) (parallel_threshold * 3))
                {
                        parallel_workers++;
                        parallel_threshold *= 3;
                        if (parallel_threshold > INT_MAX / 3)
                                break;                  /* avoid overflow */
                }
        }

        /*
         * In no case use more than max_parallel_workers_per_gather workers.
         */
        parallel_workers = Min(parallel_workers, max_parallel_workers_per_gather);

        /* If any limit was set to zero, the user doesn't want a parallel scan. */
        if (parallel_workers <= 0)
                return;

        /* Add an unordered partial path based on a parallel sequential scan. */
        add_partial_path(rel, create_seqscan_path(root, rel, NULL, parallel_workers));
}

/*
 * set_append_rel_pathlist
 *        Build access paths for an "append relation"
 */
static void
set_append_rel_pathlist(PlannerInfo *root, RelOptInfo *rel,
                                                Index rti, RangeTblEntry *rte)
{
        int                     parentRTindex = rti;
        List       *live_childrels = NIL;
        List       *subpaths = NIL;
        bool            subpaths_valid = true;
        List       *partial_subpaths = NIL;
        bool            partial_subpaths_valid = true;
        List       *all_child_pathkeys = NIL;
        List       *all_child_outers = NIL;
        ListCell   *l;

        /*
         * Generate access paths for each member relation, and remember the
         * cheapest path for each one.  Also, identify all pathkeys (orderings)
         * and parameterizations (required_outer sets) available for the member
         * relations.
         */
        foreach(l, root->append_rel_list)
        {
                AppendRelInfo *appinfo = (AppendRelInfo *) lfirst(l);
                int                     childRTindex;
                RangeTblEntry *childRTE;
                RelOptInfo *childrel;
                ListCell   *lcp;

                /* append_rel_list contains all append rels; ignore others */
                if (appinfo->parent_relid != parentRTindex)
                        continue;

                /* Re-locate the child RTE and RelOptInfo */
                childRTindex = appinfo->child_relid;
                childRTE = root->simple_rte_array[childRTindex];
                childrel = root->simple_rel_array[childRTindex];

                /*
                 * If set_append_rel_size() decided the parent appendrel was
                 * parallel-unsafe at some point after visiting this child rel, we
                 * need to propagate the unsafety marking down to the child, so that
                 * we don't generate useless partial paths for it.
                 */
                if (!rel->consider_parallel)
                        childrel->consider_parallel = false;

                /*
                 * Compute the child's access paths.
                 */
                set_rel_pathlist(root, childrel, childRTindex, childRTE);

                /*
                 * If child is dummy, ignore it.
                 */
                if (IS_DUMMY_REL(childrel))
                        continue;

                /*
                 * Child is live, so add it to the live_childrels list for use below.
                 */
                live_childrels = lappend(live_childrels, childrel);

                /*
                 * If child has an unparameterized cheapest-total path, add that to
                 * the unparameterized Append path we are constructing for the parent.
                 * If not, there's no workable unparameterized path.
                 */
                if (childrel->cheapest_total_path->param_info == NULL)
                        subpaths = accumulate_append_subpath(subpaths,
                                                                                          childrel->cheapest_total_path);
                else
                        subpaths_valid = false;

                /* Same idea, but for a partial plan. */
                if (childrel->partial_pathlist != NIL)
                        partial_subpaths = accumulate_append_subpath(partial_subpaths,
                                                                           linitial(childrel->partial_pathlist));
                else
                        partial_subpaths_valid = false;

                /*
                 * Collect lists of all the available path orderings and
                 * parameterizations for all the children.  We use these as a
                 * heuristic to indicate which sort orderings and parameterizations we
                 * should build Append and MergeAppend paths for.
                 */
                foreach(lcp, childrel->pathlist)
                {
                        Path       *childpath = (Path *) lfirst(lcp);
                        List       *childkeys = childpath->pathkeys;
                        Relids          childouter = PATH_REQ_OUTER(childpath);

                        /* Unsorted paths don't contribute to pathkey list */
                        if (childkeys != NIL)
                        {
                                ListCell   *lpk;
                                bool            found = false;

                                /* Have we already seen this ordering? */
                                foreach(lpk, all_child_pathkeys)
                                {
                                        List       *existing_pathkeys = (List *) lfirst(lpk);

                                        if (compare_pathkeys(existing_pathkeys,
                                                                                 childkeys) == PATHKEYS_EQUAL)
                                        {
                                                found = true;
                                                break;
                                        }
                                }
                                if (!found)
                                {
                                        /* No, so add it to all_child_pathkeys */
                                        all_child_pathkeys = lappend(all_child_pathkeys,
                                                                                                 childkeys);
                                }
                        }

                        /* Unparameterized paths don't contribute to param-set list */
                        if (childouter)
                        {
                                ListCell   *lco;
                                bool            found = false;

                                /* Have we already seen this param set? */
                                foreach(lco, all_child_outers)
                                {
                                        Relids          existing_outers = (Relids) lfirst(lco);

                                        if (bms_equal(existing_outers, childouter))
                                        {
                                                found = true;
                                                break;
                                        }
                                }
                                if (!found)
                                {
                                        /* No, so add it to all_child_outers */
                                        all_child_outers = lappend(all_child_outers,
                                                                                           childouter);
                                }
                        }
                }
        }

        /*
         * If we found unparameterized paths for all children, build an unordered,
         * unparameterized Append path for the rel.  (Note: this is correct even
         * if we have zero or one live subpath due to constraint exclusion.)
         */
        if (subpaths_valid)
                add_path(rel, (Path *) create_append_path(rel, subpaths, NULL, 0));

        /*
         * Consider an append of partial unordered, unparameterized partial paths.
         */
        if (partial_subpaths_valid)
        {
                AppendPath *appendpath;
                ListCell   *lc;
                int                     parallel_workers = 0;

