1 /*-------------------------------------------------------------------------
4 * The query optimizer external interface.
6 * Portions Copyright (c) 1996-2005, 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 1.191 2005/08/18 17:51:11 tgl Exp $
13 *-------------------------------------------------------------------------
20 #include "catalog/pg_operator.h"
21 #include "catalog/pg_type.h"
22 #include "executor/executor.h"
23 #include "executor/nodeAgg.h"
24 #include "miscadmin.h"
25 #include "nodes/makefuncs.h"
26 #ifdef OPTIMIZER_DEBUG
27 #include "nodes/print.h"
29 #include "optimizer/clauses.h"
30 #include "optimizer/cost.h"
31 #include "optimizer/pathnode.h"
32 #include "optimizer/paths.h"
33 #include "optimizer/planmain.h"
34 #include "optimizer/planner.h"
35 #include "optimizer/prep.h"
36 #include "optimizer/subselect.h"
37 #include "optimizer/tlist.h"
38 #include "optimizer/var.h"
39 #include "parser/parsetree.h"
40 #include "parser/parse_expr.h"
41 #include "parser/parse_oper.h"
42 #include "utils/selfuncs.h"
43 #include "utils/syscache.h"
46 ParamListInfo PlannerBoundParamList = NULL; /* current boundParams */
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_LIMIT 3
54 #define EXPRKIND_ININFO 4
57 static Node *preprocess_expression(PlannerInfo *root, Node *expr, int kind);
58 static void preprocess_qual_conditions(PlannerInfo *root, Node *jtnode);
59 static Plan *inheritance_planner(PlannerInfo *root, List *inheritlist);
60 static Plan *grouping_planner(PlannerInfo *root, double tuple_fraction);
61 static double preprocess_limit(PlannerInfo *root,
62 double tuple_fraction,
63 int *offset_est, int *count_est);
64 static bool choose_hashed_grouping(PlannerInfo *root, double tuple_fraction,
65 Path *cheapest_path, Path *sorted_path,
66 List *sort_pathkeys, List *group_pathkeys,
67 double dNumGroups, AggClauseCounts *agg_counts);
68 static bool hash_safe_grouping(PlannerInfo *root);
69 static List *make_subplanTargetList(PlannerInfo *root, List *tlist,
70 AttrNumber **groupColIdx, bool *need_tlist_eval);
71 static void locate_grouping_columns(PlannerInfo *root,
74 AttrNumber *groupColIdx);
75 static List *postprocess_setop_tlist(List *new_tlist, List *orig_tlist);
78 /*****************************************************************************
80 * Query optimizer entry point
82 *****************************************************************************/
84 planner(Query *parse, bool isCursor, int cursorOptions,
85 ParamListInfo boundParams)
87 double tuple_fraction;
89 Index save_PlannerQueryLevel;
90 List *save_PlannerParamList;
91 ParamListInfo save_PlannerBoundParamList;
94 * The planner can be called recursively (an example is when
95 * eval_const_expressions tries to pre-evaluate an SQL function). So,
96 * these global state variables must be saved and restored.
98 * Query level and the param list cannot be moved into the per-query
99 * PlannerInfo structure since their whole purpose is communication
100 * across multiple sub-queries. Also, boundParams is explicitly info
101 * from outside the query, and so is likewise better handled as a global
104 * Note we do NOT save and restore PlannerPlanId: it exists to assign
105 * unique IDs to SubPlan nodes, and we want those IDs to be unique for
106 * the life of a backend. Also, PlannerInitPlan is saved/restored in
107 * subquery_planner, not here.
109 save_PlannerQueryLevel = PlannerQueryLevel;
110 save_PlannerParamList = PlannerParamList;
111 save_PlannerBoundParamList = PlannerBoundParamList;
113 /* Initialize state for handling outer-level references and params */
114 PlannerQueryLevel = 0; /* will be 1 in top-level subquery_planner */
115 PlannerParamList = NIL;
116 PlannerBoundParamList = boundParams;
118 /* Determine what fraction of the plan is likely to be scanned */
122 * We have no real idea how many tuples the user will ultimately
123 * FETCH from a cursor, but it seems a good bet that he doesn't
124 * want 'em all. Optimize for 10% retrieval (you gotta better
125 * number? Should this be a SETtable parameter?)
127 tuple_fraction = 0.10;
131 /* Default assumption is we need all the tuples */
132 tuple_fraction = 0.0;
135 /* primary planning entry point (may recurse for subqueries) */
136 result_plan = subquery_planner(parse, tuple_fraction, NULL);
138 /* check we popped out the right number of levels */
139 Assert(PlannerQueryLevel == 0);
142 * If creating a plan for a scrollable cursor, make sure it can run
143 * backwards on demand. Add a Material node at the top at need.
145 if (isCursor && (cursorOptions & CURSOR_OPT_SCROLL))
147 if (!ExecSupportsBackwardScan(result_plan))
148 result_plan = materialize_finished_plan(result_plan);
151 /* final cleanup of the plan */
152 result_plan = set_plan_references(result_plan, parse->rtable);
154 /* executor wants to know total number of Params used overall */
155 result_plan->nParamExec = list_length(PlannerParamList);
157 /* restore state for outer planner, if any */
158 PlannerQueryLevel = save_PlannerQueryLevel;
159 PlannerParamList = save_PlannerParamList;
160 PlannerBoundParamList = save_PlannerBoundParamList;
166 /*--------------------
168 * Invokes the planner on a subquery. We recurse to here for each
169 * sub-SELECT found in the query tree.
171 * parse is the querytree produced by the parser & rewriter.
172 * tuple_fraction is the fraction of tuples we expect will be retrieved.
173 * tuple_fraction is interpreted as explained for grouping_planner, below.
175 * If subquery_pathkeys isn't NULL, it receives a list of pathkeys indicating
176 * the output sort ordering of the completed plan.
178 * Basically, this routine does the stuff that should only be done once
179 * per Query object. It then calls grouping_planner. At one time,
180 * grouping_planner could be invoked recursively on the same Query object;
181 * that's not currently true, but we keep the separation between the two
182 * routines anyway, in case we need it again someday.
184 * subquery_planner will be called recursively to handle sub-Query nodes
185 * found within the query's expressions and rangetable.
187 * Returns a query plan.
188 *--------------------
191 subquery_planner(Query *parse, double tuple_fraction,
192 List **subquery_pathkeys)
194 List *saved_initplan = PlannerInitPlan;
195 int saved_planid = PlannerPlanId;
202 /* Set up for a new level of subquery */
204 PlannerInitPlan = NIL;
206 /* Create a PlannerInfo data structure for this subquery */
207 root = makeNode(PlannerInfo);
211 * Look for IN clauses at the top level of WHERE, and transform them
212 * into joins. Note that this step only handles IN clauses originally
213 * at top level of WHERE; if we pull up any subqueries in the next
214 * step, their INs are processed just before pulling them up.
