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) 1996-2020, 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 : int refcount; /* number of remaining references */
65 : Query *ctequery; /* query to substitute */
66 : } inline_cte_walker_context;
67 :
68 :
69 : static Node *build_subplan(PlannerInfo *root, Plan *plan, PlannerInfo *subroot,
70 : List *plan_params,
71 : SubLinkType subLinkType, int subLinkId,
72 : Node *testexpr, bool adjust_testexpr,
73 : bool unknownEqFalse);
74 : static List *generate_subquery_params(PlannerInfo *root, List *tlist,
75 : List **paramIds);
76 : static List *generate_subquery_vars(PlannerInfo *root, List *tlist,
77 : Index varno);
78 : static Node *convert_testexpr(PlannerInfo *root,
79 : Node *testexpr,
80 : List *subst_nodes);
81 : static Node *convert_testexpr_mutator(Node *node,
82 : convert_testexpr_context *context);
83 : static bool subplan_is_hashable(Plan *plan);
84 : static bool testexpr_is_hashable(Node *testexpr);
85 : static bool hash_ok_operator(OpExpr *expr);
86 : static bool contain_dml(Node *node);
87 : static bool contain_dml_walker(Node *node, void *context);
88 : static bool contain_outer_selfref(Node *node);
89 : static bool contain_outer_selfref_walker(Node *node, Index *depth);
90 : static void inline_cte(PlannerInfo *root, CommonTableExpr *cte);
91 : static bool inline_cte_walker(Node *node, inline_cte_walker_context *context);
92 : static bool simplify_EXISTS_query(PlannerInfo *root, Query *query);
93 : static Query *convert_EXISTS_to_ANY(PlannerInfo *root, Query *subselect,
94 : Node **testexpr, List **paramIds);
95 : static Node *replace_correlation_vars_mutator(Node *node, PlannerInfo *root);
96 : static Node *process_sublinks_mutator(Node *node,
97 : process_sublinks_context *context);
98 : static Bitmapset *finalize_plan(PlannerInfo *root,
99 : Plan *plan,
100 : int gather_param,
101 : Bitmapset *valid_params,
102 : Bitmapset *scan_params);
103 : static bool finalize_primnode(Node *node, finalize_primnode_context *context);
104 : static bool finalize_agg_primnode(Node *node, finalize_primnode_context *context);
105 :
106 :
107 : /*
108 : * Get the datatype/typmod/collation of the first column of the plan's output.
109 : *
110 : * This information is stored for ARRAY_SUBLINK execution and for
111 : * exprType()/exprTypmod()/exprCollation(), which have no way to get at the
112 : * plan associated with a SubPlan node. We really only need the info for
113 : * EXPR_SUBLINK and ARRAY_SUBLINK subplans, but for consistency we save it
114 : * always.
115 : */
116 : static void
117 48860 : get_first_col_type(Plan *plan, Oid *coltype, int32 *coltypmod,
118 : Oid *colcollation)
119 : {
120 : /* In cases such as EXISTS, tlist might be empty; arbitrarily use VOID */
121 48860 : if (plan->targetlist)
122 : {
123 47032 : TargetEntry *tent = linitial_node(TargetEntry, plan->targetlist);
124 :
125 47032 : if (!tent->resjunk)
126 : {
127 47032 : *coltype = exprType((Node *) tent->expr);
128 47032 : *coltypmod = exprTypmod((Node *) tent->expr);
129 47032 : *colcollation = exprCollation((Node *) tent->expr);
130 47032 : return;
131 : }
132 : }
133 1828 : *coltype = VOIDOID;
134 1828 : *coltypmod = -1;
135 1828 : *colcollation = InvalidOid;
136 : }
137 :
138 : /*
139 : * Convert a SubLink (as created by the parser) into a SubPlan.
140 : *
141 : * We are given the SubLink's contained query, type, ID, and testexpr. We are
142 : * also told if this expression appears at top level of a WHERE/HAVING qual.
143 : *
144 : * Note: we assume that the testexpr has been AND/OR flattened (actually,
145 : * it's been through eval_const_expressions), but not converted to
146 : * implicit-AND form; and any SubLinks in it should already have been
147 : * converted to SubPlans. The subquery is as yet untouched, however.
148 : *
149 : * The result is whatever we need to substitute in place of the SubLink node
150 : * in the executable expression. If we're going to do the subplan as a
151 : * regular subplan, this will be the constructed SubPlan node. If we're going
152 : * to do the subplan as an InitPlan, the SubPlan node instead goes into
153 : * root->init_plans, and what we return here is an expression tree
154 : * representing the InitPlan's result: usually just a Param node representing
155 : * a single scalar result, but possibly a row comparison tree containing
156 : * multiple Param nodes, or for a MULTIEXPR subquery a simple NULL constant
157 : * (since the real output Params are elsewhere in the tree, and the MULTIEXPR
158 : * subquery itself is in a resjunk tlist entry whose value is uninteresting).
159 : */
160 : static Node *
161 46622 : make_subplan(PlannerInfo *root, Query *orig_subquery,
162 : SubLinkType subLinkType, int subLinkId,
163 : Node *testexpr, bool isTopQual)
164 : {
165 : Query *subquery;
166 46622 : bool simple_exists = false;
167 : double tuple_fraction;
168 : PlannerInfo *subroot;
169 : RelOptInfo *final_rel;
170 : Path *best_path;
171 : Plan *plan;
172 : List *plan_params;
173 : Node *result;
174 :
175 : /*
176 : * Copy the source Query node. This is a quick and dirty kluge to resolve
177 : * the fact that the parser can generate trees with multiple links to the
178 : * same sub-Query node, but the planner wants to scribble on the Query.
179 : * Try to clean this up when we do querytree redesign...
180 : */
181 46622 : subquery = copyObject(orig_subquery);
182 :
183 : /*
184 : * If it's an EXISTS subplan, we might be able to simplify it.
185 : */
186 46622 : if (subLinkType == EXISTS_SUBLINK)
187 1676 : simple_exists = simplify_EXISTS_query(root, subquery);
188 :
189 : /*
190 : * For an EXISTS subplan, tell lower-level planner to expect that only the
191 : * first tuple will be retrieved. For ALL and ANY subplans, we will be
192 : * able to stop evaluating if the test condition fails or matches, so very
193 : * often not all the tuples will be retrieved; for lack of a better idea,
194 : * specify 50% retrieval. For EXPR, MULTIEXPR, and ROWCOMPARE subplans,
195 : * use default behavior (we're only expecting one row out, anyway).
196 : *
197 : * NOTE: if you change these numbers, also change cost_subplan() in
198 : * path/costsize.c.
199 : *
200 : * XXX If an ANY subplan is uncorrelated, build_subplan may decide to hash
201 : * its output. In that case it would've been better to specify full
202 : * retrieval. At present, however, we can only check hashability after
203 : * we've made the subplan :-(. (Determining whether it'll fit in work_mem
204 : * is the really hard part.) Therefore, we don't want to be too
205 : * optimistic about the percentage of tuples retrieved, for fear of
206 : * selecting a plan that's bad for the materialization case.
207 : */
208 46622 : if (subLinkType == EXISTS_SUBLINK)
209 1676 : tuple_fraction = 1.0; /* just like a LIMIT 1 */
210 44946 : else if (subLinkType == ALL_SUBLINK ||
211 : subLinkType == ANY_SUBLINK)
212 214 : tuple_fraction = 0.5; /* 50% */
213 : else
214 44732 : tuple_fraction = 0.0; /* default behavior */
215 :
216 : /* plan_params should not be in use in current query level */
217 : Assert(root->plan_params == NIL);
218 :
219 : /* Generate Paths for the subquery */
220 46622 : subroot = subquery_planner(root->glob, subquery,
221 : root,
222 : false, tuple_fraction);
223 :
224 : /* Isolate the params needed by this specific subplan */
225 46622 : plan_params = root->plan_params;
226 46622 : 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 46622 : final_rel = fetch_upper_rel(subroot, UPPERREL_FINAL, NULL);
233 46622 : best_path = get_cheapest_fractional_path(final_rel, tuple_fraction);
234 :
235 46622 : plan = create_plan(subroot, best_path);
236 :
237 : /* And convert to SubPlan or InitPlan format. */
238 46622 : result = build_subplan(root, plan, subroot, plan_params,
239 : subLinkType, subLinkId,
240 : testexpr, true, isTopQual);
241 :
242 : /*
243 : * If it's a correlated EXISTS with an unimportant targetlist, we might be
244 : * able to transform it to the equivalent of an IN and then implement it
245 : * by hashing. We don't have enough information yet to tell which way is
246 : * likely to be better (it depends on the expected number of executions of
247 : * the EXISTS qual, and we are much too early in planning the outer query
248 : * to be able to guess that). So we generate both plans, if possible, and
249 : * leave it to the executor to decide which to use.
250 : */
251 46622 : if (simple_exists && IsA(result, SubPlan))
252 : {
253 : Node *newtestexpr;
254 : List *paramIds;
255 :
256 : /* Make a second copy of the original subquery */
257 1500 : subquery = copyObject(orig_subquery);
258 : /* and re-simplify */
259 1500 : simple_exists = simplify_EXISTS_query(root, subquery);
260 : Assert(simple_exists);
261 : /* See if it can be converted to an ANY query */
262 1500 : subquery = convert_EXISTS_to_ANY(root, subquery,
263 : &newtestexpr, ¶mIds);
264 1500 : if (subquery)
265 : {
266 : /* Generate Paths for the ANY subquery; we'll need all rows */
267 1174 : subroot = subquery_planner(root->glob, subquery,
268 : root,
269 : false, 0.0);
270 :
271 : /* Isolate the params needed by this specific subplan */
272 1174 : plan_params = root->plan_params;
273 1174 : root->plan_params = NIL;
274 :
275 : /* Select best Path and turn it into a Plan */
276 1174 : final_rel = fetch_upper_rel(subroot, UPPERREL_FINAL, NULL);
277 1174 : best_path = final_rel->cheapest_total_path;
278 :
279 1174 : plan = create_plan(subroot, best_path);
280 :
281 : /* Now we can check if it'll fit in work_mem */
282 : /* XXX can we check this at the Path stage? */
283 1174 : if (subplan_is_hashable(plan))
284 : {
285 : SubPlan *hashplan;
286 : AlternativeSubPlan *asplan;
287 :
288 : /* OK, convert to SubPlan format. */
289 1174 : hashplan = castNode(SubPlan,
290 : build_subplan(root, plan, subroot,
291 : plan_params,
292 : ANY_SUBLINK, 0,
293 : newtestexpr,
294 : false, true));
295 : /* Check we got what we expected */
296 : Assert(hashplan->parParam == NIL);
297 : Assert(hashplan->useHashTable);
298 : /* build_subplan won't have filled in paramIds */
299 1174 : hashplan->paramIds = paramIds;
300 :
301 : /* Leave it to the executor to decide which plan to use */
302 1174 : asplan = makeNode(AlternativeSubPlan);
303 1174 : asplan->subplans = list_make2(result, hashplan);
304 1174 : result = (Node *) asplan;
305 : }
306 : }
307 : }
308 :
309 46622 : 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 47796 : build_subplan(PlannerInfo *root, Plan *plan, PlannerInfo *subroot,
320 : List *plan_params,
321 : SubLinkType subLinkType, int subLinkId,
322 : Node *testexpr, bool adjust_testexpr,
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 47796 : splan = makeNode(SubPlan);
335 47796 : splan->subLinkType = subLinkType;
336 47796 : splan->testexpr = NULL;
337 47796 : splan->paramIds = NIL;
338 47796 : get_first_col_type(plan, &splan->firstColType, &splan->firstColTypmod,
339 : &splan->firstColCollation);
340 47796 : splan->useHashTable = false;
341 47796 : splan->unknownEqFalse = unknownEqFalse;
342 47796 : splan->parallel_safe = plan->parallel_safe;
343 47796 : splan->setParam = NIL;
344 47796 : splan->parParam = NIL;
345 47796 : 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 103440 : foreach(lc, plan_params)
352 : {
353 55644 : PlannerParamItem *pitem = (PlannerParamItem *) lfirst(lc);
354 55644 : Node *arg = pitem->item;
355 :
356 : /*
357 : * The Var, PlaceHolderVar, or Aggref has already been adjusted to
358 : * have the correct varlevelsup, phlevelsup, or agglevelsup.
359 : *
360 : * If it's a PlaceHolderVar or Aggref, its arguments might contain
361 : * SubLinks, which have not yet been processed (see the comments for
362 : * SS_replace_correlation_vars). Do that now.
363 : */
364 55644 : if (IsA(arg, PlaceHolderVar) ||
365 55636 : IsA(arg, Aggref))
366 40 : arg = SS_process_sublinks(root, arg, false);
367 :
368 55644 : splan->parParam = lappend_int(splan->parParam, pitem->paramId);
369 55644 : splan->args = lappend(splan->args, arg);
370 : }
371 :
372 : /*
373 : * Un-correlated or undirect correlated plans of EXISTS, EXPR, ARRAY,
374 : * ROWCOMPARE, or MULTIEXPR types can be used as initPlans. For EXISTS,
375 : * EXPR, or ARRAY, we return a Param referring to the result of evaluating
376 : * the initPlan. For ROWCOMPARE, we must modify the testexpr tree to
377 : * contain PARAM_EXEC Params instead of the PARAM_SUBLINK Params emitted
378 : * by the parser, and then return that tree. For MULTIEXPR, we return a
379 : * null constant: the resjunk targetlist item containing the SubLink does
380 : * not need to return anything useful, since the referencing Params are
381 : * elsewhere.