                /*
                 * Decide on the number of workers to request for this append path.
                 * For now, we just use the maximum value from among the members.  It
                 * might be useful to use a higher number if the Append node were
                 * smart enough to spread out the workers, but it currently isn't.
                 */
                foreach(lc, partial_subpaths)
                {
                        Path       *path = lfirst(lc);

                        parallel_workers = Max(parallel_workers, path->parallel_workers);
                }
                Assert(parallel_workers > 0);

                /* Generate a partial append path. */
                appendpath = create_append_path(rel, partial_subpaths, NULL,
                                                                                parallel_workers);
                add_partial_path(rel, (Path *) appendpath);
        }

        /*
         * Also build unparameterized MergeAppend paths based on the collected
         * list of child pathkeys.
         */
        if (subpaths_valid)
                generate_mergeappend_paths(root, rel, live_childrels,
                                                                   all_child_pathkeys);

        /*
         * Build Append paths for each parameterization seen among the child rels.
         * (This may look pretty expensive, but in most cases of practical
         * interest, the child rels will expose mostly the same parameterizations,
         * so that not that many cases actually get considered here.)
         *
         * The Append node itself cannot enforce quals, so all qual checking must
         * be done in the child paths.  This means that to have a parameterized
         * Append path, we must have the exact same parameterization for each
         * child path; otherwise some children might be failing to check the
         * moved-down quals.  To make them match up, we can try to increase the
         * parameterization of lesser-parameterized paths.
         */
        foreach(l, all_child_outers)
        {
                Relids          required_outer = (Relids) lfirst(l);
                ListCell   *lcr;

                /* Select the child paths for an Append with this parameterization */
                subpaths = NIL;
                subpaths_valid = true;
                foreach(lcr, live_childrels)
                {
                        RelOptInfo *childrel = (RelOptInfo *) lfirst(lcr);
                        Path       *subpath;

                        subpath = get_cheapest_parameterized_child_path(root,
                                                                                                                        childrel,
                                                                                                                        required_outer);
                        if (subpath == NULL)
                        {
                                /* failed to make a suitable path for this child */
                                subpaths_valid = false;
                                break;
                        }
                        subpaths = accumulate_append_subpath(subpaths, subpath);
                }

                if (subpaths_valid)
                        add_path(rel, (Path *)
                                         create_append_path(rel, subpaths, required_outer, 0));
        }
}

/*
 * generate_mergeappend_paths
 *              Generate MergeAppend paths for an append relation
 *
 * Generate a path for each ordering (pathkey list) appearing in
 * all_child_pathkeys.
 *
 * We consider both cheapest-startup and cheapest-total cases, ie, for each
 * interesting ordering, collect all the cheapest startup subpaths and all the
 * cheapest total paths, and build a MergeAppend path for each case.
 *
 * We don't currently generate any parameterized MergeAppend paths.  While
 * it would not take much more code here to do so, it's very unclear that it
 * is worth the planning cycles to investigate such paths: there's little
 * use for an ordered path on the inside of a nestloop.  In fact, it's likely
 * that the current coding of add_path would reject such paths out of hand,
 * because add_path gives no credit for sort ordering of parameterized paths,
 * and a parameterized MergeAppend is going to be more expensive than the
 * corresponding parameterized Append path.  If we ever try harder to support
 * parameterized mergejoin plans, it might be worth adding support for
 * parameterized MergeAppends to feed such joins.  (See notes in
 * optimizer/README for why that might not ever happen, though.)
 */
static void
generate_mergeappend_paths(PlannerInfo *root, RelOptInfo *rel,
                                                   List *live_childrels,
                                                   List *all_child_pathkeys)
{
        ListCell   *lcp;

        foreach(lcp, all_child_pathkeys)
        {
                List       *pathkeys = (List *) lfirst(lcp);
                List       *startup_subpaths = NIL;
                List       *total_subpaths = NIL;
                bool            startup_neq_total = false;
                ListCell   *lcr;

                /* Select the child paths for this ordering... */
                foreach(lcr, live_childrels)
                {
                        RelOptInfo *childrel = (RelOptInfo *) lfirst(lcr);
                        Path       *cheapest_startup,
                                           *cheapest_total;

                        /* Locate the right paths, if they are available. */
                        cheapest_startup =
                                get_cheapest_path_for_pathkeys(childrel->pathlist,
                                                                                           pathkeys,
                                                                                           NULL,
                                                                                           STARTUP_COST);
                        cheapest_total =
                                get_cheapest_path_for_pathkeys(childrel->pathlist,
                                                                                           pathkeys,
                                                                                           NULL,
                                                                                           TOTAL_COST);

                        /*
                         * If we can't find any paths with the right order just use the
                         * cheapest-total path; we'll have to sort it later.
                         */
                        if (cheapest_startup == NULL || cheapest_total == NULL)
                        {
                                cheapest_startup = cheapest_total =
                                        childrel->cheapest_total_path;
                                /* Assert we do have an unparameterized path for this child */
                                Assert(cheapest_total->param_info == NULL);
                        }

                        /*
                         * Notice whether we actually have different paths for the
                         * "cheapest" and "total" cases; frequently there will be no point
                         * in two create_merge_append_path() calls.
                         */
                        if (cheapest_startup != cheapest_total)
                                startup_neq_total = true;

                        startup_subpaths =
                                accumulate_append_subpath(startup_subpaths, cheapest_startup);
                        total_subpaths =
                                accumulate_append_subpath(total_subpaths, cheapest_total);
                }

                /* ... and build the MergeAppend paths */
                add_path(rel, (Path *) create_merge_append_path(root,
                                                                                                                rel,
                                                                                                                startup_subpaths,
                                                                                                                pathkeys,
                                                                                                                NULL));
                if (startup_neq_total)
                        add_path(rel, (Path *) create_merge_append_path(root,
                                                                                                                        rel,
                                                                                                                        total_subpaths,
                                                                                                                        pathkeys,
                                                                                                                        NULL));
        }
}

