subselect.c 28.6 KB
Newer Older
1 2
/*-------------------------------------------------------------------------
 *
3
 * subselect.c
4 5
 *	  Planning routines for subselects and parameters.
 *
B
Bruce Momjian 已提交
6
 * Portions Copyright (c) 1996-2002, PostgreSQL Global Development Group
B
Add:  
Bruce Momjian 已提交
7
 * Portions Copyright (c) 1994, Regents of the University of California
8 9
 *
 * IDENTIFICATION
10
 *	  $Header: /cvsroot/pgsql/src/backend/optimizer/plan/subselect.c,v 1.70 2003/02/08 20:20:54 tgl Exp $
11 12 13 14 15
 *
 *-------------------------------------------------------------------------
 */
#include "postgres.h"

16
#include "catalog/pg_operator.h"
17
#include "catalog/pg_type.h"
18
#include "miscadmin.h"
19
#include "nodes/makefuncs.h"
20
#include "nodes/params.h"
21
#include "optimizer/clauses.h"
22 23
#include "optimizer/cost.h"
#include "optimizer/planmain.h"
B
Bruce Momjian 已提交
24 25
#include "optimizer/planner.h"
#include "optimizer/subselect.h"
26
#include "optimizer/var.h"
27
#include "parser/parsetree.h"
28 29
#include "parser/parse_expr.h"
#include "parser/parse_oper.h"
30
#include "parser/parse_relation.h"
31
#include "utils/lsyscache.h"
32
#include "utils/syscache.h"
33

34

35
Index		PlannerQueryLevel;	/* level of current query */
36
List	   *PlannerInitPlan;	/* init subplans for current query */
37
List	   *PlannerParamVar;	/* to get Var from Param->paramid */
38 39

int			PlannerPlanId = 0;	/* to assign unique ID to subquery plans */
40 41 42 43 44 45 46 47 48 49 50 51

/*--------------------
 * PlannerParamVar is a list of Var nodes, wherein the n'th entry
 * (n counts from 0) corresponds to Param->paramid = n.  The Var nodes
 * are ordinary except for one thing: their varlevelsup field does NOT
 * have the usual interpretation of "subplan levels out from current".
 * Instead, it contains the absolute plan level, with the outermost
 * plan being level 1 and nested plans having higher level numbers.
 * This nonstandardness is useful because we don't have to run around
 * and update the list elements when we enter or exit a subplan
 * recursion level.  But we must pay attention not to confuse this
 * meaning with the normal meaning of varlevelsup.
52 53 54 55
 *
 * We also need to create Param slots that don't correspond to any outer Var.
 * For these, we set varno = 0 and varlevelsup = 0, so that they can't
 * accidentally match an outer Var.
56 57
 *--------------------
 */
58 59


60 61 62 63 64 65
typedef struct finalize_primnode_results
{
	List	   *paramids;		/* List of PARAM_EXEC paramids found */
} finalize_primnode_results;


66
static List *convert_sublink_opers(List *lefthand, List *operOids,
67 68
								   List *targetlist, int rtindex,
								   List **righthandIds);
69
static bool subplan_is_hashable(SubLink *slink, SubPlan *node);
70
static Node *replace_correlation_vars_mutator(Node *node, void *context);
71
static Node *process_sublinks_mutator(Node *node, bool *isTopQual);
72
static bool finalize_primnode(Node *node, finalize_primnode_results *results);
73 74


75 76 77
/*
 * Create a new entry in the PlannerParamVar list, and return its index.
 *
78 79 80 81
 * var contains the data to use, except for varlevelsup which
 * is set from the absolute level value given by varlevel.  NOTE that
 * the passed var is scribbled on and placed directly into the list!
 * Generally, caller should have just created or copied it.
82
 */
83
static int
84
new_param(Var *var, Index varlevel)
85
{
86
	var->varlevelsup = varlevel;
87

88
	PlannerParamVar = lappend(PlannerParamVar, var);
89

90
	return length(PlannerParamVar) - 1;
91 92
}

93 94 95 96
/*
 * Generate a Param node to replace the given Var,
 * which is expected to have varlevelsup > 0 (ie, it is not local).
 */
97
static Param *
98
replace_var(Var *var)
99
{
100
	List	   *ppv;
101
	Param	   *retval;
102
	Index		varlevel;
103 104
	int			i;

105 106
	Assert(var->varlevelsup > 0 && var->varlevelsup < PlannerQueryLevel);
	varlevel = PlannerQueryLevel - var->varlevelsup;
107

