Line data Source code
1 : /*-------------------------------------------------------------------------
2 : *
3 : * pathnode.c
4 : * Routines to manipulate pathlists and create path nodes
5 : *
6 : * Portions Copyright (c) 1996-2024, PostgreSQL Global Development Group
7 : * Portions Copyright (c) 1994, Regents of the University of California
8 : *
9 : *
10 : * IDENTIFICATION
11 : * src/backend/optimizer/util/pathnode.c
12 : *
13 : *-------------------------------------------------------------------------
14 : */
15 : #include "postgres.h"
16 :
17 : #include <math.h>
18 :
19 : #include "foreign/fdwapi.h"
20 : #include "miscadmin.h"
21 : #include "nodes/extensible.h"
22 : #include "optimizer/appendinfo.h"
23 : #include "optimizer/clauses.h"
24 : #include "optimizer/cost.h"
25 : #include "optimizer/optimizer.h"
26 : #include "optimizer/pathnode.h"
27 : #include "optimizer/paths.h"
28 : #include "optimizer/planmain.h"
29 : #include "optimizer/tlist.h"
30 : #include "parser/parsetree.h"
31 : #include "utils/memutils.h"
32 : #include "utils/selfuncs.h"
33 :
34 : typedef enum
35 : {
36 : COSTS_EQUAL, /* path costs are fuzzily equal */
37 : COSTS_BETTER1, /* first path is cheaper than second */
38 : COSTS_BETTER2, /* second path is cheaper than first */
39 : COSTS_DIFFERENT, /* neither path dominates the other on cost */
40 : } PathCostComparison;
41 :
42 : /*
43 : * STD_FUZZ_FACTOR is the normal fuzz factor for compare_path_costs_fuzzily.
44 : * XXX is it worth making this user-controllable? It provides a tradeoff
45 : * between planner runtime and the accuracy of path cost comparisons.
46 : */
47 : #define STD_FUZZ_FACTOR 1.01
48 :
49 : static List *translate_sub_tlist(List *tlist, int relid);
50 : static int append_total_cost_compare(const ListCell *a, const ListCell *b);
51 : static int append_startup_cost_compare(const ListCell *a, const ListCell *b);
52 : static List *reparameterize_pathlist_by_child(PlannerInfo *root,
53 : List *pathlist,
54 : RelOptInfo *child_rel);
55 : static bool pathlist_is_reparameterizable_by_child(List *pathlist,
56 : RelOptInfo *child_rel);
57 :
58 :
59 : /*****************************************************************************
60 : * MISC. PATH UTILITIES
61 : *****************************************************************************/
62 :
63 : /*
64 : * compare_path_costs
65 : * Return -1, 0, or +1 according as path1 is cheaper, the same cost,
66 : * or more expensive than path2 for the specified criterion.
67 : */
68 : int
69 838172 : compare_path_costs(Path *path1, Path *path2, CostSelector criterion)
70 : {
71 : /* Number of disabled nodes, if different, trumps all else. */
72 838172 : if (unlikely(path1->disabled_nodes != path2->disabled_nodes))
73 : {
74 2236 : if (path1->disabled_nodes < path2->disabled_nodes)
75 2236 : return -1;
76 : else
77 0 : return +1;
78 : }
79 :
80 835936 : if (criterion == STARTUP_COST)
81 : {
82 426928 : if (path1->startup_cost < path2->startup_cost)
83 254246 : return -1;
84 172682 : if (path1->startup_cost > path2->startup_cost)
85 81462 : return +1;
86 :
87 : /*
88 : * If paths have the same startup cost (not at all unlikely), order
89 : * them by total cost.
90 : */
91 91220 : if (path1->total_cost < path2->total_cost)
92 41998 : return -1;
93 49222 : if (path1->total_cost > path2->total_cost)
94 4442 : return +1;
95 : }
96 : else
97 : {
98 409008 : if (path1->total_cost < path2->total_cost)
99 383058 : return -1;
100 25950 : if (path1->total_cost > path2->total_cost)
101 1332 : return +1;
102 :
103 : /*
104 : * If paths have the same total cost, order them by startup cost.