/*
 * get_cheapest_parameterized_child_path
 *              Get cheapest path for this relation that has exactly the requested
 *              parameterization.
 *
 * Returns NULL if unable to create such a path.
 */
static Path *
get_cheapest_parameterized_child_path(PlannerInfo *root, RelOptInfo *rel,
                                                                          Relids required_outer)
{
        Path       *cheapest;
        ListCell   *lc;

        /*
         * Look up the cheapest existing path with no more than the needed
         * parameterization.  If it has exactly the needed parameterization, we're
         * done.
         */
        cheapest = get_cheapest_path_for_pathkeys(rel->pathlist,
                                                                                          NIL,
                                                                                          required_outer,
                                                                                          TOTAL_COST);
        Assert(cheapest != NULL);
        if (bms_equal(PATH_REQ_OUTER(cheapest), required_outer))
                return cheapest;

        /*
         * Otherwise, we can "reparameterize" an existing path to match the given
         * parameterization, which effectively means pushing down additional
         * joinquals to be checked within the path's scan.  However, some existing
         * paths might check the available joinquals already while others don't;
         * therefore, it's not clear which existing path will be cheapest after
         * reparameterization.  We have to go through them all and find out.
         */
        cheapest = NULL;
        foreach(lc, rel->pathlist)
        {
                Path       *path = (Path *) lfirst(lc);

                /* Can't use it if it needs more than requested parameterization */
                if (!bms_is_subset(PATH_REQ_OUTER(path), required_outer))
                        continue;

                /*
                 * Reparameterization can only increase the path's cost, so if it's
                 * already more expensive than the current cheapest, forget it.
                 */
                if (cheapest != NULL &&
                        compare_path_costs(cheapest, path, TOTAL_COST) <= 0)
                        continue;

                /* Reparameterize if needed, then recheck cost */
                if (!bms_equal(PATH_REQ_OUTER(path), required_outer))
                {
                        path = reparameterize_path(root, path, required_outer, 1.0);
                        if (path == NULL)
                                continue;               /* failed to reparameterize this one */
                        Assert(bms_equal(PATH_REQ_OUTER(path), required_outer));

                        if (cheapest != NULL &&
                                compare_path_costs(cheapest, path, TOTAL_COST) <= 0)
                                continue;
                }

                /* We have a new best path */
                cheapest = path;
        }

        /* Return the best path, or NULL if we found no suitable candidate */
        return cheapest;
}

/*
 * accumulate_append_subpath
 *              Add a subpath to the list being built for an Append or MergeAppend
 *
 * It's possible that the child is itself an Append or MergeAppend path, in
 * which case we can "cut out the middleman" and just add its child paths to
 * our own list.  (We don't try to do this earlier because we need to apply
 * both levels of transformation to the quals.)
 *
 * Note that if we omit a child MergeAppend in this way, we are effectively
 * omitting a sort step, which seems fine: if the parent is to be an Append,
 * its result would be unsorted anyway, while if the parent is to be a
 * MergeAppend, there's no point in a separate sort on a child.
 */
static List *
accumulate_append_subpath(List *subpaths, Path *path)
{
        if (IsA(path, AppendPath))
        {
                AppendPath *apath = (AppendPath *) path;

                /* list_copy is important here to avoid sharing list substructure */
                return list_concat(subpaths, list_copy(apath->subpaths));
        }
        else if (IsA(path, MergeAppendPath))
        {
                MergeAppendPath *mpath = (MergeAppendPath *) path;

                /* list_copy is important here to avoid sharing list substructure */
                return list_concat(subpaths, list_copy(mpath->subpaths));
        }
        else
                return lappend(subpaths, path);
}

/*
 * standard_join_search
 *        Find possible joinpaths for a query by successively finding ways
 *        to join component relations into join relations.
 *
 * 'levels_needed' is the number of iterations needed, ie, the number of
 *              independent jointree items in the query.  This is > 1.
 *
 * 'initial_rels' is a list of RelOptInfo nodes for each independent
 *              jointree item.  These are the components to be joined together.
 *              Note that levels_needed == list_length(initial_rels).
 *
 * Returns the final level of join relations, i.e., the relation that is
 * the result of joining all the original relations together.
 * At least one implementation path must be provided for this relation and
 * all required sub-relations.
 *
 * To support loadable plugins that modify planner behavior by changing the
 * join searching algorithm, we provide a hook variable that lets a plugin
 * replace or supplement this function.  Any such hook must return the same
 * final join relation as the standard code would, but it might have a
 * different set of implementation paths attached, and only the sub-joinrels
 * needed for these paths need have been instantiated.
 *
 * Note to plugin authors: the functions invoked during standard_join_search()
 * modify root->join_rel_list and root->join_rel_hash.  If you want to do more
 * than one join-order search, you'll probably need to save and restore the
 * original states of those data structures.  See geqo_eval() for an example.
 */
RelOptInfo *
standard_join_search(PlannerInfo *root, int levels_needed, List *initial_rels)
{
        int                     lev;
        RelOptInfo *rel;

        /*
         * This function cannot be invoked recursively within any one planning
         * problem, so join_rel_level[] can't be in use already.
         */
        Assert(root->join_rel_level == NULL);

        /*
         * We employ a simple "dynamic programming" algorithm: we first find all
         * ways to build joins of two jointree items, then all ways to build joins
         * of three items (from two-item joins and single items), then four-item
         * joins, and so on until we have considered all ways to join all the
         * items into one rel.
         *
         * root->join_rel_level[j] is a list of all the j-item rels.  Initially we
         * set root->join_rel_level[1] to represent all the single-jointree-item
         * relations.
         */
        root->join_rel_level = (List **) palloc0((levels_needed + 1) * sizeof(List *));

        root->join_rel_level[1] = initial_rels;

        for (lev = 2; lev <= levels_needed; lev++)
        {
                ListCell   *lc;

