Line data Source code
1 : /*-------------------------------------------------------------------------
2 : *
3 : * equivclass.c
4 : * Routines for managing EquivalenceClasses
5 : *
6 : * See src/backend/optimizer/README for discussion of EquivalenceClasses.
7 : *
8 : *
9 : * Portions Copyright (c) 1996-2024, PostgreSQL Global Development Group
10 : * Portions Copyright (c) 1994, Regents of the University of California
11 : *
12 : * IDENTIFICATION
13 : * src/backend/optimizer/path/equivclass.c
14 : *
15 : *-------------------------------------------------------------------------
16 : */
17 : #include "postgres.h"
18 :
19 : #include <limits.h>
20 :
21 : #include "access/stratnum.h"
22 : #include "catalog/pg_type.h"
23 : #include "nodes/makefuncs.h"
24 : #include "nodes/nodeFuncs.h"
25 : #include "optimizer/appendinfo.h"
26 : #include "optimizer/clauses.h"
27 : #include "optimizer/optimizer.h"
28 : #include "optimizer/pathnode.h"
29 : #include "optimizer/paths.h"
30 : #include "optimizer/planmain.h"
31 : #include "optimizer/restrictinfo.h"
32 : #include "rewrite/rewriteManip.h"
33 : #include "utils/lsyscache.h"
34 :
35 :
36 : static EquivalenceMember *add_eq_member(EquivalenceClass *ec,
37 : Expr *expr, Relids relids,
38 : JoinDomain *jdomain,
39 : EquivalenceMember *parent,
40 : Oid datatype);
41 : static bool is_exprlist_member(Expr *node, List *exprs);
42 : static void generate_base_implied_equalities_const(PlannerInfo *root,
43 : EquivalenceClass *ec);
44 : static void generate_base_implied_equalities_no_const(PlannerInfo *root,
45 : EquivalenceClass *ec);
46 : static void generate_base_implied_equalities_broken(PlannerInfo *root,
47 : EquivalenceClass *ec);
48 : static List *generate_join_implied_equalities_normal(PlannerInfo *root,
49 : EquivalenceClass *ec,
50 : Relids join_relids,
51 : Relids outer_relids,
52 : Relids inner_relids);
53 : static List *generate_join_implied_equalities_broken(PlannerInfo *root,
54 : EquivalenceClass *ec,
55 : Relids nominal_join_relids,
56 : Relids outer_relids,
57 : Relids nominal_inner_relids,
58 : RelOptInfo *inner_rel);
59 : static Oid select_equality_operator(EquivalenceClass *ec,
60 : Oid lefttype, Oid righttype);
61 : static RestrictInfo *create_join_clause(PlannerInfo *root,
62 : EquivalenceClass *ec, Oid opno,
63 : EquivalenceMember *leftem,
64 : EquivalenceMember *rightem,
65 : EquivalenceClass *parent_ec);
66 : static bool reconsider_outer_join_clause(PlannerInfo *root,
67 : OuterJoinClauseInfo *ojcinfo,
68 : bool outer_on_left);
69 : static bool reconsider_full_join_clause(PlannerInfo *root,
70 : OuterJoinClauseInfo *ojcinfo);
71 : static JoinDomain *find_join_domain(PlannerInfo *root, Relids relids);
72 : static Bitmapset *get_eclass_indexes_for_relids(PlannerInfo *root,
73 : Relids relids);
74 : static Bitmapset *get_common_eclass_indexes(PlannerInfo *root, Relids relids1,
75 : Relids relids2);
76 :
77 :
78 : /*
79 : * process_equivalence
80 : * The given clause has a mergejoinable operator and is not an outer-join
81 : * qualification, so its two sides can be considered equal
82 : * anywhere they are both computable; moreover that equality can be
83 : * extended transitively. Record this knowledge in the EquivalenceClass
84 : * data structure, if applicable. Returns true if successful, false if not
85 : * (in which case caller should treat the clause as ordinary, not an
86 : * equivalence).
87 : *
88 : * In some cases, although we cannot convert a clause into EquivalenceClass
89 : * knowledge, we can still modify it to a more useful form than the original.
90 : * Then, *p_restrictinfo will be replaced by a new RestrictInfo, which is what
91 : * the caller should use for further processing.
92 : *
93 : * jdomain is the join domain within which the given clause was found.
94 : * This limits the applicability of deductions from the EquivalenceClass,
95 : * as described in optimizer/README.
96 : *
97 : * We reject proposed equivalence clauses if they contain leaky functions
98 : * and have security_level above zero. The EC evaluation rules require us to
99 : * apply certain tests at certain joining levels, and we can't tolerate
100 : * delaying any test on security_level grounds. By rejecting candidate clauses
101 : * that might require security delays, we ensure it's safe to apply an EC
102 : * clause as soon as it's supposed to be applied.
103 : *
104 : * On success return, we have also initialized the clause's left_ec/right_ec
105 : * fields to point to the EquivalenceClass representing it. This saves lookup
106 : * effort later.
107 : *
108 : * Note: constructing merged EquivalenceClasses is a standard UNION-FIND
109 : * problem, for which there exist better data structures than simple lists.
110 : * If this code ever proves to be a bottleneck then it could be sped up ---
111 : * but for now, simple is beautiful.
112 : *
113 : * Note: this is only called during planner startup, not during GEQO
114 : * exploration, so we need not worry about whether we're in the right
115 : * memory context.
116 : */
117 : bool
118 247770 : process_equivalence(PlannerInfo *root,
119 : RestrictInfo **p_restrictinfo,
120 : JoinDomain *jdomain)
121 : {
122 247770 : RestrictInfo *restrictinfo = *p_restrictinfo;
123 247770 : Expr *clause = restrictinfo->clause;
124 : Oid opno,
125 : collation,
126 : item1_type,
127 : item2_type;
128 : Expr *item1;
129 : Expr *item2;
130 : Relids item1_relids,
131 : item2_relids;
132 : List *opfamilies;
133 : EquivalenceClass *ec1,
134 : *ec2;
135 : EquivalenceMember *em1,
136 : *em2;
137 : ListCell *lc1;
138 : int ec2_idx;
139 :
140 : /* Should not already be marked as having generated an eclass */
141 : Assert(restrictinfo->left_ec == NULL);
142 : Assert(restrictinfo->right_ec == NULL);
143 :
144 : /* Reject if it is potentially postponable by security considerations */
145 247770 : if (restrictinfo->security_level > 0 && !restrictinfo->leakproof)
146 202 : return false;
147 :
148 : /* Extract info from given clause */
149 : Assert(is_opclause(clause));
150 247568 : opno = ((OpExpr *) clause)->opno;
151 247568 : collation = ((OpExpr *) clause)->inputcollid;
152 247568 : item1 = (Expr *) get_leftop(clause);
153 247568 : item2 = (Expr *) get_rightop(clause);
154 247568 : item1_relids = restrictinfo->left_relids;
155 247568 : item2_relids = restrictinfo->right_relids;
156 :
157 : /*
158 : * Ensure both input expressions expose the desired collation (their types
159 : * should be OK already); see comments for canonicalize_ec_expression.
160 : */
161 247568 : item1 = canonicalize_ec_expression(item1,
162 : exprType((Node *) item1),
163 : collation);
164 247568 : item2 = canonicalize_ec_expression(item2,
165 : exprType((Node *) item2),
166 : collation);
167 :
168 : /*
169 : * Clauses of the form X=X cannot be translated into EquivalenceClasses.
170 : * We'd either end up with a single-entry EC, losing the knowledge that
171 : * the clause was present at all, or else make an EC with duplicate
172 : * entries, causing other issues.
173 : */
174 247568 : if (equal(item1, item2))
175 : {
176 : /*
177 : * If the operator is strict, then the clause can be treated as just
178 : * "X IS NOT NULL". (Since we know we are considering a top-level
179 : * qual, we can ignore the difference between FALSE and NULL results.)
180 : * It's worth making the conversion because we'll typically get a much
181 : * better selectivity estimate than we would for X=X.
182 : *
183 : * If the operator is not strict, we can't be sure what it will do
184 : * with NULLs, so don't attempt to optimize it.
185 : */
186 54 : set_opfuncid((OpExpr *) clause);
187 54 : if (func_strict(((OpExpr *) clause)->opfuncid))
188 : {
189 54 : NullTest *ntest = makeNode(NullTest);
190 :
191 54 : ntest->arg = item1;
192 54 : ntest->nulltesttype = IS_NOT_NULL;
193 54 : ntest->argisrow = false; /* correct even if composite arg */
194 54 : ntest->location = -1;
195 :
196 54 : *p_restrictinfo =
197 54 : make_restrictinfo(root,
198 : (Expr *) ntest,
199 54 : restrictinfo->is_pushed_down,
200 54 : restrictinfo->has_clone,
201 54 : restrictinfo->is_clone,
202 54 : restrictinfo->pseudoconstant,
203 : restrictinfo->security_level,
204 : NULL,
205 : restrictinfo->incompatible_relids,
206 : restrictinfo->outer_relids);
207 : }
208 54 : return false;
209 : }
210 :
211 : /*
212 : * We use the declared input types of the operator, not exprType() of the
213 : * inputs, as the nominal datatypes for opfamily lookup. This presumes
214 : * that btree operators are always registered with amoplefttype and
215 : * amoprighttype equal to their declared input types. We will need this
216 : * info anyway to build EquivalenceMember nodes, and by extracting it now
217 : * we can use type comparisons to short-circuit some equal() tests.
218 : */
219 247514 : op_input_types(opno, &item1_type, &item2_type);
220 :
221 247514 : opfamilies = restrictinfo->mergeopfamilies;
222 :
223 : /*
224 : * Sweep through the existing EquivalenceClasses looking for matches to
225 : * item1 and item2. These are the possible outcomes:
226 : *
227 : * 1. We find both in the same EC. The equivalence is already known, so
228 : * there's nothing to do.
229 : *
230 : * 2. We find both in different ECs. Merge the two ECs together.
231 : *
232 : * 3. We find just one. Add the other to its EC.
233 : *
234 : * 4. We find neither. Make a new, two-entry EC.
235 : *
236 : * Note: since all ECs are built through this process or the similar
237 : * search in get_eclass_for_sort_expr(), it's impossible that we'd match
238 : * an item in more than one existing nonvolatile EC. So it's okay to stop
239 : * at the first match.
240 : */
241 247514 : ec1 = ec2 = NULL;
242 247514 : em1 = em2 = NULL;
243 247514 : ec2_idx = -1;
244 393660 : foreach(lc1, root->eq_classes)
245 : {
246 146188 : EquivalenceClass *cur_ec = (EquivalenceClass *) lfirst(lc1);
247 : ListCell *lc2;
248 :
249 : /* Never match to a volatile EC */
250 146188 : if (cur_ec->ec_has_volatile)
251 0 : continue;
252 :
253 : /*
254 : * The collation has to match; check this first since it's cheaper
255 : * than the opfamily comparison.
256 : */
257 146188 : if (collation != cur_ec->ec_collation)
258 11472 : continue;
259 :
260 : /*
261 : * A "match" requires matching sets of btree opfamilies. Use of
262 : * equal() for this test has implications discussed in the comments
263 : * for get_mergejoin_opfamilies().
264 : */
265 134716 : if (!equal(opfamilies, cur_ec->ec_opfamilies))
266 34040 : continue;
267 :
268 304022 : foreach(lc2, cur_ec->ec_members)
269 : {
270 203388 : EquivalenceMember *cur_em = (EquivalenceMember *) lfirst(lc2);
271 :
272 : Assert(!cur_em->em_is_child); /* no children yet */
273 :
274 : /*
275 : * Match constants only within the same JoinDomain (see
276 : * optimizer/README).
277 : */
278 203388 : if (cur_em->em_is_const && cur_em->em_jdomain != jdomain)
279 3878 : continue;
280 :
281 199510 : if (!ec1 &&
282 375836 : item1_type == cur_em->em_datatype &&
283 187792 : equal(item1, cur_em->em_expr))
284 : {
285 15384 : ec1 = cur_ec;
286 15384 : em1 = cur_em;
287 15384 : if (ec2)
288 18 : break;
289 : }
290 :
291 199492 : if (!ec2 &&
292 393816 : item2_type == cur_em->em_datatype &&
293 196716 : equal(item2, cur_em->em_expr))
294 : {
295 4706 : ec2 = cur_ec;
296 4706 : ec2_idx = foreach_current_index(lc1);
297 4706 : em2 = cur_em;
298 4706 : if (ec1)
299 24 : break;
300 : }
301 : }
302 :
303 100676 : if (ec1 && ec2)
304 42 : break;
305 : }
306 :
307 : /* Sweep finished, what did we find? */
308 :
309 247514 : if (ec1 && ec2)
310 : {
311 : /* If case 1, nothing to do, except add to sources */
312 42 : if (ec1 == ec2)
313 : {
314 12 : ec1->ec_sources = lappend(ec1->ec_sources, restrictinfo);
315 12 : ec1->ec_min_security = Min(ec1->ec_min_security,
316 : restrictinfo->security_level);
317 12 : ec1->ec_max_security = Max(ec1->ec_max_security,
318 : restrictinfo->security_level);
319 : /* mark the RI as associated with this eclass */
320 12 : restrictinfo->left_ec = ec1;
321 12 : restrictinfo->right_ec = ec1;
322 : /* mark the RI as usable with this pair of EMs */
323 12 : restrictinfo->left_em = em1;
324 12 : restrictinfo->right_em = em2;
325 12 : return true;
326 : }
327 :
328 : /*
329 : * Case 2: need to merge ec1 and ec2. This should never happen after
330 : * the ECs have reached canonical state; otherwise, pathkeys could be
331 : * rendered non-canonical by the merge, and relation eclass indexes
332 : * would get broken by removal of an eq_classes list entry.
333 : */
334 30 : if (root->ec_merging_done)
335 0 : elog(ERROR, "too late to merge equivalence classes");
336 :
337 : /*
338 : * We add ec2's items to ec1, then set ec2's ec_merged link to point
339 : * to ec1 and remove ec2 from the eq_classes list. We cannot simply
340 : * delete ec2 because that could leave dangling pointers in existing
341 : * PathKeys. We leave it behind with a link so that the merged EC can
342 : * be found.
343 : */
344 30 : ec1->ec_members = list_concat(ec1->ec_members, ec2->ec_members);
345 30 : ec1->ec_sources = list_concat(ec1->ec_sources, ec2->ec_sources);
346 30 : ec1->ec_derives = list_concat(ec1->ec_derives, ec2->ec_derives);
347 30 : ec1->ec_relids = bms_join(ec1->ec_relids, ec2->ec_relids);
348 30 : ec1->ec_has_const |= ec2->ec_has_const;
349 : /* can't need to set has_volatile */
350 30 : ec1->ec_min_security = Min(ec1->ec_min_security,
351 : ec2->ec_min_security);
352 30 : ec1->ec_max_security = Max(ec1->ec_max_security,
353 : ec2->ec_max_security);
354 30 : ec2->ec_merged = ec1;
355 30 : root->eq_classes = list_delete_nth_cell(root->eq_classes, ec2_idx);
356 : /* just to avoid debugging confusion w/ dangling pointers: */
357 30 : ec2->ec_members = NIL;
358 30 : ec2->ec_sources = NIL;
359 30 : ec2->ec_derives = NIL;
360 30 : ec2->ec_relids = NULL;
361 30 : ec1->ec_sources = lappend(ec1->ec_sources, restrictinfo);
362 30 : ec1->ec_min_security = Min(ec1->ec_min_security,
363 : restrictinfo->security_level);
364 30 : ec1->ec_max_security = Max(ec1->ec_max_security,
365 : restrictinfo->security_level);
366 : /* mark the RI as associated with this eclass */
367 30 : restrictinfo->left_ec = ec1;
368 30 : restrictinfo->right_ec = ec1;
369 : /* mark the RI as usable with this pair of EMs */
370 30 : restrictinfo->left_em = em1;
371 30 : restrictinfo->right_em = em2;
372 : }
373 247472 : else if (ec1)
374 : {
375 : /* Case 3: add item2 to ec1 */
376 15342 : em2 = add_eq_member(ec1, item2, item2_relids,
377 : jdomain, NULL, item2_type);
378 15342 : ec1->ec_sources = lappend(ec1->ec_sources, restrictinfo);
379 15342 : ec1->ec_min_security = Min(ec1->ec_min_security,
380 : restrictinfo->security_level);
381 15342 : ec1->ec_max_security = Max(ec1->ec_max_security,
382 : restrictinfo->security_level);
383 : /* mark the RI as associated with this eclass */
384 15342 : restrictinfo->left_ec = ec1;
385 15342 : restrictinfo->right_ec = ec1;
386 : /* mark the RI as usable with this pair of EMs */
387 15342 : restrictinfo->left_em = em1;
388 15342 : restrictinfo->right_em = em2;
389 : }
390 232130 : else if (ec2)
391 : {
392 : /* Case 3: add item1 to ec2 */
393 4664 : em1 = add_eq_member(ec2, item1, item1_relids,
394 : jdomain, NULL, item1_type);
395 4664 : ec2->ec_sources = lappend(ec2->ec_sources, restrictinfo);
396 4664 : ec2->ec_min_security = Min(ec2->ec_min_security,
397 : restrictinfo->security_level);
398 4664 : ec2->ec_max_security = Max(ec2->ec_max_security,
399 : restrictinfo->security_level);
400 : /* mark the RI as associated with this eclass */
401 4664 : restrictinfo->left_ec = ec2;
402 4664 : restrictinfo->right_ec = ec2;
403 : /* mark the RI as usable with this pair of EMs */
404 4664 : restrictinfo->left_em = em1;
405 4664 : restrictinfo->right_em = em2;
406 : }
407 : else
408 : {
409 : /* Case 4: make a new, two-entry EC */
410 227466 : EquivalenceClass *ec = makeNode(EquivalenceClass);
411 :
412 227466 : ec->ec_opfamilies = opfamilies;
413 227466 : ec->ec_collation = collation;
414 227466 : ec->ec_members = NIL;
415 227466 : ec->ec_sources = list_make1(restrictinfo);
416 227466 : ec->ec_derives = NIL;
417 227466 : ec->ec_relids = NULL;
418 227466 : ec->ec_has_const = false;
419 227466 : ec->ec_has_volatile = false;
420 227466 : ec->ec_broken = false;
421 227466 : ec->ec_sortref = 0;
422 227466 : ec->ec_min_security = restrictinfo->security_level;
423 227466 : ec->ec_max_security = restrictinfo->security_level;
424 227466 : ec->ec_merged = NULL;
425 227466 : em1 = add_eq_member(ec, item1, item1_relids,
426 : jdomain, NULL, item1_type);
427 227466 : em2 = add_eq_member(ec, item2, item2_relids,
428 : jdomain, NULL, item2_type);
429 :
430 227466 : root->eq_classes = lappend(root->eq_classes, ec);
431 :
432 : /* mark the RI as associated with this eclass */
433 227466 : restrictinfo->left_ec = ec;
434 227466 : restrictinfo->right_ec = ec;
435 : /* mark the RI as usable with this pair of EMs */
436 227466 : restrictinfo->left_em = em1;
437 227466 : restrictinfo->right_em = em2;
438 : }
439 :
440 247502 : return true;
441 : }
442 :
443 : /*
444 : * canonicalize_ec_expression
445 : *
446 : * This function ensures that the expression exposes the expected type and
447 : * collation, so that it will be equal() to other equivalence-class expressions
448 : * that it ought to be equal() to.
