1 /*-------------------------------------------------------------------------
4 * The query optimizer external interface.
6 * Portions Copyright (c) 1996-2008, PostgreSQL Global Development Group
7 * Portions Copyright (c) 1994, Regents of the University of California
11 * $PostgreSQL: pgsql/src/backend/optimizer/plan/planner.c,v 1.232 2008/04/17 21:22:14 tgl Exp $
13 *-------------------------------------------------------------------------
20 #include "catalog/pg_operator.h"
21 #include "executor/executor.h"
22 #include "executor/nodeAgg.h"
23 #include "miscadmin.h"
24 #include "nodes/makefuncs.h"
25 #include "optimizer/clauses.h"
26 #include "optimizer/cost.h"
27 #include "optimizer/pathnode.h"
28 #include "optimizer/paths.h"
29 #include "optimizer/planmain.h"
30 #include "optimizer/planner.h"
31 #include "optimizer/prep.h"
32 #include "optimizer/subselect.h"
33 #include "optimizer/tlist.h"
34 #include "optimizer/var.h"
35 #ifdef OPTIMIZER_DEBUG
36 #include "nodes/print.h"
38 #include "parser/parse_expr.h"
39 #include "parser/parse_oper.h"
40 #include "parser/parsetree.h"
41 #include "utils/lsyscache.h"
42 #include "utils/syscache.h"
45 /* Hook for plugins to get control in planner() */
46 planner_hook_type planner_hook = NULL;
49 /* Expression kind codes for preprocess_expression */
50 #define EXPRKIND_QUAL 0
51 #define EXPRKIND_TARGET 1
52 #define EXPRKIND_RTFUNC 2
53 #define EXPRKIND_VALUES 3
54 #define EXPRKIND_LIMIT 4
55 #define EXPRKIND_ININFO 5
56 #define EXPRKIND_APPINFO 6
59 static Node *preprocess_expression(PlannerInfo *root, Node *expr, int kind);
60 static void preprocess_qual_conditions(PlannerInfo *root, Node *jtnode);
61 static Plan *inheritance_planner(PlannerInfo *root);
62 static Plan *grouping_planner(PlannerInfo *root, double tuple_fraction);
63 static bool is_dummy_plan(Plan *plan);
64 static double preprocess_limit(PlannerInfo *root,
65 double tuple_fraction,
66 int64 *offset_est, int64 *count_est);
67 static Oid *extract_grouping_ops(List *groupClause);
68 static bool choose_hashed_grouping(PlannerInfo *root,
69 double tuple_fraction, double limit_tuples,
70 Path *cheapest_path, Path *sorted_path,
71 Oid *groupOperators, double dNumGroups,
72 AggClauseCounts *agg_counts);
73 static List *make_subplanTargetList(PlannerInfo *root, List *tlist,
74 AttrNumber **groupColIdx, bool *need_tlist_eval);
75 static void locate_grouping_columns(PlannerInfo *root,
78 AttrNumber *groupColIdx);
79 static List *postprocess_setop_tlist(List *new_tlist, List *orig_tlist);
82 /*****************************************************************************
84 * Query optimizer entry point
86 * To support loadable plugins that monitor or modify planner behavior,
87 * we provide a hook variable that lets a plugin get control before and
88 * after the standard planning process. The plugin would normally call
91 * Note to plugin authors: standard_planner() scribbles on its Query input,
92 * so you'd better copy that data structure if you want to plan more than once.
94 *****************************************************************************/
96 planner(Query *parse, int cursorOptions, ParamListInfo boundParams)
101 result = (*planner_hook) (parse, cursorOptions, boundParams);
103 result = standard_planner(parse, cursorOptions, boundParams);
108 standard_planner(Query *parse, int cursorOptions, ParamListInfo boundParams)
112 double tuple_fraction;
118 /* Cursor options may come from caller or from DECLARE CURSOR stmt */
119 if (parse->utilityStmt &&
120 IsA(parse->utilityStmt, DeclareCursorStmt))
121 cursorOptions |= ((DeclareCursorStmt *) parse->utilityStmt)->options;
124 * Set up global state for this planner invocation. This data is needed
125 * across all levels of sub-Query that might exist in the given command,
126 * so we keep it in a separate struct that's linked to by each per-Query
129 glob = makeNode(PlannerGlobal);
131 glob->boundParams = boundParams;
132 glob->paramlist = NIL;
133 glob->subplans = NIL;
134 glob->subrtables = NIL;
135 glob->rewindPlanIDs = NULL;
136 glob->finalrtable = NIL;
137 glob->relationOids = NIL;
138 glob->transientPlan = false;
140 /* Determine what fraction of the plan is likely to be scanned */
141 if (cursorOptions & CURSOR_OPT_FAST_PLAN)
144 * We have no real idea how many tuples the user will ultimately FETCH
145 * from a cursor, but it seems a good bet that he doesn't want 'em
146 * all. Optimize for 10% retrieval (you gotta better number? Should
147 * this be a SETtable parameter?)
149 tuple_fraction = 0.10;
153 /* Default assumption is we need all the tuples */
154 tuple_fraction = 0.0;
157 /* primary planning entry point (may recurse for subqueries) */
158 top_plan = subquery_planner(glob, parse, 1, tuple_fraction, &root);
161 * If creating a plan for a scrollable cursor, make sure it can run
162 * backwards on demand. Add a Material node at the top at need.
164 if (cursorOptions & CURSOR_OPT_SCROLL)
166 if (!ExecSupportsBackwardScan(top_plan))
167 top_plan = materialize_finished_plan(top_plan);
170 /* final cleanup of the plan */
171 Assert(glob->finalrtable == NIL);
172 top_plan = set_plan_references(glob, top_plan, root->parse->rtable);
173 /* ... and the subplans (both regular subplans and initplans) */
174 Assert(list_length(glob->subplans) == list_length(glob->subrtables));
175 forboth(lp, glob->subplans, lr, glob->subrtables)
177 Plan *subplan = (Plan *) lfirst(lp);
178 List *subrtable = (List *) lfirst(lr);
180 lfirst(lp) = set_plan_references(glob, subplan, subrtable);
183 /* build the PlannedStmt result */
184 result = makeNode(PlannedStmt);
186 result->commandType = parse->commandType;
187 result->canSetTag = parse->canSetTag;
188 result->transientPlan = glob->transientPlan;
189 result->planTree = top_plan;
190 result->rtable = glob->finalrtable;
191 result->resultRelations = root->resultRelations;
192 result->utilityStmt = parse->utilityStmt;
193 result->intoClause = parse->intoClause;
194 result->subplans = glob->subplans;
195 result->rewindPlanIDs = glob->rewindPlanIDs;
196 result->returningLists = root->returningLists;
197 result->rowMarks = parse->rowMarks;
198 result->relationOids = glob->relationOids;
199 result->nParamExec = list_length(glob->paramlist);
205 /*--------------------
207 * Invokes the planner on a subquery. We recurse to here for each
208 * sub-SELECT found in the query tree.
