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 "miscadmin.h"
24 : #include "nodes/makefuncs.h"
25 : #include "nodes/nodeFuncs.h"
26 : #include "optimizer/clauses.h"
27 : #include "optimizer/cost.h"
28 : #include "optimizer/optimizer.h"
29 : #include "optimizer/paramassign.h"
30 : #include "optimizer/pathnode.h"
31 : #include "optimizer/planmain.h"
32 : #include "optimizer/planner.h"
33 : #include "optimizer/prep.h"
34 : #include "optimizer/subselect.h"
35 : #include "parser/parse_relation.h"
36 : #include "rewrite/rewriteManip.h"
37 : #include "utils/builtins.h"
38 : #include "utils/lsyscache.h"
39 : #include "utils/syscache.h"
40 :
41 :
42 : typedef struct convert_testexpr_context
43 : {
44 : PlannerInfo *root;
45 : List *subst_nodes; /* Nodes to substitute for Params */
46 : } convert_testexpr_context;
47 :
48 : typedef struct process_sublinks_context
49 : {
50 : PlannerInfo *root;
51 : bool isTopQual;
52 : } process_sublinks_context;
53 :
54 : typedef struct finalize_primnode_context
55 : {
56 : PlannerInfo *root;
57 : Bitmapset *paramids; /* Non-local PARAM_EXEC paramids found */
58 : } finalize_primnode_context;
59 :
60 : typedef struct inline_cte_walker_context
61 : {
62 : const char *ctename; /* name and relative level of target CTE */
63 : int levelsup;
64 : Query *ctequery; /* query to substitute */
65 : } inline_cte_walker_context;
66 :
67 :
68 : static Node *build_subplan(PlannerInfo *root, Plan *plan, Path *path,
69 : PlannerInfo *subroot, List *plan_params,
70 : SubLinkType subLinkType, int subLinkId,
71 : Node *testexpr, List *testexpr_paramids,
72 : bool unknownEqFalse);
73 : static List *generate_subquery_params(PlannerInfo *root, List *tlist,
74 : List **paramIds);
75 : static List *generate_subquery_vars(PlannerInfo *root, List *tlist,
76 : Index varno);
77 : static Node *convert_testexpr(PlannerInfo *root,
78 : Node *testexpr,
79 : List *subst_nodes);
80 : static Node *convert_testexpr_mutator(Node *node,
81 : convert_testexpr_context *context);
82 : static bool subplan_is_hashable(Plan *plan);
83 : static bool subpath_is_hashable(Path *path);
84 : static bool testexpr_is_hashable(Node *testexpr, List *param_ids);
85 : static bool test_opexpr_is_hashable(OpExpr *testexpr, List *param_ids);
86 : static bool hash_ok_operator(OpExpr *expr);
87 : static bool contain_dml(Node *node);
88 : static bool contain_dml_walker(Node *node, void *context);
89 : static bool contain_outer_selfref(Node *node);
90 : static bool contain_outer_selfref_walker(Node *node, Index *depth);
91 : static void inline_cte(PlannerInfo *root, CommonTableExpr *cte);
92 : static bool inline_cte_walker(Node *node, inline_cte_walker_context *context);
93 : static bool simplify_EXISTS_query(PlannerInfo *root, Query *query);
94 : static Query *convert_EXISTS_to_ANY(PlannerInfo *root, Query *subselect,
95 : Node **testexpr, List **paramIds);
96 : static Node *replace_correlation_vars_mutator(Node *node, PlannerInfo *root);
97 : static Node *process_sublinks_mutator(Node *node,
98 : process_sublinks_context *context);
99 : static Bitmapset *finalize_plan(PlannerInfo *root,
100 : Plan *plan,
101 : int gather_param,
102 : Bitmapset *valid_params,
103 : Bitmapset *scan_params);
104 : static bool finalize_primnode(Node *node, finalize_primnode_context *context);
105 : static bool finalize_agg_primnode(Node *node, finalize_primnode_context *context);
106 :
107 :
108 : /*
109 : * Get the datatype/typmod/collation of the first column of the plan's output.
110 : *
111 : * This information is stored for ARRAY_SUBLINK execution and for
112 : * exprType()/exprTypmod()/exprCollation(), which have no way to get at the
113 : * plan associated with a SubPlan node. We really only need the info for
114 : * EXPR_SUBLINK and ARRAY_SUBLINK subplans, but for consistency we save it
115 : * always.
116 : */
117 : static void
118 42476 : get_first_col_type(Plan *plan, Oid *coltype, int32 *coltypmod,
119 : Oid *colcollation)
120 : {
121 : /* In cases such as EXISTS, tlist might be empty; arbitrarily use VOID */
122 42476 : if (plan->targetlist)
123 : {
124 39846 : TargetEntry *tent = linitial_node(TargetEntry, plan->targetlist);
125 :
126 39846 : if (!tent->resjunk)
127 : {
128 39846 : *coltype = exprType((Node *) tent->expr);
129 39846 : *coltypmod = exprTypmod((Node *) tent->expr);
130 39846 : *colcollation = exprCollation((Node *) tent->expr);
131 39846 : return;
132 : }
133 : }
134 2630 : *coltype = VOIDOID;
135 2630 : *coltypmod = -1;
136 2630 : *colcollation = InvalidOid;
137 : }
138 :
139 : /*
140 : * Convert a SubLink (as created by the parser) into a SubPlan.
141 : *
142 : * We are given the SubLink's contained query, type, ID, and testexpr. We are
143 : * also told if this expression appears at top level of a WHERE/HAVING qual.
144 : *
145 : * Note: we assume that the testexpr has been AND/OR flattened (actually,
146 : * it's been through eval_const_expressions), but not converted to
147 : * implicit-AND form; and any SubLinks in it should already have been
148 : * converted to SubPlans. The subquery is as yet untouched, however.
149 : *
150 : * The result is whatever we need to substitute in place of the SubLink node
151 : * in the executable expression. If we're going to do the subplan as a
152 : * regular subplan, this will be the constructed SubPlan node. If we're going
153 : * to do the subplan as an InitPlan, the SubPlan node instead goes into
154 : * root->init_plans, and what we return here is an expression tree
155 : * representing the InitPlan's result: usually just a Param node representing
156 : * a single scalar result, but possibly a row comparison tree containing
157 : * multiple Param nodes, or for a MULTIEXPR subquery a simple NULL constant
158 : * (since the real output Params are elsewhere in the tree, and the MULTIEXPR
159 : * subquery itself is in a resjunk tlist entry whose value is uninteresting).
160 : */
161 : static Node *
162 38064 : make_subplan(PlannerInfo *root, Query *orig_subquery,
163 : SubLinkType subLinkType, int subLinkId,
164 : Node *testexpr, bool isTopQual)
165 : {
166 : Query *subquery;
167 38064 : bool simple_exists = false;
168 : double tuple_fraction;
169 : PlannerInfo *subroot;
170 : RelOptInfo *final_rel;
171 : Path *best_path;
172 : Plan *plan;
173 : List *plan_params;
174 : Node *result;
175 :
176 : /*
177 : * Copy the source Query node. This is a quick and dirty kluge to resolve
178 : * the fact that the parser can generate trees with multiple links to the
179 : * same sub-Query node, but the planner wants to scribble on the Query.
180 : * Try to clean this up when we do querytree redesign...
181 : */
182 38064 : subquery = copyObject(orig_subquery);
183 :
184 : /*
185 : * If it's an EXISTS subplan, we might be able to simplify it.
186 : */
187 38064 : if (subLinkType == EXISTS_SUBLINK)
188 2380 : simple_exists = simplify_EXISTS_query(root, subquery);
189 :
190 : /*
191 : * For an EXISTS subplan, tell lower-level planner to expect that only the
192 : * first tuple will be retrieved. For ALL and ANY subplans, we will be
193 : * able to stop evaluating if the test condition fails or matches, so very
194 : * often not all the tuples will be retrieved; for lack of a better idea,
195 : * specify 50% retrieval. For EXPR, MULTIEXPR, and ROWCOMPARE subplans,
196 : * use default behavior (we're only expecting one row out, anyway).
197 : *
198 : * NOTE: if you change these numbers, also change cost_subplan() in
199 : * path/costsize.c.
200 : *
201 : * XXX If an ANY subplan is uncorrelated, build_subplan may decide to hash
202 : * its output. In that case it would've been better to specify full
203 : * retrieval. At present, however, we can only check hashability after
204 : * we've made the subplan :-(. (Determining whether it'll fit in hash_mem
205 : * is the really hard part.) Therefore, we don't want to be too
206 : * optimistic about the percentage of tuples retrieved, for fear of
207 : * selecting a plan that's bad for the materialization case.
208 : */
209 38064 : if (subLinkType == EXISTS_SUBLINK)
210 2380 : tuple_fraction = 1.0; /* just like a LIMIT 1 */
211 35684 : else if (subLinkType == ALL_SUBLINK ||
212 : subLinkType == ANY_SUBLINK)
213 522 : tuple_fraction = 0.5; /* 50% */
214 : else
215 35162 : tuple_fraction = 0.0; /* default behavior */
216 :
217 : /* plan_params should not be in use in current query level */
218 : Assert(root->plan_params == NIL);
219 :
220 : /* Generate Paths for the subquery */
221 38064 : subroot = subquery_planner(root->glob, subquery, root, false,
222 : tuple_fraction, NULL);
223 :
224 : /* Isolate the params needed by this specific subplan */
225 38064 : plan_params = root->plan_params;
226 38064 : root->plan_params = NIL;
227 :
228 : /*
229 : * Select best Path and turn it into a Plan. At least for now, there
230 : * seems no reason to postpone doing that.
231 : */
232 38064 : final_rel = fetch_upper_rel(subroot, UPPERREL_FINAL, NULL);
233 38064 : best_path = get_cheapest_fractional_path(final_rel, tuple_fraction);
234 :
235 38064 : plan = create_plan(subroot, best_path);
236 :
237 : /* And convert to SubPlan or InitPlan format. */
238 38064 : result = build_subplan(root, plan, best_path,
239 : subroot, plan_params,
240 : subLinkType, subLinkId,
241 : testexpr, NIL, isTopQual);
242 :
243 : /*
244 : * If it's a correlated EXISTS with an unimportant targetlist, we might be
245 : * able to transform it to the equivalent of an IN and then implement it
246 : * by hashing. We don't have enough information yet to tell which way is
247 : * likely to be better (it depends on the expected number of executions of
248 : * the EXISTS qual, and we are much too early in planning the outer query
249 : * to be able to guess that). So we generate both plans, if possible, and
250 : * leave it to setrefs.c to decide which to use.
251 : */
252 38064 : if (simple_exists && IsA(result, SubPlan))
253 : {
254 : Node *newtestexpr;
255 : List *paramIds;
256 :
257 : /* Make a second copy of the original subquery */
258 2084 : subquery = copyObject(orig_subquery);
259 : /* and re-simplify */
260 2084 : simple_exists = simplify_EXISTS_query(root, subquery);
261 : Assert(simple_exists);
262 : /* See if it can be converted to an ANY query */
263 2084 : subquery = convert_EXISTS_to_ANY(root, subquery,
264 : &newtestexpr, ¶mIds);
265 2084 : if (subquery)
266 : {
267 : /* Generate Paths for the ANY subquery; we'll need all rows */
268 1630 : subroot = subquery_planner(root->glob, subquery, root, false, 0.0,
269 : NULL);
270 :
271 : /* Isolate the params needed by this specific subplan */
272 1630 : plan_params = root->plan_params;
273 1630 : root->plan_params = NIL;
274 :
275 : /* Select best Path */
276 1630 : final_rel = fetch_upper_rel(subroot, UPPERREL_FINAL, NULL);
277 1630 : best_path = final_rel->cheapest_total_path;
278 :
279 : /* Now we can check if it'll fit in hash_mem */
280 1630 : if (subpath_is_hashable(best_path))
281 : {
282 : SubPlan *hashplan;
283 : AlternativeSubPlan *asplan;
284 :
285 : /* OK, finish planning the ANY subquery */
286 1630 : plan = create_plan(subroot, best_path);
287 :
288 : /* ... and convert to SubPlan format */
289 1630 : hashplan = castNode(SubPlan,
290 : build_subplan(root, plan, best_path,
291 : subroot, plan_params,
292 : ANY_SUBLINK, 0,
293 : newtestexpr,
294 : paramIds,
295 : true));
296 : /* Check we got what we expected */
297 : Assert(hashplan->parParam == NIL);
298 : Assert(hashplan->useHashTable);
299 :
300 : /* Leave it to setrefs.c to decide which plan to use */
301 1630 : asplan = makeNode(AlternativeSubPlan);
302 1630 : asplan->subplans = list_make2(result, hashplan);
303 1630 : result = (Node *) asplan;
304 1630 : root->hasAlternativeSubPlans = true;
305 : }
306 : }
307 : }
308 :
309 38064 : return result;
310 : }
311 :
312 : /*
313 : * Build a SubPlan node given the raw inputs --- subroutine for make_subplan
314 : *
315 : * Returns either the SubPlan, or a replacement expression if we decide to
316 : * make it an InitPlan, as explained in the comments for make_subplan.
317 : */
318 : static Node *
319 39694 : build_subplan(PlannerInfo *root, Plan *plan, Path *path,
320 : PlannerInfo *subroot, List *plan_params,
321 : SubLinkType subLinkType, int subLinkId,
322 : Node *testexpr, List *testexpr_paramids,
323 : bool unknownEqFalse)
324 : {
325 : Node *result;
326 : SubPlan *splan;
327 : bool isInitPlan;
328 : ListCell *lc;
329 :
330 : /*
331 : * Initialize the SubPlan node. Note plan_id, plan_name, and cost fields
332 : * are set further down.
333 : */
334 39694 : splan = makeNode(SubPlan);
335 39694 : splan->subLinkType = subLinkType;
336 39694 : splan->testexpr = NULL;
337 39694 : splan->paramIds = NIL;
338 39694 : get_first_col_type(plan, &splan->firstColType, &splan->firstColTypmod,
339 : &splan->firstColCollation);
340 39694 : splan->useHashTable = false;
341 39694 : splan->unknownEqFalse = unknownEqFalse;
342 39694 : splan->parallel_safe = plan->parallel_safe;
343 39694 : splan->setParam = NIL;
344 39694 : splan->parParam = NIL;
345 39694 : splan->args = NIL;
346 :
347 : /*
348 : * Make parParam and args lists of param IDs and expressions that current
349 : * query level will pass to this child plan.
350 : */
351 83816 : foreach(lc, plan_params)
352 : {
353 44122 : PlannerParamItem *pitem = (PlannerParamItem *) lfirst(lc);
354 44122 : Node *arg = pitem->item;
355 :
356 : /*
357 : * The Var, PlaceHolderVar, Aggref, GroupingFunc, or ReturningExpr has
358 : * already been adjusted to have the correct varlevelsup, phlevelsup,
359 : * agglevelsup, or retlevelsup.
360 : *
361 : * If it's a PlaceHolderVar, Aggref, GroupingFunc, or ReturningExpr,
362 : * its arguments might contain SubLinks, which have not yet been
363 : * processed (see the comments for SS_replace_correlation_vars). Do
364 : * that now.
365 : */
366 44122 : if (IsA(arg, PlaceHolderVar) ||
367 44110 : IsA(arg, Aggref) ||
368 44058 : IsA(arg, GroupingFunc) ||
369 43994 : IsA(arg, ReturningExpr))
370 146 : arg = SS_process_sublinks(root, arg, false);
371 :
372 44122 : splan->parParam = lappend_int(splan->parParam, pitem->paramId);
373 44122 : splan->args = lappend(splan->args, arg);
374 : }
375 :
376 : /*
377 : * Un-correlated or undirect correlated plans of EXISTS, EXPR, ARRAY,
378 : * ROWCOMPARE, or MULTIEXPR types can be used as initPlans. For EXISTS,
379 : * EXPR, or ARRAY, we return a Param referring to the result of evaluating
380 : * the initPlan. For ROWCOMPARE, we must modify the testexpr tree to
381 : * contain PARAM_EXEC Params instead of the PARAM_SUBLINK Params emitted
382 : * by the parser, and then return that tree. For MULTIEXPR, we return a
383 : * null constant: the resjunk targetlist item containing the SubLink does
384 : * not need to return anything useful, since the referencing Params are
385 : * elsewhere.