449 : *
450 : * The rule for datatypes is that the exposed type should match what it would
451 : * be for an input to an operator of the EC's opfamilies; which is usually
452 : * the declared input type of the operator, but in the case of polymorphic
453 : * operators no relabeling is wanted (compare the behavior of parse_coerce.c).
454 : * Expressions coming in from quals will generally have the right type
455 : * already, but expressions coming from indexkeys may not (because they are
456 : * represented without any explicit relabel in pg_index), and the same problem
457 : * occurs for sort expressions (because the parser is likewise cavalier about
458 : * putting relabels on them). Such cases will be binary-compatible with the
459 : * real operators, so adding a RelabelType is sufficient.
460 : *
461 : * Also, the expression's exposed collation must match the EC's collation.
462 : * This is important because in comparisons like "foo < bar COLLATE baz",
463 : * only one of the expressions has the correct exposed collation as we receive
464 : * it from the parser. Forcing both of them to have it ensures that all
465 : * variant spellings of such a construct behave the same. Again, we can
466 : * stick on a RelabelType to force the right exposed collation. (It might
467 : * work to not label the collation at all in EC members, but this is risky
468 : * since some parts of the system expect exprCollation() to deliver the
469 : * right answer for a sort key.)
470 : */
471 : Expr *
472 2223516 : canonicalize_ec_expression(Expr *expr, Oid req_type, Oid req_collation)
473 : {
474 2223516 : Oid expr_type = exprType((Node *) expr);
475 :
476 : /*
477 : * For a polymorphic-input-type opclass, just keep the same exposed type.
478 : * RECORD opclasses work like polymorphic-type ones for this purpose.
479 : */
480 2223516 : if (IsPolymorphicType(req_type) || req_type == RECORDOID)
481 3452 : req_type = expr_type;
482 :
483 : /*
484 : * No work if the expression exposes the right type/collation already.
485 : */
486 4412384 : if (expr_type != req_type ||
487 2188868 : exprCollation((Node *) expr) != req_collation)
488 : {
489 : /*
490 : * If we have to change the type of the expression, set typmod to -1,
491 : * since the new type may not have the same typmod interpretation.
492 : * When we only have to change collation, preserve the exposed typmod.
493 : */
494 : int32 req_typmod;
495 :
496 35464 : if (expr_type != req_type)
497 34648 : req_typmod = -1;
498 : else
499 816 : req_typmod = exprTypmod((Node *) expr);
500 :
501 : /*
502 : * Use applyRelabelType so that we preserve const-flatness. This is
503 : * important since eval_const_expressions has already been applied.
504 : */
505 35464 : expr = (Expr *) applyRelabelType((Node *) expr,
506 : req_type, req_typmod, req_collation,
507 : COERCE_IMPLICIT_CAST, -1, false);
508 : }
509 :
510 2223516 : return expr;
511 : }
512 :
513 : /*
514 : * add_eq_member - build a new EquivalenceMember and add it to an EC
515 : */
516 : static EquivalenceMember *
517 756412 : add_eq_member(EquivalenceClass *ec, Expr *expr, Relids relids,
518 : JoinDomain *jdomain, EquivalenceMember *parent, Oid datatype)
519 : {
520 756412 : EquivalenceMember *em = makeNode(EquivalenceMember);
521 :
522 756412 : em->em_expr = expr;
523 756412 : em->em_relids = relids;
524 756412 : em->em_is_const = false;
525 756412 : em->em_is_child = (parent != NULL);
526 756412 : em->em_datatype = datatype;
527 756412 : em->em_jdomain = jdomain;
528 756412 : em->em_parent = parent;
529 :
530 756412 : if (bms_is_empty(relids))
531 : {
532 : /*
533 : * No Vars, assume it's a pseudoconstant. This is correct for entries
534 : * generated from process_equivalence(), because a WHERE clause can't
535 : * contain aggregates or SRFs, and non-volatility was checked before
536 : * process_equivalence() ever got called. But
537 : * get_eclass_for_sort_expr() has to work harder. We put the tests
538 : * there not here to save cycles in the equivalence case.
539 : */
540 : Assert(!parent);
541 187944 : em->em_is_const = true;
542 187944 : ec->ec_has_const = true;
543 : /* it can't affect ec_relids */
544 : }
545 568468 : else if (!parent) /* child members don't add to ec_relids */
546 : {
547 497310 : ec->ec_relids = bms_add_members(ec->ec_relids, relids);
548 : }
549 756412 : ec->ec_members = lappend(ec->ec_members, em);
550 :
551 756412 : return em;
552 : }
553 :
554 :
555 : /*
556 : * get_eclass_for_sort_expr
557 : * Given an expression and opfamily/collation info, find an existing
558 : * equivalence class it is a member of; if none, optionally build a new
559 : * single-member EquivalenceClass for it.
560 : *
561 : * sortref is the SortGroupRef of the originating SortGroupClause, if any,
562 : * or zero if not. (It should never be zero if the expression is volatile!)
563 : *
564 : * If rel is not NULL, it identifies a specific relation we're considering
565 : * a path for, and indicates that child EC members for that relation can be
566 : * considered. Otherwise child members are ignored. (Note: since child EC
567 : * members aren't guaranteed unique, a non-NULL value means that there could
568 : * be more than one EC that matches the expression; if so it's order-dependent
569 : * which one you get. This is annoying but it only happens in corner cases,
570 : * so for now we live with just reporting the first match. See also
571 : * generate_implied_equalities_for_column and match_pathkeys_to_index.)
572 : *
573 : * If create_it is true, we'll build a new EquivalenceClass when there is no
574 : * match. If create_it is false, we just return NULL when no match.
575 : *
576 : * This can be used safely both before and after EquivalenceClass merging;
577 : * since it never causes merging it does not invalidate any existing ECs
578 : * or PathKeys. However, ECs added after path generation has begun are
579 : * of limited usefulness, so usually it's best to create them beforehand.
580 : *
581 : * Note: opfamilies must be chosen consistently with the way
582 : * process_equivalence() would do; that is, generated from a mergejoinable
583 : * equality operator. Else we might fail to detect valid equivalences,
584 : * generating poor (but not incorrect) plans.
585 : */
586 : EquivalenceClass *
587 1655208 : get_eclass_for_sort_expr(PlannerInfo *root,
588 : Expr *expr,
589 : List *opfamilies,
590 : Oid opcintype,
591 : Oid collation,
592 : Index sortref,
593 : Relids rel,
594 : bool create_it)
595 : {
596 : JoinDomain *jdomain;
597 : Relids expr_relids;
598 : EquivalenceClass *newec;
599 : EquivalenceMember *newem;
600 : ListCell *lc1;
601 : MemoryContext oldcontext;
602 :
603 : /*
604 : * Ensure the expression exposes the correct type and collation.
605 : */
606 1655208 : expr = canonicalize_ec_expression(expr, opcintype, collation);
607 :
608 : /*
609 : * Since SortGroupClause nodes are top-level expressions (GROUP BY, ORDER
610 : * BY, etc), they can be presumed to belong to the top JoinDomain.
611 : */
612 1655208 : jdomain = linitial_node(JoinDomain, root->join_domains);
613 :
614 : /*
615 : * Scan through the existing EquivalenceClasses for a match
616 : */
617 5434344 : foreach(lc1, root->eq_classes)
618 : {
619 4700890 : EquivalenceClass *cur_ec = (EquivalenceClass *) lfirst(lc1);
620 : ListCell *lc2;
621 :
622 : /*
623 : * Never match to a volatile EC, except when we are looking at another
624 : * reference to the same volatile SortGroupClause.
625 : */
626 4700890 : if (cur_ec->ec_has_volatile &&
627 36 : (sortref == 0 || sortref != cur_ec->ec_sortref))
628 524 : continue;
629 :
630 4700366 : if (collation != cur_ec->ec_collation)
631 1282144 : continue;
632 3418222 : if (!equal(opfamilies, cur_ec->ec_opfamilies))
633 680578 : continue;
634 :
635 6210858 : foreach(lc2, cur_ec->ec_members)
636 : {
637 4394968 : EquivalenceMember *cur_em = (EquivalenceMember *) lfirst(lc2);
638 :
639 : /*
640 : * Ignore child members unless they match the request.
641 : */
642 4394968 : if (cur_em->em_is_child &&
643 294056 : !bms_equal(cur_em->em_relids, rel))
644 221640 : continue;
645 :
646 : /*
647 : * Match constants only within the same JoinDomain (see
648 : * optimizer/README).
649 : */
650 4173328 : if (cur_em->em_is_const && cur_em->em_jdomain != jdomain)
651 75382 : continue;
652 :
653 8166110 : if (opcintype == cur_em->em_datatype &&
654 4068164 : equal(expr, cur_em->em_expr))
655 : {
656 : /*
657 : * Match!
658 : *
659 : * Copy the sortref if it wasn't set yet. That may happen if
660 : * the ec was constructed from a WHERE clause, i.e. it doesn't
661 : * have a target reference at all.
662 : */
663 921754 : if (cur_ec->ec_sortref == 0 && sortref > 0)
664 8216 : cur_ec->ec_sortref = sortref;
665 921754 : return cur_ec;
666 : }
667 : }
668 : }
669 :
670 : /* No match; does caller want a NULL result? */
671 733454 : if (!create_it)
672 523138 : return NULL;
673 :
674 : /*
675 : * OK, build a new single-member EC
676 : *
677 : * Here, we must be sure that we construct the EC in the right context.
678 : */
679 210316 : oldcontext = MemoryContextSwitchTo(root->planner_cxt);
680 :
681 210316 : newec = makeNode(EquivalenceClass);
682 210316 : newec->ec_opfamilies = list_copy(opfamilies);
683 210316 : newec->ec_collation = collation;
684 210316 : newec->ec_members = NIL;
685 210316 : newec->ec_sources = NIL;
686 210316 : newec->ec_derives = NIL;
687 210316 : newec->ec_relids = NULL;
688 210316 : newec->ec_has_const = false;
689 210316 : newec->ec_has_volatile = contain_volatile_functions((Node *) expr);
690 210316 : newec->ec_broken = false;
691 210316 : newec->ec_sortref = sortref;
692 210316 : newec->ec_min_security = UINT_MAX;
693 210316 : newec->ec_max_security = 0;
694 210316 : newec->ec_merged = NULL;
695 :
696 210316 : if (newec->ec_has_volatile && sortref == 0) /* should not happen */
697 0 : elog(ERROR, "volatile EquivalenceClass has no sortref");
698 :
699 : /*
700 : * Get the precise set of relids appearing in the expression.
701 : */
702 210316 : expr_relids = pull_varnos(root, (Node *) expr);
703 :
704 210316 : newem = add_eq_member(newec, copyObject(expr), expr_relids,
705 : jdomain, NULL, opcintype);
706 :
707 : /*
708 : * add_eq_member doesn't check for volatile functions, set-returning
709 : * functions, aggregates, or window functions, but such could appear in
710 : * sort expressions; so we have to check whether its const-marking was
711 : * correct.
712 : */
713 210316 : if (newec->ec_has_const)
714 : {
715 17514 : if (newec->ec_has_volatile ||
716 17242 : expression_returns_set((Node *) expr) ||
717 16916 : contain_agg_clause((Node *) expr) ||
718 8370 : contain_window_function((Node *) expr))
719 : {
720 454 : newec->ec_has_const = false;
721 454 : newem->em_is_const = false;
722 : }
723 : }
724 :
725 210316 : root->eq_classes = lappend(root->eq_classes, newec);
726 :
727 : /*
728 : * If EC merging is already complete, we have to mop up by adding the new
729 : * EC to the eclass_indexes of the relation(s) mentioned in it.
730 : */
731 210316 : if (root->ec_merging_done)
732 : {
733 117220 : int ec_index = list_length(root->eq_classes) - 1;
734 117220 : int i = -1;
735 :
736 225302 : while ((i = bms_next_member(newec->ec_relids, i)) > 0)
737 : {
738 108082 : RelOptInfo *rel = root->simple_rel_array[i];
739 :
740 108082 : if (rel == NULL) /* must be an outer join */
741 : {
742 : Assert(bms_is_member(i, root->outer_join_rels));
743 5794 : continue;
744 : }
745 :
746 : Assert(rel->reloptkind == RELOPT_BASEREL);
747 :
748 102288 : rel->eclass_indexes = bms_add_member(rel->eclass_indexes,
749 : ec_index);
750 : }
751 : }
752 :
753 210316 : MemoryContextSwitchTo(oldcontext);
754 :
755 210316 : return newec;
756 : }
757 :
758 : /*
759 : * find_ec_member_matching_expr
760 : * Locate an EquivalenceClass member matching the given expr, if any;
761 : * return NULL if no match.
762 : *
763 : * "Matching" is defined as "equal after stripping RelabelTypes".
764 : * This is used for identifying sort expressions, and we need to allow
765 : * binary-compatible relabeling for some cases involving binary-compatible
766 : * sort operators.
767 : *
768 : * Child EC members are ignored unless they belong to given 'relids'.
769 : */
770 : EquivalenceMember *
771 263540 : find_ec_member_matching_expr(EquivalenceClass *ec,
772 : Expr *expr,
773 : Relids relids)
774 : {
775 : ListCell *lc;
776 :
777 : /* We ignore binary-compatible relabeling on both ends */
778 283134 : while (expr && IsA(expr, RelabelType))
779 19594 : expr = ((RelabelType *) expr)->arg;
780 :
781 571988 : foreach(lc, ec->ec_members)
782 : {
783 423928 : EquivalenceMember *em = (EquivalenceMember *) lfirst(lc);
784 : Expr *emexpr;
785 :
786 : /*
787 : * We shouldn't be trying to sort by an equivalence class that
788 : * contains a constant, so no need to consider such cases any further.
789 : */
790 423928 : if (em->em_is_const)
791 0 : continue;
792 :
793 : /*
794 : * Ignore child members unless they belong to the requested rel.
795 : */
796 423928 : if (em->em_is_child &&
797 144024 : !bms_is_subset(em->em_relids, relids))
798 138326 : continue;
799 :
800 : /*
801 : * Match if same expression (after stripping relabel).
802 : */
803 285602 : emexpr = em->em_expr;
804 290428 : while (emexpr && IsA(emexpr, RelabelType))
805 4826 : emexpr = ((RelabelType *) emexpr)->arg;
806 :
807 285602 : if (equal(emexpr, expr))
808 115480 : return em;
809 : }
810 :
811 148060 : return NULL;
812 : }
813 :
814 : /*
815 : * find_computable_ec_member
816 : * Locate an EquivalenceClass member that can be computed from the
817 : * expressions appearing in "exprs"; return NULL if no match.
818 : *
819 : * "exprs" can be either a list of bare expression trees, or a list of
820 : * TargetEntry nodes. Either way, it should contain Vars and possibly
821 : * Aggrefs and WindowFuncs, which are matched to the corresponding elements
822 : * of the EquivalenceClass's expressions.
823 : *
824 : * Unlike find_ec_member_matching_expr, there's no special provision here
825 : * for binary-compatible relabeling. This is intentional: if we have to
826 : * compute an expression in this way, setrefs.c is going to insist on exact
827 : * matches of Vars to the source tlist.
828 : *
829 : * Child EC members are ignored unless they belong to given 'relids'.
830 : * Also, non-parallel-safe expressions are ignored if 'require_parallel_safe'.
831 : *
832 : * Note: some callers pass root == NULL for notational reasons. This is OK
833 : * when require_parallel_safe is false.
834 : */
835 : EquivalenceMember *
836 2396 : find_computable_ec_member(PlannerInfo *root,
837 : EquivalenceClass *ec,
838 : List *exprs,
839 : Relids relids,
840 : bool require_parallel_safe)
841 : {
842 : ListCell *lc;
843 :
844 7510 : foreach(lc, ec->ec_members)
845 : {
846 5552 : EquivalenceMember *em = (EquivalenceMember *) lfirst(lc);
847 : List *exprvars;
848 : ListCell *lc2;
849 :
850 : /*
851 : * We shouldn't be trying to sort by an equivalence class that
852 : * contains a constant, so no need to consider such cases any further.
853 : */
854 5552 : if (em->em_is_const)
855 0 : continue;
856 :
857 : /*
858 : * Ignore child members unless they belong to the requested rel.
859 : */
860 5552 : if (em->em_is_child &&
861 2934 : !bms_is_subset(em->em_relids, relids))
862 2766 : continue;
863 :
864 : /*
865 : * Match if all Vars and quasi-Vars are available in "exprs".