105 : */
106 24618 : if (path1->startup_cost < path2->startup_cost)
107 1582 : return -1;
108 23036 : if (path1->startup_cost > path2->startup_cost)
109 144 : return +1;
110 : }
111 67672 : return 0;
112 : }
113 :
114 : /*
115 : * compare_fractional_path_costs
116 : * Return -1, 0, or +1 according as path1 is cheaper, the same cost,
117 : * or more expensive than path2 for fetching the specified fraction
118 : * of the total tuples.
119 : *
120 : * If fraction is <= 0 or > 1, we interpret it as 1, ie, we select the
121 : * path with the cheaper total_cost.
122 : */
123 : int
124 3442 : compare_fractional_path_costs(Path *path1, Path *path2,
125 : double fraction)
126 : {
127 : Cost cost1,
128 : cost2;
129 :
130 : /* Number of disabled nodes, if different, trumps all else. */
131 3442 : if (unlikely(path1->disabled_nodes != path2->disabled_nodes))
132 : {
133 30 : if (path1->disabled_nodes < path2->disabled_nodes)
134 30 : return -1;
135 : else
136 0 : return +1;
137 : }
138 :
139 3412 : if (fraction <= 0.0 || fraction >= 1.0)
140 2290 : return compare_path_costs(path1, path2, TOTAL_COST);
141 1122 : cost1 = path1->startup_cost +
142 1122 : fraction * (path1->total_cost - path1->startup_cost);
143 1122 : cost2 = path2->startup_cost +
144 1122 : fraction * (path2->total_cost - path2->startup_cost);
145 1122 : if (cost1 < cost2)
146 710 : return -1;
147 412 : if (cost1 > cost2)
148 412 : return +1;
149 0 : return 0;
150 : }
151 :
152 : /*
153 : * compare_path_costs_fuzzily
154 : * Compare the costs of two paths to see if either can be said to
155 : * dominate the other.
156 : *
157 : * We use fuzzy comparisons so that add_path() can avoid keeping both of
158 : * a pair of paths that really have insignificantly different cost.
159 : *
160 : * The fuzz_factor argument must be 1.0 plus delta, where delta is the
161 : * fraction of the smaller cost that is considered to be a significant
162 : * difference. For example, fuzz_factor = 1.01 makes the fuzziness limit
163 : * be 1% of the smaller cost.
164 : *
165 : * The two paths are said to have "equal" costs if both startup and total
166 : * costs are fuzzily the same. Path1 is said to be better than path2 if
167 : * it has fuzzily better startup cost and fuzzily no worse total cost,
168 : * or if it has fuzzily better total cost and fuzzily no worse startup cost.
169 : * Path2 is better than path1 if the reverse holds. Finally, if one path
170 : * is fuzzily better than the other on startup cost and fuzzily worse on
171 : * total cost, we just say that their costs are "different", since neither
172 : * dominates the other across the whole performance spectrum.
173 : *
174 : * This function also enforces a policy rule that paths for which the relevant
175 : * one of parent->consider_startup and parent->consider_param_startup is false
176 : * cannot survive comparisons solely on the grounds of good startup cost, so
177 : * we never return COSTS_DIFFERENT when that is true for the total-cost loser.
178 : * (But if total costs are fuzzily equal, we compare startup costs anyway,
179 : * in hopes of eliminating one path or the other.)
180 : */
181 : static PathCostComparison
182 3561810 : compare_path_costs_fuzzily(Path *path1, Path *path2, double fuzz_factor)
183 : {
184 : #define CONSIDER_PATH_STARTUP_COST(p) \
185 : ((p)->param_info == NULL ? (p)->parent->consider_startup : (p)->parent->consider_param_startup)
186 :
187 : /* Number of disabled nodes, if different, trumps all else. */
188 3561810 : if (unlikely(path1->disabled_nodes != path2->disabled_nodes))
189 : {
190 29304 : if (path1->disabled_nodes < path2->disabled_nodes)
191 15838 : return COSTS_BETTER1;
192 : else
193 13466 : return COSTS_BETTER2;
194 : }
195 :
196 : /*
197 : * Check total cost first since it's more likely to be different; many
198 : * paths have zero startup cost.