386 : */
387 39694 : if (splan->parParam == NIL && subLinkType == EXISTS_SUBLINK)
388 264 : {
389 : Param *prm;
390 :
391 : Assert(testexpr == NULL);
392 264 : prm = generate_new_exec_param(root, BOOLOID, -1, InvalidOid);
393 264 : splan->setParam = list_make1_int(prm->paramid);
394 264 : isInitPlan = true;
395 264 : result = (Node *) prm;
396 : }
397 39430 : else if (splan->parParam == NIL && subLinkType == EXPR_SUBLINK)
398 9276 : {
399 9276 : TargetEntry *te = linitial(plan->targetlist);
400 : Param *prm;
401 :
402 : Assert(!te->resjunk);
403 : Assert(testexpr == NULL);
404 9276 : prm = generate_new_exec_param(root,
405 9276 : exprType((Node *) te->expr),
406 9276 : exprTypmod((Node *) te->expr),
407 9276 : exprCollation((Node *) te->expr));
408 9276 : splan->setParam = list_make1_int(prm->paramid);
409 9276 : isInitPlan = true;
410 9276 : result = (Node *) prm;
411 : }
412 30154 : else if (splan->parParam == NIL && subLinkType == ARRAY_SUBLINK)
413 116 : {
414 116 : TargetEntry *te = linitial(plan->targetlist);
415 : Oid arraytype;
416 : Param *prm;
417 :
418 : Assert(!te->resjunk);
419 : Assert(testexpr == NULL);
420 116 : arraytype = get_promoted_array_type(exprType((Node *) te->expr));
421 116 : if (!OidIsValid(arraytype))
422 0 : elog(ERROR, "could not find array type for datatype %s",
423 : format_type_be(exprType((Node *) te->expr)));
424 116 : prm = generate_new_exec_param(root,
425 : arraytype,
426 116 : exprTypmod((Node *) te->expr),
427 116 : exprCollation((Node *) te->expr));
428 116 : splan->setParam = list_make1_int(prm->paramid);
429 116 : isInitPlan = true;
430 116 : result = (Node *) prm;
431 : }
432 30038 : else if (splan->parParam == NIL && subLinkType == ROWCOMPARE_SUBLINK)
433 18 : {
434 : /* Adjust the Params */
435 : List *params;
436 :
437 : Assert(testexpr != NULL);
438 18 : params = generate_subquery_params(root,
439 : plan->targetlist,
440 : &splan->paramIds);
441 18 : result = convert_testexpr(root,
442 : testexpr,
443 : params);
444 18 : splan->setParam = list_copy(splan->paramIds);
445 18 : isInitPlan = true;
446 :
447 : /*
448 : * The executable expression is returned to become part of the outer
449 : * plan's expression tree; it is not kept in the initplan node.
450 : */
451 : }
452 30020 : else if (subLinkType == MULTIEXPR_SUBLINK)
453 : {
454 : /*
455 : * Whether it's an initplan or not, it needs to set a PARAM_EXEC Param
456 : * for each output column.
457 : */
458 : List *params;
459 :
460 : Assert(testexpr == NULL);
461 132 : params = generate_subquery_params(root,
462 : plan->targetlist,
463 : &splan->setParam);
464 :
465 : /*
466 : * Save the list of replacement Params in the n'th cell of
467 : * root->multiexpr_params; setrefs.c will use it to replace
468 : * PARAM_MULTIEXPR Params.
469 : */
470 264 : while (list_length(root->multiexpr_params) < subLinkId)
471 132 : root->multiexpr_params = lappend(root->multiexpr_params, NIL);
472 132 : lc = list_nth_cell(root->multiexpr_params, subLinkId - 1);
473 : Assert(lfirst(lc) == NIL);
474 132 : lfirst(lc) = params;
475 :
476 : /* It can be an initplan if there are no parParams. */
477 132 : if (splan->parParam == NIL)
478 : {
479 30 : isInitPlan = true;
480 30 : result = (Node *) makeNullConst(RECORDOID, -1, InvalidOid);
481 : }
482 : else
483 : {
484 102 : isInitPlan = false;
485 102 : result = (Node *) splan;
486 : }
487 : }
488 : else
489 : {
490 : /*
491 : * Adjust the Params in the testexpr, unless caller already took care
492 : * of it (as indicated by passing a list of Param IDs).
493 : */
494 29888 : if (testexpr && testexpr_paramids == NIL)
495 534 : {
496 : List *params;
497 :
498 534 : params = generate_subquery_params(root,
499 : plan->targetlist,
500 : &splan->paramIds);
501 534 : splan->testexpr = convert_testexpr(root,
502 : testexpr,
503 : params);
504 : }
505 : else
506 : {
507 29354 : splan->testexpr = testexpr;
508 29354 : splan->paramIds = testexpr_paramids;
509 : }
510 :
511 : /*
512 : * We can't convert subplans of ALL_SUBLINK or ANY_SUBLINK types to
513 : * initPlans, even when they are uncorrelated or undirect correlated,
514 : * because we need to scan the output of the subplan for each outer
515 : * tuple. But if it's a not-direct-correlated IN (= ANY) test, we
516 : * might be able to use a hashtable to avoid comparing all the tuples.
517 : */
518 29888 : if (subLinkType == ANY_SUBLINK &&
519 4182 : splan->parParam == NIL &&
520 4096 : subplan_is_hashable(plan) &&
521 2048 : testexpr_is_hashable(splan->testexpr, splan->paramIds))
522 2024 : splan->useHashTable = true;
523 :
524 : /*
525 : * Otherwise, we have the option to tack a Material node onto the top
526 : * of the subplan, to reduce the cost of reading it repeatedly. This
527 : * is pointless for a direct-correlated subplan, since we'd have to
528 : * recompute its results each time anyway. For uncorrelated/undirect
529 : * correlated subplans, we add Material unless the subplan's top plan
530 : * node would materialize its output anyway. Also, if enable_material
531 : * is false, then the user does not want us to materialize anything
532 : * unnecessarily, so we don't.
533 : */
534 27864 : else if (splan->parParam == NIL && enable_material &&
535 42 : !ExecMaterializesOutput(nodeTag(plan)))
536 42 : plan = materialize_finished_plan(plan);
537 :
538 29888 : result = (Node *) splan;
539 29888 : isInitPlan = false;
540 : }
541 :
542 : /*
543 : * Add the subplan, its path, and its PlannerInfo to the global lists.
544 : */
545 39694 : root->glob->subplans = lappend(root->glob->subplans, plan);
546 39694 : root->glob->subpaths = lappend(root->glob->subpaths, path);
547 39694 : root->glob->subroots = lappend(root->glob->subroots, subroot);
548 39694 : splan->plan_id = list_length(root->glob->subplans);
549 :
550 39694 : if (isInitPlan)
551 9704 : root->init_plans = lappend(root->init_plans, splan);
552 :
553 : /*
554 : * A parameterless subplan (not initplan) should be prepared to handle
555 : * REWIND efficiently. If it has direct parameters then there's no point
556 : * since it'll be reset on each scan anyway; and if it's an initplan then
557 : * there's no point since it won't get re-run without parameter changes
558 : * anyway. The input of a hashed subplan doesn't need REWIND either.
559 : */
560 39694 : if (splan->parParam == NIL && !isInitPlan && !splan->useHashTable)
561 42 : root->glob->rewindPlanIDs = bms_add_member(root->glob->rewindPlanIDs,
562 : splan->plan_id);
563 :
564 : /* Label the subplan for EXPLAIN purposes */
565 39694 : splan->plan_name = psprintf("%s %d",
566 : isInitPlan ? "InitPlan" : "SubPlan",
567 : splan->plan_id);
568 :
569 : /* Lastly, fill in the cost estimates for use later */
570 39694 : cost_subplan(root, splan, plan);
571 :
572 39694 : return result;
573 : }
574 :
575 : /*
576 : * generate_subquery_params: build a list of Params representing the output
577 : * columns of a sublink's sub-select, given the sub-select's targetlist.
578 : *
579 : * We also return an integer list of the paramids of the Params.
580 : */
581 : static List *
582 684 : generate_subquery_params(PlannerInfo *root, List *tlist, List **paramIds)
583 : {
584 : List *result;
585 : List *ids;
586 : ListCell *lc;
587 :
588 684 : result = ids = NIL;
589 1616 : foreach(lc, tlist)
590 : {
591 932 : TargetEntry *tent = (TargetEntry *) lfirst(lc);
592 : Param *param;
593 :
594 932 : if (tent->resjunk)
595 6 : continue;
596 :
597 926 : param = generate_new_exec_param(root,
598 926 : exprType((Node *) tent->expr),
599 926 : exprTypmod((Node *) tent->expr),
600 926 : exprCollation((Node *) tent->expr));
601 926 : result = lappend(result, param);
602 926 : ids = lappend_int(ids, param->paramid);
603 : }
604 :
605 684 : *paramIds = ids;
606 684 : return result;
607 : }
608 :
609 : /*
610 : * generate_subquery_vars: build a list of Vars representing the output
611 : * columns of a sublink's sub-select, given the sub-select's targetlist.
612 : * The Vars have the specified varno (RTE index).
613 : */
614 : static List *
615 1702 : generate_subquery_vars(PlannerInfo *root, List *tlist, Index varno)
616 : {
617 : List *result;
618 : ListCell *lc;
619 :
620 1702 : result = NIL;
621 3452 : foreach(lc, tlist)
622 : {
623 1750 : TargetEntry *tent = (TargetEntry *) lfirst(lc);
624 : Var *var;
625 :
626 1750 : if (tent->resjunk)
627 0 : continue;
628 :
629 1750 : var = makeVarFromTargetEntry(varno, tent);
630 1750 : result = lappend(result, var);
631 : }
632 :
633 1702 : return result;
634 : }
635 :
636 : /*
637 : * convert_testexpr: convert the testexpr given by the parser into
638 : * actually executable form. This entails replacing PARAM_SUBLINK Params
639 : * with Params or Vars representing the results of the sub-select. The
640 : * nodes to be substituted are passed in as the List result from
641 : * generate_subquery_params or generate_subquery_vars.
642 : */
643 : static Node *
644 2254 : convert_testexpr(PlannerInfo *root,
645 : Node *testexpr,
646 : List *subst_nodes)
647 : {
648 : convert_testexpr_context context;
649 :
650 2254 : context.root = root;
651 2254 : context.subst_nodes = subst_nodes;
652 2254 : return convert_testexpr_mutator(testexpr, &context);
653 : }
654 :
655 : static Node *
656 11122 : convert_testexpr_mutator(Node *node,
657 : convert_testexpr_context *context)
658 : {
659 11122 : if (node == NULL)
660 46 : return NULL;
661 11076 : if (IsA(node, Param))
662 : {
663 2426 : Param *param = (Param *) node;
664 :
665 2426 : if (param->paramkind == PARAM_SUBLINK)
666 : {
667 4852 : if (param->paramid <= 0 ||
668 2426 : param->paramid > list_length(context->subst_nodes))
669 0 : elog(ERROR, "unexpected PARAM_SUBLINK ID: %d", param->paramid);
670 :
671 : /*
672 : * We copy the list item to avoid having doubly-linked
673 : * substructure in the modified parse tree. This is probably
674 : * unnecessary when it's a Param, but be safe.
675 : */
676 2426 : return (Node *) copyObject(list_nth(context->subst_nodes,
677 : param->paramid - 1));
678 : }
679 : }
680 8650 : if (IsA(node, SubLink))
681 : {
682 : /*
683 : * If we come across a nested SubLink, it is neither necessary nor
684 : * correct to recurse into it: any PARAM_SUBLINKs we might find inside
685 : * belong to the inner SubLink not the outer. So just return it as-is.
686 : *
687 : * This reasoning depends on the assumption that nothing will pull
688 : * subexpressions into or out of the testexpr field of a SubLink, at
689 : * least not without replacing PARAM_SUBLINKs first. If we did want
690 : * to do that we'd need to rethink the parser-output representation
691 : * altogether, since currently PARAM_SUBLINKs are only unique per
692 : * SubLink not globally across the query. The whole point of
693 : * replacing them with Vars or PARAM_EXEC nodes is to make them
694 : * globally unique before they escape from the SubLink's testexpr.
695 : *
696 : * Note: this can't happen when called during SS_process_sublinks,
697 : * because that recursively processes inner SubLinks first. It can
698 : * happen when called from convert_ANY_sublink_to_join, though.
699 : */
700 12 : return node;
701 : }
702 8638 : return expression_tree_mutator(node, convert_testexpr_mutator, context);
703 : }
704 :
705 : /*
706 : * subplan_is_hashable: can we implement an ANY subplan by hashing?
707 : *
708 : * This is not responsible for checking whether the combining testexpr
709 : * is suitable for hashing. We only look at the subquery itself.
710 : */
711 : static bool
712 2048 : subplan_is_hashable(Plan *plan)
713 : {
714 : double subquery_size;
715 :
716 : /*
717 : * The estimated size of the subquery result must fit in hash_mem. (Note:
718 : * we use heap tuple overhead here even though the tuples will actually be
719 : * stored as MinimalTuples; this provides some fudge factor for hashtable
720 : * overhead.)
721 : */
722 2048 : subquery_size = plan->plan_rows *
723 2048 : (MAXALIGN(plan->plan_width) + MAXALIGN(SizeofHeapTupleHeader));
724 2048 : if (subquery_size > get_hash_memory_limit())
725 0 : return false;
726 :
727 2048 : return true;
728 : }
729 :
730 : /*
731 : * subpath_is_hashable: can we implement an ANY subplan by hashing?
732 : *
733 : * Identical to subplan_is_hashable, but work from a Path for the subplan.
734 : */
735 : static bool
736 1630 : subpath_is_hashable(Path *path)
737 : {
738 : double subquery_size;
739 :
740 : /*
741 : * The estimated size of the subquery result must fit in hash_mem. (Note:
742 : * we use heap tuple overhead here even though the tuples will actually be
743 : * stored as MinimalTuples; this provides some fudge factor for hashtable
744 : * overhead.)
745 : */
746 1630 : subquery_size = path->rows *
747 1630 : (MAXALIGN(path->pathtarget->width) + MAXALIGN(SizeofHeapTupleHeader));
748 1630 : if (subquery_size > get_hash_memory_limit())
749 0 : return false;
750 :
751 1630 : return true;
752 : }
753 :
754 : /*
755 : * testexpr_is_hashable: is an ANY SubLink's test expression hashable?
756 : *
757 : * To identify LHS vs RHS of the hash expression, we must be given the
758 : * list of output Param IDs of the SubLink's subquery.
759 : */
760 : static bool
761 2048 : testexpr_is_hashable(Node *testexpr, List *param_ids)
762 : {
763 : /*
764 : * The testexpr must be a single OpExpr, or an AND-clause containing only
765 : * OpExprs, each of which satisfy test_opexpr_is_hashable().
766 : */
767 2048 : if (testexpr && IsA(testexpr, OpExpr))
768 : {
769 1206 : if (test_opexpr_is_hashable((OpExpr *) testexpr, param_ids))
770 1182 : return true;
771 : }
772 842 : else if (is_andclause(testexpr))
773 : {
774 : ListCell *l;
775 :
776 2526 : foreach(l, ((BoolExpr *) testexpr)->args)
777 : {
778 1684 : Node *andarg = (Node *) lfirst(l);
779 :
780 1684 : if (!IsA(andarg, OpExpr))
781 0 : return false;
782 1684 : if (!test_opexpr_is_hashable((OpExpr *) andarg, param_ids))
783 0 : return false;
784 : }
785 842 : return true;
786 : }
787 :
788 24 : return false;
789 : }
790 :
791 : static bool
792 2890 : test_opexpr_is_hashable(OpExpr *testexpr, List *param_ids)
793 : {
794 : /*
795 : * The combining operator must be hashable and strict. The need for
796 : * hashability is obvious, since we want to use hashing. Without
797 : * strictness, behavior in the presence of nulls is too unpredictable. We
798 : * actually must assume even more than plain strictness: it can't yield
799 : * NULL for non-null inputs, either (see nodeSubplan.c). However, hash
800 : * indexes and hash joins assume that too.