210 * glob is the global state for the current planner run.
211 * parse is the querytree produced by the parser & rewriter.
212 * level is the current recursion depth (1 at the top-level Query).
213 * tuple_fraction is the fraction of tuples we expect will be retrieved.
214 * tuple_fraction is interpreted as explained for grouping_planner, below.
216 * If subroot isn't NULL, we pass back the query's final PlannerInfo struct;
217 * among other things this tells the output sort ordering of the plan.
219 * Basically, this routine does the stuff that should only be done once
220 * per Query object. It then calls grouping_planner. At one time,
221 * grouping_planner could be invoked recursively on the same Query object;
222 * that's not currently true, but we keep the separation between the two
223 * routines anyway, in case we need it again someday.
225 * subquery_planner will be called recursively to handle sub-Query nodes
226 * found within the query's expressions and rangetable.
228 * Returns a query plan.
229 *--------------------
232 subquery_planner(PlannerGlobal *glob, Query *parse,
233 Index level, double tuple_fraction,
234 PlannerInfo **subroot)
236 int num_old_subplans = list_length(glob->subplans);
242 /* Create a PlannerInfo data structure for this subquery */
243 root = makeNode(PlannerInfo);
246 root->query_level = level;
247 root->planner_cxt = CurrentMemoryContext;
248 root->init_plans = NIL;
249 root->eq_classes = NIL;
250 root->in_info_list = NIL;
251 root->append_rel_list = NIL;
254 * Look for IN clauses at the top level of WHERE, and transform them into
255 * joins. Note that this step only handles IN clauses originally at top
256 * level of WHERE; if we pull up any subqueries below, their INs are
257 * processed just before pulling them up.
259 if (parse->hasSubLinks)
260 parse->jointree->quals = pull_up_IN_clauses(root,
261 parse->jointree->quals);
264 * Scan the rangetable for set-returning functions, and inline them
265 * if possible (producing subqueries that might get pulled up next).
266 * Recursion issues here are handled in the same way as for IN clauses.
268 inline_set_returning_functions(root);
271 * Check to see if any subqueries in the rangetable can be merged into
274 parse->jointree = (FromExpr *)
275 pull_up_subqueries(root, (Node *) parse->jointree, false, false);
278 * Detect whether any rangetable entries are RTE_JOIN kind; if not, we can
279 * avoid the expense of doing flatten_join_alias_vars(). Also check for
280 * outer joins --- if none, we can skip reduce_outer_joins() and some
281 * other processing. This must be done after we have done
282 * pull_up_subqueries, of course.
284 * Note: if reduce_outer_joins manages to eliminate all outer joins,
285 * root->hasOuterJoins is not reset currently. This is OK since its
286 * purpose is merely to suppress unnecessary processing in simple cases.
288 root->hasJoinRTEs = false;
289 root->hasOuterJoins = false;
290 foreach(l, parse->rtable)
292 RangeTblEntry *rte = (RangeTblEntry *) lfirst(l);
294 if (rte->rtekind == RTE_JOIN)
296 root->hasJoinRTEs = true;
297 if (IS_OUTER_JOIN(rte->jointype))
299 root->hasOuterJoins = true;
300 /* Can quit scanning once we find an outer join */
307 * Expand any rangetable entries that are inheritance sets into "append
308 * relations". This can add entries to the rangetable, but they must be
309 * plain base relations not joins, so it's OK (and marginally more
310 * efficient) to do it after checking for join RTEs. We must do it after
311 * pulling up subqueries, else we'd fail to handle inherited tables in
314 expand_inherited_tables(root);
317 * Set hasHavingQual to remember if HAVING clause is present. Needed
318 * because preprocess_expression will reduce a constant-true condition to
319 * an empty qual list ... but "HAVING TRUE" is not a semantic no-op.
321 root->hasHavingQual = (parse->havingQual != NULL);
323 /* Clear this flag; might get set in distribute_qual_to_rels */
324 root->hasPseudoConstantQuals = false;
327 * Do expression preprocessing on targetlist and quals.
329 parse->targetList = (List *)
330 preprocess_expression(root, (Node *) parse->targetList,
333 parse->returningList = (List *)
334 preprocess_expression(root, (Node *) parse->returningList,
337 preprocess_qual_conditions(root, (Node *) parse->jointree);
339 parse->havingQual = preprocess_expression(root, parse->havingQual,
342 parse->limitOffset = preprocess_expression(root, parse->limitOffset,
344 parse->limitCount = preprocess_expression(root, parse->limitCount,
347 root->in_info_list = (List *)
348 preprocess_expression(root, (Node *) root->in_info_list,
350 root->append_rel_list = (List *)
351 preprocess_expression(root, (Node *) root->append_rel_list,
354 /* Also need to preprocess expressions for function and values RTEs */
355 foreach(l, parse->rtable)
357 RangeTblEntry *rte = (RangeTblEntry *) lfirst(l);
359 if (rte->rtekind == RTE_FUNCTION)
360 rte->funcexpr = preprocess_expression(root, rte->funcexpr,
362 else if (rte->rtekind == RTE_VALUES)
363 rte->values_lists = (List *)
364 preprocess_expression(root, (Node *) rte->values_lists,
369 * In some cases we may want to transfer a HAVING clause into WHERE. We
370 * cannot do so if the HAVING clause contains aggregates (obviously) or
371 * volatile functions (since a HAVING clause is supposed to be executed
372 * only once per group). Also, it may be that the clause is so expensive
373 * to execute that we're better off doing it only once per group, despite
374 * the loss of selectivity. This is hard to estimate short of doing the
375 * entire planning process twice, so we use a heuristic: clauses
376 * containing subplans are left in HAVING. Otherwise, we move or copy the
377 * HAVING clause into WHERE, in hopes of eliminating tuples before
378 * aggregation instead of after.
380 * If the query has explicit grouping then we can simply move such a
381 * clause into WHERE; any group that fails the clause will not be in the
382 * output because none of its tuples will reach the grouping or
383 * aggregation stage. Otherwise we must have a degenerate (variable-free)
384 * HAVING clause, which we put in WHERE so that query_planner() can use it
385 * in a gating Result node, but also keep in HAVING to ensure that we
386 * don't emit a bogus aggregated row. (This could be done better, but it
387 * seems not worth optimizing.)
389 * Note that both havingQual and parse->jointree->quals are in
390 * implicitly-ANDed-list form at this point, even though they are declared
394 foreach(l, (List *) parse->havingQual)
396 Node *havingclause = (Node *) lfirst(l);
398 if (contain_agg_clause(havingclause) ||
399 contain_volatile_functions(havingclause) ||
400 contain_subplans(havingclause))
402 /* keep it in HAVING */
403 newHaving = lappend(newHaving, havingclause);
405 else if (parse->groupClause)
407 /* move it to WHERE */
408 parse->jointree->quals = (Node *)
409 lappend((List *) parse->jointree->quals, havingclause);
413 /* put a copy in WHERE, keep it in HAVING */
414 parse->jointree->quals = (Node *)
415 lappend((List *) parse->jointree->quals,
416 copyObject(havingclause));
417 newHaving = lappend(newHaving, havingclause);
420 parse->havingQual = (Node *) newHaving;
423 * If we have any outer joins, try to reduce them to plain inner joins.