108
	/*
109
	 * If there's already a PlannerParamVar entry for this same Var, just
B
Bruce Momjian 已提交
110 111 112 113 114 115
	 * use it.	NOTE: in sufficiently complex querytrees, it is possible
	 * for the same varno/varlevel to refer to different RTEs in different
	 * parts of the parsetree, so that different fields might end up
	 * sharing the same Param number.  As long as we check the vartype as
	 * well, I believe that this sort of aliasing will cause no trouble.
	 * The correct field should get stored into the Param slot at
116
	 * execution in each part of the tree.
117 118 119
	 */
	i = 0;
	foreach(ppv, PlannerParamVar)
120
	{
121
		Var		   *pvar = lfirst(ppv);
122 123 124 125 126

		if (pvar->varno == var->varno &&
			pvar->varattno == var->varattno &&
			pvar->varlevelsup == varlevel &&
			pvar->vartype == var->vartype)
127
			break;
128
		i++;
129
	}
130

131
	if (!ppv)
132 133
	{
		/* Nope, so make a new one */
134
		i = new_param((Var *) copyObject(var), varlevel);
135
	}
136

137 138 139 140
	retval = makeNode(Param);
	retval->paramkind = PARAM_EXEC;
	retval->paramid = (AttrNumber) i;
	retval->paramtype = var->vartype;
141

142
	return retval;
143 144
}

145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160
/*
 * Generate a new Param node that will not conflict with any other.
 */
static Param *
generate_new_param(Oid paramtype, int32 paramtypmod)
{
	Var		   *var = makeVar(0, 0, paramtype, paramtypmod, 0);
	Param	   *retval = makeNode(Param);

	retval->paramkind = PARAM_EXEC;
	retval->paramid = (AttrNumber) new_param(var, 0);
	retval->paramtype = paramtype;

	return retval;
}

161
/*
162 163 164
 * Convert a bare SubLink (as created by the parser) into a SubPlan.
 *
 * We are given the raw SubLink and the already-processed lefthand argument
165 166
 * list (use this instead of the SubLink's own field).  We are also told if
 * this expression appears at top level of a WHERE/HAVING qual.
167 168 169 170 171 172
 *
 * The result is whatever we need to substitute in place of the SubLink
 * node in the executable expression.  This will be either the SubPlan
 * node (if we have to do the subplan as a subplan), or a Param node
 * representing the result of an InitPlan, or possibly an AND or OR tree
 * containing InitPlan Param nodes.
173
 */
174
static Node *
175
make_subplan(SubLink *slink, List *lefthand, bool isTopQual)
176
{
177
	SubPlan	   *node = makeNode(SubPlan);
178
	Query	   *subquery = (Query *) (slink->subselect);
179
	double		tuple_fraction;
180 181 182
	Plan	   *plan;
	List	   *lst;
	Node	   *result;
183

184
	/*
B
Bruce Momjian 已提交
185 186 187 188
	 * Copy the source Query node.	This is a quick and dirty kluge to
	 * resolve the fact that the parser can generate trees with multiple
	 * links to the same sub-Query node, but the planner wants to scribble
	 * on the Query. Try to clean this up when we do querytree redesign...
189 190 191
	 */
	subquery = (Query *) copyObject(subquery);

192
	/*
193 194 195 196 197 198 199
	 * For an EXISTS subplan, tell lower-level planner to expect that only
	 * the first tuple will be retrieved.  For ALL and ANY subplans, we
	 * will be able to stop evaluating if the test condition fails, so
	 * very often not all the tuples will be retrieved; for lack of a
	 * better idea, specify 50% retrieval.	For EXPR and MULTIEXPR
	 * subplans, use default behavior (we're only expecting one row out,
	 * anyway).
200
	 *
201 202 203
	 * NOTE: if you change these numbers, also change cost_qual_eval_walker()
	 * in path/costsize.c.
	 *
204 205 206
	 * XXX If an ALL/ANY subplan is uncorrelated, we may decide to hash or
	 * materialize its result below.  In that case it would've been better to
	 * specify full retrieval.  At present, however, we can only detect
207 208 209 210 211
	 * correlation or lack of it after we've made the subplan :-(. Perhaps
	 * detection of correlation should be done as a separate step.
	 * Meanwhile, we don't want to be too optimistic about the percentage
	 * of tuples retrieved, for fear of selecting a plan that's bad for
	 * the materialization case.
212 213 214
	 */
	if (slink->subLinkType == EXISTS_SUBLINK)
		tuple_fraction = 1.0;	/* just like a LIMIT 1 */
215 216
	else if (slink->subLinkType == ALL_SUBLINK ||
			 slink->subLinkType == ANY_SUBLINK)
217
		tuple_fraction = 0.5;	/* 50% */
218 219
	else
		tuple_fraction = -1.0;	/* default behavior */
220