801 : */
802 2890 : if (!hash_ok_operator(testexpr))
803 12 : return false;
804 :
805 : /*
806 : * The left and right inputs must belong to the outer and inner queries
807 : * respectively; hence Params that will be supplied by the subquery must
808 : * not appear in the LHS, and Vars of the outer query must not appear in
809 : * the RHS. (Ordinarily, this must be true because of the way that the
810 : * parser builds an ANY SubLink's testexpr ... but inlining of functions
811 : * could have changed the expression's structure, so we have to check.
812 : * Such cases do not occur often enough to be worth trying to optimize, so
813 : * we don't worry about trying to commute the clause or anything like
814 : * that; we just need to be sure not to build an invalid plan.)
815 : */
816 2878 : if (list_length(testexpr->args) != 2)
817 0 : return false;
818 2878 : if (contain_exec_param((Node *) linitial(testexpr->args), param_ids))
819 12 : return false;
820 2866 : if (contain_var_clause((Node *) lsecond(testexpr->args)))
821 0 : return false;
822 2866 : return true;
823 : }
824 :
825 : /*
826 : * Check expression is hashable + strict
827 : *
828 : * We could use op_hashjoinable() and op_strict(), but do it like this to
829 : * avoid a redundant cache lookup.
830 : */
831 : static bool
832 8918 : hash_ok_operator(OpExpr *expr)
833 : {
834 8918 : Oid opid = expr->opno;
835 :
836 : /* quick out if not a binary operator */
837 8918 : if (list_length(expr->args) != 2)
838 0 : return false;
839 8918 : if (opid == ARRAY_EQ_OP ||
840 : opid == RECORD_EQ_OP)
841 : {
842 : /* these are strict, but must check input type to ensure hashable */
843 12 : Node *leftarg = linitial(expr->args);
844 :
845 12 : return op_hashjoinable(opid, exprType(leftarg));
846 : }
847 : else
848 : {
849 : /* else must look up the operator properties */
850 : HeapTuple tup;
851 : Form_pg_operator optup;
852 :
853 8906 : tup = SearchSysCache1(OPEROID, ObjectIdGetDatum(opid));
854 8906 : if (!HeapTupleIsValid(tup))
855 0 : elog(ERROR, "cache lookup failed for operator %u", opid);
856 8906 : optup = (Form_pg_operator) GETSTRUCT(tup);
857 8906 : if (!optup->oprcanhash || !func_strict(optup->oprcode))
858 : {
859 866 : ReleaseSysCache(tup);
860 866 : return false;
861 : }
862 8040 : ReleaseSysCache(tup);
863 8040 : return true;
864 : }
865 : }
866 :
867 :
868 : /*
869 : * SS_process_ctes: process a query's WITH list
870 : *
871 : * Consider each CTE in the WITH list and either ignore it (if it's an
872 : * unreferenced SELECT), "inline" it to create a regular sub-SELECT-in-FROM,
873 : * or convert it to an initplan.
874 : *
875 : * A side effect is to fill in root->cte_plan_ids with a list that
876 : * parallels root->parse->cteList and provides the subplan ID for
877 : * each CTE's initplan, or a dummy ID (-1) if we didn't make an initplan.
878 : */
879 : void
880 2768 : SS_process_ctes(PlannerInfo *root)
881 : {
882 : ListCell *lc;
883 :
884 : Assert(root->cte_plan_ids == NIL);
885 :
886 6680 : foreach(lc, root->parse->cteList)
887 : {
888 3918 : CommonTableExpr *cte = (CommonTableExpr *) lfirst(lc);
889 3918 : CmdType cmdType = ((Query *) cte->ctequery)->commandType;
890 : Query *subquery;
891 : PlannerInfo *subroot;
892 : RelOptInfo *final_rel;
893 : Path *best_path;
894 : Plan *plan;
895 : SubPlan *splan;
896 : int paramid;
897 :
898 : /*
899 : * Ignore SELECT CTEs that are not actually referenced anywhere.
900 : */
901 3918 : if (cte->cterefcount == 0 && cmdType == CMD_SELECT)
902 : {
903 : /* Make a dummy entry in cte_plan_ids */
904 46 : root->cte_plan_ids = lappend_int(root->cte_plan_ids, -1);
905 1530 : continue;
906 : }
907 :
908 : /*
909 : * Consider inlining the CTE (creating RTE_SUBQUERY RTE(s)) instead of
910 : * implementing it as a separately-planned CTE.
911 : *
912 : * We cannot inline if any of these conditions hold:
913 : *
914 : * 1. The user said not to (the CTEMaterializeAlways option).
915 : *
916 : * 2. The CTE is recursive.
917 : *
918 : * 3. The CTE has side-effects; this includes either not being a plain
919 : * SELECT, or containing volatile functions. Inlining might change
920 : * the side-effects, which would be bad.
921 : *
922 : * 4. The CTE is multiply-referenced and contains a self-reference to
923 : * a recursive CTE outside itself. Inlining would result in multiple
924 : * recursive self-references, which we don't support.
925 : *
926 : * Otherwise, we have an option whether to inline or not. That should
927 : * always be a win if there's just a single reference, but if the CTE
928 : * is multiply-referenced then it's unclear: inlining adds duplicate
929 : * computations, but the ability to absorb restrictions from the outer
930 : * query level could outweigh that. We do not have nearly enough
931 : * information at this point to tell whether that's true, so we let
932 : * the user express a preference. Our default behavior is to inline
933 : * only singly-referenced CTEs, but a CTE marked CTEMaterializeNever
934 : * will be inlined even if multiply referenced.
935 : *
936 : * Note: we check for volatile functions last, because that's more
937 : * expensive than the other tests needed.
938 : */
939 3872 : if ((cte->ctematerialized == CTEMaterializeNever ||
940 3824 : (cte->ctematerialized == CTEMaterializeDefault &&
941 3622 : cte->cterefcount == 1)) &&
942 2678 : !cte->cterecursive &&
943 1576 : cmdType == CMD_SELECT &&
944 1576 : !contain_dml(cte->ctequery) &&
945 1568 : (cte->cterefcount <= 1 ||
946 36 : !contain_outer_selfref(cte->ctequery)) &&
947 1556 : !contain_volatile_functions(cte->ctequery))
948 : {
949 1484 : inline_cte(root, cte);
950 : /* Make a dummy entry in cte_plan_ids */
951 1484 : root->cte_plan_ids = lappend_int(root->cte_plan_ids, -1);
952 1484 : continue;
953 : }
954 :
955 : /*
956 : * Copy the source Query node. Probably not necessary, but let's keep
957 : * this similar to make_subplan.
958 : */
959 2388 : subquery = (Query *) copyObject(cte->ctequery);
960 :
961 : /* plan_params should not be in use in current query level */
962 : Assert(root->plan_params == NIL);
963 :
964 : /*
965 : * Generate Paths for the CTE query. Always plan for full retrieval
966 : * --- we don't have enough info to predict otherwise.
967 : */
968 2388 : subroot = subquery_planner(root->glob, subquery, root,
969 2388 : cte->cterecursive, 0.0, NULL);
970 :
971 : /*
972 : * Since the current query level doesn't yet contain any RTEs, it
973 : * should not be possible for the CTE to have requested parameters of
974 : * this level.
975 : */
976 2382 : if (root->plan_params)
977 0 : elog(ERROR, "unexpected outer reference in CTE query");
978 :
979 : /*
980 : * Select best Path and turn it into a Plan. At least for now, there
981 : * seems no reason to postpone doing that.
982 : */
983 2382 : final_rel = fetch_upper_rel(subroot, UPPERREL_FINAL, NULL);
984 2382 : best_path = final_rel->cheapest_total_path;
985 :
986 2382 : plan = create_plan(subroot, best_path);
987 :
988 : /*
989 : * Make a SubPlan node for it. This is just enough unlike
990 : * build_subplan that we can't share code.
991 : *
992 : * Note plan_id, plan_name, and cost fields are set further down.
993 : */
994 2382 : splan = makeNode(SubPlan);
995 2382 : splan->subLinkType = CTE_SUBLINK;
996 2382 : splan->testexpr = NULL;
997 2382 : splan->paramIds = NIL;
998 2382 : get_first_col_type(plan, &splan->firstColType, &splan->firstColTypmod,
999 : &splan->firstColCollation);
1000 2382 : splan->useHashTable = false;
1001 2382 : splan->unknownEqFalse = false;
1002 :
1003 : /*
1004 : * CTE scans are not considered for parallelism (cf
1005 : * set_rel_consider_parallel).
1006 : */
1007 2382 : splan->parallel_safe = false;
1008 2382 : splan->setParam = NIL;
1009 2382 : splan->parParam = NIL;
1010 2382 : splan->args = NIL;
1011 :
1012 : /*
1013 : * The node can't have any inputs (since it's an initplan), so the
1014 : * parParam and args lists remain empty. (It could contain references
1015 : * to earlier CTEs' output param IDs, but CTE outputs are not
1016 : * propagated via the args list.)
1017 : */
1018 :
1019 : /*
1020 : * Assign a param ID to represent the CTE's output. No ordinary
1021 : * "evaluation" of this param slot ever happens, but we use the param
1022 : * ID for setParam/chgParam signaling just as if the CTE plan were
1023 : * returning a simple scalar output. (Also, the executor abuses the
1024 : * ParamExecData slot for this param ID for communication among
1025 : * multiple CteScan nodes that might be scanning this CTE.)
1026 : */
1027 2382 : paramid = assign_special_exec_param(root);
1028 2382 : splan->setParam = list_make1_int(paramid);
1029 :
1030 : /*
1031 : * Add the subplan, its path, and its PlannerInfo to the global lists.
1032 : */
1033 2382 : root->glob->subplans = lappend(root->glob->subplans, plan);
1034 2382 : root->glob->subpaths = lappend(root->glob->subpaths, best_path);
1035 2382 : root->glob->subroots = lappend(root->glob->subroots, subroot);
1036 2382 : splan->plan_id = list_length(root->glob->subplans);
1037 :
1038 2382 : root->init_plans = lappend(root->init_plans, splan);
1039 :
1040 2382 : root->cte_plan_ids = lappend_int(root->cte_plan_ids, splan->plan_id);
1041 :
1042 : /* Label the subplan for EXPLAIN purposes */
1043 2382 : splan->plan_name = psprintf("CTE %s", cte->ctename);
1044 :
1045 : /* Lastly, fill in the cost estimates for use later */
1046 2382 : cost_subplan(root, splan, plan);
1047 : }
1048 2762 : }
1049 :
1050 : /*
1051 : * contain_dml: is any subquery not a plain SELECT?
1052 : *
1053 : * We reject SELECT FOR UPDATE/SHARE as well as INSERT etc.
1054 : */
1055 : static bool
1056 1576 : contain_dml(Node *node)
1057 : {
1058 1576 : return contain_dml_walker(node, NULL);
1059 : }
1060 :
1061 : static bool
1062 106508 : contain_dml_walker(Node *node, void *context)
1063 : {
1064 106508 : if (node == NULL)
1065 36240 : return false;
1066 70268 : if (IsA(node, Query))
1067 : {
1068 2848 : Query *query = (Query *) node;
1069 :
1070 2848 : if (query->commandType != CMD_SELECT ||
1071 2848 : query->rowMarks != NIL)
1072 8 : return true;
1073 :
1074 2840 : return query_tree_walker(query, contain_dml_walker, context, 0);
1075 : }
1076 67420 : return expression_tree_walker(node, contain_dml_walker, context);
1077 : }
1078 :
1079 : /*
1080 : * contain_outer_selfref: is there an external recursive self-reference?
1081 : */
1082 : static bool
1083 36 : contain_outer_selfref(Node *node)
1084 : {
1085 36 : Index depth = 0;
1086 :
1087 : /*
1088 : * We should be starting with a Query, so that depth will be 1 while
1089 : * examining its immediate contents.
1090 : */
1091 : Assert(IsA(node, Query));
1092 :
1093 36 : return contain_outer_selfref_walker(node, &depth);
1094 : }
1095 :
1096 : static bool
1097 810 : contain_outer_selfref_walker(Node *node, Index *depth)
1098 : {
1099 810 : if (node == NULL)
1100 486 : return false;
1101 324 : if (IsA(node, RangeTblEntry))
1102 : {
1103 30 : RangeTblEntry *rte = (RangeTblEntry *) node;
1104 :
1105 : /*
1106 : * Check for a self-reference to a CTE that's above the Query that our
1107 : * search started at.
1108 : */
1109 30 : if (rte->rtekind == RTE_CTE &&
1110 12 : rte->self_reference &&
1111 12 : rte->ctelevelsup >= *depth)
1112 12 : return true;
1113 18 : return false; /* allow range_table_walker to continue */
1114 : }
1115 294 : if (IsA(node, Query))
1116 : {
1117 : /* Recurse into subquery, tracking nesting depth properly */
1118 42 : Query *query = (Query *) node;
1119 : bool result;
1120 :
1121 42 : (*depth)++;
1122 :
1123 42 : result = query_tree_walker(query, contain_outer_selfref_walker,
1124 : depth, QTW_EXAMINE_RTES_BEFORE);
1125 :
1126 42 : (*depth)--;
1127 :
1128 42 : return result;
1129 : }
1130 252 : return expression_tree_walker(node, contain_outer_selfref_walker, depth);
1131 : }
1132 :
1133 : /*
1134 : * inline_cte: convert RTE_CTE references to given CTE into RTE_SUBQUERYs
1135 : */
1136 : static void
1137 1484 : inline_cte(PlannerInfo *root, CommonTableExpr *cte)
1138 : {
1139 : struct inline_cte_walker_context context;
1140 :
1141 1484 : context.ctename = cte->ctename;
1142 : /* Start at levelsup = -1 because we'll immediately increment it */
1143 1484 : context.levelsup = -1;
1144 1484 : context.ctequery = castNode(Query, cte->ctequery);
1145 :
1146 1484 : (void) inline_cte_walker((Node *) root->parse, &context);
1147 1484 : }
1148 :
1149 : static bool
1150 553596 : inline_cte_walker(Node *node, inline_cte_walker_context *context)
1151 : {
1152 553596 : if (node == NULL)
1153 147426 : return false;
1154 406170 : if (IsA(node, Query))
1155 : {
1156 10790 : Query *query = (Query *) node;
1157 :
1158 10790 : context->levelsup++;
1159 :
1160 : /*
1161 : * Visit the query's RTE nodes after their contents; otherwise
1162 : * query_tree_walker would descend into the newly inlined CTE query,
1163 : * which we don't want.
1164 : */
1165 10790 : (void) query_tree_walker(query, inline_cte_walker, context,
1166 : QTW_EXAMINE_RTES_AFTER);
1167 :
1168 10790 : context->levelsup--;
1169 :
1170 10790 : return false;
1171 : }
1172 395380 : else if (IsA(node, RangeTblEntry))
1173 : {
1174 20368 : RangeTblEntry *rte = (RangeTblEntry *) node;
1175 :
1176 20368 : if (rte->rtekind == RTE_CTE &&
1177 6018 : strcmp(rte->ctename, context->ctename) == 0 &&
1178 1514 : rte->ctelevelsup == context->levelsup)
1179 : {
1180 : /*
1181 : * Found a reference to replace. Generate a copy of the CTE query
1182 : * with appropriate level adjustment for outer references (e.g.,
1183 : * to other CTEs).
1184 : */
1185 1508 : Query *newquery = copyObject(context->ctequery);
1186 :
1187 1508 : if (context->levelsup > 0)
1188 946 : IncrementVarSublevelsUp((Node *) newquery, context->levelsup, 1);
1189 :
1190 : /*
1191 : * Convert the RTE_CTE RTE into a RTE_SUBQUERY.