424 * This step is most easily done after we've done expression
427 if (root->hasOuterJoins)
428 reduce_outer_joins(root);
431 * Do the main planning. If we have an inherited target relation, that
432 * needs special processing, else go straight to grouping_planner.
434 if (parse->resultRelation &&
435 rt_fetch(parse->resultRelation, parse->rtable)->inh)
436 plan = inheritance_planner(root);
438 plan = grouping_planner(root, tuple_fraction);
441 * If any subplans were generated, or if we're inside a subplan, build
442 * initPlan list and extParam/allParam sets for plan nodes, and attach the
443 * initPlans to the top plan node.
445 if (list_length(glob->subplans) != num_old_subplans ||
446 root->query_level > 1)
447 SS_finalize_plan(root, plan);
449 /* Return internal info if caller wants it */
457 * preprocess_expression
458 * Do subquery_planner's preprocessing work for an expression,
459 * which can be a targetlist, a WHERE clause (including JOIN/ON
460 * conditions), or a HAVING clause.
463 preprocess_expression(PlannerInfo *root, Node *expr, int kind)
466 * Fall out quickly if expression is empty. This occurs often enough to
467 * be worth checking. Note that null->null is the correct conversion for
468 * implicit-AND result format, too.
474 * If the query has any join RTEs, replace join alias variables with
475 * base-relation variables. We must do this before sublink processing,
476 * else sublinks expanded out from join aliases wouldn't get processed. We
477 * can skip it in VALUES lists, however, since they can't contain any Vars
480 if (root->hasJoinRTEs && kind != EXPRKIND_VALUES)
481 expr = flatten_join_alias_vars(root, expr);
484 * Simplify constant expressions.
486 * Note: this also flattens nested AND and OR expressions into N-argument
487 * form. All processing of a qual expression after this point must be
488 * careful to maintain AND/OR flatness --- that is, do not generate a tree
489 * with AND directly under AND, nor OR directly under OR.
491 * Because this is a relatively expensive process, we skip it when the
492 * query is trivial, such as "SELECT 2+2;" or "INSERT ... VALUES()". The
493 * expression will only be evaluated once anyway, so no point in
494 * pre-simplifying; we can't execute it any faster than the executor can,
495 * and we will waste cycles copying the tree. Notice however that we
496 * still must do it for quals (to get AND/OR flatness); and if we are in a
497 * subquery we should not assume it will be done only once.
499 * For VALUES lists we never do this at all, again on the grounds that we
500 * should optimize for one-time evaluation.
502 if (kind != EXPRKIND_VALUES &&
503 (root->parse->jointree->fromlist != NIL ||
504 kind == EXPRKIND_QUAL ||
505 root->query_level > 1))
506 expr = eval_const_expressions(root, expr);
509 * If it's a qual or havingQual, canonicalize it.
511 if (kind == EXPRKIND_QUAL)
513 expr = (Node *) canonicalize_qual((Expr *) expr);
515 #ifdef OPTIMIZER_DEBUG
516 printf("After canonicalize_qual()\n");
521 /* Expand SubLinks to SubPlans */
522 if (root->parse->hasSubLinks)
523 expr = SS_process_sublinks(root, expr, (kind == EXPRKIND_QUAL));
526 * XXX do not insert anything here unless you have grokked the comments in
527 * SS_replace_correlation_vars ...
530 /* Replace uplevel vars with Param nodes (this IS possible in VALUES) */
531 if (root->query_level > 1)
532 expr = SS_replace_correlation_vars(root, expr);
535 * If it's a qual or havingQual, convert it to implicit-AND format. (We
536 * don't want to do this before eval_const_expressions, since the latter
537 * would be unable to simplify a top-level AND correctly. Also,
538 * SS_process_sublinks expects explicit-AND format.)
540 if (kind == EXPRKIND_QUAL)
541 expr = (Node *) make_ands_implicit((Expr *) expr);
547 * preprocess_qual_conditions
548 * Recursively scan the query's jointree and do subquery_planner's
549 * preprocessing work on each qual condition found therein.
552 preprocess_qual_conditions(PlannerInfo *root, Node *jtnode)
556 if (IsA(jtnode, RangeTblRef))
558 /* nothing to do here */
560 else if (IsA(jtnode, FromExpr))
562 FromExpr *f = (FromExpr *) jtnode;
565 foreach(l, f->fromlist)
566 preprocess_qual_conditions(root, lfirst(l));
568 f->quals = preprocess_expression(root, f->quals, EXPRKIND_QUAL);
570 else if (IsA(jtnode, JoinExpr))
572 JoinExpr *j = (JoinExpr *) jtnode;
574 preprocess_qual_conditions(root, j->larg);
575 preprocess_qual_conditions(root, j->rarg);
577 j->quals = preprocess_expression(root, j->quals, EXPRKIND_QUAL);
580 elog(ERROR, "unrecognized node type: %d",
581 (int) nodeTag(jtnode));
585 * inheritance_planner
586 * Generate a plan in the case where the result relation is an
589 * We have to handle this case differently from cases where a source relation
590 * is an inheritance set. Source inheritance is expanded at the bottom of the
591 * plan tree (see allpaths.c), but target inheritance has to be expanded at
592 * the top. The reason is that for UPDATE, each target relation needs a
593 * different targetlist matching its own column set. Also, for both UPDATE
594 * and DELETE, the executor needs the Append plan node at the top, else it
595 * can't keep track of which table is the current target table. Fortunately,
596 * the UPDATE/DELETE target can never be the nullable side of an outer join,
597 * so it's OK to generate the plan this way.
599 * Returns a query plan.
