1 /*-------------------------------------------------------------------------
2 *
3 * subselect.c
4 * Planning routines for subselects.
5 *
6 * This module deals with SubLinks and CTEs, but not subquery RTEs (i.e.,
7 * not sub-SELECT-in-FROM cases).
8 *
9 * Portions Copyright (c) 1996-2017, PostgreSQL Global Development Group
10 * Portions Copyright (c) 1994, Regents of the University of California
11 *
12 * IDENTIFICATION
13 * src/backend/optimizer/plan/subselect.c
14 *
15 *-------------------------------------------------------------------------
16 */
17 #include "postgres.h"
18
19 #include "access/htup_details.h"
20 #include "catalog/pg_operator.h"
21 #include "catalog/pg_type.h"
22 #include "executor/executor.h"
23 #include "miscadmin.h"
24 #include "nodes/makefuncs.h"
25 #include "nodes/nodeFuncs.h"
26 #include "optimizer/clauses.h"
27 #include "optimizer/cost.h"
28 #include "optimizer/paramassign.h"
29 #include "optimizer/pathnode.h"
30 #include "optimizer/planmain.h"
31 #include "optimizer/planner.h"
32 #include "optimizer/prep.h"
33 #include "optimizer/subselect.h"
34 #include "optimizer/var.h"
35 #include "parser/parse_relation.h"
36 #include "rewrite/rewriteManip.h"
37 #include "utils/builtins.h"
38 #include "utils/lsyscache.h"
39 #include "utils/syscache.h"
40
41
42 typedef struct convert_testexpr_context
43 {
44 PlannerInfo *root;
45 List *subst_nodes; /* Nodes to substitute for Params */
46 } convert_testexpr_context;
47
48 typedef struct process_sublinks_context
49 {
50 PlannerInfo *root;
51 bool isTopQual;
52 } process_sublinks_context;
53
54 typedef struct finalize_primnode_context
55 {
56 PlannerInfo *root;
57 Bitmapset *paramids; /* Non-local PARAM_EXEC paramids found */
58 } finalize_primnode_context;
59
60
61 static Node *build_subplan(PlannerInfo *root, Plan *plan, PlannerInfo *subroot,
62 List *plan_params,
63 SubLinkType subLinkType, int subLinkId,
64 Node *testexpr, List *testexpr_paramids,
65 bool unknownEqFalse);
66 static List *generate_subquery_params(PlannerInfo *root, List *tlist,
67 List **paramIds);
68 static List *generate_subquery_vars(PlannerInfo *root, List *tlist,
69 Index varno);
70 static Node *convert_testexpr(PlannerInfo *root,
71 Node *testexpr,
72 List *subst_nodes);
73 static Node *convert_testexpr_mutator(Node *node,
74 convert_testexpr_context *context);
75 static bool subplan_is_hashable(Plan *plan);
76 static bool testexpr_is_hashable(Node *testexpr, List *param_ids);
77 static bool test_opexpr_is_hashable(OpExpr *testexpr, List *param_ids);
78 static bool hash_ok_operator(OpExpr *expr);
79 static bool simplify_EXISTS_query(PlannerInfo *root, Query *query);
80 static Query *convert_EXISTS_to_ANY(PlannerInfo *root, Query *subselect,
81 Node **testexpr, List **paramIds);
82 static Node *replace_correlation_vars_mutator(Node *node, PlannerInfo *root);
83 static Node *process_sublinks_mutator(Node *node,
84 process_sublinks_context *context);
85 static Bitmapset *finalize_plan(PlannerInfo *root,
86 Plan *plan,
87 int gather_param,
88 Bitmapset *valid_params,
89 Bitmapset *scan_params);
90 static bool finalize_primnode(Node *node, finalize_primnode_context *context);
91 static bool finalize_agg_primnode(Node *node, finalize_primnode_context *context);
92
93
94 /*
95 * Assign a (nonnegative) PARAM_EXEC ID for a special parameter (one that
96 * is not actually used to carry a value at runtime). Such parameters are
97 * used for special runtime signaling purposes, such as connecting a
98 * recursive union node to its worktable scan node or forcing plan
99 * re-evaluation within the EvalPlanQual mechanism. No actual Param node
100 * exists with this ID, however.
101 *
102 * XXX deprecated: use assign_special_exec_param directly, instead. We are
103 * keeping this in v11 and below only to avoid API breaks.
104 */
105 int
SS_assign_special_param(PlannerInfo * root)106 SS_assign_special_param(PlannerInfo *root)
107 {
108 return assign_special_exec_param(root);
109 }
110
111 /*
112 * Get the datatype/typmod/collation of the first column of the plan's output.
113 *
114 * This information is stored for ARRAY_SUBLINK execution and for
115 * exprType()/exprTypmod()/exprCollation(), which have no way to get at the
116 * plan associated with a SubPlan node. We really only need the info for
117 * EXPR_SUBLINK and ARRAY_SUBLINK subplans, but for consistency we save it
118 * always.
119 */
120 static void
get_first_col_type(Plan * plan,Oid * coltype,int32 * coltypmod,Oid * colcollation)121 get_first_col_type(Plan *plan, Oid *coltype, int32 *coltypmod,
122 Oid *colcollation)
123 {
124 /* In cases such as EXISTS, tlist might be empty; arbitrarily use VOID */
125 if (plan->targetlist)
126 {
127 TargetEntry *tent = linitial_node(TargetEntry, plan->targetlist);
128
129 if (!tent->resjunk)
130 {
131 *coltype = exprType((Node *) tent->expr);
132 *coltypmod = exprTypmod((Node *) tent->expr);
133 *colcollation = exprCollation((Node *) tent->expr);
134 return;
135 }
136 }
137 *coltype = VOIDOID;
138 *coltypmod = -1;
139 *colcollation = InvalidOid;
140 }
141
142 /*
143 * Convert a SubLink (as created by the parser) into a SubPlan.
144 *
145 * We are given the SubLink's contained query, type, ID, and testexpr. We are
146 * also told if this expression appears at top level of a WHERE/HAVING qual.
147 *
148 * Note: we assume that the testexpr has been AND/OR flattened (actually,
149 * it's been through eval_const_expressions), but not converted to
150 * implicit-AND form; and any SubLinks in it should already have been
151 * converted to SubPlans. The subquery is as yet untouched, however.
152 *
153 * The result is whatever we need to substitute in place of the SubLink node
154 * in the executable expression. If we're going to do the subplan as a
155 * regular subplan, this will be the constructed SubPlan node. If we're going
156 * to do the subplan as an InitPlan, the SubPlan node instead goes into
157 * root->init_plans, and what we return here is an expression tree
158 * representing the InitPlan's result: usually just a Param node representing
159 * a single scalar result, but possibly a row comparison tree containing
160 * multiple Param nodes, or for a MULTIEXPR subquery a simple NULL constant
161 * (since the real output Params are elsewhere in the tree, and the MULTIEXPR
162 * subquery itself is in a resjunk tlist entry whose value is uninteresting).
163 */
164 static Node *
make_subplan(PlannerInfo * root,Query * orig_subquery,SubLinkType subLinkType,int subLinkId,Node * testexpr,bool isTopQual)165 make_subplan(PlannerInfo *root, Query *orig_subquery,
166 SubLinkType subLinkType, int subLinkId,
167 Node *testexpr, bool isTopQual)
168 {
169 Query *subquery;
170 bool simple_exists = false;
171 double tuple_fraction;
172 PlannerInfo *subroot;
173 RelOptInfo *final_rel;
174 Path *best_path;
175 Plan *plan;
176 List *plan_params;
177 Node *result;
178
179 /*
180 * Copy the source Query node. This is a quick and dirty kluge to resolve
181 * the fact that the parser can generate trees with multiple links to the
182 * same sub-Query node, but the planner wants to scribble on the Query.
183 * Try to clean this up when we do querytree redesign...
184 */
185 subquery = copyObject(orig_subquery);
186
187 /*
188 * If it's an EXISTS subplan, we might be able to simplify it.
189 */
190 if (subLinkType == EXISTS_SUBLINK)
191 simple_exists = simplify_EXISTS_query(root, subquery);
192
193 /*
194 * For an EXISTS subplan, tell lower-level planner to expect that only the
195 * first tuple will be retrieved. For ALL and ANY subplans, we will be
196 * able to stop evaluating if the test condition fails or matches, so very
197 * often not all the tuples will be retrieved; for lack of a better idea,
198 * specify 50% retrieval. For EXPR, MULTIEXPR, and ROWCOMPARE subplans,
199 * use default behavior (we're only expecting one row out, anyway).
200 *
201 * NOTE: if you change these numbers, also change cost_subplan() in
202 * path/costsize.c.
203 *
204 * XXX If an ANY subplan is uncorrelated, build_subplan may decide to hash
205 * its output. In that case it would've been better to specify full
206 * retrieval. At present, however, we can only check hashability after
207 * we've made the subplan :-(. (Determining whether it'll fit in work_mem
208 * is the really hard part.) Therefore, we don't want to be too
209 * optimistic about the percentage of tuples retrieved, for fear of
210 * selecting a plan that's bad for the materialization case.
211 */
212 if (subLinkType == EXISTS_SUBLINK)
213 tuple_fraction = 1.0; /* just like a LIMIT 1 */
214 else if (subLinkType == ALL_SUBLINK ||
215 subLinkType == ANY_SUBLINK)
216 tuple_fraction = 0.5; /* 50% */
217 else
218 tuple_fraction = 0.0; /* default behavior */
219
220 /* plan_params should not be in use in current query level */
221 Assert(root->plan_params == NIL);
222
223 /* Generate Paths for the subquery */
224 subroot = subquery_planner(root->glob, subquery,
225 root,
226 false, tuple_fraction);
227
228 /* Isolate the params needed by this specific subplan */
229 plan_params = root->plan_params;
230 root->plan_params = NIL;
231
232 /*
233 * Select best Path and turn it into a Plan. At least for now, there
234 * seems no reason to postpone doing that.
235 */
236 final_rel = fetch_upper_rel(subroot, UPPERREL_FINAL, NULL);
237 best_path = get_cheapest_fractional_path(final_rel, tuple_fraction);
238
239 plan = create_plan(subroot, best_path);
240
241 /* And convert to SubPlan or InitPlan format. */
242 result = build_subplan(root, plan, subroot, plan_params,
243 subLinkType, subLinkId,
244 testexpr, NIL, isTopQual);
245
246 /*
247 * If it's a correlated EXISTS with an unimportant targetlist, we might be
248 * able to transform it to the equivalent of an IN and then implement it
249 * by hashing. We don't have enough information yet to tell which way is
250 * likely to be better (it depends on the expected number of executions of
251 * the EXISTS qual, and we are much too early in planning the outer query
252 * to be able to guess that). So we generate both plans, if possible, and
253 * leave it to the executor to decide which to use.
254 */
255 if (simple_exists && IsA(result, SubPlan))
256 {
257 Node *newtestexpr;
258 List *paramIds;
259
260 /* Make a second copy of the original subquery */
261 subquery = copyObject(orig_subquery);
262 /* and re-simplify */
263 simple_exists = simplify_EXISTS_query(root, subquery);
264 Assert(simple_exists);
265 /* See if it can be converted to an ANY query */
266 subquery = convert_EXISTS_to_ANY(root, subquery,
267 &newtestexpr, ¶mIds);
268 if (subquery)
269 {
270 /* Generate Paths for the ANY subquery; we'll need all rows */
271 subroot = subquery_planner(root->glob, subquery,
272 root,
273 false, 0.0);
274
275 /* Isolate the params needed by this specific subplan */
276 plan_params = root->plan_params;
277 root->plan_params = NIL;
278
279 /* Select best Path and turn it into a Plan */
280 final_rel = fetch_upper_rel(subroot, UPPERREL_FINAL, NULL);
281 best_path = final_rel->cheapest_total_path;
282
283 plan = create_plan(subroot, best_path);
284
285 /* Now we can check if it'll fit in work_mem */
286 /* XXX can we check this at the Path stage? */
287 if (subplan_is_hashable(plan))
288 {
289 SubPlan *hashplan;
290 AlternativeSubPlan *asplan;
291
292 /* OK, convert to SubPlan format. */
293 hashplan = castNode(SubPlan,
294 build_subplan(root, plan, subroot,
295 plan_params,
296 ANY_SUBLINK, 0,
297 newtestexpr,
298 paramIds,
299 true));
300 /* Check we got what we expected */
301 Assert(hashplan->parParam == NIL);
302 Assert(hashplan->useHashTable);
303
304 /* Leave it to the executor to decide which plan to use */
305 asplan = makeNode(AlternativeSubPlan);
306 asplan->subplans = list_make2(result, hashplan);
307 result = (Node *) asplan;
308 }
309 }
310 }
311
312 return result;
313 }
314
315 /*
316 * Build a SubPlan node given the raw inputs --- subroutine for make_subplan
317 *
318 * Returns either the SubPlan, or a replacement expression if we decide to
319 * make it an InitPlan, as explained in the comments for make_subplan.