221
	/*
222
	 * Generate the plan for the subquery.
223
	 */
224
	node->plan = plan = subquery_planner(subquery, tuple_fraction);
225

B
Bruce Momjian 已提交
226
	node->plan_id = PlannerPlanId++;	/* Assign unique ID to this
227
										 * SubPlan */
228

229
	node->rtable = subquery->rtable;
230

231
	/*
232
	 * Initialize other fields of the SubPlan node.
233 234
	 */
	node->subLinkType = slink->subLinkType;
235
	node->useOr = slink->useOr;
236 237 238
	node->exprs = NIL;
	node->paramIds = NIL;
	node->useHashTable = false;
239 240
	/* At top level of a qual, can treat UNKNOWN the same as FALSE */
	node->unknownEqFalse = isTopQual;
241 242 243
	node->setParam = NIL;
	node->parParam = NIL;
	node->args = NIL;
244

245
	/*
B
Bruce Momjian 已提交
246 247
	 * Make parParam list of params that current query level will pass to
	 * this child plan.
248
	 */
249
	foreach(lst, plan->extParam)
250
	{
251 252
		int			paramid = lfirsti(lst);
		Var		   *var = nth(paramid, PlannerParamVar);
253

254
		/* note varlevelsup is absolute level number */
255
		if (var->varlevelsup == PlannerQueryLevel)
256
			node->parParam = lappendi(node->parParam, paramid);
257
	}
258 259

	/*
260
	 * Un-correlated or undirect correlated plans of EXISTS, EXPR, or
261 262
	 * MULTIEXPR types can be used as initPlans.  For EXISTS or EXPR, we
	 * just produce a Param referring to the result of evaluating the
263 264 265
	 * initPlan.  For MULTIEXPR, we must build an AND or OR-clause of the
	 * individual comparison operators, using the appropriate lefthand
	 * side expressions and Params for the initPlan's target items.
266
	 */
267 268
	if (node->parParam == NIL && slink->subLinkType == EXISTS_SUBLINK)
	{
269
		Param	   *prm;
270

271
		prm = generate_new_param(BOOLOID, -1);
272 273 274 275 276 277 278
		node->setParam = lappendi(node->setParam, prm->paramid);
		PlannerInitPlan = lappend(PlannerInitPlan, node);
		result = (Node *) prm;
	}
	else if (node->parParam == NIL && slink->subLinkType == EXPR_SUBLINK)
	{
		TargetEntry *te = lfirst(plan->targetlist);
279
		Param	   *prm;
280

281
		Assert(!te->resdom->resjunk);
282
		prm = generate_new_param(te->resdom->restype, te->resdom->restypmod);
283 284 285 286 287
		node->setParam = lappendi(node->setParam, prm->paramid);
		PlannerInitPlan = lappend(PlannerInitPlan, node);
		result = (Node *) prm;
	}
	else if (node->parParam == NIL && slink->subLinkType == MULTIEXPR_SUBLINK)
288
	{
289 290 291 292 293 294
		List   *exprs;

		/* Convert the lefthand exprs and oper OIDs into executable exprs */
		exprs = convert_sublink_opers(lefthand,
									  slink->operOids,
									  plan->targetlist,
295
									  0,
296 297
									  &node->paramIds);
		node->setParam = nconc(node->setParam, listCopy(node->paramIds));
298
		PlannerInitPlan = lappend(PlannerInitPlan, node);
299 300 301 302 303 304 305 306
		/*
		 * The executable expressions are returned to become part of the
		 * outer plan's expression tree; they are not kept in the initplan
		 * node.
		 */
		if (length(exprs) > 1)
			result = (Node *) (node->useOr ? make_orclause(exprs) :
							   make_andclause(exprs));
307
		else
308
			result = (Node *) lfirst(exprs);
309
	}
310
	else
311
	{
312
		List	   *args;
313