602 inheritance_planner(PlannerInfo *root)
604 Query *parse = root->parse;
605 int parentRTindex = parse->resultRelation;
606 List *subplans = NIL;
607 List *resultRelations = NIL;
608 List *returningLists = NIL;
614 foreach(l, root->append_rel_list)
616 AppendRelInfo *appinfo = (AppendRelInfo *) lfirst(l);
619 /* append_rel_list contains all append rels; ignore others */
620 if (appinfo->parent_relid != parentRTindex)
624 * Generate modified query with this rel as target. We have to be
625 * prepared to translate varnos in in_info_list as well as in the
628 memcpy(&subroot, root, sizeof(PlannerInfo));
629 subroot.parse = (Query *)
630 adjust_appendrel_attrs((Node *) parse,
632 subroot.in_info_list = (List *)
633 adjust_appendrel_attrs((Node *) root->in_info_list,
635 subroot.init_plans = NIL;
636 /* There shouldn't be any OJ info to translate, as yet */
637 Assert(subroot.oj_info_list == NIL);
640 subplan = grouping_planner(&subroot, 0.0 /* retrieve all tuples */ );
643 * If this child rel was excluded by constraint exclusion, exclude it
646 if (is_dummy_plan(subplan))
649 /* Save rtable and tlist from first rel for use below */
652 rtable = subroot.parse->rtable;
653 tlist = subplan->targetlist;
656 subplans = lappend(subplans, subplan);
658 /* Make sure any initplans from this rel get into the outer list */
659 root->init_plans = list_concat(root->init_plans, subroot.init_plans);
661 /* Build target-relations list for the executor */
662 resultRelations = lappend_int(resultRelations, appinfo->child_relid);
664 /* Build list of per-relation RETURNING targetlists */
665 if (parse->returningList)
667 Assert(list_length(subroot.returningLists) == 1);
668 returningLists = list_concat(returningLists,
669 subroot.returningLists);
673 root->resultRelations = resultRelations;
674 root->returningLists = returningLists;
676 /* Mark result as unordered (probably unnecessary) */
677 root->query_pathkeys = NIL;
680 * If we managed to exclude every child rel, return a dummy plan
684 root->resultRelations = list_make1_int(parentRTindex);
685 /* although dummy, it must have a valid tlist for executor */
686 tlist = preprocess_targetlist(root, parse->targetList);
687 return (Plan *) make_result(root,
689 (Node *) list_make1(makeBoolConst(false,
695 * Planning might have modified the rangetable, due to changes of the
696 * Query structures inside subquery RTEs. We have to ensure that this
697 * gets propagated back to the master copy. But can't do this until we
698 * are done planning, because all the calls to grouping_planner need
699 * virgin sub-Queries to work from. (We are effectively assuming that
700 * sub-Queries will get planned identically each time, or at least that
701 * the impacts on their rangetables will be the same each time.)
703 * XXX should clean this up someday
705 parse->rtable = rtable;
707 /* Suppress Append if there's only one surviving child rel */
708 if (list_length(subplans) == 1)
709 return (Plan *) linitial(subplans);
711 return (Plan *) make_append(subplans, true, tlist);
714 /*--------------------
716 * Perform planning steps related to grouping, aggregation, etc.
717 * This primarily means adding top-level processing to the basic
718 * query plan produced by query_planner.
720 * tuple_fraction is the fraction of tuples we expect will be retrieved
722 * tuple_fraction is interpreted as follows:
723 * 0: expect all tuples to be retrieved (normal case)
724 * 0 < tuple_fraction < 1: expect the given fraction of tuples available
725 * from the plan to be retrieved
726 * tuple_fraction >= 1: tuple_fraction is the absolute number of tuples
727 * expected to be retrieved (ie, a LIMIT specification)
729 * Returns a query plan. Also, root->query_pathkeys is returned as the
730 * actual output ordering of the plan (in pathkey format).
731 *--------------------
734 grouping_planner(PlannerInfo *root, double tuple_fraction)
736 Query *parse = root->parse;
737 List *tlist = parse->targetList;
738 int64 offset_est = 0;
740 double limit_tuples = -1.0;
742 List *current_pathkeys;
744 double dNumGroups = 0;
746 /* Tweak caller-supplied tuple_fraction if have LIMIT/OFFSET */
747 if (parse->limitCount || parse->limitOffset)
749 tuple_fraction = preprocess_limit(root, tuple_fraction,
750 &offset_est, &count_est);
753 * If we have a known LIMIT, and don't have an unknown OFFSET, we can
754 * estimate the effects of using a bounded sort.
756 if (count_est > 0 && offset_est >= 0)
757 limit_tuples = (double) count_est + (double) offset_est;
760 if (parse->setOperations)
762 List *set_sortclauses;
765 * If there's a top-level ORDER BY, assume we have to fetch all the
766 * tuples. This might seem too simplistic given all the hackery below
767 * to possibly avoid the sort ... but a nonzero tuple_fraction is only
768 * of use to plan_set_operations() when the setop is UNION ALL, and
769 * the result of UNION ALL is always unsorted.
771 if (parse->sortClause)
772 tuple_fraction = 0.0;
775 * Construct the plan for set operations. The result will not need
776 * any work except perhaps a top-level sort and/or LIMIT.
778 result_plan = plan_set_operations(root, tuple_fraction,
782 * Calculate pathkeys representing the sort order (if any) of the set
783 * operation's result. We have to do this before overwriting the sort
786 current_pathkeys = make_pathkeys_for_sortclauses(root,
788 result_plan->targetlist,
792 * We should not need to call preprocess_targetlist, since we must be
793 * in a SELECT query node. Instead, use the targetlist returned by
794 * plan_set_operations (since this tells whether it returned any
795 * resjunk columns!), and transfer any sort key information from the
798 Assert(parse->commandType == CMD_SELECT);
800 tlist = postprocess_setop_tlist(result_plan->targetlist, tlist);
803 * Can't handle FOR UPDATE/SHARE here (parser should have checked
804 * already, but let's make sure).
808 (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
809 errmsg("SELECT FOR UPDATE/SHARE is not allowed with UNION/INTERSECT/EXCEPT")));
812 * Calculate pathkeys that represent result ordering requirements
814 sort_pathkeys = make_pathkeys_for_sortclauses(root,
821 /* No set operations, do regular planning */
823 List *group_pathkeys;
824 AttrNumber *groupColIdx = NULL;
825 Oid *groupOperators = NULL;
826 bool need_tlist_eval = true;
832 AggClauseCounts agg_counts;
833 int numGroupCols = list_length(parse->groupClause);
834 bool use_hashed_grouping = false;
836 MemSet(&agg_counts, 0, sizeof(AggClauseCounts));
838 /* Preprocess targetlist */
839 tlist = preprocess_targetlist(root, tlist);
842 * Generate appropriate target list for subplan; may be different from
843 * tlist if grouping or aggregation is needed.
845 sub_tlist = make_subplanTargetList(root, tlist,
846 &groupColIdx, &need_tlist_eval);
849 * Calculate pathkeys that represent grouping/ordering requirements.
850 * Stash them in PlannerInfo so that query_planner can canonicalize
851 * them after EquivalenceClasses have been formed.
853 root->group_pathkeys =
854 make_pathkeys_for_sortclauses(root,
858 root->sort_pathkeys =
859 make_pathkeys_for_sortclauses(root,
865 * Will need actual number of aggregates for estimating costs.
867 * Note: we do not attempt to detect duplicate aggregates here; a
868 * somewhat-overestimated count is okay for our present purposes.
870 * Note: think not that we can turn off hasAggs if we find no aggs. It
871 * is possible for constant-expression simplification to remove all
872 * explicit references to aggs, but we still have to follow the
873 * aggregate semantics (eg, producing only one output row).
877 count_agg_clauses((Node *) tlist, &agg_counts);
878 count_agg_clauses(parse->havingQual, &agg_counts);
882 * Figure out whether we need a sorted result from query_planner.