320 */
321 static Node *
build_subplan(PlannerInfo * root,Plan * plan,PlannerInfo * subroot,List * plan_params,SubLinkType subLinkType,int subLinkId,Node * testexpr,List * testexpr_paramids,bool unknownEqFalse)322 build_subplan(PlannerInfo *root, Plan *plan, PlannerInfo *subroot,
323 List *plan_params,
324 SubLinkType subLinkType, int subLinkId,
325 Node *testexpr, List *testexpr_paramids,
326 bool unknownEqFalse)
327 {
328 Node *result;
329 SubPlan *splan;
330 bool isInitPlan;
331 ListCell *lc;
332
333 /*
334 * Initialize the SubPlan node. Note plan_id, plan_name, and cost fields
335 * are set further down.
336 */
337 splan = makeNode(SubPlan);
338 splan->subLinkType = subLinkType;
339 splan->testexpr = NULL;
340 splan->paramIds = NIL;
341 get_first_col_type(plan, &splan->firstColType, &splan->firstColTypmod,
342 &splan->firstColCollation);
343 splan->useHashTable = false;
344 splan->unknownEqFalse = unknownEqFalse;
345 splan->parallel_safe = plan->parallel_safe;
346 splan->setParam = NIL;
347 splan->parParam = NIL;
348 splan->args = NIL;
349
350 /*
351 * Make parParam and args lists of param IDs and expressions that current
352 * query level will pass to this child plan.
353 */
354 foreach(lc, plan_params)
355 {
356 PlannerParamItem *pitem = (PlannerParamItem *) lfirst(lc);
357 Node *arg = pitem->item;
358
359 /*
360 * The Var, PlaceHolderVar, or Aggref has already been adjusted to
361 * have the correct varlevelsup, phlevelsup, or agglevelsup.
362 *
363 * If it's a PlaceHolderVar or Aggref, its arguments might contain
364 * SubLinks, which have not yet been processed (see the comments for
365 * SS_replace_correlation_vars). Do that now.
366 */
367 if (IsA(arg, PlaceHolderVar) ||
368 IsA(arg, Aggref))
369 arg = SS_process_sublinks(root, arg, false);
370
371 splan->parParam = lappend_int(splan->parParam, pitem->paramId);
372 splan->args = lappend(splan->args, arg);
373 }
374
375 /*
376 * Un-correlated or undirect correlated plans of EXISTS, EXPR, ARRAY,
377 * ROWCOMPARE, or MULTIEXPR types can be used as initPlans. For EXISTS,
378 * EXPR, or ARRAY, we return a Param referring to the result of evaluating
379 * the initPlan. For ROWCOMPARE, we must modify the testexpr tree to
380 * contain PARAM_EXEC Params instead of the PARAM_SUBLINK Params emitted
381 * by the parser, and then return that tree. For MULTIEXPR, we return a
382 * null constant: the resjunk targetlist item containing the SubLink does
383 * not need to return anything useful, since the referencing Params are
384 * elsewhere.
385 */
386 if (splan->parParam == NIL && subLinkType == EXISTS_SUBLINK)
387 {
388 Param *prm;
389
390 Assert(testexpr == NULL);
391 prm = generate_new_exec_param(root, BOOLOID, -1, InvalidOid);
392 splan->setParam = list_make1_int(prm->paramid);
393 isInitPlan = true;
394 result = (Node *) prm;
395 }
396 else if (splan->parParam == NIL && subLinkType == EXPR_SUBLINK)
397 {
398 TargetEntry *te = linitial(plan->targetlist);
399 Param *prm;
400
401 Assert(!te->resjunk);
402 Assert(testexpr == NULL);
403 prm = generate_new_exec_param(root,
404 exprType((Node *) te->expr),
405 exprTypmod((Node *) te->expr),
406 exprCollation((Node *) te->expr));
407 splan->setParam = list_make1_int(prm->paramid);
408 isInitPlan = true;
409 result = (Node *) prm;
410 }
411 else if (splan->parParam == NIL && subLinkType == ARRAY_SUBLINK)
412 {
413 TargetEntry *te = linitial(plan->targetlist);
414 Oid arraytype;
415 Param *prm;
416
417 Assert(!te->resjunk);
418 Assert(testexpr == NULL);
419 arraytype = get_promoted_array_type(exprType((Node *) te->expr));
420 if (!OidIsValid(arraytype))
421 elog(ERROR, "could not find array type for datatype %s",
422 format_type_be(exprType((Node *) te->expr)));
423 prm = generate_new_exec_param(root,
424 arraytype,
425 exprTypmod((Node *) te->expr),
426 exprCollation((Node *) te->expr));
427 splan->setParam = list_make1_int(prm->paramid);
428 isInitPlan = true;
429 result = (Node *) prm;
430 }
431 else if (splan->parParam == NIL && subLinkType == ROWCOMPARE_SUBLINK)
432 {
433 /* Adjust the Params */
434 List *params;
435
436 Assert(testexpr != NULL);
437 params = generate_subquery_params(root,
438 plan->targetlist,
439 &splan->paramIds);
440 result = convert_testexpr(root,
441 testexpr,
442 params);
443 splan->setParam = list_copy(splan->paramIds);
444 isInitPlan = true;
445
446 /*
447 * The executable expression is returned to become part of the outer
448 * plan's expression tree; it is not kept in the initplan node.
449 */
450 }
451 else if (subLinkType == MULTIEXPR_SUBLINK)
452 {
453 /*
454 * Whether it's an initplan or not, it needs to set a PARAM_EXEC Param
455 * for each output column.
456 */
457 List *params;
458
459 Assert(testexpr == NULL);
460 params = generate_subquery_params(root,
461 plan->targetlist,
462 &splan->setParam);
463
464 /*
465 * Save the list of replacement Params in the n'th cell of
466 * root->multiexpr_params; setrefs.c will use it to replace
467 * PARAM_MULTIEXPR Params.
468 */
469 while (list_length(root->multiexpr_params) < subLinkId)
470 root->multiexpr_params = lappend(root->multiexpr_params, NIL);
471 lc = list_nth_cell(root->multiexpr_params, subLinkId - 1);
472 Assert(lfirst(lc) == NIL);
473 lfirst(lc) = params;
474
475 /* It can be an initplan if there are no parParams. */
476 if (splan->parParam == NIL)
477 {
478 isInitPlan = true;
479 result = (Node *) makeNullConst(RECORDOID, -1, InvalidOid);
480 }
481 else
482 {
483 isInitPlan = false;
484 result = (Node *) splan;
485 }
486 }
487 else
488 {
489 /*
490 * Adjust the Params in the testexpr, unless caller already took care
491 * of it (as indicated by passing a list of Param IDs).
492 */
493 if (testexpr && testexpr_paramids == NIL)
494 {
495 List *params;
496
497 params = generate_subquery_params(root,
498 plan->targetlist,
499 &splan->paramIds);
500 splan->testexpr = convert_testexpr(root,
501 testexpr,
502 params);
503 }
504 else
505 {
506 splan->testexpr = testexpr;
507 splan->paramIds = testexpr_paramids;
508 }
509
510 /*
511 * We can't convert subplans of ALL_SUBLINK or ANY_SUBLINK types to
512 * initPlans, even when they are uncorrelated or undirect correlated,
513 * because we need to scan the output of the subplan for each outer
514 * tuple. But if it's a not-direct-correlated IN (= ANY) test, we
515 * might be able to use a hashtable to avoid comparing all the tuples.
516 */
517 if (subLinkType == ANY_SUBLINK &&
518 splan->parParam == NIL &&
519 subplan_is_hashable(plan) &&
520 testexpr_is_hashable(splan->testexpr, splan->paramIds))
521 splan->useHashTable = true;
522
523 /*
524 * Otherwise, we have the option to tack a Material node onto the top
525 * of the subplan, to reduce the cost of reading it repeatedly. This
526 * is pointless for a direct-correlated subplan, since we'd have to
527 * recompute its results each time anyway. For uncorrelated/undirect
528 * correlated subplans, we add Material unless the subplan's top plan
529 * node would materialize its output anyway. Also, if enable_material
530 * is false, then the user does not want us to materialize anything
531 * unnecessarily, so we don't.
532 */
533 else if (splan->parParam == NIL && enable_material &&
534 !ExecMaterializesOutput(nodeTag(plan)))
535 plan = materialize_finished_plan(plan);
536
537 result = (Node *) splan;
538 isInitPlan = false;
539 }
540
541 /*
542 * Add the subplan and its PlannerInfo to the global lists.
543 */
544 root->glob->subplans = lappend(root->glob->subplans, plan);
545 root->glob->subroots = lappend(root->glob->subroots, subroot);
546 splan->plan_id = list_length(root->glob->subplans);
547
548 if (isInitPlan)
549 root->init_plans = lappend(root->init_plans, splan);
550
551 /*
552 * A parameterless subplan (not initplan) should be prepared to handle
553 * REWIND efficiently. If it has direct parameters then there's no point
554 * since it'll be reset on each scan anyway; and if it's an initplan then
555 * there's no point since it won't get re-run without parameter changes
556 * anyway. The input of a hashed subplan doesn't need REWIND either.
557 */
558 if (splan->parParam == NIL && !isInitPlan && !splan->useHashTable)
559 root->glob->rewindPlanIDs = bms_add_member(root->glob->rewindPlanIDs,
560 splan->plan_id);
561
562 /* Label the subplan for EXPLAIN purposes */
563 splan->plan_name = palloc(32 + 12 * list_length(splan->setParam));
564 sprintf(splan->plan_name, "%s %d",
565 isInitPlan ? "InitPlan" : "SubPlan",
566 splan->plan_id);
567 if (splan->setParam)
568 {
569 char *ptr = splan->plan_name + strlen(splan->plan_name);
570
571 ptr += sprintf(ptr, " (returns ");
572 foreach(lc, splan->setParam)
573 {
574 ptr += sprintf(ptr, "$%d%s",
575 lfirst_int(lc),
576 lnext(lc) ? "," : ")");
577 }
578 }
579
580 /* Lastly, fill in the cost estimates for use later */
581 cost_subplan(root, splan, plan);
582
583 return result;
584 }
585
586 /*
587 * generate_subquery_params: build a list of Params representing the output
588 * columns of a sublink's sub-select, given the sub-select's targetlist.
589 *
590 * We also return an integer list of the paramids of the Params.
591 */
592 static List *
generate_subquery_params(PlannerInfo * root,List * tlist,List ** paramIds)593 generate_subquery_params(PlannerInfo *root, List *tlist, List **paramIds)
594 {
595 List *result;
596 List *ids;
597 ListCell *lc;
598
599 result = ids = NIL;
600 foreach(lc, tlist)
601 {
602 TargetEntry *tent = (TargetEntry *) lfirst(lc);
603 Param *param;
604
605 if (tent->resjunk)
606 continue;
607
608 param = generate_new_exec_param(root,
609 exprType((Node *) tent->expr),
610 exprTypmod((Node *) tent->expr),
611 exprCollation((Node *) tent->expr));
612 result = lappend(result, param);
613 ids = lappend_int(ids, param->paramid);
614 }
615
616 *paramIds = ids;
617 return result;
618 }
619
620 /*
621 * generate_subquery_vars: build a list of Vars representing the output
622 * columns of a sublink's sub-select, given the sub-select's targetlist.
623 * The Vars have the specified varno (RTE index).
624 */
625 static List *
generate_subquery_vars(PlannerInfo * root,List * tlist,Index varno)626 generate_subquery_vars(PlannerInfo *root, List *tlist, Index varno)
627 {
628 List *result;
629 ListCell *lc;
630
631 result = NIL;
632 foreach(lc, tlist)
633 {
634 TargetEntry *tent = (TargetEntry *) lfirst(lc);
635 Var *var;
636
637 if (tent->resjunk)
638 continue;
639
640 var = makeVarFromTargetEntry(varno, tent);
641 result = lappend(result, var);
642 }
643
644 return result;
645 }
646
647 /*
648 * convert_testexpr: convert the testexpr given by the parser into
649 * actually executable form. This entails replacing PARAM_SUBLINK Params
650 * with Params or Vars representing the results of the sub-select. The
651 * nodes to be substituted are passed in as the List result from
652 * generate_subquery_params or generate_subquery_vars.
653 */
654 static Node *
convert_testexpr(PlannerInfo * root,Node * testexpr,List * subst_nodes)655 convert_testexpr(PlannerInfo *root,
656 Node *testexpr,
657 List *subst_nodes)
658 {
659 convert_testexpr_context context;
660
661 context.root = root;
662 context.subst_nodes = subst_nodes;
663 return convert_testexpr_mutator(testexpr, &context);
664 }
665
666 static Node *
convert_testexpr_mutator(Node * node,convert_testexpr_context * context)667 convert_testexpr_mutator(Node *node,
668 convert_testexpr_context *context)
669 {
670 if (node == NULL)
671 return NULL;
672 if (IsA(node, Param))
673 {
674 Param *param = (Param *) node;
675
676 if (param->paramkind == PARAM_SUBLINK)
677 {
678 if (param->paramid <= 0 ||
679 param->paramid > list_length(context->subst_nodes))
680 elog(ERROR, "unexpected PARAM_SUBLINK ID: %d", param->paramid);
681
682 /*
683 * We copy the list item to avoid having doubly-linked
684 * substructure in the modified parse tree. This is probably
685 * unnecessary when it's a Param, but be safe.