314
		/*
315 316 317
		 * We can't convert subplans of ALL_SUBLINK or ANY_SUBLINK types
		 * to initPlans, even when they are uncorrelated or undirect
		 * correlated, because we need to scan the output of the subplan
318 319 320 321 322 323 324 325 326 327 328 329 330 331
		 * for each outer tuple.  But if it's an IN (= ANY) test, we might
		 * be able to use a hashtable to avoid comparing all the tuples.
		 */
		if (subplan_is_hashable(slink, node))
			node->useHashTable = true;
		/*
		 * Otherwise, we have the option to tack a MATERIAL node onto the top
		 * of the subplan, to reduce the cost of reading it repeatedly.  This
		 * is pointless for a direct-correlated subplan, since we'd have to
		 * recompute its results each time anyway.  For uncorrelated/undirect
		 * correlated subplans, we add MATERIAL if the subplan's top plan node
		 * is anything more complicated than a plain sequential scan, and we
		 * do it even for seqscan if the qual appears selective enough to
		 * eliminate many tuples.
332 333 334 335 336 337
		 *
		 * XXX It's pretty ugly to be inserting a MATERIAL node at this
		 * point.  Since subquery_planner has already run SS_finalize_plan
		 * on the subplan tree, we have to kluge up parameter lists for
		 * the MATERIAL node.  Possibly this could be fixed by postponing
		 * SS_finalize_plan processing until setrefs.c is run.
338
		 */
339
		else if (node->parParam == NIL)
340 341 342 343 344 345
		{
			bool		use_material;

			switch (nodeTag(plan))
			{
				case T_SeqScan:
346
					if (plan->initPlan)
347 348 349 350 351 352 353
						use_material = true;
					else
					{
						Selectivity qualsel;

						qualsel = clauselist_selectivity(subquery,
														 plan->qual,
354
														 0, JOIN_INNER);
355 356 357
						/* Is 10% selectivity a good threshold?? */
						use_material = qualsel < 0.10;
					}
358 359
					break;
				case T_Material:
360
				case T_FunctionScan:
361
				case T_Sort:
362 363 364

					/*
					 * Don't add another Material node if there's one
365 366
					 * already, nor if the top node is any other type that
					 * materializes its output anyway.
367 368 369 370 371 372 373 374 375
					 */
					use_material = false;
					break;
				default:
					use_material = true;
					break;
			}
			if (use_material)
			{
B
Bruce Momjian 已提交
376
				Plan	   *matplan;
377
				Path		matpath; /* dummy for result of cost_material */
378 379

				matplan = (Plan *) make_material(plan->targetlist, plan);
380 381 382 383 384 385 386
				/* need to calculate costs */
				cost_material(&matpath,
							  plan->total_cost,
							  plan->plan_rows,
							  plan->plan_width);
				matplan->startup_cost = matpath.startup_cost;
				matplan->total_cost = matpath.total_cost;
387 388
				matplan->plan_rows = plan->plan_rows;
				matplan->plan_width = plan->plan_width;
389
				/* parameter kluge --- see comments above */
390 391 392
				matplan->extParam = listCopy(plan->extParam);
				matplan->locParam = listCopy(plan->locParam);
				node->plan = plan = matplan;
393 394 395
			}
		}

396 397 398 399
		/* Convert the lefthand exprs and oper OIDs into executable exprs */
		node->exprs = convert_sublink_opers(lefthand,
											slink->operOids,
											plan->targetlist,
400
											0,
401
											&node->paramIds);
402

403
		/*
404
		 * Make node->args from parParam.
405
		 */
406
		args = NIL;
407
		foreach(lst, node->parParam)
408
		{
409 410 411
			Var		   *var = nth(lfirsti(lst), PlannerParamVar);

			var = (Var *) copyObject(var);
412 413 414 415 416

			/*
			 * Must fix absolute-level varlevelsup from the
			 * PlannerParamVar entry.  But since var is at current subplan
			 * level, this is easy:
417
			 */
418
			var->varlevelsup = 0;
419
			args = lappend(args, var);
420
		}
421
		node->args = args;
422

423
		result = (Node *) node;
424 425 426 427 428 429
	}

	return result;
}

/*
430 431
 * convert_sublink_opers: given a lefthand-expressions list and a list of
 * operator OIDs, build a list of actually executable expressions.  The
432 433
 * righthand sides of the expressions are Params or Vars representing the
 * results of the sub-select.
434
 *
435 436 437 438 439 440
 * If rtindex is 0, we build Params to represent the sub-select outputs.
 * The paramids of the Params created are returned in the *righthandIds list.
 *
 * If rtindex is not 0, we build Vars using that rtindex as varno.  The
 * Vars themselves are returned in *righthandIds (this is a bit of a type
 * cheat, but we can get away with it).
441
 */
442
static List *
443
convert_sublink_opers(List *lefthand, List *operOids,
444 445
					  List *targetlist, int rtindex,
					  List **righthandIds)
446
{
447
	List	   *result = NIL;
448 449
	List	   *lst;