382 : */
383 47796 : if (splan->parParam == NIL && subLinkType == EXISTS_SUBLINK)
384 158 : {
385 : Param *prm;
386 :
387 : Assert(testexpr == NULL);
388 158 : prm = generate_new_exec_param(root, BOOLOID, -1, InvalidOid);
389 158 : splan->setParam = list_make1_int(prm->paramid);
390 158 : isInitPlan = true;
391 158 : result = (Node *) prm;
392 : }
393 47638 : else if (splan->parParam == NIL && subLinkType == EXPR_SUBLINK)
394 8986 : {
395 8986 : TargetEntry *te = linitial(plan->targetlist);
396 : Param *prm;
397 :
398 : Assert(!te->resjunk);
399 : Assert(testexpr == NULL);
400 8986 : prm = generate_new_exec_param(root,
401 8986 : exprType((Node *) te->expr),
402 8986 : exprTypmod((Node *) te->expr),
403 8986 : exprCollation((Node *) te->expr));
404 8986 : splan->setParam = list_make1_int(prm->paramid);
405 8986 : isInitPlan = true;
406 8986 : result = (Node *) prm;
407 : }
408 38652 : else if (splan->parParam == NIL && subLinkType == ARRAY_SUBLINK)
409 58 : {
410 58 : TargetEntry *te = linitial(plan->targetlist);
411 : Oid arraytype;
412 : Param *prm;
413 :
414 : Assert(!te->resjunk);
415 : Assert(testexpr == NULL);
416 58 : arraytype = get_promoted_array_type(exprType((Node *) te->expr));
417 58 : if (!OidIsValid(arraytype))
418 0 : elog(ERROR, "could not find array type for datatype %s",
419 : format_type_be(exprType((Node *) te->expr)));
420 58 : prm = generate_new_exec_param(root,
421 : arraytype,
422 58 : exprTypmod((Node *) te->expr),
423 58 : exprCollation((Node *) te->expr));
424 58 : splan->setParam = list_make1_int(prm->paramid);
425 58 : isInitPlan = true;
426 58 : result = (Node *) prm;
427 : }
428 38594 : else if (splan->parParam == NIL && subLinkType == ROWCOMPARE_SUBLINK)
429 0 : {
430 : /* Adjust the Params */
431 : List *params;
432 :
433 : Assert(testexpr != NULL);
434 0 : params = generate_subquery_params(root,
435 : plan->targetlist,
436 : &splan->paramIds);
437 0 : result = convert_testexpr(root,
438 : testexpr,
439 : params);
440 0 : splan->setParam = list_copy(splan->paramIds);
441 0 : isInitPlan = true;
442 :
443 : /*
444 : * The executable expression is returned to become part of the outer
445 : * plan's expression tree; it is not kept in the initplan node.
446 : */
447 : }
448 38594 : else if (subLinkType == MULTIEXPR_SUBLINK)
449 : {
450 : /*
451 : * Whether it's an initplan or not, it needs to set a PARAM_EXEC Param
452 : * for each output column.
453 : */
454 : List *params;
455 :
456 : Assert(testexpr == NULL);
457 58 : params = generate_subquery_params(root,
458 : plan->targetlist,
459 : &splan->setParam);
460 :
461 : /*
462 : * Save the list of replacement Params in the n'th cell of
463 : * root->multiexpr_params; setrefs.c will use it to replace
464 : * PARAM_MULTIEXPR Params.
465 : */
466 116 : while (list_length(root->multiexpr_params) < subLinkId)
467 58 : root->multiexpr_params = lappend(root->multiexpr_params, NIL);
468 58 : lc = list_nth_cell(root->multiexpr_params, subLinkId - 1);
469 : Assert(lfirst(lc) == NIL);
470 58 : lfirst(lc) = params;
471 :
472 : /* It can be an initplan if there are no parParams. */
473 58 : if (splan->parParam == NIL)
474 : {
475 16 : isInitPlan = true;
476 16 : result = (Node *) makeNullConst(RECORDOID, -1, InvalidOid);
477 : }
478 : else
479 : {
480 42 : isInitPlan = false;
481 42 : result = (Node *) splan;
482 : }
483 : }
484 : else
485 : {
486 : /*
487 : * Adjust the Params in the testexpr, unless caller said it's not
488 : * needed.
489 : */
490 38536 : if (testexpr && adjust_testexpr)
491 214 : {
492 : List *params;
493 :
494 214 : params = generate_subquery_params(root,
495 : plan->targetlist,
496 : &splan->paramIds);
497 214 : splan->testexpr = convert_testexpr(root,
498 : testexpr,
499 : params);
500 : }
501 : else
502 38322 : splan->testexpr = testexpr;
503 :
504 : /*
505 : * We can't convert subplans of ALL_SUBLINK or ANY_SUBLINK types to
506 : * initPlans, even when they are uncorrelated or undirect correlated,
507 : * because we need to scan the output of the subplan for each outer
508 : * tuple. But if it's a not-direct-correlated IN (= ANY) test, we
509 : * might be able to use a hashtable to avoid comparing all the tuples.
510 : */
511 38536 : if (subLinkType == ANY_SUBLINK &&
512 2704 : splan->parParam == NIL &&
513 2656 : subplan_is_hashable(plan) &&
514 1328 : testexpr_is_hashable(splan->testexpr))
515 1328 : splan->useHashTable = true;
516 :
517 : /*
518 : * Otherwise, we have the option to tack a Material node onto the top
519 : * of the subplan, to reduce the cost of reading it repeatedly. This
520 : * is pointless for a direct-correlated subplan, since we'd have to
521 : * recompute its results each time anyway. For uncorrelated/undirect
522 : * correlated subplans, we add Material unless the subplan's top plan
523 : * node would materialize its output anyway. Also, if enable_material
524 : * is false, then the user does not want us to materialize anything
525 : * unnecessarily, so we don't.
526 : */
527 37208 : else if (splan->parParam == NIL && enable_material &&
528 12 : !ExecMaterializesOutput(nodeTag(plan)))
529 12 : plan = materialize_finished_plan(plan);
530 :
531 38536 : result = (Node *) splan;
532 38536 : isInitPlan = false;
533 : }
534 :
535 : /*
536 : * Add the subplan and its PlannerInfo to the global lists.
537 : */
538 47796 : root->glob->subplans = lappend(root->glob->subplans, plan);
539 47796 : root->glob->subroots = lappend(root->glob->subroots, subroot);
540 47796 : splan->plan_id = list_length(root->glob->subplans);
541 :
542 47796 : if (isInitPlan)
543 9218 : root->init_plans = lappend(root->init_plans, splan);
544 :
545 : /*
546 : * A parameterless subplan (not initplan) should be prepared to handle
547 : * REWIND efficiently. If it has direct parameters then there's no point
548 : * since it'll be reset on each scan anyway; and if it's an initplan then
549 : * there's no point since it won't get re-run without parameter changes
550 : * anyway. The input of a hashed subplan doesn't need REWIND either.
551 : */
552 47796 : if (splan->parParam == NIL && !isInitPlan && !splan->useHashTable)
553 12 : root->glob->rewindPlanIDs = bms_add_member(root->glob->rewindPlanIDs,
554 : splan->plan_id);
555 :
556 : /* Label the subplan for EXPLAIN purposes */
557 47796 : splan->plan_name = palloc(32 + 12 * list_length(splan->setParam));
558 47796 : sprintf(splan->plan_name, "%s %d",
559 : isInitPlan ? "InitPlan" : "SubPlan",
560 : splan->plan_id);
561 47796 : if (splan->setParam)
562 : {
563 9260 : char *ptr = splan->plan_name + strlen(splan->plan_name);
564 :
565 9260 : ptr += sprintf(ptr, " (returns ");
566 18576 : foreach(lc, splan->setParam)
567 : {
568 9316 : ptr += sprintf(ptr, "$%d%s",
569 : lfirst_int(lc),
570 9316 : lnext(splan->setParam, lc) ? "," : ")");
571 : }
572 : }
573 :
574 : /* Lastly, fill in the cost estimates for use later */
575 47796 : cost_subplan(root, splan, plan);
576 :
577 47796 : return result;
578 : }
579 :
580 : /*
581 : * generate_subquery_params: build a list of Params representing the output
582 : * columns of a sublink's sub-select, given the sub-select's targetlist.
583 : *
584 : * We also return an integer list of the paramids of the Params.
585 : */
586 : static List *
587 272 : generate_subquery_params(PlannerInfo *root, List *tlist, List **paramIds)
588 : {
589 : List *result;
590 : List *ids;
591 : ListCell *lc;
592 :
593 272 : result = ids = NIL;
594 636 : foreach(lc, tlist)
595 : {
596 364 : TargetEntry *tent = (TargetEntry *) lfirst(lc);
597 : Param *param;
598 :
599 364 : if (tent->resjunk)
600 4 : continue;
601 :
602 360 : param = generate_new_exec_param(root,
603 360 : exprType((Node *) tent->expr),
604 360 : exprTypmod((Node *) tent->expr),
605 360 : exprCollation((Node *) tent->expr));
606 360 : result = lappend(result, param);
607 360 : ids = lappend_int(ids, param->paramid);
608 : }
609 :
610 272 : *paramIds = ids;
611 272 : return result;
612 : }
613 :
614 : /*
615 : * generate_subquery_vars: build a list of Vars representing the output
616 : * columns of a sublink's sub-select, given the sub-select's targetlist.
617 : * The Vars have the specified varno (RTE index).
618 : */
619 : static List *
620 584 : generate_subquery_vars(PlannerInfo *root, List *tlist, Index varno)
621 : {
622 : List *result;
623 : ListCell *lc;
624 :
625 584 : result = NIL;
626 1196 : foreach(lc, tlist)
627 : {
628 612 : TargetEntry *tent = (TargetEntry *) lfirst(lc);
629 : Var *var;
630 :
631 612 : if (tent->resjunk)
632 0 : continue;
633 :
634 612 : var = makeVarFromTargetEntry(varno, tent);
635 612 : result = lappend(result, var);
636 : }
637 :
638 584 : return result;
639 : }
640 :
641 : /*
642 : * convert_testexpr: convert the testexpr given by the parser into
643 : * actually executable form. This entails replacing PARAM_SUBLINK Params
644 : * with Params or Vars representing the results of the sub-select. The
645 : * nodes to be substituted are passed in as the List result from
646 : * generate_subquery_params or generate_subquery_vars.
647 : */
648 : static Node *
649 798 : convert_testexpr(PlannerInfo *root,
650 : Node *testexpr,
651 : List *subst_nodes)
652 : {
653 : convert_testexpr_context context;
654 :
655 798 : context.root = root;
656 798 : context.subst_nodes = subst_nodes;
657 798 : return convert_testexpr_mutator(testexpr, &context);
658 : }
659 :
660 : static Node *
661 3984 : convert_testexpr_mutator(Node *node,
662 : convert_testexpr_context *context)
663 : {
664 3984 : if (node == NULL)
665 12 : return NULL;
666 3972 : if (IsA(node, Param))
667 : {
668 858 : Param *param = (Param *) node;
669 :
670 858 : if (param->paramkind == PARAM_SUBLINK)
671 : {
672 1716 : if (param->paramid <= 0 ||
673 858 : param->paramid > list_length(context->subst_nodes))
674 0 : elog(ERROR, "unexpected PARAM_SUBLINK ID: %d", param->paramid);
675 :
676 : /*
677 : * We copy the list item to avoid having doubly-linked
678 : * substructure in the modified parse tree. This is probably
679 : * unnecessary when it's a Param, but be safe.
680 : */
681 858 : return (Node *) copyObject(list_nth(context->subst_nodes,
682 : param->paramid - 1));
683 : }
684 : }
685 3114 : if (IsA(node, SubLink))
686 : {
687 : /*
688 : * If we come across a nested SubLink, it is neither necessary nor
689 : * correct to recurse into it: any PARAM_SUBLINKs we might find inside
690 : * belong to the inner SubLink not the outer. So just return it as-is.
691 : *
692 : * This reasoning depends on the assumption that nothing will pull
693 : * subexpressions into or out of the testexpr field of a SubLink, at
694 : * least not without replacing PARAM_SUBLINKs first. If we did want
695 : * to do that we'd need to rethink the parser-output representation
696 : * altogether, since currently PARAM_SUBLINKs are only unique per
697 : * SubLink not globally across the query. The whole point of
698 : * replacing them with Vars or PARAM_EXEC nodes is to make them
699 : * globally unique before they escape from the SubLink's testexpr.
700 : *
701 : * Note: this can't happen when called during SS_process_sublinks,
702 : * because that recursively processes inner SubLinks first. It can
703 : * happen when called from convert_ANY_sublink_to_join, though.
704 : */
705 8 : return node;
706 : }
707 3106 : return expression_tree_mutator(node,
708 : convert_testexpr_mutator,
709 : (void *) context);
710 : }
711 :
712 : /*
713 : * subplan_is_hashable: can we implement an ANY subplan by hashing?
714 : */
715 : static bool
716 2502 : subplan_is_hashable(Plan *plan)
717 : {
718 : double subquery_size;
719 :
720 : /*
721 : * The estimated size of the subquery result must fit in work_mem. (Note:
722 : * we use heap tuple overhead here even though the tuples will actually be
723 : * stored as MinimalTuples; this provides some fudge factor for hashtable
724 : * overhead.)
725 : */
726 5004 : subquery_size = plan->plan_rows *
727 2502 : (MAXALIGN(plan->plan_width) + MAXALIGN(SizeofHeapTupleHeader));
728 2502 : if (subquery_size > work_mem * 1024L)
729 0 : return false;
730 :
731 2502 : return true;
732 : }
733 :
734 : /*
735 : * testexpr_is_hashable: is an ANY SubLink's test expression hashable?
736 : */
737 : static bool
738 1328 : testexpr_is_hashable(Node *testexpr)
739 : {
740 : /*
741 : * The testexpr must be a single OpExpr, or an AND-clause containing only
742 : * OpExprs.
743 : *
744 : * The combining operators must be hashable and strict. The need for
745 : * hashability is obvious, since we want to use hashing. Without
746 : * strictness, behavior in the presence of nulls is too unpredictable. We
747 : * actually must assume even more than plain strictness: they can't yield
748 : * NULL for non-null inputs, either (see nodeSubplan.c). However, hash
749 : * indexes and hash joins assume that too.
750 : */
751 1328 : if (testexpr && IsA(testexpr, OpExpr))
752 : {
753 800 : if (hash_ok_operator((OpExpr *) testexpr))
754 800 : return true;
755 : }
756 528 : else if (is_andclause(testexpr))
757 : {
758 : ListCell *l;
759 :
760 1584 : foreach(l, ((BoolExpr *) testexpr)->args)
761 : {
762 1056 : Node *andarg = (Node *) lfirst(l);
763 :
764 1056 : if (!IsA(andarg, OpExpr))
765 0 : return false;
766 1056 : if (!hash_ok_operator((OpExpr *) andarg))
767 0 : return false;
768 : }
769 528 : return true;
770 : }
771 :
772 0 : return false;
773 : }
774 :
775 : /*
776 : * Check expression is hashable + strict
777 : *
778 : * We could use op_hashjoinable() and op_strict(), but do it like this to
779 : * avoid a redundant cache lookup.