884 * If we have a GROUP BY clause, then we want a result sorted properly
885 * for grouping. Otherwise, if there is an ORDER BY clause, we want
886 * to sort by the ORDER BY clause. (Note: if we have both, and ORDER
887 * BY is a superset of GROUP BY, it would be tempting to request sort
888 * by ORDER BY --- but that might just leave us failing to exploit an
889 * available sort order at all. Needs more thought...)
891 if (parse->groupClause)
892 root->query_pathkeys = root->group_pathkeys;
893 else if (parse->sortClause)
894 root->query_pathkeys = root->sort_pathkeys;
896 root->query_pathkeys = NIL;
899 * Generate the best unsorted and presorted paths for this Query (but
900 * note there may not be any presorted path). query_planner will also
901 * estimate the number of groups in the query, and canonicalize all
904 query_planner(root, sub_tlist, tuple_fraction, limit_tuples,
905 &cheapest_path, &sorted_path, &dNumGroups);
907 group_pathkeys = root->group_pathkeys;
908 sort_pathkeys = root->sort_pathkeys;
911 * If grouping, extract the grouping operators and decide whether we
912 * want to use hashed grouping.
914 if (parse->groupClause)
916 groupOperators = extract_grouping_ops(parse->groupClause);
917 use_hashed_grouping =
918 choose_hashed_grouping(root, tuple_fraction, limit_tuples,
919 cheapest_path, sorted_path,
920 groupOperators, dNumGroups,
923 /* Also convert # groups to long int --- but 'ware overflow! */
924 numGroups = (long) Min(dNumGroups, (double) LONG_MAX);
928 * Select the best path. If we are doing hashed grouping, we will
929 * always read all the input tuples, so use the cheapest-total path.
930 * Otherwise, trust query_planner's decision about which to use.
932 if (use_hashed_grouping || !sorted_path)
933 best_path = cheapest_path;
935 best_path = sorted_path;
938 * Check to see if it's possible to optimize MIN/MAX aggregates. If
939 * so, we will forget all the work we did so far to choose a "regular"
940 * path ... but we had to do it anyway to be able to tell which way is
943 result_plan = optimize_minmax_aggregates(root,
946 if (result_plan != NULL)
949 * optimize_minmax_aggregates generated the full plan, with the
950 * right tlist, and it has no sort order.
952 current_pathkeys = NIL;
957 * Normal case --- create a plan according to query_planner's
960 bool need_sort_for_grouping = false;
962 result_plan = create_plan(root, best_path);
963 current_pathkeys = best_path->pathkeys;
965 /* Detect if we'll need an explicit sort for grouping */
966 if (parse->groupClause && !use_hashed_grouping &&
967 !pathkeys_contained_in(group_pathkeys, current_pathkeys))
969 need_sort_for_grouping = true;
971 * Always override query_planner's tlist, so that we don't
972 * sort useless data from a "physical" tlist.
974 need_tlist_eval = true;
978 * create_plan() returns a plan with just a "flat" tlist of
979 * required Vars. Usually we need to insert the sub_tlist as the
980 * tlist of the top plan node. However, we can skip that if we
981 * determined that whatever query_planner chose to return will be
987 * If the top-level plan node is one that cannot do expression
988 * evaluation, we must insert a Result node to project the
991 if (!is_projection_capable_plan(result_plan))
993 result_plan = (Plan *) make_result(root,
1001 * Otherwise, just replace the subplan's flat tlist with
1002 * the desired tlist.
1004 result_plan->targetlist = sub_tlist;
1008 * Also, account for the cost of evaluation of the sub_tlist.
1010 * Up to now, we have only been dealing with "flat" tlists,
1011 * containing just Vars. So their evaluation cost is zero
1012 * according to the model used by cost_qual_eval() (or if you
1013 * prefer, the cost is factored into cpu_tuple_cost). Thus we
1014 * can avoid accounting for tlist cost throughout
1015 * query_planner() and subroutines. But now we've inserted a
1016 * tlist that might contain actual operators, sub-selects, etc
1017 * --- so we'd better account for its cost.
1019 * Below this point, any tlist eval cost for added-on nodes
1020 * should be accounted for as we create those nodes.
1021 * Presently, of the node types we can add on, only Agg and
1022 * Group project new tlists (the rest just copy their input
1023 * tuples) --- so make_agg() and make_group() are responsible
1024 * for computing the added cost.
1026 cost_qual_eval(&tlist_cost, sub_tlist, root);
1027 result_plan->startup_cost += tlist_cost.startup;
1028 result_plan->total_cost += tlist_cost.startup +
1029 tlist_cost.per_tuple * result_plan->plan_rows;
1034 * Since we're using query_planner's tlist and not the one
1035 * make_subplanTargetList calculated, we have to refigure any
1036 * grouping-column indexes make_subplanTargetList computed.
1038 locate_grouping_columns(root, tlist, result_plan->targetlist,
1043 * Insert AGG or GROUP node if needed, plus an explicit sort step
1046 * HAVING clause, if any, becomes qual of the Agg or Group node.
1048 if (use_hashed_grouping)
1050 /* Hashed aggregate plan --- no sort needed */
1051 result_plan = (Plan *) make_agg(root,
1053 (List *) parse->havingQual,
1061 /* Hashed aggregation produces randomly-ordered results */
1062 current_pathkeys = NIL;
1064 else if (parse->hasAggs)
1066 /* Plain aggregate plan --- sort if needed */
1067 AggStrategy aggstrategy;
1069 if (parse->groupClause)
1071 if (need_sort_for_grouping)
1073 result_plan = (Plan *)
1074 make_sort_from_groupcols(root,
1078 current_pathkeys = group_pathkeys;
1080 aggstrategy = AGG_SORTED;
1083 * The AGG node will not change the sort ordering of its
1084 * groups, so current_pathkeys describes the result too.
1089 aggstrategy = AGG_PLAIN;
1090 /* Result will be only one row anyway; no sort order */
1091 current_pathkeys = NIL;
1094 result_plan = (Plan *) make_agg(root,
1096 (List *) parse->havingQual,
1105 else if (parse->groupClause)
1108 * GROUP BY without aggregation, so insert a group node (plus
1109 * the appropriate sort node, if necessary).
1111 * Add an explicit sort if we couldn't make the path come out
1112 * the way the GROUP node needs it.
1114 if (need_sort_for_grouping)
1116 result_plan = (Plan *)
1117 make_sort_from_groupcols(root,
1121 current_pathkeys = group_pathkeys;
1124 result_plan = (Plan *) make_group(root,
1126 (List *) parse->havingQual,
1132 /* The Group node won't change sort ordering */
1134 else if (root->hasHavingQual)
1137 * No aggregates, and no GROUP BY, but we have a HAVING qual.
1138 * This is a degenerate case in which we are supposed to emit
1139 * either 0 or 1 row depending on whether HAVING succeeds.