686 */
687 return (Node *) copyObject(list_nth(context->subst_nodes,
688 param->paramid - 1));
689 }
690 }
691 if (IsA(node, SubLink))
692 {
693 /*
694 * If we come across a nested SubLink, it is neither necessary nor
695 * correct to recurse into it: any PARAM_SUBLINKs we might find inside
696 * belong to the inner SubLink not the outer. So just return it as-is.
697 *
698 * This reasoning depends on the assumption that nothing will pull
699 * subexpressions into or out of the testexpr field of a SubLink, at
700 * least not without replacing PARAM_SUBLINKs first. If we did want
701 * to do that we'd need to rethink the parser-output representation
702 * altogether, since currently PARAM_SUBLINKs are only unique per
703 * SubLink not globally across the query. The whole point of
704 * replacing them with Vars or PARAM_EXEC nodes is to make them
705 * globally unique before they escape from the SubLink's testexpr.
706 *
707 * Note: this can't happen when called during SS_process_sublinks,
708 * because that recursively processes inner SubLinks first. It can
709 * happen when called from convert_ANY_sublink_to_join, though.
710 */
711 return node;
712 }
713 return expression_tree_mutator(node,
714 convert_testexpr_mutator,
715 (void *) context);
716 }
717
718 /*
719 * subplan_is_hashable: can we implement an ANY subplan by hashing?
720 */
721 static bool
subplan_is_hashable(Plan * plan)722 subplan_is_hashable(Plan *plan)
723 {
724 double subquery_size;
725
726 /*
727 * The estimated size of the subquery result must fit in work_mem. (Note:
728 * we use heap tuple overhead here even though the tuples will actually be
729 * stored as MinimalTuples; this provides some fudge factor for hashtable
730 * overhead.)
731 */
732 subquery_size = plan->plan_rows *
733 (MAXALIGN(plan->plan_width) + MAXALIGN(SizeofHeapTupleHeader));
734 if (subquery_size > work_mem * 1024L)
735 return false;
736
737 return true;
738 }
739
740 /*
741 * testexpr_is_hashable: is an ANY SubLink's test expression hashable?
742 *
743 * To identify LHS vs RHS of the hash expression, we must be given the
744 * list of output Param IDs of the SubLink's subquery.
745 */
746 static bool
testexpr_is_hashable(Node * testexpr,List * param_ids)747 testexpr_is_hashable(Node *testexpr, List *param_ids)
748 {
749 /*
750 * The testexpr must be a single OpExpr, or an AND-clause containing only
751 * OpExprs, each of which satisfy test_opexpr_is_hashable().
752 */
753 if (testexpr && IsA(testexpr, OpExpr))
754 {
755 if (test_opexpr_is_hashable((OpExpr *) testexpr, param_ids))
756 return true;
757 }
758 else if (and_clause(testexpr))
759 {
760 ListCell *l;
761
762 foreach(l, ((BoolExpr *) testexpr)->args)
763 {
764 Node *andarg = (Node *) lfirst(l);
765
766 if (!IsA(andarg, OpExpr))
767 return false;
768 if (!test_opexpr_is_hashable((OpExpr *) andarg, param_ids))
769 return false;
770 }
771 return true;
772 }
773
774 return false;
775 }
776
777 static bool
test_opexpr_is_hashable(OpExpr * testexpr,List * param_ids)778 test_opexpr_is_hashable(OpExpr *testexpr, List *param_ids)
779 {
780 /*
781 * The combining operator must be hashable and strict. The need for
782 * hashability is obvious, since we want to use hashing. Without
783 * strictness, behavior in the presence of nulls is too unpredictable. We
784 * actually must assume even more than plain strictness: it can't yield
785 * NULL for non-null inputs, either (see nodeSubplan.c). However, hash
786 * indexes and hash joins assume that too.
787 */
788 if (!hash_ok_operator(testexpr))
789 return false;
790
791 /*
792 * The left and right inputs must belong to the outer and inner queries
793 * respectively; hence Params that will be supplied by the subquery must
794 * not appear in the LHS, and Vars of the outer query must not appear in
795 * the RHS. (Ordinarily, this must be true because of the way that the
796 * parser builds an ANY SubLink's testexpr ... but inlining of functions
797 * could have changed the expression's structure, so we have to check.
798 * Such cases do not occur often enough to be worth trying to optimize, so
799 * we don't worry about trying to commute the clause or anything like
800 * that; we just need to be sure not to build an invalid plan.)
801 */
802 if (list_length(testexpr->args) != 2)
803 return false;
804 if (contain_exec_param((Node *) linitial(testexpr->args), param_ids))
805 return false;
806 if (contain_var_clause((Node *) lsecond(testexpr->args)))
807 return false;
808 return true;
809 }
810
811 /*
812 * Check expression is hashable + strict
813 *
814 * We could use op_hashjoinable() and op_strict(), but do it like this to
815 * avoid a redundant cache lookup.
816 */
817 static bool
hash_ok_operator(OpExpr * expr)818 hash_ok_operator(OpExpr *expr)
819 {
820 Oid opid = expr->opno;
821
822 /* quick out if not a binary operator */
823 if (list_length(expr->args) != 2)
824 return false;
825 if (opid == ARRAY_EQ_OP)
826 {
827 /* array_eq is strict, but must check input type to ensure hashable */
828 /* XXX record_eq will need same treatment when it becomes hashable */
829 Node *leftarg = linitial(expr->args);
830
831 return op_hashjoinable(opid, exprType(leftarg));
832 }
833 else
834 {
835 /* else must look up the operator properties */
836 HeapTuple tup;
837 Form_pg_operator optup;
838
839 tup = SearchSysCache1(OPEROID, ObjectIdGetDatum(opid));
840 if (!HeapTupleIsValid(tup))
841 elog(ERROR, "cache lookup failed for operator %u", opid);
842 optup = (Form_pg_operator) GETSTRUCT(tup);
843 if (!optup->oprcanhash || !func_strict(optup->oprcode))
844 {
845 ReleaseSysCache(tup);
846 return false;
847 }
848 ReleaseSysCache(tup);
849 return true;
850 }
851 }
852
853
854 /*
855 * SS_process_ctes: process a query's WITH list
856 *
857 * We plan each interesting WITH item and convert it to an initplan.
858 * A side effect is to fill in root->cte_plan_ids with a list that
859 * parallels root->parse->cteList and provides the subplan ID for
860 * each CTE's initplan.
861 */
862 void
SS_process_ctes(PlannerInfo * root)863 SS_process_ctes(PlannerInfo *root)
864 {
865 ListCell *lc;
866
867 Assert(root->cte_plan_ids == NIL);
868
869 foreach(lc, root->parse->cteList)
870 {
871 CommonTableExpr *cte = (CommonTableExpr *) lfirst(lc);
872 CmdType cmdType = ((Query *) cte->ctequery)->commandType;
873 Query *subquery;
874 PlannerInfo *subroot;
875 RelOptInfo *final_rel;
876 Path *best_path;
877 Plan *plan;
878 SubPlan *splan;
879 int paramid;
880
881 /*
882 * Ignore SELECT CTEs that are not actually referenced anywhere.
883 */
884 if (cte->cterefcount == 0 && cmdType == CMD_SELECT)
885 {
886 /* Make a dummy entry in cte_plan_ids */
887 root->cte_plan_ids = lappend_int(root->cte_plan_ids, -1);
888 continue;
889 }
890
891 /*
892 * Copy the source Query node. Probably not necessary, but let's keep
893 * this similar to make_subplan.
894 */
895 subquery = (Query *) copyObject(cte->ctequery);
896
897 /* plan_params should not be in use in current query level */
898 Assert(root->plan_params == NIL);
899
900 /*
901 * Generate Paths for the CTE query. Always plan for full retrieval
902 * --- we don't have enough info to predict otherwise.
903 */
904 subroot = subquery_planner(root->glob, subquery,
905 root,
906 cte->cterecursive, 0.0);
907
908 /*
909 * Since the current query level doesn't yet contain any RTEs, it
910 * should not be possible for the CTE to have requested parameters of
911 * this level.
912 */
913 if (root->plan_params)
914 elog(ERROR, "unexpected outer reference in CTE query");
915
916 /*
917 * Select best Path and turn it into a Plan. At least for now, there
918 * seems no reason to postpone doing that.
919 */
920 final_rel = fetch_upper_rel(subroot, UPPERREL_FINAL, NULL);
921 best_path = final_rel->cheapest_total_path;
922
923 plan = create_plan(subroot, best_path);
924
925 /*
926 * Make a SubPlan node for it. This is just enough unlike
927 * build_subplan that we can't share code.
928 *
929 * Note plan_id, plan_name, and cost fields are set further down.
930 */
931 splan = makeNode(SubPlan);
932 splan->subLinkType = CTE_SUBLINK;
933 splan->testexpr = NULL;
934 splan->paramIds = NIL;
935 get_first_col_type(plan, &splan->firstColType, &splan->firstColTypmod,
936 &splan->firstColCollation);
937 splan->useHashTable = false;
938 splan->unknownEqFalse = false;
939
940 /*
941 * CTE scans are not considered for parallelism (cf
942 * set_rel_consider_parallel), and even if they were, initPlans aren't
943 * parallel-safe.
944 */
945 splan->parallel_safe = false;
946 splan->setParam = NIL;
947 splan->parParam = NIL;
948 splan->args = NIL;
949
950 /*
951 * The node can't have any inputs (since it's an initplan), so the
952 * parParam and args lists remain empty. (It could contain references
953 * to earlier CTEs' output param IDs, but CTE outputs are not
954 * propagated via the args list.)
955 */
956
957 /*
958 * Assign a param ID to represent the CTE's output. No ordinary
959 * "evaluation" of this param slot ever happens, but we use the param
960 * ID for setParam/chgParam signaling just as if the CTE plan were
961 * returning a simple scalar output. (Also, the executor abuses the
962 * ParamExecData slot for this param ID for communication among
963 * multiple CteScan nodes that might be scanning this CTE.)
964 */
965 paramid = assign_special_exec_param(root);
966 splan->setParam = list_make1_int(paramid);
967
968 /*
969 * Add the subplan and its PlannerInfo to the global lists.
970 */
971 root->glob->subplans = lappend(root->glob->subplans, plan);
972 root->glob->subroots = lappend(root->glob->subroots, subroot);
973 splan->plan_id = list_length(root->glob->subplans);
974
975 root->init_plans = lappend(root->init_plans, splan);
976
977 root->cte_plan_ids = lappend_int(root->cte_plan_ids, splan->plan_id);
978
979 /* Label the subplan for EXPLAIN purposes */
980 splan->plan_name = psprintf("CTE %s", cte->ctename);
981
982 /* Lastly, fill in the cost estimates for use later */
983 cost_subplan(root, splan, plan);
984 }
985 }
986
987 /*
988 * convert_ANY_sublink_to_join: try to convert an ANY SubLink to a join
989 *
990 * The caller has found an ANY SubLink at the top level of one of the query's
991 * qual clauses, but has not checked the properties of the SubLink further.
992 * Decide whether it is appropriate to process this SubLink in join style.
993 * If so, form a JoinExpr and return it. Return NULL if the SubLink cannot
994 * be converted to a join.
995 *
996 * The only non-obvious input parameter is available_rels: this is the set
997 * of query rels that can safely be referenced in the sublink expression.
998 * (We must restrict this to avoid changing the semantics when a sublink
999 * is present in an outer join's ON qual.) The conversion must fail if
1000 * the converted qual would reference any but these parent-query relids.
1001 *
1002 * On success, the returned JoinExpr has larg = NULL and rarg = the jointree
1003 * item representing the pulled-up subquery. The caller must set larg to
1004 * represent the relation(s) on the lefthand side of the new join, and insert
1005 * the JoinExpr into the upper query's jointree at an appropriate place
1006 * (typically, where the lefthand relation(s) had been). Note that the
1007 * passed-in SubLink must also be removed from its original position in the
1008 * query quals, since the quals of the returned JoinExpr replace it.
1009 * (Notionally, we replace the SubLink with a constant TRUE, then elide the
1010 * redundant constant from the qual.)
1011 *
1012 * On success, the caller is also responsible for recursively applying
1013 * pull_up_sublinks processing to the rarg and quals of the returned JoinExpr.
1014 * (On failure, there is no need to do anything, since pull_up_sublinks will
1015 * be applied when we recursively plan the sub-select.)