780 : */
781 : static bool
782 6312 : hash_ok_operator(OpExpr *expr)
783 : {
784 6312 : Oid opid = expr->opno;
785 :
786 : /* quick out if not a binary operator */
787 6312 : if (list_length(expr->args) != 2)
788 0 : return false;
789 6312 : if (opid == ARRAY_EQ_OP)
790 : {
791 : /* array_eq is strict, but must check input type to ensure hashable */
792 : /* XXX record_eq will need same treatment when it becomes hashable */
793 0 : Node *leftarg = linitial(expr->args);
794 :
795 0 : return op_hashjoinable(opid, exprType(leftarg));
796 : }
797 : else
798 : {
799 : /* else must look up the operator properties */
800 : HeapTuple tup;
801 : Form_pg_operator optup;
802 :
803 6312 : tup = SearchSysCache1(OPEROID, ObjectIdGetDatum(opid));
804 6312 : if (!HeapTupleIsValid(tup))
805 0 : elog(ERROR, "cache lookup failed for operator %u", opid);
806 6312 : optup = (Form_pg_operator) GETSTRUCT(tup);
807 6312 : if (!optup->oprcanhash || !func_strict(optup->oprcode))
808 : {
809 324 : ReleaseSysCache(tup);
810 324 : return false;
811 : }
812 5988 : ReleaseSysCache(tup);
813 5988 : return true;
814 : }
815 : }
816 :
817 :
818 : /*
819 : * SS_process_ctes: process a query's WITH list
820 : *
821 : * Consider each CTE in the WITH list and either ignore it (if it's an
822 : * unreferenced SELECT), "inline" it to create a regular sub-SELECT-in-FROM,
823 : * or convert it to an initplan.
824 : *
825 : * A side effect is to fill in root->cte_plan_ids with a list that
826 : * parallels root->parse->cteList and provides the subplan ID for
827 : * each CTE's initplan, or a dummy ID (-1) if we didn't make an initplan.
828 : */
829 : void
830 1222 : SS_process_ctes(PlannerInfo *root)
831 : {
832 : ListCell *lc;
833 :
834 : Assert(root->cte_plan_ids == NIL);
835 :
836 2612 : foreach(lc, root->parse->cteList)
837 : {
838 1390 : CommonTableExpr *cte = (CommonTableExpr *) lfirst(lc);
839 1390 : CmdType cmdType = ((Query *) cte->ctequery)->commandType;
840 : Query *subquery;
841 : PlannerInfo *subroot;
842 : RelOptInfo *final_rel;
843 : Path *best_path;
844 : Plan *plan;
845 : SubPlan *splan;
846 : int paramid;
847 :
848 : /*
849 : * Ignore SELECT CTEs that are not actually referenced anywhere.
850 : */
851 1390 : if (cte->cterefcount == 0 && cmdType == CMD_SELECT)
852 : {
853 : /* Make a dummy entry in cte_plan_ids */
854 8 : root->cte_plan_ids = lappend_int(root->cte_plan_ids, -1);
855 8 : continue;
856 : }
857 :
858 : /*
859 : * Consider inlining the CTE (creating RTE_SUBQUERY RTE(s)) instead of
860 : * implementing it as a separately-planned CTE.
861 : *
862 : * We cannot inline if any of these conditions hold:
863 : *
864 : * 1. The user said not to (the CTEMaterializeAlways option).
865 : *
866 : * 2. The CTE is recursive.
867 : *
868 : * 3. The CTE has side-effects; this includes either not being a plain
869 : * SELECT, or containing volatile functions. Inlining might change
870 : * the side-effects, which would be bad.
871 : *
872 : * 4. The CTE is multiply-referenced and contains a self-reference to
873 : * a recursive CTE outside itself. Inlining would result in multiple
874 : * recursive self-references, which we don't support.
875 : *
876 : * Otherwise, we have an option whether to inline or not. That should
877 : * always be a win if there's just a single reference, but if the CTE
878 : * is multiply-referenced then it's unclear: inlining adds duplicate
879 : * computations, but the ability to absorb restrictions from the outer
880 : * query level could outweigh that. We do not have nearly enough
881 : * information at this point to tell whether that's true, so we let
882 : * the user express a preference. Our default behavior is to inline
883 : * only singly-referenced CTEs, but a CTE marked CTEMaterializeNever
884 : * will be inlined even if multiply referenced.
885 : *
886 : * Note: we check for volatile functions last, because that's more
887 : * expensive than the other tests needed.
888 : */
889 1382 : if ((cte->ctematerialized == CTEMaterializeNever ||
890 1362 : (cte->ctematerialized == CTEMaterializeDefault &&
891 1290 : cte->cterefcount == 1)) &&
892 1146 : !cte->cterecursive &&
893 696 : cmdType == CMD_SELECT &&
894 696 : !contain_dml(cte->ctequery) &&
895 690 : (cte->cterefcount <= 1 ||
896 12 : !contain_outer_selfref(cte->ctequery)) &&
897 682 : !contain_volatile_functions(cte->ctequery))
898 : {
899 678 : inline_cte(root, cte);
900 : /* Make a dummy entry in cte_plan_ids */
901 678 : root->cte_plan_ids = lappend_int(root->cte_plan_ids, -1);
902 678 : continue;
903 : }
904 :
905 : /*
906 : * Copy the source Query node. Probably not necessary, but let's keep
907 : * this similar to make_subplan.
908 : */
909 704 : subquery = (Query *) copyObject(cte->ctequery);
910 :
911 : /* plan_params should not be in use in current query level */
912 : Assert(root->plan_params == NIL);
913 :
914 : /*
915 : * Generate Paths for the CTE query. Always plan for full retrieval
916 : * --- we don't have enough info to predict otherwise.
917 : */
918 704 : subroot = subquery_planner(root->glob, subquery,
919 : root,
920 704 : cte->cterecursive, 0.0);
921 :
922 : /*
923 : * Since the current query level doesn't yet contain any RTEs, it
924 : * should not be possible for the CTE to have requested parameters of
925 : * this level.
926 : */
927 704 : if (root->plan_params)
928 0 : elog(ERROR, "unexpected outer reference in CTE query");
929 :
930 : /*
931 : * Select best Path and turn it into a Plan. At least for now, there
932 : * seems no reason to postpone doing that.
933 : */
934 704 : final_rel = fetch_upper_rel(subroot, UPPERREL_FINAL, NULL);
935 704 : best_path = final_rel->cheapest_total_path;
936 :
937 704 : plan = create_plan(subroot, best_path);
938 :
939 : /*
940 : * Make a SubPlan node for it. This is just enough unlike
941 : * build_subplan that we can't share code.
942 : *
943 : * Note plan_id, plan_name, and cost fields are set further down.
944 : */
945 704 : splan = makeNode(SubPlan);
946 704 : splan->subLinkType = CTE_SUBLINK;
947 704 : splan->testexpr = NULL;
948 704 : splan->paramIds = NIL;
949 704 : get_first_col_type(plan, &splan->firstColType, &splan->firstColTypmod,
950 : &splan->firstColCollation);
951 704 : splan->useHashTable = false;
952 704 : splan->unknownEqFalse = false;
953 :
954 : /*
955 : * CTE scans are not considered for parallelism (cf
956 : * set_rel_consider_parallel), and even if they were, initPlans aren't
957 : * parallel-safe.
958 : */
959 704 : splan->parallel_safe = false;
960 704 : splan->setParam = NIL;
961 704 : splan->parParam = NIL;
962 704 : splan->args = NIL;
963 :
964 : /*
965 : * The node can't have any inputs (since it's an initplan), so the
966 : * parParam and args lists remain empty. (It could contain references
967 : * to earlier CTEs' output param IDs, but CTE outputs are not
968 : * propagated via the args list.)
969 : */
970 :
971 : /*
972 : * Assign a param ID to represent the CTE's output. No ordinary
973 : * "evaluation" of this param slot ever happens, but we use the param
974 : * ID for setParam/chgParam signaling just as if the CTE plan were
975 : * returning a simple scalar output. (Also, the executor abuses the
976 : * ParamExecData slot for this param ID for communication among
977 : * multiple CteScan nodes that might be scanning this CTE.)
978 : */
979 704 : paramid = assign_special_exec_param(root);
980 704 : splan->setParam = list_make1_int(paramid);
981 :
982 : /*
983 : * Add the subplan and its PlannerInfo to the global lists.
984 : */
985 704 : root->glob->subplans = lappend(root->glob->subplans, plan);
986 704 : root->glob->subroots = lappend(root->glob->subroots, subroot);
987 704 : splan->plan_id = list_length(root->glob->subplans);
988 :
989 704 : root->init_plans = lappend(root->init_plans, splan);
990 :
991 704 : root->cte_plan_ids = lappend_int(root->cte_plan_ids, splan->plan_id);
992 :
993 : /* Label the subplan for EXPLAIN purposes */
994 704 : splan->plan_name = psprintf("CTE %s", cte->ctename);
995 :
996 : /* Lastly, fill in the cost estimates for use later */
997 704 : cost_subplan(root, splan, plan);
998 : }
999 1222 : }
1000 :
1001 : /*
1002 : * contain_dml: is any subquery not a plain SELECT?
1003 : *
1004 : * We reject SELECT FOR UPDATE/SHARE as well as INSERT etc.
1005 : */
1006 : static bool
1007 696 : contain_dml(Node *node)
1008 : {
1009 696 : return contain_dml_walker(node, NULL);
1010 : }
1011 :
1012 : static bool
1013 45400 : contain_dml_walker(Node *node, void *context)
1014 : {
1015 45400 : if (node == NULL)
1016 11070 : return false;
1017 34330 : if (IsA(node, Query))
1018 : {
1019 924 : Query *query = (Query *) node;
1020 :
1021 924 : if (query->commandType != CMD_SELECT ||
1022 924 : query->rowMarks != NIL)
1023 6 : return true;
1024 :
1025 918 : return query_tree_walker(query, contain_dml_walker, context, 0);
1026 : }
1027 33406 : return expression_tree_walker(node, contain_dml_walker, context);
1028 : }
1029 :
1030 : /*
1031 : * contain_outer_selfref: is there an external recursive self-reference?
1032 : */
1033 : static bool
1034 12 : contain_outer_selfref(Node *node)
1035 : {
1036 12 : Index depth = 0;
1037 :
1038 : /*
1039 : * We should be starting with a Query, so that depth will be 1 while
1040 : * examining its immediate contents.
1041 : */
1042 : Assert(IsA(node, Query));
1043 :
1044 12 : return contain_outer_selfref_walker(node, &depth);
1045 : }
1046 :
1047 : static bool
1048 300 : contain_outer_selfref_walker(Node *node, Index *depth)
1049 : {
1050 300 : if (node == NULL)
1051 156 : return false;
1052 144 : if (IsA(node, RangeTblEntry))
1053 : {
1054 16 : RangeTblEntry *rte = (RangeTblEntry *) node;
1055 :
1056 : /*
1057 : * Check for a self-reference to a CTE that's above the Query that our
1058 : * search started at.
1059 : */
1060 16 : if (rte->rtekind == RTE_CTE &&
1061 8 : rte->self_reference &&
1062 8 : rte->ctelevelsup >= *depth)
1063 8 : return true;
1064 8 : return false; /* allow range_table_walker to continue */
1065 : }
1066 128 : if (IsA(node, Query))
1067 : {
1068 : /* Recurse into subquery, tracking nesting depth properly */
1069 16 : Query *query = (Query *) node;
1070 : bool result;
1071 :
1072 16 : (*depth)++;
1073 :
1074 16 : result = query_tree_walker(query, contain_outer_selfref_walker,
1075 : (void *) depth, QTW_EXAMINE_RTES_BEFORE);
1076 :
1077 16 : (*depth)--;
1078 :
1079 16 : return result;
1080 : }
1081 112 : return expression_tree_walker(node, contain_outer_selfref_walker,
1082 : (void *) depth);
1083 : }
1084 :
1085 : /*
1086 : * inline_cte: convert RTE_CTE references to given CTE into RTE_SUBQUERYs
1087 : */
1088 : static void
1089 678 : inline_cte(PlannerInfo *root, CommonTableExpr *cte)
1090 : {
1091 : struct inline_cte_walker_context context;
1092 :
1093 678 : context.ctename = cte->ctename;
1094 : /* Start at levelsup = -1 because we'll immediately increment it */
1095 678 : context.levelsup = -1;
1096 678 : context.refcount = cte->cterefcount;
1097 678 : context.ctequery = castNode(Query, cte->ctequery);
1098 :
1099 678 : (void) inline_cte_walker((Node *) root->parse, &context);
1100 :
1101 : /* Assert we replaced all references */
1102 : Assert(context.refcount == 0);
1103 678 : }
1104 :
1105 : static bool
1106 123100 : inline_cte_walker(Node *node, inline_cte_walker_context *context)
1107 : {
1108 123100 : if (node == NULL)
1109 34984 : return false;
1110 88116 : if (IsA(node, Query))
1111 : {
1112 3158 : Query *query = (Query *) node;
1113 :
1114 3158 : context->levelsup++;
1115 :
1116 : /*
1117 : * Visit the query's RTE nodes after their contents; otherwise
1118 : * query_tree_walker would descend into the newly inlined CTE query,
1119 : * which we don't want.
1120 : */
1121 3158 : (void) query_tree_walker(query, inline_cte_walker, context,
1122 : QTW_EXAMINE_RTES_AFTER);
1123 :
1124 3158 : context->levelsup--;
1125 :
1126 3158 : return false;
1127 : }
1128 84958 : else if (IsA(node, RangeTblEntry))
1129 : {
1130 4948 : RangeTblEntry *rte = (RangeTblEntry *) node;
1131 :
1132 4948 : if (rte->rtekind == RTE_CTE &&
1133 1094 : strcmp(rte->ctename, context->ctename) == 0 &&
1134 682 : rte->ctelevelsup == context->levelsup)
1135 : {
1136 : /*
1137 : * Found a reference to replace. Generate a copy of the CTE query
1138 : * with appropriate level adjustment for outer references (e.g.,
1139 : * to other CTEs).
1140 : */
1141 682 : Query *newquery = copyObject(context->ctequery);
1142 :
1143 682 : if (context->levelsup > 0)
1144 518 : IncrementVarSublevelsUp((Node *) newquery, context->levelsup, 1);
1145 :
1146 : /*
1147 : * Convert the RTE_CTE RTE into a RTE_SUBQUERY.
1148 : *
1149 : * Historically, a FOR UPDATE clause has been treated as extending
1150 : * into views and subqueries, but not into CTEs. We preserve this
1151 : * distinction by not trying to push rowmarks into the new
1152 : * subquery.
1153 : */
1154 682 : rte->rtekind = RTE_SUBQUERY;
1155 682 : rte->subquery = newquery;
1156 682 : rte->security_barrier = false;
1157 :
1158 : /* Zero out CTE-specific fields */
1159 682 : rte->ctename = NULL;
1160 682 : rte->ctelevelsup = 0;
1161 682 : rte->self_reference = false;
1162 682 : rte->coltypes = NIL;
1163 682 : rte->coltypmods = NIL;
1164 682 : rte->colcollations = NIL;
1165 :
1166 : /* Count the number of replacements we've done */
1167 682 : context->refcount--;
1168 : }
1169 :
1170 4948 : return false;
1171 : }
1172 :
1173 80010 : return expression_tree_walker(node, inline_cte_walker, context);
1174 : }
1175 :
1176 :
1177 : /*
1178 : * convert_ANY_sublink_to_join: try to convert an ANY SubLink to a join
1179 : *
1180 : * The caller has found an ANY SubLink at the top level of one of the query's
1181 : * qual clauses, but has not checked the properties of the SubLink further.