216 root->in_info_list = NIL;
217 if (parse->hasSubLinks)
218 parse->jointree->quals = pull_up_IN_clauses(root,
219 parse->jointree->quals);
222 * Check to see if any subqueries in the rangetable can be merged into
225 parse->jointree = (FromExpr *)
226 pull_up_subqueries(root, (Node *) parse->jointree, false);
229 * Detect whether any rangetable entries are RTE_JOIN kind; if not, we
230 * can avoid the expense of doing flatten_join_alias_vars(). Also
231 * check for outer joins --- if none, we can skip reduce_outer_joins()
232 * and some other processing. This must be done after we have done
233 * pull_up_subqueries, of course.
235 * Note: if reduce_outer_joins manages to eliminate all outer joins,
236 * root->hasOuterJoins is not reset currently. This is OK since its
237 * purpose is merely to suppress unnecessary processing in simple cases.
239 root->hasJoinRTEs = false;
240 root->hasOuterJoins = false;
241 foreach(l, parse->rtable)
243 RangeTblEntry *rte = (RangeTblEntry *) lfirst(l);
245 if (rte->rtekind == RTE_JOIN)
247 root->hasJoinRTEs = true;
248 if (IS_OUTER_JOIN(rte->jointype))
250 root->hasOuterJoins = true;
251 /* Can quit scanning once we find an outer join */
258 * Set hasHavingQual to remember if HAVING clause is present. Needed
259 * because preprocess_expression will reduce a constant-true condition
260 * to an empty qual list ... but "HAVING TRUE" is not a semantic no-op.
262 root->hasHavingQual = (parse->havingQual != NULL);
265 * Do expression preprocessing on targetlist and quals.
267 parse->targetList = (List *)
268 preprocess_expression(root, (Node *) parse->targetList,
271 preprocess_qual_conditions(root, (Node *) parse->jointree);
273 parse->havingQual = preprocess_expression(root, parse->havingQual,
276 parse->limitOffset = preprocess_expression(root, parse->limitOffset,
278 parse->limitCount = preprocess_expression(root, parse->limitCount,
281 root->in_info_list = (List *)
282 preprocess_expression(root, (Node *) root->in_info_list,
285 /* Also need to preprocess expressions for function RTEs */
286 foreach(l, parse->rtable)
288 RangeTblEntry *rte = (RangeTblEntry *) lfirst(l);
290 if (rte->rtekind == RTE_FUNCTION)
291 rte->funcexpr = preprocess_expression(root, rte->funcexpr,
296 * In some cases we may want to transfer a HAVING clause into WHERE.
297 * We cannot do so if the HAVING clause contains aggregates (obviously)
298 * or volatile functions (since a HAVING clause is supposed to be executed
299 * only once per group). Also, it may be that the clause is so expensive
300 * to execute that we're better off doing it only once per group, despite
301 * the loss of selectivity. This is hard to estimate short of doing the
302 * entire planning process twice, so we use a heuristic: clauses
303 * containing subplans are left in HAVING. Otherwise, we move or copy
304 * the HAVING clause into WHERE, in hopes of eliminating tuples before
305 * aggregation instead of after.
307 * If the query has explicit grouping then we can simply move such a
308 * clause into WHERE; any group that fails the clause will not be
309 * in the output because none of its tuples will reach the grouping
310 * or aggregation stage. Otherwise we must have a degenerate
311 * (variable-free) HAVING clause, which we put in WHERE so that
312 * query_planner() can use it in a gating Result node, but also keep
313 * in HAVING to ensure that we don't emit a bogus aggregated row.
314 * (This could be done better, but it seems not worth optimizing.)
316 * Note that both havingQual and parse->jointree->quals are in
317 * implicitly-ANDed-list form at this point, even though they are
318 * declared as Node *.
321 foreach(l, (List *) parse->havingQual)
323 Node *havingclause = (Node *) lfirst(l);
325 if (contain_agg_clause(havingclause) ||
326 contain_volatile_functions(havingclause) ||
327 contain_subplans(havingclause))
329 /* keep it in HAVING */
330 newHaving = lappend(newHaving, havingclause);
332 else if (parse->groupClause)
334 /* move it to WHERE */
335 parse->jointree->quals = (Node *)
336 lappend((List *) parse->jointree->quals, havingclause);
340 /* put a copy in WHERE, keep it in HAVING */
341 parse->jointree->quals = (Node *)
342 lappend((List *) parse->jointree->quals,
343 copyObject(havingclause));
344 newHaving = lappend(newHaving, havingclause);
347 parse->havingQual = (Node *) newHaving;
350 * If we have any outer joins, try to reduce them to plain inner
351 * joins. This step is most easily done after we've done expression
354 if (root->hasOuterJoins)
355 reduce_outer_joins(root);
358 * See if we can simplify the jointree; opportunities for this may
359 * come from having pulled up subqueries, or from flattening explicit
360 * JOIN syntax. We must do this after flattening JOIN alias
361 * variables, since eliminating explicit JOIN nodes from the jointree
362 * will cause get_relids_for_join() to fail. But it should happen
363 * after reduce_outer_joins, anyway.
365 parse->jointree = (FromExpr *)
366 simplify_jointree(root, (Node *) parse->jointree);
369 * Do the main planning. If we have an inherited target relation,
370 * that needs special processing, else go straight to
373 if (parse->resultRelation &&
374 (lst = expand_inherited_rtentry(root, parse->resultRelation)) != NIL)
375 plan = inheritance_planner(root, lst);
377 plan = grouping_planner(root, tuple_fraction);
380 * If any subplans were generated, or if we're inside a subplan, build
381 * initPlan list and extParam/allParam sets for plan nodes, and attach
382 * the initPlans to the top plan node.
384 if (PlannerPlanId != saved_planid || PlannerQueryLevel > 1)
385 SS_finalize_plan(plan, parse->rtable);
387 /* Return sort ordering info if caller wants it */
388 if (subquery_pathkeys)
389 *subquery_pathkeys = root->query_pathkeys;
391 /* Return to outer subquery context */
393 PlannerInitPlan = saved_initplan;
394 /* we do NOT restore PlannerPlanId; that's not an oversight! */
400 * preprocess_expression
401 * Do subquery_planner's preprocessing work for an expression,
402 * which can be a targetlist, a WHERE clause (including JOIN/ON
403 * conditions), or a HAVING clause.