1192 : *
1193 : * Historically, a FOR UPDATE clause has been treated as extending
1194 : * into views and subqueries, but not into CTEs. We preserve this
1195 : * distinction by not trying to push rowmarks into the new
1196 : * subquery.
1197 : */
1198 1508 : rte->rtekind = RTE_SUBQUERY;
1199 1508 : rte->subquery = newquery;
1200 1508 : rte->security_barrier = false;
1201 :
1202 : /* Zero out CTE-specific fields */
1203 1508 : rte->ctename = NULL;
1204 1508 : rte->ctelevelsup = 0;
1205 1508 : rte->self_reference = false;
1206 1508 : rte->coltypes = NIL;
1207 1508 : rte->coltypmods = NIL;
1208 1508 : rte->colcollations = NIL;
1209 : }
1210 :
1211 20368 : return false;
1212 : }
1213 :
1214 375012 : return expression_tree_walker(node, inline_cte_walker, context);
1215 : }
1216 :
1217 :
1218 : /*
1219 : * convert_ANY_sublink_to_join: try to convert an ANY SubLink to a join
1220 : *
1221 : * The caller has found an ANY SubLink at the top level of one of the query's
1222 : * qual clauses, but has not checked the properties of the SubLink further.
1223 : * Decide whether it is appropriate to process this SubLink in join style.
1224 : * If so, form a JoinExpr and return it. Return NULL if the SubLink cannot
1225 : * be converted to a join.
1226 : *
1227 : * The only non-obvious input parameter is available_rels: this is the set
1228 : * of query rels that can safely be referenced in the sublink expression.
1229 : * (We must restrict this to avoid changing the semantics when a sublink
1230 : * is present in an outer join's ON qual.) The conversion must fail if
1231 : * the converted qual would reference any but these parent-query relids.
1232 : *
1233 : * On success, the returned JoinExpr has larg = NULL and rarg = the jointree
1234 : * item representing the pulled-up subquery. The caller must set larg to
1235 : * represent the relation(s) on the lefthand side of the new join, and insert
1236 : * the JoinExpr into the upper query's jointree at an appropriate place
1237 : * (typically, where the lefthand relation(s) had been). Note that the
1238 : * passed-in SubLink must also be removed from its original position in the
1239 : * query quals, since the quals of the returned JoinExpr replace it.
1240 : * (Notionally, we replace the SubLink with a constant TRUE, then elide the
1241 : * redundant constant from the qual.)
1242 : *
1243 : * On success, the caller is also responsible for recursively applying
1244 : * pull_up_sublinks processing to the rarg and quals of the returned JoinExpr.
1245 : * (On failure, there is no need to do anything, since pull_up_sublinks will
1246 : * be applied when we recursively plan the sub-select.)
1247 : *
1248 : * Side effects of a successful conversion include adding the SubLink's
1249 : * subselect to the query's rangetable, so that it can be referenced in
1250 : * the JoinExpr's rarg.
1251 : */
1252 : JoinExpr *
1253 1814 : convert_ANY_sublink_to_join(PlannerInfo *root, SubLink *sublink,
1254 : Relids available_rels)
1255 : {
1256 : JoinExpr *result;
1257 1814 : Query *parse = root->parse;
1258 1814 : Query *subselect = (Query *) sublink->subselect;
1259 : Relids upper_varnos;
1260 : int rtindex;
1261 : ParseNamespaceItem *nsitem;
1262 : RangeTblEntry *rte;
1263 : RangeTblRef *rtr;
1264 : List *subquery_vars;
1265 : Node *quals;
1266 : ParseState *pstate;
1267 : Relids sub_ref_outer_relids;
1268 : bool use_lateral;
1269 :
1270 : Assert(sublink->subLinkType == ANY_SUBLINK);
1271 :
1272 : /*
1273 : * If the sub-select contains any Vars of the parent query, we treat it as
1274 : * LATERAL. (Vars from higher levels don't matter here.)
1275 : */
1276 1814 : sub_ref_outer_relids = pull_varnos_of_level(NULL, (Node *) subselect, 1);
1277 1814 : use_lateral = !bms_is_empty(sub_ref_outer_relids);
1278 :
1279 : /*
1280 : * Can't convert if the sub-select contains parent-level Vars of relations
1281 : * not in available_rels.
1282 : */
1283 1814 : if (!bms_is_subset(sub_ref_outer_relids, available_rels))
1284 12 : return NULL;
1285 :
1286 : /*
1287 : * The test expression must contain some Vars of the parent query, else
1288 : * it's not gonna be a join. (Note that it won't have Vars referring to
1289 : * the subquery, rather Params.)
1290 : */
1291 1802 : upper_varnos = pull_varnos(root, sublink->testexpr);
1292 1802 : if (bms_is_empty(upper_varnos))
1293 12 : return NULL;
1294 :
1295 : /*
1296 : * However, it can't refer to anything outside available_rels.
1297 : */
1298 1790 : if (!bms_is_subset(upper_varnos, available_rels))
1299 24 : return NULL;
1300 :
1301 : /*
1302 : * The combining operators and left-hand expressions mustn't be volatile.
1303 : */
1304 1766 : if (contain_volatile_functions(sublink->testexpr))
1305 64 : return NULL;
1306 :
1307 : /* Create a dummy ParseState for addRangeTableEntryForSubquery */
1308 1702 : pstate = make_parsestate(NULL);
1309 :
1310 : /*
1311 : * Okay, pull up the sub-select into upper range table.
1312 : *
1313 : * We rely here on the assumption that the outer query has no references
1314 : * to the inner (necessarily true, other than the Vars that we build
1315 : * below). Therefore this is a lot easier than what pull_up_subqueries has
1316 : * to go through.
1317 : */
1318 1702 : nsitem = addRangeTableEntryForSubquery(pstate,
1319 : subselect,
1320 : makeAlias("ANY_subquery", NIL),
1321 : use_lateral,
1322 : false);
1323 1702 : rte = nsitem->p_rte;
1324 1702 : parse->rtable = lappend(parse->rtable, rte);
1325 1702 : rtindex = list_length(parse->rtable);
1326 :
1327 : /*
1328 : * Form a RangeTblRef for the pulled-up sub-select.
1329 : */
1330 1702 : rtr = makeNode(RangeTblRef);
1331 1702 : rtr->rtindex = rtindex;
1332 :
1333 : /*
1334 : * Build a list of Vars representing the subselect outputs.
1335 : */
1336 1702 : subquery_vars = generate_subquery_vars(root,
1337 : subselect->targetList,
1338 : rtindex);
1339 :
1340 : /*
1341 : * Build the new join's qual expression, replacing Params with these Vars.
1342 : */
1343 1702 : quals = convert_testexpr(root, sublink->testexpr, subquery_vars);
1344 :
1345 : /*
1346 : * And finally, build the JoinExpr node.
1347 : */
1348 1702 : result = makeNode(JoinExpr);
1349 1702 : result->jointype = JOIN_SEMI;
1350 1702 : result->isNatural = false;
1351 1702 : result->larg = NULL; /* caller must fill this in */
1352 1702 : result->rarg = (Node *) rtr;
1353 1702 : result->usingClause = NIL;
1354 1702 : result->join_using_alias = NULL;
1355 1702 : result->quals = quals;
1356 1702 : result->alias = NULL;
1357 1702 : result->rtindex = 0; /* we don't need an RTE for it */
1358 :
1359 1702 : return result;
1360 : }
1361 :
1362 : /*
1363 : * convert_EXISTS_sublink_to_join: try to convert an EXISTS SubLink to a join
1364 : *
1365 : * The API of this function is identical to convert_ANY_sublink_to_join's,
1366 : * except that we also support the case where the caller has found NOT EXISTS,
1367 : * so we need an additional input parameter "under_not".
1368 : */
1369 : JoinExpr *
1370 3594 : convert_EXISTS_sublink_to_join(PlannerInfo *root, SubLink *sublink,
1371 : bool under_not, Relids available_rels)
1372 : {
1373 : JoinExpr *result;
1374 3594 : Query *parse = root->parse;
1375 3594 : Query *subselect = (Query *) sublink->subselect;
1376 : Node *whereClause;
1377 : int rtoffset;
1378 : int varno;
1379 : Relids clause_varnos;
1380 : Relids upper_varnos;
1381 :
1382 : Assert(sublink->subLinkType == EXISTS_SUBLINK);
1383 :
1384 : /*
1385 : * Can't flatten if it contains WITH. (We could arrange to pull up the
1386 : * WITH into the parent query's cteList, but that risks changing the
1387 : * semantics, since a WITH ought to be executed once per associated query
1388 : * call.) Note that convert_ANY_sublink_to_join doesn't have to reject
1389 : * this case, since it just produces a subquery RTE that doesn't have to
1390 : * get flattened into the parent query.
1391 : */
1392 3594 : if (subselect->cteList)
1393 0 : return NULL;
1394 :
1395 : /*
1396 : * Copy the subquery so we can modify it safely (see comments in
1397 : * make_subplan).
1398 : */
1399 3594 : subselect = copyObject(subselect);
1400 :
1401 : /*
1402 : * See if the subquery can be simplified based on the knowledge that it's
1403 : * being used in EXISTS(). If we aren't able to get rid of its
1404 : * targetlist, we have to fail, because the pullup operation leaves us
1405 : * with noplace to evaluate the targetlist.
1406 : */
1407 3594 : if (!simplify_EXISTS_query(root, subselect))
1408 32 : return NULL;
1409 :
1410 : /*
1411 : * Separate out the WHERE clause. (We could theoretically also remove
1412 : * top-level plain JOIN/ON clauses, but it's probably not worth the
1413 : * trouble.)
1414 : */
1415 3562 : whereClause = subselect->jointree->quals;
1416 3562 : subselect->jointree->quals = NULL;
1417 :
1418 : /*
1419 : * The rest of the sub-select must not refer to any Vars of the parent
1420 : * query. (Vars of higher levels should be okay, though.)
1421 : */
1422 3562 : if (contain_vars_of_level((Node *) subselect, 1))
1423 0 : return NULL;
1424 :
1425 : /*
1426 : * On the other hand, the WHERE clause must contain some Vars of the
1427 : * parent query, else it's not gonna be a join.
1428 : */
1429 3562 : if (!contain_vars_of_level(whereClause, 1))
1430 98 : return NULL;
1431 :
1432 : /*
1433 : * We don't risk optimizing if the WHERE clause is volatile, either.
1434 : */
1435 3464 : if (contain_volatile_functions(whereClause))
1436 0 : return NULL;
1437 :
1438 : /*
1439 : * The subquery must have a nonempty jointree, but we can make it so.
1440 : */
1441 3464 : replace_empty_jointree(subselect);
1442 :
1443 : /*
1444 : * Prepare to pull up the sub-select into top range table.
1445 : *
1446 : * We rely here on the assumption that the outer query has no references
1447 : * to the inner (necessarily true). Therefore this is a lot easier than
1448 : * what pull_up_subqueries has to go through.
1449 : *
1450 : * In fact, it's even easier than what convert_ANY_sublink_to_join has to
1451 : * do. The machinations of simplify_EXISTS_query ensured that there is
1452 : * nothing interesting in the subquery except an rtable and jointree, and
1453 : * even the jointree FromExpr no longer has quals. So we can just append
1454 : * the rtable to our own and use the FromExpr in our jointree. But first,
1455 : * adjust all level-zero varnos in the subquery to account for the rtable
1456 : * merger.
1457 : */
1458 3464 : rtoffset = list_length(parse->rtable);
1459 3464 : OffsetVarNodes((Node *) subselect, rtoffset, 0);
1460 3464 : OffsetVarNodes(whereClause, rtoffset, 0);
1461 :
1462 : /*
1463 : * Upper-level vars in subquery will now be one level closer to their
1464 : * parent than before; in particular, anything that had been level 1
1465 : * becomes level zero.
1466 : */
1467 3464 : IncrementVarSublevelsUp((Node *) subselect, -1, 1);
1468 3464 : IncrementVarSublevelsUp(whereClause, -1, 1);
1469 :
1470 : /*
1471 : * Now that the WHERE clause is adjusted to match the parent query
1472 : * environment, we can easily identify all the level-zero rels it uses.
1473 : * The ones <= rtoffset belong to the upper query; the ones > rtoffset do
1474 : * not.
1475 : */
1476 3464 : clause_varnos = pull_varnos(root, whereClause);
1477 3464 : upper_varnos = NULL;
1478 3464 : varno = -1;
1479 10416 : while ((varno = bms_next_member(clause_varnos, varno)) >= 0)
1480 : {
1481 6952 : if (varno <= rtoffset)
1482 3488 : upper_varnos = bms_add_member(upper_varnos, varno);
1483 : }
1484 3464 : bms_free(clause_varnos);
1485 : Assert(!bms_is_empty(upper_varnos));
1486 :
1487 : /*
1488 : * Now that we've got the set of upper-level varnos, we can make the last
1489 : * check: only available_rels can be referenced.
1490 : */
1491 3464 : if (!bms_is_subset(upper_varnos, available_rels))
1492 32 : return NULL;
1493 :
1494 : /*
1495 : * Now we can attach the modified subquery rtable to the parent. This also
1496 : * adds subquery's RTEPermissionInfos into the upper query.
1497 : */
1498 3432 : CombineRangeTables(&parse->rtable, &parse->rteperminfos,
1499 : subselect->rtable, subselect->rteperminfos);
1500 :
1501 : /*
1502 : * And finally, build the JoinExpr node.
1503 : */
1504 3432 : result = makeNode(JoinExpr);
1505 3432 : result->jointype = under_not ? JOIN_ANTI : JOIN_SEMI;
1506 3432 : result->isNatural = false;
1507 3432 : result->larg = NULL; /* caller must fill this in */
1508 : /* flatten out the FromExpr node if it's useless */
1509 3432 : if (list_length(subselect->jointree->fromlist) == 1)
1510 3414 : result->rarg = (Node *) linitial(subselect->jointree->fromlist);
1511 : else
1512 18 : result->rarg = (Node *) subselect->jointree;
1513 3432 : result->usingClause = NIL;
1514 3432 : result->join_using_alias = NULL;
1515 3432 : result->quals = whereClause;
1516 3432 : result->alias = NULL;
1517 3432 : result->rtindex = 0; /* we don't need an RTE for it */
1518 :
1519 3432 : return result;
1520 : }
1521 :
1522 : /*
1523 : * simplify_EXISTS_query: remove any useless stuff in an EXISTS's subquery
1524 : *
1525 : * The only thing that matters about an EXISTS query is whether it returns
1526 : * zero or more than zero rows. Therefore, we can remove certain SQL features
1527 : * that won't affect that. The only part that is really likely to matter in
1528 : * typical usage is simplifying the targetlist: it's a common habit to write
1529 : * "SELECT * FROM" even though there is no need to evaluate any columns.
1530 : *
1531 : * Note: by suppressing the targetlist we could cause an observable behavioral
1532 : * change, namely that any errors that might occur in evaluating the tlist
1533 : * won't occur, nor will other side-effects of volatile functions. This seems
1534 : * unlikely to bother anyone in practice.
1535 : *
1536 : * Returns true if was able to discard the targetlist, else false.
1537 : */
1538 : static bool
1539 8058 : simplify_EXISTS_query(PlannerInfo *root, Query *query)
1540 : {
1541 : ListCell *lc;
1542 :
1543 : /*
1544 : * We don't try to simplify at all if the query uses set operations,
1545 : * aggregates, grouping sets, SRFs, modifying CTEs, HAVING, OFFSET, or FOR
1546 : * UPDATE/SHARE; none of these seem likely in normal usage and their
1547 : * possible effects are complex. (Note: we could ignore an "OFFSET 0"
1548 : * clause, but that traditionally is used as an optimization fence, so we
1549 : * don't.)