1182 : * Decide whether it is appropriate to process this SubLink in join style.
1183 : * If so, form a JoinExpr and return it. Return NULL if the SubLink cannot
1184 : * be converted to a join.
1185 : *
1186 : * The only non-obvious input parameter is available_rels: this is the set
1187 : * of query rels that can safely be referenced in the sublink expression.
1188 : * (We must restrict this to avoid changing the semantics when a sublink
1189 : * is present in an outer join's ON qual.) The conversion must fail if
1190 : * the converted qual would reference any but these parent-query relids.
1191 : *
1192 : * On success, the returned JoinExpr has larg = NULL and rarg = the jointree
1193 : * item representing the pulled-up subquery. The caller must set larg to
1194 : * represent the relation(s) on the lefthand side of the new join, and insert
1195 : * the JoinExpr into the upper query's jointree at an appropriate place
1196 : * (typically, where the lefthand relation(s) had been). Note that the
1197 : * passed-in SubLink must also be removed from its original position in the
1198 : * query quals, since the quals of the returned JoinExpr replace it.
1199 : * (Notionally, we replace the SubLink with a constant TRUE, then elide the
1200 : * redundant constant from the qual.)
1201 : *
1202 : * On success, the caller is also responsible for recursively applying
1203 : * pull_up_sublinks processing to the rarg and quals of the returned JoinExpr.
1204 : * (On failure, there is no need to do anything, since pull_up_sublinks will
1205 : * be applied when we recursively plan the sub-select.)
1206 : *
1207 : * Side effects of a successful conversion include adding the SubLink's
1208 : * subselect to the query's rangetable, so that it can be referenced in
1209 : * the JoinExpr's rarg.
1210 : */
1211 : JoinExpr *
1212 632 : convert_ANY_sublink_to_join(PlannerInfo *root, SubLink *sublink,
1213 : Relids available_rels)
1214 : {
1215 : JoinExpr *result;
1216 632 : Query *parse = root->parse;
1217 632 : Query *subselect = (Query *) sublink->subselect;
1218 : Relids upper_varnos;
1219 : int rtindex;
1220 : ParseNamespaceItem *nsitem;
1221 : RangeTblEntry *rte;
1222 : RangeTblRef *rtr;
1223 : List *subquery_vars;
1224 : Node *quals;
1225 : ParseState *pstate;
1226 :
1227 : Assert(sublink->subLinkType == ANY_SUBLINK);
1228 :
1229 : /*
1230 : * The sub-select must not refer to any Vars of the parent query. (Vars of
1231 : * higher levels should be okay, though.)
1232 : */
1233 632 : if (contain_vars_of_level((Node *) subselect, 1))
1234 40 : return NULL;
1235 :
1236 : /*
1237 : * The test expression must contain some Vars of the parent query, else
1238 : * it's not gonna be a join. (Note that it won't have Vars referring to
1239 : * the subquery, rather Params.)
1240 : */
1241 592 : upper_varnos = pull_varnos(sublink->testexpr);
1242 592 : if (bms_is_empty(upper_varnos))
1243 8 : return NULL;
1244 :
1245 : /*
1246 : * However, it can't refer to anything outside available_rels.
1247 : */
1248 584 : if (!bms_is_subset(upper_varnos, available_rels))
1249 0 : return NULL;
1250 :
1251 : /*
1252 : * The combining operators and left-hand expressions mustn't be volatile.
1253 : */
1254 584 : if (contain_volatile_functions(sublink->testexpr))
1255 0 : return NULL;
1256 :
1257 : /* Create a dummy ParseState for addRangeTableEntryForSubquery */
1258 584 : pstate = make_parsestate(NULL);
1259 :
1260 : /*
1261 : * Okay, pull up the sub-select into upper range table.
1262 : *
1263 : * We rely here on the assumption that the outer query has no references
1264 : * to the inner (necessarily true, other than the Vars that we build
1265 : * below). Therefore this is a lot easier than what pull_up_subqueries has
1266 : * to go through.
1267 : */
1268 584 : nsitem = addRangeTableEntryForSubquery(pstate,
1269 : subselect,
1270 : makeAlias("ANY_subquery", NIL),
1271 : false,
1272 : false);
1273 584 : rte = nsitem->p_rte;
1274 584 : parse->rtable = lappend(parse->rtable, rte);
1275 584 : rtindex = list_length(parse->rtable);
1276 :
1277 : /*
1278 : * Form a RangeTblRef for the pulled-up sub-select.
1279 : */
1280 584 : rtr = makeNode(RangeTblRef);
1281 584 : rtr->rtindex = rtindex;
1282 :
1283 : /*
1284 : * Build a list of Vars representing the subselect outputs.
1285 : */
1286 584 : subquery_vars = generate_subquery_vars(root,
1287 : subselect->targetList,
1288 : rtindex);
1289 :
1290 : /*
1291 : * Build the new join's qual expression, replacing Params with these Vars.
1292 : */
1293 584 : quals = convert_testexpr(root, sublink->testexpr, subquery_vars);
1294 :
1295 : /*
1296 : * And finally, build the JoinExpr node.
1297 : */
1298 584 : result = makeNode(JoinExpr);
1299 584 : result->jointype = JOIN_SEMI;
1300 584 : result->isNatural = false;
1301 584 : result->larg = NULL; /* caller must fill this in */
1302 584 : result->rarg = (Node *) rtr;
1303 584 : result->usingClause = NIL;
1304 584 : result->quals = quals;
1305 584 : result->alias = NULL;
1306 584 : result->rtindex = 0; /* we don't need an RTE for it */
1307 :
1308 584 : return result;
1309 : }
1310 :
1311 : /*
1312 : * convert_EXISTS_sublink_to_join: try to convert an EXISTS SubLink to a join
1313 : *
1314 : * The API of this function is identical to convert_ANY_sublink_to_join's,
1315 : * except that we also support the case where the caller has found NOT EXISTS,
1316 : * so we need an additional input parameter "under_not".
1317 : */
1318 : JoinExpr *
1319 8034 : convert_EXISTS_sublink_to_join(PlannerInfo *root, SubLink *sublink,
1320 : bool under_not, Relids available_rels)
1321 : {
1322 : JoinExpr *result;
1323 8034 : Query *parse = root->parse;
1324 8034 : Query *subselect = (Query *) sublink->subselect;
1325 : Node *whereClause;
1326 : int rtoffset;
1327 : int varno;
1328 : Relids clause_varnos;
1329 : Relids upper_varnos;
1330 :
1331 : Assert(sublink->subLinkType == EXISTS_SUBLINK);
1332 :
1333 : /*
1334 : * Can't flatten if it contains WITH. (We could arrange to pull up the
1335 : * WITH into the parent query's cteList, but that risks changing the
1336 : * semantics, since a WITH ought to be executed once per associated query
1337 : * call.) Note that convert_ANY_sublink_to_join doesn't have to reject
1338 : * this case, since it just produces a subquery RTE that doesn't have to
1339 : * get flattened into the parent query.
1340 : */
1341 8034 : if (subselect->cteList)
1342 0 : return NULL;
1343 :
1344 : /*
1345 : * Copy the subquery so we can modify it safely (see comments in
1346 : * make_subplan).
1347 : */
1348 8034 : subselect = copyObject(subselect);
1349 :
1350 : /*
1351 : * See if the subquery can be simplified based on the knowledge that it's
1352 : * being used in EXISTS(). If we aren't able to get rid of its
1353 : * targetlist, we have to fail, because the pullup operation leaves us
1354 : * with noplace to evaluate the targetlist.
1355 : */
1356 8034 : if (!simplify_EXISTS_query(root, subselect))
1357 18 : return NULL;
1358 :
1359 : /*
1360 : * Separate out the WHERE clause. (We could theoretically also remove
1361 : * top-level plain JOIN/ON clauses, but it's probably not worth the
1362 : * trouble.)
1363 : */
1364 8016 : whereClause = subselect->jointree->quals;
1365 8016 : subselect->jointree->quals = NULL;
1366 :
1367 : /*
1368 : * The rest of the sub-select must not refer to any Vars of the parent
1369 : * query. (Vars of higher levels should be okay, though.)
1370 : */
1371 8016 : if (contain_vars_of_level((Node *) subselect, 1))
1372 4 : return NULL;
1373 :
1374 : /*
1375 : * On the other hand, the WHERE clause must contain some Vars of the
1376 : * parent query, else it's not gonna be a join.
1377 : */
1378 8012 : if (!contain_vars_of_level(whereClause, 1))
1379 40 : return NULL;
1380 :
1381 : /*
1382 : * We don't risk optimizing if the WHERE clause is volatile, either.
1383 : */
1384 7972 : if (contain_volatile_functions(whereClause))
1385 0 : return NULL;
1386 :
1387 : /*
1388 : * The subquery must have a nonempty jointree, but we can make it so.
1389 : */
1390 7972 : replace_empty_jointree(subselect);
1391 :
1392 : /*
1393 : * Prepare to pull up the sub-select into top range table.
1394 : *
1395 : * We rely here on the assumption that the outer query has no references
1396 : * to the inner (necessarily true). Therefore this is a lot easier than
1397 : * what pull_up_subqueries has to go through.
1398 : *
1399 : * In fact, it's even easier than what convert_ANY_sublink_to_join has to
1400 : * do. The machinations of simplify_EXISTS_query ensured that there is
1401 : * nothing interesting in the subquery except an rtable and jointree, and
1402 : * even the jointree FromExpr no longer has quals. So we can just append
1403 : * the rtable to our own and use the FromExpr in our jointree. But first,
1404 : * adjust all level-zero varnos in the subquery to account for the rtable
1405 : * merger.
1406 : */
1407 7972 : rtoffset = list_length(parse->rtable);
1408 7972 : OffsetVarNodes((Node *) subselect, rtoffset, 0);
1409 7972 : OffsetVarNodes(whereClause, rtoffset, 0);
1410 :
1411 : /*
1412 : * Upper-level vars in subquery will now be one level closer to their
1413 : * parent than before; in particular, anything that had been level 1
1414 : * becomes level zero.
1415 : */
1416 7972 : IncrementVarSublevelsUp((Node *) subselect, -1, 1);
1417 7972 : IncrementVarSublevelsUp(whereClause, -1, 1);
1418 :
1419 : /*
1420 : * Now that the WHERE clause is adjusted to match the parent query
1421 : * environment, we can easily identify all the level-zero rels it uses.
1422 : * The ones <= rtoffset belong to the upper query; the ones > rtoffset do
1423 : * not.
1424 : */
1425 7972 : clause_varnos = pull_varnos(whereClause);
1426 7972 : upper_varnos = NULL;
1427 23932 : while ((varno = bms_first_member(clause_varnos)) >= 0)
1428 : {
1429 15960 : if (varno <= rtoffset)
1430 7988 : upper_varnos = bms_add_member(upper_varnos, varno);
1431 : }
1432 7972 : bms_free(clause_varnos);
1433 : Assert(!bms_is_empty(upper_varnos));
1434 :
1435 : /*
1436 : * Now that we've got the set of upper-level varnos, we can make the last
1437 : * check: only available_rels can be referenced.
1438 : */
1439 7972 : if (!bms_is_subset(upper_varnos, available_rels))
1440 0 : return NULL;
1441 :
1442 : /* Now we can attach the modified subquery rtable to the parent */
1443 7972 : parse->rtable = list_concat(parse->rtable, subselect->rtable);
1444 :
1445 : /*
1446 : * And finally, build the JoinExpr node.
1447 : */
1448 7972 : result = makeNode(JoinExpr);
1449 7972 : result->jointype = under_not ? JOIN_ANTI : JOIN_SEMI;
1450 7972 : result->isNatural = false;
1451 7972 : result->larg = NULL; /* caller must fill this in */
1452 : /* flatten out the FromExpr node if it's useless */
1453 7972 : if (list_length(subselect->jointree->fromlist) == 1)
1454 7968 : result->rarg = (Node *) linitial(subselect->jointree->fromlist);
1455 : else
1456 4 : result->rarg = (Node *) subselect->jointree;
1457 7972 : result->usingClause = NIL;
1458 7972 : result->quals = whereClause;
1459 7972 : result->alias = NULL;
1460 7972 : result->rtindex = 0; /* we don't need an RTE for it */
1461 :
1462 7972 : return result;
1463 : }
1464 :
1465 : /*
1466 : * simplify_EXISTS_query: remove any useless stuff in an EXISTS's subquery
1467 : *
1468 : * The only thing that matters about an EXISTS query is whether it returns
1469 : * zero or more than zero rows. Therefore, we can remove certain SQL features
1470 : * that won't affect that. The only part that is really likely to matter in
1471 : * typical usage is simplifying the targetlist: it's a common habit to write
1472 : * "SELECT * FROM" even though there is no need to evaluate any columns.
1473 : *
1474 : * Note: by suppressing the targetlist we could cause an observable behavioral
1475 : * change, namely that any errors that might occur in evaluating the tlist
1476 : * won't occur, nor will other side-effects of volatile functions. This seems
1477 : * unlikely to bother anyone in practice.
1478 : *
1479 : * Returns true if was able to discard the targetlist, else false.
1480 : */
1481 : static bool
1482 11210 : simplify_EXISTS_query(PlannerInfo *root, Query *query)
1483 : {
1484 : /*
1485 : * We don't try to simplify at all if the query uses set operations,
1486 : * aggregates, grouping sets, SRFs, modifying CTEs, HAVING, OFFSET, or FOR
1487 : * UPDATE/SHARE; none of these seem likely in normal usage and their
1488 : * possible effects are complex. (Note: we could ignore an "OFFSET 0"
1489 : * clause, but that traditionally is used as an optimization fence, so we
1490 : * don't.)
1491 : */
1492 11210 : if (query->commandType != CMD_SELECT ||
1493 11210 : query->setOperations ||
1494 11210 : query->hasAggs ||
1495 11210 : query->groupingSets ||
1496 11210 : query->hasWindowFuncs ||
1497 11210 : query->hasTargetSRFs ||
1498 11210 : query->hasModifyingCTE ||
1499 11210 : query->havingQual ||
1500 11210 : query->limitOffset ||
1501 11210 : query->rowMarks)
1502 28 : return false;
1503 :
1504 : /*
1505 : * LIMIT with a constant positive (or NULL) value doesn't affect the
1506 : * semantics of EXISTS, so let's ignore such clauses. This is worth doing
1507 : * because people accustomed to certain other DBMSes may be in the habit
1508 : * of writing EXISTS(SELECT ... LIMIT 1) as an optimization. If there's a
1509 : * LIMIT with anything else as argument, though, we can't simplify.
