Line data Source code
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) 1996-2025, 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 "executor/nodeSubplan.h"
24 : #include "miscadmin.h"
25 : #include "nodes/makefuncs.h"
26 : #include "nodes/nodeFuncs.h"
27 : #include "optimizer/clauses.h"
28 : #include "optimizer/cost.h"
29 : #include "optimizer/optimizer.h"
30 : #include "optimizer/paramassign.h"
31 : #include "optimizer/pathnode.h"
32 : #include "optimizer/planmain.h"
33 : #include "optimizer/planner.h"
34 : #include "optimizer/prep.h"
35 : #include "optimizer/subselect.h"
36 : #include "parser/parse_relation.h"
37 : #include "rewrite/rewriteManip.h"
38 : #include "utils/builtins.h"
39 : #include "utils/lsyscache.h"
40 : #include "utils/syscache.h"
41 :
42 :
43 : typedef struct convert_testexpr_context
44 : {
45 : PlannerInfo *root;
46 : List *subst_nodes; /* Nodes to substitute for Params */
47 : } convert_testexpr_context;
48 :
49 : typedef struct process_sublinks_context
50 : {
51 : PlannerInfo *root;
52 : bool isTopQual;
53 : } process_sublinks_context;
54 :
55 : typedef struct finalize_primnode_context
56 : {
57 : PlannerInfo *root;
58 : Bitmapset *paramids; /* Non-local PARAM_EXEC paramids found */
59 : } finalize_primnode_context;
60 :
61 : typedef struct inline_cte_walker_context
62 : {
63 : const char *ctename; /* name and relative level of target CTE */
64 : int levelsup;
65 : Query *ctequery; /* query to substitute */
66 : } inline_cte_walker_context;
67 :
68 :
69 : static Node *build_subplan(PlannerInfo *root, Plan *plan, Path *path,
70 : PlannerInfo *subroot, List *plan_params,
71 : SubLinkType subLinkType, int subLinkId,
72 : Node *testexpr, List *testexpr_paramids,
73 : bool unknownEqFalse);
74 : static List *generate_subquery_params(PlannerInfo *root, List *tlist,
75 : List **paramIds);
76 : static List *generate_subquery_vars(PlannerInfo *root, List *tlist,
77 : Index varno);
78 : static Node *convert_testexpr(PlannerInfo *root,
79 : Node *testexpr,
80 : List *subst_nodes);
81 : static Node *convert_testexpr_mutator(Node *node,
82 : convert_testexpr_context *context);
83 : static bool subplan_is_hashable(Plan *plan, bool unknownEqFalse);
84 : static bool subpath_is_hashable(Path *path, bool unknownEqFalse);
85 : static bool testexpr_is_hashable(Node *testexpr, List *param_ids);
86 : static bool test_opexpr_is_hashable(OpExpr *testexpr, List *param_ids);
87 : static bool hash_ok_operator(OpExpr *expr);
88 : static bool contain_dml(Node *node);
89 : static bool contain_dml_walker(Node *node, void *context);
90 : static bool contain_outer_selfref(Node *node);
91 : static bool contain_outer_selfref_walker(Node *node, Index *depth);
92 : static void inline_cte(PlannerInfo *root, CommonTableExpr *cte);
93 : static bool inline_cte_walker(Node *node, inline_cte_walker_context *context);
94 : static bool simplify_EXISTS_query(PlannerInfo *root, Query *query);
95 : static Query *convert_EXISTS_to_ANY(PlannerInfo *root, Query *subselect,
96 : Node **testexpr, List **paramIds);
97 : static Node *replace_correlation_vars_mutator(Node *node, PlannerInfo *root);
98 : static Node *process_sublinks_mutator(Node *node,
99 : process_sublinks_context *context);
100 : static Bitmapset *finalize_plan(PlannerInfo *root,
101 : Plan *plan,
102 : int gather_param,
103 : Bitmapset *valid_params,
104 : Bitmapset *scan_params);
105 : static bool finalize_primnode(Node *node, finalize_primnode_context *context);
106 : static bool finalize_agg_primnode(Node *node, finalize_primnode_context *context);
107 : static const char *sublinktype_to_string(SubLinkType subLinkType);
108 :
109 :
110 : /*
111 : * Get the datatype/typmod/collation of the first column of the plan's output.
112 : *
113 : * This information is stored for ARRAY_SUBLINK execution and for
114 : * exprType()/exprTypmod()/exprCollation(), which have no way to get at the
115 : * plan associated with a SubPlan node. We really only need the info for
116 : * EXPR_SUBLINK and ARRAY_SUBLINK subplans, but for consistency we save it
117 : * always.
118 : */
119 : static void
120 43604 : get_first_col_type(Plan *plan, Oid *coltype, int32 *coltypmod,
121 : Oid *colcollation)
122 : {
123 : /* In cases such as EXISTS, tlist might be empty; arbitrarily use VOID */
124 43604 : if (plan->targetlist)
125 : {
126 40836 : TargetEntry *tent = linitial_node(TargetEntry, plan->targetlist);
127 :
128 40836 : if (!tent->resjunk)
129 : {
130 40836 : *coltype = exprType((Node *) tent->expr);
131 40836 : *coltypmod = exprTypmod((Node *) tent->expr);
132 40836 : *colcollation = exprCollation((Node *) tent->expr);
133 40836 : return;
134 : }
135 : }
136 2768 : *coltype = VOIDOID;
137 2768 : *coltypmod = -1;
138 2768 : *colcollation = InvalidOid;
139 : }
140 :
141 : /*
142 : * Convert a SubLink (as created by the parser) into a SubPlan.
143 : *
144 : * We are given the SubLink's contained query, type, ID, and testexpr. We are
145 : * also told if this expression appears at top level of a WHERE/HAVING qual.
146 : *
147 : * Note: we assume that the testexpr has been AND/OR flattened (actually,
148 : * it's been through eval_const_expressions), but not converted to
149 : * implicit-AND form; and any SubLinks in it should already have been
150 : * converted to SubPlans. The subquery is as yet untouched, however.
151 : *
152 : * The result is whatever we need to substitute in place of the SubLink node
153 : * in the executable expression. If we're going to do the subplan as a
154 : * regular subplan, this will be the constructed SubPlan node. If we're going
155 : * to do the subplan as an InitPlan, the SubPlan node instead goes into
156 : * root->init_plans, and what we return here is an expression tree
157 : * representing the InitPlan's result: usually just a Param node representing
158 : * a single scalar result, but possibly a row comparison tree containing
159 : * multiple Param nodes, or for a MULTIEXPR subquery a simple NULL constant
160 : * (since the real output Params are elsewhere in the tree, and the MULTIEXPR
161 : * subquery itself is in a resjunk tlist entry whose value is uninteresting).
162 : */
163 : static Node *
164 39012 : make_subplan(PlannerInfo *root, Query *orig_subquery,
165 : SubLinkType subLinkType, int subLinkId,
166 : Node *testexpr, bool isTopQual)
167 : {
168 : Query *subquery;
169 39012 : bool simple_exists = false;
170 : double tuple_fraction;
171 : PlannerInfo *subroot;
172 : RelOptInfo *final_rel;
173 : Path *best_path;
174 : Plan *plan;
175 : List *plan_params;
176 : Node *result;
177 39012 : const char *sublinkstr = sublinktype_to_string(subLinkType);
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 39012 : subquery = copyObject(orig_subquery);
186 :
187 : /*
188 : * If it's an EXISTS subplan, we might be able to simplify it.
189 : */
190 39012 : if (subLinkType == EXISTS_SUBLINK)
191 2508 : 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 hash_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 39012 : if (subLinkType == EXISTS_SUBLINK)
213 2508 : tuple_fraction = 1.0; /* just like a LIMIT 1 */
214 36504 : else if (subLinkType == ALL_SUBLINK ||
215 : subLinkType == ANY_SUBLINK)
216 564 : tuple_fraction = 0.5; /* 50% */
217 : else
218 35940 : 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 39012 : subroot = subquery_planner(root->glob, subquery,
225 : choose_plan_name(root->glob, sublinkstr, true),
226 : root, false, tuple_fraction, NULL);
227 :
228 : /* Isolate the params needed by this specific subplan */
229 39012 : plan_params = root->plan_params;
230 39012 : 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 39012 : final_rel = fetch_upper_rel(subroot, UPPERREL_FINAL, NULL);
237 39012 : best_path = get_cheapest_fractional_path(final_rel, tuple_fraction);
238 :
239 39012 : plan = create_plan(subroot, best_path);
240 :
241 : /* And convert to SubPlan or InitPlan format. */
242 39012 : result = build_subplan(root, plan, best_path,
243 : subroot, plan_params,
244 : subLinkType, subLinkId,
245 : testexpr, NIL, isTopQual);
246 :
247 : /*
248 : * If it's a correlated EXISTS with an unimportant targetlist, we might be
249 : * able to transform it to the equivalent of an IN and then implement it
250 : * by hashing. We don't have enough information yet to tell which way is
251 : * likely to be better (it depends on the expected number of executions of
252 : * the EXISTS qual, and we are much too early in planning the outer query
253 : * to be able to guess that). So we generate both plans, if possible, and
254 : * leave it to setrefs.c to decide which to use.
255 : */
256 39012 : if (simple_exists && IsA(result, SubPlan))
257 : {
258 : Node *newtestexpr;
259 : List *paramIds;
260 :
261 : /* Make a second copy of the original subquery */
262 2174 : subquery = copyObject(orig_subquery);
263 : /* and re-simplify */
264 2174 : simple_exists = simplify_EXISTS_query(root, subquery);
265 : Assert(simple_exists);
266 : /* See if it can be converted to an ANY query */
267 2174 : subquery = convert_EXISTS_to_ANY(root, subquery,
268 : &newtestexpr, ¶mIds);
269 2174 : if (subquery)
270 : {
271 : char *plan_name;
272 :
273 : /* Generate Paths for the ANY subquery; we'll need all rows */
274 1698 : plan_name = choose_plan_name(root->glob, sublinkstr, true);
275 1698 : subroot = subquery_planner(root->glob, subquery, plan_name,
276 : root, false, 0.0, NULL);
277 :
278 : /* Isolate the params needed by this specific subplan */
279 1698 : plan_params = root->plan_params;
280 1698 : root->plan_params = NIL;
281 :
282 : /* Select best Path */
283 1698 : final_rel = fetch_upper_rel(subroot, UPPERREL_FINAL, NULL);
284 1698 : best_path = final_rel->cheapest_total_path;
285 :
286 : /* Now we can check if it'll fit in hash_mem */
287 1698 : if (subpath_is_hashable(best_path, true))
288 : {
289 : SubPlan *hashplan;
290 : AlternativeSubPlan *asplan;
291 :
292 : /* OK, finish planning the ANY subquery */
293 1692 : plan = create_plan(subroot, best_path);
294 :
295 : /* ... and convert to SubPlan format */
296 1692 : hashplan = castNode(SubPlan,
297 : build_subplan(root, plan, best_path,
298 : subroot, plan_params,
299 : ANY_SUBLINK, 0,
300 : newtestexpr,
301 : paramIds,
302 : true));
303 : /* Check we got what we expected */
304 : Assert(hashplan->parParam == NIL);
305 : Assert(hashplan->useHashTable);
306 :
307 : /* Leave it to setrefs.c to decide which plan to use */
308 1692 : asplan = makeNode(AlternativeSubPlan);
309 1692 : asplan->subplans = list_make2(result, hashplan);
310 1692 : result = (Node *) asplan;
311 1692 : root->hasAlternativeSubPlans = true;
312 : }
313 : }
314 : }
315 :
316 39012 : return result;
317 : }
318 :
319 : /*
320 : * Build a SubPlan node given the raw inputs --- subroutine for make_subplan
321 : *
322 : * Returns either the SubPlan, or a replacement expression if we decide to
323 : * make it an InitPlan, as explained in the comments for make_subplan.
324 : */
325 : static Node *
326 40704 : build_subplan(PlannerInfo *root, Plan *plan, Path *path,
327 : PlannerInfo *subroot, List *plan_params,
328 : SubLinkType subLinkType, int subLinkId,
329 : Node *testexpr, List *testexpr_paramids,
330 : bool unknownEqFalse)
331 : {
332 : Node *result;
333 : SubPlan *splan;
334 : ListCell *lc;
335 :
336 : /*
337 : * Initialize the SubPlan node.
338 : *
339 : * Note: plan_id and cost fields are set further down.
340 : */
341 40704 : splan = makeNode(SubPlan);
342 40704 : splan->subLinkType = subLinkType;
343 40704 : splan->plan_name = subroot->plan_name;
344 40704 : splan->testexpr = NULL;
345 40704 : splan->paramIds = NIL;
346 40704 : get_first_col_type(plan, &splan->firstColType, &splan->firstColTypmod,
347 : &splan->firstColCollation);
348 40704 : splan->useHashTable = false;
349 40704 : splan->unknownEqFalse = unknownEqFalse;
350 40704 : splan->parallel_safe = plan->parallel_safe;
351 40704 : splan->setParam = NIL;
352 40704 : splan->parParam = NIL;
353 40704 : splan->args = NIL;
354 :
355 : /*
356 : * Make parParam and args lists of param IDs and expressions that current
357 : * query level will pass to this child plan.
358 : */
359 85266 : foreach(lc, plan_params)
360 : {
361 44562 : PlannerParamItem *pitem = (PlannerParamItem *) lfirst(lc);
362 44562 : Node *arg = pitem->item;
363 :
364 : /*
365 : * The Var, PlaceHolderVar, Aggref, GroupingFunc, or ReturningExpr has
366 : * already been adjusted to have the correct varlevelsup, phlevelsup,
367 : * agglevelsup, or retlevelsup.
368 : *
369 : * If it's a PlaceHolderVar, Aggref, GroupingFunc, or ReturningExpr,
370 : * its arguments might contain SubLinks, which have not yet been
371 : * processed (see the comments for SS_replace_correlation_vars). Do
372 : * that now.
373 : */
374 44562 : if (IsA(arg, PlaceHolderVar) ||
375 44550 : IsA(arg, Aggref) ||
376 44498 : IsA(arg, GroupingFunc) ||
377 44434 : IsA(arg, ReturningExpr))
378 146 : arg = SS_process_sublinks(root, arg, false);
379 :
380 44562 : splan->parParam = lappend_int(splan->parParam, pitem->paramId);
381 44562 : splan->args = lappend(splan->args, arg);
382 : }
383 :
384 : /*
385 : * Un-correlated or undirect correlated plans of EXISTS, EXPR, ARRAY,
386 : * ROWCOMPARE, or MULTIEXPR types can be used as initPlans. For EXISTS,
387 : * EXPR, or ARRAY, we return a Param referring to the result of evaluating
388 : * the initPlan. For ROWCOMPARE, we must modify the testexpr tree to
389 : * contain PARAM_EXEC Params instead of the PARAM_SUBLINK Params emitted
390 : * by the parser, and then return that tree. For MULTIEXPR, we return a
391 : * null constant: the resjunk targetlist item containing the SubLink does
392 : * not need to return anything useful, since the referencing Params are
393 : * elsewhere.
394 : */
395 40704 : if (splan->parParam == NIL && subLinkType == EXISTS_SUBLINK)
396 302 : {
397 : Param *prm;
398 :
399 : Assert(testexpr == NULL);
400 302 : prm = generate_new_exec_param(root, BOOLOID, -1, InvalidOid);
401 302 : splan->setParam = list_make1_int(prm->paramid);
402 302 : splan->isInitPlan = true;
403 302 : result = (Node *) prm;
404 : }
405 40402 : else if (splan->parParam == NIL && subLinkType == EXPR_SUBLINK)
406 9806 : {
407 9806 : TargetEntry *te = linitial(plan->targetlist);
408 : Param *prm;
409 :
410 : Assert(!te->resjunk);
411 : Assert(testexpr == NULL);
412 9806 : prm = generate_new_exec_param(root,
413 9806 : exprType((Node *) te->expr),
414 9806 : exprTypmod((Node *) te->expr),
415 9806 : exprCollation((Node *) te->expr));
416 9806 : splan->setParam = list_make1_int(prm->paramid);
417 9806 : splan->isInitPlan = true;
418 9806 : result = (Node *) prm;
419 : }
420 30596 : else if (splan->parParam == NIL && subLinkType == ARRAY_SUBLINK)
421 122 : {
422 122 : TargetEntry *te = linitial(plan->targetlist);
423 : Oid arraytype;
424 : Param *prm;
425 :
426 : Assert(!te->resjunk);
427 : Assert(testexpr == NULL);
428 122 : arraytype = get_promoted_array_type(exprType((Node *) te->expr));
429 122 : if (!OidIsValid(arraytype))
430 0 : elog(ERROR, "could not find array type for datatype %s",
431 : format_type_be(exprType((Node *) te->expr)));
432 122 : prm = generate_new_exec_param(root,
433 : arraytype,
434 122 : exprTypmod((Node *) te->expr),
435 122 : exprCollation((Node *) te->expr));
436 122 : splan->setParam = list_make1_int(prm->paramid);
437 122 : splan->isInitPlan = true;
438 122 : result = (Node *) prm;
439 : }
440 30474 : else if (splan->parParam == NIL && subLinkType == ROWCOMPARE_SUBLINK)
441 18 : {
442 : /* Adjust the Params */
443 : List *params;
444 :
445 : Assert(testexpr != NULL);
446 18 : params = generate_subquery_params(root,
447 : plan->targetlist,
448 : &splan->paramIds);
449 18 : result = convert_testexpr(root,
450 : testexpr,
451 : params);
452 18 : splan->setParam = list_copy(splan->paramIds);
453 18 : splan->isInitPlan = true;
454 :
455 : /*
456 : * The executable expression is returned to become part of the outer
457 : * plan's expression tree; it is not kept in the initplan node.