406 preprocess_expression(PlannerInfo *root, Node *expr, int kind)
409 * Fall out quickly if expression is empty. This occurs often enough
410 * to be worth checking. Note that null->null is the correct conversion
411 * for implicit-AND result format, too.
417 * If the query has any join RTEs, replace join alias variables with
418 * base-relation variables. We must do this before sublink processing,
419 * else sublinks expanded out from join aliases wouldn't get
422 if (root->hasJoinRTEs)
423 expr = flatten_join_alias_vars(root, expr);
426 * Simplify constant expressions.
428 * Note: this also flattens nested AND and OR expressions into N-argument
429 * form. All processing of a qual expression after this point must be
430 * careful to maintain AND/OR flatness --- that is, do not generate a tree
431 * with AND directly under AND, nor OR directly under OR.
433 * Because this is a relatively expensive process, we skip it when the
434 * query is trivial, such as "SELECT 2+2;" or "INSERT ... VALUES()".
435 * The expression will only be evaluated once anyway, so no point in
436 * pre-simplifying; we can't execute it any faster than the executor can,
437 * and we will waste cycles copying the tree. Notice however that we
438 * still must do it for quals (to get AND/OR flatness); and if we are
439 * in a subquery we should not assume it will be done only once.
441 if (root->parse->jointree->fromlist != NIL ||
442 kind == EXPRKIND_QUAL ||
443 PlannerQueryLevel > 1)
444 expr = eval_const_expressions(expr);
447 * If it's a qual or havingQual, canonicalize it.
449 if (kind == EXPRKIND_QUAL)
451 expr = (Node *) canonicalize_qual((Expr *) expr);
453 #ifdef OPTIMIZER_DEBUG
454 printf("After canonicalize_qual()\n");
459 /* Expand SubLinks to SubPlans */
460 if (root->parse->hasSubLinks)
461 expr = SS_process_sublinks(expr, (kind == EXPRKIND_QUAL));
464 * XXX do not insert anything here unless you have grokked the
465 * comments in SS_replace_correlation_vars ...
468 /* Replace uplevel vars with Param nodes */
469 if (PlannerQueryLevel > 1)
470 expr = SS_replace_correlation_vars(expr);
473 * If it's a qual or havingQual, convert it to implicit-AND format.
474 * (We don't want to do this before eval_const_expressions, since the
475 * latter would be unable to simplify a top-level AND correctly. Also,
476 * SS_process_sublinks expects explicit-AND format.)
478 if (kind == EXPRKIND_QUAL)
479 expr = (Node *) make_ands_implicit((Expr *) expr);
485 * preprocess_qual_conditions
486 * Recursively scan the query's jointree and do subquery_planner's
487 * preprocessing work on each qual condition found therein.
490 preprocess_qual_conditions(PlannerInfo *root, Node *jtnode)
494 if (IsA(jtnode, RangeTblRef))
496 /* nothing to do here */
498 else if (IsA(jtnode, FromExpr))
500 FromExpr *f = (FromExpr *) jtnode;
503 foreach(l, f->fromlist)
504 preprocess_qual_conditions(root, lfirst(l));
506 f->quals = preprocess_expression(root, f->quals, EXPRKIND_QUAL);
508 else if (IsA(jtnode, JoinExpr))
510 JoinExpr *j = (JoinExpr *) jtnode;
512 preprocess_qual_conditions(root, j->larg);
513 preprocess_qual_conditions(root, j->rarg);
515 j->quals = preprocess_expression(root, j->quals, EXPRKIND_QUAL);
518 elog(ERROR, "unrecognized node type: %d",
519 (int) nodeTag(jtnode));
522 /*--------------------
523 * inheritance_planner
524 * Generate a plan in the case where the result relation is an
527 * We have to handle this case differently from cases where a source
528 * relation is an inheritance set. Source inheritance is expanded at
529 * the bottom of the plan tree (see allpaths.c), but target inheritance
530 * has to be expanded at the top. The reason is that for UPDATE, each
531 * target relation needs a different targetlist matching its own column
532 * set. (This is not so critical for DELETE, but for simplicity we treat
533 * inherited DELETE the same way.) Fortunately, the UPDATE/DELETE target
534 * can never be the nullable side of an outer join, so it's OK to generate
537 * inheritlist is an integer list of RT indexes for the result relation set.
539 * Returns a query plan.
540 *--------------------
543 inheritance_planner(PlannerInfo *root, List *inheritlist)
545 Query *parse = root->parse;
546 int parentRTindex = parse->resultRelation;
547 Oid parentOID = getrelid(parentRTindex, parse->rtable);
548 int mainrtlength = list_length(parse->rtable);
549 List *subplans = NIL;
553 foreach(l, inheritlist)
555 int childRTindex = lfirst_int(l);
556 Oid childOID = getrelid(childRTindex, parse->rtable);
561 * Generate modified query with this rel as target. We have to
562 * be prepared to translate varnos in in_info_list as well as in
565 memcpy(&subroot, root, sizeof(PlannerInfo));
566 subroot.parse = (Query *)
567 adjust_inherited_attrs((Node *) parse,
568 parentRTindex, parentOID,
569 childRTindex, childOID);
570 subroot.in_info_list = (List *)
571 adjust_inherited_attrs((Node *) root->in_info_list,
572 parentRTindex, parentOID,
573 childRTindex, childOID);
576 subplan = grouping_planner(&subroot, 0.0 /* retrieve all tuples */ );
578 subplans = lappend(subplans, subplan);
581 * XXX my goodness this next bit is ugly. Really need to think about
582 * ways to rein in planner's habit of scribbling on its input.
584 * Planning of the subquery might have modified the rangetable,
585 * either by addition of RTEs due to expansion of inherited source
586 * tables, or by changes of the Query structures inside subquery
587 * RTEs. We have to ensure that this gets propagated back to the
588 * master copy. However, if we aren't done planning yet, we also
589 * need to ensure that subsequent calls to grouping_planner have
590 * virgin sub-Queries to work from. So, if we are at the last
591 * list entry, just copy the subquery rangetable back to the master
592 * copy; if we are not, then extend the master copy by adding
593 * whatever the subquery added. (We assume these added entries
594 * will go untouched by the future grouping_planner calls. We are
595 * also effectively assuming that sub-Queries will get planned
596 * identically each time, or at least that the impacts on their
597 * rangetables will be the same each time. Did I say this is ugly?)