1550 : */
1551 8058 : if (query->commandType != CMD_SELECT ||
1552 8058 : query->setOperations ||
1553 8058 : query->hasAggs ||
1554 8058 : query->groupingSets ||
1555 8058 : query->hasWindowFuncs ||
1556 8058 : query->hasTargetSRFs ||
1557 8058 : query->hasModifyingCTE ||
1558 8058 : query->havingQual ||
1559 8058 : query->limitOffset ||
1560 8034 : query->rowMarks)
1561 52 : return false;
1562 :
1563 : /*
1564 : * LIMIT with a constant positive (or NULL) value doesn't affect the
1565 : * semantics of EXISTS, so let's ignore such clauses. This is worth doing
1566 : * because people accustomed to certain other DBMSes may be in the habit
1567 : * of writing EXISTS(SELECT ... LIMIT 1) as an optimization. If there's a
1568 : * LIMIT with anything else as argument, though, we can't simplify.
1569 : */
1570 8006 : if (query->limitCount)
1571 : {
1572 : /*
1573 : * The LIMIT clause has not yet been through eval_const_expressions,
1574 : * so we have to apply that here. It might seem like this is a waste
1575 : * of cycles, since the only case plausibly worth worrying about is
1576 : * "LIMIT 1" ... but what we'll actually see is "LIMIT int8(1::int4)",
1577 : * so we have to fold constants or we're not going to recognize it.
1578 : */
1579 24 : Node *node = eval_const_expressions(root, query->limitCount);
1580 : Const *limit;
1581 :
1582 : /* Might as well update the query if we simplified the clause. */
1583 24 : query->limitCount = node;
1584 :
1585 24 : if (!IsA(node, Const))
1586 0 : return false;
1587 :
1588 24 : limit = (Const *) node;
1589 : Assert(limit->consttype == INT8OID);
1590 24 : if (!limit->constisnull && DatumGetInt64(limit->constvalue) <= 0)
1591 12 : return false;
1592 :
1593 : /* Whether or not the targetlist is safe, we can drop the LIMIT. */
1594 12 : query->limitCount = NULL;
1595 : }
1596 :
1597 : /*
1598 : * Otherwise, we can throw away the targetlist, as well as any GROUP,
1599 : * WINDOW, DISTINCT, and ORDER BY clauses; none of those clauses will
1600 : * change a nonzero-rows result to zero rows or vice versa. (Furthermore,
1601 : * since our parsetree representation of these clauses depends on the
1602 : * targetlist, we'd better throw them away if we drop the targetlist.)
1603 : */
1604 7994 : query->targetList = NIL;
1605 7994 : query->groupClause = NIL;
1606 7994 : query->windowClause = NIL;
1607 7994 : query->distinctClause = NIL;
1608 7994 : query->sortClause = NIL;
1609 7994 : query->hasDistinctOn = false;
1610 :
1611 : /*
1612 : * Since we have thrown away the GROUP BY clauses, we'd better remove the
1613 : * RTE_GROUP RTE and clear the hasGroupRTE flag.
1614 : */
1615 16276 : foreach(lc, query->rtable)
1616 : {
1617 8288 : RangeTblEntry *rte = lfirst_node(RangeTblEntry, lc);
1618 :
1619 : /*
1620 : * Remove the RTE_GROUP RTE and clear the hasGroupRTE flag. (Since
1621 : * we'll exit the foreach loop immediately, we don't bother with
1622 : * foreach_delete_current.)
1623 : */
1624 8288 : if (rte->rtekind == RTE_GROUP)
1625 : {
1626 : Assert(query->hasGroupRTE);
1627 6 : query->rtable = list_delete_cell(query->rtable, lc);
1628 6 : query->hasGroupRTE = false;
1629 6 : break;
1630 : }
1631 : }
1632 :
1633 7994 : return true;
1634 : }
1635 :
1636 : /*
1637 : * convert_EXISTS_to_ANY: try to convert EXISTS to a hashable ANY sublink
1638 : *
1639 : * The subselect is expected to be a fresh copy that we can munge up,
1640 : * and to have been successfully passed through simplify_EXISTS_query.
1641 : *
1642 : * On success, the modified subselect is returned, and we store a suitable
1643 : * upper-level test expression at *testexpr, plus a list of the subselect's
1644 : * output Params at *paramIds. (The test expression is already Param-ified
1645 : * and hence need not go through convert_testexpr, which is why we have to
1646 : * deal with the Param IDs specially.)
1647 : *
1648 : * On failure, returns NULL.
1649 : */
1650 : static Query *
1651 2084 : convert_EXISTS_to_ANY(PlannerInfo *root, Query *subselect,
1652 : Node **testexpr, List **paramIds)
1653 : {
1654 : Node *whereClause;
1655 : List *leftargs,
1656 : *rightargs,
1657 : *opids,
1658 : *opcollations,
1659 : *newWhere,
1660 : *tlist,
1661 : *testlist,
1662 : *paramids;
1663 : ListCell *lc,
1664 : *rc,
1665 : *oc,
1666 : *cc;
1667 : AttrNumber resno;
1668 :
1669 : /*
1670 : * Query must not require a targetlist, since we have to insert a new one.
1671 : * Caller should have dealt with the case already.
1672 : */
1673 : Assert(subselect->targetList == NIL);
1674 :
1675 : /*
1676 : * Separate out the WHERE clause. (We could theoretically also remove
1677 : * top-level plain JOIN/ON clauses, but it's probably not worth the
1678 : * trouble.)
1679 : */
1680 2084 : whereClause = subselect->jointree->quals;
1681 2084 : subselect->jointree->quals = NULL;
1682 :
1683 : /*
1684 : * The rest of the sub-select must not refer to any Vars of the parent
1685 : * query. (Vars of higher levels should be okay, though.)
1686 : *
1687 : * Note: we need not check for Aggrefs separately because we know the
1688 : * sub-select is as yet unoptimized; any uplevel Aggref must therefore
1689 : * contain an uplevel Var reference. This is not the case below ...
1690 : */
1691 2084 : if (contain_vars_of_level((Node *) subselect, 1))
1692 6 : return NULL;
1693 :
1694 : /*
1695 : * We don't risk optimizing if the WHERE clause is volatile, either.
1696 : */
1697 2078 : if (contain_volatile_functions(whereClause))
1698 0 : return NULL;
1699 :
1700 : /*
1701 : * Clean up the WHERE clause by doing const-simplification etc on it.
1702 : * Aside from simplifying the processing we're about to do, this is
1703 : * important for being able to pull chunks of the WHERE clause up into the
1704 : * parent query. Since we are invoked partway through the parent's
1705 : * preprocess_expression() work, earlier steps of preprocess_expression()
1706 : * wouldn't get applied to the pulled-up stuff unless we do them here. For
1707 : * the parts of the WHERE clause that get put back into the child query,
1708 : * this work is partially duplicative, but it shouldn't hurt.
1709 : *
1710 : * Note: we do not run flatten_join_alias_vars. This is OK because any
1711 : * parent aliases were flattened already, and we're not going to pull any
1712 : * child Vars (of any description) into the parent.
1713 : *
1714 : * Note: passing the parent's root to eval_const_expressions is
1715 : * technically wrong, but we can get away with it since only the
1716 : * boundParams (if any) are used, and those would be the same in a
1717 : * subroot.
1718 : */
1719 2078 : whereClause = eval_const_expressions(root, whereClause);
1720 2078 : whereClause = (Node *) canonicalize_qual((Expr *) whereClause, false);
1721 2078 : whereClause = (Node *) make_ands_implicit((Expr *) whereClause);
1722 :
1723 : /*
1724 : * We now have a flattened implicit-AND list of clauses, which we try to
1725 : * break apart into "outervar = innervar" hash clauses. Anything that
1726 : * can't be broken apart just goes back into the newWhere list. Note that
1727 : * we aren't trying hard yet to ensure that we have only outer or only
1728 : * inner on each side; we'll check that if we get to the end.
1729 : */
1730 2078 : leftargs = rightargs = opids = opcollations = newWhere = NIL;
1731 8250 : foreach(lc, (List *) whereClause)
1732 : {
1733 6172 : OpExpr *expr = (OpExpr *) lfirst(lc);
1734 :
1735 10092 : if (IsA(expr, OpExpr) &&
1736 3920 : hash_ok_operator(expr))
1737 : {
1738 3054 : Node *leftarg = (Node *) linitial(expr->args);
1739 3054 : Node *rightarg = (Node *) lsecond(expr->args);
1740 :
1741 3054 : if (contain_vars_of_level(leftarg, 1))
1742 : {
1743 388 : leftargs = lappend(leftargs, leftarg);
1744 388 : rightargs = lappend(rightargs, rightarg);
1745 388 : opids = lappend_oid(opids, expr->opno);
1746 388 : opcollations = lappend_oid(opcollations, expr->inputcollid);
1747 388 : continue;
1748 : }
1749 2666 : if (contain_vars_of_level(rightarg, 1))
1750 : {
1751 : /*
1752 : * We must commute the clause to put the outer var on the
1753 : * left, because the hashing code in nodeSubplan.c expects
1754 : * that. This probably shouldn't ever fail, since hashable
1755 : * operators ought to have commutators, but be paranoid.
1756 : */
1757 2108 : expr->opno = get_commutator(expr->opno);
1758 2108 : if (OidIsValid(expr->opno) && hash_ok_operator(expr))
1759 : {
1760 2108 : leftargs = lappend(leftargs, rightarg);
1761 2108 : rightargs = lappend(rightargs, leftarg);
1762 2108 : opids = lappend_oid(opids, expr->opno);
1763 2108 : opcollations = lappend_oid(opcollations, expr->inputcollid);
1764 2108 : continue;
1765 : }
1766 : /* If no commutator, no chance to optimize the WHERE clause */
1767 0 : return NULL;
1768 : }
1769 : }
1770 : /* Couldn't handle it as a hash clause */
1771 3676 : newWhere = lappend(newWhere, expr);
1772 : }
1773 :
1774 : /*
1775 : * If we didn't find anything we could convert, fail.
1776 : */
1777 2078 : if (leftargs == NIL)
1778 358 : return NULL;
1779 :
1780 : /*
1781 : * There mustn't be any parent Vars or Aggs in the stuff that we intend to
1782 : * put back into the child query. Note: you might think we don't need to
1783 : * check for Aggs separately, because an uplevel Agg must contain an
1784 : * uplevel Var in its argument. But it is possible that the uplevel Var
1785 : * got optimized away by eval_const_expressions. Consider
1786 : *
1787 : * SUM(CASE WHEN false THEN uplevelvar ELSE 0 END)
1788 : */
1789 3350 : if (contain_vars_of_level((Node *) newWhere, 1) ||
1790 1630 : contain_vars_of_level((Node *) rightargs, 1))
1791 90 : return NULL;
1792 1672 : if (root->parse->hasAggs &&
1793 84 : (contain_aggs_of_level((Node *) newWhere, 1) ||
1794 42 : contain_aggs_of_level((Node *) rightargs, 1)))
1795 0 : return NULL;
1796 :
1797 : /*
1798 : * And there can't be any child Vars in the stuff we intend to pull up.
1799 : * (Note: we'd need to check for child Aggs too, except we know the child
1800 : * has no aggs at all because of simplify_EXISTS_query's check. The same
1801 : * goes for window functions.)
1802 : */
1803 1630 : if (contain_vars_of_level((Node *) leftargs, 0))
1804 0 : return NULL;
1805 :
1806 : /*
1807 : * Also reject sublinks in the stuff we intend to pull up. (It might be
1808 : * possible to support this, but doesn't seem worth the complication.)
1809 : */
1810 1630 : if (contain_subplans((Node *) leftargs))
1811 0 : return NULL;
1812 :
1813 : /*
1814 : * Okay, adjust the sublevelsup in the stuff we're pulling up.
1815 : */
1816 1630 : IncrementVarSublevelsUp((Node *) leftargs, -1, 1);
1817 :
1818 : /*
1819 : * Put back any child-level-only WHERE clauses.
1820 : */
1821 1630 : if (newWhere)
1822 1426 : subselect->jointree->quals = (Node *) make_ands_explicit(newWhere);
1823 :
1824 : /*
1825 : * Build a new targetlist for the child that emits the expressions we
1826 : * need. Concurrently, build a testexpr for the parent using Params to
1827 : * reference the child outputs. (Since we generate Params directly here,
1828 : * there will be no need to convert the testexpr in build_subplan.)
1829 : */
1830 1630 : tlist = testlist = paramids = NIL;
1831 1630 : resno = 1;
1832 4036 : forfour(lc, leftargs, rc, rightargs, oc, opids, cc, opcollations)
1833 : {
1834 2406 : Node *leftarg = (Node *) lfirst(lc);
1835 2406 : Node *rightarg = (Node *) lfirst(rc);
1836 2406 : Oid opid = lfirst_oid(oc);
1837 2406 : Oid opcollation = lfirst_oid(cc);
1838 : Param *param;
1839 :
1840 2406 : param = generate_new_exec_param(root,
1841 : exprType(rightarg),
1842 : exprTypmod(rightarg),
1843 : exprCollation(rightarg));
1844 2406 : tlist = lappend(tlist,
1845 2406 : makeTargetEntry((Expr *) rightarg,
1846 2406 : resno++,
1847 : NULL,
1848 : false));
1849 2406 : testlist = lappend(testlist,
1850 2406 : make_opclause(opid, BOOLOID, false,
1851 : (Expr *) leftarg, (Expr *) param,
1852 : InvalidOid, opcollation));
1853 2406 : paramids = lappend_int(paramids, param->paramid);
1854 : }
1855 :
1856 : /* Put everything where it should go, and we're done */
1857 1630 : subselect->targetList = tlist;
1858 1630 : *testexpr = (Node *) make_ands_explicit(testlist);
1859 1630 : *paramIds = paramids;
1860 :
1861 1630 : return subselect;
1862 : }
1863 :
1864 :
1865 : /*
1866 : * Replace correlation vars (uplevel vars) with Params.
1867 : *
1868 : * Uplevel PlaceHolderVars, aggregates, GROUPING() expressions,
1869 : * MergeSupportFuncs, and ReturningExprs are replaced, too.
1870 : *
1871 : * Note: it is critical that this runs immediately after SS_process_sublinks.
1872 : * Since we do not recurse into the arguments of uplevel PHVs and aggregates,
1873 : * they will get copied to the appropriate subplan args list in the parent
1874 : * query with uplevel vars not replaced by Params, but only adjusted in level
1875 : * (see replace_outer_placeholdervar and replace_outer_agg). That's exactly
1876 : * what we want for the vars of the parent level --- but if a PHV's or
1877 : * aggregate's argument contains any further-up variables, they have to be
1878 : * replaced with Params in their turn. That will happen when the parent level
1879 : * runs SS_replace_correlation_vars. Therefore it must do so after expanding
1880 : * its sublinks to subplans. And we don't want any steps in between, else
1881 : * those steps would never get applied to the argument expressions, either in
1882 : * the parent or the child level.
1883 : *
1884 : * Another fairly tricky thing going on here is the handling of SubLinks in
1885 : * the arguments of uplevel PHVs/aggregates. Those are not touched inside the
1886 : * intermediate query level, either. Instead, SS_process_sublinks recurses on
1887 : * them after copying the PHV or Aggref expression into the parent plan level
1888 : * (this is actually taken care of in build_subplan).