458 : */
459 : }
460 30456 : else if (subLinkType == MULTIEXPR_SUBLINK)
461 : {
462 : /*
463 : * Whether it's an initplan or not, it needs to set a PARAM_EXEC Param
464 : * for each output column.
465 : */
466 : List *params;
467 :
468 : Assert(testexpr == NULL);
469 132 : params = generate_subquery_params(root,
470 : plan->targetlist,
471 : &splan->setParam);
472 :
473 : /*
474 : * Save the list of replacement Params in the n'th cell of
475 : * root->multiexpr_params; setrefs.c will use it to replace
476 : * PARAM_MULTIEXPR Params.
477 : */
478 264 : while (list_length(root->multiexpr_params) < subLinkId)
479 132 : root->multiexpr_params = lappend(root->multiexpr_params, NIL);
480 132 : lc = list_nth_cell(root->multiexpr_params, subLinkId - 1);
481 : Assert(lfirst(lc) == NIL);
482 132 : lfirst(lc) = params;
483 :
484 : /* It can be an initplan if there are no parParams. */
485 132 : if (splan->parParam == NIL)
486 : {
487 30 : splan->isInitPlan = true;
488 30 : result = (Node *) makeNullConst(RECORDOID, -1, InvalidOid);
489 : }
490 : else
491 : {
492 102 : splan->isInitPlan = false;
493 102 : result = (Node *) splan;
494 : }
495 : }
496 : else
497 : {
498 : /*
499 : * Adjust the Params in the testexpr, unless caller already took care
500 : * of it (as indicated by passing a list of Param IDs).
501 : */
502 30324 : if (testexpr && testexpr_paramids == NIL)
503 576 : {
504 : List *params;
505 :
506 576 : params = generate_subquery_params(root,
507 : plan->targetlist,
508 : &splan->paramIds);
509 576 : splan->testexpr = convert_testexpr(root,
510 : testexpr,
511 : params);
512 : }
513 : else
514 : {
515 29748 : splan->testexpr = testexpr;
516 29748 : splan->paramIds = testexpr_paramids;
517 : }
518 :
519 : /*
520 : * We can't convert subplans of ALL_SUBLINK or ANY_SUBLINK types to
521 : * initPlans, even when they are uncorrelated or undirect correlated,
522 : * because we need to scan the output of the subplan for each outer
523 : * tuple. But if it's a not-direct-correlated IN (= ANY) test, we
524 : * might be able to use a hashtable to avoid comparing all the tuples.
525 : */
526 30324 : if (subLinkType == ANY_SUBLINK &&
527 4372 : splan->parParam == NIL &&
528 4268 : subplan_is_hashable(plan, unknownEqFalse) &&
529 2134 : testexpr_is_hashable(splan->testexpr, splan->paramIds))
530 2110 : splan->useHashTable = true;
531 :
532 : /*
533 : * Otherwise, we have the option to tack a Material node onto the top
534 : * of the subplan, to reduce the cost of reading it repeatedly. This
535 : * is pointless for a direct-correlated subplan, since we'd have to
536 : * recompute its results each time anyway. For uncorrelated/undirect
537 : * correlated subplans, we add Material unless the subplan's top plan
538 : * node would materialize its output anyway. Also, if enable_material
539 : * is false, then the user does not want us to materialize anything
540 : * unnecessarily, so we don't.
541 : */
542 28214 : else if (splan->parParam == NIL && enable_material &&
543 42 : !ExecMaterializesOutput(nodeTag(plan)))
544 42 : plan = materialize_finished_plan(plan);
545 :
546 30324 : result = (Node *) splan;
547 30324 : splan->isInitPlan = false;
548 : }
549 :
550 : /*
551 : * Add the subplan, its path, and its PlannerInfo to the global lists.
552 : */
553 40704 : root->glob->subplans = lappend(root->glob->subplans, plan);
554 40704 : root->glob->subpaths = lappend(root->glob->subpaths, path);
555 40704 : root->glob->subroots = lappend(root->glob->subroots, subroot);
556 40704 : splan->plan_id = list_length(root->glob->subplans);
557 :
558 40704 : if (splan->isInitPlan)
559 10278 : root->init_plans = lappend(root->init_plans, splan);
560 :
561 : /*
562 : * A parameterless subplan (not initplan) should be prepared to handle
563 : * REWIND efficiently. If it has direct parameters then there's no point
564 : * since it'll be reset on each scan anyway; and if it's an initplan then
565 : * there's no point since it won't get re-run without parameter changes
566 : * anyway. The input of a hashed subplan doesn't need REWIND either.
567 : */
568 40704 : if (splan->parParam == NIL && !splan->isInitPlan && !splan->useHashTable)
569 42 : root->glob->rewindPlanIDs = bms_add_member(root->glob->rewindPlanIDs,
570 : splan->plan_id);
571 :
572 : /* Lastly, fill in the cost estimates for use later */
573 40704 : cost_subplan(root, splan, plan);
574 :
575 40704 : return result;
576 : }
577 :
578 : /*
579 : * generate_subquery_params: build a list of Params representing the output
580 : * columns of a sublink's sub-select, given the sub-select's targetlist.
581 : *
582 : * We also return an integer list of the paramids of the Params.
583 : */
584 : static List *
585 726 : generate_subquery_params(PlannerInfo *root, List *tlist, List **paramIds)
586 : {
587 : List *result;
588 : List *ids;
589 : ListCell *lc;
590 :
591 726 : result = ids = NIL;
592 1700 : foreach(lc, tlist)
593 : {
594 974 : TargetEntry *tent = (TargetEntry *) lfirst(lc);
595 : Param *param;
596 :
597 974 : if (tent->resjunk)
598 6 : continue;
599 :
600 968 : param = generate_new_exec_param(root,
601 968 : exprType((Node *) tent->expr),
602 968 : exprTypmod((Node *) tent->expr),
603 968 : exprCollation((Node *) tent->expr));
604 968 : result = lappend(result, param);
605 968 : ids = lappend_int(ids, param->paramid);
606 : }
607 :
608 726 : *paramIds = ids;
609 726 : return result;
610 : }
611 :
612 : /*
613 : * generate_subquery_vars: build a list of Vars representing the output
614 : * columns of a sublink's sub-select, given the sub-select's targetlist.
615 : * The Vars have the specified varno (RTE index).
616 : */
617 : static List *
618 4476 : generate_subquery_vars(PlannerInfo *root, List *tlist, Index varno)
619 : {
620 : List *result;
621 : ListCell *lc;
622 :
623 4476 : result = NIL;
624 9030 : foreach(lc, tlist)
625 : {
626 4554 : TargetEntry *tent = (TargetEntry *) lfirst(lc);
627 : Var *var;
628 :
629 4554 : if (tent->resjunk)
630 0 : continue;
631 :
632 4554 : var = makeVarFromTargetEntry(varno, tent);
633 4554 : result = lappend(result, var);
634 : }
635 :
636 4476 : return result;
637 : }
638 :
639 : /*
640 : * convert_testexpr: convert the testexpr given by the parser into
641 : * actually executable form. This entails replacing PARAM_SUBLINK Params
642 : * with Params or Vars representing the results of the sub-select. The
643 : * nodes to be substituted are passed in as the List result from
644 : * generate_subquery_params or generate_subquery_vars.
645 : */
646 : static Node *
647 5346 : convert_testexpr(PlannerInfo *root,
648 : Node *testexpr,
649 : List *subst_nodes)
650 : {
651 : convert_testexpr_context context;
652 :
653 5346 : context.root = root;
654 5346 : context.subst_nodes = subst_nodes;
655 5346 : return convert_testexpr_mutator(testexpr, &context);
656 : }
657 :
658 : static Node *
659 25572 : convert_testexpr_mutator(Node *node,
660 : convert_testexpr_context *context)
661 : {
662 25572 : if (node == NULL)
663 94 : return NULL;
664 25478 : if (IsA(node, Param))
665 : {
666 5554 : Param *param = (Param *) node;
667 :
668 5554 : if (param->paramkind == PARAM_SUBLINK)
669 : {
670 11096 : if (param->paramid <= 0 ||
671 5548 : param->paramid > list_length(context->subst_nodes))
672 0 : elog(ERROR, "unexpected PARAM_SUBLINK ID: %d", param->paramid);
673 :
674 : /*
675 : * We copy the list item to avoid having doubly-linked
676 : * substructure in the modified parse tree. This is probably
677 : * unnecessary when it's a Param, but be safe.
678 : */
679 5548 : return (Node *) copyObject(list_nth(context->subst_nodes,
680 : param->paramid - 1));
681 : }
682 : }
683 19930 : if (IsA(node, SubLink))
684 : {
685 : /*
686 : * If we come across a nested SubLink, it is neither necessary nor
687 : * correct to recurse into it: any PARAM_SUBLINKs we might find inside
688 : * belong to the inner SubLink not the outer. So just return it as-is.
689 : *
690 : * This reasoning depends on the assumption that nothing will pull
691 : * subexpressions into or out of the testexpr field of a SubLink, at
692 : * least not without replacing PARAM_SUBLINKs first. If we did want
693 : * to do that we'd need to rethink the parser-output representation
694 : * altogether, since currently PARAM_SUBLINKs are only unique per
695 : * SubLink not globally across the query. The whole point of
696 : * replacing them with Vars or PARAM_EXEC nodes is to make them
697 : * globally unique before they escape from the SubLink's testexpr.
698 : *
699 : * Note: this can't happen when called during SS_process_sublinks,
700 : * because that recursively processes inner SubLinks first. It can
701 : * happen when called from convert_ANY_sublink_to_join, though.
702 : */
703 12 : return node;
704 : }
705 19918 : return expression_tree_mutator(node, convert_testexpr_mutator, context);
706 : }
707 :
708 : /*
709 : * subplan_is_hashable: can we implement an ANY subplan by hashing?
710 : *
711 : * This is not responsible for checking whether the combining testexpr
712 : * is suitable for hashing. We only look at the subquery itself.
713 : */
714 : static bool
715 2134 : subplan_is_hashable(Plan *plan, bool unknownEqFalse)
716 : {
717 : Size hashtablesize;
718 :
719 : /*
720 : * The estimated size of the hashtable holding the subquery result must
721 : * fit in hash_mem. (Note: reject on equality, to ensure that an estimate
722 : * of SIZE_MAX disables hashing regardless of the hash_mem limit.)
723 : */
724 2134 : hashtablesize = EstimateSubplanHashTableSpace(plan->plan_rows,
725 2134 : plan->plan_width,
726 : unknownEqFalse);
727 2134 : if (hashtablesize >= get_hash_memory_limit())
728 0 : return false;
729 :
730 2134 : return true;
731 : }
732 :
733 : /*
734 : * subpath_is_hashable: can we implement an ANY subplan by hashing?
735 : *
736 : * Identical to subplan_is_hashable, but work from a Path for the subplan.
737 : */
738 : static bool
739 1698 : subpath_is_hashable(Path *path, bool unknownEqFalse)
740 : {
741 : Size hashtablesize;
742 :
743 : /*
744 : * The estimated size of the hashtable holding the subquery result must
745 : * fit in hash_mem. (Note: reject on equality, to ensure that an estimate
746 : * of SIZE_MAX disables hashing regardless of the hash_mem limit.)
747 : */
748 1698 : hashtablesize = EstimateSubplanHashTableSpace(path->rows,
749 1698 : path->pathtarget->width,
750 : unknownEqFalse);
751 1698 : if (hashtablesize >= get_hash_memory_limit())
752 6 : return false;
753 :
754 1692 : return true;
755 : }
756 :
757 : /*
758 : * testexpr_is_hashable: is an ANY SubLink's test expression hashable?
759 : *
760 : * To identify LHS vs RHS of the hash expression, we must be given the
761 : * list of output Param IDs of the SubLink's subquery.
762 : */
763 : static bool
764 2134 : testexpr_is_hashable(Node *testexpr, List *param_ids)
765 : {
766 : /*
767 : * The testexpr must be a single OpExpr, or an AND-clause containing only
768 : * OpExprs, each of which satisfy test_opexpr_is_hashable().
769 : */
770 2134 : if (testexpr && IsA(testexpr, OpExpr))
771 : {
772 1268 : if (test_opexpr_is_hashable((OpExpr *) testexpr, param_ids))
773 1244 : return true;
774 : }
775 866 : else if (is_andclause(testexpr))
776 : {
777 : ListCell *l;
778 :
779 2598 : foreach(l, ((BoolExpr *) testexpr)->args)
780 : {
781 1732 : Node *andarg = (Node *) lfirst(l);
782 :
783 1732 : if (!IsA(andarg, OpExpr))
784 0 : return false;
785 1732 : if (!test_opexpr_is_hashable((OpExpr *) andarg, param_ids))
786 0 : return false;
787 : }
788 866 : return true;
789 : }
790 :
791 24 : return false;
792 : }
793 :
794 : static bool
795 3000 : test_opexpr_is_hashable(OpExpr *testexpr, List *param_ids)
796 : {
797 : /*
798 : * The combining operator must be hashable and strict. The need for
799 : * hashability is obvious, since we want to use hashing. Without
800 : * strictness, behavior in the presence of nulls is too unpredictable. We
801 : * actually must assume even more than plain strictness: it can't yield
802 : * NULL for non-null inputs, either (see nodeSubplan.c). However, hash
803 : * indexes and hash joins assume that too.
804 : */
805 3000 : if (!hash_ok_operator(testexpr))
806 12 : return false;
807 :
808 : /*
809 : * The left and right inputs must belong to the outer and inner queries
810 : * respectively; hence Params that will be supplied by the subquery must
811 : * not appear in the LHS, and Vars of the outer query must not appear in
812 : * the RHS. (Ordinarily, this must be true because of the way that the
813 : * parser builds an ANY SubLink's testexpr ... but inlining of functions
814 : * could have changed the expression's structure, so we have to check.
815 : * Such cases do not occur often enough to be worth trying to optimize, so
816 : * we don't worry about trying to commute the clause or anything like
817 : * that; we just need to be sure not to build an invalid plan.)
818 : */
819 2988 : if (list_length(testexpr->args) != 2)
820 0 : return false;
821 2988 : if (contain_exec_param((Node *) linitial(testexpr->args), param_ids))
822 12 : return false;
823 2976 : if (contain_var_clause((Node *) lsecond(testexpr->args)))
824 0 : return false;
825 2976 : return true;
826 : }
827 :
828 : /*
829 : * Check expression is hashable + strict
830 : *
831 : * We could use op_hashjoinable() and op_strict(), but do it like this to
832 : * avoid a redundant cache lookup.
833 : */
834 : static bool
835 9262 : hash_ok_operator(OpExpr *expr)
836 : {
837 9262 : Oid opid = expr->opno;
838 :
839 : /* quick out if not a binary operator */
840 9262 : if (list_length(expr->args) != 2)
841 0 : return false;
842 9262 : if (opid == ARRAY_EQ_OP ||
843 : opid == RECORD_EQ_OP)
844 : {
845 : /* these are strict, but must check input type to ensure hashable */
846 12 : Node *leftarg = linitial(expr->args);
847 :
848 12 : return op_hashjoinable(opid, exprType(leftarg));
849 : }
850 : else
851 : {
852 : /* else must look up the operator properties */
853 : HeapTuple tup;
854 : Form_pg_operator optup;
855 :
856 9250 : tup = SearchSysCache1(OPEROID, ObjectIdGetDatum(opid));
857 9250 : if (!HeapTupleIsValid(tup))
858 0 : elog(ERROR, "cache lookup failed for operator %u", opid);
859 9250 : optup = (Form_pg_operator) GETSTRUCT(tup);
860 9250 : if (!optup->oprcanhash || !func_strict(optup->oprcode))
861 : {
862 846 : ReleaseSysCache(tup);
863 846 : return false;
864 : }
865 8404 : ReleaseSysCache(tup);
866 8404 : return true;
867 : }
868 : }
869 :
870 :
871 : /*
872 : * SS_process_ctes: process a query's WITH list
873 : *
874 : * Consider each CTE in the WITH list and either ignore it (if it's an
875 : * unreferenced SELECT), "inline" it to create a regular sub-SELECT-in-FROM,
876 : * or convert it to an initplan.