1140 * Furthermore, there cannot be any variables in either HAVING
1141 * or the targetlist, so we actually do not need the FROM
1142 * table at all! We can just throw away the plan-so-far and
1143 * generate a Result node. This is a sufficiently unusual
1144 * corner case that it's not worth contorting the structure of
1145 * this routine to avoid having to generate the plan in the
1148 result_plan = (Plan *) make_result(root,
1153 } /* end of non-minmax-aggregate case */
1154 } /* end of if (setOperations) */
1157 * If we were not able to make the plan come out in the right order, add
1158 * an explicit sort step.
1160 if (parse->sortClause)
1162 if (!pathkeys_contained_in(sort_pathkeys, current_pathkeys))
1164 result_plan = (Plan *) make_sort_from_pathkeys(root,
1168 current_pathkeys = sort_pathkeys;
1173 * If there is a DISTINCT clause, add the UNIQUE node.
1175 if (parse->distinctClause)
1177 result_plan = (Plan *) make_unique(result_plan, parse->distinctClause);
1180 * If there was grouping or aggregation, leave plan_rows as-is (ie,
1181 * assume the result was already mostly unique). If not, use the
1182 * number of distinct-groups calculated by query_planner.
1184 if (!parse->groupClause && !root->hasHavingQual && !parse->hasAggs)
1185 result_plan->plan_rows = dNumGroups;
1189 * Finally, if there is a LIMIT/OFFSET clause, add the LIMIT node.
1191 if (parse->limitCount || parse->limitOffset)
1193 result_plan = (Plan *) make_limit(result_plan,
1201 * Deal with the RETURNING clause if any. It's convenient to pass the
1202 * returningList through setrefs.c now rather than at top level (if we
1203 * waited, handling inherited UPDATE/DELETE would be much harder).
1205 if (parse->returningList)
1209 Assert(parse->resultRelation);
1210 rlist = set_returning_clause_references(root->glob,
1211 parse->returningList,
1213 parse->resultRelation);
1214 root->returningLists = list_make1(rlist);
1217 root->returningLists = NIL;
1219 /* Compute result-relations list if needed */
1220 if (parse->resultRelation)
1221 root->resultRelations = list_make1_int(parse->resultRelation);
1223 root->resultRelations = NIL;
1226 * Return the actual output ordering in query_pathkeys for possible use by
1227 * an outer query level.
1229 root->query_pathkeys = current_pathkeys;
1235 * Detect whether a plan node is a "dummy" plan created when a relation
1236 * is deemed not to need scanning due to constraint exclusion.
1238 * Currently, such dummy plans are Result nodes with constant FALSE
1242 is_dummy_plan(Plan *plan)
1244 if (IsA(plan, Result))
1246 List *rcqual = (List *) ((Result *) plan)->resconstantqual;
1248 if (list_length(rcqual) == 1)
1250 Const *constqual = (Const *) linitial(rcqual);
1252 if (constqual && IsA(constqual, Const))
1254 if (!constqual->constisnull &&
1255 !DatumGetBool(constqual->constvalue))
1264 * preprocess_limit - do pre-estimation for LIMIT and/or OFFSET clauses
1266 * We try to estimate the values of the LIMIT/OFFSET clauses, and pass the
1267 * results back in *count_est and *offset_est. These variables are set to
1268 * 0 if the corresponding clause is not present, and -1 if it's present
1269 * but we couldn't estimate the value for it. (The "0" convention is OK
1270 * for OFFSET but a little bit bogus for LIMIT: effectively we estimate
1271 * LIMIT 0 as though it were LIMIT 1. But this is in line with the planner's
1272 * usual practice of never estimating less than one row.) These values will
1273 * be passed to make_limit, which see if you change this code.
1275 * The return value is the suitably adjusted tuple_fraction to use for
1276 * planning the query. This adjustment is not overridable, since it reflects
1277 * plan actions that grouping_planner() will certainly take, not assumptions
1281 preprocess_limit(PlannerInfo *root, double tuple_fraction,
1282 int64 *offset_est, int64 *count_est)
1284 Query *parse = root->parse;
1286 double limit_fraction;
1288 /* Should not be called unless LIMIT or OFFSET */
1289 Assert(parse->limitCount || parse->limitOffset);
1292 * Try to obtain the clause values. We use estimate_expression_value
1293 * primarily because it can sometimes do something useful with Params.
1295 if (parse->limitCount)
1297 est = estimate_expression_value(root, parse->limitCount);
1298 if (est && IsA(est, Const))
1300 if (((Const *) est)->constisnull)
1302 /* NULL indicates LIMIT ALL, ie, no limit */
1303 *count_est = 0; /* treat as not present */
1307 *count_est = DatumGetInt64(((Const *) est)->constvalue);
1308 if (*count_est <= 0)
1309 *count_est = 1; /* force to at least 1 */
1313 *count_est = -1; /* can't estimate */
1316 *count_est = 0; /* not present */
1318 if (parse->limitOffset)
1320 est = estimate_expression_value(root, parse->limitOffset);
1321 if (est && IsA(est, Const))
1323 if (((Const *) est)->constisnull)
1325 /* Treat NULL as no offset; the executor will too */
1326 *offset_est = 0; /* treat as not present */
1330 *offset_est = DatumGetInt64(((Const *) est)->constvalue);
1331 if (*offset_est < 0)
1332 *offset_est = 0; /* less than 0 is same as 0 */
1336 *offset_est = -1; /* can't estimate */
1339 *offset_est = 0; /* not present */
1341 if (*count_est != 0)
1344 * A LIMIT clause limits the absolute number of tuples returned.
1345 * However, if it's not a constant LIMIT then we have to guess; for
1346 * lack of a better idea, assume 10% of the plan's result is wanted.
1348 if (*count_est < 0 || *offset_est < 0)
1350 /* LIMIT or OFFSET is an expression ... punt ... */
1351 limit_fraction = 0.10;
1355 /* LIMIT (plus OFFSET, if any) is max number of tuples needed */
1356 limit_fraction = (double) *count_est + (double) *offset_est;
1360 * If we have absolute limits from both caller and LIMIT, use the
1361 * smaller value; likewise if they are both fractional. If one is
1362 * fractional and the other absolute, we can't easily determine which
1363 * is smaller, but we use the heuristic that the absolute will usually
1366 if (tuple_fraction >= 1.0)
1368 if (limit_fraction >= 1.0)
1371 tuple_fraction = Min(tuple_fraction, limit_fraction);
1375 /* caller absolute, limit fractional; use caller's value */
1378 else if (tuple_fraction > 0.0)
1380 if (limit_fraction >= 1.0)
1382 /* caller fractional, limit absolute; use limit */
1383 tuple_fraction = limit_fraction;
1387 /* both fractional */
1388 tuple_fraction = Min(tuple_fraction, limit_fraction);
1393 /* no info from caller, just use limit */
1394 tuple_fraction = limit_fraction;
1397 else if (*offset_est != 0 && tuple_fraction > 0.0)
1400 * We have an OFFSET but no LIMIT. This acts entirely differently
1401 * from the LIMIT case: here, we need to increase rather than decrease
1402 * the caller's tuple_fraction, because the OFFSET acts to cause more
1403 * tuples to be fetched instead of fewer. This only matters if we got
1404 * a tuple_fraction > 0, however.