1510 : */
1511 11182 : if (query->limitCount)
1512 : {
1513 : /*
1514 : * The LIMIT clause has not yet been through eval_const_expressions,
1515 : * so we have to apply that here. It might seem like this is a waste
1516 : * of cycles, since the only case plausibly worth worrying about is
1517 : * "LIMIT 1" ... but what we'll actually see is "LIMIT int8(1::int4)",
1518 : * so we have to fold constants or we're not going to recognize it.
1519 : */
1520 16 : Node *node = eval_const_expressions(root, query->limitCount);
1521 : Const *limit;
1522 :
1523 : /* Might as well update the query if we simplified the clause. */
1524 16 : query->limitCount = node;
1525 :
1526 16 : if (!IsA(node, Const))
1527 0 : return false;
1528 :
1529 16 : limit = (Const *) node;
1530 : Assert(limit->consttype == INT8OID);
1531 16 : if (!limit->constisnull && DatumGetInt64(limit->constvalue) <= 0)
1532 8 : return false;
1533 :
1534 : /* Whether or not the targetlist is safe, we can drop the LIMIT. */
1535 8 : query->limitCount = NULL;
1536 : }
1537 :
1538 : /*
1539 : * Otherwise, we can throw away the targetlist, as well as any GROUP,
1540 : * WINDOW, DISTINCT, and ORDER BY clauses; none of those clauses will
1541 : * change a nonzero-rows result to zero rows or vice versa. (Furthermore,
1542 : * since our parsetree representation of these clauses depends on the
1543 : * targetlist, we'd better throw them away if we drop the targetlist.)
1544 : */
1545 11174 : query->targetList = NIL;
1546 11174 : query->groupClause = NIL;
1547 11174 : query->windowClause = NIL;
1548 11174 : query->distinctClause = NIL;
1549 11174 : query->sortClause = NIL;
1550 11174 : query->hasDistinctOn = false;
1551 :
1552 11174 : return true;
1553 : }
1554 :
1555 : /*
1556 : * convert_EXISTS_to_ANY: try to convert EXISTS to a hashable ANY sublink
1557 : *
1558 : * The subselect is expected to be a fresh copy that we can munge up,
1559 : * and to have been successfully passed through simplify_EXISTS_query.
1560 : *
1561 : * On success, the modified subselect is returned, and we store a suitable
1562 : * upper-level test expression at *testexpr, plus a list of the subselect's
1563 : * output Params at *paramIds. (The test expression is already Param-ified
1564 : * and hence need not go through convert_testexpr, which is why we have to
1565 : * deal with the Param IDs specially.)
1566 : *
1567 : * On failure, returns NULL.
1568 : */
1569 : static Query *
1570 1500 : convert_EXISTS_to_ANY(PlannerInfo *root, Query *subselect,
1571 : Node **testexpr, List **paramIds)
1572 : {
1573 : Node *whereClause;
1574 : List *leftargs,
1575 : *rightargs,
1576 : *opids,
1577 : *opcollations,
1578 : *newWhere,
1579 : *tlist,
1580 : *testlist,
1581 : *paramids;
1582 : ListCell *lc,
1583 : *rc,
1584 : *oc,
1585 : *cc;
1586 : AttrNumber resno;
1587 :
1588 : /*
1589 : * Query must not require a targetlist, since we have to insert a new one.
1590 : * Caller should have dealt with the case already.
1591 : */
1592 : Assert(subselect->targetList == NIL);
1593 :
1594 : /*
1595 : * Separate out the WHERE clause. (We could theoretically also remove
1596 : * top-level plain JOIN/ON clauses, but it's probably not worth the
1597 : * trouble.)
1598 : */
1599 1500 : whereClause = subselect->jointree->quals;
1600 1500 : subselect->jointree->quals = NULL;
1601 :
1602 : /*
1603 : * The rest of the sub-select must not refer to any Vars of the parent
1604 : * query. (Vars of higher levels should be okay, though.)
1605 : *
1606 : * Note: we need not check for Aggrefs separately because we know the
1607 : * sub-select is as yet unoptimized; any uplevel Aggref must therefore
1608 : * contain an uplevel Var reference. This is not the case below ...
1609 : */
1610 1500 : if (contain_vars_of_level((Node *) subselect, 1))
1611 166 : return NULL;
1612 :
1613 : /*
1614 : * We don't risk optimizing if the WHERE clause is volatile, either.
1615 : */
1616 1334 : if (contain_volatile_functions(whereClause))
1617 0 : return NULL;
1618 :
1619 : /*
1620 : * Clean up the WHERE clause by doing const-simplification etc on it.
1621 : * Aside from simplifying the processing we're about to do, this is
1622 : * important for being able to pull chunks of the WHERE clause up into the
1623 : * parent query. Since we are invoked partway through the parent's
1624 : * preprocess_expression() work, earlier steps of preprocess_expression()
1625 : * wouldn't get applied to the pulled-up stuff unless we do them here. For
1626 : * the parts of the WHERE clause that get put back into the child query,
1627 : * this work is partially duplicative, but it shouldn't hurt.
1628 : *
1629 : * Note: we do not run flatten_join_alias_vars. This is OK because any
1630 : * parent aliases were flattened already, and we're not going to pull any
1631 : * child Vars (of any description) into the parent.
1632 : *
1633 : * Note: passing the parent's root to eval_const_expressions is
1634 : * technically wrong, but we can get away with it since only the
1635 : * boundParams (if any) are used, and those would be the same in a
1636 : * subroot.
1637 : */
1638 1334 : whereClause = eval_const_expressions(root, whereClause);
1639 1334 : whereClause = (Node *) canonicalize_qual((Expr *) whereClause, false);
1640 1334 : whereClause = (Node *) make_ands_implicit((Expr *) whereClause);
1641 :
1642 : /*
1643 : * We now have a flattened implicit-AND list of clauses, which we try to
1644 : * break apart into "outervar = innervar" hash clauses. Anything that
1645 : * can't be broken apart just goes back into the newWhere list. Note that
1646 : * we aren't trying hard yet to ensure that we have only outer or only
1647 : * inner on each side; we'll check that if we get to the end.
1648 : */
1649 1334 : leftargs = rightargs = opids = opcollations = newWhere = NIL;
1650 5444 : foreach(lc, (List *) whereClause)
1651 : {
1652 4110 : OpExpr *expr = (OpExpr *) lfirst(lc);
1653 :
1654 7068 : if (IsA(expr, OpExpr) &&
1655 2958 : hash_ok_operator(expr))
1656 : {
1657 2634 : Node *leftarg = (Node *) linitial(expr->args);
1658 2634 : Node *rightarg = (Node *) lsecond(expr->args);
1659 :
1660 2634 : if (contain_vars_of_level(leftarg, 1))
1661 : {
1662 172 : leftargs = lappend(leftargs, leftarg);
1663 172 : rightargs = lappend(rightargs, rightarg);
1664 172 : opids = lappend_oid(opids, expr->opno);
1665 172 : opcollations = lappend_oid(opcollations, expr->inputcollid);
1666 172 : continue;
1667 : }
1668 2462 : if (contain_vars_of_level(rightarg, 1))
1669 : {
1670 : /*
1671 : * We must commute the clause to put the outer var on the
1672 : * left, because the hashing code in nodeSubplan.c expects
1673 : * that. This probably shouldn't ever fail, since hashable
1674 : * operators ought to have commutators, but be paranoid.
1675 : */
1676 1498 : expr->opno = get_commutator(expr->opno);
1677 1498 : if (OidIsValid(expr->opno) && hash_ok_operator(expr))
1678 : {
1679 1498 : leftargs = lappend(leftargs, rightarg);
1680 1498 : rightargs = lappend(rightargs, leftarg);
1681 1498 : opids = lappend_oid(opids, expr->opno);
1682 1498 : opcollations = lappend_oid(opcollations, expr->inputcollid);
1683 1498 : continue;
1684 : }
1685 : /* If no commutator, no chance to optimize the WHERE clause */
1686 0 : return NULL;
1687 : }
1688 : }
1689 : /* Couldn't handle it as a hash clause */
1690 2440 : newWhere = lappend(newWhere, expr);
1691 : }
1692 :
1693 : /*
1694 : * If we didn't find anything we could convert, fail.
1695 : */
1696 1334 : if (leftargs == NIL)
1697 160 : return NULL;
1698 :
1699 : /*
1700 : * There mustn't be any parent Vars or Aggs in the stuff that we intend to
1701 : * put back into the child query. Note: you might think we don't need to
1702 : * check for Aggs separately, because an uplevel Agg must contain an
1703 : * uplevel Var in its argument. But it is possible that the uplevel Var
1704 : * got optimized away by eval_const_expressions. Consider
1705 : *
1706 : * SUM(CASE WHEN false THEN uplevelvar ELSE 0 END)
1707 : */
1708 2348 : if (contain_vars_of_level((Node *) newWhere, 1) ||
1709 1174 : contain_vars_of_level((Node *) rightargs, 1))
1710 0 : return NULL;
1711 1186 : if (root->parse->hasAggs &&
1712 24 : (contain_aggs_of_level((Node *) newWhere, 1) ||
1713 12 : contain_aggs_of_level((Node *) rightargs, 1)))
1714 0 : return NULL;
1715 :
1716 : /*
1717 : * And there can't be any child Vars in the stuff we intend to pull up.
1718 : * (Note: we'd need to check for child Aggs too, except we know the child
1719 : * has no aggs at all because of simplify_EXISTS_query's check. The same
1720 : * goes for window functions.)
1721 : */
1722 1174 : if (contain_vars_of_level((Node *) leftargs, 0))
1723 0 : return NULL;
1724 :
1725 : /*
1726 : * Also reject sublinks in the stuff we intend to pull up. (It might be
1727 : * possible to support this, but doesn't seem worth the complication.)
1728 : */
1729 1174 : if (contain_subplans((Node *) leftargs))
1730 0 : return NULL;
1731 :
1732 : /*
1733 : * Okay, adjust the sublevelsup in the stuff we're pulling up.
1734 : */
1735 1174 : IncrementVarSublevelsUp((Node *) leftargs, -1, 1);
1736 :
1737 : /*
1738 : * Put back any child-level-only WHERE clauses.
1739 : */
1740 1174 : if (newWhere)
1741 1070 : subselect->jointree->quals = (Node *) make_ands_explicit(newWhere);
1742 :
1743 : /*
1744 : * Build a new targetlist for the child that emits the expressions we
1745 : * need. Concurrently, build a testexpr for the parent using Params to
1746 : * reference the child outputs. (Since we generate Params directly here,
1747 : * there will be no need to convert the testexpr in build_subplan.)
1748 : */
1749 1174 : tlist = testlist = paramids = NIL;
1750 1174 : resno = 1;
1751 2844 : forfour(lc, leftargs, rc, rightargs, oc, opids, cc, opcollations)
1752 : {
1753 1670 : Node *leftarg = (Node *) lfirst(lc);
1754 1670 : Node *rightarg = (Node *) lfirst(rc);
1755 1670 : Oid opid = lfirst_oid(oc);
1756 1670 : Oid opcollation = lfirst_oid(cc);
1757 : Param *param;
1758 :
1759 1670 : param = generate_new_exec_param(root,
1760 : exprType(rightarg),
1761 : exprTypmod(rightarg),
1762 : exprCollation(rightarg));
1763 1670 : tlist = lappend(tlist,
1764 1670 : makeTargetEntry((Expr *) rightarg,
1765 1670 : resno++,
1766 : NULL,
1767 : false));
1768 1670 : testlist = lappend(testlist,
1769 1670 : make_opclause(opid, BOOLOID, false,
1770 : (Expr *) leftarg, (Expr *) param,
1771 : InvalidOid, opcollation));
1772 1670 : paramids = lappend_int(paramids, param->paramid);
1773 : }
1774 :
1775 : /* Put everything where it should go, and we're done */
1776 1174 : subselect->targetList = tlist;
1777 1174 : *testexpr = (Node *) make_ands_explicit(testlist);
1778 1174 : *paramIds = paramids;
1779 :
1780 1174 : return subselect;
1781 : }
1782 :
1783 :
1784 : /*
1785 : * Replace correlation vars (uplevel vars) with Params.
1786 : *
1787 : * Uplevel PlaceHolderVars and aggregates are replaced, too.
1788 : *
1789 : * Note: it is critical that this runs immediately after SS_process_sublinks.
1790 : * Since we do not recurse into the arguments of uplevel PHVs and aggregates,
1791 : * they will get copied to the appropriate subplan args list in the parent
1792 : * query with uplevel vars not replaced by Params, but only adjusted in level
1793 : * (see replace_outer_placeholdervar and replace_outer_agg). That's exactly
1794 : * what we want for the vars of the parent level --- but if a PHV's or
1795 : * aggregate's argument contains any further-up variables, they have to be
1796 : * replaced with Params in their turn. That will happen when the parent level
1797 : * runs SS_replace_correlation_vars. Therefore it must do so after expanding
1798 : * its sublinks to subplans. And we don't want any steps in between, else
1799 : * those steps would never get applied to the argument expressions, either in
1800 : * the parent or the child level.
1801 : *
1802 : * Another fairly tricky thing going on here is the handling of SubLinks in
1803 : * the arguments of uplevel PHVs/aggregates. Those are not touched inside the
1804 : * intermediate query level, either. Instead, SS_process_sublinks recurses on
1805 : * them after copying the PHV or Aggref expression into the parent plan level
1806 : * (this is actually taken care of in build_subplan).
1807 : */
1808 : Node *
1809 135604 : SS_replace_correlation_vars(PlannerInfo *root, Node *expr)
1810 : {
1811 : /* No setup needed for tree walk, so away we go */
1812 135604 : return replace_correlation_vars_mutator(expr, root);
1813 : }
1814 :
1815 : static Node *
1816 1045426 : replace_correlation_vars_mutator(Node *node, PlannerInfo *root)
1817 : {
1818 1045426 : if (node == NULL)
1819 41700 : return NULL;
1820 1003726 : if (IsA(node, Var))
1821 : {
1822 258404 : if (((Var *) node)->varlevelsup > 0)
1823 64996 : return (Node *) replace_outer_var(root, (Var *) node);
1824 : }
1825 938730 : if (IsA(node, PlaceHolderVar))
1826 : {
1827 28 : if (((PlaceHolderVar *) node)->phlevelsup > 0)
1828 28 : return (Node *) replace_outer_placeholdervar(root,
1829 : (PlaceHolderVar *) node);
1830 : }
1831 938702 : if (IsA(node, Aggref))
1832 : {
1833 11512 : if (((Aggref *) node)->agglevelsup > 0)
1834 32 : return (Node *) replace_outer_agg(root, (Aggref *) node);
1835 : }
1836 938670 : if (IsA(node, GroupingFunc))
1837 : {
1838 32 : if (((GroupingFunc *) node)->agglevelsup > 0)
1839 16 : return (Node *) replace_outer_grouping(root, (GroupingFunc *) node);
1840 : }
1841 938654 : return expression_tree_mutator(node,
1842 : replace_correlation_vars_mutator,
1843 : (void *) root);
1844 : }
1845 :
1846 : /*
1847 : * Expand SubLinks to SubPlans in the given expression.