877 : *
878 : * A side effect is to fill in root->cte_plan_ids with a list that
879 : * parallels root->parse->cteList and provides the subplan ID for
880 : * each CTE's initplan, or a dummy ID (-1) if we didn't make an initplan.
881 : */
882 : void
883 2886 : SS_process_ctes(PlannerInfo *root)
884 : {
885 : ListCell *lc;
886 :
887 : Assert(root->cte_plan_ids == NIL);
888 :
889 6962 : foreach(lc, root->parse->cteList)
890 : {
891 4082 : CommonTableExpr *cte = (CommonTableExpr *) lfirst(lc);
892 4082 : CmdType cmdType = ((Query *) cte->ctequery)->commandType;
893 : Query *subquery;
894 : PlannerInfo *subroot;
895 : RelOptInfo *final_rel;
896 : Path *best_path;
897 : Plan *plan;
898 : SubPlan *splan;
899 : int paramid;
900 :
901 : /*
902 : * Ignore SELECT CTEs that are not actually referenced anywhere.
903 : */
904 4082 : if (cte->cterefcount == 0 && cmdType == CMD_SELECT)
905 : {
906 : /* Make a dummy entry in cte_plan_ids */
907 52 : root->cte_plan_ids = lappend_int(root->cte_plan_ids, -1);
908 1576 : continue;
909 : }
910 :
911 : /*
912 : * Consider inlining the CTE (creating RTE_SUBQUERY RTE(s)) instead of
913 : * implementing it as a separately-planned CTE.
914 : *
915 : * We cannot inline if any of these conditions hold:
916 : *
917 : * 1. The user said not to (the CTEMaterializeAlways option).
918 : *
919 : * 2. The CTE is recursive.
920 : *
921 : * 3. The CTE has side-effects; this includes either not being a plain
922 : * SELECT, or containing volatile functions. Inlining might change
923 : * the side-effects, which would be bad.
924 : *
925 : * 4. The CTE is multiply-referenced and contains a self-reference to
926 : * a recursive CTE outside itself. Inlining would result in multiple
927 : * recursive self-references, which we don't support.
928 : *
929 : * Otherwise, we have an option whether to inline or not. That should
930 : * always be a win if there's just a single reference, but if the CTE
931 : * is multiply-referenced then it's unclear: inlining adds duplicate
932 : * computations, but the ability to absorb restrictions from the outer
933 : * query level could outweigh that. We do not have nearly enough
934 : * information at this point to tell whether that's true, so we let
935 : * the user express a preference. Our default behavior is to inline
936 : * only singly-referenced CTEs, but a CTE marked CTEMaterializeNever
937 : * will be inlined even if multiply referenced.
938 : *
939 : * Note: we check for volatile functions last, because that's more
940 : * expensive than the other tests needed.
941 : */
942 4030 : if ((cte->ctematerialized == CTEMaterializeNever ||
943 3982 : (cte->ctematerialized == CTEMaterializeDefault &&
944 3780 : cte->cterefcount == 1)) &&
945 2754 : !cte->cterecursive &&
946 1616 : cmdType == CMD_SELECT &&
947 1616 : !contain_dml(cte->ctequery) &&
948 1608 : (cte->cterefcount <= 1 ||
949 36 : !contain_outer_selfref(cte->ctequery)) &&
950 1596 : !contain_volatile_functions(cte->ctequery))
951 : {
952 1524 : inline_cte(root, cte);
953 : /* Make a dummy entry in cte_plan_ids */
954 1524 : root->cte_plan_ids = lappend_int(root->cte_plan_ids, -1);
955 1524 : continue;
956 : }
957 :
958 : /*
959 : * Copy the source Query node. Probably not necessary, but let's keep
960 : * this similar to make_subplan.
961 : */
962 2506 : subquery = (Query *) copyObject(cte->ctequery);
963 :
964 : /* plan_params should not be in use in current query level */
965 : Assert(root->plan_params == NIL);
966 :
967 : /*
968 : * Generate Paths for the CTE query. Always plan for full retrieval
969 : * --- we don't have enough info to predict otherwise.
970 : */
971 2506 : subroot = subquery_planner(root->glob, subquery,
972 2506 : choose_plan_name(root->glob, cte->ctename, false),
973 2506 : root, cte->cterecursive, 0.0, NULL);
974 :
975 : /*
976 : * Since the current query level doesn't yet contain any RTEs, it
977 : * should not be possible for the CTE to have requested parameters of
978 : * this level.
979 : */
980 2500 : if (root->plan_params)
981 0 : elog(ERROR, "unexpected outer reference in CTE query");
982 :
983 : /*
984 : * Select best Path and turn it into a Plan. At least for now, there
985 : * seems no reason to postpone doing that.
986 : */
987 2500 : final_rel = fetch_upper_rel(subroot, UPPERREL_FINAL, NULL);
988 2500 : best_path = final_rel->cheapest_total_path;
989 :
990 2500 : plan = create_plan(subroot, best_path);
991 :
992 : /*
993 : * Make a SubPlan node for it. This is just enough unlike
994 : * build_subplan that we can't share code.
995 : *
996 : * Note: plan_id and cost fields are set further down.
997 : */
998 2500 : splan = makeNode(SubPlan);
999 2500 : splan->subLinkType = CTE_SUBLINK;
1000 2500 : splan->plan_name = subroot->plan_name;
1001 2500 : splan->testexpr = NULL;
1002 2500 : splan->paramIds = NIL;
1003 2500 : get_first_col_type(plan, &splan->firstColType, &splan->firstColTypmod,
1004 : &splan->firstColCollation);
1005 2500 : splan->useHashTable = false;
1006 2500 : splan->unknownEqFalse = false;
1007 :
1008 : /*
1009 : * CTE scans are not considered for parallelism (cf
1010 : * set_rel_consider_parallel).
1011 : */
1012 2500 : splan->parallel_safe = false;
1013 2500 : splan->setParam = NIL;
1014 2500 : splan->parParam = NIL;
1015 2500 : splan->args = NIL;
1016 :
1017 : /*
1018 : * The node can't have any inputs (since it's an initplan), so the
1019 : * parParam and args lists remain empty. (It could contain references
1020 : * to earlier CTEs' output param IDs, but CTE outputs are not
1021 : * propagated via the args list.)
1022 : */
1023 :
1024 : /*
1025 : * Assign a param ID to represent the CTE's output. No ordinary
1026 : * "evaluation" of this param slot ever happens, but we use the param
1027 : * ID for setParam/chgParam signaling just as if the CTE plan were
1028 : * returning a simple scalar output. (Also, the executor abuses the
1029 : * ParamExecData slot for this param ID for communication among
1030 : * multiple CteScan nodes that might be scanning this CTE.)
1031 : */
1032 2500 : paramid = assign_special_exec_param(root);
1033 2500 : splan->setParam = list_make1_int(paramid);
1034 :
1035 : /*
1036 : * Add the subplan, its path, and its PlannerInfo to the global lists.
1037 : */
1038 2500 : root->glob->subplans = lappend(root->glob->subplans, plan);
1039 2500 : root->glob->subpaths = lappend(root->glob->subpaths, best_path);
1040 2500 : root->glob->subroots = lappend(root->glob->subroots, subroot);
1041 2500 : splan->plan_id = list_length(root->glob->subplans);
1042 :
1043 2500 : root->init_plans = lappend(root->init_plans, splan);
1044 :
1045 2500 : root->cte_plan_ids = lappend_int(root->cte_plan_ids, splan->plan_id);
1046 :
1047 : /* Lastly, fill in the cost estimates for use later */
1048 2500 : cost_subplan(root, splan, plan);
1049 : }
1050 2880 : }
1051 :
1052 : /*
1053 : * contain_dml: is any subquery not a plain SELECT?
1054 : *
1055 : * We reject SELECT FOR UPDATE/SHARE as well as INSERT etc.
1056 : */
1057 : static bool
1058 1616 : contain_dml(Node *node)
1059 : {
1060 1616 : return contain_dml_walker(node, NULL);
1061 : }
1062 :
1063 : static bool
1064 108760 : contain_dml_walker(Node *node, void *context)
1065 : {
1066 108760 : if (node == NULL)
1067 37640 : return false;
1068 71120 : if (IsA(node, Query))
1069 : {
1070 2966 : Query *query = (Query *) node;
1071 :
1072 2966 : if (query->commandType != CMD_SELECT ||
1073 2966 : query->rowMarks != NIL)
1074 8 : return true;
1075 :
1076 2958 : return query_tree_walker(query, contain_dml_walker, context, 0);
1077 : }
1078 68154 : return expression_tree_walker(node, contain_dml_walker, context);
1079 : }
1080 :
1081 : /*
1082 : * contain_outer_selfref: is there an external recursive self-reference?
1083 : */
1084 : static bool
1085 36 : contain_outer_selfref(Node *node)
1086 : {
1087 36 : Index depth = 0;
1088 :
1089 : /*
1090 : * We should be starting with a Query, so that depth will be 1 while
1091 : * examining its immediate contents.
1092 : */
1093 : Assert(IsA(node, Query));
1094 :
1095 36 : return contain_outer_selfref_walker(node, &depth);
1096 : }
1097 :
1098 : static bool
1099 810 : contain_outer_selfref_walker(Node *node, Index *depth)
1100 : {
1101 810 : if (node == NULL)
1102 486 : return false;
1103 324 : if (IsA(node, RangeTblEntry))
1104 : {
1105 30 : RangeTblEntry *rte = (RangeTblEntry *) node;
1106 :
1107 : /*
1108 : * Check for a self-reference to a CTE that's above the Query that our
1109 : * search started at.
1110 : */
1111 30 : if (rte->rtekind == RTE_CTE &&
1112 12 : rte->self_reference &&
1113 12 : rte->ctelevelsup >= *depth)
1114 12 : return true;
1115 18 : return false; /* allow range_table_walker to continue */
1116 : }
1117 294 : if (IsA(node, Query))
1118 : {
1119 : /* Recurse into subquery, tracking nesting depth properly */
1120 42 : Query *query = (Query *) node;
1121 : bool result;
1122 :
1123 42 : (*depth)++;
1124 :
1125 42 : result = query_tree_walker(query, contain_outer_selfref_walker,
1126 : depth, QTW_EXAMINE_RTES_BEFORE);
1127 :
1128 42 : (*depth)--;
1129 :
1130 42 : return result;
1131 : }
1132 252 : return expression_tree_walker(node, contain_outer_selfref_walker, depth);
1133 : }
1134 :
1135 : /*
1136 : * inline_cte: convert RTE_CTE references to given CTE into RTE_SUBQUERYs
1137 : */
1138 : static void
1139 1524 : inline_cte(PlannerInfo *root, CommonTableExpr *cte)
1140 : {
1141 : struct inline_cte_walker_context context;
1142 :
1143 1524 : context.ctename = cte->ctename;
1144 : /* Start at levelsup = -1 because we'll immediately increment it */
1145 1524 : context.levelsup = -1;
1146 1524 : context.ctequery = castNode(Query, cte->ctequery);
1147 :
1148 1524 : (void) inline_cte_walker((Node *) root->parse, &context);
1149 1524 : }
1150 :
1151 : static bool
1152 568776 : inline_cte_walker(Node *node, inline_cte_walker_context *context)
1153 : {
1154 568776 : if (node == NULL)
1155 151846 : return false;
1156 416930 : if (IsA(node, Query))
1157 : {
1158 11116 : Query *query = (Query *) node;
1159 :
1160 11116 : context->levelsup++;
1161 :
1162 : /*
1163 : * Visit the query's RTE nodes after their contents; otherwise
1164 : * query_tree_walker would descend into the newly inlined CTE query,
1165 : * which we don't want.
1166 : */
1167 11116 : (void) query_tree_walker(query, inline_cte_walker, context,
1168 : QTW_EXAMINE_RTES_AFTER);
1169 :
1170 11116 : context->levelsup--;
1171 :
1172 11116 : return false;
1173 : }
1174 405814 : else if (IsA(node, RangeTblEntry))
1175 : {
1176 20966 : RangeTblEntry *rte = (RangeTblEntry *) node;
1177 :
1178 20966 : if (rte->rtekind == RTE_CTE &&
1179 6160 : strcmp(rte->ctename, context->ctename) == 0 &&
1180 1554 : rte->ctelevelsup == context->levelsup)
1181 : {
1182 : /*
1183 : * Found a reference to replace. Generate a copy of the CTE query
1184 : * with appropriate level adjustment for outer references (e.g.,
1185 : * to other CTEs).
1186 : */
1187 1548 : Query *newquery = copyObject(context->ctequery);
1188 :
1189 1548 : if (context->levelsup > 0)
1190 984 : IncrementVarSublevelsUp((Node *) newquery, context->levelsup, 1);
1191 :
1192 : /*
1193 : * Convert the RTE_CTE RTE into a RTE_SUBQUERY.
1194 : *
1195 : * Historically, a FOR UPDATE clause has been treated as extending
1196 : * into views and subqueries, but not into CTEs. We preserve this
1197 : * distinction by not trying to push rowmarks into the new
1198 : * subquery.
1199 : */
1200 1548 : rte->rtekind = RTE_SUBQUERY;
1201 1548 : rte->subquery = newquery;
1202 1548 : rte->security_barrier = false;
1203 :
1204 : /* Zero out CTE-specific fields */
1205 1548 : rte->ctename = NULL;
1206 1548 : rte->ctelevelsup = 0;
1207 1548 : rte->self_reference = false;
1208 1548 : rte->coltypes = NIL;
1209 1548 : rte->coltypmods = NIL;
1210 1548 : rte->colcollations = NIL;
1211 : }
1212 :
1213 20966 : return false;
1214 : }
1215 :
1216 384848 : return expression_tree_walker(node, inline_cte_walker, context);
1217 : }
1218 :
1219 : /*
1220 : * Attempt to transform 'testexpr' over the VALUES subquery into
1221 : * a ScalarArrayOpExpr. We currently support the transformation only when
1222 : * it ends up with a constant array. Otherwise, the evaluation of non-hashed
1223 : * SAOP might be slower than the corresponding Hash Join with VALUES.
1224 : *
1225 : * Return transformed ScalarArrayOpExpr or NULL if transformation isn't
1226 : * allowed.
1227 : */
1228 : ScalarArrayOpExpr *
1229 4666 : convert_VALUES_to_ANY(PlannerInfo *root, Node *testexpr, Query *values)
1230 : {
1231 : RangeTblEntry *rte;
1232 : Node *leftop;
1233 : Node *rightop;
1234 : Oid opno;
1235 : ListCell *lc;
1236 : Oid inputcollid;
1237 4666 : List *exprs = NIL;
1238 :
1239 : /*
1240 : * Check we have a binary operator over a single-column subquery with no
1241 : * joins and no LIMIT/OFFSET/ORDER BY clauses.
1242 : */
1243 9226 : if (!IsA(testexpr, OpExpr) ||
1244 9120 : list_length(((OpExpr *) testexpr)->args) != 2 ||
1245 4560 : list_length(values->targetList) > 1 ||
1246 4560 : values->limitCount != NULL ||
1247 4548 : values->limitOffset != NULL ||
1248 9042 : values->sortClause != NIL ||
1249 4518 : list_length(values->rtable) != 1)
1250 3754 : return NULL;
1251 :
1252 912 : rte = linitial_node(RangeTblEntry, values->rtable);
1253 912 : leftop = linitial(((OpExpr *) testexpr)->args);
1254 912 : rightop = lsecond(((OpExpr *) testexpr)->args);
1255 912 : opno = ((OpExpr *) testexpr)->opno;
1256 912 : inputcollid = ((OpExpr *) testexpr)->inputcollid;
1257 :
1258 : /*
1259 : * Also, check that only RTE corresponds to VALUES; the list of values has
1260 : * at least two items and no volatile functions.