199 : */
200 3532506 : if (path1->total_cost > path2->total_cost * fuzz_factor)
201 : {
202 : /* path1 fuzzily worse on total cost */
203 1823586 : if (CONSIDER_PATH_STARTUP_COST(path1) &&
204 122642 : path2->startup_cost > path1->startup_cost * fuzz_factor)
205 : {
206 : /* ... but path2 fuzzily worse on startup, so DIFFERENT */
207 83894 : return COSTS_DIFFERENT;
208 : }
209 : /* else path2 dominates */
210 1739692 : return COSTS_BETTER2;
211 : }
212 1708920 : if (path2->total_cost > path1->total_cost * fuzz_factor)
213 : {
214 : /* path2 fuzzily worse on total cost */
215 890774 : if (CONSIDER_PATH_STARTUP_COST(path2) &&
216 52344 : path1->startup_cost > path2->startup_cost * fuzz_factor)
217 : {
218 : /* ... but path1 fuzzily worse on startup, so DIFFERENT */
219 34862 : return COSTS_DIFFERENT;
220 : }
221 : /* else path1 dominates */
222 855912 : return COSTS_BETTER1;
223 : }
224 : /* fuzzily the same on total cost ... */
225 818146 : if (path1->startup_cost > path2->startup_cost * fuzz_factor)
226 : {
227 : /* ... but path1 fuzzily worse on startup, so path2 wins */
228 319472 : return COSTS_BETTER2;
229 : }
230 498674 : if (path2->startup_cost > path1->startup_cost * fuzz_factor)
231 : {
232 : /* ... but path2 fuzzily worse on startup, so path1 wins */
233 54622 : return COSTS_BETTER1;
234 : }
235 : /* fuzzily the same on both costs */
236 444052 : return COSTS_EQUAL;
237 :
238 : #undef CONSIDER_PATH_STARTUP_COST
239 : }
240 :
241 : /*
242 : * set_cheapest
243 : * Find the minimum-cost paths from among a relation's paths,
244 : * and save them in the rel's cheapest-path fields.
245 : *
246 : * cheapest_total_path is normally the cheapest-total-cost unparameterized
247 : * path; but if there are no unparameterized paths, we assign it to be the
248 : * best (cheapest least-parameterized) parameterized path. However, only
249 : * unparameterized paths are considered candidates for cheapest_startup_path,
250 : * so that will be NULL if there are no unparameterized paths.
251 : *
252 : * The cheapest_parameterized_paths list collects all parameterized paths
253 : * that have survived the add_path() tournament for this relation. (Since
254 : * add_path ignores pathkeys for a parameterized path, these will be paths
255 : * that have best cost or best row count for their parameterization. We
256 : * may also have both a parallel-safe and a non-parallel-safe path in some
257 : * cases for the same parameterization in some cases, but this should be
258 : * relatively rare since, most typically, all paths for the same relation
259 : * will be parallel-safe or none of them will.)
260 : *
261 : * cheapest_parameterized_paths always includes the cheapest-total
262 : * unparameterized path, too, if there is one; the users of that list find
263 : * it more convenient if that's included.
264 : *
265 : * This is normally called only after we've finished constructing the path
266 : * list for the rel node.
267 : */
268 : void
269 1960602 : set_cheapest(RelOptInfo *parent_rel)
270 : {
271 : Path *cheapest_startup_path;
272 : Path *cheapest_total_path;
273 : Path *best_param_path;
274 : List *parameterized_paths;
275 : ListCell *p;
276 :
277 : Assert(IsA(parent_rel, RelOptInfo));
278 :
279 1960602 : if (parent_rel->pathlist == NIL)
280 0 : elog(ERROR, "could not devise a query plan for the given query");
281 :
282 1960602 : cheapest_startup_path = cheapest_total_path = best_param_path = NULL;
283 1960602 : parameterized_paths = NIL;
284 :
285 4362282 : foreach(p, parent_rel->pathlist)
286 : {
287 2401680 : Path *path = (Path *) lfirst(p);
288 : int cmp;
289 :
290 2401680 : if (path->param_info)
291 : {
292 : /* Parameterized path, so add it to parameterized_paths */
293 118776 : parameterized_paths = lappend(parameterized_paths, path);
294 :
295 : /*
296 : * If we have an unparameterized cheapest-total, we no longer care
297 : * about finding the best parameterized path, so move on.