450
	*righthandIds = NIL;
451 452

	foreach(lst, operOids)
453
	{
454 455
		Oid			opid = (Oid) lfirsti(lst);
		Node	   *leftop = lfirst(lefthand);
456
		TargetEntry *te = lfirst(targetlist);
457
		Node	   *rightop;
458 459 460 461 462
		Operator	tup;
		Form_pg_operator opform;
		Node	   *left,
				   *right;

463 464
		Assert(!te->resdom->resjunk);

465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486
		if (rtindex)
		{
			/* Make the Var node representing the subplan's result */
			rightop = (Node *) makeVar(rtindex,
									   te->resdom->resno,
									   te->resdom->restype,
									   te->resdom->restypmod,
									   0);
			/* Record it for caller */
			*righthandIds = lappend(*righthandIds, rightop);
		}
		else
		{
			/* Make the Param node representing the subplan's result */
			Param	   *prm;

			prm = generate_new_param(te->resdom->restype,
									 te->resdom->restypmod);
			/* Record its ID */
			*righthandIds = lappendi(*righthandIds, prm->paramid);
			rightop = (Node *) prm;
		}
487

488
		/* Look up the operator to get its declared input types */
489
		tup = SearchSysCache(OPEROID,
490
							 ObjectIdGetDatum(opid),
491 492
							 0, 0, 0);
		if (!HeapTupleIsValid(tup))
493
			elog(ERROR, "cache lookup failed for operator %u", opid);
494 495
		opform = (Form_pg_operator) GETSTRUCT(tup);

496
		/*
497 498 499 500
		 * Make the expression node.
		 *
		 * Note: we use make_operand in case runtime type conversion
		 * function calls must be inserted for this operator!
501
		 */
502
		left = make_operand(leftop, exprType(leftop), opform->oprleft);
503
		right = make_operand(rightop, te->resdom->restype, opform->oprright);
504 505 506 507 508 509
		result = lappend(result,
						 make_opclause(opid,
									   opform->oprresult,
									   false, /* set-result not allowed */
									   (Expr *) left,
									   (Expr *) right));
510

511 512
		ReleaseSysCache(tup);

513
		lefthand = lnext(lefthand);
514
		targetlist = lnext(targetlist);
515
	}
516

517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558
	return result;
}

/*
 * subplan_is_hashable: decide whether we can implement a subplan by hashing
 *
 * Caution: the SubPlan node is not completely filled in yet.  We can rely
 * on its plan and parParam fields, however.
 */
static bool
subplan_is_hashable(SubLink *slink, SubPlan *node)
{
	double		subquery_size;
	List	   *opids;

	/*
	 * The sublink type must be "= ANY" --- that is, an IN operator.
	 * (We require the operator name to be unqualified, which may be
	 * overly paranoid, or may not be.)  XXX since we also check that the
	 * operators are hashable, the test on operator name may be redundant?
	 */
	if (slink->subLinkType != ANY_SUBLINK)
		return false;
	if (length(slink->operName) != 1 ||
		strcmp(strVal(lfirst(slink->operName)), "=") != 0)
		return false;
	/*
	 * The subplan must not have any direct correlation vars --- else we'd
	 * have to recompute its output each time, so that the hashtable wouldn't
	 * gain anything.
	 */
	if (node->parParam != NIL)
		return false;
	/*
	 * The estimated size of the subquery result must fit in SortMem.
	 * (XXX what about hashtable overhead?)
	 */
	subquery_size = node->plan->plan_rows *
		(MAXALIGN(node->plan->plan_width) + MAXALIGN(sizeof(HeapTupleData)));
	if (subquery_size > SortMem * 1024L)
		return false;
	/*
559 560 561 562 563 564 565 566 567 568 569
	 * The combining operators must be hashable, strict, and self-commutative.
	 * The need for hashability is obvious, since we want to use hashing.
	 * Without strictness, behavior in the presence of nulls is too
	 * unpredictable.  (We actually must assume even more than plain
	 * strictness, see nodeSubplan.c for details.)  And commutativity ensures
	 * that the left and right datatypes are the same; this allows us to
	 * assume that the combining operators are equality for the righthand
	 * datatype, so that they can be used to compare righthand tuples as
	 * well as comparing lefthand to righthand tuples.  (This last restriction
	 * could be relaxed by using two different sets of operators with the
	 * hash table, but there is no obvious usefulness to that at present.)
570 571 572 573 574 575 576 577 578 579 580 581 582
	 */
	foreach(opids, slink->operOids)
	{
		Oid			opid = (Oid) lfirsti(opids);
		HeapTuple	tup;
		Form_pg_operator optup;

		tup = SearchSysCache(OPEROID,
							 ObjectIdGetDatum(opid),
							 0, 0, 0);
		if (!HeapTupleIsValid(tup))
			elog(ERROR, "cache lookup failed for operator %u", opid);
		optup = (Form_pg_operator) GETSTRUCT(tup);
583 584
		if (!optup->oprcanhash || optup->oprcom != opid ||
			!func_strict(optup->oprcode))
585 586 587 588 589 590 591
		{
			ReleaseSysCache(tup);
			return false;
		}
		ReleaseSysCache(tup);
	}
	return true;
592 593
}