1016 *
1017 * Side effects of a successful conversion include adding the SubLink's
1018 * subselect to the query's rangetable, so that it can be referenced in
1019 * the JoinExpr's rarg.
1020 */
1021 JoinExpr *
convert_ANY_sublink_to_join(PlannerInfo * root,SubLink * sublink,Relids available_rels)1022 convert_ANY_sublink_to_join(PlannerInfo *root, SubLink *sublink,
1023 Relids available_rels)
1024 {
1025 JoinExpr *result;
1026 Query *parse = root->parse;
1027 Query *subselect = (Query *) sublink->subselect;
1028 Relids upper_varnos;
1029 int rtindex;
1030 RangeTblEntry *rte;
1031 RangeTblRef *rtr;
1032 List *subquery_vars;
1033 Node *quals;
1034 ParseState *pstate;
1035
1036 Assert(sublink->subLinkType == ANY_SUBLINK);
1037
1038 /*
1039 * The sub-select must not refer to any Vars of the parent query. (Vars of
1040 * higher levels should be okay, though.)
1041 */
1042 if (contain_vars_of_level((Node *) subselect, 1))
1043 return NULL;
1044
1045 /*
1046 * The test expression must contain some Vars of the parent query, else
1047 * it's not gonna be a join. (Note that it won't have Vars referring to
1048 * the subquery, rather Params.)
1049 */
1050 upper_varnos = pull_varnos(sublink->testexpr);
1051 if (bms_is_empty(upper_varnos))
1052 return NULL;
1053
1054 /*
1055 * However, it can't refer to anything outside available_rels.
1056 */
1057 if (!bms_is_subset(upper_varnos, available_rels))
1058 return NULL;
1059
1060 /*
1061 * The combining operators and left-hand expressions mustn't be volatile.
1062 */
1063 if (contain_volatile_functions(sublink->testexpr))
1064 return NULL;
1065
1066 /* Create a dummy ParseState for addRangeTableEntryForSubquery */
1067 pstate = make_parsestate(NULL);
1068
1069 /*
1070 * Okay, pull up the sub-select into upper range table.
1071 *
1072 * We rely here on the assumption that the outer query has no references
1073 * to the inner (necessarily true, other than the Vars that we build
1074 * below). Therefore this is a lot easier than what pull_up_subqueries has
1075 * to go through.
1076 */
1077 rte = addRangeTableEntryForSubquery(pstate,
1078 subselect,
1079 makeAlias("ANY_subquery", NIL),
1080 false,
1081 false);
1082 parse->rtable = lappend(parse->rtable, rte);
1083 rtindex = list_length(parse->rtable);
1084
1085 /*
1086 * Form a RangeTblRef for the pulled-up sub-select.
1087 */
1088 rtr = makeNode(RangeTblRef);
1089 rtr->rtindex = rtindex;
1090
1091 /*
1092 * Build a list of Vars representing the subselect outputs.
1093 */
1094 subquery_vars = generate_subquery_vars(root,
1095 subselect->targetList,
1096 rtindex);
1097
1098 /*
1099 * Build the new join's qual expression, replacing Params with these Vars.
1100 */
1101 quals = convert_testexpr(root, sublink->testexpr, subquery_vars);
1102
1103 /*
1104 * And finally, build the JoinExpr node.
1105 */
1106 result = makeNode(JoinExpr);
1107 result->jointype = JOIN_SEMI;
1108 result->isNatural = false;
1109 result->larg = NULL; /* caller must fill this in */
1110 result->rarg = (Node *) rtr;
1111 result->usingClause = NIL;
1112 result->quals = quals;
1113 result->alias = NULL;
1114 result->rtindex = 0; /* we don't need an RTE for it */
1115
1116 return result;
1117 }
1118
1119 /*
1120 * convert_EXISTS_sublink_to_join: try to convert an EXISTS SubLink to a join
1121 *
1122 * The API of this function is identical to convert_ANY_sublink_to_join's,
1123 * except that we also support the case where the caller has found NOT EXISTS,
1124 * so we need an additional input parameter "under_not".
1125 */
1126 JoinExpr *
convert_EXISTS_sublink_to_join(PlannerInfo * root,SubLink * sublink,bool under_not,Relids available_rels)1127 convert_EXISTS_sublink_to_join(PlannerInfo *root, SubLink *sublink,
1128 bool under_not, Relids available_rels)
1129 {
1130 JoinExpr *result;
1131 Query *parse = root->parse;
1132 Query *subselect = (Query *) sublink->subselect;
1133 Node *whereClause;
1134 int rtoffset;
1135 int varno;
1136 Relids clause_varnos;
1137 Relids upper_varnos;
1138
1139 Assert(sublink->subLinkType == EXISTS_SUBLINK);
1140
1141 /*
1142 * Can't flatten if it contains WITH. (We could arrange to pull up the
1143 * WITH into the parent query's cteList, but that risks changing the
1144 * semantics, since a WITH ought to be executed once per associated query
1145 * call.) Note that convert_ANY_sublink_to_join doesn't have to reject
1146 * this case, since it just produces a subquery RTE that doesn't have to
1147 * get flattened into the parent query.
1148 */
1149 if (subselect->cteList)
1150 return NULL;
1151
1152 /*
1153 * Copy the subquery so we can modify it safely (see comments in
1154 * make_subplan).
1155 */
1156 subselect = copyObject(subselect);
1157
1158 /*
1159 * See if the subquery can be simplified based on the knowledge that it's
1160 * being used in EXISTS(). If we aren't able to get rid of its
1161 * targetlist, we have to fail, because the pullup operation leaves us
1162 * with noplace to evaluate the targetlist.
1163 */
1164 if (!simplify_EXISTS_query(root, subselect))
1165 return NULL;
1166
1167 /*
1168 * The subquery must have a nonempty jointree, else we won't have a join.
1169 */
1170 if (subselect->jointree->fromlist == NIL)
1171 return NULL;
1172
1173 /*
1174 * Separate out the WHERE clause. (We could theoretically also remove
1175 * top-level plain JOIN/ON clauses, but it's probably not worth the
1176 * trouble.)
1177 */
1178 whereClause = subselect->jointree->quals;
1179 subselect->jointree->quals = NULL;
1180
1181 /*
1182 * The rest of the sub-select must not refer to any Vars of the parent
1183 * query. (Vars of higher levels should be okay, though.)
1184 */
1185 if (contain_vars_of_level((Node *) subselect, 1))
1186 return NULL;
1187
1188 /*
1189 * On the other hand, the WHERE clause must contain some Vars of the
1190 * parent query, else it's not gonna be a join.
1191 */
1192 if (!contain_vars_of_level(whereClause, 1))
1193 return NULL;
1194
1195 /*
1196 * We don't risk optimizing if the WHERE clause is volatile, either.
1197 */
1198 if (contain_volatile_functions(whereClause))
1199 return NULL;
1200
1201 /*
1202 * Prepare to pull up the sub-select into top range table.
1203 *
1204 * We rely here on the assumption that the outer query has no references
1205 * to the inner (necessarily true). Therefore this is a lot easier than
1206 * what pull_up_subqueries has to go through.
1207 *
1208 * In fact, it's even easier than what convert_ANY_sublink_to_join has to
1209 * do. The machinations of simplify_EXISTS_query ensured that there is
1210 * nothing interesting in the subquery except an rtable and jointree, and
1211 * even the jointree FromExpr no longer has quals. So we can just append
1212 * the rtable to our own and use the FromExpr in our jointree. But first,
1213 * adjust all level-zero varnos in the subquery to account for the rtable
1214 * merger.
1215 */
1216 rtoffset = list_length(parse->rtable);
1217 OffsetVarNodes((Node *) subselect, rtoffset, 0);
1218 OffsetVarNodes(whereClause, rtoffset, 0);
1219
1220 /*
1221 * Upper-level vars in subquery will now be one level closer to their
1222 * parent than before; in particular, anything that had been level 1
1223 * becomes level zero.
1224 */
1225 IncrementVarSublevelsUp((Node *) subselect, -1, 1);
1226 IncrementVarSublevelsUp(whereClause, -1, 1);
1227
1228 /*
1229 * Now that the WHERE clause is adjusted to match the parent query
1230 * environment, we can easily identify all the level-zero rels it uses.
1231 * The ones <= rtoffset belong to the upper query; the ones > rtoffset do
1232 * not.
1233 */
1234 clause_varnos = pull_varnos(whereClause);
1235 upper_varnos = NULL;
1236 while ((varno = bms_first_member(clause_varnos)) >= 0)
1237 {
1238 if (varno <= rtoffset)
1239 upper_varnos = bms_add_member(upper_varnos, varno);
1240 }
1241 bms_free(clause_varnos);
1242 Assert(!bms_is_empty(upper_varnos));
1243
1244 /*
1245 * Now that we've got the set of upper-level varnos, we can make the last
1246 * check: only available_rels can be referenced.
1247 */
1248 if (!bms_is_subset(upper_varnos, available_rels))
1249 return NULL;
1250
1251 /* Now we can attach the modified subquery rtable to the parent */
1252 parse->rtable = list_concat(parse->rtable, subselect->rtable);
1253
1254 /*
1255 * And finally, build the JoinExpr node.
1256 */
1257 result = makeNode(JoinExpr);
1258 result->jointype = under_not ? JOIN_ANTI : JOIN_SEMI;
1259 result->isNatural = false;
1260 result->larg = NULL; /* caller must fill this in */
1261 /* flatten out the FromExpr node if it's useless */
1262 if (list_length(subselect->jointree->fromlist) == 1)
1263 result->rarg = (Node *) linitial(subselect->jointree->fromlist);
1264 else
1265 result->rarg = (Node *) subselect->jointree;
1266 result->usingClause = NIL;
1267 result->quals = whereClause;
1268 result->alias = NULL;
1269 result->rtindex = 0; /* we don't need an RTE for it */
1270
1271 return result;
1272 }
1273
1274 /*
1275 * simplify_EXISTS_query: remove any useless stuff in an EXISTS's subquery
1276 *
1277 * The only thing that matters about an EXISTS query is whether it returns
1278 * zero or more than zero rows. Therefore, we can remove certain SQL features
1279 * that won't affect that. The only part that is really likely to matter in
1280 * typical usage is simplifying the targetlist: it's a common habit to write
1281 * "SELECT * FROM" even though there is no need to evaluate any columns.
1282 *
1283 * Note: by suppressing the targetlist we could cause an observable behavioral
1284 * change, namely that any errors that might occur in evaluating the tlist
1285 * won't occur, nor will other side-effects of volatile functions. This seems
1286 * unlikely to bother anyone in practice.
1287 *
1288 * Returns TRUE if was able to discard the targetlist, else FALSE.
1289 */
1290 static bool
simplify_EXISTS_query(PlannerInfo * root,Query * query)1291 simplify_EXISTS_query(PlannerInfo *root, Query *query)
1292 {
1293 /*
1294 * We don't try to simplify at all if the query uses set operations,
1295 * aggregates, grouping sets, SRFs, modifying CTEs, HAVING, OFFSET, or FOR
1296 * UPDATE/SHARE; none of these seem likely in normal usage and their
1297 * possible effects are complex. (Note: we could ignore an "OFFSET 0"
1298 * clause, but that traditionally is used as an optimization fence, so we
1299 * don't.)
1300 */
1301 if (query->commandType != CMD_SELECT ||
1302 query->setOperations ||
1303 query->hasAggs ||
1304 query->groupingSets ||
1305 query->hasWindowFuncs ||
1306 query->hasTargetSRFs ||
1307 query->hasModifyingCTE ||
1308 query->havingQual ||
1309 query->limitOffset ||
1310 query->rowMarks)
1311 return false;
1312
1313 /*
1314 * LIMIT with a constant positive (or NULL) value doesn't affect the
1315 * semantics of EXISTS, so let's ignore such clauses. This is worth doing
1316 * because people accustomed to certain other DBMSes may be in the habit
1317 * of writing EXISTS(SELECT ... LIMIT 1) as an optimization. If there's a
1318 * LIMIT with anything else as argument, though, we can't simplify.
1319 */
1320 if (query->limitCount)
1321 {
1322 /*
1323 * The LIMIT clause has not yet been through eval_const_expressions,
1324 * so we have to apply that here. It might seem like this is a waste
1325 * of cycles, since the only case plausibly worth worrying about is
1326 * "LIMIT 1" ... but what we'll actually see is "LIMIT int8(1::int4)",
1327 * so we have to fold constants or we're not going to recognize it.