1406 * As above, use 10% if OFFSET is present but unestimatable.
1408 if (*offset_est < 0)
1409 limit_fraction = 0.10;
1411 limit_fraction = (double) *offset_est;
1414 * If we have absolute counts from both caller and OFFSET, add them
1415 * together; likewise if they are both fractional. If one is
1416 * fractional and the other absolute, we want to take the larger, and
1417 * we heuristically assume that's the fractional one.
1419 if (tuple_fraction >= 1.0)
1421 if (limit_fraction >= 1.0)
1423 /* both absolute, so add them together */
1424 tuple_fraction += limit_fraction;
1428 /* caller absolute, limit fractional; use limit */
1429 tuple_fraction = limit_fraction;
1434 if (limit_fraction >= 1.0)
1436 /* caller fractional, limit absolute; use caller's value */
1440 /* both fractional, so add them together */
1441 tuple_fraction += limit_fraction;
1442 if (tuple_fraction >= 1.0)
1443 tuple_fraction = 0.0; /* assume fetch all */
1448 return tuple_fraction;
1452 * extract_grouping_ops - make an array of the equality operator OIDs
1453 * for the GROUP BY clause
1456 extract_grouping_ops(List *groupClause)
1458 int numCols = list_length(groupClause);
1460 Oid *groupOperators;
1463 groupOperators = (Oid *) palloc(sizeof(Oid) * numCols);
1465 foreach(glitem, groupClause)
1467 GroupClause *groupcl = (GroupClause *) lfirst(glitem);
1469 groupOperators[colno] = get_equality_op_for_ordering_op(groupcl->sortop);
1470 if (!OidIsValid(groupOperators[colno])) /* shouldn't happen */
1471 elog(ERROR, "could not find equality operator for ordering operator %u",
1476 return groupOperators;
1480 * choose_hashed_grouping - should we use hashed grouping?
1483 choose_hashed_grouping(PlannerInfo *root,
1484 double tuple_fraction, double limit_tuples,
1485 Path *cheapest_path, Path *sorted_path,
1486 Oid *groupOperators, double dNumGroups,
1487 AggClauseCounts *agg_counts)
1489 int numGroupCols = list_length(root->parse->groupClause);
1490 double cheapest_path_rows;
1491 int cheapest_path_width;
1493 List *current_pathkeys;
1499 * Check can't-do-it conditions, including whether the grouping operators
1500 * are hashjoinable. (We assume hashing is OK if they are marked
1501 * oprcanhash. If there isn't actually a supporting hash function, the
1502 * executor will complain at runtime.)
1504 * Executor doesn't support hashed aggregation with DISTINCT aggregates.
1505 * (Doing so would imply storing *all* the input values in the hash table,
1506 * which seems like a certain loser.)
1508 if (!enable_hashagg)
1510 if (agg_counts->numDistinctAggs != 0)
1512 for (i = 0; i < numGroupCols; i++)
1514 if (!op_hashjoinable(groupOperators[i]))
1519 * Don't do it if it doesn't look like the hashtable will fit into
1522 * Beware here of the possibility that cheapest_path->parent is NULL. This
1523 * could happen if user does something silly like SELECT 'foo' GROUP BY 1;
1525 if (cheapest_path->parent)
1527 cheapest_path_rows = cheapest_path->parent->rows;
1528 cheapest_path_width = cheapest_path->parent->width;
1532 cheapest_path_rows = 1; /* assume non-set result */
1533 cheapest_path_width = 100; /* arbitrary */
1536 /* Estimate per-hash-entry space at tuple width... */
1537 hashentrysize = MAXALIGN(cheapest_path_width) + MAXALIGN(sizeof(MinimalTupleData));
1538 /* plus space for pass-by-ref transition values... */
1539 hashentrysize += agg_counts->transitionSpace;
1540 /* plus the per-hash-entry overhead */
1541 hashentrysize += hash_agg_entry_size(agg_counts->numAggs);
1543 if (hashentrysize * dNumGroups > work_mem * 1024L)
1547 * See if the estimated cost is no more than doing it the other way. While
1548 * avoiding the need for sorted input is usually a win, the fact that the
1549 * output won't be sorted may be a loss; so we need to do an actual cost
1552 * We need to consider cheapest_path + hashagg [+ final sort] versus
1553 * either cheapest_path [+ sort] + group or agg [+ final sort] or
1554 * presorted_path + group or agg [+ final sort] where brackets indicate a
1555 * step that may not be needed. We assume query_planner() will have
1556 * returned a presorted path only if it's a winner compared to
1557 * cheapest_path for this purpose.
1559 * These path variables are dummies that just hold cost fields; we don't
1560 * make actual Paths for these steps.
1562 cost_agg(&hashed_p, root, AGG_HASHED, agg_counts->numAggs,
1563 numGroupCols, dNumGroups,
1564 cheapest_path->startup_cost, cheapest_path->total_cost,
1565 cheapest_path_rows);
1566 /* Result of hashed agg is always unsorted */
1567 if (root->sort_pathkeys)
1568 cost_sort(&hashed_p, root, root->sort_pathkeys, hashed_p.total_cost,
1569 dNumGroups, cheapest_path_width, limit_tuples);
1573 sorted_p.startup_cost = sorted_path->startup_cost;
1574 sorted_p.total_cost = sorted_path->total_cost;
1575 current_pathkeys = sorted_path->pathkeys;
1579 sorted_p.startup_cost = cheapest_path->startup_cost;
1580 sorted_p.total_cost = cheapest_path->total_cost;
1581 current_pathkeys = cheapest_path->pathkeys;
1583 if (!pathkeys_contained_in(root->group_pathkeys, current_pathkeys))
1585 cost_sort(&sorted_p, root, root->group_pathkeys, sorted_p.total_cost,
1586 cheapest_path_rows, cheapest_path_width, -1.0);
1587 current_pathkeys = root->group_pathkeys;
1590 if (root->parse->hasAggs)
1591 cost_agg(&sorted_p, root, AGG_SORTED, agg_counts->numAggs,
1592 numGroupCols, dNumGroups,
1593 sorted_p.startup_cost, sorted_p.total_cost,
1594 cheapest_path_rows);
1596 cost_group(&sorted_p, root, numGroupCols, dNumGroups,
1597 sorted_p.startup_cost, sorted_p.total_cost,
1598 cheapest_path_rows);
1599 /* The Agg or Group node will preserve ordering */
1600 if (root->sort_pathkeys &&
1601 !pathkeys_contained_in(root->sort_pathkeys, current_pathkeys))
1602 cost_sort(&sorted_p, root, root->sort_pathkeys, sorted_p.total_cost,
1603 dNumGroups, cheapest_path_width, limit_tuples);
1606 * Now make the decision using the top-level tuple fraction. First we
1607 * have to convert an absolute count (LIMIT) into fractional form.