594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609
/*
 * convert_IN_to_join: can we convert an IN SubLink to join style?
 *
 * The caller has found a SubLink at the top level of WHERE, but has not
 * checked the properties of the SubLink at all.  Decide whether it is
 * appropriate to process this SubLink in join style.  If not, return NULL.
 * If so, build the qual clause(s) to replace the SubLink, and return them.
 *
 * Side effects of a successful conversion include adding the SubLink's
 * subselect to the query's rangetable and adding an InClauseInfo node to
 * its in_info_list.
 */
Node *
convert_IN_to_join(Query *parse, SubLink *sublink)
{
	Query	   *subselect = (Query *) sublink->subselect;
610
	Relids		left_varnos;
611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637
	int			rtindex;
	RangeTblEntry *rte;
	RangeTblRef *rtr;
	InClauseInfo  *ininfo;
	List	   *exprs;

	/*
	 * The sublink type must be "= ANY" --- that is, an IN operator.
	 * (We require the operator name to be unqualified, which may be
	 * overly paranoid, or may not be.)
	 */
	if (sublink->subLinkType != ANY_SUBLINK)
		return NULL;
	if (length(sublink->operName) != 1 ||
		strcmp(strVal(lfirst(sublink->operName)), "=") != 0)
		return NULL;
	/*
	 * The sub-select must not refer to any Vars of the parent query.
	 * (Vars of higher levels should be okay, though.)
	 */
	if (contain_vars_of_level((Node *) subselect, 1))
		return NULL;
	/*
	 * The left-hand expressions must contain some Vars of the current
	 * query, else it's not gonna be a join.
	 */
	left_varnos = pull_varnos((Node *) sublink->lefthand);
638
	if (bms_is_empty(left_varnos))
639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668
		return NULL;
	/*
	 * The left-hand expressions mustn't be volatile.  (Perhaps we should
	 * test the combining operators, too?  We'd only need to point the
	 * function directly at the sublink ...)
	 */
	if (contain_volatile_functions((Node *) sublink->lefthand))
		return NULL;
	/*
	 * Okay, pull up the sub-select into top range table and jointree.
	 *
	 * We rely here on the assumption that the outer query has no references
	 * to the inner (necessarily true, other than the Vars that we build
	 * below).  Therefore this is a lot easier than what pull_up_subqueries
	 * has to go through.
	 */
	rte = addRangeTableEntryForSubquery(NULL,
										subselect,
										makeAlias("IN_subquery", NIL),
										false);
	parse->rtable = lappend(parse->rtable, rte);
	rtindex = length(parse->rtable);
	rtr = makeNode(RangeTblRef);
	rtr->rtindex = rtindex;
	parse->jointree->fromlist = lappend(parse->jointree->fromlist, rtr);
	/*
	 * Now build the InClauseInfo node.
	 */
	ininfo = makeNode(InClauseInfo);
	ininfo->lefthand = left_varnos;
669
	ininfo->righthand = bms_make_singleton(rtindex);
670 671 672 673 674 675 676 677 678 679 680 681 682 683
	parse->in_info_list = lcons(ininfo, parse->in_info_list);
	/*
	 * Build the result qual expressions.  As a side effect,
	 * ininfo->sub_targetlist is filled with a list of the Vars
	 * representing the subselect outputs.
	 */
	exprs = convert_sublink_opers(sublink->lefthand,
								  sublink->operOids,
								  subselect->targetList,
								  rtindex,
								  &ininfo->sub_targetlist);
	return (Node *) make_ands_explicit(exprs);
}

684 685
/*
 * Replace correlation vars (uplevel vars) with Params.
686
 */
687
Node *
688
SS_replace_correlation_vars(Node *expr)
689
{
690 691 692
	/* No setup needed for tree walk, so away we go */
	return replace_correlation_vars_mutator(expr, NULL);
}
693