866 : */
867 2786 : exprvars = pull_var_clause((Node *) em->em_expr,
868 : PVC_INCLUDE_AGGREGATES |
869 : PVC_INCLUDE_WINDOWFUNCS |
870 : PVC_INCLUDE_PLACEHOLDERS);
871 3376 : foreach(lc2, exprvars)
872 : {
873 2914 : if (!is_exprlist_member(lfirst(lc2), exprs))
874 2324 : break;
875 : }
876 2786 : list_free(exprvars);
877 2786 : if (lc2)
878 2324 : continue; /* we hit a non-available Var */
879 :
880 : /*
881 : * If requested, reject expressions that are not parallel-safe. We
882 : * check this last because it's a rather expensive test.
883 : */
884 462 : if (require_parallel_safe &&
885 122 : !is_parallel_safe(root, (Node *) em->em_expr))
886 24 : continue;
887 :
888 438 : return em; /* found usable expression */
889 : }
890 :
891 1958 : return NULL;
892 : }
893 :
894 : /*
895 : * is_exprlist_member
896 : * Subroutine for find_computable_ec_member: is "node" in "exprs"?
897 : *
898 : * Per the requirements of that function, "exprs" might or might not have
899 : * TargetEntry superstructure.
900 : */
901 : static bool
902 2914 : is_exprlist_member(Expr *node, List *exprs)
903 : {
904 : ListCell *lc;
905 :
906 8124 : foreach(lc, exprs)
907 : {
908 5800 : Expr *expr = (Expr *) lfirst(lc);
909 :
910 5800 : if (expr && IsA(expr, TargetEntry))
911 1172 : expr = ((TargetEntry *) expr)->expr;
912 :
913 5800 : if (equal(node, expr))
914 590 : return true;
915 : }
916 2324 : return false;
917 : }
918 :
919 : /*
920 : * relation_can_be_sorted_early
921 : * Can this relation be sorted on this EC before the final output step?
922 : *
923 : * To succeed, we must find an EC member that prepare_sort_from_pathkeys knows
924 : * how to sort on, given the rel's reltarget as input. There are also a few
925 : * additional constraints based on the fact that the desired sort will be done
926 : * "early", within the scan/join part of the plan. Also, non-parallel-safe
927 : * expressions are ignored if 'require_parallel_safe'.
928 : *
929 : * At some point we might want to return the identified EquivalenceMember,
930 : * but for now, callers only want to know if there is one.
931 : */
932 : bool
933 9994 : relation_can_be_sorted_early(PlannerInfo *root, RelOptInfo *rel,
934 : EquivalenceClass *ec, bool require_parallel_safe)
935 : {
936 9994 : PathTarget *target = rel->reltarget;
937 : EquivalenceMember *em;
938 : ListCell *lc;
939 :
940 : /*
941 : * Reject volatile ECs immediately; such sorts must always be postponed.
942 : */
943 9994 : if (ec->ec_has_volatile)
944 72 : return false;
945 :
946 : /*
947 : * Try to find an EM directly matching some reltarget member.
948 : */
949 20286 : foreach(lc, target->exprs)
950 : {
951 18230 : Expr *targetexpr = (Expr *) lfirst(lc);
952 :
953 18230 : em = find_ec_member_matching_expr(ec, targetexpr, rel->relids);
954 18230 : if (!em)
955 10364 : continue;
956 :
957 : /*
958 : * Reject expressions involving set-returning functions, as those
959 : * can't be computed early either. (Note: this test and the following
960 : * one are effectively checking properties of targetexpr, so there's
961 : * no point in asking whether some other EC member would be better.)
962 : */
963 7866 : if (expression_returns_set((Node *) em->em_expr))
964 0 : continue;
965 :
966 : /*
967 : * If requested, reject expressions that are not parallel-safe. We
968 : * check this last because it's a rather expensive test.
969 : */
970 7866 : if (require_parallel_safe &&
971 7866 : !is_parallel_safe(root, (Node *) em->em_expr))
972 0 : continue;
973 :
974 7866 : return true;
975 : }
976 :
977 : /*
978 : * Try to find an expression computable from the reltarget.
979 : */
980 2056 : em = find_computable_ec_member(root, ec, target->exprs, rel->relids,
981 : require_parallel_safe);
982 2056 : if (!em)
983 1958 : return false;
984 :
985 : /*
986 : * Reject expressions involving set-returning functions, as those can't be
987 : * computed early either. (There's no point in looking for another EC
988 : * member in this case; since SRFs can't appear in WHERE, they cannot
989 : * belong to multi-member ECs.)
990 : */
991 98 : if (expression_returns_set((Node *) em->em_expr))
992 12 : return false;
993 :
994 86 : return true;
995 : }
996 :
997 : /*
998 : * generate_base_implied_equalities
999 : * Generate any restriction clauses that we can deduce from equivalence
1000 : * classes.
1001 : *
1002 : * When an EC contains pseudoconstants, our strategy is to generate
1003 : * "member = const1" clauses where const1 is the first constant member, for
1004 : * every other member (including other constants). If we are able to do this
1005 : * then we don't need any "var = var" comparisons because we've successfully
1006 : * constrained all the vars at their points of creation. If we fail to
1007 : * generate any of these clauses due to lack of cross-type operators, we fall
1008 : * back to the "ec_broken" strategy described below. (XXX if there are
1009 : * multiple constants of different types, it's possible that we might succeed
1010 : * in forming all the required clauses if we started from a different const
1011 : * member; but this seems a sufficiently hokey corner case to not be worth
1012 : * spending lots of cycles on.)
1013 : *
1014 : * For ECs that contain no pseudoconstants, we generate derived clauses
1015 : * "member1 = member2" for each pair of members belonging to the same base
1016 : * relation (actually, if there are more than two for the same base relation,
1017 : * we only need enough clauses to link each to each other). This provides
1018 : * the base case for the recursion: each row emitted by a base relation scan
1019 : * will constrain all computable members of the EC to be equal. As each
1020 : * join path is formed, we'll add additional derived clauses on-the-fly
1021 : * to maintain this invariant (see generate_join_implied_equalities).
1022 : *
1023 : * If the opfamilies used by the EC do not provide complete sets of cross-type
1024 : * equality operators, it is possible that we will fail to generate a clause
1025 : * that must be generated to maintain the invariant. (An example: given
1026 : * "WHERE a.x = b.y AND b.y = a.z", the scheme breaks down if we cannot
1027 : * generate "a.x = a.z" as a restriction clause for A.) In this case we mark
1028 : * the EC "ec_broken" and fall back to regurgitating its original source
1029 : * RestrictInfos at appropriate times. We do not try to retract any derived
1030 : * clauses already generated from the broken EC, so the resulting plan could
1031 : * be poor due to bad selectivity estimates caused by redundant clauses. But
1032 : * the correct solution to that is to fix the opfamilies ...
1033 : *
1034 : * Equality clauses derived by this function are passed off to
1035 : * process_implied_equality (in plan/initsplan.c) to be inserted into the
1036 : * restrictinfo datastructures. Note that this must be called after initial
1037 : * scanning of the quals and before Path construction begins.
1038 : *
1039 : * We make no attempt to avoid generating duplicate RestrictInfos here: we
1040 : * don't search ec_sources or ec_derives for matches. It doesn't really
1041 : * seem worth the trouble to do so.
1042 : */
1043 : void
1044 276694 : generate_base_implied_equalities(PlannerInfo *root)
1045 : {
1046 : int ec_index;
1047 : ListCell *lc;
1048 :
1049 : /*
1050 : * At this point, we're done absorbing knowledge of equivalences in the
1051 : * query, so no further EC merging should happen, and ECs remaining in the
1052 : * eq_classes list can be considered canonical. (But note that it's still
1053 : * possible for new single-member ECs to be added through
1054 : * get_eclass_for_sort_expr().)
1055 : */
1056 276694 : root->ec_merging_done = true;
1057 :
1058 276694 : ec_index = 0;
1059 597226 : foreach(lc, root->eq_classes)
1060 : {
1061 320532 : EquivalenceClass *ec = (EquivalenceClass *) lfirst(lc);
1062 320532 : bool can_generate_joinclause = false;
1063 : int i;
1064 :
1065 : Assert(ec->ec_merged == NULL); /* else shouldn't be in list */
1066 : Assert(!ec->ec_broken); /* not yet anyway... */
1067 :
1068 : /*
1069 : * Generate implied equalities that are restriction clauses.
1070 : * Single-member ECs won't generate any deductions, either here or at
1071 : * the join level.
1072 : */
1073 320532 : if (list_length(ec->ec_members) > 1)
1074 : {
1075 229068 : if (ec->ec_has_const)
1076 178988 : generate_base_implied_equalities_const(root, ec);
1077 : else
1078 50080 : generate_base_implied_equalities_no_const(root, ec);
1079 :
1080 : /* Recover if we failed to generate required derived clauses */
1081 229068 : if (ec->ec_broken)
1082 30 : generate_base_implied_equalities_broken(root, ec);
1083 :
1084 : /* Detect whether this EC might generate join clauses */
1085 229068 : can_generate_joinclause =
1086 229068 : (bms_membership(ec->ec_relids) == BMS_MULTIPLE);
1087 : }
1088 :
1089 : /*
1090 : * Mark the base rels cited in each eclass (which should all exist by
1091 : * now) with the eq_classes indexes of all eclasses mentioning them.
1092 : * This will let us avoid searching in subsequent lookups. While
1093 : * we're at it, we can mark base rels that have pending eclass joins;
1094 : * this is a cheap version of has_relevant_eclass_joinclause().
1095 : */
1096 320532 : i = -1;
1097 712242 : while ((i = bms_next_member(ec->ec_relids, i)) > 0)
1098 : {
1099 391710 : RelOptInfo *rel = root->simple_rel_array[i];
1100 :
1101 391710 : if (rel == NULL) /* must be an outer join */
1102 : {
1103 : Assert(bms_is_member(i, root->outer_join_rels));
1104 5064 : continue;
1105 : }
1106 :
1107 : Assert(rel->reloptkind == RELOPT_BASEREL);
1108 :
1109 386646 : rel->eclass_indexes = bms_add_member(rel->eclass_indexes,
1110 : ec_index);
1111 :
1112 386646 : if (can_generate_joinclause)
1113 132072 : rel->has_eclass_joins = true;
1114 : }
1115 :
1116 320532 : ec_index++;
1117 : }
1118 276694 : }
1119 :
1120 : /*
1121 : * generate_base_implied_equalities when EC contains pseudoconstant(s)
1122 : */
1123 : static void
1124 178988 : generate_base_implied_equalities_const(PlannerInfo *root,
1125 : EquivalenceClass *ec)
1126 : {
1127 178988 : EquivalenceMember *const_em = NULL;
1128 : ListCell *lc;
1129 :
1130 : /*
1131 : * In the trivial case where we just had one "var = const" clause, push
1132 : * the original clause back into the main planner machinery. There is
1133 : * nothing to be gained by doing it differently, and we save the effort to
1134 : * re-build and re-analyze an equality clause that will be exactly
1135 : * equivalent to the old one.
1136 : */
1137 341202 : if (list_length(ec->ec_members) == 2 &&
1138 162214 : list_length(ec->ec_sources) == 1)
1139 : {
1140 162214 : RestrictInfo *restrictinfo = (RestrictInfo *) linitial(ec->ec_sources);
1141 :
1142 162214 : distribute_restrictinfo_to_rels(root, restrictinfo);
1143 162214 : return;
1144 : }
1145 :
1146 : /*
1147 : * Find the constant member to use. We prefer an actual constant to
1148 : * pseudo-constants (such as Params), because the constraint exclusion
1149 : * machinery might be able to exclude relations on the basis of generated
1150 : * "var = const" equalities, but "var = param" won't work for that.
1151 : */
1152 37624 : foreach(lc, ec->ec_members)
1153 : {
1154 37574 : EquivalenceMember *cur_em = (EquivalenceMember *) lfirst(lc);
1155 :
1156 37574 : if (cur_em->em_is_const)
1157 : {
1158 16780 : const_em = cur_em;
1159 16780 : if (IsA(cur_em->em_expr, Const))
1160 16724 : break;
1161 : }
1162 : }
1163 : Assert(const_em != NULL);
1164 :
1165 : /* Generate a derived equality against each other member */
1166 67216 : foreach(lc, ec->ec_members)
1167 : {
1168 50472 : EquivalenceMember *cur_em = (EquivalenceMember *) lfirst(lc);
1169 : Oid eq_op;
1170 : RestrictInfo *rinfo;
1171 :
1172 : Assert(!cur_em->em_is_child); /* no children yet */
1173 50472 : if (cur_em == const_em)
1174 16750 : continue;
1175 33722 : eq_op = select_equality_operator(ec,
1176 : cur_em->em_datatype,
1177 : const_em->em_datatype);
1178 33722 : if (!OidIsValid(eq_op))
1179 : {
1180 : /* failed... */
1181 30 : ec->ec_broken = true;
1182 30 : break;
1183 : }
1184 :
1185 : /*
1186 : * We use the constant's em_jdomain as qualscope, so that if the
1187 : * generated clause is variable-free (i.e, both EMs are consts) it
1188 : * will be enforced at the join domain level.
1189 : */
1190 33692 : rinfo = process_implied_equality(root, eq_op, ec->ec_collation,
1191 : cur_em->em_expr, const_em->em_expr,
1192 33692 : const_em->em_jdomain->jd_relids,
1193 : ec->ec_min_security,
1194 33692 : cur_em->em_is_const);
1195 :
1196 : /*
1197 : * If the clause didn't degenerate to a constant, fill in the correct
1198 : * markings for a mergejoinable clause, and save it in ec_derives. (We
1199 : * will not re-use such clauses directly, but selectivity estimation
1200 : * may consult the list later. Note that this use of ec_derives does
1201 : * not overlap with its use for join clauses, since we never generate
1202 : * join clauses from an ec_has_const eclass.)
1203 : */
1204 33692 : if (rinfo && rinfo->mergeopfamilies)
1205 : {
1206 : /* it's not redundant, so don't set parent_ec */
1207 33560 : rinfo->left_ec = rinfo->right_ec = ec;
1208 33560 : rinfo->left_em = cur_em;
1209 33560 : rinfo->right_em = const_em;
1210 33560 : ec->ec_derives = lappend(ec->ec_derives, rinfo);
1211 : }
1212 : }
1213 : }
1214 :
1215 : /*
1216 : * generate_base_implied_equalities when EC contains no pseudoconstants
1217 : */
1218 : static void
1219 50080 : generate_base_implied_equalities_no_const(PlannerInfo *root,
1220 : EquivalenceClass *ec)
1221 : {
1222 : EquivalenceMember **prev_ems;
1223 : ListCell *lc;
1224 :
1225 : /*
1226 : * We scan the EC members once and track the last-seen member for each
1227 : * base relation. When we see another member of the same base relation,
1228 : * we generate "prev_em = cur_em". This results in the minimum number of
1229 : * derived clauses, but it's possible that it will fail when a different
1230 : * ordering would succeed. XXX FIXME: use a UNION-FIND algorithm similar
1231 : * to the way we build merged ECs. (Use a list-of-lists for each rel.)
1232 : */
1233 : prev_ems = (EquivalenceMember **)
1234 50080 : palloc0(root->simple_rel_array_size * sizeof(EquivalenceMember *));
1235 :
1236 151708 : foreach(lc, ec->ec_members)
1237 : {
1238 101628 : EquivalenceMember *cur_em = (EquivalenceMember *) lfirst(lc);
1239 : int relid;
1240 :
1241 : Assert(!cur_em->em_is_child); /* no children yet */
1242 101628 : if (!bms_get_singleton_member(cur_em->em_relids, &relid))
1243 186 : continue;
1244 : Assert(relid < root->simple_rel_array_size);
1245 :
1246 101442 : if (prev_ems[relid] != NULL)
1247 : {
1248 310 : EquivalenceMember *prev_em = prev_ems[relid];
1249 : Oid eq_op;
1250 : RestrictInfo *rinfo;
1251 :
1252 310 : eq_op = select_equality_operator(ec,
1253 : prev_em->em_datatype,
1254 : cur_em->em_datatype);
1255 310 : if (!OidIsValid(eq_op))
1256 : {
1257 : /* failed... */
1258 0 : ec->ec_broken = true;
1259 0 : break;
1260 : }
1261 :
1262 : /*
1263 : * The expressions aren't constants, so the passed qualscope will
1264 : * never be used to place the generated clause. We just need to
1265 : * be sure it covers both expressions, which em_relids should do.
1266 : */
1267 310 : rinfo = process_implied_equality(root, eq_op, ec->ec_collation,
1268 : prev_em->em_expr, cur_em->em_expr,
1269 : cur_em->em_relids,
1270 : ec->ec_min_security,
1271 : false);
1272 :
1273 : /*
1274 : * If the clause didn't degenerate to a constant, fill in the
1275 : * correct markings for a mergejoinable clause. We don't put it
1276 : * in ec_derives however; we don't currently need to re-find such
1277 : * clauses, and we don't want to clutter that list with non-join
1278 : * clauses.
1279 : */
1280 310 : if (rinfo && rinfo->mergeopfamilies)
1281 : {
1282 : /* it's not redundant, so don't set parent_ec */
1283 310 : rinfo->left_ec = rinfo->right_ec = ec;
1284 310 : rinfo->left_em = prev_em;
1285 310 : rinfo->right_em = cur_em;
1286 : }
1287 : }
1288 101442 : prev_ems[relid] = cur_em;
1289 : }
1290 :
1291 50080 : pfree(prev_ems);
1292 :
1293 : /*
1294 : * We also have to make sure that all the Vars used in the member clauses
1295 : * will be available at any join node we might try to reference them at.
1296 : * For the moment we force all the Vars to be available at all join nodes
1297 : * for this eclass. Perhaps this could be improved by doing some
1298 : * pre-analysis of which members we prefer to join, but it's no worse than
1299 : * what happened in the pre-8.3 code.