1261 : */
1262 1044 : if (rte->rtekind != RTE_VALUES ||
1263 252 : list_length(rte->values_lists) < 2 ||
1264 120 : contain_volatile_functions((Node *) rte->values_lists))
1265 792 : return NULL;
1266 :
1267 360 : foreach(lc, rte->values_lists)
1268 : {
1269 276 : List *elem = lfirst(lc);
1270 276 : Node *value = linitial(elem);
1271 :
1272 : /*
1273 : * Prepare an evaluation of the right side of the operator with
1274 : * substitution of the given value.
1275 : */
1276 276 : value = convert_testexpr(root, rightop, list_make1(value));
1277 :
1278 : /*
1279 : * Try to evaluate constant expressions. We could get Const as a
1280 : * result.
1281 : */
1282 276 : value = eval_const_expressions(root, value);
1283 :
1284 : /*
1285 : * As we only support constant output arrays, all the items must also
1286 : * be constant.
1287 : */
1288 276 : if (!IsA(value, Const))
1289 36 : return NULL;
1290 :
1291 240 : exprs = lappend(exprs, value);
1292 : }
1293 :
1294 : /* Finally, build ScalarArrayOpExpr at the top of the 'exprs' list. */
1295 84 : return make_SAOP_expr(opno, leftop, exprType(rightop),
1296 84 : linitial_oid(rte->colcollations), inputcollid,
1297 : exprs, false);
1298 : }
1299 :
1300 : /*
1301 : * convert_ANY_sublink_to_join: try to convert an ANY SubLink to a join
1302 : *
1303 : * The caller has found an ANY SubLink at the top level of one of the query's
1304 : * qual clauses, but has not checked the properties of the SubLink further.
1305 : * Decide whether it is appropriate to process this SubLink in join style.
1306 : * If so, form a JoinExpr and return it. Return NULL if the SubLink cannot
1307 : * be converted to a join.
1308 : *
1309 : * The only non-obvious input parameter is available_rels: this is the set
1310 : * of query rels that can safely be referenced in the sublink expression.
1311 : * (We must restrict this to avoid changing the semantics when a sublink
1312 : * is present in an outer join's ON qual.) The conversion must fail if
1313 : * the converted qual would reference any but these parent-query relids.
1314 : *
1315 : * On success, the returned JoinExpr has larg = NULL and rarg = the jointree
1316 : * item representing the pulled-up subquery. The caller must set larg to
1317 : * represent the relation(s) on the lefthand side of the new join, and insert
1318 : * the JoinExpr into the upper query's jointree at an appropriate place
1319 : * (typically, where the lefthand relation(s) had been). Note that the
1320 : * passed-in SubLink must also be removed from its original position in the
1321 : * query quals, since the quals of the returned JoinExpr replace it.
1322 : * (Notionally, we replace the SubLink with a constant TRUE, then elide the
1323 : * redundant constant from the qual.)
1324 : *
1325 : * On success, the caller is also responsible for recursively applying
1326 : * pull_up_sublinks processing to the rarg and quals of the returned JoinExpr.
1327 : * (On failure, there is no need to do anything, since pull_up_sublinks will
1328 : * be applied when we recursively plan the sub-select.)
1329 : *
1330 : * Side effects of a successful conversion include adding the SubLink's
1331 : * subselect to the query's rangetable, so that it can be referenced in
1332 : * the JoinExpr's rarg.
1333 : */
1334 : JoinExpr *
1335 4588 : convert_ANY_sublink_to_join(PlannerInfo *root, SubLink *sublink,
1336 : Relids available_rels)
1337 : {
1338 : JoinExpr *result;
1339 4588 : Query *parse = root->parse;
1340 4588 : Query *subselect = (Query *) sublink->subselect;
1341 : Relids upper_varnos;
1342 : int rtindex;
1343 : ParseNamespaceItem *nsitem;
1344 : RangeTblEntry *rte;
1345 : RangeTblRef *rtr;
1346 : List *subquery_vars;
1347 : Node *quals;
1348 : ParseState *pstate;
1349 : Relids sub_ref_outer_relids;
1350 : bool use_lateral;
1351 :
1352 : Assert(sublink->subLinkType == ANY_SUBLINK);
1353 :
1354 : /*
1355 : * If the sub-select contains any Vars of the parent query, we treat it as
1356 : * LATERAL. (Vars from higher levels don't matter here.)
1357 : */
1358 4588 : sub_ref_outer_relids = pull_varnos_of_level(NULL, (Node *) subselect, 1);
1359 4588 : use_lateral = !bms_is_empty(sub_ref_outer_relids);
1360 :
1361 : /*
1362 : * Can't convert if the sub-select contains parent-level Vars of relations
1363 : * not in available_rels.
1364 : */
1365 4588 : if (!bms_is_subset(sub_ref_outer_relids, available_rels))
1366 12 : return NULL;
1367 :
1368 : /*
1369 : * The test expression must contain some Vars of the parent query, else
1370 : * it's not gonna be a join. (Note that it won't have Vars referring to
1371 : * the subquery, rather Params.)
1372 : */
1373 4576 : upper_varnos = pull_varnos(root, sublink->testexpr);
1374 4576 : if (bms_is_empty(upper_varnos))
1375 12 : return NULL;
1376 :
1377 : /*
1378 : * However, it can't refer to anything outside available_rels.
1379 : */
1380 4564 : if (!bms_is_subset(upper_varnos, available_rels))
1381 24 : return NULL;
1382 :
1383 : /*
1384 : * The combining operators and left-hand expressions mustn't be volatile.
1385 : */
1386 4540 : if (contain_volatile_functions(sublink->testexpr))
1387 64 : return NULL;
1388 :
1389 : /* Create a dummy ParseState for addRangeTableEntryForSubquery */
1390 4476 : pstate = make_parsestate(NULL);
1391 :
1392 : /*
1393 : * Okay, pull up the sub-select into upper range table.
1394 : *
1395 : * We rely here on the assumption that the outer query has no references
1396 : * to the inner (necessarily true, other than the Vars that we build
1397 : * below). Therefore this is a lot easier than what pull_up_subqueries has
1398 : * to go through.
1399 : */
1400 4476 : nsitem = addRangeTableEntryForSubquery(pstate,
1401 : subselect,
1402 : NULL,
1403 : use_lateral,
1404 : false);
1405 4476 : rte = nsitem->p_rte;
1406 4476 : parse->rtable = lappend(parse->rtable, rte);
1407 4476 : rtindex = list_length(parse->rtable);
1408 :
1409 : /*
1410 : * Form a RangeTblRef for the pulled-up sub-select.
1411 : */
1412 4476 : rtr = makeNode(RangeTblRef);
1413 4476 : rtr->rtindex = rtindex;
1414 :
1415 : /*
1416 : * Build a list of Vars representing the subselect outputs.
1417 : */
1418 4476 : subquery_vars = generate_subquery_vars(root,
1419 : subselect->targetList,
1420 : rtindex);
1421 :
1422 : /*
1423 : * Build the new join's qual expression, replacing Params with these Vars.
1424 : */
1425 4476 : quals = convert_testexpr(root, sublink->testexpr, subquery_vars);
1426 :
1427 : /*
1428 : * And finally, build the JoinExpr node.
1429 : */
1430 4476 : result = makeNode(JoinExpr);
1431 4476 : result->jointype = JOIN_SEMI;
1432 4476 : result->isNatural = false;
1433 4476 : result->larg = NULL; /* caller must fill this in */
1434 4476 : result->rarg = (Node *) rtr;
1435 4476 : result->usingClause = NIL;
1436 4476 : result->join_using_alias = NULL;
1437 4476 : result->quals = quals;
1438 4476 : result->alias = NULL;
1439 4476 : result->rtindex = 0; /* we don't need an RTE for it */
1440 :
1441 4476 : return result;
1442 : }
1443 :
1444 : /*
1445 : * convert_EXISTS_sublink_to_join: try to convert an EXISTS SubLink to a join
1446 : *
1447 : * The API of this function is identical to convert_ANY_sublink_to_join's,
1448 : * except that we also support the case where the caller has found NOT EXISTS,
1449 : * so we need an additional input parameter "under_not".
1450 : */
1451 : JoinExpr *
1452 3706 : convert_EXISTS_sublink_to_join(PlannerInfo *root, SubLink *sublink,
1453 : bool under_not, Relids available_rels)
1454 : {
1455 : JoinExpr *result;
1456 3706 : Query *parse = root->parse;
1457 3706 : Query *subselect = (Query *) sublink->subselect;
1458 : Node *whereClause;
1459 : PlannerInfo subroot;
1460 : int rtoffset;
1461 : int varno;
1462 : Relids clause_varnos;
1463 : Relids upper_varnos;
1464 :
1465 : Assert(sublink->subLinkType == EXISTS_SUBLINK);
1466 :
1467 : /*
1468 : * Can't flatten if it contains WITH. (We could arrange to pull up the
1469 : * WITH into the parent query's cteList, but that risks changing the
1470 : * semantics, since a WITH ought to be executed once per associated query
1471 : * call.) Note that convert_ANY_sublink_to_join doesn't have to reject
1472 : * this case, since it just produces a subquery RTE that doesn't have to
1473 : * get flattened into the parent query.
1474 : */
1475 3706 : if (subselect->cteList)
1476 0 : return NULL;
1477 :
1478 : /*
1479 : * Copy the subquery so we can modify it safely (see comments in
1480 : * make_subplan).
1481 : */
1482 3706 : subselect = copyObject(subselect);
1483 :
1484 : /*
1485 : * See if the subquery can be simplified based on the knowledge that it's
1486 : * being used in EXISTS(). If we aren't able to get rid of its
1487 : * targetlist, we have to fail, because the pullup operation leaves us
1488 : * with noplace to evaluate the targetlist.
1489 : */
1490 3706 : if (!simplify_EXISTS_query(root, subselect))
1491 32 : return NULL;
1492 :
1493 : /*
1494 : * Separate out the WHERE clause. (We could theoretically also remove
1495 : * top-level plain JOIN/ON clauses, but it's probably not worth the
1496 : * trouble.)
1497 : */
1498 3674 : whereClause = subselect->jointree->quals;
1499 3674 : subselect->jointree->quals = NULL;
1500 :
1501 : /*
1502 : * The rest of the sub-select must not refer to any Vars of the parent
1503 : * query. (Vars of higher levels should be okay, though.)
1504 : */
1505 3674 : if (contain_vars_of_level((Node *) subselect, 1))
1506 0 : return NULL;
1507 :
1508 : /*
1509 : * On the other hand, the WHERE clause must contain some Vars of the
1510 : * parent query, else it's not gonna be a join.
1511 : */
1512 3674 : if (!contain_vars_of_level(whereClause, 1))
1513 102 : return NULL;
1514 :
1515 : /*
1516 : * We don't risk optimizing if the WHERE clause is volatile, either.
1517 : */
1518 3572 : if (contain_volatile_functions(whereClause))
1519 0 : return NULL;
1520 :
1521 : /*
1522 : * Scan the rangetable for relation RTEs and retrieve the necessary
1523 : * catalog information for each relation. Using this information, clear
1524 : * the inh flag for any relation that has no children, collect not-null
1525 : * attribute numbers for any relation that has column not-null
1526 : * constraints, and expand virtual generated columns for any relation that
1527 : * contains them.
1528 : *
1529 : * Note: we construct up an entirely dummy PlannerInfo for use here. This
1530 : * is fine because only the "glob" and "parse" links will be used in this
1531 : * case.
1532 : *
1533 : * Note: we temporarily assign back the WHERE clause so that any virtual
1534 : * generated column references within it can be expanded. It should be
1535 : * separated out again afterward.
1536 : */
1537 332196 : MemSet(&subroot, 0, sizeof(subroot));
1538 3572 : subroot.type = T_PlannerInfo;
1539 3572 : subroot.glob = root->glob;
1540 3572 : subroot.parse = subselect;
1541 3572 : subselect->jointree->quals = whereClause;
1542 3572 : subselect = preprocess_relation_rtes(&subroot);
1543 :
1544 : /*
1545 : * Now separate out the WHERE clause again.
1546 : */
1547 3572 : whereClause = subselect->jointree->quals;
1548 3572 : subselect->jointree->quals = NULL;
1549 :
1550 : /*
1551 : * The subquery must have a nonempty jointree, but we can make it so.
1552 : */
1553 3572 : replace_empty_jointree(subselect);
1554 :
1555 : /*
1556 : * Prepare to pull up the sub-select into top range table.
1557 : *
1558 : * We rely here on the assumption that the outer query has no references
1559 : * to the inner (necessarily true). Therefore this is a lot easier than
1560 : * what pull_up_subqueries has to go through.
1561 : *
1562 : * In fact, it's even easier than what convert_ANY_sublink_to_join has to
1563 : * do. The machinations of simplify_EXISTS_query ensured that there is
1564 : * nothing interesting in the subquery except an rtable and jointree, and
1565 : * even the jointree FromExpr no longer has quals. So we can just append
1566 : * the rtable to our own and use the FromExpr in our jointree. But first,
1567 : * adjust all level-zero varnos in the subquery to account for the rtable
1568 : * merger.
1569 : */
1570 3572 : rtoffset = list_length(parse->rtable);
1571 3572 : OffsetVarNodes((Node *) subselect, rtoffset, 0);
1572 3572 : OffsetVarNodes(whereClause, rtoffset, 0);
1573 :
1574 : /*
1575 : * Upper-level vars in subquery will now be one level closer to their
1576 : * parent than before; in particular, anything that had been level 1
1577 : * becomes level zero.
1578 : */
1579 3572 : IncrementVarSublevelsUp((Node *) subselect, -1, 1);
1580 3572 : IncrementVarSublevelsUp(whereClause, -1, 1);
1581 :
1582 : /*
1583 : * Now that the WHERE clause is adjusted to match the parent query
1584 : * environment, we can easily identify all the level-zero rels it uses.
1585 : * The ones <= rtoffset belong to the upper query; the ones > rtoffset do
1586 : * not.
1587 : */
1588 3572 : clause_varnos = pull_varnos(root, whereClause);
1589 3572 : upper_varnos = NULL;
1590 3572 : varno = -1;
1591 10788 : while ((varno = bms_next_member(clause_varnos, varno)) >= 0)
1592 : {
1593 7216 : if (varno <= rtoffset)
1594 3620 : upper_varnos = bms_add_member(upper_varnos, varno);
1595 : }
1596 3572 : bms_free(clause_varnos);
1597 : Assert(!bms_is_empty(upper_varnos));
1598 :
1599 : /*
1600 : * Now that we've got the set of upper-level varnos, we can make the last
1601 : * check: only available_rels can be referenced.
1602 : */
1603 3572 : if (!bms_is_subset(upper_varnos, available_rels))
1604 44 : return NULL;
1605 :
1606 : /*
1607 : * Now we can attach the modified subquery rtable to the parent. This also
1608 : * adds subquery's RTEPermissionInfos into the upper query.
1609 : */
1610 3528 : CombineRangeTables(&parse->rtable, &parse->rteperminfos,
1611 : subselect->rtable, subselect->rteperminfos);
1612 :
1613 : /*
1614 : * And finally, build the JoinExpr node.
1615 : */
1616 3528 : result = makeNode(JoinExpr);
1617 3528 : result->jointype = under_not ? JOIN_ANTI : JOIN_SEMI;
1618 3528 : result->isNatural = false;
1619 3528 : result->larg = NULL; /* caller must fill this in */
1620 : /* flatten out the FromExpr node if it's useless */
1621 3528 : if (list_length(subselect->jointree->fromlist) == 1)
1622 3510 : result->rarg = (Node *) linitial(subselect->jointree->fromlist);
1623 : else
1624 18 : result->rarg = (Node *) subselect->jointree;
1625 3528 : result->usingClause = NIL;
1626 3528 : result->join_using_alias = NULL;
1627 3528 : result->quals = whereClause;
1628 3528 : result->alias = NULL;
1629 3528 : result->rtindex = 0; /* we don't need an RTE for it */
1630 :
1631 3528 : return result;
1632 : }
1633 :
1634 : /*
1635 : * simplify_EXISTS_query: remove any useless stuff in an EXISTS's subquery
1636 : *
1637 : * The only thing that matters about an EXISTS query is whether it returns
1638 : * zero or more than zero rows. Therefore, we can remove certain SQL features
1639 : * that won't affect that. The only part that is really likely to matter in
1640 : * typical usage is simplifying the targetlist: it's a common habit to write
1641 : * "SELECT * FROM" even though there is no need to evaluate any columns.
1642 : *
1643 : * Note: by suppressing the targetlist we could cause an observable behavioral
1644 : * change, namely that any errors that might occur in evaluating the tlist
1645 : * won't occur, nor will other side-effects of volatile functions. This seems
1646 : * unlikely to bother anyone in practice.