599 if (lnext(l) == NULL)
600 parse->rtable = subroot.parse->rtable;
603 int subrtlength = list_length(subroot.parse->rtable);
605 if (subrtlength > mainrtlength)
609 subrt = list_copy_tail(subroot.parse->rtable, mainrtlength);
610 parse->rtable = list_concat(parse->rtable, subrt);
611 mainrtlength = subrtlength;
615 /* Save preprocessed tlist from first rel for use in Append */
617 tlist = subplan->targetlist;
620 /* Save the target-relations list for the executor, too */
621 parse->resultRelations = inheritlist;
623 /* Mark result as unordered (probably unnecessary) */
624 root->query_pathkeys = NIL;
626 return (Plan *) make_append(subplans, true, tlist);
629 /*--------------------
631 * Perform planning steps related to grouping, aggregation, etc.
632 * This primarily means adding top-level processing to the basic
633 * query plan produced by query_planner.
635 * tuple_fraction is the fraction of tuples we expect will be retrieved
637 * tuple_fraction is interpreted as follows:
638 * 0: expect all tuples to be retrieved (normal case)
639 * 0 < tuple_fraction < 1: expect the given fraction of tuples available
640 * from the plan to be retrieved
641 * tuple_fraction >= 1: tuple_fraction is the absolute number of tuples
642 * expected to be retrieved (ie, a LIMIT specification)
644 * Returns a query plan. Also, root->query_pathkeys is returned as the
645 * actual output ordering of the plan (in pathkey format).
646 *--------------------
649 grouping_planner(PlannerInfo *root, double tuple_fraction)
651 Query *parse = root->parse;
652 List *tlist = parse->targetList;
656 List *current_pathkeys;
659 /* Tweak caller-supplied tuple_fraction if have LIMIT/OFFSET */
660 if (parse->limitCount || parse->limitOffset)
661 tuple_fraction = preprocess_limit(root, tuple_fraction,
662 &offset_est, &count_est);
664 if (parse->setOperations)
666 List *set_sortclauses;
669 * If there's a top-level ORDER BY, assume we have to fetch all
670 * the tuples. This might seem too simplistic given all the
671 * hackery below to possibly avoid the sort ... but a nonzero
672 * tuple_fraction is only of use to plan_set_operations() when
673 * the setop is UNION ALL, and the result of UNION ALL is always
676 if (parse->sortClause)
677 tuple_fraction = 0.0;
680 * Construct the plan for set operations. The result will not
681 * need any work except perhaps a top-level sort and/or LIMIT.
683 result_plan = plan_set_operations(root, tuple_fraction,
687 * Calculate pathkeys representing the sort order (if any) of the
688 * set operation's result. We have to do this before overwriting
689 * the sort key information...
691 current_pathkeys = make_pathkeys_for_sortclauses(set_sortclauses,
692 result_plan->targetlist);
693 current_pathkeys = canonicalize_pathkeys(root, current_pathkeys);
696 * We should not need to call preprocess_targetlist, since we must
697 * be in a SELECT query node. Instead, use the targetlist
698 * returned by plan_set_operations (since this tells whether it
699 * returned any resjunk columns!), and transfer any sort key
700 * information from the original tlist.
702 Assert(parse->commandType == CMD_SELECT);
704 tlist = postprocess_setop_tlist(result_plan->targetlist, tlist);
707 * Can't handle FOR UPDATE/SHARE here (parser should have checked
708 * already, but let's make sure).
712 (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
713 errmsg("SELECT FOR UPDATE/SHARE is not allowed with UNION/INTERSECT/EXCEPT")));
716 * Calculate pathkeys that represent result ordering requirements
718 sort_pathkeys = make_pathkeys_for_sortclauses(parse->sortClause,
720 sort_pathkeys = canonicalize_pathkeys(root, sort_pathkeys);
724 /* No set operations, do regular planning */
726 List *group_pathkeys;
727 AttrNumber *groupColIdx = NULL;
728 bool need_tlist_eval = true;
730 double sub_tuple_fraction;
734 double dNumGroups = 0;
736 AggClauseCounts agg_counts;
737 int numGroupCols = list_length(parse->groupClause);
738 bool use_hashed_grouping = false;
740 MemSet(&agg_counts, 0, sizeof(AggClauseCounts));
742 /* Preprocess targetlist */
743 tlist = preprocess_targetlist(root, tlist);
746 * Generate appropriate target list for subplan; may be different
747 * from tlist if grouping or aggregation is needed.
749 sub_tlist = make_subplanTargetList(root, tlist,
750 &groupColIdx, &need_tlist_eval);
753 * Calculate pathkeys that represent grouping/ordering
756 group_pathkeys = make_pathkeys_for_sortclauses(parse->groupClause,
758 sort_pathkeys = make_pathkeys_for_sortclauses(parse->sortClause,
762 * Will need actual number of aggregates for estimating costs.
764 * Note: we do not attempt to detect duplicate aggregates here; a
765 * somewhat-overestimated count is okay for our present purposes.
767 * Note: think not that we can turn off hasAggs if we find no aggs.
768 * It is possible for constant-expression simplification to remove
769 * all explicit references to aggs, but we still have to follow
770 * the aggregate semantics (eg, producing only one output row).
774 count_agg_clauses((Node *) tlist, &agg_counts);
775 count_agg_clauses(parse->havingQual, &agg_counts);
779 * Figure out whether we need a sorted result from query_planner.
781 * If we have a GROUP BY clause, then we want a result sorted
782 * properly for grouping. Otherwise, if there is an ORDER BY
783 * clause, we want to sort by the ORDER BY clause. (Note: if we
784 * have both, and ORDER BY is a superset of GROUP BY, it would be
785 * tempting to request sort by ORDER BY --- but that might just
786 * leave us failing to exploit an available sort order at all.
787 * Needs more thought...)
789 if (parse->groupClause)
790 root->query_pathkeys = group_pathkeys;
791 else if (parse->sortClause)
792 root->query_pathkeys = sort_pathkeys;
794 root->query_pathkeys = NIL;
797 * With grouping or aggregation, the tuple fraction to pass to
798 * query_planner() may be different from what it is at top level.
800 sub_tuple_fraction = tuple_fraction;
802 if (parse->groupClause)
805 * In GROUP BY mode, we have the little problem that we don't
806 * really know how many input tuples will be needed to make a
807 * group, so we can't translate an output LIMIT count into an
808 * input count. For lack of a better idea, assume 25% of the
809 * input data will be processed if there is any output limit.
810 * However, if the caller gave us a fraction rather than an
811 * absolute count, we can keep using that fraction (which
812 * amounts to assuming that all the groups are about the same
815 if (sub_tuple_fraction >= 1.0)
816 sub_tuple_fraction = 0.25;
819 * If both GROUP BY and ORDER BY are specified, we will need
820 * two levels of sort --- and, therefore, certainly need to
821 * read all the input tuples --- unless ORDER BY is a subset
822 * of GROUP BY. (We have not yet canonicalized the pathkeys,
823 * so must use the slower noncanonical comparison method.)