1300 : */
1301 151708 : foreach(lc, ec->ec_members)
1302 : {
1303 101628 : EquivalenceMember *cur_em = (EquivalenceMember *) lfirst(lc);
1304 101628 : List *vars = pull_var_clause((Node *) cur_em->em_expr,
1305 : PVC_RECURSE_AGGREGATES |
1306 : PVC_RECURSE_WINDOWFUNCS |
1307 : PVC_INCLUDE_PLACEHOLDERS);
1308 :
1309 101628 : add_vars_to_targetlist(root, vars, ec->ec_relids);
1310 101628 : list_free(vars);
1311 : }
1312 50080 : }
1313 :
1314 : /*
1315 : * generate_base_implied_equalities cleanup after failure
1316 : *
1317 : * What we must do here is push any zero- or one-relation source RestrictInfos
1318 : * of the EC back into the main restrictinfo datastructures. Multi-relation
1319 : * clauses will be regurgitated later by generate_join_implied_equalities().
1320 : * (We do it this way to maintain continuity with the case that ec_broken
1321 : * becomes set only after we've gone up a join level or two.) However, for
1322 : * an EC that contains constants, we can adopt a simpler strategy and just
1323 : * throw back all the source RestrictInfos immediately; that works because
1324 : * we know that such an EC can't become broken later. (This rule justifies
1325 : * ignoring ec_has_const ECs in generate_join_implied_equalities, even when
1326 : * they are broken.)
1327 : */
1328 : static void
1329 30 : generate_base_implied_equalities_broken(PlannerInfo *root,
1330 : EquivalenceClass *ec)
1331 : {
1332 : ListCell *lc;
1333 :
1334 96 : foreach(lc, ec->ec_sources)
1335 : {
1336 66 : RestrictInfo *restrictinfo = (RestrictInfo *) lfirst(lc);
1337 :
1338 66 : if (ec->ec_has_const ||
1339 0 : bms_membership(restrictinfo->required_relids) != BMS_MULTIPLE)
1340 66 : distribute_restrictinfo_to_rels(root, restrictinfo);
1341 : }
1342 30 : }
1343 :
1344 :
1345 : /*
1346 : * generate_join_implied_equalities
1347 : * Generate any join clauses that we can deduce from equivalence classes.
1348 : *
1349 : * At a join node, we must enforce restriction clauses sufficient to ensure
1350 : * that all equivalence-class members computable at that node are equal.
1351 : * Since the set of clauses to enforce can vary depending on which subset
1352 : * relations are the inputs, we have to compute this afresh for each join
1353 : * relation pair. Hence a fresh List of RestrictInfo nodes is built and
1354 : * passed back on each call.
1355 : *
1356 : * In addition to its use at join nodes, this can be applied to generate
1357 : * eclass-based join clauses for use in a parameterized scan of a base rel.
1358 : * The reason for the asymmetry of specifying the inner rel as a RelOptInfo
1359 : * and the outer rel by Relids is that this usage occurs before we have
1360 : * built any join RelOptInfos.
1361 : *
1362 : * An annoying special case for parameterized scans is that the inner rel can
1363 : * be an appendrel child (an "other rel"). In this case we must generate
1364 : * appropriate clauses using child EC members. add_child_rel_equivalences
1365 : * must already have been done for the child rel.
1366 : *
1367 : * The results are sufficient for use in merge, hash, and plain nestloop join
1368 : * methods. We do not worry here about selecting clauses that are optimal
1369 : * for use in a parameterized indexscan. indxpath.c makes its own selections
1370 : * of clauses to use, and if the ones we pick here are redundant with those,
1371 : * the extras will be eliminated at createplan time, using the parent_ec
1372 : * markers that we provide (see is_redundant_derived_clause()).
1373 : *
1374 : * Because the same join clauses are likely to be needed multiple times as
1375 : * we consider different join paths, we avoid generating multiple copies:
1376 : * whenever we select a particular pair of EquivalenceMembers to join,
1377 : * we check to see if the pair matches any original clause (in ec_sources)
1378 : * or previously-built clause (in ec_derives). This saves memory and allows
1379 : * re-use of information cached in RestrictInfos. We also avoid generating
1380 : * commutative duplicates, i.e. if the algorithm selects "a.x = b.y" but
1381 : * we already have "b.y = a.x", we return the existing clause.
1382 : *
1383 : * If we are considering an outer join, sjinfo is the associated OJ info,
1384 : * otherwise it can be NULL.
1385 : *
1386 : * join_relids should always equal bms_union(outer_relids, inner_rel->relids)
1387 : * plus whatever add_outer_joins_to_relids() would add. We could simplify
1388 : * this function's API by computing it internally, but most callers have the
1389 : * value at hand anyway.
1390 : */
1391 : List *
1392 432066 : generate_join_implied_equalities(PlannerInfo *root,
1393 : Relids join_relids,
1394 : Relids outer_relids,
1395 : RelOptInfo *inner_rel,
1396 : SpecialJoinInfo *sjinfo)
1397 : {
1398 432066 : List *result = NIL;
1399 432066 : Relids inner_relids = inner_rel->relids;
1400 : Relids nominal_inner_relids;
1401 : Relids nominal_join_relids;
1402 : Bitmapset *matching_ecs;
1403 : int i;
1404 :
1405 : /* If inner rel is a child, extra setup work is needed */
1406 432066 : if (IS_OTHER_REL(inner_rel))
1407 : {
1408 : Assert(!bms_is_empty(inner_rel->top_parent_relids));
1409 :
1410 : /* Fetch relid set for the topmost parent rel */
1411 6832 : nominal_inner_relids = inner_rel->top_parent_relids;
1412 : /* ECs will be marked with the parent's relid, not the child's */
1413 6832 : nominal_join_relids = bms_union(outer_relids, nominal_inner_relids);
1414 6832 : nominal_join_relids = add_outer_joins_to_relids(root,
1415 : nominal_join_relids,
1416 : sjinfo,
1417 : NULL);
1418 : }
1419 : else
1420 : {
1421 425234 : nominal_inner_relids = inner_relids;
1422 425234 : nominal_join_relids = join_relids;
1423 : }
1424 :
1425 : /*
1426 : * Examine all potentially-relevant eclasses.
1427 : *
1428 : * If we are considering an outer join, we must include "join" clauses
1429 : * that mention either input rel plus the outer join's relid; these
1430 : * represent post-join filter clauses that have to be applied at this
1431 : * join. We don't have infrastructure that would let us identify such
1432 : * eclasses cheaply, so just fall back to considering all eclasses
1433 : * mentioning anything in nominal_join_relids.
1434 : *
1435 : * At inner joins, we can be smarter: only consider eclasses mentioning
1436 : * both input rels.
1437 : */
1438 432066 : if (sjinfo && sjinfo->ojrelid != 0)
1439 92962 : matching_ecs = get_eclass_indexes_for_relids(root, nominal_join_relids);
1440 : else
1441 339104 : matching_ecs = get_common_eclass_indexes(root, nominal_inner_relids,
1442 : outer_relids);
1443 :
1444 432066 : i = -1;
1445 1263610 : while ((i = bms_next_member(matching_ecs, i)) >= 0)
1446 : {
1447 831544 : EquivalenceClass *ec = (EquivalenceClass *) list_nth(root->eq_classes, i);
1448 831544 : List *sublist = NIL;
1449 :
1450 : /* ECs containing consts do not need any further enforcement */
1451 831544 : if (ec->ec_has_const)
1452 116932 : continue;
1453 :
1454 : /* Single-member ECs won't generate any deductions */
1455 714612 : if (list_length(ec->ec_members) <= 1)
1456 428046 : continue;
1457 :
1458 : /* Sanity check that this eclass overlaps the join */
1459 : Assert(bms_overlap(ec->ec_relids, nominal_join_relids));
1460 :
1461 286566 : if (!ec->ec_broken)
1462 286242 : sublist = generate_join_implied_equalities_normal(root,
1463 : ec,
1464 : join_relids,
1465 : outer_relids,
1466 : inner_relids);
1467 :
1468 : /* Recover if we failed to generate required derived clauses */
1469 286566 : if (ec->ec_broken)
1470 360 : sublist = generate_join_implied_equalities_broken(root,
1471 : ec,
1472 : nominal_join_relids,
1473 : outer_relids,
1474 : nominal_inner_relids,
1475 : inner_rel);
1476 :
1477 286566 : result = list_concat(result, sublist);
1478 : }
1479 :
1480 432066 : return result;
1481 : }
1482 :
1483 : /*
1484 : * generate_join_implied_equalities_for_ecs
1485 : * As above, but consider only the listed ECs.
1486 : *
1487 : * For the sole current caller, we can assume sjinfo == NULL, that is we are
1488 : * not interested in outer-join filter clauses. This might need to change
1489 : * in future.
1490 : */
1491 : List *
1492 890 : generate_join_implied_equalities_for_ecs(PlannerInfo *root,
1493 : List *eclasses,
1494 : Relids join_relids,
1495 : Relids outer_relids,
1496 : RelOptInfo *inner_rel)
1497 : {
1498 890 : List *result = NIL;
1499 890 : Relids inner_relids = inner_rel->relids;
1500 : Relids nominal_inner_relids;
1501 : Relids nominal_join_relids;
1502 : ListCell *lc;
1503 :
1504 : /* If inner rel is a child, extra setup work is needed */
1505 890 : if (IS_OTHER_REL(inner_rel))
1506 : {
1507 : Assert(!bms_is_empty(inner_rel->top_parent_relids));
1508 :
1509 : /* Fetch relid set for the topmost parent rel */
1510 0 : nominal_inner_relids = inner_rel->top_parent_relids;
1511 : /* ECs will be marked with the parent's relid, not the child's */
1512 0 : nominal_join_relids = bms_union(outer_relids, nominal_inner_relids);
1513 : }
1514 : else
1515 : {
1516 890 : nominal_inner_relids = inner_relids;
1517 890 : nominal_join_relids = join_relids;
1518 : }
1519 :
1520 1828 : foreach(lc, eclasses)
1521 : {
1522 938 : EquivalenceClass *ec = (EquivalenceClass *) lfirst(lc);
1523 938 : List *sublist = NIL;
1524 :
1525 : /* ECs containing consts do not need any further enforcement */
1526 938 : if (ec->ec_has_const)
1527 0 : continue;
1528 :
1529 : /* Single-member ECs won't generate any deductions */
1530 938 : if (list_length(ec->ec_members) <= 1)
1531 0 : continue;
1532 :
1533 : /* We can quickly ignore any that don't overlap the join, too */
1534 938 : if (!bms_overlap(ec->ec_relids, nominal_join_relids))
1535 0 : continue;
1536 :
1537 938 : if (!ec->ec_broken)
1538 938 : sublist = generate_join_implied_equalities_normal(root,
1539 : ec,
1540 : join_relids,
1541 : outer_relids,
1542 : inner_relids);
1543 :
1544 : /* Recover if we failed to generate required derived clauses */
1545 938 : if (ec->ec_broken)
1546 0 : sublist = generate_join_implied_equalities_broken(root,
1547 : ec,
1548 : nominal_join_relids,
1549 : outer_relids,
1550 : nominal_inner_relids,
1551 : inner_rel);
1552 :
1553 938 : result = list_concat(result, sublist);
1554 : }
1555 :
1556 890 : return result;
1557 : }
1558 :
1559 : /*
1560 : * generate_join_implied_equalities for a still-valid EC
1561 : */
1562 : static List *
1563 287180 : generate_join_implied_equalities_normal(PlannerInfo *root,
1564 : EquivalenceClass *ec,
1565 : Relids join_relids,
1566 : Relids outer_relids,
1567 : Relids inner_relids)
1568 : {
1569 287180 : List *result = NIL;
1570 287180 : List *new_members = NIL;
1571 287180 : List *outer_members = NIL;
1572 287180 : List *inner_members = NIL;
1573 : ListCell *lc1;
1574 :
1575 : /*
1576 : * First, scan the EC to identify member values that are computable at the
1577 : * outer rel, at the inner rel, or at this relation but not in either
1578 : * input rel. The outer-rel members should already be enforced equal,
1579 : * likewise for the inner-rel members. We'll need to create clauses to
1580 : * enforce that any newly computable members are all equal to each other
1581 : * as well as to at least one input member, plus enforce at least one
1582 : * outer-rel member equal to at least one inner-rel member.
1583 : */
1584 988332 : foreach(lc1, ec->ec_members)
1585 : {
1586 701152 : EquivalenceMember *cur_em = (EquivalenceMember *) lfirst(lc1);
1587 :
1588 : /*
1589 : * We don't need to check explicitly for child EC members. This test
1590 : * against join_relids will cause them to be ignored except when
1591 : * considering a child inner rel, which is what we want.
1592 : */
1593 701152 : if (!bms_is_subset(cur_em->em_relids, join_relids))
1594 126374 : continue; /* not computable yet, or wrong child */
1595 :
1596 574778 : if (bms_is_subset(cur_em->em_relids, outer_relids))
1597 331602 : outer_members = lappend(outer_members, cur_em);
1598 243176 : else if (bms_is_subset(cur_em->em_relids, inner_relids))
1599 241580 : inner_members = lappend(inner_members, cur_em);
1600 : else
1601 1596 : new_members = lappend(new_members, cur_em);
1602 : }
1603 :
1604 : /*
1605 : * First, select the joinclause if needed. We can equate any one outer
1606 : * member to any one inner member, but we have to find a datatype
1607 : * combination for which an opfamily member operator exists. If we have
1608 : * choices, we prefer simple Var members (possibly with RelabelType) since
1609 : * these are (a) cheapest to compute at runtime and (b) most likely to
1610 : * have useful statistics. Also, prefer operators that are also
1611 : * hashjoinable.
1612 : */
1613 287180 : if (outer_members && inner_members)
1614 : {
1615 229126 : EquivalenceMember *best_outer_em = NULL;
1616 229126 : EquivalenceMember *best_inner_em = NULL;
1617 229126 : Oid best_eq_op = InvalidOid;
1618 229126 : int best_score = -1;
1619 : RestrictInfo *rinfo;
1620 :
1621 242842 : foreach(lc1, outer_members)
1622 : {
1623 229198 : EquivalenceMember *outer_em = (EquivalenceMember *) lfirst(lc1);
1624 : ListCell *lc2;
1625 :
1626 242926 : foreach(lc2, inner_members)
1627 : {
1628 229210 : EquivalenceMember *inner_em = (EquivalenceMember *) lfirst(lc2);
1629 : Oid eq_op;
1630 : int score;
1631 :
1632 229210 : eq_op = select_equality_operator(ec,
1633 : outer_em->em_datatype,
1634 : inner_em->em_datatype);
1635 229210 : if (!OidIsValid(eq_op))
1636 36 : continue;
1637 229174 : score = 0;
1638 229174 : if (IsA(outer_em->em_expr, Var) ||
1639 14874 : (IsA(outer_em->em_expr, RelabelType) &&
1640 3346 : IsA(((RelabelType *) outer_em->em_expr)->arg, Var)))
1641 217598 : score++;
1642 229174 : if (IsA(inner_em->em_expr, Var) ||
1643 5398 : (IsA(inner_em->em_expr, RelabelType) &&
1644 2904 : IsA(((RelabelType *) inner_em->em_expr)->arg, Var)))
1645 226662 : score++;
1646 229174 : if (op_hashjoinable(eq_op,
1647 229174 : exprType((Node *) outer_em->em_expr)))
1648 229096 : score++;
1649 229174 : if (score > best_score)
1650 : {
1651 229090 : best_outer_em = outer_em;
1652 229090 : best_inner_em = inner_em;
1653 229090 : best_eq_op = eq_op;
1654 229090 : best_score = score;
1655 229090 : if (best_score == 3)
1656 215482 : break; /* no need to look further */
1657 : }
1658 : }
1659 229198 : if (best_score == 3)
1660 215482 : break; /* no need to look further */
1661 : }
1662 229126 : if (best_score < 0)
1663 : {
1664 : /* failed... */
1665 36 : ec->ec_broken = true;
1666 36 : return NIL;
1667 : }
1668 :
1669 : /*
1670 : * Create clause, setting parent_ec to mark it as redundant with other
1671 : * joinclauses
1672 : */
1673 229090 : rinfo = create_join_clause(root, ec, best_eq_op,
1674 : best_outer_em, best_inner_em,
1675 : ec);
1676 :
1677 229090 : result = lappend(result, rinfo);
1678 : }
1679 :
1680 : /*
1681 : * Now deal with building restrictions for any expressions that involve
1682 : * Vars from both sides of the join. We have to equate all of these to
1683 : * each other as well as to at least one old member (if any).
1684 : *
1685 : * XXX as in generate_base_implied_equalities_no_const, we could be a lot
1686 : * smarter here to avoid unnecessary failures in cross-type situations.
1687 : * For now, use the same left-to-right method used there.
1688 : */
1689 287144 : if (new_members)
1690 : {
1691 1560 : List *old_members = list_concat(outer_members, inner_members);
1692 1560 : EquivalenceMember *prev_em = NULL;
1693 : RestrictInfo *rinfo;
1694 :
1695 : /* For now, arbitrarily take the first old_member as the one to use */
1696 1560 : if (old_members)
1697 1098 : new_members = lappend(new_members, linitial(old_members));
1698 :
1699 4254 : foreach(lc1, new_members)
1700 : {
1701 2694 : EquivalenceMember *cur_em = (EquivalenceMember *) lfirst(lc1);
1702 :
1703 2694 : if (prev_em != NULL)
1704 : {
1705 : Oid eq_op;
1706 :
1707 1134 : eq_op = select_equality_operator(ec,
1708 : prev_em->em_datatype,
1709 : cur_em->em_datatype);
1710 1134 : if (!OidIsValid(eq_op))
1711 : {
1712 : /* failed... */
1713 0 : ec->ec_broken = true;
1714 0 : return NIL;
1715 : }
1716 : /* do NOT set parent_ec, this qual is not redundant! */
1717 1134 : rinfo = create_join_clause(root, ec, eq_op,
1718 : prev_em, cur_em,
1719 : NULL);
1720 :
1721 1134 : result = lappend(result, rinfo);
1722 : }
1723 2694 : prev_em = cur_em;
1724 : }
1725 : }
1726 :
1727 287144 : return result;
1728 : }
1729 :
1730 : /*
1731 : * generate_join_implied_equalities cleanup after failure
1732 : *
1733 : * Return any original RestrictInfos that are enforceable at this join.