1328 */
1329 Node *node = eval_const_expressions(root, query->limitCount);
1330 Const *limit;
1331
1332 /* Might as well update the query if we simplified the clause. */
1333 query->limitCount = node;
1334
1335 if (!IsA(node, Const))
1336 return false;
1337
1338 limit = (Const *) node;
1339 Assert(limit->consttype == INT8OID);
1340 if (!limit->constisnull && DatumGetInt64(limit->constvalue) <= 0)
1341 return false;
1342
1343 /* Whether or not the targetlist is safe, we can drop the LIMIT. */
1344 query->limitCount = NULL;
1345 }
1346
1347 /*
1348 * Otherwise, we can throw away the targetlist, as well as any GROUP,
1349 * WINDOW, DISTINCT, and ORDER BY clauses; none of those clauses will
1350 * change a nonzero-rows result to zero rows or vice versa. (Furthermore,
1351 * since our parsetree representation of these clauses depends on the
1352 * targetlist, we'd better throw them away if we drop the targetlist.)
1353 */
1354 query->targetList = NIL;
1355 query->groupClause = NIL;
1356 query->windowClause = NIL;
1357 query->distinctClause = NIL;
1358 query->sortClause = NIL;
1359 query->hasDistinctOn = false;
1360
1361 return true;
1362 }
1363
1364 /*
1365 * convert_EXISTS_to_ANY: try to convert EXISTS to a hashable ANY sublink
1366 *
1367 * The subselect is expected to be a fresh copy that we can munge up,
1368 * and to have been successfully passed through simplify_EXISTS_query.
1369 *
1370 * On success, the modified subselect is returned, and we store a suitable
1371 * upper-level test expression at *testexpr, plus a list of the subselect's
1372 * output Params at *paramIds. (The test expression is already Param-ified
1373 * and hence need not go through convert_testexpr, which is why we have to
1374 * deal with the Param IDs specially.)
1375 *
1376 * On failure, returns NULL.
1377 */
1378 static Query *
convert_EXISTS_to_ANY(PlannerInfo * root,Query * subselect,Node ** testexpr,List ** paramIds)1379 convert_EXISTS_to_ANY(PlannerInfo *root, Query *subselect,
1380 Node **testexpr, List **paramIds)
1381 {
1382 Node *whereClause;
1383 List *leftargs,
1384 *rightargs,
1385 *opids,
1386 *opcollations,
1387 *newWhere,
1388 *tlist,
1389 *testlist,
1390 *paramids;
1391 ListCell *lc,
1392 *rc,
1393 *oc,
1394 *cc;
1395 AttrNumber resno;
1396
1397 /*
1398 * Query must not require a targetlist, since we have to insert a new one.
1399 * Caller should have dealt with the case already.
1400 */
1401 Assert(subselect->targetList == NIL);
1402
1403 /*
1404 * Separate out the WHERE clause. (We could theoretically also remove
1405 * top-level plain JOIN/ON clauses, but it's probably not worth the
1406 * trouble.)
1407 */
1408 whereClause = subselect->jointree->quals;
1409 subselect->jointree->quals = NULL;
1410
1411 /*
1412 * The rest of the sub-select must not refer to any Vars of the parent
1413 * query. (Vars of higher levels should be okay, though.)
1414 *
1415 * Note: we need not check for Aggrefs separately because we know the
1416 * sub-select is as yet unoptimized; any uplevel Aggref must therefore
1417 * contain an uplevel Var reference. This is not the case below ...
1418 */
1419 if (contain_vars_of_level((Node *) subselect, 1))
1420 return NULL;
1421
1422 /*
1423 * We don't risk optimizing if the WHERE clause is volatile, either.
1424 */
1425 if (contain_volatile_functions(whereClause))
1426 return NULL;
1427
1428 /*
1429 * Clean up the WHERE clause by doing const-simplification etc on it.
1430 * Aside from simplifying the processing we're about to do, this is
1431 * important for being able to pull chunks of the WHERE clause up into the
1432 * parent query. Since we are invoked partway through the parent's
1433 * preprocess_expression() work, earlier steps of preprocess_expression()
1434 * wouldn't get applied to the pulled-up stuff unless we do them here. For
1435 * the parts of the WHERE clause that get put back into the child query,
1436 * this work is partially duplicative, but it shouldn't hurt.
1437 *
1438 * Note: we do not run flatten_join_alias_vars. This is OK because any
1439 * parent aliases were flattened already, and we're not going to pull any
1440 * child Vars (of any description) into the parent.
1441 *
1442 * Note: passing the parent's root to eval_const_expressions is
1443 * technically wrong, but we can get away with it since only the
1444 * boundParams (if any) are used, and those would be the same in a
1445 * subroot.
1446 */
1447 whereClause = eval_const_expressions(root, whereClause);
1448 whereClause = (Node *) canonicalize_qual_ext((Expr *) whereClause, false);
1449 whereClause = (Node *) make_ands_implicit((Expr *) whereClause);
1450
1451 /*
1452 * We now have a flattened implicit-AND list of clauses, which we try to
1453 * break apart into "outervar = innervar" hash clauses. Anything that
1454 * can't be broken apart just goes back into the newWhere list. Note that
1455 * we aren't trying hard yet to ensure that we have only outer or only
1456 * inner on each side; we'll check that if we get to the end.
1457 */
1458 leftargs = rightargs = opids = opcollations = newWhere = NIL;
1459 foreach(lc, (List *) whereClause)
1460 {
1461 OpExpr *expr = (OpExpr *) lfirst(lc);
1462
1463 if (IsA(expr, OpExpr) &&
1464 hash_ok_operator(expr))
1465 {
1466 Node *leftarg = (Node *) linitial(expr->args);
1467 Node *rightarg = (Node *) lsecond(expr->args);
1468
1469 if (contain_vars_of_level(leftarg, 1))
1470 {
1471 leftargs = lappend(leftargs, leftarg);
1472 rightargs = lappend(rightargs, rightarg);
1473 opids = lappend_oid(opids, expr->opno);
1474 opcollations = lappend_oid(opcollations, expr->inputcollid);
1475 continue;
1476 }
1477 if (contain_vars_of_level(rightarg, 1))
1478 {
1479 /*
1480 * We must commute the clause to put the outer var on the
1481 * left, because the hashing code in nodeSubplan.c expects
1482 * that. This probably shouldn't ever fail, since hashable
1483 * operators ought to have commutators, but be paranoid.
1484 */
1485 expr->opno = get_commutator(expr->opno);
1486 if (OidIsValid(expr->opno) && hash_ok_operator(expr))
1487 {
1488 leftargs = lappend(leftargs, rightarg);
1489 rightargs = lappend(rightargs, leftarg);
1490 opids = lappend_oid(opids, expr->opno);
1491 opcollations = lappend_oid(opcollations, expr->inputcollid);
1492 continue;
1493 }
1494 /* If no commutator, no chance to optimize the WHERE clause */
1495 return NULL;
1496 }
1497 }
1498 /* Couldn't handle it as a hash clause */
1499 newWhere = lappend(newWhere, expr);
1500 }
1501
1502 /*
1503 * If we didn't find anything we could convert, fail.
1504 */
1505 if (leftargs == NIL)
1506 return NULL;
1507
1508 /*
1509 * There mustn't be any parent Vars or Aggs in the stuff that we intend to
1510 * put back into the child query. Note: you might think we don't need to
1511 * check for Aggs separately, because an uplevel Agg must contain an
1512 * uplevel Var in its argument. But it is possible that the uplevel Var
1513 * got optimized away by eval_const_expressions. Consider
1514 *
1515 * SUM(CASE WHEN false THEN uplevelvar ELSE 0 END)
1516 */
1517 if (contain_vars_of_level((Node *) newWhere, 1) ||
1518 contain_vars_of_level((Node *) rightargs, 1))
1519 return NULL;
1520 if (root->parse->hasAggs &&
1521 (contain_aggs_of_level((Node *) newWhere, 1) ||
1522 contain_aggs_of_level((Node *) rightargs, 1)))
1523 return NULL;
1524
1525 /*
1526 * And there can't be any child Vars in the stuff we intend to pull up.
1527 * (Note: we'd need to check for child Aggs too, except we know the child
1528 * has no aggs at all because of simplify_EXISTS_query's check. The same
1529 * goes for window functions.)
1530 */
1531 if (contain_vars_of_level((Node *) leftargs, 0))
1532 return NULL;
1533
1534 /*
1535 * Also reject sublinks in the stuff we intend to pull up. (It might be
1536 * possible to support this, but doesn't seem worth the complication.)
1537 */
1538 if (contain_subplans((Node *) leftargs))
1539 return NULL;
1540
1541 /*
1542 * Okay, adjust the sublevelsup in the stuff we're pulling up.
1543 */
1544 IncrementVarSublevelsUp((Node *) leftargs, -1, 1);
1545
1546 /*
1547 * Put back any child-level-only WHERE clauses.
1548 */
1549 if (newWhere)
1550 subselect->jointree->quals = (Node *) make_ands_explicit(newWhere);
1551
1552 /*
1553 * Build a new targetlist for the child that emits the expressions we
1554 * need. Concurrently, build a testexpr for the parent using Params to
1555 * reference the child outputs. (Since we generate Params directly here,
1556 * there will be no need to convert the testexpr in build_subplan.)
1557 */
1558 tlist = testlist = paramids = NIL;
1559 resno = 1;
1560 /* there's no "forfour" so we have to chase one of the lists manually */
1561 cc = list_head(opcollations);
1562 forthree(lc, leftargs, rc, rightargs, oc, opids)
1563 {
1564 Node *leftarg = (Node *) lfirst(lc);
1565 Node *rightarg = (Node *) lfirst(rc);
1566 Oid opid = lfirst_oid(oc);
1567 Oid opcollation = lfirst_oid(cc);
1568 Param *param;
1569
1570 cc = lnext(cc);
1571 param = generate_new_exec_param(root,
1572 exprType(rightarg),
1573 exprTypmod(rightarg),
1574 exprCollation(rightarg));
1575 tlist = lappend(tlist,
1576 makeTargetEntry((Expr *) rightarg,
1577 resno++,
1578 NULL,
1579 false));
1580 testlist = lappend(testlist,
1581 make_opclause(opid, BOOLOID, false,
1582 (Expr *) leftarg, (Expr *) param,
1583 InvalidOid, opcollation));
1584 paramids = lappend_int(paramids, param->paramid);
1585 }
1586
1587 /* Put everything where it should go, and we're done */
1588 subselect->targetList = tlist;
1589 *testexpr = (Node *) make_ands_explicit(testlist);
1590 *paramIds = paramids;
1591
1592 return subselect;
1593 }
1594
1595
1596 /*
1597 * Replace correlation vars (uplevel vars) with Params.
1598 *
1599 * Uplevel PlaceHolderVars and aggregates are replaced, too.
1600 *
1601 * Note: it is critical that this runs immediately after SS_process_sublinks.
1602 * Since we do not recurse into the arguments of uplevel PHVs and aggregates,
1603 * they will get copied to the appropriate subplan args list in the parent
1604 * query with uplevel vars not replaced by Params, but only adjusted in level
1605 * (see replace_outer_placeholdervar and replace_outer_agg). That's exactly
1606 * what we want for the vars of the parent level --- but if a PHV's or
1607 * aggregate's argument contains any further-up variables, they have to be
1608 * replaced with Params in their turn. That will happen when the parent level
1609 * runs SS_replace_correlation_vars. Therefore it must do so after expanding
1610 * its sublinks to subplans. And we don't want any steps in between, else
1611 * those steps would never get applied to the argument expressions, either in
1612 * the parent or the child level.
1613 *
1614 * Another fairly tricky thing going on here is the handling of SubLinks in
1615 * the arguments of uplevel PHVs/aggregates. Those are not touched inside the
1616 * intermediate query level, either. Instead, SS_process_sublinks recurses on
1617 * them after copying the PHV or Aggref expression into the parent plan level
1618 * (this is actually taken care of in build_subplan).
1619 */
1620 Node *
SS_replace_correlation_vars(PlannerInfo * root,Node * expr)1621 SS_replace_correlation_vars(PlannerInfo *root, Node *expr)
1622 {
1623 /* No setup needed for tree walk, so away we go */
1624 return replace_correlation_vars_mutator(expr, root);
1625 }
1626
1627 static Node *
replace_correlation_vars_mutator(Node * node,PlannerInfo * root)1628 replace_correlation_vars_mutator(Node *node, PlannerInfo *root)
1629 {
1630 if (node == NULL)
1631 return NULL;
1632 if (IsA(node, Var))
1633 {
1634 if (((Var *) node)->varlevelsup > 0)
1635 return (Node *) replace_outer_var(root, (Var *) node);
1636 }
1637 if (IsA(node, PlaceHolderVar))
1638 {
1639 if (((PlaceHolderVar *) node)->phlevelsup > 0)
1640 return (Node *) replace_outer_placeholdervar(root,
1641 (PlaceHolderVar *) node);
1642 }
1643 if (IsA(node, Aggref))
1644 {
1645 if (((Aggref *) node)->agglevelsup > 0)
1646 return (Node *) replace_outer_agg(root, (Aggref *) node);
1647 }
1648 if (IsA(node, GroupingFunc))
1649 {
1650 if (((GroupingFunc *) node)->agglevelsup > 0)
1651 return (Node *) replace_outer_grouping(root, (GroupingFunc *) node);
1652 }
1653 return expression_tree_mutator(node,
1654 replace_correlation_vars_mutator,
1655 (void *) root);
1656 }
1657
1658 /*
1659 * Expand SubLinks to SubPlans in the given expression.