                /*
                 * Determine all possible pairs of relations to be joined at this
                 * level, and build paths for making each one from every available
                 * pair of lower-level relations.
                 */
                join_search_one_level(root, lev);

                /*
                 * Run generate_gather_paths() for each just-processed joinrel.  We
                 * could not do this earlier because both regular and partial paths
                 * can get added to a particular joinrel at multiple times within
                 * join_search_one_level.  After that, we're done creating paths for
                 * the joinrel, so run set_cheapest().
                 */
                foreach(lc, root->join_rel_level[lev])
                {
                        rel = (RelOptInfo *) lfirst(lc);

                        /* Create GatherPaths for any useful partial paths for rel */
                        generate_gather_paths(root, rel);

                        /* Find and save the cheapest paths for this rel */
                        set_cheapest(rel);

#ifdef OPTIMIZER_DEBUG
                        debug_print_rel(root, rel);
#endif
                }
        }

        /*
         * We should have a single rel at the final level.
         */
        if (root->join_rel_level[levels_needed] == NIL)
                elog(ERROR, "failed to build any %d-way joins", levels_needed);
        Assert(list_length(root->join_rel_level[levels_needed]) == 1);

        rel = (RelOptInfo *) linitial(root->join_rel_level[levels_needed]);

        root->join_rel_level = NULL;

        return rel;
}

/*
 * join_search_one_level
 *        Consider ways to produce join relations containing exactly 'level'
 *        jointree items.  (This is one step of the dynamic-programming method
 *        embodied in standard_join_search.)  Join rel nodes for each feasible
 *        combination of lower-level rels are created and returned in a list.
 *        Implementation paths are created for each such joinrel, too.
 *
 * level: level of rels we want to make this time
 * root->join_rel_level[j], 1 <= j < level, is a list of rels containing j items
 *
 * The result is returned in root->join_rel_level[level].
 */
void
join_search_one_level(PlannerInfo *root, int level)
{
        List      **joinrels = root->join_rel_level;
        ListCell   *r;
        int                     k;

        Assert(joinrels[level] == NIL);

        /* Set join_cur_level so that new joinrels are added to proper list */
        root->join_cur_level = level;

        /*
         * First, consider left-sided and right-sided plans, in which rels of
         * exactly level-1 member relations are joined against initial relations.
         * We prefer to join using join clauses, but if we find a rel of level-1
         * members that has no join clauses, we will generate Cartesian-product
         * joins against all initial rels not already contained in it.
         */
        foreach(r, joinrels[level - 1])
        {
                RelOptInfo *old_rel = (RelOptInfo *) lfirst(r);

                if (old_rel->joininfo != NIL || old_rel->has_eclass_joins ||
                        has_join_restriction(root, old_rel))
                {
                        /*
                         * There are join clauses or join order restrictions relevant to
                         * this rel, so consider joins between this rel and (only) those
                         * initial rels it is linked to by a clause or restriction.
                         *
                         * At level 2 this condition is symmetric, so there is no need to
                         * look at initial rels before this one in the list; we already
                         * considered such joins when we were at the earlier rel.  (The
                         * mirror-image joins are handled automatically by make_join_rel.)
                         * In later passes (level > 2), we join rels of the previous level
                         * to each initial rel they don't already include but have a join
                         * clause or restriction with.
                         */
                        ListCell   *other_rels;

                        if (level == 2)         /* consider remaining initial rels */
                                other_rels = lnext(r);
                        else    /* consider all initial rels */
                                other_rels = list_head(joinrels[1]);

                        make_rels_by_clause_joins(root,
                                                                          old_rel,
                                                                          other_rels);
                }
                else
                {
                        /*
                         * Oops, we have a relation that is not joined to any other
                         * relation, either directly or by join-order restrictions.
                         * Cartesian product time.
                         *
                         * We consider a cartesian product with each not-already-included
                         * initial rel, whether it has other join clauses or not.  At
                         * level 2, if there are two or more clauseless initial rels, we
                         * will redundantly consider joining them in both directions; but
                         * such cases aren't common enough to justify adding complexity to
                         * avoid the duplicated effort.
                         */
                        make_rels_by_clauseless_joins(root,
                                                                                  old_rel,
                                                                                  list_head(joinrels[1]));
                }
        }

        /*
         * Now, consider "bushy plans" in which relations of k initial rels are
         * joined to relations of level-k initial rels, for 2 <= k <= level-2.
         *
         * We only consider bushy-plan joins for pairs of rels where there is a
         * suitable join clause (or join order restriction), in order to avoid
         * unreasonable growth of planning time.
         */
        for (k = 2;; k++)
        {
                int                     other_level = level - k;

                /*
                 * Since make_join_rel(x, y) handles both x,y and y,x cases, we only
                 * need to go as far as the halfway point.
                 */
                if (k > other_level)
                        break;

                foreach(r, joinrels[k])
                {
                        RelOptInfo *old_rel = (RelOptInfo *) lfirst(r);
                        ListCell   *other_rels;
                        ListCell   *r2;

                        /*
                         * We can ignore relations without join clauses here, unless they
                         * participate in join-order restrictions --- then we might have
                         * to force a bushy join plan.
                         */
                        if (old_rel->joininfo == NIL && !old_rel->has_eclass_joins &&
                                !has_join_restriction(root, old_rel))
                                continue;

                        if (k == other_level)
                                other_rels = lnext(r);  /* only consider remaining rels */
                        else
                                other_rels = list_head(joinrels[other_level]);