1848 : *
1849 : * The isQual argument tells whether or not this expression is a WHERE/HAVING
1850 : * qualifier expression. If it is, any sublinks appearing at top level need
1851 : * not distinguish FALSE from UNKNOWN return values.
1852 : */
1853 : Node *
1854 109490 : SS_process_sublinks(PlannerInfo *root, Node *expr, bool isQual)
1855 : {
1856 : process_sublinks_context context;
1857 :
1858 109490 : context.root = root;
1859 109490 : context.isTopQual = isQual;
1860 109490 : return process_sublinks_mutator(expr, &context);
1861 : }
1862 :
1863 : static Node *
1864 1729246 : process_sublinks_mutator(Node *node, process_sublinks_context *context)
1865 : {
1866 : process_sublinks_context locContext;
1867 :
1868 1729246 : locContext.root = context->root;
1869 :
1870 1729246 : if (node == NULL)
1871 88958 : return NULL;
1872 1640288 : if (IsA(node, SubLink))
1873 : {
1874 46622 : SubLink *sublink = (SubLink *) node;
1875 : Node *testexpr;
1876 :
1877 : /*
1878 : * First, recursively process the lefthand-side expressions, if any.
1879 : * They're not top-level anymore.
1880 : */
1881 46622 : locContext.isTopQual = false;
1882 46622 : testexpr = process_sublinks_mutator(sublink->testexpr, &locContext);
1883 :
1884 : /*
1885 : * Now build the SubPlan node and make the expr to return.
1886 : */
1887 46622 : return make_subplan(context->root,
1888 46622 : (Query *) sublink->subselect,
1889 : sublink->subLinkType,
1890 : sublink->subLinkId,
1891 : testexpr,
1892 46622 : context->isTopQual);
1893 : }
1894 :
1895 : /*
1896 : * Don't recurse into the arguments of an outer PHV or aggregate here. Any
1897 : * SubLinks in the arguments have to be dealt with at the outer query
1898 : * level; they'll be handled when build_subplan collects the PHV or Aggref
1899 : * into the arguments to be passed down to the current subplan.
1900 : */
1901 1593666 : if (IsA(node, PlaceHolderVar))
1902 : {
1903 64 : if (((PlaceHolderVar *) node)->phlevelsup > 0)
1904 0 : return node;
1905 : }
1906 1593602 : else if (IsA(node, Aggref))
1907 : {
1908 6322 : if (((Aggref *) node)->agglevelsup > 0)
1909 12 : return node;
1910 : }
1911 :
1912 : /*
1913 : * We should never see a SubPlan expression in the input (since this is
1914 : * the very routine that creates 'em to begin with). We shouldn't find
1915 : * ourselves invoked directly on a Query, either.
1916 : */
1917 : Assert(!IsA(node, SubPlan));
1918 : Assert(!IsA(node, AlternativeSubPlan));
1919 : Assert(!IsA(node, Query));
1920 :
1921 : /*
1922 : * Because make_subplan() could return an AND or OR clause, we have to
1923 : * take steps to preserve AND/OR flatness of a qual. We assume the input
1924 : * has been AND/OR flattened and so we need no recursion here.
1925 : *
1926 : * (Due to the coding here, we will not get called on the List subnodes of
1927 : * an AND; and the input is *not* yet in implicit-AND format. So no check
1928 : * is needed for a bare List.)
1929 : *
1930 : * Anywhere within the top-level AND/OR clause structure, we can tell
1931 : * make_subplan() that NULL and FALSE are interchangeable. So isTopQual
1932 : * propagates down in both cases. (Note that this is unlike the meaning
1933 : * of "top level qual" used in most other places in Postgres.)
1934 : */
1935 1593654 : if (is_andclause(node))
1936 : {
1937 23816 : List *newargs = NIL;
1938 : ListCell *l;
1939 :
1940 : /* Still at qual top-level */
1941 23816 : locContext.isTopQual = context->isTopQual;
1942 :
1943 81882 : foreach(l, ((BoolExpr *) node)->args)
1944 : {
1945 : Node *newarg;
1946 :
1947 58066 : newarg = process_sublinks_mutator(lfirst(l), &locContext);
1948 58066 : if (is_andclause(newarg))
1949 0 : newargs = list_concat(newargs, ((BoolExpr *) newarg)->args);
1950 : else
1951 58066 : newargs = lappend(newargs, newarg);
1952 : }
1953 23816 : return (Node *) make_andclause(newargs);
1954 : }
1955 :
1956 1569838 : if (is_orclause(node))
1957 : {
1958 2912 : List *newargs = NIL;
1959 : ListCell *l;
1960 :
1961 : /* Still at qual top-level */
1962 2912 : locContext.isTopQual = context->isTopQual;
1963 :
1964 9518 : foreach(l, ((BoolExpr *) node)->args)
1965 : {
1966 : Node *newarg;
1967 :
1968 6606 : newarg = process_sublinks_mutator(lfirst(l), &locContext);
1969 6606 : if (is_orclause(newarg))
1970 0 : newargs = list_concat(newargs, ((BoolExpr *) newarg)->args);
1971 : else
1972 6606 : newargs = lappend(newargs, newarg);
1973 : }
1974 2912 : return (Node *) make_orclause(newargs);
1975 : }
1976 :
1977 : /*
1978 : * If we recurse down through anything other than an AND or OR node, we
1979 : * are definitely not at top qual level anymore.
1980 : */
1981 1566926 : locContext.isTopQual = false;
1982 :
1983 1566926 : return expression_tree_mutator(node,
1984 : process_sublinks_mutator,
1985 : (void *) &locContext);
1986 : }
1987 :
1988 : /*
1989 : * SS_identify_outer_params - identify the Params available from outer levels
1990 : *
1991 : * This must be run after SS_replace_correlation_vars and SS_process_sublinks
1992 : * processing is complete in a given query level as well as all of its
1993 : * descendant levels (which means it's most practical to do it at the end of
1994 : * processing the query level). We compute the set of paramIds that outer
1995 : * levels will make available to this level+descendants, and record it in
1996 : * root->outer_params for use while computing extParam/allParam sets in final
1997 : * plan cleanup. (We can't just compute it then, because the upper levels'
1998 : * plan_params lists are transient and will be gone by then.)
1999 : */
2000 : void
2001 318348 : SS_identify_outer_params(PlannerInfo *root)
2002 : {
2003 : Bitmapset *outer_params;
2004 : PlannerInfo *proot;
2005 : ListCell *l;
2006 :
2007 : /*
2008 : * If no parameters have been assigned anywhere in the tree, we certainly
2009 : * don't need to do anything here.
2010 : */
2011 318348 : if (root->glob->paramExecTypes == NIL)
2012 171364 : return;
2013 :
2014 : /*
2015 : * Scan all query levels above this one to see which parameters are due to
2016 : * be available from them, either because lower query levels have
2017 : * requested them (via plan_params) or because they will be available from
2018 : * initPlans of those levels.
2019 : */
2020 146984 : outer_params = NULL;
2021 191862 : for (proot = root->parent_root; proot != NULL; proot = proot->parent_root)
2022 : {
2023 : /* Include ordinary Var/PHV/Aggref params */
2024 102474 : foreach(l, proot->plan_params)
2025 : {
2026 57596 : PlannerParamItem *pitem = (PlannerParamItem *) lfirst(l);
2027 :
2028 57596 : outer_params = bms_add_member(outer_params, pitem->paramId);
2029 : }
2030 : /* Include any outputs of outer-level initPlans */
2031 46120 : foreach(l, proot->init_plans)
2032 : {
2033 1242 : SubPlan *initsubplan = (SubPlan *) lfirst(l);
2034 : ListCell *l2;
2035 :
2036 2484 : foreach(l2, initsubplan->setParam)
2037 : {
2038 1242 : outer_params = bms_add_member(outer_params, lfirst_int(l2));
2039 : }
2040 : }
2041 : /* Include worktable ID, if a recursive query is being planned */
2042 44878 : if (proot->wt_param_id >= 0)
2043 748 : outer_params = bms_add_member(outer_params, proot->wt_param_id);
2044 : }
2045 146984 : root->outer_params = outer_params;
2046 : }
2047 :
2048 : /*
2049 : * SS_charge_for_initplans - account for initplans in Path costs & parallelism
2050 : *
2051 : * If any initPlans have been created in the current query level, they will
2052 : * get attached to the Plan tree created from whichever Path we select from
2053 : * the given rel. Increment all that rel's Paths' costs to account for them,
2054 : * and make sure the paths get marked as parallel-unsafe, since we can't
2055 : * currently transmit initPlans to parallel workers.
2056 : *
2057 : * This is separate from SS_attach_initplans because we might conditionally
2058 : * create more initPlans during create_plan(), depending on which Path we
2059 : * select. However, Paths that would generate such initPlans are expected
2060 : * to have included their cost already.
2061 : */
2062 : void
2063 318348 : SS_charge_for_initplans(PlannerInfo *root, RelOptInfo *final_rel)
2064 : {
2065 : Cost initplan_cost;
2066 : ListCell *lc;
2067 :
2068 : /* Nothing to do if no initPlans */
2069 318348 : if (root->init_plans == NIL)
2070 308776 : return;
2071 :
2072 : /*
2073 : * Compute the cost increment just once, since it will be the same for all
2074 : * Paths. We assume each initPlan gets run once during top plan startup.
2075 : * This is a conservative overestimate, since in fact an initPlan might be
2076 : * executed later than plan startup, or even not at all.
2077 : */
2078 9572 : initplan_cost = 0;
2079 19494 : foreach(lc, root->init_plans)
2080 : {
2081 9922 : SubPlan *initsubplan = (SubPlan *) lfirst(lc);
2082 :
2083 9922 : initplan_cost += initsubplan->startup_cost + initsubplan->per_call_cost;
2084 : }
2085 :
2086 : /*
2087 : * Now adjust the costs and parallel_safe flags.
2088 : */
2089 19180 : foreach(lc, final_rel->pathlist)
2090 : {
2091 9608 : Path *path = (Path *) lfirst(lc);
2092 :
2093 9608 : path->startup_cost += initplan_cost;
2094 9608 : path->total_cost += initplan_cost;
2095 9608 : path->parallel_safe = false;
2096 : }
2097 :
2098 : /*
2099 : * Forget about any partial paths and clear consider_parallel, too;
2100 : * they're not usable if we attached an initPlan.
2101 : */
2102 9572 : final_rel->partial_pathlist = NIL;
2103 9572 : final_rel->consider_parallel = false;
2104 :
2105 : /* We needn't do set_cheapest() here, caller will do it */
2106 : }
2107 :
2108 : /*
2109 : * SS_attach_initplans - attach initplans to topmost plan node
2110 : *
2111 : * Attach any initplans created in the current query level to the specified
2112 : * plan node, which should normally be the topmost node for the query level.
2113 : * (In principle the initPlans could go in any node at or above where they're
2114 : * referenced; but there seems no reason to put them any lower than the
2115 : * topmost node, so we don't bother to track exactly where they came from.)
2116 : * We do not touch the plan node's cost; the initplans should have been
2117 : * accounted for in path costing.
2118 : */
2119 : void
2120 317682 : SS_attach_initplans(PlannerInfo *root, Plan *plan)
2121 : {
2122 317682 : plan->initPlan = root->init_plans;
2123 317682 : }
2124 :
2125 : /*
2126 : * SS_finalize_plan - do final parameter processing for a completed Plan.
2127 : *
2128 : * This recursively computes the extParam and allParam sets for every Plan
2129 : * node in the given plan tree. (Oh, and RangeTblFunction.funcparams too.)
2130 : *
2131 : * We assume that SS_finalize_plan has already been run on any initplans or
2132 : * subplans the plan tree could reference.
2133 : */
2134 : void
2135 162526 : SS_finalize_plan(PlannerInfo *root, Plan *plan)
2136 : {
2137 : /* No setup needed, just recurse through plan tree. */
2138 162526 : (void) finalize_plan(root, plan, -1, root->outer_params, NULL);
2139 162526 : }
2140 :
2141 : /*
2142 : * Recursive processing of all nodes in the plan tree
2143 : *
2144 : * gather_param is the rescan_param of an ancestral Gather/GatherMerge,
2145 : * or -1 if there is none.
2146 : *
2147 : * valid_params is the set of param IDs supplied by outer plan levels
2148 : * that are valid to reference in this plan node or its children.
2149 : *
2150 : * scan_params is a set of param IDs to force scan plan nodes to reference.
2151 : * This is for EvalPlanQual support, and is always NULL at the top of the
2152 : * recursion.
2153 : *
2154 : * The return value is the computed allParam set for the given Plan node.
2155 : * This is just an internal notational convenience: we can add a child
2156 : * plan's allParams to the set of param IDs of interest to this level
2157 : * in the same statement that recurses to that child.
2158 : *
2159 : * Do not scribble on caller's values of valid_params or scan_params!
2160 : *
2161 : * Note: although we attempt to deal with initPlans anywhere in the tree, the
2162 : * logic is not really right. The problem is that a plan node might return an
2163 : * output Param of its initPlan as a targetlist item, in which case it's valid
2164 : * for the parent plan level to reference that same Param; the parent's usage
2165 : * will be converted into a Var referencing the child plan node by setrefs.c.
2166 : * But this function would see the parent's reference as out of scope and
2167 : * complain about it. For now, this does not matter because the planner only
2168 : * attaches initPlans to the topmost plan node in a query level, so the case
2169 : * doesn't arise. If we ever merge this processing into setrefs.c, maybe it
2170 : * can be handled more cleanly.
2171 : */
2172 : static Bitmapset *
2173 1060404 : finalize_plan(PlannerInfo *root, Plan *plan,
2174 : int gather_param,
2175 : Bitmapset *valid_params,
2176 : Bitmapset *scan_params)
2177 : {
2178 : finalize_primnode_context context;
2179 : int locally_added_param;
2180 : Bitmapset *nestloop_params;
2181 : Bitmapset *initExtParam;
2182 : Bitmapset *initSetParam;
2183 : Bitmapset *child_params;
2184 : ListCell *l;
2185 :
2186 1060404 : if (plan == NULL)
2187 653790 : return NULL;
2188 :
2189 406614 : context.root = root;
2190 406614 : context.paramids = NULL; /* initialize set to empty */
2191 406614 : locally_added_param = -1; /* there isn't one */
2192 406614 : nestloop_params = NULL; /* there aren't any */
2193 :
2194 : /*
2195 : * Examine any initPlans to determine the set of external params they
2196 : * reference and the set of output params they supply. (We assume
2197 : * SS_finalize_plan was run on them already.)