1660 *
1661 * The isQual argument tells whether or not this expression is a WHERE/HAVING
1662 * qualifier expression. If it is, any sublinks appearing at top level need
1663 * not distinguish FALSE from UNKNOWN return values.
1664 */
1665 Node *
SS_process_sublinks(PlannerInfo * root,Node * expr,bool isQual)1666 SS_process_sublinks(PlannerInfo *root, Node *expr, bool isQual)
1667 {
1668 process_sublinks_context context;
1669
1670 context.root = root;
1671 context.isTopQual = isQual;
1672 return process_sublinks_mutator(expr, &context);
1673 }
1674
1675 static Node *
process_sublinks_mutator(Node * node,process_sublinks_context * context)1676 process_sublinks_mutator(Node *node, process_sublinks_context *context)
1677 {
1678 process_sublinks_context locContext;
1679
1680 locContext.root = context->root;
1681
1682 if (node == NULL)
1683 return NULL;
1684 if (IsA(node, SubLink))
1685 {
1686 SubLink *sublink = (SubLink *) node;
1687 Node *testexpr;
1688
1689 /*
1690 * First, recursively process the lefthand-side expressions, if any.
1691 * They're not top-level anymore.
1692 */
1693 locContext.isTopQual = false;
1694 testexpr = process_sublinks_mutator(sublink->testexpr, &locContext);
1695
1696 /*
1697 * Now build the SubPlan node and make the expr to return.
1698 */
1699 return make_subplan(context->root,
1700 (Query *) sublink->subselect,
1701 sublink->subLinkType,
1702 sublink->subLinkId,
1703 testexpr,
1704 context->isTopQual);
1705 }
1706
1707 /*
1708 * Don't recurse into the arguments of an outer PHV or aggregate here. Any
1709 * SubLinks in the arguments have to be dealt with at the outer query
1710 * level; they'll be handled when build_subplan collects the PHV or Aggref
1711 * into the arguments to be passed down to the current subplan.
1712 */
1713 if (IsA(node, PlaceHolderVar))
1714 {
1715 if (((PlaceHolderVar *) node)->phlevelsup > 0)
1716 return node;
1717 }
1718 else if (IsA(node, Aggref))
1719 {
1720 if (((Aggref *) node)->agglevelsup > 0)
1721 return node;
1722 }
1723
1724 /*
1725 * We should never see a SubPlan expression in the input (since this is
1726 * the very routine that creates 'em to begin with). We shouldn't find
1727 * ourselves invoked directly on a Query, either.
1728 */
1729 Assert(!IsA(node, SubPlan));
1730 Assert(!IsA(node, AlternativeSubPlan));
1731 Assert(!IsA(node, Query));
1732
1733 /*
1734 * Because make_subplan() could return an AND or OR clause, we have to
1735 * take steps to preserve AND/OR flatness of a qual. We assume the input
1736 * has been AND/OR flattened and so we need no recursion here.
1737 *
1738 * (Due to the coding here, we will not get called on the List subnodes of
1739 * an AND; and the input is *not* yet in implicit-AND format. So no check
1740 * is needed for a bare List.)
1741 *
1742 * Anywhere within the top-level AND/OR clause structure, we can tell
1743 * make_subplan() that NULL and FALSE are interchangeable. So isTopQual
1744 * propagates down in both cases. (Note that this is unlike the meaning
1745 * of "top level qual" used in most other places in Postgres.)
1746 */
1747 if (and_clause(node))
1748 {
1749 List *newargs = NIL;
1750 ListCell *l;
1751
1752 /* Still at qual top-level */
1753 locContext.isTopQual = context->isTopQual;
1754
1755 foreach(l, ((BoolExpr *) node)->args)
1756 {
1757 Node *newarg;
1758
1759 newarg = process_sublinks_mutator(lfirst(l), &locContext);
1760 if (and_clause(newarg))
1761 newargs = list_concat(newargs, ((BoolExpr *) newarg)->args);
1762 else
1763 newargs = lappend(newargs, newarg);
1764 }
1765 return (Node *) make_andclause(newargs);
1766 }
1767
1768 if (or_clause(node))
1769 {
1770 List *newargs = NIL;
1771 ListCell *l;
1772
1773 /* Still at qual top-level */
1774 locContext.isTopQual = context->isTopQual;
1775
1776 foreach(l, ((BoolExpr *) node)->args)
1777 {
1778 Node *newarg;
1779
1780 newarg = process_sublinks_mutator(lfirst(l), &locContext);
1781 if (or_clause(newarg))
1782 newargs = list_concat(newargs, ((BoolExpr *) newarg)->args);
1783 else
1784 newargs = lappend(newargs, newarg);
1785 }
1786 return (Node *) make_orclause(newargs);
1787 }
1788
1789 /*
1790 * If we recurse down through anything other than an AND or OR node, we
1791 * are definitely not at top qual level anymore.
1792 */
1793 locContext.isTopQual = false;
1794
1795 return expression_tree_mutator(node,
1796 process_sublinks_mutator,
1797 (void *) &locContext);
1798 }
1799
1800 /*
1801 * SS_identify_outer_params - identify the Params available from outer levels
1802 *
1803 * This must be run after SS_replace_correlation_vars and SS_process_sublinks
1804 * processing is complete in a given query level as well as all of its
1805 * descendant levels (which means it's most practical to do it at the end of
1806 * processing the query level). We compute the set of paramIds that outer
1807 * levels will make available to this level+descendants, and record it in
1808 * root->outer_params for use while computing extParam/allParam sets in final
1809 * plan cleanup. (We can't just compute it then, because the upper levels'
1810 * plan_params lists are transient and will be gone by then.)
1811 */
1812 void
SS_identify_outer_params(PlannerInfo * root)1813 SS_identify_outer_params(PlannerInfo *root)
1814 {
1815 Bitmapset *outer_params;
1816 PlannerInfo *proot;
1817 ListCell *l;
1818
1819 /*
1820 * If no parameters have been assigned anywhere in the tree, we certainly
1821 * don't need to do anything here.
1822 */
1823 if (root->glob->nParamExec == 0)
1824 return;
1825
1826 /*
1827 * Scan all query levels above this one to see which parameters are due to
1828 * be available from them, either because lower query levels have
1829 * requested them (via plan_params) or because they will be available from
1830 * initPlans of those levels.
1831 */
1832 outer_params = NULL;
1833 for (proot = root->parent_root; proot != NULL; proot = proot->parent_root)
1834 {
1835 /* Include ordinary Var/PHV/Aggref params */
1836 foreach(l, proot->plan_params)
1837 {
1838 PlannerParamItem *pitem = (PlannerParamItem *) lfirst(l);
1839
1840 outer_params = bms_add_member(outer_params, pitem->paramId);
1841 }
1842 /* Include any outputs of outer-level initPlans */
1843 foreach(l, proot->init_plans)
1844 {
1845 SubPlan *initsubplan = (SubPlan *) lfirst(l);
1846 ListCell *l2;
1847
1848 foreach(l2, initsubplan->setParam)
1849 {
1850 outer_params = bms_add_member(outer_params, lfirst_int(l2));
1851 }
1852 }
1853 /* Include worktable ID, if a recursive query is being planned */
1854 if (proot->wt_param_id >= 0)
1855 outer_params = bms_add_member(outer_params, proot->wt_param_id);
1856 }
1857 root->outer_params = outer_params;
1858 }
1859
1860 /*
1861 * SS_charge_for_initplans - account for initplans in Path costs & parallelism
1862 *
1863 * If any initPlans have been created in the current query level, they will
1864 * get attached to the Plan tree created from whichever Path we select from
1865 * the given rel. Increment all that rel's Paths' costs to account for them,
1866 * and make sure the paths get marked as parallel-unsafe, since we can't
1867 * currently transmit initPlans to parallel workers.
1868 *
1869 * This is separate from SS_attach_initplans because we might conditionally
1870 * create more initPlans during create_plan(), depending on which Path we
1871 * select. However, Paths that would generate such initPlans are expected
1872 * to have included their cost already.
1873 */
1874 void
SS_charge_for_initplans(PlannerInfo * root,RelOptInfo * final_rel)1875 SS_charge_for_initplans(PlannerInfo *root, RelOptInfo *final_rel)
1876 {
1877 Cost initplan_cost;
1878 ListCell *lc;
1879
1880 /* Nothing to do if no initPlans */
1881 if (root->init_plans == NIL)
1882 return;
1883
1884 /*
1885 * Compute the cost increment just once, since it will be the same for all
1886 * Paths. We assume each initPlan gets run once during top plan startup.
1887 * This is a conservative overestimate, since in fact an initPlan might be
1888 * executed later than plan startup, or even not at all.
1889 */
1890 initplan_cost = 0;
1891 foreach(lc, root->init_plans)
1892 {
1893 SubPlan *initsubplan = (SubPlan *) lfirst(lc);
1894
1895 initplan_cost += initsubplan->startup_cost + initsubplan->per_call_cost;
1896 }
1897
1898 /*
1899 * Now adjust the costs and parallel_safe flags.
1900 */
1901 foreach(lc, final_rel->pathlist)
1902 {
1903 Path *path = (Path *) lfirst(lc);
1904
1905 path->startup_cost += initplan_cost;
1906 path->total_cost += initplan_cost;
1907 path->parallel_safe = false;
1908 }
1909
1910 /* We needn't do set_cheapest() here, caller will do it */
1911 }
1912
1913 /*
1914 * SS_attach_initplans - attach initplans to topmost plan node
1915 *
1916 * Attach any initplans created in the current query level to the specified
1917 * plan node, which should normally be the topmost node for the query level.
1918 * (In principle the initPlans could go in any node at or above where they're
1919 * referenced; but there seems no reason to put them any lower than the
1920 * topmost node, so we don't bother to track exactly where they came from.)
1921 * We do not touch the plan node's cost; the initplans should have been
1922 * accounted for in path costing.
1923 */
1924 void
SS_attach_initplans(PlannerInfo * root,Plan * plan)1925 SS_attach_initplans(PlannerInfo *root, Plan *plan)
1926 {
1927 plan->initPlan = root->init_plans;
1928 }
1929
1930 /*
1931 * SS_finalize_plan - do final parameter processing for a completed Plan.
1932 *
1933 * This recursively computes the extParam and allParam sets for every Plan
1934 * node in the given plan tree. (Oh, and RangeTblFunction.funcparams too.)
1935 *
1936 * We assume that SS_finalize_plan has already been run on any initplans or
1937 * subplans the plan tree could reference.
1938 */
1939 void
SS_finalize_plan(PlannerInfo * root,Plan * plan)1940 SS_finalize_plan(PlannerInfo *root, Plan *plan)
1941 {
1942 /* No setup needed, just recurse through plan tree. */
1943 (void) finalize_plan(root, plan, -1, root->outer_params, NULL);
1944 }
1945
1946 /*
1947 * Recursive processing of all nodes in the plan tree
1948 *
1949 * gather_param is the rescan_param of an ancestral Gather/GatherMerge,
1950 * or -1 if there is none.
1951 *
1952 * valid_params is the set of param IDs supplied by outer plan levels
1953 * that are valid to reference in this plan node or its children.
1954 *
1955 * scan_params is a set of param IDs to force scan plan nodes to reference.
1956 * This is for EvalPlanQual support, and is always NULL at the top of the
1957 * recursion.
1958 *
1959 * The return value is the computed allParam set for the given Plan node.
1960 * This is just an internal notational convenience: we can add a child
1961 * plan's allParams to the set of param IDs of interest to this level
1962 * in the same statement that recurses to that child.
1963 *
1964 * Do not scribble on caller's values of valid_params or scan_params!
1965 *
1966 * Note: although we attempt to deal with initPlans anywhere in the tree, the
1967 * logic is not really right. The problem is that a plan node might return an
1968 * output Param of its initPlan as a targetlist item, in which case it's valid
1969 * for the parent plan level to reference that same Param; the parent's usage
1970 * will be converted into a Var referencing the child plan node by setrefs.c.
1971 * But this function would see the parent's reference as out of scope and
1972 * complain about it. For now, this does not matter because the planner only
1973 * attaches initPlans to the topmost plan node in a query level, so the case
1974 * doesn't arise. If we ever merge this processing into setrefs.c, maybe it
1975 * can be handled more cleanly.