1647 : *
1648 : * Returns true if was able to discard the targetlist, else false.
1649 : */
1650 : static bool
1651 8388 : simplify_EXISTS_query(PlannerInfo *root, Query *query)
1652 : {
1653 : ListCell *lc;
1654 :
1655 : /*
1656 : * We don't try to simplify at all if the query uses set operations,
1657 : * aggregates, grouping sets, SRFs, modifying CTEs, HAVING, OFFSET, or FOR
1658 : * UPDATE/SHARE; none of these seem likely in normal usage and their
1659 : * possible effects are complex. (Note: we could ignore an "OFFSET 0"
1660 : * clause, but that traditionally is used as an optimization fence, so we
1661 : * don't.)
1662 : */
1663 8388 : if (query->commandType != CMD_SELECT ||
1664 8388 : query->setOperations ||
1665 8388 : query->hasAggs ||
1666 8388 : query->groupingSets ||
1667 8388 : query->hasWindowFuncs ||
1668 8388 : query->hasTargetSRFs ||
1669 8388 : query->hasModifyingCTE ||
1670 8388 : query->havingQual ||
1671 8388 : query->limitOffset ||
1672 8364 : query->rowMarks)
1673 52 : return false;
1674 :
1675 : /*
1676 : * LIMIT with a constant positive (or NULL) value doesn't affect the
1677 : * semantics of EXISTS, so let's ignore such clauses. This is worth doing
1678 : * because people accustomed to certain other DBMSes may be in the habit
1679 : * of writing EXISTS(SELECT ... LIMIT 1) as an optimization. If there's a
1680 : * LIMIT with anything else as argument, though, we can't simplify.
1681 : */
1682 8336 : if (query->limitCount)
1683 : {
1684 : /*
1685 : * The LIMIT clause has not yet been through eval_const_expressions,
1686 : * so we have to apply that here. It might seem like this is a waste
1687 : * of cycles, since the only case plausibly worth worrying about is
1688 : * "LIMIT 1" ... but what we'll actually see is "LIMIT int8(1::int4)",
1689 : * so we have to fold constants or we're not going to recognize it.
1690 : */
1691 24 : Node *node = eval_const_expressions(root, query->limitCount);
1692 : Const *limit;
1693 :
1694 : /* Might as well update the query if we simplified the clause. */
1695 24 : query->limitCount = node;
1696 :
1697 24 : if (!IsA(node, Const))
1698 0 : return false;
1699 :
1700 24 : limit = (Const *) node;
1701 : Assert(limit->consttype == INT8OID);
1702 24 : if (!limit->constisnull && DatumGetInt64(limit->constvalue) <= 0)
1703 12 : return false;
1704 :
1705 : /* Whether or not the targetlist is safe, we can drop the LIMIT. */
1706 12 : query->limitCount = NULL;
1707 : }
1708 :
1709 : /*
1710 : * Otherwise, we can throw away the targetlist, as well as any GROUP,
1711 : * WINDOW, DISTINCT, and ORDER BY clauses; none of those clauses will
1712 : * change a nonzero-rows result to zero rows or vice versa. (Furthermore,
1713 : * since our parsetree representation of these clauses depends on the
1714 : * targetlist, we'd better throw them away if we drop the targetlist.)
1715 : */
1716 8324 : query->targetList = NIL;
1717 8324 : query->groupClause = NIL;
1718 8324 : query->windowClause = NIL;
1719 8324 : query->distinctClause = NIL;
1720 8324 : query->sortClause = NIL;
1721 8324 : query->hasDistinctOn = false;
1722 :
1723 : /*
1724 : * Since we have thrown away the GROUP BY clauses, we'd better remove the
1725 : * RTE_GROUP RTE and clear the hasGroupRTE flag.
1726 : */
1727 17082 : foreach(lc, query->rtable)
1728 : {
1729 8764 : RangeTblEntry *rte = lfirst_node(RangeTblEntry, lc);
1730 :
1731 : /*
1732 : * Remove the RTE_GROUP RTE and clear the hasGroupRTE flag. (Since
1733 : * we'll exit the foreach loop immediately, we don't bother with
1734 : * foreach_delete_current.)
1735 : */
1736 8764 : if (rte->rtekind == RTE_GROUP)
1737 : {
1738 : Assert(query->hasGroupRTE);
1739 6 : query->rtable = list_delete_cell(query->rtable, lc);
1740 6 : query->hasGroupRTE = false;
1741 6 : break;
1742 : }
1743 : }
1744 :
1745 8324 : return true;
1746 : }
1747 :
1748 : /*
1749 : * convert_EXISTS_to_ANY: try to convert EXISTS to a hashable ANY sublink
1750 : *
1751 : * The subselect is expected to be a fresh copy that we can munge up,
1752 : * and to have been successfully passed through simplify_EXISTS_query.
1753 : *
1754 : * On success, the modified subselect is returned, and we store a suitable
1755 : * upper-level test expression at *testexpr, plus a list of the subselect's
1756 : * output Params at *paramIds. (The test expression is already Param-ified
1757 : * and hence need not go through convert_testexpr, which is why we have to
1758 : * deal with the Param IDs specially.)
1759 : *
1760 : * On failure, returns NULL.
1761 : */
1762 : static Query *
1763 2174 : convert_EXISTS_to_ANY(PlannerInfo *root, Query *subselect,
1764 : Node **testexpr, List **paramIds)
1765 : {
1766 : Node *whereClause;
1767 : PlannerInfo subroot;
1768 : List *leftargs,
1769 : *rightargs,
1770 : *opids,
1771 : *opcollations,
1772 : *newWhere,
1773 : *tlist,
1774 : *testlist,
1775 : *paramids;
1776 : ListCell *lc,
1777 : *rc,
1778 : *oc,
1779 : *cc;
1780 : AttrNumber resno;
1781 :
1782 : /*
1783 : * Query must not require a targetlist, since we have to insert a new one.
1784 : * Caller should have dealt with the case already.
1785 : */
1786 : Assert(subselect->targetList == NIL);
1787 :
1788 : /*
1789 : * Separate out the WHERE clause. (We could theoretically also remove
1790 : * top-level plain JOIN/ON clauses, but it's probably not worth the
1791 : * trouble.)
1792 : */
1793 2174 : whereClause = subselect->jointree->quals;
1794 2174 : subselect->jointree->quals = NULL;
1795 :
1796 : /*
1797 : * The rest of the sub-select must not refer to any Vars of the parent
1798 : * query. (Vars of higher levels should be okay, though.)
1799 : *
1800 : * Note: we need not check for Aggrefs separately because we know the
1801 : * sub-select is as yet unoptimized; any uplevel Aggref must therefore
1802 : * contain an uplevel Var reference. This is not the case below ...
1803 : */
1804 2174 : if (contain_vars_of_level((Node *) subselect, 1))
1805 6 : return NULL;
1806 :
1807 : /*
1808 : * We don't risk optimizing if the WHERE clause is volatile, either.
1809 : */
1810 2168 : if (contain_volatile_functions(whereClause))
1811 0 : return NULL;
1812 :
1813 : /*
1814 : * Clean up the WHERE clause by doing const-simplification etc on it.
1815 : * Aside from simplifying the processing we're about to do, this is
1816 : * important for being able to pull chunks of the WHERE clause up into the
1817 : * parent query. Since we are invoked partway through the parent's
1818 : * preprocess_expression() work, earlier steps of preprocess_expression()
1819 : * wouldn't get applied to the pulled-up stuff unless we do them here. For
1820 : * the parts of the WHERE clause that get put back into the child query,
1821 : * this work is partially duplicative, but it shouldn't hurt.
1822 : *
1823 : * Note: we do not run flatten_join_alias_vars. This is OK because any
1824 : * parent aliases were flattened already, and we're not going to pull any
1825 : * child Vars (of any description) into the parent.
1826 : *
1827 : * Note: we construct up an entirely dummy PlannerInfo to pass to
1828 : * eval_const_expressions. This is fine because only the "glob" and
1829 : * "parse" links are used by eval_const_expressions.
1830 : */
1831 201624 : MemSet(&subroot, 0, sizeof(subroot));
1832 2168 : subroot.type = T_PlannerInfo;
1833 2168 : subroot.glob = root->glob;
1834 2168 : subroot.parse = subselect;
1835 2168 : whereClause = eval_const_expressions(&subroot, whereClause);
1836 2168 : whereClause = (Node *) canonicalize_qual((Expr *) whereClause, false);
1837 2168 : whereClause = (Node *) make_ands_implicit((Expr *) whereClause);
1838 :
1839 : /*
1840 : * We now have a flattened implicit-AND list of clauses, which we try to
1841 : * break apart into "outervar = innervar" hash clauses. Anything that
1842 : * can't be broken apart just goes back into the newWhere list. Note that
1843 : * we aren't trying hard yet to ensure that we have only outer or only
1844 : * inner on each side; we'll check that if we get to the end.
1845 : */
1846 2168 : leftargs = rightargs = opids = opcollations = newWhere = NIL;
1847 8594 : foreach(lc, (List *) whereClause)
1848 : {
1849 6426 : OpExpr *expr = (OpExpr *) lfirst(lc);
1850 :
1851 10496 : if (IsA(expr, OpExpr) &&
1852 4070 : hash_ok_operator(expr))
1853 : {
1854 3224 : Node *leftarg = (Node *) linitial(expr->args);
1855 3224 : Node *rightarg = (Node *) lsecond(expr->args);
1856 :
1857 3224 : if (contain_vars_of_level(leftarg, 1))
1858 : {
1859 406 : leftargs = lappend(leftargs, leftarg);
1860 406 : rightargs = lappend(rightargs, rightarg);
1861 406 : opids = lappend_oid(opids, expr->opno);
1862 406 : opcollations = lappend_oid(opcollations, expr->inputcollid);
1863 406 : continue;
1864 : }
1865 2818 : if (contain_vars_of_level(rightarg, 1))
1866 : {
1867 : /*
1868 : * We must commute the clause to put the outer var on the
1869 : * left, because the hashing code in nodeSubplan.c expects
1870 : * that. This probably shouldn't ever fail, since hashable
1871 : * operators ought to have commutators, but be paranoid.
1872 : */
1873 2192 : expr->opno = get_commutator(expr->opno);
1874 2192 : if (OidIsValid(expr->opno) && hash_ok_operator(expr))
1875 : {
1876 2192 : leftargs = lappend(leftargs, rightarg);
1877 2192 : rightargs = lappend(rightargs, leftarg);
1878 2192 : opids = lappend_oid(opids, expr->opno);
1879 2192 : opcollations = lappend_oid(opcollations, expr->inputcollid);
1880 2192 : continue;
1881 : }
1882 : /* If no commutator, no chance to optimize the WHERE clause */
1883 0 : return NULL;
1884 : }
1885 : }
1886 : /* Couldn't handle it as a hash clause */
1887 3828 : newWhere = lappend(newWhere, expr);
1888 : }
1889 :
1890 : /*
1891 : * If we didn't find anything we could convert, fail.
1892 : */
1893 2168 : if (leftargs == NIL)
1894 370 : return NULL;
1895 :
1896 : /*
1897 : * There mustn't be any parent Vars or Aggs in the stuff that we intend to
1898 : * put back into the child query. Note: you might think we don't need to
1899 : * check for Aggs separately, because an uplevel Agg must contain an
1900 : * uplevel Var in its argument. But it is possible that the uplevel Var
1901 : * got optimized away by eval_const_expressions. Consider
1902 : *
1903 : * SUM(CASE WHEN false THEN uplevelvar ELSE 0 END)
1904 : */
1905 3496 : if (contain_vars_of_level((Node *) newWhere, 1) ||
1906 1698 : contain_vars_of_level((Node *) rightargs, 1))
1907 100 : return NULL;
1908 1740 : if (root->parse->hasAggs &&
1909 84 : (contain_aggs_of_level((Node *) newWhere, 1) ||
1910 42 : contain_aggs_of_level((Node *) rightargs, 1)))
1911 0 : return NULL;
1912 :
1913 : /*
1914 : * And there can't be any child Vars in the stuff we intend to pull up.
1915 : * (Note: we'd need to check for child Aggs too, except we know the child
1916 : * has no aggs at all because of simplify_EXISTS_query's check. The same
1917 : * goes for window functions.)
1918 : */
1919 1698 : if (contain_vars_of_level((Node *) leftargs, 0))
1920 0 : return NULL;
1921 :
1922 : /*
1923 : * Also reject sublinks in the stuff we intend to pull up. (It might be
1924 : * possible to support this, but doesn't seem worth the complication.)
1925 : */
1926 1698 : if (contain_subplans((Node *) leftargs))
1927 0 : return NULL;
1928 :
1929 : /*
1930 : * Okay, adjust the sublevelsup in the stuff we're pulling up.
1931 : */
1932 1698 : IncrementVarSublevelsUp((Node *) leftargs, -1, 1);
1933 :
1934 : /*
1935 : * Put back any child-level-only WHERE clauses.
1936 : */
1937 1698 : if (newWhere)
1938 1494 : subselect->jointree->quals = (Node *) make_ands_explicit(newWhere);
1939 :
1940 : /*
1941 : * Build a new targetlist for the child that emits the expressions we
1942 : * need. Concurrently, build a testexpr for the parent using Params to
1943 : * reference the child outputs. (Since we generate Params directly here,
1944 : * there will be no need to convert the testexpr in build_subplan.)
1945 : */
1946 1698 : tlist = testlist = paramids = NIL;
1947 1698 : resno = 1;
1948 4196 : forfour(lc, leftargs, rc, rightargs, oc, opids, cc, opcollations)
1949 : {
1950 2498 : Node *leftarg = (Node *) lfirst(lc);
1951 2498 : Node *rightarg = (Node *) lfirst(rc);
1952 2498 : Oid opid = lfirst_oid(oc);
1953 2498 : Oid opcollation = lfirst_oid(cc);
1954 : Param *param;
1955 :
1956 2498 : param = generate_new_exec_param(root,
1957 : exprType(rightarg),
1958 : exprTypmod(rightarg),
1959 : exprCollation(rightarg));
1960 2498 : tlist = lappend(tlist,
1961 2498 : makeTargetEntry((Expr *) rightarg,
1962 2498 : resno++,
1963 : NULL,
1964 : false));
1965 2498 : testlist = lappend(testlist,
1966 2498 : make_opclause(opid, BOOLOID, false,
1967 : (Expr *) leftarg, (Expr *) param,
1968 : InvalidOid, opcollation));
1969 2498 : paramids = lappend_int(paramids, param->paramid);
1970 : }
1971 :
1972 : /* Put everything where it should go, and we're done */
1973 1698 : subselect->targetList = tlist;
1974 1698 : *testexpr = (Node *) make_ands_explicit(testlist);
1975 1698 : *paramIds = paramids;
1976 :
1977 1698 : return subselect;
1978 : }
1979 :
1980 :
1981 : /*
1982 : * Replace correlation vars (uplevel vars) with Params.
1983 : *
1984 : * Uplevel PlaceHolderVars, aggregates, GROUPING() expressions,
1985 : * MergeSupportFuncs, and ReturningExprs are replaced, too.
1986 : *
1987 : * Note: it is critical that this runs immediately after SS_process_sublinks.
1988 : * Since we do not recurse into the arguments of uplevel PHVs and aggregates,
1989 : * they will get copied to the appropriate subplan args list in the parent
1990 : * query with uplevel vars not replaced by Params, but only adjusted in level
1991 : * (see replace_outer_placeholdervar and replace_outer_agg). That's exactly
1992 : * what we want for the vars of the parent level --- but if a PHV's or
1993 : * aggregate's argument contains any further-up variables, they have to be
1994 : * replaced with Params in their turn. That will happen when the parent level
1995 : * runs SS_replace_correlation_vars. Therefore it must do so after expanding
1996 : * its sublinks to subplans. And we don't want any steps in between, else
1997 : * those steps would never get applied to the argument expressions, either in
1998 : * the parent or the child level.
1999 : *
2000 : * Another fairly tricky thing going on here is the handling of SubLinks in
2001 : * the arguments of uplevel PHVs/aggregates. Those are not touched inside the
2002 : * intermediate query level, either. Instead, SS_process_sublinks recurses on
2003 : * them after copying the PHV or Aggref expression into the parent plan level
2004 : * (this is actually taken care of in build_subplan).