1889 : */
1890 : Node *
1891 153070 : SS_replace_correlation_vars(PlannerInfo *root, Node *expr)
1892 : {
1893 : /* No setup needed for tree walk, so away we go */
1894 153070 : return replace_correlation_vars_mutator(expr, root);
1895 : }
1896 :
1897 : static Node *
1898 1673826 : replace_correlation_vars_mutator(Node *node, PlannerInfo *root)
1899 : {
1900 1673826 : if (node == NULL)
1901 84374 : return NULL;
1902 1589452 : if (IsA(node, Var))
1903 : {
1904 435596 : if (((Var *) node)->varlevelsup > 0)
1905 52512 : return (Node *) replace_outer_var(root, (Var *) node);
1906 : }
1907 1536940 : if (IsA(node, PlaceHolderVar))
1908 : {
1909 84 : if (((PlaceHolderVar *) node)->phlevelsup > 0)
1910 42 : return (Node *) replace_outer_placeholdervar(root,
1911 : (PlaceHolderVar *) node);
1912 : }
1913 1536898 : if (IsA(node, Aggref))
1914 : {
1915 8542 : if (((Aggref *) node)->agglevelsup > 0)
1916 52 : return (Node *) replace_outer_agg(root, (Aggref *) node);
1917 : }
1918 1536846 : if (IsA(node, GroupingFunc))
1919 : {
1920 90 : if (((GroupingFunc *) node)->agglevelsup > 0)
1921 64 : return (Node *) replace_outer_grouping(root, (GroupingFunc *) node);
1922 : }
1923 1536782 : if (IsA(node, MergeSupportFunc))
1924 : {
1925 36 : if (root->parse->commandType != CMD_MERGE)
1926 6 : return (Node *) replace_outer_merge_support(root,
1927 : (MergeSupportFunc *) node);
1928 : }
1929 1536776 : if (IsA(node, ReturningExpr))
1930 : {
1931 18 : if (((ReturningExpr *) node)->retlevelsup > 0)
1932 18 : return (Node *) replace_outer_returning(root,
1933 : (ReturningExpr *) node);
1934 : }
1935 1536758 : return expression_tree_mutator(node, replace_correlation_vars_mutator, root);
1936 : }
1937 :
1938 : /*
1939 : * Expand SubLinks to SubPlans in the given expression.
1940 : *
1941 : * The isQual argument tells whether or not this expression is a WHERE/HAVING
1942 : * qualifier expression. If it is, any sublinks appearing at top level need
1943 : * not distinguish FALSE from UNKNOWN return values.
1944 : */
1945 : Node *
1946 91626 : SS_process_sublinks(PlannerInfo *root, Node *expr, bool isQual)
1947 : {
1948 : process_sublinks_context context;
1949 :
1950 91626 : context.root = root;
1951 91626 : context.isTopQual = isQual;
1952 91626 : return process_sublinks_mutator(expr, &context);
1953 : }
1954 :
1955 : static Node *
1956 1243370 : process_sublinks_mutator(Node *node, process_sublinks_context *context)
1957 : {
1958 : process_sublinks_context locContext;
1959 :
1960 1243370 : locContext.root = context->root;
1961 :
1962 1243370 : if (node == NULL)
1963 57238 : return NULL;
1964 1186132 : if (IsA(node, SubLink))
1965 : {
1966 38064 : SubLink *sublink = (SubLink *) node;
1967 : Node *testexpr;
1968 :
1969 : /*
1970 : * First, recursively process the lefthand-side expressions, if any.
1971 : * They're not top-level anymore.
1972 : */
1973 38064 : locContext.isTopQual = false;
1974 38064 : testexpr = process_sublinks_mutator(sublink->testexpr, &locContext);
1975 :
1976 : /*
1977 : * Now build the SubPlan node and make the expr to return.
1978 : */
1979 38064 : return make_subplan(context->root,
1980 38064 : (Query *) sublink->subselect,
1981 : sublink->subLinkType,
1982 : sublink->subLinkId,
1983 : testexpr,
1984 38064 : context->isTopQual);
1985 : }
1986 :
1987 : /*
1988 : * Don't recurse into the arguments of an outer PHV, Aggref, GroupingFunc,
1989 : * or ReturningExpr here. Any SubLinks in the arguments have to be dealt
1990 : * with at the outer query level; they'll be handled when build_subplan
1991 : * collects the PHV, Aggref, GroupingFunc, or ReturningExpr into the
1992 : * arguments to be passed down to the current subplan.
1993 : */
1994 1148068 : if (IsA(node, PlaceHolderVar))
1995 : {
1996 210 : if (((PlaceHolderVar *) node)->phlevelsup > 0)
1997 0 : return node;
1998 : }
1999 1147858 : else if (IsA(node, Aggref))
2000 : {
2001 584 : if (((Aggref *) node)->agglevelsup > 0)
2002 18 : return node;
2003 : }
2004 1147274 : else if (IsA(node, GroupingFunc))
2005 : {
2006 160 : if (((GroupingFunc *) node)->agglevelsup > 0)
2007 36 : return node;
2008 : }
2009 1147114 : else if (IsA(node, ReturningExpr))
2010 : {
2011 198 : if (((ReturningExpr *) node)->retlevelsup > 0)
2012 6 : return node;
2013 : }
2014 :
2015 : /*
2016 : * We should never see a SubPlan expression in the input (since this is
2017 : * the very routine that creates 'em to begin with). We shouldn't find
2018 : * ourselves invoked directly on a Query, either.
2019 : */
2020 : Assert(!IsA(node, SubPlan));
2021 : Assert(!IsA(node, AlternativeSubPlan));
2022 : Assert(!IsA(node, Query));
2023 :
2024 : /*
2025 : * Because make_subplan() could return an AND or OR clause, we have to
2026 : * take steps to preserve AND/OR flatness of a qual. We assume the input
2027 : * has been AND/OR flattened and so we need no recursion here.
2028 : *
2029 : * (Due to the coding here, we will not get called on the List subnodes of
2030 : * an AND; and the input is *not* yet in implicit-AND format. So no check
2031 : * is needed for a bare List.)
2032 : *
2033 : * Anywhere within the top-level AND/OR clause structure, we can tell
2034 : * make_subplan() that NULL and FALSE are interchangeable. So isTopQual
2035 : * propagates down in both cases. (Note that this is unlike the meaning
2036 : * of "top level qual" used in most other places in Postgres.)
2037 : */
2038 1148008 : if (is_andclause(node))
2039 : {
2040 16400 : List *newargs = NIL;
2041 : ListCell *l;
2042 :
2043 : /* Still at qual top-level */
2044 16400 : locContext.isTopQual = context->isTopQual;
2045 :
2046 61820 : foreach(l, ((BoolExpr *) node)->args)
2047 : {
2048 : Node *newarg;
2049 :
2050 45420 : newarg = process_sublinks_mutator(lfirst(l), &locContext);
2051 45420 : if (is_andclause(newarg))
2052 0 : newargs = list_concat(newargs, ((BoolExpr *) newarg)->args);
2053 : else
2054 45420 : newargs = lappend(newargs, newarg);
2055 : }
2056 16400 : return (Node *) make_andclause(newargs);
2057 : }
2058 :
2059 1131608 : if (is_orclause(node))
2060 : {
2061 2212 : List *newargs = NIL;
2062 : ListCell *l;
2063 :
2064 : /* Still at qual top-level */
2065 2212 : locContext.isTopQual = context->isTopQual;
2066 :
2067 7900 : foreach(l, ((BoolExpr *) node)->args)
2068 : {
2069 : Node *newarg;
2070 :
2071 5688 : newarg = process_sublinks_mutator(lfirst(l), &locContext);
2072 5688 : if (is_orclause(newarg))
2073 0 : newargs = list_concat(newargs, ((BoolExpr *) newarg)->args);
2074 : else
2075 5688 : newargs = lappend(newargs, newarg);
2076 : }
2077 2212 : return (Node *) make_orclause(newargs);
2078 : }
2079 :
2080 : /*
2081 : * If we recurse down through anything other than an AND or OR node, we
2082 : * are definitely not at top qual level anymore.
2083 : */
2084 1129396 : locContext.isTopQual = false;
2085 :
2086 1129396 : return expression_tree_mutator(node,
2087 : process_sublinks_mutator,
2088 : &locContext);
2089 : }
2090 :
2091 : /*
2092 : * SS_identify_outer_params - identify the Params available from outer levels
2093 : *
2094 : * This must be run after SS_replace_correlation_vars and SS_process_sublinks
2095 : * processing is complete in a given query level as well as all of its
2096 : * descendant levels (which means it's most practical to do it at the end of
2097 : * processing the query level). We compute the set of paramIds that outer
2098 : * levels will make available to this level+descendants, and record it in
2099 : * root->outer_params for use while computing extParam/allParam sets in final
2100 : * plan cleanup. (We can't just compute it then, because the upper levels'
2101 : * plan_params lists are transient and will be gone by then.)
2102 : */
2103 : void
2104 542762 : SS_identify_outer_params(PlannerInfo *root)
2105 : {
2106 : Bitmapset *outer_params;
2107 : PlannerInfo *proot;
2108 : ListCell *l;
2109 :
2110 : /*
2111 : * If no parameters have been assigned anywhere in the tree, we certainly
2112 : * don't need to do anything here.
2113 : */
2114 542762 : if (root->glob->paramExecTypes == NIL)
2115 370882 : return;
2116 :
2117 : /*
2118 : * Scan all query levels above this one to see which parameters are due to
2119 : * be available from them, either because lower query levels have
2120 : * requested them (via plan_params) or because they will be available from
2121 : * initPlans of those levels.
2122 : */
2123 171880 : outer_params = NULL;
2124 227092 : for (proot = root->parent_root; proot != NULL; proot = proot->parent_root)
2125 : {
2126 : /*
2127 : * Include ordinary Var/PHV/Aggref/GroupingFunc/ReturningExpr params.
2128 : */
2129 100636 : foreach(l, proot->plan_params)
2130 : {
2131 45424 : PlannerParamItem *pitem = (PlannerParamItem *) lfirst(l);
2132 :
2133 45424 : outer_params = bms_add_member(outer_params, pitem->paramId);
2134 : }
2135 : /* Include any outputs of outer-level initPlans */
2136 61092 : foreach(l, proot->init_plans)
2137 : {
2138 5880 : SubPlan *initsubplan = (SubPlan *) lfirst(l);
2139 : ListCell *l2;
2140 :
2141 11760 : foreach(l2, initsubplan->setParam)
2142 : {
2143 5880 : outer_params = bms_add_member(outer_params, lfirst_int(l2));
2144 : }
2145 : }
2146 : /* Include worktable ID, if a recursive query is being planned */
2147 55212 : if (proot->wt_param_id >= 0)
2148 2994 : outer_params = bms_add_member(outer_params, proot->wt_param_id);
2149 : }
2150 171880 : root->outer_params = outer_params;
2151 : }
2152 :
2153 : /*
2154 : * SS_charge_for_initplans - account for initplans in Path costs & parallelism
2155 : *
2156 : * If any initPlans have been created in the current query level, they will
2157 : * get attached to the Plan tree created from whichever Path we select from
2158 : * the given rel. Increment all that rel's Paths' costs to account for them,
2159 : * and if any of the initPlans are parallel-unsafe, mark all the rel's Paths
2160 : * parallel-unsafe as well.
2161 : *
2162 : * This is separate from SS_attach_initplans because we might conditionally
2163 : * create more initPlans during create_plan(), depending on which Path we
2164 : * select. However, Paths that would generate such initPlans are expected
2165 : * to have included their cost and parallel-safety effects already.
2166 : */
2167 : void
2168 542762 : SS_charge_for_initplans(PlannerInfo *root, RelOptInfo *final_rel)
2169 : {
2170 : Cost initplan_cost;
2171 : bool unsafe_initplans;
2172 : ListCell *lc;
2173 :
2174 : /* Nothing to do if no initPlans */
2175 542762 : if (root->init_plans == NIL)
2176 531846 : return;
2177 :
2178 : /*
2179 : * Compute the cost increment just once, since it will be the same for all
2180 : * Paths. Also check for parallel-unsafe initPlans.
2181 : */
2182 10916 : SS_compute_initplan_cost(root->init_plans,
2183 : &initplan_cost, &unsafe_initplans);
2184 :
2185 : /*
2186 : * Now adjust the costs and parallel_safe flags.
2187 : */
2188 22002 : foreach(lc, final_rel->pathlist)
2189 : {
2190 11086 : Path *path = (Path *) lfirst(lc);
2191 :
2192 11086 : path->startup_cost += initplan_cost;
2193 11086 : path->total_cost += initplan_cost;
2194 11086 : if (unsafe_initplans)
2195 6512 : path->parallel_safe = false;
2196 : }
2197 :
2198 : /*
2199 : * Adjust partial paths' costs too, or forget them entirely if we must
2200 : * consider the rel parallel-unsafe.
2201 : */
2202 10916 : if (unsafe_initplans)
2203 : {
2204 6436 : final_rel->partial_pathlist = NIL;
2205 6436 : final_rel->consider_parallel = false;
2206 : }
2207 : else
2208 : {
2209 4492 : foreach(lc, final_rel->partial_pathlist)
2210 : {
2211 12 : Path *path = (Path *) lfirst(lc);
2212 :
2213 12 : path->startup_cost += initplan_cost;
2214 12 : path->total_cost += initplan_cost;
2215 : }
2216 : }
2217 :
2218 : /* We needn't do set_cheapest() here, caller will do it */
2219 : }
2220 :
2221 : /*
2222 : * SS_compute_initplan_cost - count up the cost delta for some initplans
2223 : *
2224 : * The total cost returned in *initplan_cost_p should be added to both the
2225 : * startup and total costs of the plan node the initplans get attached to.
2226 : * We also report whether any of the initplans are not parallel-safe.
2227 : *
2228 : * The primary user of this is SS_charge_for_initplans, but it's also
2229 : * used in adjusting costs when we move initplans to another plan node.
2230 : */
2231 : void
2232 11118 : SS_compute_initplan_cost(List *init_plans,
2233 : Cost *initplan_cost_p,
2234 : bool *unsafe_initplans_p)
2235 : {
2236 : Cost initplan_cost;
2237 : bool unsafe_initplans;
2238 : ListCell *lc;
2239 :
2240 : /*
2241 : * We assume each initPlan gets run once during top plan startup. This is
2242 : * a conservative overestimate, since in fact an initPlan might be
2243 : * executed later than plan startup, or even not at all.
2244 : */
2245 11118 : initplan_cost = 0;
2246 11118 : unsafe_initplans = false;
2247 23264 : foreach(lc, init_plans)
2248 : {
2249 12146 : SubPlan *initsubplan = lfirst_node(SubPlan, lc);
2250 :
2251 12146 : initplan_cost += initsubplan->startup_cost + initsubplan->per_call_cost;
2252 12146 : if (!initsubplan->parallel_safe)
2253 7410 : unsafe_initplans = true;
2254 : }
2255 11118 : *initplan_cost_p = initplan_cost;
2256 11118 : *unsafe_initplans_p = unsafe_initplans;
2257 11118 : }
2258 :
2259 : /*
2260 : * SS_attach_initplans - attach initplans to topmost plan node
2261 : *
2262 : * Attach any initplans created in the current query level to the specified
2263 : * plan node, which should normally be the topmost node for the query level.
2264 : * (In principle the initPlans could go in any node at or above where they're
2265 : * referenced; but there seems no reason to put them any lower than the
2266 : * topmost node, so we don't bother to track exactly where they came from.)
2267 : *
2268 : * We do not touch the plan node's cost or parallel_safe flag. The initplans
2269 : * must have been accounted for in SS_charge_for_initplans, or by any later
2270 : * code that adds initplans via SS_make_initplan_from_plan.
2271 : */
2272 : void
2273 541498 : SS_attach_initplans(PlannerInfo *root, Plan *plan)
2274 : {
2275 541498 : plan->initPlan = root->init_plans;
2276 541498 : }
2277 :
2278 : /*
2279 : * SS_finalize_plan - do final parameter processing for a completed Plan.
2280 : *
2281 : * This recursively computes the extParam and allParam sets for every Plan
2282 : * node in the given plan tree. (Oh, and RangeTblFunction.funcparams too.)
2283 : *
2284 : * We assume that SS_finalize_plan has already been run on any initplans or
2285 : * subplans the plan tree could reference.
2286 : */
2287 : void
2288 198290 : SS_finalize_plan(PlannerInfo *root, Plan *plan)
2289 : {
2290 : /* No setup needed, just recurse through plan tree. */
2291 198290 : (void) finalize_plan(root, plan, -1, root->outer_params, NULL);
2292 198290 : }
2293 :
2294 : /*
2295 : * Recursive processing of all nodes in the plan tree
2296 : *
2297 : * gather_param is the rescan_param of an ancestral Gather/GatherMerge,
2298 : * or -1 if there is none.
2299 : *
2300 : * valid_params is the set of param IDs supplied by outer plan levels
2301 : * that are valid to reference in this plan node or its children.
2302 : *
2303 : * scan_params is a set of param IDs to force scan plan nodes to reference.