1976 */
1977 static Bitmapset *
finalize_plan(PlannerInfo * root,Plan * plan,int gather_param,Bitmapset * valid_params,Bitmapset * scan_params)1978 finalize_plan(PlannerInfo *root, Plan *plan,
1979 int gather_param,
1980 Bitmapset *valid_params,
1981 Bitmapset *scan_params)
1982 {
1983 finalize_primnode_context context;
1984 int locally_added_param;
1985 Bitmapset *nestloop_params;
1986 Bitmapset *initExtParam;
1987 Bitmapset *initSetParam;
1988 Bitmapset *child_params;
1989 ListCell *l;
1990
1991 if (plan == NULL)
1992 return NULL;
1993
1994 context.root = root;
1995 context.paramids = NULL; /* initialize set to empty */
1996 locally_added_param = -1; /* there isn't one */
1997 nestloop_params = NULL; /* there aren't any */
1998
1999 /*
2000 * Examine any initPlans to determine the set of external params they
2001 * reference and the set of output params they supply. (We assume
2002 * SS_finalize_plan was run on them already.)
2003 */
2004 initExtParam = initSetParam = NULL;
2005 foreach(l, plan->initPlan)
2006 {
2007 SubPlan *initsubplan = (SubPlan *) lfirst(l);
2008 Plan *initplan = planner_subplan_get_plan(root, initsubplan);
2009 ListCell *l2;
2010
2011 initExtParam = bms_add_members(initExtParam, initplan->extParam);
2012 foreach(l2, initsubplan->setParam)
2013 {
2014 initSetParam = bms_add_member(initSetParam, lfirst_int(l2));
2015 }
2016 }
2017
2018 /* Any setParams are validly referenceable in this node and children */
2019 if (initSetParam)
2020 valid_params = bms_union(valid_params, initSetParam);
2021
2022 /*
2023 * When we call finalize_primnode, context.paramids sets are automatically
2024 * merged together. But when recursing to self, we have to do it the hard
2025 * way. We want the paramids set to include params in subplans as well as
2026 * at this level.
2027 */
2028
2029 /* Find params in targetlist and qual */
2030 finalize_primnode((Node *) plan->targetlist, &context);
2031 finalize_primnode((Node *) plan->qual, &context);
2032
2033 /*
2034 * If it's a parallel-aware scan node, mark it as dependent on the parent
2035 * Gather/GatherMerge's rescan Param.
2036 */
2037 if (plan->parallel_aware)
2038 {
2039 if (gather_param < 0)
2040 elog(ERROR, "parallel-aware plan node is not below a Gather");
2041 context.paramids =
2042 bms_add_member(context.paramids, gather_param);
2043 }
2044
2045 /* Check additional node-type-specific fields */
2046 switch (nodeTag(plan))
2047 {
2048 case T_Result:
2049 finalize_primnode(((Result *) plan)->resconstantqual,
2050 &context);
2051 break;
2052
2053 case T_SeqScan:
2054 context.paramids = bms_add_members(context.paramids, scan_params);
2055 break;
2056
2057 case T_SampleScan:
2058 finalize_primnode((Node *) ((SampleScan *) plan)->tablesample,
2059 &context);
2060 context.paramids = bms_add_members(context.paramids, scan_params);
2061 break;
2062
2063 case T_IndexScan:
2064 finalize_primnode((Node *) ((IndexScan *) plan)->indexqual,
2065 &context);
2066 finalize_primnode((Node *) ((IndexScan *) plan)->indexorderby,
2067 &context);
2068
2069 /*
2070 * we need not look at indexqualorig, since it will have the same
2071 * param references as indexqual. Likewise, we can ignore
2072 * indexorderbyorig.
2073 */
2074 context.paramids = bms_add_members(context.paramids, scan_params);
2075 break;
2076
2077 case T_IndexOnlyScan:
2078 finalize_primnode((Node *) ((IndexOnlyScan *) plan)->indexqual,
2079 &context);
2080 finalize_primnode((Node *) ((IndexOnlyScan *) plan)->indexorderby,
2081 &context);
2082
2083 /*
2084 * we need not look at indextlist, since it cannot contain Params.
2085 */
2086 context.paramids = bms_add_members(context.paramids, scan_params);
2087 break;
2088
2089 case T_BitmapIndexScan:
2090 finalize_primnode((Node *) ((BitmapIndexScan *) plan)->indexqual,
2091 &context);
2092
2093 /*
2094 * we need not look at indexqualorig, since it will have the same
2095 * param references as indexqual.
2096 */
2097 break;
2098
2099 case T_BitmapHeapScan:
2100 finalize_primnode((Node *) ((BitmapHeapScan *) plan)->bitmapqualorig,
2101 &context);
2102 context.paramids = bms_add_members(context.paramids, scan_params);
2103 break;
2104
2105 case T_TidScan:
2106 finalize_primnode((Node *) ((TidScan *) plan)->tidquals,
2107 &context);
2108 context.paramids = bms_add_members(context.paramids, scan_params);
2109 break;
2110
2111 case T_SubqueryScan:
2112 {
2113 SubqueryScan *sscan = (SubqueryScan *) plan;
2114 RelOptInfo *rel;
2115
2116 /* We must run SS_finalize_plan on the subquery */
2117 rel = find_base_rel(root, sscan->scan.scanrelid);
2118 SS_finalize_plan(rel->subroot, sscan->subplan);
2119
2120 /* Now we can add its extParams to the parent's params */
2121 context.paramids = bms_add_members(context.paramids,
2122 sscan->subplan->extParam);
2123 /* We need scan_params too, though */
2124 context.paramids = bms_add_members(context.paramids,
2125 scan_params);
2126 }
2127 break;
2128
2129 case T_FunctionScan:
2130 {
2131 FunctionScan *fscan = (FunctionScan *) plan;
2132 ListCell *lc;
2133
2134 /*
2135 * Call finalize_primnode independently on each function
2136 * expression, so that we can record which params are
2137 * referenced in each, in order to decide which need
2138 * re-evaluating during rescan.
2139 */
2140 foreach(lc, fscan->functions)
2141 {
2142 RangeTblFunction *rtfunc = (RangeTblFunction *) lfirst(lc);
2143 finalize_primnode_context funccontext;
2144
2145 funccontext = context;
2146 funccontext.paramids = NULL;
2147
2148 finalize_primnode(rtfunc->funcexpr, &funccontext);
2149
2150 /* remember results for execution */
2151 rtfunc->funcparams = funccontext.paramids;
2152
2153 /* add the function's params to the overall set */
2154 context.paramids = bms_add_members(context.paramids,
2155 funccontext.paramids);
2156 }
2157
2158 context.paramids = bms_add_members(context.paramids,
2159 scan_params);
2160 }
2161 break;
2162
2163 case T_TableFuncScan:
2164 finalize_primnode((Node *) ((TableFuncScan *) plan)->tablefunc,
2165 &context);
2166 context.paramids = bms_add_members(context.paramids, scan_params);
2167 break;
2168
2169 case T_ValuesScan:
2170 finalize_primnode((Node *) ((ValuesScan *) plan)->values_lists,
2171 &context);
2172 context.paramids = bms_add_members(context.paramids, scan_params);
2173 break;
2174
2175 case T_CteScan:
2176 {
2177 /*
2178 * You might think we should add the node's cteParam to
2179 * paramids, but we shouldn't because that param is just a
2180 * linkage mechanism for multiple CteScan nodes for the same
2181 * CTE; it is never used for changed-param signaling. What we
2182 * have to do instead is to find the referenced CTE plan and
2183 * incorporate its external paramids, so that the correct
2184 * things will happen if the CTE references outer-level
2185 * variables. See test cases for bug #4902. (We assume
2186 * SS_finalize_plan was run on the CTE plan already.)
2187 */
2188 int plan_id = ((CteScan *) plan)->ctePlanId;
2189 Plan *cteplan;
2190
2191 /* so, do this ... */
2192 if (plan_id < 1 || plan_id > list_length(root->glob->subplans))
2193 elog(ERROR, "could not find plan for CteScan referencing plan ID %d",
2194 plan_id);
2195 cteplan = (Plan *) list_nth(root->glob->subplans, plan_id - 1);
2196 context.paramids =
2197 bms_add_members(context.paramids, cteplan->extParam);
2198
2199 #ifdef NOT_USED
2200 /* ... but not this */
2201 context.paramids =
2202 bms_add_member(context.paramids,
2203 ((CteScan *) plan)->cteParam);
2204 #endif
2205
2206 context.paramids = bms_add_members(context.paramids,
2207 scan_params);
2208 }
2209 break;
2210
2211 case T_WorkTableScan:
2212 context.paramids =
2213 bms_add_member(context.paramids,
2214 ((WorkTableScan *) plan)->wtParam);
2215 context.paramids = bms_add_members(context.paramids, scan_params);
2216 break;
2217
2218 case T_NamedTuplestoreScan:
2219 context.paramids = bms_add_members(context.paramids, scan_params);
2220 break;
2221
2222 case T_ForeignScan:
2223 {
2224 ForeignScan *fscan = (ForeignScan *) plan;
2225
2226 finalize_primnode((Node *) fscan->fdw_exprs,
2227 &context);
2228 finalize_primnode((Node *) fscan->fdw_recheck_quals,
2229 &context);
2230
2231 /* We assume fdw_scan_tlist cannot contain Params */
2232 context.paramids = bms_add_members(context.paramids,
2233 scan_params);
2234 }
2235 break;
2236
2237 case T_CustomScan:
2238 {
2239 CustomScan *cscan = (CustomScan *) plan;
2240 ListCell *lc;
2241
2242 finalize_primnode((Node *) cscan->custom_exprs,
2243 &context);
2244 /* We assume custom_scan_tlist cannot contain Params */
2245 context.paramids =
2246 bms_add_members(context.paramids, scan_params);
2247
2248 /* child nodes if any */
2249 foreach(lc, cscan->custom_plans)
2250 {
2251 context.paramids =
2252 bms_add_members(context.paramids,
2253 finalize_plan(root,
2254 (Plan *) lfirst(lc),
2255 gather_param,
2256 valid_params,
2257 scan_params));
2258 }
2259 }
2260 break;
2261
2262 case T_ModifyTable:
2263 {
2264 ModifyTable *mtplan = (ModifyTable *) plan;
2265 ListCell *l;
2266
2267 /* Force descendant scan nodes to reference epqParam */
2268 locally_added_param = mtplan->epqParam;
2269 valid_params = bms_add_member(bms_copy(valid_params),
2270 locally_added_param);
2271 scan_params = bms_add_member(bms_copy(scan_params),
2272 locally_added_param);
2273 finalize_primnode((Node *) mtplan->returningLists,
2274 &context);
2275 finalize_primnode((Node *) mtplan->onConflictSet,
2276 &context);
2277 finalize_primnode((Node *) mtplan->onConflictWhere,
2278 &context);
2279 /* exclRelTlist contains only Vars, doesn't need examination */
2280 foreach(l, mtplan->plans)
2281 {
2282 context.paramids =
2283 bms_add_members(context.paramids,
2284 finalize_plan(root,
2285 (Plan *) lfirst(l),
2286 gather_param,
2287 valid_params,
2288 scan_params));
2289 }
2290 }
2291 break;
2292
2293 case T_Append:
2294 {
2295 ListCell *l;
2296
2297 foreach(l, ((Append *) plan)->appendplans)
2298 {
2299 context.paramids =
2300 bms_add_members(context.paramids,
2301 finalize_plan(root,
2302 (Plan *) lfirst(l),
2303 gather_param,
2304 valid_params,
2305 scan_params));
2306 }
2307 }
2308 break;
2309
2310 case T_MergeAppend:
2311 {
2312 ListCell *l;
2313
2314 foreach(l, ((MergeAppend *) plan)->mergeplans)
2315 {
2316 context.paramids =
2317 bms_add_members(context.paramids,
2318 finalize_plan(root,
2319 (Plan *) lfirst(l),
2320 gather_param,
2321 valid_params,
2322 scan_params));
2323 }
2324 }
2325 break;
2326
2327 case T_BitmapAnd:
2328 {
2329 ListCell *l;
2330
2331 foreach(l, ((BitmapAnd *) plan)->bitmapplans)
2332 {
2333 context.paramids =
2334 bms_add_members(context.paramids,
2335 finalize_plan(root,
2336 (Plan *) lfirst(l),
2337 gather_param,
2338 valid_params,
2339 scan_params));
2340 }
2341 }
2342 break;
2343
2344 case T_BitmapOr:
2345 {
2346 ListCell *l;
2347
2348 foreach(l, ((BitmapOr *) plan)->bitmapplans)
2349 {
2350 context.paramids =
2351 bms_add_members(context.paramids,
2352 finalize_plan(root,
2353 (Plan *) lfirst(l),
2354 gather_param,
2355 valid_params,
2356 scan_params));
2357 }
2358 }
2359 break;
2360
2361 case T_NestLoop:
2362 {
2363 ListCell *l;
2364
2365 finalize_primnode((Node *) ((Join *) plan)->joinqual,
2366 &context);
2367 /* collect set of params that will be passed to right child */
2368 foreach(l, ((NestLoop *) plan)->nestParams)
2369 {
2370 NestLoopParam *nlp = (NestLoopParam *) lfirst(l);
2371
2372 nestloop_params = bms_add_member(nestloop_params,
2373 nlp->paramno);
2374 }
2375 }
2376 break;
2377
2378 case T_MergeJoin:
2379 finalize_primnode((Node *) ((Join *) plan)->joinqual,
2380 &context);
2381 finalize_primnode((Node *) ((MergeJoin *) plan)->mergeclauses,
2382 &context);
2383 break;
2384
2385 case T_HashJoin:
2386 finalize_primnode((Node *) ((Join *) plan)->joinqual,
2387 &context);
2388 finalize_primnode((Node *) ((HashJoin *) plan)->hashclauses,
2389 &context);
2390 break;
2391
2392 case T_Limit:
2393 finalize_primnode(((Limit *) plan)->limitOffset,
2394 &context);
2395 finalize_primnode(((Limit *) plan)->limitCount,
2396 &context);
2397 break;
2398
2399 case T_RecursiveUnion:
2400 /* child nodes are allowed to reference wtParam */
2401 locally_added_param = ((RecursiveUnion *) plan)->wtParam;
2402 valid_params = bms_add_member(bms_copy(valid_params),
2403 locally_added_param);
2404 /* wtParam does *not* get added to scan_params */
2405 break;
2406
2407 case T_LockRows:
2408 /* Force descendant scan nodes to reference epqParam */
2409 locally_added_param = ((LockRows *) plan)->epqParam;
2410 valid_params = bms_add_member(bms_copy(valid_params),
2411 locally_added_param);
2412 scan_params = bms_add_member(bms_copy(scan_params),
2413 locally_added_param);
2414 break;
2415
2416 case T_Agg:
2417 {
2418 Agg *agg = (Agg *) plan;
2419
2420 /*
2421 * AGG_HASHED plans need to know which Params are referenced
2422 * in aggregate calls. Do a separate scan to identify them.