694 695 696 697 698 699
static Node *
replace_correlation_vars_mutator(Node *node, void *context)
{
	if (node == NULL)
		return NULL;
	if (IsA(node, Var))
700
	{
701 702
		if (((Var *) node)->varlevelsup > 0)
			return (Node *) replace_var((Var *) node);
703
	}
704 705 706
	return expression_tree_mutator(node,
								   replace_correlation_vars_mutator,
								   context);
707 708
}

709 710
/*
 * Expand SubLinks to SubPlans in the given expression.
711 712 713 714
 *
 * The isQual argument tells whether or not this expression is a WHERE/HAVING
 * qualifier expression.  If it is, any sublinks appearing at top level need
 * not distinguish FALSE from UNKNOWN return values.
715
 */
716
Node *
717
SS_process_sublinks(Node *expr, bool isQual)
718
{
719 720
	/* The only context needed is the initial are-we-in-a-qual flag */
	return process_sublinks_mutator(expr, &isQual);
721 722 723
}

static Node *
724
process_sublinks_mutator(Node *node, bool *isTopQual)
725
{
726 727
	bool	locTopQual;

728
	if (node == NULL)
729
		return NULL;
730
	if (IsA(node, SubLink))
731
	{
732
		SubLink    *sublink = (SubLink *) node;
733
		List	   *lefthand;
734

735
		/*
736
		 * First, recursively process the lefthand-side expressions, if any.
737
		 */
738
		locTopQual = false;
739
		lefthand = (List *)
740
			process_sublinks_mutator((Node *) sublink->lefthand, &locTopQual);
741 742 743
		/*
		 * Now build the SubPlan node and make the expr to return.
		 */
744
		return make_subplan(sublink, lefthand, *isTopQual);
745
	}
746

747
	/*
748 749 750
	 * We should never see a SubPlan expression in the input (since this is
	 * the very routine that creates 'em to begin with).  We shouldn't find
	 * ourselves invoked directly on a Query, either.
751
	 */
752
	Assert(!is_subplan(node));
753
	Assert(!IsA(node, Query));
754

755 756 757 758 759 760 761 762 763
	/*
	 * If we recurse down through anything other than a List node, we are
	 * definitely not at top qual level anymore.
	 */
	if (IsA(node, List))
		locTopQual = *isTopQual;
	else
		locTopQual = false;

764 765
	return expression_tree_mutator(node,
								   process_sublinks_mutator,
766
								   (void *) &locTopQual);
767 768
}

769 770 771 772 773 774
/*
 * SS_finalize_plan - do final sublink processing for a completed Plan.
 *
 * This recursively computes and sets the extParam and locParam lists
 * for every Plan node in the given tree.
 */
775
List *
776
SS_finalize_plan(Plan *plan, List *rtable)
777
{
778 779 780
	List	   *extParam = NIL;
	List	   *locParam = NIL;
	finalize_primnode_results results;
781 782 783
	List	   *lst;

	if (plan == NULL)
784
		return NIL;
785

786
	results.paramids = NIL;		/* initialize list to NIL */
787

788 789
	/*
	 * When we call finalize_primnode, results.paramids lists are
790 791
	 * automatically merged together.  But when recursing to self, we have
	 * to do it the hard way.  We want the paramids list to include params
792
	 * in subplans as well as at this level.
793 794
	 */

795
	/* Find params in targetlist and qual */
796
	finalize_primnode((Node *) plan->targetlist, &results);
797
	finalize_primnode((Node *) plan->qual, &results);
798

799
	/* Check additional node-type-specific fields */
800 801 802
	switch (nodeTag(plan))
	{
		case T_Result:
803 804
			finalize_primnode(((Result *) plan)->resconstantqual,
							  &results);
805 806
			break;

807 808 809 810 811 812
		case T_IndexScan:
			finalize_primnode((Node *) ((IndexScan *) plan)->indxqual,
							  &results);

			/*
			 * we need not look at indxqualorig, since it will have the
813
			 * same param references as indxqual.
814 815 816 817 818 819
			 */
			break;

		case T_TidScan:
			finalize_primnode((Node *) ((TidScan *) plan)->tideval,
							  &results);
820
			break;
821

822
		case T_SubqueryScan:
B
Bruce Momjian 已提交
823

824
			/*
B
Bruce Momjian 已提交
825 826 827 828 829
			 * In a SubqueryScan, SS_finalize_plan has already been run on
			 * the subplan by the inner invocation of subquery_planner, so
			 * there's no need to do it again.  Instead, just pull out the
			 * subplan's extParams list, which represents the params it
			 * needs from my level and higher levels.
830
			 */
831
			results.paramids = set_unioni(results.paramids,
B
Bruce Momjian 已提交
832
							 ((SubqueryScan *) plan)->subplan->extParam);
833 834
			break;