1734 : *
1735 : * In the case of a child inner relation, we have to translate the
1736 : * original RestrictInfos from parent to child Vars.
1737 : */
1738 : static List *
1739 360 : generate_join_implied_equalities_broken(PlannerInfo *root,
1740 : EquivalenceClass *ec,
1741 : Relids nominal_join_relids,
1742 : Relids outer_relids,
1743 : Relids nominal_inner_relids,
1744 : RelOptInfo *inner_rel)
1745 : {
1746 360 : List *result = NIL;
1747 : ListCell *lc;
1748 :
1749 984 : foreach(lc, ec->ec_sources)
1750 : {
1751 624 : RestrictInfo *restrictinfo = (RestrictInfo *) lfirst(lc);
1752 624 : Relids clause_relids = restrictinfo->required_relids;
1753 :
1754 624 : if (bms_is_subset(clause_relids, nominal_join_relids) &&
1755 336 : !bms_is_subset(clause_relids, outer_relids) &&
1756 312 : !bms_is_subset(clause_relids, nominal_inner_relids))
1757 312 : result = lappend(result, restrictinfo);
1758 : }
1759 :
1760 : /*
1761 : * If we have to translate, just brute-force apply adjust_appendrel_attrs
1762 : * to all the RestrictInfos at once. This will result in returning
1763 : * RestrictInfos that are not listed in ec_derives, but there shouldn't be
1764 : * any duplication, and it's a sufficiently narrow corner case that we
1765 : * shouldn't sweat too much over it anyway.
1766 : *
1767 : * Since inner_rel might be an indirect descendant of the baserel
1768 : * mentioned in the ec_sources clauses, we have to be prepared to apply
1769 : * multiple levels of Var translation.
1770 : */
1771 360 : if (IS_OTHER_REL(inner_rel) && result != NIL)
1772 162 : result = (List *) adjust_appendrel_attrs_multilevel(root,
1773 : (Node *) result,
1774 : inner_rel,
1775 162 : inner_rel->top_parent);
1776 :
1777 360 : return result;
1778 : }
1779 :
1780 :
1781 : /*
1782 : * select_equality_operator
1783 : * Select a suitable equality operator for comparing two EC members
1784 : *
1785 : * Returns InvalidOid if no operator can be found for this datatype combination
1786 : */
1787 : static Oid
1788 348906 : select_equality_operator(EquivalenceClass *ec, Oid lefttype, Oid righttype)
1789 : {
1790 : ListCell *lc;
1791 :
1792 348972 : foreach(lc, ec->ec_opfamilies)
1793 : {
1794 348906 : Oid opfamily = lfirst_oid(lc);
1795 : Oid opno;
1796 :
1797 348906 : opno = get_opfamily_member(opfamily, lefttype, righttype,
1798 : BTEqualStrategyNumber);
1799 348906 : if (!OidIsValid(opno))
1800 66 : continue;
1801 : /* If no barrier quals in query, don't worry about leaky operators */
1802 348840 : if (ec->ec_max_security == 0)
1803 348840 : return opno;
1804 : /* Otherwise, insist that selected operators be leakproof */
1805 428 : if (get_func_leakproof(get_opcode(opno)))
1806 428 : return opno;
1807 : }
1808 66 : return InvalidOid;
1809 : }
1810 :
1811 :
1812 : /*
1813 : * create_join_clause
1814 : * Find or make a RestrictInfo comparing the two given EC members
1815 : * with the given operator (or, possibly, its commutator, because
1816 : * the ordering of the operands in the result is not guaranteed).
1817 : *
1818 : * parent_ec is either equal to ec (if the clause is a potentially-redundant
1819 : * join clause) or NULL (if not). We have to treat this as part of the
1820 : * match requirements --- it's possible that a clause comparing the same two
1821 : * EMs is a join clause in one join path and a restriction clause in another.
1822 : */
1823 : static RestrictInfo *
1824 316780 : create_join_clause(PlannerInfo *root,
1825 : EquivalenceClass *ec, Oid opno,
1826 : EquivalenceMember *leftem,
1827 : EquivalenceMember *rightem,
1828 : EquivalenceClass *parent_ec)
1829 : {
1830 : RestrictInfo *rinfo;
1831 316780 : RestrictInfo *parent_rinfo = NULL;
1832 : ListCell *lc;
1833 : MemoryContext oldcontext;
1834 :
1835 : /*
1836 : * Search to see if we already built a RestrictInfo for this pair of
1837 : * EquivalenceMembers. We can use either original source clauses or
1838 : * previously-derived clauses, and a commutator clause is acceptable.
1839 : *
1840 : * We used to verify that opno matches, but that seems redundant: even if
1841 : * it's not identical, it'd better have the same effects, or the operator
1842 : * families we're using are broken.
1843 : */
1844 687654 : foreach(lc, ec->ec_sources)
1845 : {
1846 371846 : rinfo = (RestrictInfo *) lfirst(lc);
1847 371846 : if (rinfo->left_em == leftem &&
1848 168980 : rinfo->right_em == rightem &&
1849 150158 : rinfo->parent_ec == parent_ec)
1850 972 : return rinfo;
1851 371744 : if (rinfo->left_em == rightem &&
1852 162282 : rinfo->right_em == leftem &&
1853 146988 : rinfo->parent_ec == parent_ec)
1854 870 : return rinfo;
1855 : }
1856 :
1857 407214 : foreach(lc, ec->ec_derives)
1858 : {
1859 350852 : rinfo = (RestrictInfo *) lfirst(lc);
1860 350852 : if (rinfo->left_em == leftem &&
1861 137290 : rinfo->right_em == rightem &&
1862 122528 : rinfo->parent_ec == parent_ec)
1863 259446 : return rinfo;
1864 228330 : if (rinfo->left_em == rightem &&
1865 145200 : rinfo->right_em == leftem &&
1866 136924 : rinfo->parent_ec == parent_ec)
1867 136924 : return rinfo;
1868 : }
1869 :
1870 : /*
1871 : * Not there, so build it, in planner context so we can re-use it. (Not
1872 : * important in normal planning, but definitely so in GEQO.)
1873 : */
1874 56362 : oldcontext = MemoryContextSwitchTo(root->planner_cxt);
1875 :
1876 : /*
1877 : * If either EM is a child, recursively create the corresponding
1878 : * parent-to-parent clause, so that we can duplicate its rinfo_serial.
1879 : */
1880 56362 : if (leftem->em_is_child || rightem->em_is_child)
1881 : {
1882 3694 : EquivalenceMember *leftp = leftem->em_parent ? leftem->em_parent : leftem;
1883 3694 : EquivalenceMember *rightp = rightem->em_parent ? rightem->em_parent : rightem;
1884 :
1885 3694 : parent_rinfo = create_join_clause(root, ec, opno,
1886 : leftp, rightp,
1887 : parent_ec);
1888 : }
1889 :
1890 56362 : rinfo = build_implied_join_equality(root,
1891 : opno,
1892 : ec->ec_collation,
1893 : leftem->em_expr,
1894 : rightem->em_expr,
1895 56362 : bms_union(leftem->em_relids,
1896 56362 : rightem->em_relids),
1897 : ec->ec_min_security);
1898 :
1899 : /*
1900 : * If either EM is a child, force the clause's clause_relids to include
1901 : * the relid(s) of the child rel. In normal cases it would already, but
1902 : * not if we are considering appendrel child relations with pseudoconstant
1903 : * translated variables (i.e., UNION ALL sub-selects with constant output
1904 : * items). We must do this so that join_clause_is_movable_into() will
1905 : * think that the clause should be evaluated at the correct place.
1906 : */
1907 56362 : if (leftem->em_is_child)
1908 3190 : rinfo->clause_relids = bms_add_members(rinfo->clause_relids,
1909 3190 : leftem->em_relids);
1910 56362 : if (rightem->em_is_child)
1911 504 : rinfo->clause_relids = bms_add_members(rinfo->clause_relids,
1912 504 : rightem->em_relids);
1913 :
1914 : /* If it's a child clause, copy the parent's rinfo_serial */
1915 56362 : if (parent_rinfo)
1916 3694 : rinfo->rinfo_serial = parent_rinfo->rinfo_serial;
1917 :
1918 : /* Mark the clause as redundant, or not */
1919 56362 : rinfo->parent_ec = parent_ec;
1920 :
1921 : /*
1922 : * We know the correct values for left_ec/right_ec, ie this particular EC,
1923 : * so we can just set them directly instead of forcing another lookup.
1924 : */
1925 56362 : rinfo->left_ec = ec;
1926 56362 : rinfo->right_ec = ec;
1927 :
1928 : /* Mark it as usable with these EMs */
1929 56362 : rinfo->left_em = leftem;
1930 56362 : rinfo->right_em = rightem;
1931 : /* and save it for possible re-use */
1932 56362 : ec->ec_derives = lappend(ec->ec_derives, rinfo);
1933 :
1934 56362 : MemoryContextSwitchTo(oldcontext);
1935 :
1936 56362 : return rinfo;
1937 : }
1938 :
1939 :
1940 : /*
1941 : * reconsider_outer_join_clauses
1942 : * Re-examine any outer-join clauses that were set aside by
1943 : * distribute_qual_to_rels(), and see if we can derive any
1944 : * EquivalenceClasses from them. Then, if they were not made
1945 : * redundant, push them out into the regular join-clause lists.
1946 : *
1947 : * When we have mergejoinable clauses A = B that are outer-join clauses,
1948 : * we can't blindly combine them with other clauses A = C to deduce B = C,
1949 : * since in fact the "equality" A = B won't necessarily hold above the
1950 : * outer join (one of the variables might be NULL instead). Nonetheless
1951 : * there are cases where we can add qual clauses using transitivity.
1952 : *
1953 : * One case that we look for here is an outer-join clause OUTERVAR = INNERVAR
1954 : * for which there is also an equivalence clause OUTERVAR = CONSTANT.
1955 : * It is safe and useful to push a clause INNERVAR = CONSTANT into the
1956 : * evaluation of the inner (nullable) relation, because any inner rows not
1957 : * meeting this condition will not contribute to the outer-join result anyway.
1958 : * (Any outer rows they could join to will be eliminated by the pushed-down
1959 : * equivalence clause.)
1960 : *
1961 : * Note that the above rule does not work for full outer joins; nor is it
1962 : * very interesting to consider cases where the generated equivalence clause
1963 : * would involve relations outside the outer join, since such clauses couldn't
1964 : * be pushed into the inner side's scan anyway. So the restriction to
1965 : * outervar = pseudoconstant is not really giving up anything.
1966 : *
1967 : * For full-join cases, we can only do something useful if it's a FULL JOIN
1968 : * USING and a merged column has an equivalence MERGEDVAR = CONSTANT.
1969 : * By the time it gets here, the merged column will look like
1970 : * COALESCE(LEFTVAR, RIGHTVAR)
1971 : * and we will have a full-join clause LEFTVAR = RIGHTVAR that we can match
1972 : * the COALESCE expression to. In this situation we can push LEFTVAR = CONSTANT
1973 : * and RIGHTVAR = CONSTANT into the input relations, since any rows not
1974 : * meeting these conditions cannot contribute to the join result.
1975 : *
1976 : * Again, there isn't any traction to be gained by trying to deal with
1977 : * clauses comparing a mergedvar to a non-pseudoconstant. So we can make
1978 : * use of the EquivalenceClasses to search for matching variables that were
1979 : * equivalenced to constants. The interesting outer-join clauses were
1980 : * accumulated for us by distribute_qual_to_rels.
1981 : *
1982 : * When we find one of these cases, we implement the changes we want by
1983 : * generating a new equivalence clause INNERVAR = CONSTANT (or LEFTVAR, etc)
1984 : * and pushing it into the EquivalenceClass structures. This is because we
1985 : * may already know that INNERVAR is equivalenced to some other var(s), and
1986 : * we'd like the constant to propagate to them too. Note that it would be
1987 : * unsafe to merge any existing EC for INNERVAR with the OUTERVAR's EC ---
1988 : * that could result in propagating constant restrictions from
1989 : * INNERVAR to OUTERVAR, which would be very wrong.
1990 : *
1991 : * It's possible that the INNERVAR is also an OUTERVAR for some other
1992 : * outer-join clause, in which case the process can be repeated. So we repeat
1993 : * looping over the lists of clauses until no further deductions can be made.
1994 : * Whenever we do make a deduction, we remove the generating clause from the
1995 : * lists, since we don't want to make the same deduction twice.
1996 : *
1997 : * If we don't find any match for a set-aside outer join clause, we must
1998 : * throw it back into the regular joinclause processing by passing it to
1999 : * distribute_restrictinfo_to_rels(). If we do generate a derived clause,
2000 : * however, the outer-join clause is redundant. We must still put some
2001 : * clause into the regular processing, because otherwise the join will be
2002 : * seen as a clauseless join and avoided during join order searching.
2003 : * We handle this by generating a constant-TRUE clause that is marked with
2004 : * the same required_relids etc as the removed outer-join clause, thus
2005 : * making it a join clause between the correct relations.
2006 : */
2007 : void
2008 278302 : reconsider_outer_join_clauses(PlannerInfo *root)
2009 : {
2010 : bool found;
2011 : ListCell *cell;
2012 :
2013 : /* Outer loop repeats until we find no more deductions */
2014 : do
2015 : {
2016 278302 : found = false;
2017 :
2018 : /* Process the LEFT JOIN clauses */
2019 310204 : foreach(cell, root->left_join_clauses)
2020 : {
2021 31902 : OuterJoinClauseInfo *ojcinfo = (OuterJoinClauseInfo *) lfirst(cell);
2022 :
2023 31902 : if (reconsider_outer_join_clause(root, ojcinfo, true))
2024 : {
2025 696 : RestrictInfo *rinfo = ojcinfo->rinfo;
2026 :
2027 696 : found = true;
2028 : /* remove it from the list */
2029 696 : root->left_join_clauses =
2030 696 : foreach_delete_current(root->left_join_clauses, cell);
2031 : /* throw back a dummy replacement clause (see notes above) */
2032 696 : rinfo = make_restrictinfo(root,
2033 696 : (Expr *) makeBoolConst(true, false),
2034 696 : rinfo->is_pushed_down,
2035 696 : rinfo->has_clone,
2036 696 : rinfo->is_clone,
2037 : false, /* pseudoconstant */
2038 : 0, /* security_level */
2039 : rinfo->required_relids,
2040 : rinfo->incompatible_relids,
2041 : rinfo->outer_relids);
2042 696 : distribute_restrictinfo_to_rels(root, rinfo);
2043 : }
2044 : }
2045 :
2046 : /* Process the RIGHT JOIN clauses */
2047 298524 : foreach(cell, root->right_join_clauses)
2048 : {
2049 20222 : OuterJoinClauseInfo *ojcinfo = (OuterJoinClauseInfo *) lfirst(cell);
2050 :
2051 20222 : if (reconsider_outer_join_clause(root, ojcinfo, false))
2052 : {
2053 918 : RestrictInfo *rinfo = ojcinfo->rinfo;
2054 :
2055 918 : found = true;
2056 : /* remove it from the list */
2057 918 : root->right_join_clauses =
2058 918 : foreach_delete_current(root->right_join_clauses, cell);
2059 : /* throw back a dummy replacement clause (see notes above) */
2060 918 : rinfo = make_restrictinfo(root,
2061 918 : (Expr *) makeBoolConst(true, false),
2062 918 : rinfo->is_pushed_down,
2063 918 : rinfo->has_clone,
2064 918 : rinfo->is_clone,
2065 : false, /* pseudoconstant */
2066 : 0, /* security_level */
2067 : rinfo->required_relids,
2068 : rinfo->incompatible_relids,
2069 : rinfo->outer_relids);
2070 918 : distribute_restrictinfo_to_rels(root, rinfo);
2071 : }
2072 : }
2073 :
2074 : /* Process the FULL JOIN clauses */
2075 279524 : foreach(cell, root->full_join_clauses)
2076 : {
2077 1222 : OuterJoinClauseInfo *ojcinfo = (OuterJoinClauseInfo *) lfirst(cell);
2078 :
2079 1222 : if (reconsider_full_join_clause(root, ojcinfo))
2080 : {
2081 6 : RestrictInfo *rinfo = ojcinfo->rinfo;
2082 :
2083 6 : found = true;
2084 : /* remove it from the list */
2085 6 : root->full_join_clauses =
2086 6 : foreach_delete_current(root->full_join_clauses, cell);
2087 : /* throw back a dummy replacement clause (see notes above) */
2088 6 : rinfo = make_restrictinfo(root,
2089 6 : (Expr *) makeBoolConst(true, false),
2090 6 : rinfo->is_pushed_down,
2091 6 : rinfo->has_clone,
2092 6 : rinfo->is_clone,
2093 : false, /* pseudoconstant */
2094 : 0, /* security_level */
2095 : rinfo->required_relids,
2096 : rinfo->incompatible_relids,
2097 : rinfo->outer_relids);
2098 6 : distribute_restrictinfo_to_rels(root, rinfo);
2099 : }
2100 : }
2101 278302 : } while (found);
2102 :
2103 : /* Now, any remaining clauses have to be thrown back */
2104 307378 : foreach(cell, root->left_join_clauses)
2105 : {
2106 30684 : OuterJoinClauseInfo *ojcinfo = (OuterJoinClauseInfo *) lfirst(cell);
2107 :
2108 30684 : distribute_restrictinfo_to_rels(root, ojcinfo->rinfo);
2109 : }
2110 295032 : foreach(cell, root->right_join_clauses)
2111 : {
2112 18338 : OuterJoinClauseInfo *ojcinfo = (OuterJoinClauseInfo *) lfirst(cell);
2113 :
2114 18338 : distribute_restrictinfo_to_rels(root, ojcinfo->rinfo);
2115 : }
2116 277910 : foreach(cell, root->full_join_clauses)
2117 : {
2118 1216 : OuterJoinClauseInfo *ojcinfo = (OuterJoinClauseInfo *) lfirst(cell);
2119 :
2120 1216 : distribute_restrictinfo_to_rels(root, ojcinfo->rinfo);
2121 : }
2122 276694 : }
2123 :
2124 : /*
2125 : * reconsider_outer_join_clauses for a single LEFT/RIGHT JOIN clause
2126 : *
2127 : * Returns true if we were able to propagate a constant through the clause.