2304 : * This is for EvalPlanQual support, and is always NULL at the top of the
2305 : * recursion.
2306 : *
2307 : * The return value is the computed allParam set for the given Plan node.
2308 : * This is just an internal notational convenience: we can add a child
2309 : * plan's allParams to the set of param IDs of interest to this level
2310 : * in the same statement that recurses to that child.
2311 : *
2312 : * Do not scribble on caller's values of valid_params or scan_params!
2313 : *
2314 : * Note: although we attempt to deal with initPlans anywhere in the tree, the
2315 : * logic is not really right. The problem is that a plan node might return an
2316 : * output Param of its initPlan as a targetlist item, in which case it's valid
2317 : * for the parent plan level to reference that same Param; the parent's usage
2318 : * will be converted into a Var referencing the child plan node by setrefs.c.
2319 : * But this function would see the parent's reference as out of scope and
2320 : * complain about it. For now, this does not matter because the planner only
2321 : * attaches initPlans to the topmost plan node in a query level, so the case
2322 : * doesn't arise. If we ever merge this processing into setrefs.c, maybe it
2323 : * can be handled more cleanly.
2324 : */
2325 : static Bitmapset *
2326 1489238 : finalize_plan(PlannerInfo *root, Plan *plan,
2327 : int gather_param,
2328 : Bitmapset *valid_params,
2329 : Bitmapset *scan_params)
2330 : {
2331 : finalize_primnode_context context;
2332 : int locally_added_param;
2333 : Bitmapset *nestloop_params;
2334 : Bitmapset *initExtParam;
2335 : Bitmapset *initSetParam;
2336 : Bitmapset *child_params;
2337 : ListCell *l;
2338 :
2339 1489238 : if (plan == NULL)
2340 866368 : return NULL;
2341 :
2342 622870 : context.root = root;
2343 622870 : context.paramids = NULL; /* initialize set to empty */
2344 622870 : locally_added_param = -1; /* there isn't one */
2345 622870 : nestloop_params = NULL; /* there aren't any */
2346 :
2347 : /*
2348 : * Examine any initPlans to determine the set of external params they
2349 : * reference and the set of output params they supply. (We assume
2350 : * SS_finalize_plan was run on them already.)
2351 : */
2352 622870 : initExtParam = initSetParam = NULL;
2353 635356 : foreach(l, plan->initPlan)
2354 : {
2355 12486 : SubPlan *initsubplan = (SubPlan *) lfirst(l);
2356 12486 : Plan *initplan = planner_subplan_get_plan(root, initsubplan);
2357 : ListCell *l2;
2358 :
2359 12486 : initExtParam = bms_add_members(initExtParam, initplan->extParam);
2360 25020 : foreach(l2, initsubplan->setParam)
2361 : {
2362 12534 : initSetParam = bms_add_member(initSetParam, lfirst_int(l2));
2363 : }
2364 : }
2365 :
2366 : /* Any setParams are validly referenceable in this node and children */
2367 622870 : if (initSetParam)
2368 11268 : valid_params = bms_union(valid_params, initSetParam);
2369 :
2370 : /*
2371 : * When we call finalize_primnode, context.paramids sets are automatically
2372 : * merged together. But when recursing to self, we have to do it the hard
2373 : * way. We want the paramids set to include params in subplans as well as
2374 : * at this level.
2375 : */
2376 :
2377 : /* Find params in targetlist and qual */
2378 622870 : finalize_primnode((Node *) plan->targetlist, &context);
2379 622870 : finalize_primnode((Node *) plan->qual, &context);
2380 :
2381 : /*
2382 : * If it's a parallel-aware scan node, mark it as dependent on the parent
2383 : * Gather/GatherMerge's rescan Param.
2384 : */
2385 622870 : if (plan->parallel_aware)
2386 : {
2387 2540 : if (gather_param < 0)
2388 0 : elog(ERROR, "parallel-aware plan node is not below a Gather");
2389 2540 : context.paramids =
2390 2540 : bms_add_member(context.paramids, gather_param);
2391 : }
2392 :
2393 : /* Check additional node-type-specific fields */
2394 622870 : switch (nodeTag(plan))
2395 : {
2396 76752 : case T_Result:
2397 76752 : finalize_primnode(((Result *) plan)->resconstantqual,
2398 : &context);
2399 76752 : break;
2400 :
2401 89736 : case T_SeqScan:
2402 89736 : context.paramids = bms_add_members(context.paramids, scan_params);
2403 89736 : break;
2404 :
2405 104 : case T_SampleScan:
2406 104 : finalize_primnode((Node *) ((SampleScan *) plan)->tablesample,
2407 : &context);
2408 104 : context.paramids = bms_add_members(context.paramids, scan_params);
2409 104 : break;
2410 :
2411 96840 : case T_IndexScan:
2412 96840 : finalize_primnode((Node *) ((IndexScan *) plan)->indexqual,
2413 : &context);
2414 96840 : finalize_primnode((Node *) ((IndexScan *) plan)->indexorderby,
2415 : &context);
2416 :
2417 : /*
2418 : * we need not look at indexqualorig, since it will have the same
2419 : * param references as indexqual. Likewise, we can ignore
2420 : * indexorderbyorig.
2421 : */
2422 96840 : context.paramids = bms_add_members(context.paramids, scan_params);
2423 96840 : break;
2424 :
2425 7122 : case T_IndexOnlyScan:
2426 7122 : finalize_primnode((Node *) ((IndexOnlyScan *) plan)->indexqual,
2427 : &context);
2428 7122 : finalize_primnode((Node *) ((IndexOnlyScan *) plan)->recheckqual,
2429 : &context);
2430 7122 : finalize_primnode((Node *) ((IndexOnlyScan *) plan)->indexorderby,
2431 : &context);
2432 :
2433 : /*
2434 : * we need not look at indextlist, since it cannot contain Params.
2435 : */
2436 7122 : context.paramids = bms_add_members(context.paramids, scan_params);
2437 7122 : break;
2438 :
2439 8830 : case T_BitmapIndexScan:
2440 8830 : finalize_primnode((Node *) ((BitmapIndexScan *) plan)->indexqual,
2441 : &context);
2442 :
2443 : /*
2444 : * we need not look at indexqualorig, since it will have the same
2445 : * param references as indexqual.
2446 : */
2447 8830 : break;
2448 :
2449 8480 : case T_BitmapHeapScan:
2450 8480 : finalize_primnode((Node *) ((BitmapHeapScan *) plan)->bitmapqualorig,
2451 : &context);
2452 8480 : context.paramids = bms_add_members(context.paramids, scan_params);
2453 8480 : break;
2454 :
2455 604 : case T_TidScan:
2456 604 : finalize_primnode((Node *) ((TidScan *) plan)->tidquals,
2457 : &context);
2458 604 : context.paramids = bms_add_members(context.paramids, scan_params);
2459 604 : break;
2460 :
2461 34 : case T_TidRangeScan:
2462 34 : finalize_primnode((Node *) ((TidRangeScan *) plan)->tidrangequals,
2463 : &context);
2464 34 : context.paramids = bms_add_members(context.paramids, scan_params);
2465 34 : break;
2466 :
2467 19198 : case T_SubqueryScan:
2468 : {
2469 19198 : SubqueryScan *sscan = (SubqueryScan *) plan;
2470 : RelOptInfo *rel;
2471 : Bitmapset *subquery_params;
2472 :
2473 : /* We must run finalize_plan on the subquery */
2474 19198 : rel = find_base_rel(root, sscan->scan.scanrelid);
2475 19198 : subquery_params = rel->subroot->outer_params;
2476 19198 : if (gather_param >= 0)
2477 24 : subquery_params = bms_add_member(bms_copy(subquery_params),
2478 : gather_param);
2479 19198 : finalize_plan(rel->subroot, sscan->subplan, gather_param,
2480 : subquery_params, NULL);
2481 :
2482 : /* Now we can add its extParams to the parent's params */
2483 38396 : context.paramids = bms_add_members(context.paramids,
2484 19198 : sscan->subplan->extParam);
2485 : /* We need scan_params too, though */
2486 19198 : context.paramids = bms_add_members(context.paramids,
2487 : scan_params);
2488 : }
2489 19198 : break;
2490 :
2491 23242 : case T_FunctionScan:
2492 : {
2493 23242 : FunctionScan *fscan = (FunctionScan *) plan;
2494 : ListCell *lc;
2495 :
2496 : /*
2497 : * Call finalize_primnode independently on each function
2498 : * expression, so that we can record which params are
2499 : * referenced in each, in order to decide which need
2500 : * re-evaluating during rescan.
2501 : */
2502 46508 : foreach(lc, fscan->functions)
2503 : {
2504 23266 : RangeTblFunction *rtfunc = (RangeTblFunction *) lfirst(lc);
2505 : finalize_primnode_context funccontext;
2506 :
2507 23266 : funccontext = context;
2508 23266 : funccontext.paramids = NULL;
2509 :
2510 23266 : finalize_primnode(rtfunc->funcexpr, &funccontext);
2511 :
2512 : /* remember results for execution */
2513 23266 : rtfunc->funcparams = funccontext.paramids;
2514 :
2515 : /* add the function's params to the overall set */
2516 23266 : context.paramids = bms_add_members(context.paramids,
2517 23266 : funccontext.paramids);
2518 : }
2519 :
2520 23242 : context.paramids = bms_add_members(context.paramids,
2521 : scan_params);
2522 : }
2523 23242 : break;
2524 :
2525 234 : case T_TableFuncScan:
2526 234 : finalize_primnode((Node *) ((TableFuncScan *) plan)->tablefunc,
2527 : &context);
2528 234 : context.paramids = bms_add_members(context.paramids, scan_params);
2529 234 : break;
2530 :
2531 5624 : case T_ValuesScan:
2532 5624 : finalize_primnode((Node *) ((ValuesScan *) plan)->values_lists,
2533 : &context);
2534 5624 : context.paramids = bms_add_members(context.paramids, scan_params);
2535 5624 : break;
2536 :
2537 3986 : case T_CteScan:
2538 : {
2539 : /*
2540 : * You might think we should add the node's cteParam to
2541 : * paramids, but we shouldn't because that param is just a
2542 : * linkage mechanism for multiple CteScan nodes for the same
2543 : * CTE; it is never used for changed-param signaling. What we
2544 : * have to do instead is to find the referenced CTE plan and
2545 : * incorporate its external paramids, so that the correct
2546 : * things will happen if the CTE references outer-level
2547 : * variables. See test cases for bug #4902. (We assume
2548 : * SS_finalize_plan was run on the CTE plan already.)
2549 : */
2550 3986 : int plan_id = ((CteScan *) plan)->ctePlanId;
2551 : Plan *cteplan;
2552 :
2553 : /* so, do this ... */
2554 3986 : if (plan_id < 1 || plan_id > list_length(root->glob->subplans))
2555 0 : elog(ERROR, "could not find plan for CteScan referencing plan ID %d",
2556 : plan_id);
2557 3986 : cteplan = (Plan *) list_nth(root->glob->subplans, plan_id - 1);
2558 3986 : context.paramids =
2559 3986 : bms_add_members(context.paramids, cteplan->extParam);
2560 :
2561 : #ifdef NOT_USED
2562 : /* ... but not this */
2563 : context.paramids =
2564 : bms_add_member(context.paramids,
2565 : ((CteScan *) plan)->cteParam);
2566 : #endif
2567 :
2568 3986 : context.paramids = bms_add_members(context.paramids,
2569 : scan_params);
2570 : }
2571 3986 : break;
2572 :
2573 896 : case T_WorkTableScan:
2574 896 : context.paramids =
2575 896 : bms_add_member(context.paramids,
2576 : ((WorkTableScan *) plan)->wtParam);
2577 896 : context.paramids = bms_add_members(context.paramids, scan_params);
2578 896 : break;
2579 :
2580 390 : case T_NamedTuplestoreScan:
2581 390 : context.paramids = bms_add_members(context.paramids, scan_params);
2582 390 : break;
2583 :
2584 788 : case T_ForeignScan:
2585 : {
2586 788 : ForeignScan *fscan = (ForeignScan *) plan;
2587 :
2588 788 : finalize_primnode((Node *) fscan->fdw_exprs,
2589 : &context);
2590 788 : finalize_primnode((Node *) fscan->fdw_recheck_quals,
2591 : &context);
2592 :
2593 : /* We assume fdw_scan_tlist cannot contain Params */
2594 788 : context.paramids = bms_add_members(context.paramids,
2595 : scan_params);
2596 : }
2597 788 : break;
2598 :
2599 0 : case T_CustomScan:
2600 : {
2601 0 : CustomScan *cscan = (CustomScan *) plan;
2602 : ListCell *lc;
2603 :
2604 0 : finalize_primnode((Node *) cscan->custom_exprs,
2605 : &context);
2606 : /* We assume custom_scan_tlist cannot contain Params */
2607 0 : context.paramids =
2608 0 : bms_add_members(context.paramids, scan_params);
2609 :
2610 : /* child nodes if any */
2611 0 : foreach(lc, cscan->custom_plans)
2612 : {
2613 0 : context.paramids =
2614 0 : bms_add_members(context.paramids,
2615 0 : finalize_plan(root,
2616 0 : (Plan *) lfirst(lc),
2617 : gather_param,
2618 : valid_params,
2619 : scan_params));
2620 : }
2621 : }
2622 0 : break;
2623 :
2624 91890 : case T_ModifyTable:
2625 : {
2626 91890 : ModifyTable *mtplan = (ModifyTable *) plan;
2627 :
2628 : /* Force descendant scan nodes to reference epqParam */
2629 91890 : locally_added_param = mtplan->epqParam;
2630 91890 : valid_params = bms_add_member(bms_copy(valid_params),
2631 : locally_added_param);
2632 91890 : scan_params = bms_add_member(bms_copy(scan_params),
2633 : locally_added_param);
2634 91890 : finalize_primnode((Node *) mtplan->returningLists,
2635 : &context);
2636 91890 : finalize_primnode((Node *) mtplan->onConflictSet,
2637 : &context);
2638 91890 : finalize_primnode((Node *) mtplan->onConflictWhere,
2639 : &context);
2640 : /* exclRelTlist contains only Vars, doesn't need examination */
2641 : }
2642 91890 : break;
2643 :
2644 10108 : case T_Append:
2645 : {
2646 35070 : foreach(l, ((Append *) plan)->appendplans)
2647 : {
2648 24962 : context.paramids =
2649 24962 : bms_add_members(context.paramids,
2650 24962 : finalize_plan(root,
2651 24962 : (Plan *) lfirst(l),
2652 : gather_param,
2653 : valid_params,
2654 : scan_params));
2655 : }
2656 : }
2657 10108 : break;
2658 :
2659 108 : case T_MergeAppend:
2660 : {
2661 456 : foreach(l, ((MergeAppend *) plan)->mergeplans)
2662 : {
2663 348 : context.paramids =
2664 348 : bms_add_members(context.paramids,
2665 348 : finalize_plan(root,
2666 348 : (Plan *) lfirst(l),
2667 : gather_param,
2668 : valid_params,
2669 : scan_params));
2670 : }
2671 : }
2672 108 : break;
2673 :
2674 150 : case T_BitmapAnd:
2675 : {
2676 450 : foreach(l, ((BitmapAnd *) plan)->bitmapplans)
2677 : {
2678 300 : context.paramids =
2679 300 : bms_add_members(context.paramids,
2680 300 : finalize_plan(root,
2681 300 : (Plan *) lfirst(l),
2682 : gather_param,
2683 : valid_params,
2684 : scan_params));
2685 : }
2686 : }
2687 150 : break;
2688 :
2689 200 : case T_BitmapOr:
2690 : {
2691 600 : foreach(l, ((BitmapOr *) plan)->bitmapplans)
2692 : {
2693 400 : context.paramids =
2694 400 : bms_add_members(context.paramids,
2695 400 : finalize_plan(root,
2696 400 : (Plan *) lfirst(l),
2697 : gather_param,
2698 : valid_params,
2699 : scan_params));
2700 : }
2701 : }
2702 200 : break;
2703 :
2704 67612 : case T_NestLoop:
2705 : {
2706 67612 : finalize_primnode((Node *) ((Join *) plan)->joinqual,
2707 : &context);
2708 : /* collect set of params that will be passed to right child */
2709 118368 : foreach(l, ((NestLoop *) plan)->nestParams)
2710 : {
2711 50756 : NestLoopParam *nlp = (NestLoopParam *) lfirst(l);
2712 :
2713 50756 : nestloop_params = bms_add_member(nestloop_params,
2714 : nlp->paramno);
2715 : }
2716 : }
2717 67612 : break;
2718 :
2719 4288 : case T_MergeJoin:
2720 4288 : finalize_primnode((Node *) ((Join *) plan)->joinqual,
2721 : &context);
2722 4288 : finalize_primnode((Node *) ((MergeJoin *) plan)->mergeclauses,
2723 : &context);
2724 4288 : break;
2725 :
2726 18166 : case T_HashJoin:
2727 18166 : finalize_primnode((Node *) ((Join *) plan)->joinqual,
2728 : &context);
2729 18166 : finalize_primnode((Node *) ((HashJoin *) plan)->hashclauses,
2730 : &context);
2731 18166 : break;
2732 :
2733 18166 : case T_Hash:
2734 18166 : finalize_primnode((Node *) ((Hash *) plan)->hashkeys,
2735 : &context);
2736 18166 : break;
2737 :
2738 2208 : case T_Limit:
2739 2208 : finalize_primnode(((Limit *) plan)->limitOffset,
2740 : &context);
2741 2208 : finalize_primnode(((Limit *) plan)->limitCount,
2742 : &context);
2743 2208 : break;
2744 :
2745 896 : case T_RecursiveUnion:
2746 : /* child nodes are allowed to reference wtParam */
2747 896 : locally_added_param = ((RecursiveUnion *) plan)->wtParam;
2748 896 : valid_params = bms_add_member(bms_copy(valid_params),
2749 : locally_added_param);
2750 : /* wtParam does *not* get added to scan_params */
2751 896 : break;
2752 :
2753 7646 : case T_LockRows:
2754 : /* Force descendant scan nodes to reference epqParam */
2755 7646 : locally_added_param = ((LockRows *) plan)->epqParam;
2756 7646 : valid_params = bms_add_member(bms_copy(valid_params),
2757 : locally_added_param);
2758 7646 : scan_params = bms_add_member(bms_copy(scan_params),
2759 : locally_added_param);
2760 7646 : break;
2761 :
2762 10134 : case T_Agg:
2763 : {
2764 10134 : Agg *agg = (Agg *) plan;
2765 :
2766 : /*
2767 : * AGG_HASHED plans need to know which Params are referenced
2768 : * in aggregate calls. Do a separate scan to identify them.