2198 : */
2199 406614 : initExtParam = initSetParam = NULL;
2200 416896 : foreach(l, plan->initPlan)
2201 : {
2202 10282 : SubPlan *initsubplan = (SubPlan *) lfirst(l);
2203 10282 : Plan *initplan = planner_subplan_get_plan(root, initsubplan);
2204 : ListCell *l2;
2205 :
2206 10282 : initExtParam = bms_add_members(initExtParam, initplan->extParam);
2207 20580 : foreach(l2, initsubplan->setParam)
2208 : {
2209 10298 : initSetParam = bms_add_member(initSetParam, lfirst_int(l2));
2210 : }
2211 : }
2212 :
2213 : /* Any setParams are validly referenceable in this node and children */
2214 406614 : if (initSetParam)
2215 9900 : valid_params = bms_union(valid_params, initSetParam);
2216 :
2217 : /*
2218 : * When we call finalize_primnode, context.paramids sets are automatically
2219 : * merged together. But when recursing to self, we have to do it the hard
2220 : * way. We want the paramids set to include params in subplans as well as
2221 : * at this level.
2222 : */
2223 :
2224 : /* Find params in targetlist and qual */
2225 406614 : finalize_primnode((Node *) plan->targetlist, &context);
2226 406614 : finalize_primnode((Node *) plan->qual, &context);
2227 :
2228 : /*
2229 : * If it's a parallel-aware scan node, mark it as dependent on the parent
2230 : * Gather/GatherMerge's rescan Param.
2231 : */
2232 406614 : if (plan->parallel_aware)
2233 : {
2234 1276 : if (gather_param < 0)
2235 0 : elog(ERROR, "parallel-aware plan node is not below a Gather");
2236 1276 : context.paramids =
2237 1276 : bms_add_member(context.paramids, gather_param);
2238 : }
2239 :
2240 : /* Check additional node-type-specific fields */
2241 406614 : switch (nodeTag(plan))
2242 : {
2243 48198 : case T_Result:
2244 48198 : finalize_primnode(((Result *) plan)->resconstantqual,
2245 : &context);
2246 48198 : break;
2247 :
2248 68690 : case T_SeqScan:
2249 68690 : context.paramids = bms_add_members(context.paramids, scan_params);
2250 68690 : break;
2251 :
2252 36 : case T_SampleScan:
2253 36 : finalize_primnode((Node *) ((SampleScan *) plan)->tablesample,
2254 : &context);
2255 36 : context.paramids = bms_add_members(context.paramids, scan_params);
2256 36 : break;
2257 :
2258 57758 : case T_IndexScan:
2259 57758 : finalize_primnode((Node *) ((IndexScan *) plan)->indexqual,
2260 : &context);
2261 57758 : finalize_primnode((Node *) ((IndexScan *) plan)->indexorderby,
2262 : &context);
2263 :
2264 : /*
2265 : * we need not look at indexqualorig, since it will have the same
2266 : * param references as indexqual. Likewise, we can ignore
2267 : * indexorderbyorig.
2268 : */
2269 57758 : context.paramids = bms_add_members(context.paramids, scan_params);
2270 57758 : break;
2271 :
2272 2090 : case T_IndexOnlyScan:
2273 2090 : finalize_primnode((Node *) ((IndexOnlyScan *) plan)->indexqual,
2274 : &context);
2275 2090 : finalize_primnode((Node *) ((IndexOnlyScan *) plan)->indexorderby,
2276 : &context);
2277 :
2278 : /*
2279 : * we need not look at indextlist, since it cannot contain Params.
2280 : */
2281 2090 : context.paramids = bms_add_members(context.paramids, scan_params);
2282 2090 : break;
2283 :
2284 9366 : case T_BitmapIndexScan:
2285 9366 : finalize_primnode((Node *) ((BitmapIndexScan *) plan)->indexqual,
2286 : &context);
2287 :
2288 : /*
2289 : * we need not look at indexqualorig, since it will have the same
2290 : * param references as indexqual.
2291 : */
2292 9366 : break;
2293 :
2294 9306 : case T_BitmapHeapScan:
2295 9306 : finalize_primnode((Node *) ((BitmapHeapScan *) plan)->bitmapqualorig,
2296 : &context);
2297 9306 : context.paramids = bms_add_members(context.paramids, scan_params);
2298 9306 : break;
2299 :
2300 382 : case T_TidScan:
2301 382 : finalize_primnode((Node *) ((TidScan *) plan)->tidquals,
2302 : &context);
2303 382 : context.paramids = bms_add_members(context.paramids, scan_params);
2304 382 : break;
2305 :
2306 4482 : case T_SubqueryScan:
2307 : {
2308 4482 : SubqueryScan *sscan = (SubqueryScan *) plan;
2309 : RelOptInfo *rel;
2310 : Bitmapset *subquery_params;
2311 :
2312 : /* We must run finalize_plan on the subquery */
2313 4482 : rel = find_base_rel(root, sscan->scan.scanrelid);
2314 4482 : subquery_params = rel->subroot->outer_params;
2315 4482 : if (gather_param >= 0)
2316 8 : subquery_params = bms_add_member(bms_copy(subquery_params),
2317 : gather_param);
2318 4482 : finalize_plan(rel->subroot, sscan->subplan, gather_param,
2319 : subquery_params, NULL);
2320 :
2321 : /* Now we can add its extParams to the parent's params */
2322 8964 : context.paramids = bms_add_members(context.paramids,
2323 4482 : sscan->subplan->extParam);
2324 : /* We need scan_params too, though */
2325 4482 : context.paramids = bms_add_members(context.paramids,
2326 : scan_params);
2327 : }
2328 4482 : break;
2329 :
2330 23226 : case T_FunctionScan:
2331 : {
2332 23226 : FunctionScan *fscan = (FunctionScan *) plan;
2333 : ListCell *lc;
2334 :
2335 : /*
2336 : * Call finalize_primnode independently on each function
2337 : * expression, so that we can record which params are
2338 : * referenced in each, in order to decide which need
2339 : * re-evaluating during rescan.
2340 : */
2341 46468 : foreach(lc, fscan->functions)
2342 : {
2343 23242 : RangeTblFunction *rtfunc = (RangeTblFunction *) lfirst(lc);
2344 : finalize_primnode_context funccontext;
2345 :
2346 23242 : funccontext = context;
2347 23242 : funccontext.paramids = NULL;
2348 :
2349 23242 : finalize_primnode(rtfunc->funcexpr, &funccontext);
2350 :
2351 : /* remember results for execution */
2352 23242 : rtfunc->funcparams = funccontext.paramids;
2353 :
2354 : /* add the function's params to the overall set */
2355 23242 : context.paramids = bms_add_members(context.paramids,
2356 23242 : funccontext.paramids);
2357 : }
2358 :
2359 23226 : context.paramids = bms_add_members(context.paramids,
2360 : scan_params);
2361 : }
2362 23226 : break;
2363 :
2364 96 : case T_TableFuncScan:
2365 96 : finalize_primnode((Node *) ((TableFuncScan *) plan)->tablefunc,
2366 : &context);
2367 96 : context.paramids = bms_add_members(context.paramids, scan_params);
2368 96 : break;
2369 :
2370 2828 : case T_ValuesScan:
2371 2828 : finalize_primnode((Node *) ((ValuesScan *) plan)->values_lists,
2372 : &context);
2373 2828 : context.paramids = bms_add_members(context.paramids, scan_params);
2374 2828 : break;
2375 :
2376 892 : case T_CteScan:
2377 : {
2378 : /*
2379 : * You might think we should add the node's cteParam to
2380 : * paramids, but we shouldn't because that param is just a
2381 : * linkage mechanism for multiple CteScan nodes for the same
2382 : * CTE; it is never used for changed-param signaling. What we
2383 : * have to do instead is to find the referenced CTE plan and
2384 : * incorporate its external paramids, so that the correct
2385 : * things will happen if the CTE references outer-level
2386 : * variables. See test cases for bug #4902. (We assume
2387 : * SS_finalize_plan was run on the CTE plan already.)
2388 : */
2389 892 : int plan_id = ((CteScan *) plan)->ctePlanId;
2390 : Plan *cteplan;
2391 :
2392 : /* so, do this ... */
2393 892 : if (plan_id < 1 || plan_id > list_length(root->glob->subplans))
2394 0 : elog(ERROR, "could not find plan for CteScan referencing plan ID %d",
2395 : plan_id);
2396 892 : cteplan = (Plan *) list_nth(root->glob->subplans, plan_id - 1);
2397 892 : context.paramids =
2398 892 : bms_add_members(context.paramids, cteplan->extParam);
2399 :
2400 : #ifdef NOT_USED
2401 : /* ... but not this */
2402 : context.paramids =
2403 : bms_add_member(context.paramids,
2404 : ((CteScan *) plan)->cteParam);
2405 : #endif
2406 :
2407 892 : context.paramids = bms_add_members(context.paramids,
2408 : scan_params);
2409 : }
2410 892 : break;
2411 :
2412 332 : case T_WorkTableScan:
2413 332 : context.paramids =
2414 332 : bms_add_member(context.paramids,
2415 : ((WorkTableScan *) plan)->wtParam);
2416 332 : context.paramids = bms_add_members(context.paramids, scan_params);
2417 332 : break;
2418 :
2419 208 : case T_NamedTuplestoreScan:
2420 208 : context.paramids = bms_add_members(context.paramids, scan_params);
2421 208 : break;
2422 :
2423 510 : case T_ForeignScan:
2424 : {
2425 510 : ForeignScan *fscan = (ForeignScan *) plan;
2426 :
2427 510 : finalize_primnode((Node *) fscan->fdw_exprs,
2428 : &context);
2429 510 : finalize_primnode((Node *) fscan->fdw_recheck_quals,
2430 : &context);
2431 :
2432 : /* We assume fdw_scan_tlist cannot contain Params */
2433 510 : context.paramids = bms_add_members(context.paramids,
2434 : scan_params);
2435 : }
2436 510 : break;
2437 :
2438 0 : case T_CustomScan:
2439 0 : {
2440 0 : CustomScan *cscan = (CustomScan *) plan;
2441 : ListCell *lc;
2442 :
2443 0 : finalize_primnode((Node *) cscan->custom_exprs,
2444 : &context);
2445 : /* We assume custom_scan_tlist cannot contain Params */
2446 0 : context.paramids =
2447 0 : bms_add_members(context.paramids, scan_params);
2448 :
2449 : /* child nodes if any */
2450 0 : foreach(lc, cscan->custom_plans)
2451 : {
2452 0 : context.paramids =
2453 0 : bms_add_members(context.paramids,
2454 0 : finalize_plan(root,
2455 0 : (Plan *) lfirst(lc),
2456 : gather_param,
2457 : valid_params,
2458 : scan_params));
2459 : }
2460 : }
2461 0 : break;
2462 :
2463 72980 : case T_ModifyTable:
2464 72980 : {
2465 72980 : ModifyTable *mtplan = (ModifyTable *) plan;
2466 : ListCell *l;
2467 :
2468 : /* Force descendant scan nodes to reference epqParam */
2469 72980 : locally_added_param = mtplan->epqParam;
2470 72980 : valid_params = bms_add_member(bms_copy(valid_params),
2471 : locally_added_param);
2472 72980 : scan_params = bms_add_member(bms_copy(scan_params),
2473 : locally_added_param);
2474 72980 : finalize_primnode((Node *) mtplan->returningLists,
2475 : &context);
2476 72980 : finalize_primnode((Node *) mtplan->onConflictSet,
2477 : &context);
2478 72980 : finalize_primnode((Node *) mtplan->onConflictWhere,
2479 : &context);
2480 : /* exclRelTlist contains only Vars, doesn't need examination */
2481 147158 : foreach(l, mtplan->plans)
2482 : {
2483 74178 : context.paramids =
2484 74178 : bms_add_members(context.paramids,
2485 74178 : finalize_plan(root,
2486 74178 : (Plan *) lfirst(l),
2487 : gather_param,
2488 : valid_params,
2489 : scan_params));
2490 : }
2491 : }
2492 72980 : break;
2493 :
2494 2288 : case T_Append:
2495 2288 : {
2496 : ListCell *l;
2497 :
2498 7954 : foreach(l, ((Append *) plan)->appendplans)
2499 : {
2500 5666 : context.paramids =
2501 5666 : bms_add_members(context.paramids,
2502 5666 : finalize_plan(root,
2503 5666 : (Plan *) lfirst(l),
2504 : gather_param,
2505 : valid_params,
2506 : scan_params));
2507 : }
2508 : }
2509 2288 : break;
2510 :
2511 60 : case T_MergeAppend:
2512 60 : {
2513 : ListCell *l;
2514 :
2515 264 : foreach(l, ((MergeAppend *) plan)->mergeplans)
2516 : {
2517 204 : context.paramids =
2518 204 : bms_add_members(context.paramids,
2519 204 : finalize_plan(root,
2520 204 : (Plan *) lfirst(l),
2521 : gather_param,
2522 : valid_params,
2523 : scan_params));
2524 : }
2525 : }
2526 60 : break;
2527 :
2528 28 : case T_BitmapAnd:
2529 28 : {
2530 : ListCell *l;
2531 :
2532 84 : foreach(l, ((BitmapAnd *) plan)->bitmapplans)
2533 : {
2534 56 : context.paramids =
2535 56 : bms_add_members(context.paramids,
2536 56 : finalize_plan(root,
2537 56 : (Plan *) lfirst(l),
2538 : gather_param,
2539 : valid_params,
2540 : scan_params));
2541 : }
2542 : }
2543 28 : break;
2544 :
2545 32 : case T_BitmapOr:
2546 32 : {
2547 : ListCell *l;
2548 :
2549 96 : foreach(l, ((BitmapOr *) plan)->bitmapplans)
2550 : {
2551 64 : context.paramids =
2552 64 : bms_add_members(context.paramids,
2553 64 : finalize_plan(root,
2554 64 : (Plan *) lfirst(l),
2555 : gather_param,
2556 : valid_params,
2557 : scan_params));
2558 : }
2559 : }
2560 32 : break;
2561 :
2562 28772 : case T_NestLoop:
2563 28772 : {
2564 : ListCell *l;
2565 :
2566 28772 : finalize_primnode((Node *) ((Join *) plan)->joinqual,
2567 : &context);
2568 : /* collect set of params that will be passed to right child */
2569 49276 : foreach(l, ((NestLoop *) plan)->nestParams)
2570 : {
2571 20504 : NestLoopParam *nlp = (NestLoopParam *) lfirst(l);
2572 :
2573 20504 : nestloop_params = bms_add_member(nestloop_params,
2574 : nlp->paramno);
2575 : }
2576 : }
2577 28772 : break;
2578 :
2579 1040 : case T_MergeJoin:
2580 1040 : finalize_primnode((Node *) ((Join *) plan)->joinqual,
2581 : &context);
2582 1040 : finalize_primnode((Node *) ((MergeJoin *) plan)->mergeclauses,
2583 : &context);
2584 1040 : break;
2585 :
2586 15126 : case T_HashJoin:
2587 15126 : finalize_primnode((Node *) ((Join *) plan)->joinqual,
2588 : &context);
2589 15126 : finalize_primnode((Node *) ((HashJoin *) plan)->hashclauses,
2590 : &context);
2591 15126 : break;
2592 :
2593 1672 : case T_Limit:
2594 1672 : finalize_primnode(((Limit *) plan)->limitOffset,
2595 : &context);
2596 1672 : finalize_primnode(((Limit *) plan)->limitCount,
2597 : &context);
2598 1672 : break;
2599 :
2600 332 : case T_RecursiveUnion:
2601 : /* child nodes are allowed to reference wtParam */
2602 332 : locally_added_param = ((RecursiveUnion *) plan)->wtParam;
2603 332 : valid_params = bms_add_member(bms_copy(valid_params),
2604 : locally_added_param);
2605 : /* wtParam does *not* get added to scan_params */
2606 332 : break;
2607 :
2608 4876 : case T_LockRows:
2609 : /* Force descendant scan nodes to reference epqParam */
2610 4876 : locally_added_param = ((LockRows *) plan)->epqParam;
2611 4876 : valid_params = bms_add_member(bms_copy(valid_params),
2612 : locally_added_param);
2613 4876 : scan_params = bms_add_member(bms_copy(scan_params),
2614 : locally_added_param);
2615 4876 : break;
2616 :
2617 13200 : case T_Agg:
2618 12292 : {
2619 13200 : Agg *agg = (Agg *) plan;
2620 :
2621 : /*
2622 : * AGG_HASHED plans need to know which Params are referenced
2623 : * in aggregate calls. Do a separate scan to identify them.