2128 : */
2129 : static bool
2130 52124 : reconsider_outer_join_clause(PlannerInfo *root, OuterJoinClauseInfo *ojcinfo,
2131 : bool outer_on_left)
2132 : {
2133 52124 : RestrictInfo *rinfo = ojcinfo->rinfo;
2134 52124 : SpecialJoinInfo *sjinfo = ojcinfo->sjinfo;
2135 : Expr *outervar,
2136 : *innervar;
2137 : Oid opno,
2138 : collation,
2139 : left_type,
2140 : right_type,
2141 : inner_datatype;
2142 : Relids inner_relids;
2143 : ListCell *lc1;
2144 :
2145 : Assert(is_opclause(rinfo->clause));
2146 52124 : opno = ((OpExpr *) rinfo->clause)->opno;
2147 52124 : collation = ((OpExpr *) rinfo->clause)->inputcollid;
2148 :
2149 : /* Extract needed info from the clause */
2150 52124 : op_input_types(opno, &left_type, &right_type);
2151 52124 : if (outer_on_left)
2152 : {
2153 31902 : outervar = (Expr *) get_leftop(rinfo->clause);
2154 31902 : innervar = (Expr *) get_rightop(rinfo->clause);
2155 31902 : inner_datatype = right_type;
2156 31902 : inner_relids = rinfo->right_relids;
2157 : }
2158 : else
2159 : {
2160 20222 : outervar = (Expr *) get_rightop(rinfo->clause);
2161 20222 : innervar = (Expr *) get_leftop(rinfo->clause);
2162 20222 : inner_datatype = left_type;
2163 20222 : inner_relids = rinfo->left_relids;
2164 : }
2165 :
2166 : /* Scan EquivalenceClasses for a match to outervar */
2167 333112 : foreach(lc1, root->eq_classes)
2168 : {
2169 282602 : EquivalenceClass *cur_ec = (EquivalenceClass *) lfirst(lc1);
2170 : bool match;
2171 : ListCell *lc2;
2172 :
2173 : /* Ignore EC unless it contains pseudoconstants */
2174 282602 : if (!cur_ec->ec_has_const)
2175 222304 : continue;
2176 : /* Never match to a volatile EC */
2177 60298 : if (cur_ec->ec_has_volatile)
2178 0 : continue;
2179 : /* It has to match the outer-join clause as to semantics, too */
2180 60298 : if (collation != cur_ec->ec_collation)
2181 2202 : continue;
2182 58096 : if (!equal(rinfo->mergeopfamilies, cur_ec->ec_opfamilies))
2183 13576 : continue;
2184 : /* Does it contain a match to outervar? */
2185 44520 : match = false;
2186 137206 : foreach(lc2, cur_ec->ec_members)
2187 : {
2188 94300 : EquivalenceMember *cur_em = (EquivalenceMember *) lfirst(lc2);
2189 :
2190 : Assert(!cur_em->em_is_child); /* no children yet */
2191 94300 : if (equal(outervar, cur_em->em_expr))
2192 : {
2193 1614 : match = true;
2194 1614 : break;
2195 : }
2196 : }
2197 44520 : if (!match)
2198 42906 : continue; /* no match, so ignore this EC */
2199 :
2200 : /*
2201 : * Yes it does! Try to generate a clause INNERVAR = CONSTANT for each
2202 : * CONSTANT in the EC. Note that we must succeed with at least one
2203 : * constant before we can decide to throw away the outer-join clause.
2204 : */
2205 1614 : match = false;
2206 5808 : foreach(lc2, cur_ec->ec_members)
2207 : {
2208 4194 : EquivalenceMember *cur_em = (EquivalenceMember *) lfirst(lc2);
2209 : Oid eq_op;
2210 : RestrictInfo *newrinfo;
2211 : JoinDomain *jdomain;
2212 :
2213 4194 : if (!cur_em->em_is_const)
2214 2538 : continue; /* ignore non-const members */
2215 1656 : eq_op = select_equality_operator(cur_ec,
2216 : inner_datatype,
2217 : cur_em->em_datatype);
2218 1656 : if (!OidIsValid(eq_op))
2219 0 : continue; /* can't generate equality */
2220 1656 : newrinfo = build_implied_join_equality(root,
2221 : eq_op,
2222 : cur_ec->ec_collation,
2223 : innervar,
2224 : cur_em->em_expr,
2225 : bms_copy(inner_relids),
2226 : cur_ec->ec_min_security);
2227 : /* This equality holds within the OJ's child JoinDomain */
2228 1656 : jdomain = find_join_domain(root, sjinfo->syn_righthand);
2229 1656 : if (process_equivalence(root, &newrinfo, jdomain))
2230 1656 : match = true;
2231 : }
2232 :
2233 : /*
2234 : * If we were able to equate INNERVAR to any constant, report success.
2235 : * Otherwise, fall out of the search loop, since we know the OUTERVAR
2236 : * appears in at most one EC.
2237 : */
2238 1614 : if (match)
2239 1614 : return true;
2240 : else
2241 0 : break;
2242 : }
2243 :
2244 50510 : return false; /* failed to make any deduction */
2245 : }
2246 :
2247 : /*
2248 : * reconsider_outer_join_clauses for a single FULL JOIN clause
2249 : *
2250 : * Returns true if we were able to propagate a constant through the clause.
2251 : */
2252 : static bool
2253 1222 : reconsider_full_join_clause(PlannerInfo *root, OuterJoinClauseInfo *ojcinfo)
2254 : {
2255 1222 : RestrictInfo *rinfo = ojcinfo->rinfo;
2256 1222 : SpecialJoinInfo *sjinfo = ojcinfo->sjinfo;
2257 1222 : Relids fjrelids = bms_make_singleton(sjinfo->ojrelid);
2258 : Expr *leftvar;
2259 : Expr *rightvar;
2260 : Oid opno,
2261 : collation,
2262 : left_type,
2263 : right_type;
2264 : Relids left_relids,
2265 : right_relids;
2266 : ListCell *lc1;
2267 :
2268 : /* Extract needed info from the clause */
2269 : Assert(is_opclause(rinfo->clause));
2270 1222 : opno = ((OpExpr *) rinfo->clause)->opno;
2271 1222 : collation = ((OpExpr *) rinfo->clause)->inputcollid;
2272 1222 : op_input_types(opno, &left_type, &right_type);
2273 1222 : leftvar = (Expr *) get_leftop(rinfo->clause);
2274 1222 : rightvar = (Expr *) get_rightop(rinfo->clause);
2275 1222 : left_relids = rinfo->left_relids;
2276 1222 : right_relids = rinfo->right_relids;
2277 :
2278 6246 : foreach(lc1, root->eq_classes)
2279 : {
2280 5030 : EquivalenceClass *cur_ec = (EquivalenceClass *) lfirst(lc1);
2281 5030 : EquivalenceMember *coal_em = NULL;
2282 : bool match;
2283 : bool matchleft;
2284 : bool matchright;
2285 : ListCell *lc2;
2286 5030 : int coal_idx = -1;
2287 :
2288 : /* Ignore EC unless it contains pseudoconstants */
2289 5030 : if (!cur_ec->ec_has_const)
2290 4734 : continue;
2291 : /* Never match to a volatile EC */
2292 296 : if (cur_ec->ec_has_volatile)
2293 0 : continue;
2294 : /* It has to match the outer-join clause as to semantics, too */
2295 296 : if (collation != cur_ec->ec_collation)
2296 36 : continue;
2297 260 : if (!equal(rinfo->mergeopfamilies, cur_ec->ec_opfamilies))
2298 0 : continue;
2299 :
2300 : /*
2301 : * Does it contain a COALESCE(leftvar, rightvar) construct?
2302 : *
2303 : * We can assume the COALESCE() inputs are in the same order as the
2304 : * join clause, since both were automatically generated in the cases
2305 : * we care about.
2306 : *
2307 : * XXX currently this may fail to match in cross-type cases because
2308 : * the COALESCE will contain typecast operations while the join clause
2309 : * may not (if there is a cross-type mergejoin operator available for
2310 : * the two column types). Is it OK to strip implicit coercions from
2311 : * the COALESCE arguments?
2312 : */
2313 260 : match = false;
2314 762 : foreach(lc2, cur_ec->ec_members)
2315 : {
2316 508 : coal_em = (EquivalenceMember *) lfirst(lc2);
2317 : Assert(!coal_em->em_is_child); /* no children yet */
2318 508 : if (IsA(coal_em->em_expr, CoalesceExpr))
2319 : {
2320 18 : CoalesceExpr *cexpr = (CoalesceExpr *) coal_em->em_expr;
2321 : Node *cfirst;
2322 : Node *csecond;
2323 :
2324 18 : if (list_length(cexpr->args) != 2)
2325 12 : continue;
2326 6 : cfirst = (Node *) linitial(cexpr->args);
2327 6 : csecond = (Node *) lsecond(cexpr->args);
2328 :
2329 : /*
2330 : * The COALESCE arguments will be marked as possibly nulled by
2331 : * the full join, while we wish to generate clauses that apply
2332 : * to the join's inputs. So we must strip the join from the
2333 : * nullingrels fields of cfirst/csecond before comparing them
2334 : * to leftvar/rightvar. (Perhaps with a less hokey
2335 : * representation for FULL JOIN USING output columns, this
2336 : * wouldn't be needed?)
2337 : */
2338 6 : cfirst = remove_nulling_relids(cfirst, fjrelids, NULL);
2339 6 : csecond = remove_nulling_relids(csecond, fjrelids, NULL);
2340 :
2341 6 : if (equal(leftvar, cfirst) && equal(rightvar, csecond))
2342 : {
2343 6 : coal_idx = foreach_current_index(lc2);
2344 6 : match = true;
2345 6 : break;
2346 : }
2347 : }
2348 : }
2349 260 : if (!match)
2350 254 : continue; /* no match, so ignore this EC */
2351 :
2352 : /*
2353 : * Yes it does! Try to generate clauses LEFTVAR = CONSTANT and
2354 : * RIGHTVAR = CONSTANT for each CONSTANT in the EC. Note that we must
2355 : * succeed with at least one constant for each var before we can
2356 : * decide to throw away the outer-join clause.
2357 : */
2358 6 : matchleft = matchright = false;
2359 18 : foreach(lc2, cur_ec->ec_members)
2360 : {
2361 12 : EquivalenceMember *cur_em = (EquivalenceMember *) lfirst(lc2);
2362 : Oid eq_op;
2363 : RestrictInfo *newrinfo;
2364 : JoinDomain *jdomain;
2365 :
2366 12 : if (!cur_em->em_is_const)
2367 6 : continue; /* ignore non-const members */
2368 6 : eq_op = select_equality_operator(cur_ec,
2369 : left_type,
2370 : cur_em->em_datatype);
2371 6 : if (OidIsValid(eq_op))
2372 : {
2373 6 : newrinfo = build_implied_join_equality(root,
2374 : eq_op,
2375 : cur_ec->ec_collation,
2376 : leftvar,
2377 : cur_em->em_expr,
2378 : bms_copy(left_relids),
2379 : cur_ec->ec_min_security);
2380 : /* This equality holds within the lefthand child JoinDomain */
2381 6 : jdomain = find_join_domain(root, sjinfo->syn_lefthand);
2382 6 : if (process_equivalence(root, &newrinfo, jdomain))
2383 6 : matchleft = true;
2384 : }
2385 6 : eq_op = select_equality_operator(cur_ec,
2386 : right_type,
2387 : cur_em->em_datatype);
2388 6 : if (OidIsValid(eq_op))
2389 : {
2390 6 : newrinfo = build_implied_join_equality(root,
2391 : eq_op,
2392 : cur_ec->ec_collation,
2393 : rightvar,
2394 : cur_em->em_expr,
2395 : bms_copy(right_relids),
2396 : cur_ec->ec_min_security);
2397 : /* This equality holds within the righthand child JoinDomain */
2398 6 : jdomain = find_join_domain(root, sjinfo->syn_righthand);
2399 6 : if (process_equivalence(root, &newrinfo, jdomain))
2400 6 : matchright = true;
2401 : }
2402 : }
2403 :
2404 : /*
2405 : * If we were able to equate both vars to constants, we're done, and
2406 : * we can throw away the full-join clause as redundant. Moreover, we
2407 : * can remove the COALESCE entry from the EC, since the added
2408 : * restrictions ensure it will always have the expected value. (We
2409 : * don't bother trying to update ec_relids or ec_sources.)
2410 : */
2411 6 : if (matchleft && matchright)
2412 : {
2413 6 : cur_ec->ec_members = list_delete_nth_cell(cur_ec->ec_members, coal_idx);
2414 6 : return true;
2415 : }
2416 :
2417 : /*
2418 : * Otherwise, fall out of the search loop, since we know the COALESCE
2419 : * appears in at most one EC (XXX might stop being true if we allow
2420 : * stripping of coercions above?)
2421 : */
2422 0 : break;
2423 : }
2424 :
2425 1216 : return false; /* failed to make any deduction */
2426 : }
2427 :
2428 : /*
2429 : * find_join_domain
2430 : * Find the highest JoinDomain enclosed within the given relid set.
2431 : *
2432 : * (We could avoid this search at the cost of complicating APIs elsewhere,
2433 : * which doesn't seem worth it.)
2434 : */
2435 : static JoinDomain *
2436 1668 : find_join_domain(PlannerInfo *root, Relids relids)
2437 : {
2438 : ListCell *lc;
2439 :
2440 3426 : foreach(lc, root->join_domains)
2441 : {
2442 3426 : JoinDomain *jdomain = (JoinDomain *) lfirst(lc);
2443 :
2444 3426 : if (bms_is_subset(jdomain->jd_relids, relids))
2445 1668 : return jdomain;
2446 : }
2447 0 : elog(ERROR, "failed to find appropriate JoinDomain");
2448 : return NULL; /* keep compiler quiet */
2449 : }
2450 :
2451 :
2452 : /*
2453 : * exprs_known_equal
2454 : * Detect whether two expressions are known equal due to equivalence
2455 : * relationships.
2456 : *
2457 : * Actually, this only shows that the expressions are equal according
2458 : * to some opfamily's notion of equality --- but we only use it for
2459 : * selectivity estimation, so a fuzzy idea of equality is OK.
2460 : *
2461 : * Note: does not bother to check for "equal(item1, item2)"; caller must
2462 : * check that case if it's possible to pass identical items.
2463 : */
2464 : bool
2465 3240 : exprs_known_equal(PlannerInfo *root, Node *item1, Node *item2)
2466 : {
2467 : ListCell *lc1;
2468 :
2469 19744 : foreach(lc1, root->eq_classes)
2470 : {
2471 16624 : EquivalenceClass *ec = (EquivalenceClass *) lfirst(lc1);
2472 16624 : bool item1member = false;
2473 16624 : bool item2member = false;
2474 : ListCell *lc2;
2475 :
2476 : /* Never match to a volatile EC */
2477 16624 : if (ec->ec_has_volatile)
2478 0 : continue;
2479 :
2480 53116 : foreach(lc2, ec->ec_members)
2481 : {
2482 36612 : EquivalenceMember *em = (EquivalenceMember *) lfirst(lc2);
2483 :
2484 36612 : if (em->em_is_child)
2485 12756 : continue; /* ignore children here */
2486 23856 : if (equal(item1, em->em_expr))
2487 1618 : item1member = true;
2488 22238 : else if (equal(item2, em->em_expr))
2489 1762 : item2member = true;
2490 : /* Exit as soon as equality is proven */
2491 23856 : if (item1member && item2member)
2492 120 : return true;
2493 : }
2494 : }
2495 3120 : return false;
2496 : }
2497 :
2498 :
2499 : /*
2500 : * match_eclasses_to_foreign_key_col
2501 : * See whether a foreign key column match is proven by any eclass.
2502 : *
2503 : * If the referenced and referencing Vars of the fkey's colno'th column are
2504 : * known equal due to any eclass, return that eclass; otherwise return NULL.
2505 : * (In principle there might be more than one matching eclass if multiple
2506 : * collations are involved, but since collation doesn't matter for equality,
2507 : * we ignore that fine point here.) This is much like exprs_known_equal,
2508 : * except that we insist on the comparison operator matching the eclass, so
2509 : * that the result is definite not approximate.
2510 : *
2511 : * On success, we also set fkinfo->eclass[colno] to the matching eclass,
2512 : * and set fkinfo->fk_eclass_member[colno] to the eclass member for the
2513 : * referencing Var.