825 if (parse->groupClause && parse->sortClause &&
826 !noncanonical_pathkeys_contained_in(sort_pathkeys,
828 sub_tuple_fraction = 0.0;
830 else if (parse->hasAggs)
833 * Ungrouped aggregate will certainly want all the input
836 sub_tuple_fraction = 0.0;
838 else if (parse->distinctClause)
841 * SELECT DISTINCT, like GROUP, will absorb an unpredictable
842 * number of input tuples per output tuple. Handle the same
845 if (sub_tuple_fraction >= 1.0)
846 sub_tuple_fraction = 0.25;
850 * Generate the best unsorted and presorted paths for this Query
851 * (but note there may not be any presorted path).
853 query_planner(root, sub_tlist, sub_tuple_fraction,
854 &cheapest_path, &sorted_path);
857 * We couldn't canonicalize group_pathkeys and sort_pathkeys
858 * before running query_planner(), so do it now.
860 group_pathkeys = canonicalize_pathkeys(root, group_pathkeys);
861 sort_pathkeys = canonicalize_pathkeys(root, sort_pathkeys);
864 * If grouping, estimate the number of groups. (We can't do this
865 * until after running query_planner(), either.) Then decide
866 * whether we want to use hashed grouping.
868 if (parse->groupClause)
871 double cheapest_path_rows;
874 * Beware of the possibility that cheapest_path->parent is NULL.
875 * This could happen if user does something silly like
876 * SELECT 'foo' GROUP BY 1;
878 if (cheapest_path->parent)
879 cheapest_path_rows = cheapest_path->parent->rows;
881 cheapest_path_rows = 1; /* assume non-set result */
883 groupExprs = get_sortgrouplist_exprs(parse->groupClause,
885 dNumGroups = estimate_num_groups(root,
888 /* Also want it as a long int --- but 'ware overflow! */
889 numGroups = (long) Min(dNumGroups, (double) LONG_MAX);
891 use_hashed_grouping =
892 choose_hashed_grouping(root, tuple_fraction,
893 cheapest_path, sorted_path,
894 sort_pathkeys, group_pathkeys,
895 dNumGroups, &agg_counts);
899 * Select the best path. If we are doing hashed grouping, we will
900 * always read all the input tuples, so use the cheapest-total
901 * path. Otherwise, trust query_planner's decision about which to use.
903 if (use_hashed_grouping || !sorted_path)
904 best_path = cheapest_path;
906 best_path = sorted_path;
909 * Check to see if it's possible to optimize MIN/MAX aggregates.
910 * If so, we will forget all the work we did so far to choose a
911 * "regular" path ... but we had to do it anyway to be able to
912 * tell which way is cheaper.
914 result_plan = optimize_minmax_aggregates(root,
917 if (result_plan != NULL)
920 * optimize_minmax_aggregates generated the full plan, with
921 * the right tlist, and it has no sort order.
923 current_pathkeys = NIL;
928 * Normal case --- create a plan according to query_planner's
931 result_plan = create_plan(root, best_path);
932 current_pathkeys = best_path->pathkeys;
935 * create_plan() returns a plan with just a "flat" tlist of
936 * required Vars. Usually we need to insert the sub_tlist as the
937 * tlist of the top plan node. However, we can skip that if we
938 * determined that whatever query_planner chose to return will be
944 * If the top-level plan node is one that cannot do expression
945 * evaluation, we must insert a Result node to project the
948 if (!is_projection_capable_plan(result_plan))
950 result_plan = (Plan *) make_result(sub_tlist, NULL,
956 * Otherwise, just replace the subplan's flat tlist with
959 result_plan->targetlist = sub_tlist;
963 * Also, account for the cost of evaluation of the sub_tlist.
965 * Up to now, we have only been dealing with "flat" tlists,
966 * containing just Vars. So their evaluation cost is zero
967 * according to the model used by cost_qual_eval() (or if you
968 * prefer, the cost is factored into cpu_tuple_cost). Thus we
969 * can avoid accounting for tlist cost throughout
970 * query_planner() and subroutines. But now we've inserted a
971 * tlist that might contain actual operators, sub-selects, etc
972 * --- so we'd better account for its cost.
974 * Below this point, any tlist eval cost for added-on nodes
975 * should be accounted for as we create those nodes.
976 * Presently, of the node types we can add on, only Agg and
977 * Group project new tlists (the rest just copy their input
978 * tuples) --- so make_agg() and make_group() are responsible
979 * for computing the added cost.
981 cost_qual_eval(&tlist_cost, sub_tlist);
982 result_plan->startup_cost += tlist_cost.startup;
983 result_plan->total_cost += tlist_cost.startup +
984 tlist_cost.per_tuple * result_plan->plan_rows;
989 * Since we're using query_planner's tlist and not the one
990 * make_subplanTargetList calculated, we have to refigure any
991 * grouping-column indexes make_subplanTargetList computed.
993 locate_grouping_columns(root, tlist, result_plan->targetlist,
998 * Insert AGG or GROUP node if needed, plus an explicit sort step
1001 * HAVING clause, if any, becomes qual of the Agg or Group node.
1003 if (use_hashed_grouping)
1005 /* Hashed aggregate plan --- no sort needed */
1006 result_plan = (Plan *) make_agg(root,
1008 (List *) parse->havingQual,
1015 /* Hashed aggregation produces randomly-ordered results */
1016 current_pathkeys = NIL;
1018 else if (parse->hasAggs)
1020 /* Plain aggregate plan --- sort if needed */
1021 AggStrategy aggstrategy;
1023 if (parse->groupClause)
1025 if (!pathkeys_contained_in(group_pathkeys,
1028 result_plan = (Plan *)
1029 make_sort_from_groupcols(root,
1033 current_pathkeys = group_pathkeys;
1035 aggstrategy = AGG_SORTED;
1038 * The AGG node will not change the sort ordering of its
1039 * groups, so current_pathkeys describes the result too.
1044 aggstrategy = AGG_PLAIN;
1045 /* Result will be only one row anyway; no sort order */
1046 current_pathkeys = NIL;
1049 result_plan = (Plan *) make_agg(root,
1051 (List *) parse->havingQual,
1059 else if (parse->groupClause)
1062 * GROUP BY without aggregation, so insert a group node (plus
1063 * the appropriate sort node, if necessary).
1065 * Add an explicit sort if we couldn't make the path come
1066 * out the way the GROUP node needs it.