                        for_each_cell(r2, other_rels)
                        {
                                RelOptInfo *new_rel = (RelOptInfo *) lfirst(r2);

                                if (!bms_overlap(old_rel->relids, new_rel->relids))
                                {
                                        /*
                                         * OK, we can build a rel of the right level from this
                                         * pair of rels.  Do so if there is at least one relevant
                                         * join clause or join order restriction.
                                         */
                                        if (have_relevant_joinclause(root, old_rel, new_rel) ||
                                                have_join_order_restriction(root, old_rel, new_rel))
                                        {
                                                (void) make_join_rel(root, old_rel, new_rel);
                                        }
                                }
                        }
                }
        }

        /*----------
         * Last-ditch effort: if we failed to find any usable joins so far, force
         * a set of cartesian-product joins to be generated.  This handles the
         * special case where all the available rels have join clauses but we
         * cannot use any of those clauses yet.  This can only happen when we are
         * considering a join sub-problem (a sub-joinlist) and all the rels in the
         * sub-problem have only join clauses with rels outside the sub-problem.
         * An example is
         *
         *              SELECT ... FROM a INNER JOIN b ON TRUE, c, d, ...
         *              WHERE a.w = c.x and b.y = d.z;
         *
         * If the "a INNER JOIN b" sub-problem does not get flattened into the
         * upper level, we must be willing to make a cartesian join of a and b;
         * but the code above will not have done so, because it thought that both
         * a and b have joinclauses.  We consider only left-sided and right-sided
         * cartesian joins in this case (no bushy).
         *----------
         */
        if (joinrels[level] == NIL)
        {
                /*
                 * This loop is just like the first one, except we always call
                 * make_rels_by_clauseless_joins().
                 */
                foreach(r, joinrels[level - 1])
                {
                        RelOptInfo *old_rel = (RelOptInfo *) lfirst(r);

                        make_rels_by_clauseless_joins(root,
                                                                                  old_rel,
                                                                                  list_head(joinrels[1]));
                }

                /*----------
                 * When special joins are involved, there may be no legal way
                 * to make an N-way join for some values of N.  For example consider
                 *
                 * SELECT ... FROM t1 WHERE
                 *       x IN (SELECT ... FROM t2,t3 WHERE ...) AND
                 *       y IN (SELECT ... FROM t4,t5 WHERE ...)
                 *
                 * We will flatten this query to a 5-way join problem, but there are
                 * no 4-way joins that join_is_legal() will consider legal.  We have
                 * to accept failure at level 4 and go on to discover a workable
                 * bushy plan at level 5.
                 *
                 * However, if there are no special joins and no lateral references
                 * then join_is_legal() should never fail, and so the following sanity
                 * check is useful.
                 *----------
                 */
                if (joinrels[level] == NIL &&
                        root->join_info_list == NIL &&
                        !root->hasLateralRTEs)
                        elog(ERROR, "failed to build any %d-way joins", level);
        }
}

/*
 * make_rels_by_clause_joins
 *        Build joins between the given relation 'old_rel' and other relations
 *        that participate in join clauses that 'old_rel' also participates in
 *        (or participate in join-order restrictions with it).
 *        The join rels are returned in root->join_rel_level[join_cur_level].
 *
 * Note: at levels above 2 we will generate the same joined relation in
 * multiple ways --- for example (a join b) join c is the same RelOptInfo as
 * (b join c) join a, though the second case will add a different set of Paths
 * to it.  This is the reason for using the join_rel_level mechanism, which
 * automatically ensures that each new joinrel is only added to the list once.
 *
 * 'old_rel' is the relation entry for the relation to be joined
 * 'other_rels': the first cell in a linked list containing the other
 * rels to be considered for joining
 *
 * Currently, this is only used with initial rels in other_rels, but it
 * will work for joining to joinrels too.
 */
static void
make_rels_by_clause_joins(PlannerInfo *root,
                                                  RelOptInfo *old_rel,
                                                  ListCell *other_rels)
{
        ListCell   *l;

        for_each_cell(l, other_rels)
        {
                RelOptInfo *other_rel = (RelOptInfo *) lfirst(l);

                if (!bms_overlap(old_rel->relids, other_rel->relids) &&
                        (have_relevant_joinclause(root, old_rel, other_rel) ||
                         have_join_order_restriction(root, old_rel, other_rel)))
                {
                        (void) make_join_rel(root, old_rel, other_rel);
                }
        }
}

/*
 * make_rels_by_clauseless_joins
 *        Given a relation 'old_rel' and a list of other relations
 *        'other_rels', create a join relation between 'old_rel' and each
 *        member of 'other_rels' that isn't already included in 'old_rel'.
 *        The join rels are returned in root->join_rel_level[join_cur_level].
 *
 * 'old_rel' is the relation entry for the relation to be joined
 * 'other_rels': the first cell of a linked list containing the
 * other rels to be considered for joining
 *
 * Currently, this is only used with initial rels in other_rels, but it would
 * work for joining to joinrels too.
 */
static void
make_rels_by_clauseless_joins(PlannerInfo *root,
                                                          RelOptInfo *old_rel,
                                                          ListCell *other_rels)
{
        ListCell   *l;

        for_each_cell(l, other_rels)
        {
                RelOptInfo *other_rel = (RelOptInfo *) lfirst(l);

                if (!bms_overlap(other_rel->relids, old_rel->relids))
                {
                        (void) make_join_rel(root, old_rel, other_rel);
                }
        }
}