835 836 837
		case T_FunctionScan:
			{
				RangeTblEntry *rte;
838

839 840 841 842 843 844 845 846 847 848
				rte = rt_fetch(((FunctionScan *) plan)->scan.scanrelid,
							   rtable);
				Assert(rte->rtekind == RTE_FUNCTION);
				finalize_primnode(rte->funcexpr, &results);
			}
			break;

		case T_Append:
			foreach(lst, ((Append *) plan)->appendplans)
				results.paramids = set_unioni(results.paramids,
B
Bruce Momjian 已提交
849 850
								   SS_finalize_plan((Plan *) lfirst(lst),
													rtable));
851 852
			break;

853 854 855 856 857
		case T_NestLoop:
			finalize_primnode((Node *) ((Join *) plan)->joinqual,
							  &results);
			break;

858
		case T_MergeJoin:
859 860
			finalize_primnode((Node *) ((Join *) plan)->joinqual,
							  &results);
861 862
			finalize_primnode((Node *) ((MergeJoin *) plan)->mergeclauses,
							  &results);
863 864 865
			break;

		case T_HashJoin:
866 867
			finalize_primnode((Node *) ((Join *) plan)->joinqual,
							  &results);
868 869
			finalize_primnode((Node *) ((HashJoin *) plan)->hashclauses,
							  &results);
870
			break;
871

872
		case T_Hash:
873
			finalize_primnode((Node *) ((Hash *) plan)->hashkeys,
874
							  &results);
875 876 877 878 879 880 881
			break;

		case T_Agg:
		case T_SeqScan:
		case T_Material:
		case T_Sort:
		case T_Unique:
882
		case T_SetOp:
883
		case T_Limit:
884 885
		case T_Group:
			break;
886

887
		default:
888 889
			elog(ERROR, "SS_finalize_plan: node %d unsupported",
				 nodeTag(plan));
890
	}
891

892
	/* Process left and right child plans, if any */
893
	results.paramids = set_unioni(results.paramids,
894 895
								  SS_finalize_plan(plan->lefttree,
												   rtable));
896
	results.paramids = set_unioni(results.paramids,
897 898
								  SS_finalize_plan(plan->righttree,
												   rtable));
899

900
	/* Now we have all the paramids */
901

902
	foreach(lst, results.paramids)
903
	{
904 905
		int			paramid = lfirsti(lst);
		Var		   *var = nth(paramid, PlannerParamVar);
906

907
		/* note varlevelsup is absolute level number */
908
		if (var->varlevelsup < PlannerQueryLevel)
909
			extParam = lappendi(extParam, paramid);
910
		else if (var->varlevelsup > PlannerQueryLevel)
911
			elog(ERROR, "SS_finalize_plan: plan shouldn't reference subplan's variable");
912 913
		else
		{
914
			Assert(var->varno == 0 && var->varattno == 0);
915
			locParam = lappendi(locParam, paramid);
916 917
		}
	}
918

919 920 921
	plan->extParam = extParam;
	plan->locParam = locParam;

922
	return results.paramids;
923
}
924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966

/*
 * finalize_primnode: build lists of params appearing
 * in the given expression tree.  NOTE: items are added to list passed in,
 * so caller must initialize list to NIL before first call!
 */
static bool
finalize_primnode(Node *node, finalize_primnode_results *results)
{
	if (node == NULL)
		return false;
	if (IsA(node, Param))
	{
		if (((Param *) node)->paramkind == PARAM_EXEC)
		{
			int			paramid = (int) ((Param *) node)->paramid;

			if (!intMember(paramid, results->paramids))
				results->paramids = lconsi(paramid, results->paramids);
		}
		return false;			/* no more to do here */
	}
	if (is_subplan(node))
	{
		SubPlan	   *subplan = (SubPlan *) node;
		List	   *lst;

		/* Check extParam list for params to add to paramids */
		foreach(lst, subplan->plan->extParam)
		{
			int			paramid = lfirsti(lst);
			Var		   *var = nth(paramid, PlannerParamVar);

			/* note varlevelsup is absolute level number */
			if (var->varlevelsup < PlannerQueryLevel &&
				!intMember(paramid, results->paramids))
				results->paramids = lconsi(paramid, results->paramids);
		}
		/* fall through to recurse into subplan args */
	}
	return expression_tree_walker(node, finalize_primnode,
								  (void *) results);
}