2514 : */
2515 : EquivalenceClass *
2516 2162 : match_eclasses_to_foreign_key_col(PlannerInfo *root,
2517 : ForeignKeyOptInfo *fkinfo,
2518 : int colno)
2519 : {
2520 2162 : Index var1varno = fkinfo->con_relid;
2521 2162 : AttrNumber var1attno = fkinfo->conkey[colno];
2522 2162 : Index var2varno = fkinfo->ref_relid;
2523 2162 : AttrNumber var2attno = fkinfo->confkey[colno];
2524 2162 : Oid eqop = fkinfo->conpfeqop[colno];
2525 2162 : RelOptInfo *rel1 = root->simple_rel_array[var1varno];
2526 2162 : RelOptInfo *rel2 = root->simple_rel_array[var2varno];
2527 2162 : List *opfamilies = NIL; /* compute only if needed */
2528 : Bitmapset *matching_ecs;
2529 : int i;
2530 :
2531 : /* Consider only eclasses mentioning both relations */
2532 : Assert(root->ec_merging_done);
2533 : Assert(IS_SIMPLE_REL(rel1));
2534 : Assert(IS_SIMPLE_REL(rel2));
2535 2162 : matching_ecs = bms_intersect(rel1->eclass_indexes,
2536 2162 : rel2->eclass_indexes);
2537 :
2538 2162 : i = -1;
2539 2258 : while ((i = bms_next_member(matching_ecs, i)) >= 0)
2540 : {
2541 438 : EquivalenceClass *ec = (EquivalenceClass *) list_nth(root->eq_classes,
2542 : i);
2543 438 : EquivalenceMember *item1_em = NULL;
2544 438 : EquivalenceMember *item2_em = NULL;
2545 : ListCell *lc2;
2546 :
2547 : /* Never match to a volatile EC */
2548 438 : if (ec->ec_has_volatile)
2549 0 : continue;
2550 : /* Note: it seems okay to match to "broken" eclasses here */
2551 :
2552 1074 : foreach(lc2, ec->ec_members)
2553 : {
2554 978 : EquivalenceMember *em = (EquivalenceMember *) lfirst(lc2);
2555 : Var *var;
2556 :
2557 978 : if (em->em_is_child)
2558 0 : continue; /* ignore children here */
2559 :
2560 : /* EM must be a Var, possibly with RelabelType */
2561 978 : var = (Var *) em->em_expr;
2562 978 : while (var && IsA(var, RelabelType))
2563 0 : var = (Var *) ((RelabelType *) var)->arg;
2564 978 : if (!(var && IsA(var, Var)))
2565 6 : continue;
2566 :
2567 : /* Match? */
2568 972 : if (var->varno == var1varno && var->varattno == var1attno)
2569 342 : item1_em = em;
2570 630 : else if (var->varno == var2varno && var->varattno == var2attno)
2571 342 : item2_em = em;
2572 :
2573 : /* Have we found both PK and FK column in this EC? */
2574 972 : if (item1_em && item2_em)
2575 : {
2576 : /*
2577 : * Succeed if eqop matches EC's opfamilies. We could test
2578 : * this before scanning the members, but it's probably cheaper
2579 : * to test for member matches first.
2580 : */
2581 342 : if (opfamilies == NIL) /* compute if we didn't already */
2582 342 : opfamilies = get_mergejoin_opfamilies(eqop);
2583 342 : if (equal(opfamilies, ec->ec_opfamilies))
2584 : {
2585 342 : fkinfo->eclass[colno] = ec;
2586 342 : fkinfo->fk_eclass_member[colno] = item2_em;
2587 342 : return ec;
2588 : }
2589 : /* Otherwise, done with this EC, move on to the next */
2590 0 : break;
2591 : }
2592 : }
2593 : }
2594 1820 : return NULL;
2595 : }
2596 :
2597 : /*
2598 : * find_derived_clause_for_ec_member
2599 : * Search for a previously-derived clause mentioning the given EM.
2600 : *
2601 : * The eclass should be an ec_has_const EC, of which the EM is a non-const
2602 : * member. This should ensure there is just one derived clause mentioning
2603 : * the EM (and equating it to a constant).
2604 : * Returns NULL if no such clause can be found.
2605 : */
2606 : RestrictInfo *
2607 6 : find_derived_clause_for_ec_member(EquivalenceClass *ec,
2608 : EquivalenceMember *em)
2609 : {
2610 : ListCell *lc;
2611 :
2612 : Assert(ec->ec_has_const);
2613 : Assert(!em->em_is_const);
2614 6 : foreach(lc, ec->ec_derives)
2615 : {
2616 6 : RestrictInfo *rinfo = (RestrictInfo *) lfirst(lc);
2617 :
2618 : /*
2619 : * generate_base_implied_equalities_const will have put non-const
2620 : * members on the left side of derived clauses.
2621 : */
2622 6 : if (rinfo->left_em == em)
2623 6 : return rinfo;
2624 : }
2625 0 : return NULL;
2626 : }
2627 :
2628 :
2629 : /*
2630 : * add_child_rel_equivalences
2631 : * Search for EC members that reference the root parent of child_rel, and
2632 : * add transformed members referencing the child_rel.
2633 : *
2634 : * Note that this function won't be called at all unless we have at least some
2635 : * reason to believe that the EC members it generates will be useful.
2636 : *
2637 : * parent_rel and child_rel could be derived from appinfo, but since the
2638 : * caller has already computed them, we might as well just pass them in.
2639 : *
2640 : * The passed-in AppendRelInfo is not used when the parent_rel is not a
2641 : * top-level baserel, since it shows the mapping from the parent_rel but
2642 : * we need to translate EC expressions that refer to the top-level parent.
2643 : * Using it is faster than using adjust_appendrel_attrs_multilevel(), though,
2644 : * so we prefer it when we can.
2645 : */
2646 : void
2647 21618 : add_child_rel_equivalences(PlannerInfo *root,
2648 : AppendRelInfo *appinfo,
2649 : RelOptInfo *parent_rel,
2650 : RelOptInfo *child_rel)
2651 : {
2652 21618 : Relids top_parent_relids = child_rel->top_parent_relids;
2653 21618 : Relids child_relids = child_rel->relids;
2654 : int i;
2655 :
2656 : /*
2657 : * EC merging should be complete already, so we can use the parent rel's
2658 : * eclass_indexes to avoid searching all of root->eq_classes.
2659 : */
2660 : Assert(root->ec_merging_done);
2661 : Assert(IS_SIMPLE_REL(parent_rel));
2662 :
2663 21618 : i = -1;
2664 64088 : while ((i = bms_next_member(parent_rel->eclass_indexes, i)) >= 0)
2665 : {
2666 42470 : EquivalenceClass *cur_ec = (EquivalenceClass *) list_nth(root->eq_classes, i);
2667 : int num_members;
2668 :
2669 : /*
2670 : * If this EC contains a volatile expression, then generating child
2671 : * EMs would be downright dangerous, so skip it. We rely on a
2672 : * volatile EC having only one EM.
2673 : */
2674 42470 : if (cur_ec->ec_has_volatile)
2675 0 : continue;
2676 :
2677 : /* Sanity check eclass_indexes only contain ECs for parent_rel */
2678 : Assert(bms_is_subset(top_parent_relids, cur_ec->ec_relids));
2679 :
2680 : /*
2681 : * We don't use foreach() here because there's no point in scanning
2682 : * newly-added child members, so we can stop after the last
2683 : * pre-existing EC member.
2684 : */
2685 42470 : num_members = list_length(cur_ec->ec_members);
2686 198936 : for (int pos = 0; pos < num_members; pos++)
2687 : {
2688 156466 : EquivalenceMember *cur_em = (EquivalenceMember *) list_nth(cur_ec->ec_members, pos);
2689 :
2690 156466 : if (cur_em->em_is_const)
2691 3168 : continue; /* ignore consts here */
2692 :
2693 : /*
2694 : * We consider only original EC members here, not
2695 : * already-transformed child members. Otherwise, if some original
2696 : * member expression references more than one appendrel, we'd get
2697 : * an O(N^2) explosion of useless derived expressions for
2698 : * combinations of children. (But add_child_join_rel_equivalences
2699 : * may add targeted combinations for partitionwise-join purposes.)
2700 : */
2701 153298 : if (cur_em->em_is_child)
2702 97280 : continue; /* ignore children here */
2703 :
2704 : /*
2705 : * Consider only members that reference and can be computed at
2706 : * child's topmost parent rel. In particular we want to exclude
2707 : * parent-rel Vars that have nonempty varnullingrels. Translating
2708 : * those might fail, if the transformed expression wouldn't be a
2709 : * simple Var; and in any case it wouldn't produce a member that
2710 : * has any use in creating plans for the child rel.
2711 : */
2712 56018 : if (bms_is_subset(cur_em->em_relids, top_parent_relids) &&
2713 40034 : !bms_is_empty(cur_em->em_relids))
2714 : {
2715 : /* OK, generate transformed child version */
2716 : Expr *child_expr;
2717 : Relids new_relids;
2718 :
2719 40034 : if (parent_rel->reloptkind == RELOPT_BASEREL)
2720 : {
2721 : /* Simple single-level transformation */
2722 : child_expr = (Expr *)
2723 32504 : adjust_appendrel_attrs(root,
2724 32504 : (Node *) cur_em->em_expr,
2725 : 1, &appinfo);
2726 : }
2727 : else
2728 : {
2729 : /* Must do multi-level transformation */
2730 : child_expr = (Expr *)
2731 7530 : adjust_appendrel_attrs_multilevel(root,
2732 7530 : (Node *) cur_em->em_expr,
2733 : child_rel,
2734 7530 : child_rel->top_parent);
2735 : }
2736 :
2737 : /*
2738 : * Transform em_relids to match. Note we do *not* do
2739 : * pull_varnos(child_expr) here, as for example the
2740 : * transformation might have substituted a constant, but we
2741 : * don't want the child member to be marked as constant.
2742 : */
2743 40034 : new_relids = bms_difference(cur_em->em_relids,
2744 : top_parent_relids);
2745 40034 : new_relids = bms_add_members(new_relids, child_relids);
2746 :
2747 40034 : (void) add_eq_member(cur_ec, child_expr, new_relids,
2748 : cur_em->em_jdomain,
2749 : cur_em, cur_em->em_datatype);
2750 :
2751 : /* Record this EC index for the child rel */
2752 40034 : child_rel->eclass_indexes = bms_add_member(child_rel->eclass_indexes, i);
2753 : }
2754 : }
2755 : }
2756 21618 : }
2757 :
2758 : /*
2759 : * add_child_join_rel_equivalences
2760 : * Like add_child_rel_equivalences(), but for joinrels
2761 : *
2762 : * Here we find the ECs relevant to the top parent joinrel and add transformed
2763 : * member expressions that refer to this child joinrel.
2764 : *
2765 : * Note that this function won't be called at all unless we have at least some
2766 : * reason to believe that the EC members it generates will be useful.
2767 : */
2768 : void
2769 4180 : add_child_join_rel_equivalences(PlannerInfo *root,
2770 : int nappinfos, AppendRelInfo **appinfos,
2771 : RelOptInfo *parent_joinrel,
2772 : RelOptInfo *child_joinrel)
2773 : {
2774 4180 : Relids top_parent_relids = child_joinrel->top_parent_relids;
2775 4180 : Relids child_relids = child_joinrel->relids;
2776 : Bitmapset *matching_ecs;
2777 : MemoryContext oldcontext;
2778 : int i;
2779 :
2780 : Assert(IS_JOIN_REL(child_joinrel) && IS_JOIN_REL(parent_joinrel));
2781 :
2782 : /* We need consider only ECs that mention the parent joinrel */
2783 4180 : matching_ecs = get_eclass_indexes_for_relids(root, top_parent_relids);
2784 :
2785 : /*
2786 : * If we're being called during GEQO join planning, we still have to
2787 : * create any new EC members in the main planner context, to avoid having
2788 : * a corrupt EC data structure after the GEQO context is reset. This is
2789 : * problematic since we'll leak memory across repeated GEQO cycles. For
2790 : * now, though, bloat is better than crash. If it becomes a real issue
2791 : * we'll have to do something to avoid generating duplicate EC members.
2792 : */
2793 4180 : oldcontext = MemoryContextSwitchTo(root->planner_cxt);
2794 :
2795 4180 : i = -1;
2796 19686 : while ((i = bms_next_member(matching_ecs, i)) >= 0)
2797 : {
2798 15506 : EquivalenceClass *cur_ec = (EquivalenceClass *) list_nth(root->eq_classes, i);
2799 : int num_members;
2800 :
2801 : /*
2802 : * If this EC contains a volatile expression, then generating child
2803 : * EMs would be downright dangerous, so skip it. We rely on a
2804 : * volatile EC having only one EM.
2805 : */
2806 15506 : if (cur_ec->ec_has_volatile)
2807 0 : continue;
2808 :
2809 : /* Sanity check on get_eclass_indexes_for_relids result */
2810 : Assert(bms_overlap(top_parent_relids, cur_ec->ec_relids));
2811 :
2812 : /*
2813 : * We don't use foreach() here because there's no point in scanning
2814 : * newly-added child members, so we can stop after the last
2815 : * pre-existing EC member.
2816 : */
2817 15506 : num_members = list_length(cur_ec->ec_members);
2818 103442 : for (int pos = 0; pos < num_members; pos++)
2819 : {
2820 87936 : EquivalenceMember *cur_em = (EquivalenceMember *) list_nth(cur_ec->ec_members, pos);
2821 :
2822 87936 : if (cur_em->em_is_const)
2823 2232 : continue; /* ignore consts here */
2824 :
2825 : /*
2826 : * We consider only original EC members here, not
2827 : * already-transformed child members.
2828 : */
2829 85704 : if (cur_em->em_is_child)
2830 66074 : continue; /* ignore children here */
2831 :
2832 : /*
2833 : * We may ignore expressions that reference a single baserel,
2834 : * because add_child_rel_equivalences should have handled them.
2835 : */
2836 19630 : if (bms_membership(cur_em->em_relids) != BMS_MULTIPLE)
2837 17008 : continue;
2838 :
2839 : /* Does this member reference child's topmost parent rel? */
2840 2622 : if (bms_overlap(cur_em->em_relids, top_parent_relids))
2841 : {
2842 : /* Yes, generate transformed child version */
2843 : Expr *child_expr;
2844 : Relids new_relids;
2845 :
2846 2622 : if (parent_joinrel->reloptkind == RELOPT_JOINREL)
2847 : {
2848 : /* Simple single-level transformation */
2849 : child_expr = (Expr *)
2850 2526 : adjust_appendrel_attrs(root,
2851 2526 : (Node *) cur_em->em_expr,
2852 : nappinfos, appinfos);
2853 : }
2854 : else
2855 : {
2856 : /* Must do multi-level transformation */
2857 : Assert(parent_joinrel->reloptkind == RELOPT_OTHER_JOINREL);
2858 : child_expr = (Expr *)
2859 96 : adjust_appendrel_attrs_multilevel(root,
2860 96 : (Node *) cur_em->em_expr,
2861 : child_joinrel,
2862 96 : child_joinrel->top_parent);
2863 : }
2864 :
2865 : /*
2866 : * Transform em_relids to match. Note we do *not* do
2867 : * pull_varnos(child_expr) here, as for example the
2868 : * transformation might have substituted a constant, but we
2869 : * don't want the child member to be marked as constant.
2870 : */
2871 2622 : new_relids = bms_difference(cur_em->em_relids,
2872 : top_parent_relids);
2873 2622 : new_relids = bms_add_members(new_relids, child_relids);
2874 :
2875 2622 : (void) add_eq_member(cur_ec, child_expr, new_relids,
2876 : cur_em->em_jdomain,
2877 : cur_em, cur_em->em_datatype);
2878 : }
2879 : }
2880 : }
2881 :
2882 4180 : MemoryContextSwitchTo(oldcontext);
2883 4180 : }
2884 :
2885 : /*
2886 : * add_setop_child_rel_equivalences
2887 : * Add equivalence members for each non-resjunk target in 'child_tlist'
2888 : * to the EquivalenceClass in the corresponding setop_pathkey's pk_eclass.
2889 : *
2890 : * 'root' is the PlannerInfo belonging to the top-level set operation.
2891 : * 'child_rel' is the RelOptInfo of the child relation we're adding
2892 : * EquivalenceMembers for.
2893 : * 'child_tlist' is the target list for the setop child relation. The target
2894 : * list expressions are what we add as EquivalenceMembers.
2895 : * 'setop_pathkeys' is a list of PathKeys which must contain an entry for each
2896 : * non-resjunk target in 'child_tlist'.
2897 : */
2898 : void
2899 9554 : add_setop_child_rel_equivalences(PlannerInfo *root, RelOptInfo *child_rel,
2900 : List *child_tlist, List *setop_pathkeys)
2901 : {
2902 : ListCell *lc;
2903 9554 : ListCell *lc2 = list_head(setop_pathkeys);
2904 :
2905 38056 : foreach(lc, child_tlist)
2906 : {
2907 28502 : TargetEntry *tle = lfirst_node(TargetEntry, lc);
2908 : EquivalenceMember *parent_em;
2909 : PathKey *pk;
2910 :
2911 28502 : if (tle->resjunk)
2912 0 : continue;
2913 :
2914 28502 : if (lc2 == NULL)
2915 0 : elog(ERROR, "too few pathkeys for set operation");
2916 :
2917 28502 : pk = lfirst_node(PathKey, lc2);
2918 28502 : parent_em = linitial(pk->pk_eclass->ec_members);
2919 :
2920 : /*
2921 : * We can safely pass the parent member as the first member in the
2922 : * ec_members list as this is added first in generate_union_paths,
2923 : * likewise, the JoinDomain can be that of the initial member of the
2924 : * Pathkey's EquivalenceClass.
2925 : */
2926 28502 : add_eq_member(pk->pk_eclass,
2927 : tle->expr,
2928 : child_rel->relids,
2929 : parent_em->em_jdomain,
2930 : parent_em,
2931 28502 : exprType((Node *) tle->expr));
2932 :
2933 28502 : lc2 = lnext(setop_pathkeys, lc2);
2934 : }
2935 :
2936 : /*
2937 : * transformSetOperationStmt() ensures that the targetlist never contains
2938 : * any resjunk columns, so all eclasses that exist in 'root' must have
2939 : * received a new member in the loop above. Add them to the child_rel's
2940 : * eclass_indexes.
2941 : */
2942 9554 : child_rel->eclass_indexes = bms_add_range(child_rel->eclass_indexes, 0,
2943 9554 : list_length(root->eq_classes) - 1);
2944 9554 : }
2945 :
2946 :
2947 : /*
2948 : * generate_implied_equalities_for_column
2949 : * Create EC-derived joinclauses usable with a specific column.
2950 : *
2951 : * This is used by indxpath.c to extract potentially indexable joinclauses
2952 : * from ECs, and can be used by foreign data wrappers for similar purposes.