/*
 * join_is_legal
 *         Determine whether a proposed join is legal given the query's
 *         join order constraints; and if it is, determine the join type.
 *
 * Caller must supply not only the two rels, but the union of their relids.
 * (We could simplify the API by computing joinrelids locally, but this
 * would be redundant work in the normal path through make_join_rel.)
 *
 * On success, *sjinfo_p is set to NULL if this is to be a plain inner join,
 * else it's set to point to the associated SpecialJoinInfo node.  Also,
 * *reversed_p is set TRUE if the given relations need to be swapped to
 * match the SpecialJoinInfo node.
 */
static bool
join_is_legal(PlannerInfo *root, RelOptInfo *rel1, RelOptInfo *rel2,
                          Relids joinrelids,
                          SpecialJoinInfo **sjinfo_p, bool *reversed_p)
{
        SpecialJoinInfo *match_sjinfo;
        bool            reversed;
        bool            unique_ified;
        bool            must_be_leftjoin;
        ListCell   *l;

        /*
         * Ensure output params are set on failure return.  This is just to
         * suppress uninitialized-variable warnings from overly anal compilers.
         */
        *sjinfo_p = NULL;
        *reversed_p = false;

        /*
         * If we have any special joins, the proposed join might be illegal; and
         * in any case we have to determine its join type.  Scan the join info
         * list for matches and conflicts.
         */
        match_sjinfo = NULL;
        reversed = false;
        unique_ified = false;
        must_be_leftjoin = false;

        foreach(l, root->join_info_list)
        {
                SpecialJoinInfo *sjinfo = (SpecialJoinInfo *) lfirst(l);

                /*
                 * This special join is not relevant unless its RHS overlaps the
                 * proposed join.  (Check this first as a fast path for dismissing
                 * most irrelevant SJs quickly.)
                 */
                if (!bms_overlap(sjinfo->min_righthand, joinrelids))
                        continue;

                /*
                 * Also, not relevant if proposed join is fully contained within RHS
                 * (ie, we're still building up the RHS).
                 */
                if (bms_is_subset(joinrelids, sjinfo->min_righthand))
                        continue;

                /*
                 * Also, not relevant if SJ is already done within either input.
                 */
                if (bms_is_subset(sjinfo->min_lefthand, rel1->relids) &&
                        bms_is_subset(sjinfo->min_righthand, rel1->relids))
                        continue;
                if (bms_is_subset(sjinfo->min_lefthand, rel2->relids) &&
                        bms_is_subset(sjinfo->min_righthand, rel2->relids))
                        continue;

                /*
                 * If it's a semijoin and we already joined the RHS to any other rels
                 * within either input, then we must have unique-ified the RHS at that
                 * point (see below).  Therefore the semijoin is no longer relevant in
                 * this join path.
                 */
                if (sjinfo->jointype == JOIN_SEMI)
                {
                        if (bms_is_subset(sjinfo->syn_righthand, rel1->relids) &&
                                !bms_equal(sjinfo->syn_righthand, rel1->relids))
                                continue;
                        if (bms_is_subset(sjinfo->syn_righthand, rel2->relids) &&
                                !bms_equal(sjinfo->syn_righthand, rel2->relids))
                                continue;
                }

                /*
                 * If one input contains min_lefthand and the other contains
                 * min_righthand, then we can perform the SJ at this join.
                 *
                 * Reject if we get matches to more than one SJ; that implies we're
                 * considering something that's not really valid.
                 */
                if (bms_is_subset(sjinfo->min_lefthand, rel1->relids) &&
                        bms_is_subset(sjinfo->min_righthand, rel2->relids))
                {
                        if (match_sjinfo)
                                return false;   /* invalid join path */
                        match_sjinfo = sjinfo;
                        reversed = false;
                }
                else if (bms_is_subset(sjinfo->min_lefthand, rel2->relids) &&
                                 bms_is_subset(sjinfo->min_righthand, rel1->relids))
                {
                        if (match_sjinfo)
                                return false;   /* invalid join path */
                        match_sjinfo = sjinfo;
                        reversed = true;
                }
                else if (sjinfo->jointype == JOIN_SEMI &&
                                 bms_equal(sjinfo->syn_righthand, rel2->relids) &&
                                 create_unique_path(root, rel2, rel2->cheapest_total_path,
                                                                        sjinfo) != NULL)
                {
                        /*----------
                         * For a semijoin, we can join the RHS to anything else by
                         * unique-ifying the RHS (if the RHS can be unique-ified).
                         * We will only get here if we have the full RHS but less
                         * than min_lefthand on the LHS.
                         *
                         * The reason to consider such a join path is exemplified by
                         *      SELECT ... FROM a,b WHERE (a.x,b.y) IN (SELECT c1,c2 FROM c)
                         * If we insist on doing this as a semijoin we will first have
                         * to form the cartesian product of A*B.  But if we unique-ify
                         * C then the semijoin becomes a plain innerjoin and we can join
                         * in any order, eg C to A and then to B.  When C is much smaller
                         * than A and B this can be a huge win.  So we allow C to be
                         * joined to just A or just B here, and then make_join_rel has
                         * to handle the case properly.
                         *
                         * Note that actually we'll allow unique-ified C to be joined to
                         * some other relation D here, too.  That is legal, if usually not
                         * very sane, and this routine is only concerned with legality not
                         * with whether the join is good strategy.
                         *----------
                         */
                        if (match_sjinfo)
                                return false;   /* invalid join path */
                        match_sjinfo = sjinfo;
                        reversed = false;
                        unique_ified = true;
                }
                else if (sjinfo->jointype == JOIN_SEMI &&
                                 bms_equal(sjinfo->syn_righthand, rel1->relids) &&
                                 create_unique_path(root, rel1, rel1->cheapest_total_path,
                                                                        sjinfo) != NULL)
                {
                        /* Reversed semijoin case */
                        if (match_sjinfo)
                                return false;   /* invalid join path */
                        match_sjinfo = sjinfo;
                        reversed = true;
                        unique_ified = true;
                }
                else
                {
                        /*
                         * Otherwise, the proposed join overlaps the RHS but isn't a valid
                         * implementation of this SJ.  But don't panic quite yet: the RHS
                         * violation might have occurred previously, in one or both input
                         * relations, in which case we must have previously decided that
                         * it was OK to commute some other SJ with this one.  If we need
                         * to perform this join to finish building up the RHS, rejecting
                         * it could lead to not finding any plan at all.  (This can occur
                         * because of the heuristics elsewhere in this file that postpone
                         * clauseless joins: we might not consider doing a clauseless join
                         * within the RHS until after we've performed other, validly
                         * commutable SJs with one or both sides of the clauseless join.)
                         * This consideration boils down to the rule that if both inputs
                         * overlap the RHS, we can allow the join --- they are either
                         * fully within the RHS, or represent previously-allowed joins to
                         * rels outside it.
                         */
                        if (bms_overlap(rel1->relids, sjinfo->min_righthand) &&
                                bms_overlap(rel2->relids, sjinfo->min_righthand))
                                continue;               /* assume valid previous violation of RHS */