2005 : */
2006 : Node *
2007 173298 : SS_replace_correlation_vars(PlannerInfo *root, Node *expr)
2008 : {
2009 : /* No setup needed for tree walk, so away we go */
2010 173298 : return replace_correlation_vars_mutator(expr, root);
2011 : }
2012 :
2013 : static Node *
2014 1530396 : replace_correlation_vars_mutator(Node *node, PlannerInfo *root)
2015 : {
2016 1530396 : if (node == NULL)
2017 62784 : return NULL;
2018 1467612 : if (IsA(node, Var))
2019 : {
2020 405920 : if (((Var *) node)->varlevelsup > 0)
2021 53082 : return (Node *) replace_outer_var(root, (Var *) node);
2022 : }
2023 1414530 : if (IsA(node, PlaceHolderVar))
2024 : {
2025 102 : if (((PlaceHolderVar *) node)->phlevelsup > 0)
2026 60 : return (Node *) replace_outer_placeholdervar(root,
2027 : (PlaceHolderVar *) node);
2028 : }
2029 1414470 : if (IsA(node, Aggref))
2030 : {
2031 8770 : if (((Aggref *) node)->agglevelsup > 0)
2032 52 : return (Node *) replace_outer_agg(root, (Aggref *) node);
2033 : }
2034 1414418 : if (IsA(node, GroupingFunc))
2035 : {
2036 90 : if (((GroupingFunc *) node)->agglevelsup > 0)
2037 64 : return (Node *) replace_outer_grouping(root, (GroupingFunc *) node);
2038 : }
2039 1414354 : if (IsA(node, MergeSupportFunc))
2040 : {
2041 36 : if (root->parse->commandType != CMD_MERGE)
2042 6 : return (Node *) replace_outer_merge_support(root,
2043 : (MergeSupportFunc *) node);
2044 : }
2045 1414348 : if (IsA(node, ReturningExpr))
2046 : {
2047 18 : if (((ReturningExpr *) node)->retlevelsup > 0)
2048 18 : return (Node *) replace_outer_returning(root,
2049 : (ReturningExpr *) node);
2050 : }
2051 1414330 : return expression_tree_mutator(node, replace_correlation_vars_mutator, root);
2052 : }
2053 :
2054 : /*
2055 : * Expand SubLinks to SubPlans in the given expression.
2056 : *
2057 : * The isQual argument tells whether or not this expression is a WHERE/HAVING
2058 : * qualifier expression. If it is, any sublinks appearing at top level need
2059 : * not distinguish FALSE from UNKNOWN return values.
2060 : */
2061 : Node *
2062 104876 : SS_process_sublinks(PlannerInfo *root, Node *expr, bool isQual)
2063 : {
2064 : process_sublinks_context context;
2065 :
2066 104876 : context.root = root;
2067 104876 : context.isTopQual = isQual;
2068 104876 : return process_sublinks_mutator(expr, &context);
2069 : }
2070 :
2071 : static Node *
2072 1370002 : process_sublinks_mutator(Node *node, process_sublinks_context *context)
2073 : {
2074 : process_sublinks_context locContext;
2075 :
2076 1370002 : locContext.root = context->root;
2077 :
2078 1370002 : if (node == NULL)
2079 59312 : return NULL;
2080 1310690 : if (IsA(node, SubLink))
2081 : {
2082 39012 : SubLink *sublink = (SubLink *) node;
2083 : Node *testexpr;
2084 :
2085 : /*
2086 : * First, recursively process the lefthand-side expressions, if any.
2087 : * They're not top-level anymore.
2088 : */
2089 39012 : locContext.isTopQual = false;
2090 39012 : testexpr = process_sublinks_mutator(sublink->testexpr, &locContext);
2091 :
2092 : /*
2093 : * Now build the SubPlan node and make the expr to return.
2094 : */
2095 39012 : return make_subplan(context->root,
2096 39012 : (Query *) sublink->subselect,
2097 : sublink->subLinkType,
2098 : sublink->subLinkId,
2099 : testexpr,
2100 39012 : context->isTopQual);
2101 : }
2102 :
2103 : /*
2104 : * Don't recurse into the arguments of an outer PHV, Aggref, GroupingFunc,
2105 : * or ReturningExpr here. Any SubLinks in the arguments have to be dealt
2106 : * with at the outer query level; they'll be handled when build_subplan
2107 : * collects the PHV, Aggref, GroupingFunc, or ReturningExpr into the
2108 : * arguments to be passed down to the current subplan.
2109 : */
2110 1271678 : if (IsA(node, PlaceHolderVar))
2111 : {
2112 238 : if (((PlaceHolderVar *) node)->phlevelsup > 0)
2113 12 : return node;
2114 : }
2115 1271440 : else if (IsA(node, Aggref))
2116 : {
2117 610 : if (((Aggref *) node)->agglevelsup > 0)
2118 18 : return node;
2119 : }
2120 1270830 : else if (IsA(node, GroupingFunc))
2121 : {
2122 160 : if (((GroupingFunc *) node)->agglevelsup > 0)
2123 36 : return node;
2124 : }
2125 1270670 : else if (IsA(node, ReturningExpr))
2126 : {
2127 198 : if (((ReturningExpr *) node)->retlevelsup > 0)
2128 6 : return node;
2129 : }
2130 :
2131 : /*
2132 : * We should never see a SubPlan expression in the input (since this is
2133 : * the very routine that creates 'em to begin with). We shouldn't find
2134 : * ourselves invoked directly on a Query, either.
2135 : */
2136 : Assert(!IsA(node, SubPlan));
2137 : Assert(!IsA(node, AlternativeSubPlan));
2138 : Assert(!IsA(node, Query));
2139 :
2140 : /*
2141 : * Because make_subplan() could return an AND or OR clause, we have to
2142 : * take steps to preserve AND/OR flatness of a qual. We assume the input
2143 : * has been AND/OR flattened and so we need no recursion here.
2144 : *
2145 : * (Due to the coding here, we will not get called on the List subnodes of
2146 : * an AND; and the input is *not* yet in implicit-AND format. So no check
2147 : * is needed for a bare List.)
2148 : *
2149 : * Anywhere within the top-level AND/OR clause structure, we can tell
2150 : * make_subplan() that NULL and FALSE are interchangeable. So isTopQual
2151 : * propagates down in both cases. (Note that this is unlike the meaning
2152 : * of "top level qual" used in most other places in Postgres.)
2153 : */
2154 1271606 : if (is_andclause(node))
2155 : {
2156 19934 : List *newargs = NIL;
2157 : ListCell *l;
2158 :
2159 : /* Still at qual top-level */
2160 19934 : locContext.isTopQual = context->isTopQual;
2161 :
2162 72776 : foreach(l, ((BoolExpr *) node)->args)
2163 : {
2164 : Node *newarg;
2165 :
2166 52842 : newarg = process_sublinks_mutator(lfirst(l), &locContext);
2167 52842 : if (is_andclause(newarg))
2168 0 : newargs = list_concat(newargs, ((BoolExpr *) newarg)->args);
2169 : else
2170 52842 : newargs = lappend(newargs, newarg);
2171 : }
2172 19934 : return (Node *) make_andclause(newargs);
2173 : }
2174 :
2175 1251672 : if (is_orclause(node))
2176 : {
2177 2242 : List *newargs = NIL;
2178 : ListCell *l;
2179 :
2180 : /* Still at qual top-level */
2181 2242 : locContext.isTopQual = context->isTopQual;
2182 :
2183 8026 : foreach(l, ((BoolExpr *) node)->args)
2184 : {
2185 : Node *newarg;
2186 :
2187 5784 : newarg = process_sublinks_mutator(lfirst(l), &locContext);
2188 5784 : if (is_orclause(newarg))
2189 0 : newargs = list_concat(newargs, ((BoolExpr *) newarg)->args);
2190 : else
2191 5784 : newargs = lappend(newargs, newarg);
2192 : }
2193 2242 : return (Node *) make_orclause(newargs);
2194 : }
2195 :
2196 : /*
2197 : * If we recurse down through anything other than an AND or OR node, we
2198 : * are definitely not at top qual level anymore.
2199 : */
2200 1249430 : locContext.isTopQual = false;
2201 :
2202 1249430 : return expression_tree_mutator(node,
2203 : process_sublinks_mutator,
2204 : &locContext);
2205 : }
2206 :
2207 : /*
2208 : * SS_identify_outer_params - identify the Params available from outer levels
2209 : *
2210 : * This must be run after SS_replace_correlation_vars and SS_process_sublinks
2211 : * processing is complete in a given query level as well as all of its
2212 : * descendant levels (which means it's most practical to do it at the end of
2213 : * processing the query level). We compute the set of paramIds that outer
2214 : * levels will make available to this level+descendants, and record it in
2215 : * root->outer_params for use while computing extParam/allParam sets in final
2216 : * plan cleanup. (We can't just compute it then, because the upper levels'
2217 : * plan_params lists are transient and will be gone by then.)
2218 : */
2219 : void
2220 523376 : SS_identify_outer_params(PlannerInfo *root)
2221 : {
2222 : Bitmapset *outer_params;
2223 : PlannerInfo *proot;
2224 : ListCell *l;
2225 :
2226 : /*
2227 : * If no parameters have been assigned anywhere in the tree, we certainly
2228 : * don't need to do anything here.
2229 : */
2230 523376 : if (root->glob->paramExecTypes == NIL)
2231 354346 : return;
2232 :
2233 : /*
2234 : * Scan all query levels above this one to see which parameters are due to
2235 : * be available from them, either because lower query levels have
2236 : * requested them (via plan_params) or because they will be available from
2237 : * initPlans of those levels.
2238 : */
2239 169030 : outer_params = NULL;
2240 225976 : for (proot = root->parent_root; proot != NULL; proot = proot->parent_root)
2241 : {
2242 : /*
2243 : * Include ordinary Var/PHV/Aggref/GroupingFunc/ReturningExpr params.
2244 : */
2245 102906 : foreach(l, proot->plan_params)
2246 : {
2247 45960 : PlannerParamItem *pitem = (PlannerParamItem *) lfirst(l);
2248 :
2249 45960 : outer_params = bms_add_member(outer_params, pitem->paramId);
2250 : }
2251 : /* Include any outputs of outer-level initPlans */
2252 63220 : foreach(l, proot->init_plans)
2253 : {
2254 6274 : SubPlan *initsubplan = (SubPlan *) lfirst(l);
2255 : ListCell *l2;
2256 :
2257 12548 : foreach(l2, initsubplan->setParam)
2258 : {
2259 6274 : outer_params = bms_add_member(outer_params, lfirst_int(l2));
2260 : }
2261 : }
2262 : /* Include worktable ID, if a recursive query is being planned */
2263 56946 : if (proot->wt_param_id >= 0)
2264 3240 : outer_params = bms_add_member(outer_params, proot->wt_param_id);
2265 : }
2266 169030 : root->outer_params = outer_params;
2267 : }
2268 :
2269 : /*
2270 : * SS_charge_for_initplans - account for initplans in Path costs & parallelism
2271 : *
2272 : * If any initPlans have been created in the current query level, they will
2273 : * get attached to the Plan tree created from whichever Path we select from
2274 : * the given rel. Increment all that rel's Paths' costs to account for them,
2275 : * and if any of the initPlans are parallel-unsafe, mark all the rel's Paths
2276 : * parallel-unsafe as well.
2277 : *
2278 : * This is separate from SS_attach_initplans because we might conditionally
2279 : * create more initPlans during create_plan(), depending on which Path we
2280 : * select. However, Paths that would generate such initPlans are expected
2281 : * to have included their cost and parallel-safety effects already.
2282 : */
2283 : void
2284 523376 : SS_charge_for_initplans(PlannerInfo *root, RelOptInfo *final_rel)
2285 : {
2286 : Cost initplan_cost;
2287 : bool unsafe_initplans;
2288 : ListCell *lc;
2289 :
2290 : /* Nothing to do if no initPlans */
2291 523376 : if (root->init_plans == NIL)
2292 511826 : return;
2293 :
2294 : /*
2295 : * Compute the cost increment just once, since it will be the same for all
2296 : * Paths. Also check for parallel-unsafe initPlans.
2297 : */
2298 11550 : SS_compute_initplan_cost(root->init_plans,
2299 : &initplan_cost, &unsafe_initplans);
2300 :
2301 : /*
2302 : * Now adjust the costs and parallel_safe flags.
2303 : */
2304 23274 : foreach(lc, final_rel->pathlist)
2305 : {
2306 11724 : Path *path = (Path *) lfirst(lc);
2307 :
2308 11724 : path->startup_cost += initplan_cost;
2309 11724 : path->total_cost += initplan_cost;
2310 11724 : if (unsafe_initplans)
2311 6782 : path->parallel_safe = false;
2312 : }
2313 :
2314 : /*
2315 : * Adjust partial paths' costs too, or forget them entirely if we must
2316 : * consider the rel parallel-unsafe.
2317 : */
2318 11550 : if (unsafe_initplans)
2319 : {
2320 6702 : final_rel->partial_pathlist = NIL;
2321 6702 : final_rel->consider_parallel = false;
2322 : }
2323 : else
2324 : {
2325 4860 : foreach(lc, final_rel->partial_pathlist)
2326 : {
2327 12 : Path *path = (Path *) lfirst(lc);
2328 :
2329 12 : path->startup_cost += initplan_cost;
2330 12 : path->total_cost += initplan_cost;
2331 : }
2332 : }
2333 :
2334 : /* We needn't do set_cheapest() here, caller will do it */
2335 : }
2336 :
2337 : /*
2338 : * SS_compute_initplan_cost - count up the cost delta for some initplans
2339 : *
2340 : * The total cost returned in *initplan_cost_p should be added to both the
2341 : * startup and total costs of the plan node the initplans get attached to.
2342 : * We also report whether any of the initplans are not parallel-safe.
2343 : *
2344 : * The primary user of this is SS_charge_for_initplans, but it's also
2345 : * used in adjusting costs when we move initplans to another plan node.
2346 : */
2347 : void
2348 11794 : SS_compute_initplan_cost(List *init_plans,
2349 : Cost *initplan_cost_p,
2350 : bool *unsafe_initplans_p)
2351 : {
2352 : Cost initplan_cost;
2353 : bool unsafe_initplans;
2354 : ListCell *lc;
2355 :
2356 : /*
2357 : * We assume each initPlan gets run once during top plan startup. This is
2358 : * a conservative overestimate, since in fact an initPlan might be
2359 : * executed later than plan startup, or even not at all.
2360 : */
2361 11794 : initplan_cost = 0;
2362 11794 : unsafe_initplans = false;
2363 24638 : foreach(lc, init_plans)
2364 : {
2365 12844 : SubPlan *initsubplan = lfirst_node(SubPlan, lc);
2366 :
2367 12844 : initplan_cost += initsubplan->startup_cost + initsubplan->per_call_cost;
2368 12844 : if (!initsubplan->parallel_safe)
2369 7700 : unsafe_initplans = true;
2370 : }
2371 11794 : *initplan_cost_p = initplan_cost;
2372 11794 : *unsafe_initplans_p = unsafe_initplans;
2373 11794 : }
2374 :
2375 : /*
2376 : * SS_attach_initplans - attach initplans to topmost plan node
2377 : *
2378 : * Attach any initplans created in the current query level to the specified
2379 : * plan node, which should normally be the topmost node for the query level.
2380 : * (In principle the initPlans could go in any node at or above where they're
2381 : * referenced; but there seems no reason to put them any lower than the
2382 : * topmost node, so we don't bother to track exactly where they came from.)
2383 : *
2384 : * We do not touch the plan node's cost or parallel_safe flag. The initplans
2385 : * must have been accounted for in SS_charge_for_initplans, or by any later
2386 : * code that adds initplans via SS_make_initplan_from_plan.
2387 : */
2388 : void
2389 521994 : SS_attach_initplans(PlannerInfo *root, Plan *plan)
2390 : {
2391 521994 : plan->initPlan = root->init_plans;
2392 521994 : }
2393 :
2394 : /*
2395 : * SS_finalize_plan - do final parameter processing for a completed Plan.
2396 : *
2397 : * This recursively computes the extParam and allParam sets for every Plan
2398 : * node in the given plan tree. (Oh, and RangeTblFunction.funcparams too.)
2399 : *
2400 : * We assume that SS_finalize_plan has already been run on any initplans or
2401 : * subplans the plan tree could reference.
2402 : */
2403 : void
2404 195580 : SS_finalize_plan(PlannerInfo *root, Plan *plan)
2405 : {
2406 : /* No setup needed, just recurse through plan tree. */
2407 195580 : (void) finalize_plan(root, plan, -1, root->outer_params, NULL);
2408 195580 : }
2409 :
2410 : /*
2411 : * Recursive processing of all nodes in the plan tree
2412 : *
2413 : * gather_param is the rescan_param of an ancestral Gather/GatherMerge,
2414 : * or -1 if there is none.
2415 : *
2416 : * valid_params is the set of param IDs supplied by outer plan levels
2417 : * that are valid to reference in this plan node or its children.