1068 if (!pathkeys_contained_in(group_pathkeys, current_pathkeys))
1070 result_plan = (Plan *)
1071 make_sort_from_groupcols(root,
1075 current_pathkeys = group_pathkeys;
1078 result_plan = (Plan *) make_group(root,
1080 (List *) parse->havingQual,
1085 /* The Group node won't change sort ordering */
1087 else if (root->hasHavingQual)
1090 * No aggregates, and no GROUP BY, but we have a HAVING qual.
1091 * This is a degenerate case in which we are supposed to emit
1092 * either 0 or 1 row depending on whether HAVING succeeds.
1093 * Furthermore, there cannot be any variables in either HAVING
1094 * or the targetlist, so we actually do not need the FROM table
1095 * at all! We can just throw away the plan-so-far and generate
1096 * a Result node. This is a sufficiently unusual corner case
1097 * that it's not worth contorting the structure of this routine
1098 * to avoid having to generate the plan in the first place.
1100 result_plan = (Plan *) make_result(tlist,
1104 } /* end of non-minmax-aggregate case */
1105 } /* end of if (setOperations) */
1108 * If we were not able to make the plan come out in the right order,
1109 * add an explicit sort step.
1111 if (parse->sortClause)
1113 if (!pathkeys_contained_in(sort_pathkeys, current_pathkeys))
1115 result_plan = (Plan *)
1116 make_sort_from_sortclauses(root,
1119 current_pathkeys = sort_pathkeys;
1124 * If there is a DISTINCT clause, add the UNIQUE node.
1126 if (parse->distinctClause)
1128 result_plan = (Plan *) make_unique(result_plan, parse->distinctClause);
1131 * If there was grouping or aggregation, leave plan_rows as-is
1132 * (ie, assume the result was already mostly unique). If not,
1133 * it's reasonable to assume the UNIQUE filter has effects
1134 * comparable to GROUP BY.
1136 if (!parse->groupClause && !root->hasHavingQual && !parse->hasAggs)
1138 List *distinctExprs;
1140 distinctExprs = get_sortgrouplist_exprs(parse->distinctClause,
1142 result_plan->plan_rows = estimate_num_groups(root,
1144 result_plan->plan_rows);
1149 * Finally, if there is a LIMIT/OFFSET clause, add the LIMIT node.
1151 if (parse->limitCount || parse->limitOffset)
1153 result_plan = (Plan *) make_limit(result_plan,
1161 * Return the actual output ordering in query_pathkeys for possible
1162 * use by an outer query level.
1164 root->query_pathkeys = current_pathkeys;
1170 * preprocess_limit - do pre-estimation for LIMIT and/or OFFSET clauses
1172 * We try to estimate the values of the LIMIT/OFFSET clauses, and pass the
1173 * results back in *count_est and *offset_est. These variables are set to
1174 * 0 if the corresponding clause is not present, and -1 if it's present
1175 * but we couldn't estimate the value for it. (The "0" convention is OK
1176 * for OFFSET but a little bit bogus for LIMIT: effectively we estimate
1177 * LIMIT 0 as though it were LIMIT 1. But this is in line with the planner's
1178 * usual practice of never estimating less than one row.) These values will
1179 * be passed to make_limit, which see if you change this code.
1181 * The return value is the suitably adjusted tuple_fraction to use for
1182 * planning the query. This adjustment is not overridable, since it reflects
1183 * plan actions that grouping_planner() will certainly take, not assumptions
1187 preprocess_limit(PlannerInfo *root, double tuple_fraction,
1188 int *offset_est, int *count_est)
1190 Query *parse = root->parse;
1192 double limit_fraction;
1194 /* Should not be called unless LIMIT or OFFSET */
1195 Assert(parse->limitCount || parse->limitOffset);
1198 * Try to obtain the clause values. We use estimate_expression_value
1199 * primarily because it can sometimes do something useful with Params.
1201 if (parse->limitCount)
1203 est = estimate_expression_value(parse->limitCount);
1204 if (est && IsA(est, Const))
1206 if (((Const *) est)->constisnull)
1208 /* NULL indicates LIMIT ALL, ie, no limit */
1209 *count_est = 0; /* treat as not present */
1213 *count_est = DatumGetInt32(((Const *) est)->constvalue);
1214 if (*count_est <= 0)
1215 *count_est = 1; /* force to at least 1 */
1219 *count_est = -1; /* can't estimate */
1222 *count_est = 0; /* not present */
1224 if (parse->limitOffset)
1226 est = estimate_expression_value(parse->limitOffset);
1227 if (est && IsA(est, Const))
1229 if (((Const *) est)->constisnull)
1231 /* Treat NULL as no offset; the executor will too */
1232 *offset_est = 0; /* treat as not present */
1236 *offset_est = DatumGetInt32(((Const *) est)->constvalue);
1237 if (*offset_est < 0)
1238 *offset_est = 0; /* less than 0 is same as 0 */
1242 *offset_est = -1; /* can't estimate */
1245 *offset_est = 0; /* not present */
1247 if (*count_est != 0)
1250 * A LIMIT clause limits the absolute number of tuples returned.
1251 * However, if it's not a constant LIMIT then we have to guess; for
1252 * lack of a better idea, assume 10% of the plan's result is wanted.
1254 if (*count_est < 0 || *offset_est < 0)
1256 /* LIMIT or OFFSET is an expression ... punt ... */
1257 limit_fraction = 0.10;
1261 /* LIMIT (plus OFFSET, if any) is max number of tuples needed */
1262 limit_fraction = (double) *count_est + (double) *offset_est;
1266 * If we have absolute limits from both caller and LIMIT, use the
1267 * smaller value; likewise if they are both fractional. If one is
1268 * fractional and the other absolute, we can't easily determine which
1269 * is smaller, but we use the heuristic that the absolute will usually
1272 if (tuple_fraction >= 1.0)
1274 if (limit_fraction >= 1.0)
1277 tuple_fraction = Min(tuple_fraction, limit_fraction);
1281 /* caller absolute, limit fractional; use caller's value */
1284 else if (tuple_fraction > 0.0)
1286 if (limit_fraction >= 1.0)
1288 /* caller fractional, limit absolute; use limit */
1289 tuple_fraction = limit_fraction;
1293 /* both fractional */
1294 tuple_fraction = Min(tuple_fraction, limit_fraction);
1299 /* no info from caller, just use limit */
1300 tuple_fraction = limit_fraction;
1303 else if (*offset_est != 0 && tuple_fraction > 0.0)
1306 * We have an OFFSET but no LIMIT. This acts entirely differently
1307 * from the LIMIT case: here, we need to increase rather than
1308 * decrease the caller's tuple_fraction, because the OFFSET acts
1309 * to cause more tuples to be fetched instead of fewer. This only
1310 * matters if we got a tuple_fraction > 0, however.