                        /*
                         * The proposed join could still be legal, but only if we're
                         * allowed to associate it into the RHS of this SJ.  That means
                         * this SJ must be a LEFT join (not SEMI or ANTI, and certainly
                         * not FULL) and the proposed join must not overlap the LHS.
                         */
                        if (sjinfo->jointype != JOIN_LEFT ||
                                bms_overlap(joinrelids, sjinfo->min_lefthand))
                                return false;   /* invalid join path */

                        /*
                         * To be valid, the proposed join must be a LEFT join; otherwise
                         * it can't associate into this SJ's RHS.  But we may not yet have
                         * found the SpecialJoinInfo matching the proposed join, so we
                         * can't test that yet.  Remember the requirement for later.
                         */
                        must_be_leftjoin = true;
                }
        }

        /*
         * Fail if violated any SJ's RHS and didn't match to a LEFT SJ: the
         * proposed join can't associate into an SJ's RHS.
         *
         * Also, fail if the proposed join's predicate isn't strict; we're
         * essentially checking to see if we can apply outer-join identity 3, and
         * that's a requirement.  (This check may be redundant with checks in
         * make_outerjoininfo, but I'm not quite sure, and it's cheap to test.)
         */
        if (must_be_leftjoin &&
                (match_sjinfo == NULL ||
                 match_sjinfo->jointype != JOIN_LEFT ||
                 !match_sjinfo->lhs_strict))
                return false;                   /* invalid join path */

        /*
         * We also have to check for constraints imposed by LATERAL references.
         */
        if (root->hasLateralRTEs)
        {
                bool            lateral_fwd;
                bool            lateral_rev;
                Relids          join_lateral_rels;

                /*
                 * The proposed rels could each contain lateral references to the
                 * other, in which case the join is impossible.  If there are lateral
                 * references in just one direction, then the join has to be done with
                 * a nestloop with the lateral referencer on the inside.  If the join
                 * matches an SJ that cannot be implemented by such a nestloop, the
                 * join is impossible.
                 *
                 * Also, if the lateral reference is only indirect, we should reject
                 * the join; whatever rel(s) the reference chain goes through must be
                 * joined to first.
                 *
                 * Another case that might keep us from building a valid plan is the
                 * implementation restriction described by have_dangerous_phv().
                 */
                lateral_fwd = bms_overlap(rel1->relids, rel2->lateral_relids);
                lateral_rev = bms_overlap(rel2->relids, rel1->lateral_relids);
                if (lateral_fwd && lateral_rev)
                        return false;           /* have lateral refs in both directions */
                if (lateral_fwd)
                {
                        /* has to be implemented as nestloop with rel1 on left */
                        if (match_sjinfo &&
                                (reversed ||
                                 unique_ified ||
                                 match_sjinfo->jointype == JOIN_FULL))
                                return false;   /* not implementable as nestloop */
                        /* check there is a direct reference from rel2 to rel1 */
                        if (!bms_overlap(rel1->relids, rel2->direct_lateral_relids))
                                return false;   /* only indirect refs, so reject */
                        /* check we won't have a dangerous PHV */
                        if (have_dangerous_phv(root, rel1->relids, rel2->lateral_relids))
                                return false;   /* might be unable to handle required PHV */
                }
                else if (lateral_rev)
                {
                        /* has to be implemented as nestloop with rel2 on left */
                        if (match_sjinfo &&
                                (!reversed ||
                                 unique_ified ||
                                 match_sjinfo->jointype == JOIN_FULL))
                                return false;   /* not implementable as nestloop */
                        /* check there is a direct reference from rel1 to rel2 */
                        if (!bms_overlap(rel2->relids, rel1->direct_lateral_relids))
                                return false;   /* only indirect refs, so reject */
                        /* check we won't have a dangerous PHV */
                        if (have_dangerous_phv(root, rel2->relids, rel1->lateral_relids))
                                return false;   /* might be unable to handle required PHV */
                }

                /*
                 * LATERAL references could also cause problems later on if we accept
                 * this join: if the join's minimum parameterization includes any rels
                 * that would have to be on the inside of an outer join with this join
                 * rel, then it's never going to be possible to build the complete
                 * query using this join.  We should reject this join not only because
                 * it'll save work, but because if we don't, the clauseless-join
                 * heuristics might think that legality of this join means that some
                 * other join rel need not be formed, and that could lead to failure
                 * to find any plan at all.  We have to consider not only rels that
                 * are directly on the inner side of an OJ with the joinrel, but also
                 * ones that are indirectly so, so search to find all such rels.
                 */
                join_lateral_rels = min_join_parameterization(root, joinrelids,
                                                                                                          rel1, rel2);
                if (join_lateral_rels)
                {
                        Relids          join_plus_rhs = bms_copy(joinrelids);
                        bool            more;

                        do
                        {
                                more = false;
                                foreach(l, root->join_info_list)
                                {
                                        SpecialJoinInfo *sjinfo = (SpecialJoinInfo *) lfirst(l);