2418 : *
2419 : * scan_params is a set of param IDs to force scan plan nodes to reference.
2420 : * This is for EvalPlanQual support, and is always NULL at the top of the
2421 : * recursion.
2422 : *
2423 : * The return value is the computed allParam set for the given Plan node.
2424 : * This is just an internal notational convenience: we can add a child
2425 : * plan's allParams to the set of param IDs of interest to this level
2426 : * in the same statement that recurses to that child.
2427 : *
2428 : * Do not scribble on caller's values of valid_params or scan_params!
2429 : *
2430 : * Note: although we attempt to deal with initPlans anywhere in the tree, the
2431 : * logic is not really right. The problem is that a plan node might return an
2432 : * output Param of its initPlan as a targetlist item, in which case it's valid
2433 : * for the parent plan level to reference that same Param; the parent's usage
2434 : * will be converted into a Var referencing the child plan node by setrefs.c.
2435 : * But this function would see the parent's reference as out of scope and
2436 : * complain about it. For now, this does not matter because the planner only
2437 : * attaches initPlans to the topmost plan node in a query level, so the case
2438 : * doesn't arise. If we ever merge this processing into setrefs.c, maybe it
2439 : * can be handled more cleanly.
2440 : */
2441 : static Bitmapset *
2442 1550680 : finalize_plan(PlannerInfo *root, Plan *plan,
2443 : int gather_param,
2444 : Bitmapset *valid_params,
2445 : Bitmapset *scan_params)
2446 : {
2447 : finalize_primnode_context context;
2448 : int locally_added_param;
2449 : Bitmapset *nestloop_params;
2450 : Bitmapset *initExtParam;
2451 : Bitmapset *initSetParam;
2452 : Bitmapset *child_params;
2453 : ListCell *l;
2454 :
2455 1550680 : if (plan == NULL)
2456 901740 : return NULL;
2457 :
2458 648940 : context.root = root;
2459 648940 : context.paramids = NULL; /* initialize set to empty */
2460 648940 : locally_added_param = -1; /* there isn't one */
2461 648940 : nestloop_params = NULL; /* there aren't any */
2462 :
2463 : /*
2464 : * Examine any initPlans to determine the set of external params they
2465 : * reference and the set of output params they supply. (We assume
2466 : * SS_finalize_plan was run on them already.)
2467 : */
2468 648940 : initExtParam = initSetParam = NULL;
2469 662118 : foreach(l, plan->initPlan)
2470 : {
2471 13178 : SubPlan *initsubplan = (SubPlan *) lfirst(l);
2472 13178 : Plan *initplan = planner_subplan_get_plan(root, initsubplan);
2473 : ListCell *l2;
2474 :
2475 13178 : initExtParam = bms_add_members(initExtParam, initplan->extParam);
2476 26404 : foreach(l2, initsubplan->setParam)
2477 : {
2478 13226 : initSetParam = bms_add_member(initSetParam, lfirst_int(l2));
2479 : }
2480 : }
2481 :
2482 : /* Any setParams are validly referenceable in this node and children */
2483 648940 : if (initSetParam)
2484 11902 : valid_params = bms_union(valid_params, initSetParam);
2485 :
2486 : /*
2487 : * When we call finalize_primnode, context.paramids sets are automatically
2488 : * merged together. But when recursing to self, we have to do it the hard
2489 : * way. We want the paramids set to include params in subplans as well as
2490 : * at this level.
2491 : */
2492 :
2493 : /* Find params in targetlist and qual */
2494 648940 : finalize_primnode((Node *) plan->targetlist, &context);
2495 648940 : finalize_primnode((Node *) plan->qual, &context);
2496 :
2497 : /*
2498 : * If it's a parallel-aware scan node, mark it as dependent on the parent
2499 : * Gather/GatherMerge's rescan Param.
2500 : */
2501 648940 : if (plan->parallel_aware)
2502 : {
2503 2624 : if (gather_param < 0)
2504 0 : elog(ERROR, "parallel-aware plan node is not below a Gather");
2505 2624 : context.paramids =
2506 2624 : bms_add_member(context.paramids, gather_param);
2507 : }
2508 :
2509 : /* Check additional node-type-specific fields */
2510 648940 : switch (nodeTag(plan))
2511 : {
2512 71996 : case T_Result:
2513 71996 : finalize_primnode(((Result *) plan)->resconstantqual,
2514 : &context);
2515 71996 : break;
2516 :
2517 100794 : case T_SeqScan:
2518 100794 : context.paramids = bms_add_members(context.paramids, scan_params);
2519 100794 : break;
2520 :
2521 104 : case T_SampleScan:
2522 104 : finalize_primnode((Node *) ((SampleScan *) plan)->tablesample,
2523 : &context);
2524 104 : context.paramids = bms_add_members(context.paramids, scan_params);
2525 104 : break;
2526 :
2527 98262 : case T_IndexScan:
2528 98262 : finalize_primnode((Node *) ((IndexScan *) plan)->indexqual,
2529 : &context);
2530 98262 : finalize_primnode((Node *) ((IndexScan *) plan)->indexorderby,
2531 : &context);
2532 :
2533 : /*
2534 : * we need not look at indexqualorig, since it will have the same
2535 : * param references as indexqual. Likewise, we can ignore
2536 : * indexorderbyorig.
2537 : */
2538 98262 : context.paramids = bms_add_members(context.paramids, scan_params);
2539 98262 : break;
2540 :
2541 9370 : case T_IndexOnlyScan:
2542 9370 : finalize_primnode((Node *) ((IndexOnlyScan *) plan)->indexqual,
2543 : &context);
2544 9370 : finalize_primnode((Node *) ((IndexOnlyScan *) plan)->recheckqual,
2545 : &context);
2546 9370 : finalize_primnode((Node *) ((IndexOnlyScan *) plan)->indexorderby,
2547 : &context);
2548 :
2549 : /*
2550 : * we need not look at indextlist, since it cannot contain Params.
2551 : */
2552 9370 : context.paramids = bms_add_members(context.paramids, scan_params);
2553 9370 : break;
2554 :
2555 8496 : case T_BitmapIndexScan:
2556 8496 : finalize_primnode((Node *) ((BitmapIndexScan *) plan)->indexqual,
2557 : &context);
2558 :
2559 : /*
2560 : * we need not look at indexqualorig, since it will have the same
2561 : * param references as indexqual.
2562 : */
2563 8496 : break;
2564 :
2565 8100 : case T_BitmapHeapScan:
2566 8100 : finalize_primnode((Node *) ((BitmapHeapScan *) plan)->bitmapqualorig,
2567 : &context);
2568 8100 : context.paramids = bms_add_members(context.paramids, scan_params);
2569 8100 : break;
2570 :
2571 624 : case T_TidScan:
2572 624 : finalize_primnode((Node *) ((TidScan *) plan)->tidquals,
2573 : &context);
2574 624 : context.paramids = bms_add_members(context.paramids, scan_params);
2575 624 : break;
2576 :
2577 38 : case T_TidRangeScan:
2578 38 : finalize_primnode((Node *) ((TidRangeScan *) plan)->tidrangequals,
2579 : &context);
2580 38 : context.paramids = bms_add_members(context.paramids, scan_params);
2581 38 : break;
2582 :
2583 24826 : case T_SubqueryScan:
2584 : {
2585 24826 : SubqueryScan *sscan = (SubqueryScan *) plan;
2586 : RelOptInfo *rel;
2587 : Bitmapset *subquery_params;
2588 :
2589 : /* We must run finalize_plan on the subquery */
2590 24826 : rel = find_base_rel(root, sscan->scan.scanrelid);
2591 24826 : subquery_params = rel->subroot->outer_params;
2592 24826 : if (gather_param >= 0)
2593 24 : subquery_params = bms_add_member(bms_copy(subquery_params),
2594 : gather_param);
2595 24826 : finalize_plan(rel->subroot, sscan->subplan, gather_param,
2596 : subquery_params, NULL);
2597 :
2598 : /* Now we can add its extParams to the parent's params */
2599 49652 : context.paramids = bms_add_members(context.paramids,
2600 24826 : sscan->subplan->extParam);
2601 : /* We need scan_params too, though */
2602 24826 : context.paramids = bms_add_members(context.paramids,
2603 : scan_params);
2604 : }
2605 24826 : break;
2606 :
2607 24012 : case T_FunctionScan:
2608 : {
2609 24012 : FunctionScan *fscan = (FunctionScan *) plan;
2610 : ListCell *lc;
2611 :
2612 : /*
2613 : * Call finalize_primnode independently on each function
2614 : * expression, so that we can record which params are
2615 : * referenced in each, in order to decide which need
2616 : * re-evaluating during rescan.
2617 : */
2618 48236 : foreach(lc, fscan->functions)
2619 : {
2620 24224 : RangeTblFunction *rtfunc = (RangeTblFunction *) lfirst(lc);
2621 : finalize_primnode_context funccontext;
2622 :
2623 24224 : funccontext = context;
2624 24224 : funccontext.paramids = NULL;
2625 :
2626 24224 : finalize_primnode(rtfunc->funcexpr, &funccontext);
2627 :
2628 : /* remember results for execution */
2629 24224 : rtfunc->funcparams = funccontext.paramids;
2630 :
2631 : /* add the function's params to the overall set */
2632 24224 : context.paramids = bms_add_members(context.paramids,
2633 24224 : funccontext.paramids);
2634 : }
2635 :
2636 24012 : context.paramids = bms_add_members(context.paramids,
2637 : scan_params);
2638 : }
2639 24012 : break;
2640 :
2641 234 : case T_TableFuncScan:
2642 234 : finalize_primnode((Node *) ((TableFuncScan *) plan)->tablefunc,
2643 : &context);
2644 234 : context.paramids = bms_add_members(context.paramids, scan_params);
2645 234 : break;
2646 :
2647 5772 : case T_ValuesScan:
2648 5772 : finalize_primnode((Node *) ((ValuesScan *) plan)->values_lists,
2649 : &context);
2650 5772 : context.paramids = bms_add_members(context.paramids, scan_params);
2651 5772 : break;
2652 :
2653 4252 : case T_CteScan:
2654 : {
2655 : /*
2656 : * You might think we should add the node's cteParam to
2657 : * paramids, but we shouldn't because that param is just a
2658 : * linkage mechanism for multiple CteScan nodes for the same
2659 : * CTE; it is never used for changed-param signaling. What we
2660 : * have to do instead is to find the referenced CTE plan and
2661 : * incorporate its external paramids, so that the correct
2662 : * things will happen if the CTE references outer-level
2663 : * variables. See test cases for bug #4902. (We assume
2664 : * SS_finalize_plan was run on the CTE plan already.)
2665 : */
2666 4252 : int plan_id = ((CteScan *) plan)->ctePlanId;
2667 : Plan *cteplan;
2668 :
2669 : /* so, do this ... */
2670 4252 : if (plan_id < 1 || plan_id > list_length(root->glob->subplans))
2671 0 : elog(ERROR, "could not find plan for CteScan referencing plan ID %d",
2672 : plan_id);
2673 4252 : cteplan = (Plan *) list_nth(root->glob->subplans, plan_id - 1);
2674 4252 : context.paramids =
2675 4252 : bms_add_members(context.paramids, cteplan->extParam);
2676 :
2677 : #ifdef NOT_USED
2678 : /* ... but not this */
2679 : context.paramids =
2680 : bms_add_member(context.paramids,
2681 : ((CteScan *) plan)->cteParam);
2682 : #endif
2683 :
2684 4252 : context.paramids = bms_add_members(context.paramids,
2685 : scan_params);
2686 : }
2687 4252 : break;
2688 :
2689 928 : case T_WorkTableScan:
2690 928 : context.paramids =
2691 928 : bms_add_member(context.paramids,
2692 : ((WorkTableScan *) plan)->wtParam);
2693 928 : context.paramids = bms_add_members(context.paramids, scan_params);
2694 928 : break;
2695 :
2696 402 : case T_NamedTuplestoreScan:
2697 402 : context.paramids = bms_add_members(context.paramids, scan_params);
2698 402 : break;
2699 :
2700 844 : case T_ForeignScan:
2701 : {
2702 844 : ForeignScan *fscan = (ForeignScan *) plan;
2703 :
2704 844 : finalize_primnode((Node *) fscan->fdw_exprs,
2705 : &context);
2706 844 : finalize_primnode((Node *) fscan->fdw_recheck_quals,
2707 : &context);
2708 :
2709 : /* We assume fdw_scan_tlist cannot contain Params */
2710 844 : context.paramids = bms_add_members(context.paramids,
2711 : scan_params);
2712 : }
2713 844 : break;
2714 :
2715 0 : case T_CustomScan:
2716 : {
2717 0 : CustomScan *cscan = (CustomScan *) plan;
2718 : ListCell *lc;
2719 :
2720 0 : finalize_primnode((Node *) cscan->custom_exprs,
2721 : &context);
2722 : /* We assume custom_scan_tlist cannot contain Params */
2723 0 : context.paramids =
2724 0 : bms_add_members(context.paramids, scan_params);
2725 :
2726 : /* child nodes if any */
2727 0 : foreach(lc, cscan->custom_plans)
2728 : {
2729 0 : context.paramids =
2730 0 : bms_add_members(context.paramids,
2731 0 : finalize_plan(root,
2732 0 : (Plan *) lfirst(lc),
2733 : gather_param,
2734 : valid_params,
2735 : scan_params));
2736 : }
2737 : }
2738 0 : break;
2739 :
2740 87094 : case T_ModifyTable:
2741 : {
2742 87094 : ModifyTable *mtplan = (ModifyTable *) plan;
2743 :
2744 : /* Force descendant scan nodes to reference epqParam */
2745 87094 : locally_added_param = mtplan->epqParam;
2746 87094 : valid_params = bms_add_member(bms_copy(valid_params),
2747 : locally_added_param);
2748 87094 : scan_params = bms_add_member(bms_copy(scan_params),
2749 : locally_added_param);
2750 87094 : finalize_primnode((Node *) mtplan->returningLists,
2751 : &context);
2752 87094 : finalize_primnode((Node *) mtplan->onConflictSet,
2753 : &context);
2754 87094 : finalize_primnode((Node *) mtplan->onConflictWhere,
2755 : &context);
2756 : /* exclRelTlist contains only Vars, doesn't need examination */
2757 : }
2758 87094 : break;
2759 :
2760 11144 : case T_Append:
2761 : {
2762 42296 : foreach(l, ((Append *) plan)->appendplans)
2763 : {
2764 31152 : context.paramids =
2765 31152 : bms_add_members(context.paramids,
2766 31152 : finalize_plan(root,
2767 31152 : (Plan *) lfirst(l),
2768 : gather_param,
2769 : valid_params,
2770 : scan_params));
2771 : }
2772 : }
2773 11144 : break;
2774 :
2775 144 : case T_MergeAppend:
2776 : {
2777 594 : foreach(l, ((MergeAppend *) plan)->mergeplans)
2778 : {
2779 450 : context.paramids =
2780 450 : bms_add_members(context.paramids,
2781 450 : finalize_plan(root,
2782 450 : (Plan *) lfirst(l),
2783 : gather_param,
2784 : valid_params,
2785 : scan_params));
2786 : }
2787 : }
2788 144 : break;
2789 :
2790 170 : case T_BitmapAnd:
2791 : {
2792 510 : foreach(l, ((BitmapAnd *) plan)->bitmapplans)
2793 : {
2794 340 : context.paramids =
2795 340 : bms_add_members(context.paramids,
2796 340 : finalize_plan(root,
2797 340 : (Plan *) lfirst(l),
2798 : gather_param,
2799 : valid_params,
2800 : scan_params));
2801 : }
2802 : }
2803 170 : break;
2804 :
2805 226 : case T_BitmapOr:
2806 : {
2807 678 : foreach(l, ((BitmapOr *) plan)->bitmapplans)
2808 : {
2809 452 : context.paramids =
2810 452 : bms_add_members(context.paramids,
2811 452 : finalize_plan(root,
2812 452 : (Plan *) lfirst(l),
2813 : gather_param,
2814 : valid_params,
2815 : scan_params));
2816 : }
2817 : }
2818 226 : break;
2819 :
2820 72222 : case T_NestLoop:
2821 : {
2822 72222 : finalize_primnode((Node *) ((Join *) plan)->joinqual,
2823 : &context);
2824 : /* collect set of params that will be passed to right child */
2825 127610 : foreach(l, ((NestLoop *) plan)->nestParams)
2826 : {
2827 55388 : NestLoopParam *nlp = (NestLoopParam *) lfirst(l);
2828 :
2829 55388 : nestloop_params = bms_add_member(nestloop_params,
2830 : nlp->paramno);
2831 : }
2832 : }
2833 72222 : break;
2834 :
2835 4190 : case T_MergeJoin:
2836 4190 : finalize_primnode((Node *) ((Join *) plan)->joinqual,
2837 : &context);
2838 4190 : finalize_primnode((Node *) ((MergeJoin *) plan)->mergeclauses,
2839 : &context);
2840 4190 : break;
2841 :
2842 21592 : case T_HashJoin:
2843 21592 : finalize_primnode((Node *) ((Join *) plan)->joinqual,
2844 : &context);
2845 21592 : finalize_primnode((Node *) ((HashJoin *) plan)->hashclauses,
2846 : &context);
2847 21592 : break;
2848 :
2849 21592 : case T_Hash:
2850 21592 : finalize_primnode((Node *) ((Hash *) plan)->hashkeys,
2851 : &context);
2852 21592 : break;
2853 :
2854 2298 : case T_Limit:
2855 2298 : finalize_primnode(((Limit *) plan)->limitOffset,
2856 : &context);
2857 2298 : finalize_primnode(((Limit *) plan)->limitCount,
2858 : &context);
2859 2298 : break;
2860 :
2861 928 : case T_RecursiveUnion:
2862 : /* child nodes are allowed to reference wtParam */
2863 928 : locally_added_param = ((RecursiveUnion *) plan)->wtParam;
2864 928 : valid_params = bms_add_member(bms_copy(valid_params),
2865 : locally_added_param);
2866 : /* wtParam does *not* get added to scan_params */
2867 928 : break;
2868 :
2869 7898 : case T_LockRows:
2870 : /* Force descendant scan nodes to reference epqParam */
2871 7898 : locally_added_param = ((LockRows *) plan)->epqParam;
2872 7898 : valid_params = bms_add_member(bms_copy(valid_params),
2873 : locally_added_param);
2874 7898 : scan_params = bms_add_member(bms_copy(scan_params),
2875 : locally_added_param);
2876 7898 : break;
2877 :
2878 11288 : case T_Agg:
2879 : {
2880 11288 : Agg *agg = (Agg *) plan;
2881 :
2882 : /*
2883 : * AGG_HASHED plans need to know which Params are referenced
2884 : * in aggregate calls. Do a separate scan to identify them.