2624 : */
2625 13200 : if (agg->aggstrategy == AGG_HASHED)
2626 : {
2627 : finalize_primnode_context aggcontext;
2628 :
2629 908 : aggcontext.root = root;
2630 908 : aggcontext.paramids = NULL;
2631 908 : finalize_agg_primnode((Node *) agg->plan.targetlist,
2632 : &aggcontext);
2633 908 : finalize_agg_primnode((Node *) agg->plan.qual,
2634 : &aggcontext);
2635 908 : agg->aggParams = aggcontext.paramids;
2636 : }
2637 : }
2638 13200 : break;
2639 :
2640 40 : case T_WindowAgg:
2641 40 : finalize_primnode(((WindowAgg *) plan)->startOffset,
2642 : &context);
2643 40 : finalize_primnode(((WindowAgg *) plan)->endOffset,
2644 : &context);
2645 40 : break;
2646 :
2647 500 : case T_Gather:
2648 : /* child nodes are allowed to reference rescan_param, if any */
2649 500 : locally_added_param = ((Gather *) plan)->rescan_param;
2650 500 : if (locally_added_param >= 0)
2651 : {
2652 500 : valid_params = bms_add_member(bms_copy(valid_params),
2653 : locally_added_param);
2654 :
2655 : /*
2656 : * We currently don't support nested Gathers. The issue so
2657 : * far as this function is concerned would be how to identify
2658 : * which child nodes depend on which Gather.
2659 : */
2660 : Assert(gather_param < 0);
2661 : /* Pass down rescan_param to child parallel-aware nodes */
2662 500 : gather_param = locally_added_param;
2663 : }
2664 : /* rescan_param does *not* get added to scan_params */
2665 500 : break;
2666 :
2667 152 : case T_GatherMerge:
2668 : /* child nodes are allowed to reference rescan_param, if any */
2669 152 : locally_added_param = ((GatherMerge *) plan)->rescan_param;
2670 152 : if (locally_added_param >= 0)
2671 : {
2672 152 : valid_params = bms_add_member(bms_copy(valid_params),
2673 : locally_added_param);
2674 :
2675 : /*
2676 : * We currently don't support nested Gathers. The issue so
2677 : * far as this function is concerned would be how to identify
2678 : * which child nodes depend on which Gather.
2679 : */
2680 : Assert(gather_param < 0);
2681 : /* Pass down rescan_param to child parallel-aware nodes */
2682 152 : gather_param = locally_added_param;
2683 : }
2684 : /* rescan_param does *not* get added to scan_params */
2685 152 : break;
2686 :
2687 37116 : case T_ProjectSet:
2688 : case T_Hash:
2689 : case T_Material:
2690 : case T_Sort:
2691 : case T_IncrementalSort:
2692 : case T_Unique:
2693 : case T_SetOp:
2694 : case T_Group:
2695 : /* no node-type-specific fields need fixing */
2696 37116 : break;
2697 :
2698 0 : default:
2699 0 : elog(ERROR, "unrecognized node type: %d",
2700 : (int) nodeTag(plan));
2701 : }
2702 :
2703 : /* Process left and right child plans, if any */
2704 406614 : child_params = finalize_plan(root,
2705 406614 : plan->lefttree,
2706 : gather_param,
2707 : valid_params,
2708 : scan_params);
2709 406614 : context.paramids = bms_add_members(context.paramids, child_params);
2710 :
2711 406614 : if (nestloop_params)
2712 : {
2713 : /* right child can reference nestloop_params as well as valid_params */
2714 19196 : child_params = finalize_plan(root,
2715 19196 : plan->righttree,
2716 : gather_param,
2717 : bms_union(nestloop_params, valid_params),
2718 : scan_params);
2719 : /* ... and they don't count as parameters used at my level */
2720 19196 : child_params = bms_difference(child_params, nestloop_params);
2721 19196 : bms_free(nestloop_params);
2722 : }
2723 : else
2724 : {
2725 : /* easy case */
2726 387418 : child_params = finalize_plan(root,
2727 387418 : plan->righttree,
2728 : gather_param,
2729 : valid_params,
2730 : scan_params);
2731 : }
2732 406614 : context.paramids = bms_add_members(context.paramids, child_params);
2733 :
2734 : /*
2735 : * Any locally generated parameter doesn't count towards its generating
2736 : * plan node's external dependencies. (Note: if we changed valid_params
2737 : * and/or scan_params, we leak those bitmapsets; not worth the notational
2738 : * trouble to clean them up.)
2739 : */
2740 406614 : if (locally_added_param >= 0)
2741 : {
2742 78840 : context.paramids = bms_del_member(context.paramids,
2743 : locally_added_param);
2744 : }
2745 :
2746 : /* Now we have all the paramids referenced in this node and children */
2747 :
2748 406614 : if (!bms_is_subset(context.paramids, valid_params))
2749 0 : elog(ERROR, "plan should not reference subplan's variable");
2750 :
2751 : /*
2752 : * The plan node's allParam and extParam fields should include all its
2753 : * referenced paramids, plus contributions from any child initPlans.
2754 : * However, any setParams of the initPlans should not be present in the
2755 : * parent node's extParams, only in its allParams. (It's possible that
2756 : * some initPlans have extParams that are setParams of other initPlans.)
2757 : */
2758 :
2759 : /* allParam must include initplans' extParams and setParams */
2760 406614 : plan->allParam = bms_union(context.paramids, initExtParam);
2761 406614 : plan->allParam = bms_add_members(plan->allParam, initSetParam);
2762 : /* extParam must include any initplan extParams */
2763 406614 : plan->extParam = bms_union(context.paramids, initExtParam);
2764 : /* but not any initplan setParams */
2765 406614 : plan->extParam = bms_del_members(plan->extParam, initSetParam);
2766 :
2767 : /*
2768 : * For speed at execution time, make sure extParam/allParam are actually
2769 : * NULL if they are empty sets.
2770 : */
2771 406614 : if (bms_is_empty(plan->extParam))
2772 248346 : plan->extParam = NULL;
2773 406614 : if (bms_is_empty(plan->allParam))
2774 238604 : plan->allParam = NULL;
2775 :
2776 406614 : return plan->allParam;
2777 : }
2778 :
2779 : /*
2780 : * finalize_primnode: add IDs of all PARAM_EXEC params appearing in the given
2781 : * expression tree to the result set.
2782 : */
2783 : static bool
2784 7444554 : finalize_primnode(Node *node, finalize_primnode_context *context)
2785 : {
2786 7444554 : if (node == NULL)
2787 884458 : return false;
2788 6560096 : if (IsA(node, Param))
2789 : {
2790 100758 : if (((Param *) node)->paramkind == PARAM_EXEC)
2791 : {
2792 99106 : int paramid = ((Param *) node)->paramid;
2793 :
2794 99106 : context->paramids = bms_add_member(context->paramids, paramid);
2795 : }
2796 100758 : return false; /* no more to do here */
2797 : }
2798 6459338 : if (IsA(node, SubPlan))
2799 : {
2800 48294 : SubPlan *subplan = (SubPlan *) node;
2801 48294 : Plan *plan = planner_subplan_get_plan(context->root, subplan);
2802 : ListCell *lc;
2803 : Bitmapset *subparamids;
2804 :
2805 : /* Recurse into the testexpr, but not into the Plan */
2806 48294 : finalize_primnode(subplan->testexpr, context);
2807 :
2808 : /*
2809 : * Remove any param IDs of output parameters of the subplan that were
2810 : * referenced in the testexpr. These are not interesting for
2811 : * parameter change signaling since we always re-evaluate the subplan.
2812 : * Note that this wouldn't work too well if there might be uses of the
2813 : * same param IDs elsewhere in the plan, but that can't happen because
2814 : * generate_new_exec_param never tries to merge params.
2815 : */
2816 50234 : foreach(lc, subplan->paramIds)
2817 : {
2818 1940 : context->paramids = bms_del_member(context->paramids,
2819 : lfirst_int(lc));
2820 : }
2821 :
2822 : /* Also examine args list */
2823 48294 : finalize_primnode((Node *) subplan->args, context);
2824 :
2825 : /*
2826 : * Add params needed by the subplan to paramids, but excluding those
2827 : * we will pass down to it. (We assume SS_finalize_plan was run on
2828 : * the subplan already.)
2829 : */
2830 48294 : subparamids = bms_copy(plan->extParam);
2831 116904 : foreach(lc, subplan->parParam)
2832 : {
2833 68610 : subparamids = bms_del_member(subparamids, lfirst_int(lc));
2834 : }
2835 48294 : context->paramids = bms_join(context->paramids, subparamids);
2836 :
2837 48294 : return false; /* no more to do here */
2838 : }
2839 6411044 : return expression_tree_walker(node, finalize_primnode,
2840 : (void *) context);
2841 : }
2842 :
2843 : /*
2844 : * finalize_agg_primnode: find all Aggref nodes in the given expression tree,
2845 : * and add IDs of all PARAM_EXEC params appearing within their aggregated
2846 : * arguments to the result set.
2847 : */
2848 : static bool
2849 6028 : finalize_agg_primnode(Node *node, finalize_primnode_context *context)
2850 : {
2851 6028 : if (node == NULL)
2852 876 : return false;
2853 5152 : if (IsA(node, Aggref))
2854 : {
2855 632 : Aggref *agg = (Aggref *) node;
2856 :
2857 : /* we should not consider the direct arguments, if any */
2858 632 : finalize_primnode((Node *) agg->args, context);
2859 632 : finalize_primnode((Node *) agg->aggfilter, context);
2860 632 : return false; /* there can't be any Aggrefs below here */
2861 : }
2862 4520 : return expression_tree_walker(node, finalize_agg_primnode,
2863 : (void *) context);
2864 : }
2865 :
2866 : /*
2867 : * SS_make_initplan_output_param - make a Param for an initPlan's output
2868 : *
2869 : * The plan is expected to return a scalar value of the given type/collation.
2870 : *
2871 : * Note that in some cases the initplan may not ever appear in the finished
2872 : * plan tree. If that happens, we'll have wasted a PARAM_EXEC slot, which
2873 : * is no big deal.
2874 : */
2875 : Param *
2876 422 : SS_make_initplan_output_param(PlannerInfo *root,
2877 : Oid resulttype, int32 resulttypmod,
2878 : Oid resultcollation)
2879 : {
2880 422 : return generate_new_exec_param(root, resulttype,
2881 : resulttypmod, resultcollation);
2882 : }
2883 :
2884 : /*
2885 : * SS_make_initplan_from_plan - given a plan tree, make it an InitPlan
2886 : *
2887 : * We build an EXPR_SUBLINK SubPlan node and put it into the initplan
2888 : * list for the outer query level. A Param that represents the initplan's
2889 : * output has already been assigned using SS_make_initplan_output_param.
2890 : */
2891 : void
2892 360 : SS_make_initplan_from_plan(PlannerInfo *root,
2893 : PlannerInfo *subroot, Plan *plan,
2894 : Param *prm)
2895 : {
2896 : SubPlan *node;
2897 :
2898 : /*
2899 : * Add the subplan and its PlannerInfo to the global lists.
2900 : */
2901 360 : root->glob->subplans = lappend(root->glob->subplans, plan);
2902 360 : root->glob->subroots = lappend(root->glob->subroots, subroot);
2903 :
2904 : /*
2905 : * Create a SubPlan node and add it to the outer list of InitPlans. Note
2906 : * it has to appear after any other InitPlans it might depend on (see
2907 : * comments in ExecReScan).
2908 : */
2909 360 : node = makeNode(SubPlan);
2910 360 : node->subLinkType = EXPR_SUBLINK;
2911 360 : node->plan_id = list_length(root->glob->subplans);
2912 360 : node->plan_name = psprintf("InitPlan %d (returns $%d)",
2913 : node->plan_id, prm->paramid);
2914 360 : get_first_col_type(plan, &node->firstColType, &node->firstColTypmod,
2915 : &node->firstColCollation);
2916 360 : node->setParam = list_make1_int(prm->paramid);
2917 :
2918 360 : root->init_plans = lappend(root->init_plans, node);
2919 :
2920 : /*
2921 : * The node can't have any inputs (since it's an initplan), so the
2922 : * parParam and args lists remain empty.
2923 : */
2924 :
2925 : /* Set costs of SubPlan using info from the plan tree */
2926 360 : cost_subplan(subroot, node, plan);
2927 360 : }
|