2423 */
2424 if (agg->aggstrategy == AGG_HASHED)
2425 {
2426 finalize_primnode_context aggcontext;
2427
2428 aggcontext.root = root;
2429 aggcontext.paramids = NULL;
2430 finalize_agg_primnode((Node *) agg->plan.targetlist,
2431 &aggcontext);
2432 finalize_agg_primnode((Node *) agg->plan.qual,
2433 &aggcontext);
2434 agg->aggParams = aggcontext.paramids;
2435 }
2436 }
2437 break;
2438
2439 case T_WindowAgg:
2440 finalize_primnode(((WindowAgg *) plan)->startOffset,
2441 &context);
2442 finalize_primnode(((WindowAgg *) plan)->endOffset,
2443 &context);
2444 break;
2445
2446 case T_Gather:
2447 /* child nodes are allowed to reference rescan_param, if any */
2448 locally_added_param = ((Gather *) plan)->rescan_param;
2449 if (locally_added_param >= 0)
2450 {
2451 valid_params = bms_add_member(bms_copy(valid_params),
2452 locally_added_param);
2453
2454 /*
2455 * We currently don't support nested Gathers. The issue so
2456 * far as this function is concerned would be how to identify
2457 * which child nodes depend on which Gather.
2458 */
2459 Assert(gather_param < 0);
2460 /* Pass down rescan_param to child parallel-aware nodes */
2461 gather_param = locally_added_param;
2462 }
2463 /* rescan_param does *not* get added to scan_params */
2464 break;
2465
2466 case T_GatherMerge:
2467 /* child nodes are allowed to reference rescan_param, if any */
2468 locally_added_param = ((GatherMerge *) plan)->rescan_param;
2469 if (locally_added_param >= 0)
2470 {
2471 valid_params = bms_add_member(bms_copy(valid_params),
2472 locally_added_param);
2473
2474 /*
2475 * We currently don't support nested Gathers. The issue so
2476 * far as this function is concerned would be how to identify
2477 * which child nodes depend on which Gather.
2478 */
2479 Assert(gather_param < 0);
2480 /* Pass down rescan_param to child parallel-aware nodes */
2481 gather_param = locally_added_param;
2482 }
2483 /* rescan_param does *not* get added to scan_params */
2484 break;
2485
2486 case T_ProjectSet:
2487 case T_Hash:
2488 case T_Material:
2489 case T_Sort:
2490 case T_Unique:
2491 case T_SetOp:
2492 case T_Group:
2493 /* no node-type-specific fields need fixing */
2494 break;
2495
2496 default:
2497 elog(ERROR, "unrecognized node type: %d",
2498 (int) nodeTag(plan));
2499 }
2500
2501 /* Process left and right child plans, if any */
2502 child_params = finalize_plan(root,
2503 plan->lefttree,
2504 gather_param,
2505 valid_params,
2506 scan_params);
2507 context.paramids = bms_add_members(context.paramids, child_params);
2508
2509 if (nestloop_params)
2510 {
2511 /* right child can reference nestloop_params as well as valid_params */
2512 child_params = finalize_plan(root,
2513 plan->righttree,
2514 gather_param,
2515 bms_union(nestloop_params, valid_params),
2516 scan_params);
2517 /* ... and they don't count as parameters used at my level */
2518 child_params = bms_difference(child_params, nestloop_params);
2519 bms_free(nestloop_params);
2520 }
2521 else
2522 {
2523 /* easy case */
2524 child_params = finalize_plan(root,
2525 plan->righttree,
2526 gather_param,
2527 valid_params,
2528 scan_params);
2529 }
2530 context.paramids = bms_add_members(context.paramids, child_params);
2531
2532 /*
2533 * Any locally generated parameter doesn't count towards its generating
2534 * plan node's external dependencies. (Note: if we changed valid_params
2535 * and/or scan_params, we leak those bitmapsets; not worth the notational
2536 * trouble to clean them up.)
2537 */
2538 if (locally_added_param >= 0)
2539 {
2540 context.paramids = bms_del_member(context.paramids,
2541 locally_added_param);
2542 }
2543
2544 /* Now we have all the paramids referenced in this node and children */
2545
2546 if (!bms_is_subset(context.paramids, valid_params))
2547 elog(ERROR, "plan should not reference subplan's variable");
2548
2549 /*
2550 * The plan node's allParam and extParam fields should include all its
2551 * referenced paramids, plus contributions from any child initPlans.
2552 * However, any setParams of the initPlans should not be present in the
2553 * parent node's extParams, only in its allParams. (It's possible that
2554 * some initPlans have extParams that are setParams of other initPlans.)
2555 */
2556
2557 /* allParam must include initplans' extParams and setParams */
2558 plan->allParam = bms_union(context.paramids, initExtParam);
2559 plan->allParam = bms_add_members(plan->allParam, initSetParam);
2560 /* extParam must include any initplan extParams */
2561 plan->extParam = bms_union(context.paramids, initExtParam);
2562 /* but not any initplan setParams */
2563 plan->extParam = bms_del_members(plan->extParam, initSetParam);
2564
2565 /*
2566 * For speed at execution time, make sure extParam/allParam are actually
2567 * NULL if they are empty sets.
2568 */
2569 if (bms_is_empty(plan->extParam))
2570 plan->extParam = NULL;
2571 if (bms_is_empty(plan->allParam))
2572 plan->allParam = NULL;
2573
2574 return plan->allParam;
2575 }
2576
2577 /*
2578 * finalize_primnode: add IDs of all PARAM_EXEC params appearing in the given
2579 * expression tree to the result set.
2580 */
2581 static bool
finalize_primnode(Node * node,finalize_primnode_context * context)2582 finalize_primnode(Node *node, finalize_primnode_context *context)
2583 {
2584 if (node == NULL)
2585 return false;
2586 if (IsA(node, Param))
2587 {
2588 if (((Param *) node)->paramkind == PARAM_EXEC)
2589 {
2590 int paramid = ((Param *) node)->paramid;
2591
2592 context->paramids = bms_add_member(context->paramids, paramid);
2593 }
2594 return false; /* no more to do here */
2595 }
2596 if (IsA(node, SubPlan))
2597 {
2598 SubPlan *subplan = (SubPlan *) node;
2599 Plan *plan = planner_subplan_get_plan(context->root, subplan);
2600 ListCell *lc;
2601 Bitmapset *subparamids;
2602
2603 /* Recurse into the testexpr, but not into the Plan */
2604 finalize_primnode(subplan->testexpr, context);
2605
2606 /*
2607 * Remove any param IDs of output parameters of the subplan that were
2608 * referenced in the testexpr. These are not interesting for
2609 * parameter change signaling since we always re-evaluate the subplan.
2610 * Note that this wouldn't work too well if there might be uses of the
2611 * same param IDs elsewhere in the plan, but that can't happen because
2612 * generate_new_exec_param never tries to merge params.
2613 */
2614 foreach(lc, subplan->paramIds)
2615 {
2616 context->paramids = bms_del_member(context->paramids,
2617 lfirst_int(lc));
2618 }
2619
2620 /* Also examine args list */
2621 finalize_primnode((Node *) subplan->args, context);
2622
2623 /*
2624 * Add params needed by the subplan to paramids, but excluding those
2625 * we will pass down to it. (We assume SS_finalize_plan was run on
2626 * the subplan already.)
2627 */
2628 subparamids = bms_copy(plan->extParam);
2629 foreach(lc, subplan->parParam)
2630 {
2631 subparamids = bms_del_member(subparamids, lfirst_int(lc));
2632 }
2633 context->paramids = bms_join(context->paramids, subparamids);
2634
2635 return false; /* no more to do here */
2636 }
2637 return expression_tree_walker(node, finalize_primnode,
2638 (void *) context);
2639 }
2640
2641 /*
2642 * finalize_agg_primnode: find all Aggref nodes in the given expression tree,
2643 * and add IDs of all PARAM_EXEC params appearing within their aggregated
2644 * arguments to the result set.
2645 */
2646 static bool
finalize_agg_primnode(Node * node,finalize_primnode_context * context)2647 finalize_agg_primnode(Node *node, finalize_primnode_context *context)
2648 {
2649 if (node == NULL)
2650 return false;
2651 if (IsA(node, Aggref))
2652 {
2653 Aggref *agg = (Aggref *) node;
2654
2655 /* we should not consider the direct arguments, if any */
2656 finalize_primnode((Node *) agg->args, context);
2657 finalize_primnode((Node *) agg->aggfilter, context);
2658 return false; /* there can't be any Aggrefs below here */
2659 }
2660 return expression_tree_walker(node, finalize_agg_primnode,
2661 (void *) context);
2662 }
2663
2664 /*
2665 * SS_make_initplan_output_param - make a Param for an initPlan's output
2666 *
2667 * The plan is expected to return a scalar value of the given type/collation.
2668 *
2669 * Note that in some cases the initplan may not ever appear in the finished
2670 * plan tree. If that happens, we'll have wasted a PARAM_EXEC slot, which
2671 * is no big deal.
2672 */
2673 Param *
SS_make_initplan_output_param(PlannerInfo * root,Oid resulttype,int32 resulttypmod,Oid resultcollation)2674 SS_make_initplan_output_param(PlannerInfo *root,
2675 Oid resulttype, int32 resulttypmod,
2676 Oid resultcollation)
2677 {
2678 return generate_new_exec_param(root, resulttype,
2679 resulttypmod, resultcollation);
2680 }
2681
2682 /*
2683 * SS_make_initplan_from_plan - given a plan tree, make it an InitPlan
2684 *
2685 * We build an EXPR_SUBLINK SubPlan node and put it into the initplan
2686 * list for the outer query level. A Param that represents the initplan's
2687 * output has already been assigned using SS_make_initplan_output_param.
2688 */
2689 void
SS_make_initplan_from_plan(PlannerInfo * root,PlannerInfo * subroot,Plan * plan,Param * prm)2690 SS_make_initplan_from_plan(PlannerInfo *root,
2691 PlannerInfo *subroot, Plan *plan,
2692 Param *prm)
2693 {
2694 SubPlan *node;
2695
2696 /*
2697 * Add the subplan and its PlannerInfo to the global lists.
2698 */
2699 root->glob->subplans = lappend(root->glob->subplans, plan);
2700 root->glob->subroots = lappend(root->glob->subroots, subroot);
2701
2702 /*
2703 * Create a SubPlan node and add it to the outer list of InitPlans. Note
2704 * it has to appear after any other InitPlans it might depend on (see
2705 * comments in ExecReScan).
2706 */
2707 node = makeNode(SubPlan);
2708 node->subLinkType = EXPR_SUBLINK;
2709 node->plan_id = list_length(root->glob->subplans);
2710 node->plan_name = psprintf("InitPlan %d (returns $%d)",
2711 node->plan_id, prm->paramid);
2712 get_first_col_type(plan, &node->firstColType, &node->firstColTypmod,
2713 &node->firstColCollation);
2714 node->setParam = list_make1_int(prm->paramid);
2715
2716 root->init_plans = lappend(root->init_plans, node);
2717
2718 /*
2719 * The node can't have any inputs (since it's an initplan), so the
2720 * parParam and args lists remain empty.
2721 */
2722
2723 /* Set costs of SubPlan using info from the plan tree */
2724 cost_subplan(subroot, node, plan);
2725 }
2726