1312 * As above, use 10% if OFFSET is present but unestimatable.
1314 if (*offset_est < 0)
1315 limit_fraction = 0.10;
1317 limit_fraction = (double) *offset_est;
1320 * If we have absolute counts from both caller and OFFSET, add them
1321 * together; likewise if they are both fractional. If one is
1322 * fractional and the other absolute, we want to take the larger,
1323 * and we heuristically assume that's the fractional one.
1325 if (tuple_fraction >= 1.0)
1327 if (limit_fraction >= 1.0)
1329 /* both absolute, so add them together */
1330 tuple_fraction += limit_fraction;
1334 /* caller absolute, limit fractional; use limit */
1335 tuple_fraction = limit_fraction;
1340 if (limit_fraction >= 1.0)
1342 /* caller fractional, limit absolute; use caller's value */
1346 /* both fractional, so add them together */
1347 tuple_fraction += limit_fraction;
1348 if (tuple_fraction >= 1.0)
1349 tuple_fraction = 0.0; /* assume fetch all */
1354 return tuple_fraction;
1358 * choose_hashed_grouping - should we use hashed grouping?
1361 choose_hashed_grouping(PlannerInfo *root, double tuple_fraction,
1362 Path *cheapest_path, Path *sorted_path,
1363 List *sort_pathkeys, List *group_pathkeys,
1364 double dNumGroups, AggClauseCounts *agg_counts)
1366 int numGroupCols = list_length(root->parse->groupClause);
1367 double cheapest_path_rows;
1368 int cheapest_path_width;
1370 List *current_pathkeys;
1375 * Check can't-do-it conditions, including whether the grouping operators
1378 * Executor doesn't support hashed aggregation with DISTINCT aggregates.
1379 * (Doing so would imply storing *all* the input values in the hash table,
1380 * which seems like a certain loser.)
1382 if (!enable_hashagg)
1384 if (agg_counts->numDistinctAggs != 0)
1386 if (!hash_safe_grouping(root))
1390 * Don't do it if it doesn't look like the hashtable will fit into
1393 * Beware here of the possibility that cheapest_path->parent is NULL.
1394 * This could happen if user does something silly like
1395 * SELECT 'foo' GROUP BY 1;
1397 if (cheapest_path->parent)
1399 cheapest_path_rows = cheapest_path->parent->rows;
1400 cheapest_path_width = cheapest_path->parent->width;
1404 cheapest_path_rows = 1; /* assume non-set result */
1405 cheapest_path_width = 100; /* arbitrary */
1408 /* Estimate per-hash-entry space at tuple width... */
1409 hashentrysize = cheapest_path_width;
1410 /* plus space for pass-by-ref transition values... */
1411 hashentrysize += agg_counts->transitionSpace;
1412 /* plus the per-hash-entry overhead */
1413 hashentrysize += hash_agg_entry_size(agg_counts->numAggs);
1415 if (hashentrysize * dNumGroups > work_mem * 1024L)
1419 * See if the estimated cost is no more than doing it the other way.
1420 * While avoiding the need for sorted input is usually a win, the fact
1421 * that the output won't be sorted may be a loss; so we need to do an
1422 * actual cost comparison.
1424 * We need to consider
1425 * cheapest_path + hashagg [+ final sort]
1427 * cheapest_path [+ sort] + group or agg [+ final sort]
1429 * presorted_path + group or agg [+ final sort]
1430 * where brackets indicate a step that may not be needed. We assume
1431 * query_planner() will have returned a presorted path only if it's a
1432 * winner compared to cheapest_path for this purpose.
1434 * These path variables are dummies that just hold cost fields; we don't
1435 * make actual Paths for these steps.
1437 cost_agg(&hashed_p, root, AGG_HASHED, agg_counts->numAggs,
1438 numGroupCols, dNumGroups,
1439 cheapest_path->startup_cost, cheapest_path->total_cost,
1440 cheapest_path_rows);
1441 /* Result of hashed agg is always unsorted */
1443 cost_sort(&hashed_p, root, sort_pathkeys, hashed_p.total_cost,
1444 dNumGroups, cheapest_path_width);
1448 sorted_p.startup_cost = sorted_path->startup_cost;
1449 sorted_p.total_cost = sorted_path->total_cost;
1450 current_pathkeys = sorted_path->pathkeys;
1454 sorted_p.startup_cost = cheapest_path->startup_cost;
1455 sorted_p.total_cost = cheapest_path->total_cost;
1456 current_pathkeys = cheapest_path->pathkeys;
1458 if (!pathkeys_contained_in(group_pathkeys,
1461 cost_sort(&sorted_p, root, group_pathkeys, sorted_p.total_cost,
1462 cheapest_path_rows, cheapest_path_width);
1463 current_pathkeys = group_pathkeys;
1466 if (root->parse->hasAggs)
1467 cost_agg(&sorted_p, root, AGG_SORTED, agg_counts->numAggs,
1468 numGroupCols, dNumGroups,
1469 sorted_p.startup_cost, sorted_p.total_cost,
1470 cheapest_path_rows);
1472 cost_group(&sorted_p, root, numGroupCols, dNumGroups,
1473 sorted_p.startup_cost, sorted_p.total_cost,
1474 cheapest_path_rows);
1475 /* The Agg or Group node will preserve ordering */
1476 if (sort_pathkeys &&
1477 !pathkeys_contained_in(sort_pathkeys, current_pathkeys))
1478 cost_sort(&sorted_p, root, sort_pathkeys, sorted_p.total_cost,
1479 dNumGroups, cheapest_path_width);
1482 * Now make the decision using the top-level tuple fraction. First we
1483 * have to convert an absolute count (LIMIT) into fractional form.
1485 if (tuple_fraction >= 1.0)
1486 tuple_fraction /= dNumGroups;
1488 if (compare_fractional_path_costs(&hashed_p, &sorted_p,
1489 tuple_fraction) < 0)
1491 /* Hashed is cheaper, so use it */
1498 * hash_safe_grouping - are grouping operators hashable?
1500 * We assume hashed aggregation will work if the datatype's equality operator
1501 * is marked hashjoinable.
1504 hash_safe_grouping(PlannerInfo *root)
1508 foreach(gl, root->parse->groupClause)
1510 GroupClause *grpcl = (GroupClause *) lfirst(gl);
1511 TargetEntry *tle = get_sortgroupclause_tle(grpcl,
1512 root->parse->targetList);
1516 optup = equality_oper(exprType((Node *) tle->expr), true);
1519 oprcanhash = ((Form_pg_operator) GETSTRUCT(optup))->oprcanhash;
1520 ReleaseSysCache(optup);
1528 * make_subplanTargetList
1529 * Generate appropriate target list when grouping is required.