2953 : * We assume that only expressions in Vars of a single table are of interest,
2954 : * but the caller provides a callback function to identify exactly which
2955 : * such expressions it would like to know about.
2956 : *
2957 : * We assume that any given table/index column could appear in only one EC.
2958 : * (This should be true in all but the most pathological cases, and if it
2959 : * isn't, we stop on the first match anyway.) Therefore, what we return
2960 : * is a redundant list of clauses equating the table/index column to each of
2961 : * the other-relation values it is known to be equal to. Any one of
2962 : * these clauses can be used to create a parameterized path, and there
2963 : * is no value in using more than one. (But it *is* worthwhile to create
2964 : * a separate parameterized path for each one, since that leads to different
2965 : * join orders.)
2966 : *
2967 : * The caller can pass a Relids set of rels we aren't interested in joining
2968 : * to, so as to save the work of creating useless clauses.
2969 : */
2970 : List *
2971 467236 : generate_implied_equalities_for_column(PlannerInfo *root,
2972 : RelOptInfo *rel,
2973 : ec_matches_callback_type callback,
2974 : void *callback_arg,
2975 : Relids prohibited_rels)
2976 : {
2977 467236 : List *result = NIL;
2978 467236 : bool is_child_rel = (rel->reloptkind == RELOPT_OTHER_MEMBER_REL);
2979 : Relids parent_relids;
2980 : int i;
2981 :
2982 : /* Should be OK to rely on eclass_indexes */
2983 : Assert(root->ec_merging_done);
2984 :
2985 : /* Indexes are available only on base or "other" member relations. */
2986 : Assert(IS_SIMPLE_REL(rel));
2987 :
2988 : /* If it's a child rel, we'll need to know what its parent(s) are */
2989 467236 : if (is_child_rel)
2990 10526 : parent_relids = find_childrel_parents(root, rel);
2991 : else
2992 456710 : parent_relids = NULL; /* not used, but keep compiler quiet */
2993 :
2994 467236 : i = -1;
2995 1274090 : while ((i = bms_next_member(rel->eclass_indexes, i)) >= 0)
2996 : {
2997 886018 : EquivalenceClass *cur_ec = (EquivalenceClass *) list_nth(root->eq_classes, i);
2998 : EquivalenceMember *cur_em;
2999 : ListCell *lc2;
3000 :
3001 : /* Sanity check eclass_indexes only contain ECs for rel */
3002 : Assert(is_child_rel || bms_is_subset(rel->relids, cur_ec->ec_relids));
3003 :
3004 : /*
3005 : * Won't generate joinclauses if const or single-member (the latter
3006 : * test covers the volatile case too)
3007 : */
3008 886018 : if (cur_ec->ec_has_const || list_length(cur_ec->ec_members) <= 1)
3009 488144 : continue;
3010 :
3011 : /*
3012 : * Scan members, looking for a match to the target column. Note that
3013 : * child EC members are considered, but only when they belong to the
3014 : * target relation. (Unlike regular members, the same expression
3015 : * could be a child member of more than one EC. Therefore, it's
3016 : * potentially order-dependent which EC a child relation's target
3017 : * column gets matched to. This is annoying but it only happens in
3018 : * corner cases, so for now we live with just reporting the first
3019 : * match. See also get_eclass_for_sort_expr.)
3020 : */
3021 397874 : cur_em = NULL;
3022 1189256 : foreach(lc2, cur_ec->ec_members)
3023 : {
3024 870994 : cur_em = (EquivalenceMember *) lfirst(lc2);
3025 1269746 : if (bms_equal(cur_em->em_relids, rel->relids) &&
3026 398752 : callback(root, rel, cur_ec, cur_em, callback_arg))
3027 79612 : break;
3028 791382 : cur_em = NULL;
3029 : }
3030 :
3031 397874 : if (!cur_em)
3032 318262 : continue;
3033 :
3034 : /*
3035 : * Found our match. Scan the other EC members and attempt to generate
3036 : * joinclauses.
3037 : */
3038 263072 : foreach(lc2, cur_ec->ec_members)
3039 : {
3040 183460 : EquivalenceMember *other_em = (EquivalenceMember *) lfirst(lc2);
3041 : Oid eq_op;
3042 : RestrictInfo *rinfo;
3043 :
3044 183460 : if (other_em->em_is_child)
3045 20468 : continue; /* ignore children here */
3046 :
3047 : /* Make sure it'll be a join to a different rel */
3048 249236 : if (other_em == cur_em ||
3049 86244 : bms_overlap(other_em->em_relids, rel->relids))
3050 77080 : continue;
3051 :
3052 : /* Forget it if caller doesn't want joins to this rel */
3053 85912 : if (bms_overlap(other_em->em_relids, prohibited_rels))
3054 138 : continue;
3055 :
3056 : /*
3057 : * Also, if this is a child rel, avoid generating a useless join
3058 : * to its parent rel(s).
3059 : */
3060 91960 : if (is_child_rel &&
3061 6186 : bms_overlap(parent_relids, other_em->em_relids))
3062 2912 : continue;
3063 :
3064 82862 : eq_op = select_equality_operator(cur_ec,
3065 : cur_em->em_datatype,
3066 : other_em->em_datatype);
3067 82862 : if (!OidIsValid(eq_op))
3068 0 : continue;
3069 :
3070 : /* set parent_ec to mark as redundant with other joinclauses */
3071 82862 : rinfo = create_join_clause(root, cur_ec, eq_op,
3072 : cur_em, other_em,
3073 : cur_ec);
3074 :
3075 82862 : result = lappend(result, rinfo);
3076 : }
3077 :
3078 : /*
3079 : * If somehow we failed to create any join clauses, we might as well
3080 : * keep scanning the ECs for another match. But if we did make any,
3081 : * we're done, because we don't want to return non-redundant clauses.
3082 : */
3083 79612 : if (result)
3084 79164 : break;
3085 : }
3086 :
3087 467236 : return result;
3088 : }
3089 :
3090 : /*
3091 : * have_relevant_eclass_joinclause
3092 : * Detect whether there is an EquivalenceClass that could produce
3093 : * a joinclause involving the two given relations.
3094 : *
3095 : * This is essentially a very cut-down version of
3096 : * generate_join_implied_equalities(). Note it's OK to occasionally say "yes"
3097 : * incorrectly. Hence we don't bother with details like whether the lack of a
3098 : * cross-type operator might prevent the clause from actually being generated.
3099 : * False negatives are not always fatal either: they will discourage, but not
3100 : * completely prevent, investigation of particular join pathways.
3101 : */
3102 : bool
3103 141852 : have_relevant_eclass_joinclause(PlannerInfo *root,
3104 : RelOptInfo *rel1, RelOptInfo *rel2)
3105 : {
3106 : Bitmapset *matching_ecs;
3107 : int i;
3108 :
3109 : /*
3110 : * Examine only eclasses mentioning both rel1 and rel2.
3111 : *
3112 : * Note that we do not consider the possibility of an eclass generating
3113 : * "join" clauses that mention just one of the rels plus an outer join
3114 : * that could be formed from them. Although such clauses must be
3115 : * correctly enforced when we form the outer join, they don't seem like
3116 : * sufficient reason to prioritize this join over other ones. The join
3117 : * ordering rules will force the join to be made when necessary.
3118 : */
3119 141852 : matching_ecs = get_common_eclass_indexes(root, rel1->relids,
3120 : rel2->relids);
3121 :
3122 141852 : i = -1;
3123 141852 : while ((i = bms_next_member(matching_ecs, i)) >= 0)
3124 : {
3125 118058 : EquivalenceClass *ec = (EquivalenceClass *) list_nth(root->eq_classes,
3126 : i);
3127 :
3128 : /*
3129 : * Sanity check that get_common_eclass_indexes gave only ECs
3130 : * containing both rels.
3131 : */
3132 : Assert(bms_overlap(rel1->relids, ec->ec_relids));
3133 : Assert(bms_overlap(rel2->relids, ec->ec_relids));
3134 :
3135 : /*
3136 : * Won't generate joinclauses if single-member (this test covers the
3137 : * volatile case too)
3138 : */
3139 118058 : if (list_length(ec->ec_members) <= 1)
3140 0 : continue;
3141 :
3142 : /*
3143 : * We do not need to examine the individual members of the EC, because
3144 : * all that we care about is whether each rel overlaps the relids of
3145 : * at least one member, and get_common_eclass_indexes() and the single
3146 : * member check above are sufficient to prove that. (As with
3147 : * have_relevant_joinclause(), it is not necessary that the EC be able
3148 : * to form a joinclause relating exactly the two given rels, only that
3149 : * it be able to form a joinclause mentioning both, and this will
3150 : * surely be true if both of them overlap ec_relids.)
3151 : *
3152 : * Note we don't test ec_broken; if we did, we'd need a separate code
3153 : * path to look through ec_sources. Checking the membership anyway is
3154 : * OK as a possibly-overoptimistic heuristic.
3155 : *
3156 : * We don't test ec_has_const either, even though a const eclass won't
3157 : * generate real join clauses. This is because if we had "WHERE a.x =
3158 : * b.y and a.x = 42", it is worth considering a join between a and b,
3159 : * since the join result is likely to be small even though it'll end
3160 : * up being an unqualified nestloop.
3161 : */
3162 :
3163 118058 : return true;
3164 : }
3165 :
3166 23794 : return false;
3167 : }
3168 :
3169 :
3170 : /*
3171 : * has_relevant_eclass_joinclause
3172 : * Detect whether there is an EquivalenceClass that could produce
3173 : * a joinclause involving the given relation and anything else.
3174 : *
3175 : * This is the same as have_relevant_eclass_joinclause with the other rel
3176 : * implicitly defined as "everything else in the query".
3177 : */
3178 : bool
3179 177424 : has_relevant_eclass_joinclause(PlannerInfo *root, RelOptInfo *rel1)
3180 : {
3181 : Bitmapset *matched_ecs;
3182 : int i;
3183 :
3184 : /* Examine only eclasses mentioning rel1 */
3185 177424 : matched_ecs = get_eclass_indexes_for_relids(root, rel1->relids);
3186 :
3187 177424 : i = -1;
3188 643158 : while ((i = bms_next_member(matched_ecs, i)) >= 0)
3189 : {
3190 524048 : EquivalenceClass *ec = (EquivalenceClass *) list_nth(root->eq_classes,
3191 : i);
3192 :
3193 : /*
3194 : * Won't generate joinclauses if single-member (this test covers the
3195 : * volatile case too)
3196 : */
3197 524048 : if (list_length(ec->ec_members) <= 1)
3198 262790 : continue;
3199 :
3200 : /*
3201 : * Per the comment in have_relevant_eclass_joinclause, it's sufficient
3202 : * to find an EC that mentions both this rel and some other rel.
3203 : */
3204 261258 : if (!bms_is_subset(ec->ec_relids, rel1->relids))
3205 58314 : return true;
3206 : }
3207 :
3208 119110 : return false;
3209 : }
3210 :
3211 :
3212 : /*
3213 : * eclass_useful_for_merging
3214 : * Detect whether the EC could produce any mergejoinable join clauses
3215 : * against the specified relation.
3216 : *
3217 : * This is just a heuristic test and doesn't have to be exact; it's better
3218 : * to say "yes" incorrectly than "no". Hence we don't bother with details
3219 : * like whether the lack of a cross-type operator might prevent the clause
3220 : * from actually being generated.
3221 : */
3222 : bool
3223 610482 : eclass_useful_for_merging(PlannerInfo *root,
3224 : EquivalenceClass *eclass,
3225 : RelOptInfo *rel)
3226 : {
3227 : Relids relids;
3228 : ListCell *lc;
3229 :
3230 : Assert(!eclass->ec_merged);
3231 :
3232 : /*
3233 : * Won't generate joinclauses if const or single-member (the latter test
3234 : * covers the volatile case too)
3235 : */
3236 610482 : if (eclass->ec_has_const || list_length(eclass->ec_members) <= 1)
3237 59370 : return false;
3238 :
3239 : /*
3240 : * Note we don't test ec_broken; if we did, we'd need a separate code path
3241 : * to look through ec_sources. Checking the members anyway is OK as a
3242 : * possibly-overoptimistic heuristic.
3243 : */
3244 :
3245 : /* If specified rel is a child, we must consider the topmost parent rel */
3246 551112 : if (IS_OTHER_REL(rel))
3247 : {
3248 : Assert(!bms_is_empty(rel->top_parent_relids));
3249 10434 : relids = rel->top_parent_relids;
3250 : }
3251 : else
3252 540678 : relids = rel->relids;
3253 :
3254 : /* If rel already includes all members of eclass, no point in searching */
3255 551112 : if (bms_is_subset(eclass->ec_relids, relids))
3256 211708 : return false;
3257 :
3258 : /* To join, we need a member not in the given rel */
3259 527118 : foreach(lc, eclass->ec_members)
3260 : {
3261 526596 : EquivalenceMember *cur_em = (EquivalenceMember *) lfirst(lc);
3262 :
3263 526596 : if (cur_em->em_is_child)
3264 0 : continue; /* ignore children here */
3265 :
3266 526596 : if (!bms_overlap(cur_em->em_relids, relids))
3267 338882 : return true;
3268 : }
3269 :
3270 522 : return false;
3271 : }
3272 :
3273 :
3274 : /*
3275 : * is_redundant_derived_clause
3276 : * Test whether rinfo is derived from same EC as any clause in clauselist;
3277 : * if so, it can be presumed to represent a condition that's redundant
3278 : * with that member of the list.
3279 : */
3280 : bool
3281 72 : is_redundant_derived_clause(RestrictInfo *rinfo, List *clauselist)
3282 : {
3283 72 : EquivalenceClass *parent_ec = rinfo->parent_ec;
3284 : ListCell *lc;
3285 :
3286 : /* Fail if it's not a potentially-redundant clause from some EC */
3287 72 : if (parent_ec == NULL)
3288 72 : return false;
3289 :
3290 0 : foreach(lc, clauselist)
3291 : {
3292 0 : RestrictInfo *otherrinfo = (RestrictInfo *) lfirst(lc);
3293 :
3294 0 : if (otherrinfo->parent_ec == parent_ec)
3295 0 : return true;
3296 : }
3297 :
3298 0 : return false;
3299 : }
3300 :
3301 : /*
3302 : * is_redundant_with_indexclauses
3303 : * Test whether rinfo is redundant with any clause in the IndexClause
3304 : * list. Here, for convenience, we test both simple identity and
3305 : * whether it is derived from the same EC as any member of the list.
3306 : */
3307 : bool
3308 1156074 : is_redundant_with_indexclauses(RestrictInfo *rinfo, List *indexclauses)
3309 : {
3310 1156074 : EquivalenceClass *parent_ec = rinfo->parent_ec;
3311 : ListCell *lc;
3312 :
3313 1591530 : foreach(lc, indexclauses)
3314 : {
3315 1192392 : IndexClause *iclause = lfirst_node(IndexClause, lc);
3316 1192392 : RestrictInfo *otherrinfo = iclause->rinfo;
3317 :
3318 : /* If indexclause is lossy, it won't enforce the condition exactly */
3319 1192392 : if (iclause->lossy)
3320 36594 : continue;
3321 :
3322 : /* Match if it's same clause (pointer equality should be enough) */
3323 1155798 : if (rinfo == otherrinfo)
3324 756936 : return true;
3325 : /* Match if derived from same EC */
3326 399150 : if (parent_ec && otherrinfo->parent_ec == parent_ec)
3327 288 : return true;
3328 :
3329 : /*
3330 : * No need to look at the derived clauses in iclause->indexquals; they
3331 : * couldn't match if the parent clause didn't.
3332 : */
3333 : }
3334 :
3335 399138 : return false;
3336 : }
3337 :
3338 : /*
3339 : * get_eclass_indexes_for_relids
3340 : * Build and return a Bitmapset containing the indexes into root's
3341 : * eq_classes list for all eclasses that mention any of these relids
3342 : */
3343 : static Bitmapset *
3344 857152 : get_eclass_indexes_for_relids(PlannerInfo *root, Relids relids)
3345 : {
3346 857152 : Bitmapset *ec_indexes = NULL;
3347 857152 : int i = -1;
3348 :
3349 : /* Should be OK to rely on eclass_indexes */
3350 : Assert(root->ec_merging_done);
3351 :
3352 2790274 : while ((i = bms_next_member(relids, i)) > 0)
3353 : {
3354 1933122 : RelOptInfo *rel = root->simple_rel_array[i];
3355 :
3356 1933122 : if (rel == NULL) /* must be an outer join */
3357 : {
3358 : Assert(bms_is_member(i, root->outer_join_rels));
3359 303970 : continue;
3360 : }
3361 :
3362 1629152 : ec_indexes = bms_add_members(ec_indexes, rel->eclass_indexes);
3363 : }
3364 857152 : return ec_indexes;
3365 : }
3366 :
3367 : /*
3368 : * get_common_eclass_indexes
3369 : * Build and return a Bitmapset containing the indexes into root's
3370 : * eq_classes list for all eclasses that mention rels in both
3371 : * relids1 and relids2.
3372 : */
3373 : static Bitmapset *
3374 480956 : get_common_eclass_indexes(PlannerInfo *root, Relids relids1, Relids relids2)
3375 : {
3376 : Bitmapset *rel1ecs;
3377 : Bitmapset *rel2ecs;
3378 : int relid;
3379 :
3380 480956 : rel1ecs = get_eclass_indexes_for_relids(root, relids1);
3381 :
3382 : /*
3383 : * We can get away with just using the relation's eclass_indexes directly
3384 : * when relids2 is a singleton set.
3385 : */
3386 480956 : if (bms_get_singleton_member(relids2, &relid))
3387 379326 : rel2ecs = root->simple_rel_array[relid]->eclass_indexes;
3388 : else
3389 101630 : rel2ecs = get_eclass_indexes_for_relids(root, relids2);
3390 :
3391 : /* Calculate and return the common EC indexes, recycling the left input. */
3392 480956 : return bms_int_members(rel1ecs, rel2ecs);
3393 : }
|