                                        /* ignore full joins --- their ordering is predetermined */
                                        if (sjinfo->jointype == JOIN_FULL)
                                                continue;

                                        if (bms_overlap(sjinfo->min_lefthand, join_plus_rhs) &&
                                                !bms_is_subset(sjinfo->min_righthand, join_plus_rhs))
                                        {
                                                join_plus_rhs = bms_add_members(join_plus_rhs,
                                                                                                          sjinfo->min_righthand);
                                                more = true;
                                        }
                                }
                        } while (more);
                        if (bms_overlap(join_plus_rhs, join_lateral_rels))
                                return false;   /* will not be able to join to some RHS rel */
                }
        }

        /* Otherwise, it's a valid join */
        *sjinfo_p = match_sjinfo;
        *reversed_p = reversed;
        return true;
}

/*
 * has_join_restriction
 *              Detect whether the specified relation has join-order restrictions,
 *              due to being inside an outer join or an IN (sub-SELECT),
 *              or participating in any LATERAL references or multi-rel PHVs.
 *
 * Essentially, this tests whether have_join_order_restriction() could
 * succeed with this rel and some other one.  It's OK if we sometimes
 * say "true" incorrectly.  (Therefore, we don't bother with the relatively
 * expensive has_legal_joinclause test.)
 */
static bool
has_join_restriction(PlannerInfo *root, RelOptInfo *rel)
{
        ListCell   *l;

        if (rel->lateral_relids != NULL || rel->lateral_referencers != NULL)
                return true;

        foreach(l, root->placeholder_list)
        {
                PlaceHolderInfo *phinfo = (PlaceHolderInfo *) lfirst(l);

                if (bms_is_subset(rel->relids, phinfo->ph_eval_at) &&
                        !bms_equal(rel->relids, phinfo->ph_eval_at))
                        return true;
        }

        foreach(l, root->join_info_list)
        {
                SpecialJoinInfo *sjinfo = (SpecialJoinInfo *) lfirst(l);

                /* ignore full joins --- other mechanisms preserve their ordering */
                if (sjinfo->jointype == JOIN_FULL)
                        continue;

                /* ignore if SJ is already contained in rel */
                if (bms_is_subset(sjinfo->min_lefthand, rel->relids) &&
                        bms_is_subset(sjinfo->min_righthand, rel->relids))
                        continue;

                /* restricted if it overlaps LHS or RHS, but doesn't contain SJ */
                if (bms_overlap(sjinfo->min_lefthand, rel->relids) ||
                        bms_overlap(sjinfo->min_righthand, rel->relids))
                        return true;
        }

        return false;
}

/*
 * is_dummy_rel --- has relation been proven empty?
 */
static bool
is_dummy_rel(RelOptInfo *rel)
{
        return IS_DUMMY_REL(rel);
}

/*
 * Mark a relation as proven empty.
 *
 * During GEQO planning, this can get invoked more than once on the same
 * baserel struct, so it's worth checking to see if the rel is already marked
 * dummy.
 *
 * Also, when called during GEQO join planning, we are in a short-lived
 * memory context.  We must make sure that the dummy path attached to a
 * baserel survives the GEQO cycle, else the baserel is trashed for future
 * GEQO cycles.  On the other hand, when we are marking a joinrel during GEQO,
 * we don't want the dummy path to clutter the main planning context.  Upshot
 * is that the best solution is to explicitly make the dummy path in the same
 * context the given RelOptInfo is in.
 */
static void
mark_dummy_rel(RelOptInfo *rel)
{
        MemoryContext oldcontext;

        /* Already marked? */
        if (is_dummy_rel(rel))
                return;

        /* No, so choose correct context to make the dummy path in */
        oldcontext = MemoryContextSwitchTo(GetMemoryChunkContext(rel));

        /* Set dummy size estimate */
        rel->rows = 0;

        /* Evict any previously chosen paths */
        rel->pathlist = NIL;
        rel->partial_pathlist = NIL;

        /* Set up the dummy path */
        add_path(rel, (Path *) create_append_path(rel, NIL,
                                                                                          rel->lateral_relids,
                                                                                          0));

        /* Set or update cheapest_total_path and related fields */
        set_cheapest(rel);

        MemoryContextSwitchTo(oldcontext);
}

/*
 * restriction_is_constant_false --- is a restrictlist just false?
 *
 * In cases where a qual is provably constant false, eval_const_expressions
 * will generally have thrown away anything that's ANDed with it.  In outer
 * join situations this will leave us computing cartesian products only to
 * decide there's no match for an outer row, which is pretty stupid.  So,
 * we need to detect the case.
 *
 * If only_pushed_down is true, then consider only quals that are pushed-down
 * from the point of view of the joinrel.
 */
static bool
restriction_is_constant_false(List *restrictlist,
                                                          RelOptInfo *joinrel,
                                                          bool only_pushed_down)
{
        ListCell   *lc;

        /*
         * Despite the above comment, the restriction list we see here might
         * possibly have other members besides the FALSE constant, since other
         * quals could get "pushed down" to the outer join level.  So we check
         * each member of the list.
         */
        foreach(lc, restrictlist)
        {
                RestrictInfo *rinfo = (RestrictInfo *) lfirst(lc);

                Assert(IsA(rinfo, RestrictInfo));
                if (only_pushed_down && !RINFO_IS_PUSHED_DOWN(rinfo, joinrel->relids))
                        continue;

                if (rinfo->clause && IsA(rinfo->clause, Const))
                {
                        Const      *con = (Const *) rinfo->clause;

                        /* constant NULL is as good as constant FALSE for our purposes */
                        if (con->constisnull)
                                return true;
                        if (!DatumGetBool(con->constvalue))
                                return true;
                }
        }
        return false;
}