1609 if (tuple_fraction >= 1.0)
1610 tuple_fraction /= dNumGroups;
1612 if (compare_fractional_path_costs(&hashed_p, &sorted_p,
1613 tuple_fraction) < 0)
1615 /* Hashed is cheaper, so use it */
1622 * make_subplanTargetList
1623 * Generate appropriate target list when grouping is required.
1625 * When grouping_planner inserts Aggregate, Group, or Result plan nodes
1626 * above the result of query_planner, we typically want to pass a different
1627 * target list to query_planner than the outer plan nodes should have.
1628 * This routine generates the correct target list for the subplan.
1630 * The initial target list passed from the parser already contains entries
1631 * for all ORDER BY and GROUP BY expressions, but it will not have entries
1632 * for variables used only in HAVING clauses; so we need to add those
1633 * variables to the subplan target list. Also, we flatten all expressions
1634 * except GROUP BY items into their component variables; the other expressions
1635 * will be computed by the inserted nodes rather than by the subplan.
1636 * For example, given a query like
1637 * SELECT a+b,SUM(c+d) FROM table GROUP BY a+b;
1638 * we want to pass this targetlist to the subplan:
1640 * where the a+b target will be used by the Sort/Group steps, and the
1641 * other targets will be used for computing the final results. (In the
1642 * above example we could theoretically suppress the a and b targets and
1643 * pass down only c,d,a+b, but it's not really worth the trouble to
1644 * eliminate simple var references from the subplan. We will avoid doing
1645 * the extra computation to recompute a+b at the outer level; see
1646 * fix_upper_expr() in setrefs.c.)
1648 * If we are grouping or aggregating, *and* there are no non-Var grouping
1649 * expressions, then the returned tlist is effectively dummy; we do not
1650 * need to force it to be evaluated, because all the Vars it contains
1651 * should be present in the output of query_planner anyway.
1653 * 'tlist' is the query's target list.
1654 * 'groupColIdx' receives an array of column numbers for the GROUP BY
1655 * expressions (if there are any) in the subplan's target list.
1656 * 'need_tlist_eval' is set true if we really need to evaluate the
1659 * The result is the targetlist to be passed to the subplan.
1663 make_subplanTargetList(PlannerInfo *root,
1665 AttrNumber **groupColIdx,
1666 bool *need_tlist_eval)
1668 Query *parse = root->parse;
1673 *groupColIdx = NULL;
1676 * If we're not grouping or aggregating, there's nothing to do here;
1677 * query_planner should receive the unmodified target list.
1679 if (!parse->hasAggs && !parse->groupClause && !root->hasHavingQual)
1681 *need_tlist_eval = true;
1686 * Otherwise, start with a "flattened" tlist (having just the vars
1687 * mentioned in the targetlist and HAVING qual --- but not upper- level
1688 * Vars; they will be replaced by Params later on).
1690 sub_tlist = flatten_tlist(tlist);
1691 extravars = pull_var_clause(parse->havingQual, false);
1692 sub_tlist = add_to_flat_tlist(sub_tlist, extravars);
1693 list_free(extravars);
1694 *need_tlist_eval = false; /* only eval if not flat tlist */
1697 * If grouping, create sub_tlist entries for all GROUP BY expressions
1698 * (GROUP BY items that are simple Vars should be in the list already),
1699 * and make an array showing where the group columns are in the sub_tlist.
1701 numCols = list_length(parse->groupClause);
1705 AttrNumber *grpColIdx;
1708 grpColIdx = (AttrNumber *) palloc(sizeof(AttrNumber) * numCols);
1709 *groupColIdx = grpColIdx;
1711 foreach(gl, parse->groupClause)
1713 GroupClause *grpcl = (GroupClause *) lfirst(gl);
1714 Node *groupexpr = get_sortgroupclause_expr(grpcl, tlist);
1715 TargetEntry *te = NULL;
1718 /* Find or make a matching sub_tlist entry */
1719 foreach(sl, sub_tlist)
1721 te = (TargetEntry *) lfirst(sl);
1722 if (equal(groupexpr, te->expr))
1727 te = makeTargetEntry((Expr *) groupexpr,
1728 list_length(sub_tlist) + 1,
1731 sub_tlist = lappend(sub_tlist, te);
1732 *need_tlist_eval = true; /* it's not flat anymore */
1735 /* and save its resno */
1736 grpColIdx[keyno++] = te->resno;
1744 * locate_grouping_columns
1745 * Locate grouping columns in the tlist chosen by query_planner.
1747 * This is only needed if we don't use the sub_tlist chosen by
1748 * make_subplanTargetList. We have to forget the column indexes found
1749 * by that routine and re-locate the grouping vars in the real sub_tlist.
1752 locate_grouping_columns(PlannerInfo *root,
1755 AttrNumber *groupColIdx)
1761 * No work unless grouping.
1763 if (!root->parse->groupClause)
1765 Assert(groupColIdx == NULL);
1768 Assert(groupColIdx != NULL);
1770 foreach(gl, root->parse->groupClause)
1772 GroupClause *grpcl = (GroupClause *) lfirst(gl);
1773 Node *groupexpr = get_sortgroupclause_expr(grpcl, tlist);
1774 TargetEntry *te = NULL;
1777 foreach(sl, sub_tlist)
1779 te = (TargetEntry *) lfirst(sl);
1780 if (equal(groupexpr, te->expr))
1784 elog(ERROR, "failed to locate grouping columns");
1786 groupColIdx[keyno++] = te->resno;
1791 * postprocess_setop_tlist
1792 * Fix up targetlist returned by plan_set_operations().
1794 * We need to transpose sort key info from the orig_tlist into new_tlist.
1795 * NOTE: this would not be good enough if we supported resjunk sort keys
1796 * for results of set operations --- then, we'd need to project a whole
1797 * new tlist to evaluate the resjunk columns. For now, just ereport if we
1798 * find any resjunk columns in orig_tlist.
1801 postprocess_setop_tlist(List *new_tlist, List *orig_tlist)
1804 ListCell *orig_tlist_item = list_head(orig_tlist);
1806 foreach(l, new_tlist)
1808 TargetEntry *new_tle = (TargetEntry *) lfirst(l);
1809 TargetEntry *orig_tle;
1811 /* ignore resjunk columns in setop result */
1812 if (new_tle->resjunk)
1815 Assert(orig_tlist_item != NULL);
1816 orig_tle = (TargetEntry *) lfirst(orig_tlist_item);
1817 orig_tlist_item = lnext(orig_tlist_item);
1818 if (orig_tle->resjunk) /* should not happen */
1819 elog(ERROR, "resjunk output columns are not implemented");
1820 Assert(new_tle->resno == orig_tle->resno);
1821 new_tle->ressortgroupref = orig_tle->ressortgroupref;
1823 if (orig_tlist_item != NULL)
1824 elog(ERROR, "resjunk output columns are not implemented");