2769 : */
2770 10134 : if (agg->aggstrategy == AGG_HASHED)
2771 : {
2772 : finalize_primnode_context aggcontext;
2773 :
2774 1324 : aggcontext.root = root;
2775 1324 : aggcontext.paramids = NULL;
2776 1324 : finalize_agg_primnode((Node *) agg->plan.targetlist,
2777 : &aggcontext);
2778 1324 : finalize_agg_primnode((Node *) agg->plan.qual,
2779 : &aggcontext);
2780 1324 : agg->aggParams = aggcontext.paramids;
2781 : }
2782 : }
2783 10134 : break;
2784 :
2785 122 : case T_WindowAgg:
2786 122 : finalize_primnode(((WindowAgg *) plan)->startOffset,
2787 : &context);
2788 122 : finalize_primnode(((WindowAgg *) plan)->endOffset,
2789 : &context);
2790 122 : break;
2791 :
2792 956 : case T_Gather:
2793 : /* child nodes are allowed to reference rescan_param, if any */
2794 956 : locally_added_param = ((Gather *) plan)->rescan_param;
2795 956 : if (locally_added_param >= 0)
2796 : {
2797 950 : valid_params = bms_add_member(bms_copy(valid_params),
2798 : locally_added_param);
2799 :
2800 : /*
2801 : * We currently don't support nested Gathers. The issue so
2802 : * far as this function is concerned would be how to identify
2803 : * which child nodes depend on which Gather.
2804 : */
2805 : Assert(gather_param < 0);
2806 : /* Pass down rescan_param to child parallel-aware nodes */
2807 950 : gather_param = locally_added_param;
2808 : }
2809 : /* rescan_param does *not* get added to scan_params */
2810 956 : break;
2811 :
2812 330 : case T_GatherMerge:
2813 : /* child nodes are allowed to reference rescan_param, if any */
2814 330 : locally_added_param = ((GatherMerge *) plan)->rescan_param;
2815 330 : if (locally_added_param >= 0)
2816 : {
2817 330 : valid_params = bms_add_member(bms_copy(valid_params),
2818 : locally_added_param);
2819 :
2820 : /*
2821 : * We currently don't support nested Gathers. The issue so
2822 : * far as this function is concerned would be how to identify
2823 : * which child nodes depend on which Gather.
2824 : */
2825 : Assert(gather_param < 0);
2826 : /* Pass down rescan_param to child parallel-aware nodes */
2827 330 : gather_param = locally_added_param;
2828 : }
2829 : /* rescan_param does *not* get added to scan_params */
2830 330 : break;
2831 :
2832 1604 : case T_Memoize:
2833 1604 : finalize_primnode((Node *) ((Memoize *) plan)->param_exprs,
2834 : &context);
2835 1604 : break;
2836 :
2837 45426 : case T_ProjectSet:
2838 : case T_Material:
2839 : case T_Sort:
2840 : case T_IncrementalSort:
2841 : case T_Unique:
2842 : case T_SetOp:
2843 : case T_Group:
2844 : /* no node-type-specific fields need fixing */
2845 45426 : break;
2846 :
2847 0 : default:
2848 0 : elog(ERROR, "unrecognized node type: %d",
2849 : (int) nodeTag(plan));
2850 : }
2851 :
2852 : /* Process left and right child plans, if any */
2853 622870 : child_params = finalize_plan(root,
2854 622870 : plan->lefttree,
2855 : gather_param,
2856 : valid_params,
2857 : scan_params);
2858 622870 : context.paramids = bms_add_members(context.paramids, child_params);
2859 :
2860 622870 : if (nestloop_params)
2861 : {
2862 : /* right child can reference nestloop_params as well as valid_params */
2863 45570 : child_params = finalize_plan(root,
2864 45570 : plan->righttree,
2865 : gather_param,
2866 : bms_union(nestloop_params, valid_params),
2867 : scan_params);
2868 : /* ... and they don't count as parameters used at my level */
2869 45570 : child_params = bms_difference(child_params, nestloop_params);
2870 45570 : bms_free(nestloop_params);
2871 : }
2872 : else
2873 : {
2874 : /* easy case */
2875 577300 : child_params = finalize_plan(root,
2876 577300 : plan->righttree,
2877 : gather_param,
2878 : valid_params,
2879 : scan_params);
2880 : }
2881 622870 : context.paramids = bms_add_members(context.paramids, child_params);
2882 :
2883 : /*
2884 : * Any locally generated parameter doesn't count towards its generating
2885 : * plan node's external dependencies. (Note: if we changed valid_params
2886 : * and/or scan_params, we leak those bitmapsets; not worth the notational
2887 : * trouble to clean them up.)
2888 : */
2889 622870 : if (locally_added_param >= 0)
2890 : {
2891 101712 : context.paramids = bms_del_member(context.paramids,
2892 : locally_added_param);
2893 : }
2894 :
2895 : /* Now we have all the paramids referenced in this node and children */
2896 :
2897 622870 : if (!bms_is_subset(context.paramids, valid_params))
2898 0 : elog(ERROR, "plan should not reference subplan's variable");
2899 :
2900 : /*
2901 : * The plan node's allParam and extParam fields should include all its
2902 : * referenced paramids, plus contributions from any child initPlans.
2903 : * However, any setParams of the initPlans should not be present in the
2904 : * parent node's extParams, only in its allParams. (It's possible that
2905 : * some initPlans have extParams that are setParams of other initPlans.)
2906 : */
2907 :
2908 : /* allParam must include initplans' extParams and setParams */
2909 622870 : plan->allParam = bms_union(context.paramids, initExtParam);
2910 622870 : plan->allParam = bms_add_members(plan->allParam, initSetParam);
2911 : /* extParam must include any initplan extParams */
2912 622870 : plan->extParam = bms_union(context.paramids, initExtParam);
2913 : /* but not any initplan setParams */
2914 622870 : plan->extParam = bms_del_members(plan->extParam, initSetParam);
2915 :
2916 622870 : return plan->allParam;
2917 : }
2918 :
2919 : /*
2920 : * finalize_primnode: add IDs of all PARAM_EXEC params that appear (or will
2921 : * appear) in the given expression tree to the result set.
2922 : */
2923 : static bool
2924 10761522 : finalize_primnode(Node *node, finalize_primnode_context *context)
2925 : {
2926 10761522 : if (node == NULL)
2927 1271398 : return false;
2928 9490124 : if (IsA(node, Param))
2929 : {
2930 128726 : if (((Param *) node)->paramkind == PARAM_EXEC)
2931 : {
2932 124900 : int paramid = ((Param *) node)->paramid;
2933 :
2934 124900 : context->paramids = bms_add_member(context->paramids, paramid);
2935 : }
2936 128726 : return false; /* no more to do here */
2937 : }
2938 9361398 : else if (IsA(node, Aggref))
2939 : {
2940 : /*
2941 : * Check to see if the aggregate will be replaced by a Param
2942 : * referencing a subquery output during setrefs.c. If so, we must
2943 : * account for that Param here. (For various reasons, it's not
2944 : * convenient to perform that substitution earlier than setrefs.c, nor
2945 : * to perform this processing after setrefs.c. Thus we need a wart
2946 : * here.)
2947 : */
2948 13378 : Aggref *aggref = (Aggref *) node;
2949 : Param *aggparam;
2950 :
2951 13378 : aggparam = find_minmax_agg_replacement_param(context->root, aggref);
2952 13378 : if (aggparam != NULL)
2953 556 : context->paramids = bms_add_member(context->paramids,
2954 : aggparam->paramid);
2955 : /* Fall through to examine the agg's arguments */
2956 : }
2957 9348020 : else if (IsA(node, SubPlan))
2958 : {
2959 36910 : SubPlan *subplan = (SubPlan *) node;
2960 36910 : Plan *plan = planner_subplan_get_plan(context->root, subplan);
2961 : ListCell *lc;
2962 : Bitmapset *subparamids;
2963 :
2964 : /* Recurse into the testexpr, but not into the Plan */
2965 36910 : finalize_primnode(subplan->testexpr, context);
2966 :
2967 : /*
2968 : * Remove any param IDs of output parameters of the subplan that were
2969 : * referenced in the testexpr. These are not interesting for
2970 : * parameter change signaling since we always re-evaluate the subplan.
2971 : * Note that this wouldn't work too well if there might be uses of the
2972 : * same param IDs elsewhere in the plan, but that can't happen because
2973 : * generate_new_exec_param never tries to merge params.
2974 : */
2975 40012 : foreach(lc, subplan->paramIds)
2976 : {
2977 3102 : context->paramids = bms_del_member(context->paramids,
2978 : lfirst_int(lc));
2979 : }
2980 :
2981 : /* Also examine args list */
2982 36910 : finalize_primnode((Node *) subplan->args, context);
2983 :
2984 : /*
2985 : * Add params needed by the subplan to paramids, but excluding those
2986 : * we will pass down to it. (We assume SS_finalize_plan was run on
2987 : * the subplan already.)
2988 : */
2989 36910 : subparamids = bms_copy(plan->extParam);
2990 89204 : foreach(lc, subplan->parParam)
2991 : {
2992 52294 : subparamids = bms_del_member(subparamids, lfirst_int(lc));
2993 : }
2994 36910 : context->paramids = bms_join(context->paramids, subparamids);
2995 :
2996 36910 : return false; /* no more to do here */
2997 : }
2998 9324488 : return expression_tree_walker(node, finalize_primnode, context);
2999 : }
3000 :
3001 : /*
3002 : * finalize_agg_primnode: find all Aggref nodes in the given expression tree,
3003 : * and add IDs of all PARAM_EXEC params appearing within their aggregated
3004 : * arguments to the result set.
3005 : */
3006 : static bool
3007 10566 : finalize_agg_primnode(Node *node, finalize_primnode_context *context)
3008 : {
3009 10566 : if (node == NULL)
3010 1402 : return false;
3011 9164 : if (IsA(node, Aggref))
3012 : {
3013 1144 : Aggref *agg = (Aggref *) node;
3014 :
3015 : /* we should not consider the direct arguments, if any */
3016 1144 : finalize_primnode((Node *) agg->args, context);
3017 1144 : finalize_primnode((Node *) agg->aggfilter, context);
3018 1144 : return false; /* there can't be any Aggrefs below here */
3019 : }
3020 8020 : return expression_tree_walker(node, finalize_agg_primnode, context);
3021 : }
3022 :
3023 : /*
3024 : * SS_make_initplan_output_param - make a Param for an initPlan's output
3025 : *
3026 : * The plan is expected to return a scalar value of the given type/collation.
3027 : *
3028 : * Note that in some cases the initplan may not ever appear in the finished
3029 : * plan tree. If that happens, we'll have wasted a PARAM_EXEC slot, which
3030 : * is no big deal.
3031 : */
3032 : Param *
3033 446 : SS_make_initplan_output_param(PlannerInfo *root,
3034 : Oid resulttype, int32 resulttypmod,
3035 : Oid resultcollation)
3036 : {
3037 446 : return generate_new_exec_param(root, resulttype,
3038 : resulttypmod, resultcollation);
3039 : }
3040 :
3041 : /*
3042 : * SS_make_initplan_from_plan - given a plan tree, make it an InitPlan
3043 : *
3044 : * We build an EXPR_SUBLINK SubPlan node and put it into the initplan
3045 : * list for the outer query level. A Param that represents the initplan's
3046 : * output has already been assigned using SS_make_initplan_output_param.
3047 : */
3048 : void
3049 400 : SS_make_initplan_from_plan(PlannerInfo *root,
3050 : PlannerInfo *subroot, Plan *plan,
3051 : Param *prm)
3052 : {
3053 : SubPlan *node;
3054 :
3055 : /*
3056 : * Add the subplan and its PlannerInfo, as well as a dummy path entry, to
3057 : * the global lists. Ideally we'd save a real path, but right now our
3058 : * sole caller doesn't build a path that exactly matches the plan. Since
3059 : * we're not currently going to need the path for an initplan, it's not
3060 : * worth requiring construction of such a path.
3061 : */
3062 400 : root->glob->subplans = lappend(root->glob->subplans, plan);
3063 400 : root->glob->subpaths = lappend(root->glob->subpaths, NULL);
3064 400 : root->glob->subroots = lappend(root->glob->subroots, subroot);
3065 :
3066 : /*
3067 : * Create a SubPlan node and add it to the outer list of InitPlans. Note
3068 : * it has to appear after any other InitPlans it might depend on (see
3069 : * comments in ExecReScan).
3070 : */
3071 400 : node = makeNode(SubPlan);
3072 400 : node->subLinkType = EXPR_SUBLINK;
3073 400 : node->plan_id = list_length(root->glob->subplans);
3074 400 : node->plan_name = psprintf("InitPlan %d", node->plan_id);
3075 400 : get_first_col_type(plan, &node->firstColType, &node->firstColTypmod,
3076 : &node->firstColCollation);
3077 400 : node->parallel_safe = plan->parallel_safe;
3078 400 : node->setParam = list_make1_int(prm->paramid);
3079 :
3080 400 : root->init_plans = lappend(root->init_plans, node);
3081 :
3082 : /*
3083 : * The node can't have any inputs (since it's an initplan), so the
3084 : * parParam and args lists remain empty.
3085 : */
3086 :
3087 : /* Set costs of SubPlan using info from the plan tree */
3088 400 : cost_subplan(subroot, node, plan);
3089 400 : }
|