2885 : */
2886 11288 : if (agg->aggstrategy == AGG_HASHED)
2887 : {
2888 : finalize_primnode_context aggcontext;
2889 :
2890 1938 : aggcontext.root = root;
2891 1938 : aggcontext.paramids = NULL;
2892 1938 : finalize_agg_primnode((Node *) agg->plan.targetlist,
2893 : &aggcontext);
2894 1938 : finalize_agg_primnode((Node *) agg->plan.qual,
2895 : &aggcontext);
2896 1938 : agg->aggParams = aggcontext.paramids;
2897 : }
2898 : }
2899 11288 : break;
2900 :
2901 122 : case T_WindowAgg:
2902 122 : finalize_primnode(((WindowAgg *) plan)->startOffset,
2903 : &context);
2904 122 : finalize_primnode(((WindowAgg *) plan)->endOffset,
2905 : &context);
2906 122 : break;
2907 :
2908 962 : case T_Gather:
2909 : /* child nodes are allowed to reference rescan_param, if any */
2910 962 : locally_added_param = ((Gather *) plan)->rescan_param;
2911 962 : if (locally_added_param >= 0)
2912 : {
2913 956 : valid_params = bms_add_member(bms_copy(valid_params),
2914 : locally_added_param);
2915 :
2916 : /*
2917 : * We currently don't support nested Gathers. The issue so
2918 : * far as this function is concerned would be how to identify
2919 : * which child nodes depend on which Gather.
2920 : */
2921 : Assert(gather_param < 0);
2922 : /* Pass down rescan_param to child parallel-aware nodes */
2923 956 : gather_param = locally_added_param;
2924 : }
2925 : /* rescan_param does *not* get added to scan_params */
2926 962 : break;
2927 :
2928 354 : case T_GatherMerge:
2929 : /* child nodes are allowed to reference rescan_param, if any */
2930 354 : locally_added_param = ((GatherMerge *) plan)->rescan_param;
2931 354 : if (locally_added_param >= 0)
2932 : {
2933 354 : valid_params = bms_add_member(bms_copy(valid_params),
2934 : locally_added_param);
2935 :
2936 : /*
2937 : * We currently don't support nested Gathers. The issue so
2938 : * far as this function is concerned would be how to identify
2939 : * which child nodes depend on which Gather.
2940 : */
2941 : Assert(gather_param < 0);
2942 : /* Pass down rescan_param to child parallel-aware nodes */
2943 354 : gather_param = locally_added_param;
2944 : }
2945 : /* rescan_param does *not* get added to scan_params */
2946 354 : break;
2947 :
2948 1948 : case T_Memoize:
2949 1948 : finalize_primnode((Node *) ((Memoize *) plan)->param_exprs,
2950 : &context);
2951 1948 : break;
2952 :
2953 45714 : case T_ProjectSet:
2954 : case T_Material:
2955 : case T_Sort:
2956 : case T_IncrementalSort:
2957 : case T_Unique:
2958 : case T_SetOp:
2959 : case T_Group:
2960 : /* no node-type-specific fields need fixing */
2961 45714 : break;
2962 :
2963 0 : default:
2964 0 : elog(ERROR, "unrecognized node type: %d",
2965 : (int) nodeTag(plan));
2966 : }
2967 :
2968 : /* Process left and right child plans, if any */
2969 648940 : child_params = finalize_plan(root,
2970 648940 : plan->lefttree,
2971 : gather_param,
2972 : valid_params,
2973 : scan_params);
2974 648940 : context.paramids = bms_add_members(context.paramids, child_params);
2975 :
2976 648940 : if (nestloop_params)
2977 : {
2978 : /* right child can reference nestloop_params as well as valid_params */
2979 49018 : child_params = finalize_plan(root,
2980 49018 : plan->righttree,
2981 : gather_param,
2982 : bms_union(nestloop_params, valid_params),
2983 : scan_params);
2984 : /* ... and they don't count as parameters used at my level */
2985 49018 : child_params = bms_difference(child_params, nestloop_params);
2986 49018 : bms_free(nestloop_params);
2987 : }
2988 : else
2989 : {
2990 : /* easy case */
2991 599922 : child_params = finalize_plan(root,
2992 599922 : plan->righttree,
2993 : gather_param,
2994 : valid_params,
2995 : scan_params);
2996 : }
2997 648940 : context.paramids = bms_add_members(context.paramids, child_params);
2998 :
2999 : /*
3000 : * Any locally generated parameter doesn't count towards its generating
3001 : * plan node's external dependencies. (Note: if we changed valid_params
3002 : * and/or scan_params, we leak those bitmapsets; not worth the notational
3003 : * trouble to clean them up.)
3004 : */
3005 648940 : if (locally_added_param >= 0)
3006 : {
3007 97230 : context.paramids = bms_del_member(context.paramids,
3008 : locally_added_param);
3009 : }
3010 :
3011 : /* Now we have all the paramids referenced in this node and children */
3012 :
3013 648940 : if (!bms_is_subset(context.paramids, valid_params))
3014 0 : elog(ERROR, "plan should not reference subplan's variable");
3015 :
3016 : /*
3017 : * The plan node's allParam and extParam fields should include all its
3018 : * referenced paramids, plus contributions from any child initPlans.
3019 : * However, any setParams of the initPlans should not be present in the
3020 : * parent node's extParams, only in its allParams. (It's possible that
3021 : * some initPlans have extParams that are setParams of other initPlans.)
3022 : */
3023 :
3024 : /* allParam must include initplans' extParams and setParams */
3025 648940 : plan->allParam = bms_union(context.paramids, initExtParam);
3026 648940 : plan->allParam = bms_add_members(plan->allParam, initSetParam);
3027 : /* extParam must include any initplan extParams */
3028 648940 : plan->extParam = bms_union(context.paramids, initExtParam);
3029 : /* but not any initplan setParams */
3030 648940 : plan->extParam = bms_del_members(plan->extParam, initSetParam);
3031 :
3032 648940 : return plan->allParam;
3033 : }
3034 :
3035 : /*
3036 : * finalize_primnode: add IDs of all PARAM_EXEC params that appear (or will
3037 : * appear) in the given expression tree to the result set.
3038 : */
3039 : static bool
3040 10810296 : finalize_primnode(Node *node, finalize_primnode_context *context)
3041 : {
3042 10810296 : if (node == NULL)
3043 1268582 : return false;
3044 9541714 : if (IsA(node, Param))
3045 : {
3046 135628 : if (((Param *) node)->paramkind == PARAM_EXEC)
3047 : {
3048 133310 : int paramid = ((Param *) node)->paramid;
3049 :
3050 133310 : context->paramids = bms_add_member(context->paramids, paramid);
3051 : }
3052 135628 : return false; /* no more to do here */
3053 : }
3054 9406086 : else if (IsA(node, Aggref))
3055 : {
3056 : /*
3057 : * Check to see if the aggregate will be replaced by a Param
3058 : * referencing a subquery output during setrefs.c. If so, we must
3059 : * account for that Param here. (For various reasons, it's not
3060 : * convenient to perform that substitution earlier than setrefs.c, nor
3061 : * to perform this processing after setrefs.c. Thus we need a wart
3062 : * here.)
3063 : */
3064 15554 : Aggref *aggref = (Aggref *) node;
3065 : Param *aggparam;
3066 :
3067 15554 : aggparam = find_minmax_agg_replacement_param(context->root, aggref);
3068 15554 : if (aggparam != NULL)
3069 556 : context->paramids = bms_add_member(context->paramids,
3070 : aggparam->paramid);
3071 : /* Fall through to examine the agg's arguments */
3072 : }
3073 9390532 : else if (IsA(node, SubPlan))
3074 : {
3075 37070 : SubPlan *subplan = (SubPlan *) node;
3076 37070 : Plan *plan = planner_subplan_get_plan(context->root, subplan);
3077 : ListCell *lc;
3078 : Bitmapset *subparamids;
3079 :
3080 : /* Recurse into the testexpr, but not into the Plan */
3081 37070 : finalize_primnode(subplan->testexpr, context);
3082 :
3083 : /*
3084 : * Remove any param IDs of output parameters of the subplan that were
3085 : * referenced in the testexpr. These are not interesting for
3086 : * parameter change signaling since we always re-evaluate the subplan.
3087 : * Note that this wouldn't work too well if there might be uses of the
3088 : * same param IDs elsewhere in the plan, but that can't happen because
3089 : * generate_new_exec_param never tries to merge params.
3090 : */
3091 40300 : foreach(lc, subplan->paramIds)
3092 : {
3093 3230 : context->paramids = bms_del_member(context->paramids,
3094 : lfirst_int(lc));
3095 : }
3096 :
3097 : /* Also examine args list */
3098 37070 : finalize_primnode((Node *) subplan->args, context);
3099 :
3100 : /*
3101 : * Add params needed by the subplan to paramids, but excluding those
3102 : * we will pass down to it. (We assume SS_finalize_plan was run on
3103 : * the subplan already.)
3104 : */
3105 37070 : subparamids = bms_copy(plan->extParam);
3106 89458 : foreach(lc, subplan->parParam)
3107 : {
3108 52388 : subparamids = bms_del_member(subparamids, lfirst_int(lc));
3109 : }
3110 37070 : context->paramids = bms_join(context->paramids, subparamids);
3111 :
3112 37070 : return false; /* no more to do here */
3113 : }
3114 9369016 : return expression_tree_walker(node, finalize_primnode, context);
3115 : }
3116 :
3117 : /*
3118 : * finalize_agg_primnode: find all Aggref nodes in the given expression tree,
3119 : * and add IDs of all PARAM_EXEC params appearing within their aggregated
3120 : * arguments to the result set.
3121 : */
3122 : static bool
3123 16418 : finalize_agg_primnode(Node *node, finalize_primnode_context *context)
3124 : {
3125 16418 : if (node == NULL)
3126 2016 : return false;
3127 14402 : if (IsA(node, Aggref))
3128 : {
3129 1162 : Aggref *agg = (Aggref *) node;
3130 :
3131 : /* we should not consider the direct arguments, if any */
3132 1162 : finalize_primnode((Node *) agg->args, context);
3133 1162 : finalize_primnode((Node *) agg->aggfilter, context);
3134 1162 : return false; /* there can't be any Aggrefs below here */
3135 : }
3136 13240 : return expression_tree_walker(node, finalize_agg_primnode, context);
3137 : }
3138 :
3139 : /*
3140 : * SS_make_initplan_output_param - make a Param for an initPlan's output
3141 : *
3142 : * The plan is expected to return a scalar value of the given type/collation.
3143 : *
3144 : * Note that in some cases the initplan may not ever appear in the finished
3145 : * plan tree. If that happens, we'll have wasted a PARAM_EXEC slot, which
3146 : * is no big deal.
3147 : */
3148 : Param *
3149 446 : SS_make_initplan_output_param(PlannerInfo *root,
3150 : Oid resulttype, int32 resulttypmod,
3151 : Oid resultcollation)
3152 : {
3153 446 : return generate_new_exec_param(root, resulttype,
3154 : resulttypmod, resultcollation);
3155 : }
3156 :
3157 : /*
3158 : * SS_make_initplan_from_plan - given a plan tree, make it an InitPlan
3159 : *
3160 : * We build an EXPR_SUBLINK SubPlan node and put it into the initplan
3161 : * list for the outer query level. A Param that represents the initplan's
3162 : * output has already been assigned using SS_make_initplan_output_param.
3163 : */
3164 : void
3165 400 : SS_make_initplan_from_plan(PlannerInfo *root,
3166 : PlannerInfo *subroot, Plan *plan,
3167 : Param *prm)
3168 : {
3169 : SubPlan *node;
3170 :
3171 : /*
3172 : * Add the subplan and its PlannerInfo, as well as a dummy path entry, to
3173 : * the global lists. Ideally we'd save a real path, but right now our
3174 : * sole caller doesn't build a path that exactly matches the plan. Since
3175 : * we're not currently going to need the path for an initplan, it's not
3176 : * worth requiring construction of such a path.
3177 : */
3178 400 : root->glob->subplans = lappend(root->glob->subplans, plan);
3179 400 : root->glob->subpaths = lappend(root->glob->subpaths, NULL);
3180 400 : root->glob->subroots = lappend(root->glob->subroots, subroot);
3181 :
3182 : /*
3183 : * Create a SubPlan node and add it to the outer list of InitPlans. Note
3184 : * it has to appear after any other InitPlans it might depend on (see
3185 : * comments in ExecReScan).
3186 : */
3187 400 : node = makeNode(SubPlan);
3188 400 : node->subLinkType = EXPR_SUBLINK;
3189 400 : node->plan_id = list_length(root->glob->subplans);
3190 400 : node->plan_name = subroot->plan_name;
3191 400 : node->isInitPlan = true;
3192 400 : get_first_col_type(plan, &node->firstColType, &node->firstColTypmod,
3193 : &node->firstColCollation);
3194 400 : node->parallel_safe = plan->parallel_safe;
3195 400 : node->setParam = list_make1_int(prm->paramid);
3196 :
3197 400 : root->init_plans = lappend(root->init_plans, node);
3198 :
3199 : /*
3200 : * The node can't have any inputs (since it's an initplan), so the
3201 : * parParam and args lists remain empty.
3202 : */
3203 :
3204 : /* Set costs of SubPlan using info from the plan tree */
3205 400 : cost_subplan(subroot, node, plan);
3206 400 : }
3207 :
3208 : /*
3209 : * Get a string equivalent of a given subLinkType.
3210 : */
3211 : static const char *
3212 39012 : sublinktype_to_string(SubLinkType subLinkType)
3213 : {
3214 39012 : switch (subLinkType)
3215 : {
3216 2508 : case EXISTS_SUBLINK:
3217 2508 : return "exists";
3218 18 : case ALL_SUBLINK:
3219 18 : return "all";
3220 546 : case ANY_SUBLINK:
3221 546 : return "any";
3222 30 : case ROWCOMPARE_SUBLINK:
3223 30 : return "rowcompare";
3224 27590 : case EXPR_SUBLINK:
3225 27590 : return "expr";
3226 132 : case MULTIEXPR_SUBLINK:
3227 132 : return "multiexpr";
3228 8188 : case ARRAY_SUBLINK:
3229 8188 : return "array";
3230 0 : case CTE_SUBLINK:
3231 0 : return "cte";
3232 : }
3233 : Assert(false);
3234 0 : return "???";
3235 : }
|