1531 * When grouping_planner inserts Aggregate, Group, or Result plan nodes
1532 * above the result of query_planner, we typically want to pass a different
1533 * target list to query_planner than the outer plan nodes should have.
1534 * This routine generates the correct target list for the subplan.
1536 * The initial target list passed from the parser already contains entries
1537 * for all ORDER BY and GROUP BY expressions, but it will not have entries
1538 * for variables used only in HAVING clauses; so we need to add those
1539 * variables to the subplan target list. Also, we flatten all expressions
1540 * except GROUP BY items into their component variables; the other expressions
1541 * will be computed by the inserted nodes rather than by the subplan.
1542 * For example, given a query like
1543 * SELECT a+b,SUM(c+d) FROM table GROUP BY a+b;
1544 * we want to pass this targetlist to the subplan:
1546 * where the a+b target will be used by the Sort/Group steps, and the
1547 * other targets will be used for computing the final results. (In the
1548 * above example we could theoretically suppress the a and b targets and
1549 * pass down only c,d,a+b, but it's not really worth the trouble to
1550 * eliminate simple var references from the subplan. We will avoid doing
1551 * the extra computation to recompute a+b at the outer level; see
1552 * replace_vars_with_subplan_refs() in setrefs.c.)
1554 * If we are grouping or aggregating, *and* there are no non-Var grouping
1555 * expressions, then the returned tlist is effectively dummy; we do not
1556 * need to force it to be evaluated, because all the Vars it contains
1557 * should be present in the output of query_planner anyway.
1559 * 'tlist' is the query's target list.
1560 * 'groupColIdx' receives an array of column numbers for the GROUP BY
1561 * expressions (if there are any) in the subplan's target list.
1562 * 'need_tlist_eval' is set true if we really need to evaluate the
1565 * The result is the targetlist to be passed to the subplan.
1569 make_subplanTargetList(PlannerInfo *root,
1571 AttrNumber **groupColIdx,
1572 bool *need_tlist_eval)
1574 Query *parse = root->parse;
1579 *groupColIdx = NULL;
1582 * If we're not grouping or aggregating, there's nothing to do here;
1583 * query_planner should receive the unmodified target list.
1585 if (!parse->hasAggs && !parse->groupClause && !root->hasHavingQual)
1587 *need_tlist_eval = true;
1592 * Otherwise, start with a "flattened" tlist (having just the vars
1593 * mentioned in the targetlist and HAVING qual --- but not upper-
1594 * level Vars; they will be replaced by Params later on).
1596 sub_tlist = flatten_tlist(tlist);
1597 extravars = pull_var_clause(parse->havingQual, false);
1598 sub_tlist = add_to_flat_tlist(sub_tlist, extravars);
1599 list_free(extravars);
1600 *need_tlist_eval = false; /* only eval if not flat tlist */
1603 * If grouping, create sub_tlist entries for all GROUP BY expressions
1604 * (GROUP BY items that are simple Vars should be in the list
1605 * already), and make an array showing where the group columns are in
1608 numCols = list_length(parse->groupClause);
1612 AttrNumber *grpColIdx;
1615 grpColIdx = (AttrNumber *) palloc(sizeof(AttrNumber) * numCols);
1616 *groupColIdx = grpColIdx;
1618 foreach(gl, parse->groupClause)
1620 GroupClause *grpcl = (GroupClause *) lfirst(gl);
1621 Node *groupexpr = get_sortgroupclause_expr(grpcl, tlist);
1622 TargetEntry *te = NULL;
1625 /* Find or make a matching sub_tlist entry */
1626 foreach(sl, sub_tlist)
1628 te = (TargetEntry *) lfirst(sl);
1629 if (equal(groupexpr, te->expr))
1634 te = makeTargetEntry((Expr *) groupexpr,
1635 list_length(sub_tlist) + 1,
1638 sub_tlist = lappend(sub_tlist, te);
1639 *need_tlist_eval = true; /* it's not flat anymore */
1642 /* and save its resno */
1643 grpColIdx[keyno++] = te->resno;
1651 * locate_grouping_columns
1652 * Locate grouping columns in the tlist chosen by query_planner.
1654 * This is only needed if we don't use the sub_tlist chosen by
1655 * make_subplanTargetList. We have to forget the column indexes found
1656 * by that routine and re-locate the grouping vars in the real sub_tlist.
1659 locate_grouping_columns(PlannerInfo *root,
1662 AttrNumber *groupColIdx)
1668 * No work unless grouping.
1670 if (!root->parse->groupClause)
1672 Assert(groupColIdx == NULL);
1675 Assert(groupColIdx != NULL);
1677 foreach(gl, root->parse->groupClause)
1679 GroupClause *grpcl = (GroupClause *) lfirst(gl);
1680 Node *groupexpr = get_sortgroupclause_expr(grpcl, tlist);
1681 TargetEntry *te = NULL;
1684 foreach(sl, sub_tlist)
1686 te = (TargetEntry *) lfirst(sl);
1687 if (equal(groupexpr, te->expr))
1691 elog(ERROR, "failed to locate grouping columns");
1693 groupColIdx[keyno++] = te->resno;
1698 * postprocess_setop_tlist
1699 * Fix up targetlist returned by plan_set_operations().
1701 * We need to transpose sort key info from the orig_tlist into new_tlist.
1702 * NOTE: this would not be good enough if we supported resjunk sort keys
1703 * for results of set operations --- then, we'd need to project a whole
1704 * new tlist to evaluate the resjunk columns. For now, just ereport if we
1705 * find any resjunk columns in orig_tlist.
1708 postprocess_setop_tlist(List *new_tlist, List *orig_tlist)
1711 ListCell *orig_tlist_item = list_head(orig_tlist);
1713 foreach(l, new_tlist)
1715 TargetEntry *new_tle = (TargetEntry *) lfirst(l);
1716 TargetEntry *orig_tle;
1718 /* ignore resjunk columns in setop result */
1719 if (new_tle->resjunk)
1722 Assert(orig_tlist_item != NULL);
1723 orig_tle = (TargetEntry *) lfirst(orig_tlist_item);
1724 orig_tlist_item = lnext(orig_tlist_item);
1725 if (orig_tle->resjunk) /* should not happen */
1726 elog(ERROR, "resjunk output columns are not implemented");
1727 Assert(new_tle->resno == orig_tle->resno);
1728 new_tle->ressortgroupref = orig_tle->ressortgroupref;
1730 if (orig_tlist_item != NULL)
1731 elog(ERROR, "resjunk output columns are not implemented");