298 : */
299 118776 : if (cheapest_total_path)
300 23462 : continue;
301 :
302 : /*
303 : * Otherwise, track the best parameterized path, which is the one
304 : * with least total cost among those of the minimum
305 : * parameterization.
306 : */
307 95314 : if (best_param_path == NULL)
308 87980 : best_param_path = path;
309 : else
310 : {
311 7334 : switch (bms_subset_compare(PATH_REQ_OUTER(path),
312 7334 : PATH_REQ_OUTER(best_param_path)))
313 : {
314 54 : case BMS_EQUAL:
315 : /* keep the cheaper one */
316 54 : if (compare_path_costs(path, best_param_path,
317 : TOTAL_COST) < 0)
318 0 : best_param_path = path;
319 54 : break;
320 346 : case BMS_SUBSET1:
321 : /* new path is less-parameterized */
322 346 : best_param_path = path;
323 346 : break;
324 4 : case BMS_SUBSET2:
325 : /* old path is less-parameterized, keep it */
326 4 : break;
327 6930 : case BMS_DIFFERENT:
328 :
329 : /*
330 : * This means that neither path has the least possible
331 : * parameterization for the rel. We'll sit on the old
332 : * path until something better comes along.
333 : */
334 6930 : break;
335 : }
336 95314 : }
337 : }
338 : else
339 : {
340 : /* Unparameterized path, so consider it for cheapest slots */
341 2282904 : if (cheapest_total_path == NULL)
342 : {
343 1949080 : cheapest_startup_path = cheapest_total_path = path;
344 1949080 : continue;
345 : }
346 :
347 : /*
348 : * If we find two paths of identical costs, try to keep the
349 : * better-sorted one. The paths might have unrelated sort
350 : * orderings, in which case we can only guess which might be
351 : * better to keep, but if one is superior then we definitely
352 : * should keep that one.
353 : */
354 333824 : cmp = compare_path_costs(cheapest_startup_path, path, STARTUP_COST);
355 333824 : if (cmp > 0 ||
356 270 : (cmp == 0 &&
357 270 : compare_pathkeys(cheapest_startup_path->pathkeys,
358 : path->pathkeys) == PATHKEYS_BETTER2))
359 64356 : cheapest_startup_path = path;
360 :
361 333824 : cmp = compare_path_costs(cheapest_total_path, path, TOTAL_COST);
362 333824 : if (cmp > 0 ||
363 0 : (cmp == 0 &&
364 0 : compare_pathkeys(cheapest_total_path->pathkeys,
365 : path->pathkeys) == PATHKEYS_BETTER2))
366 0 : cheapest_total_path = path;
367 : }
368 : }
369 :
370 : /* Add cheapest unparameterized path, if any, to parameterized_paths */
371 1960602 : if (cheapest_total_path)
372 1949080 : parameterized_paths = lcons(cheapest_total_path, parameterized_paths);
373 :
374 : /*
375 : * If there is no unparameterized path, use the best parameterized path as
376 : * cheapest_total_path (but not as cheapest_startup_path).
377 : */
378 1960602 : if (cheapest_total_path == NULL)
379 11522 : cheapest_total_path = best_param_path;
380 : Assert(cheapest_total_path != NULL);
381 :
382 1960602 : parent_rel->cheapest_startup_path = cheapest_startup_path;
383 1960602 : parent_rel->cheapest_total_path = cheapest_total_path;
384 1960602 : parent_rel->cheapest_unique_path = NULL; /* computed only if needed */
385 1960602 : parent_rel->cheapest_parameterized_paths = parameterized_paths;
386 1960602 : }
387 :
388 : /*
389 : * add_path
390 : * Consider a potential implementation path for the specified parent rel,
391 : * and add it to the rel's pathlist if it is worthy of consideration.
392 : *
393 : * A path is worthy if it has a better sort order (better pathkeys) or
394 : * cheaper cost (as defined below), or generates fewer rows, than any
395 : * existing path that has the same or superset parameterization rels. We
396 : * also consider parallel-safe paths more worthy than others.
397 : *
398 : * Cheaper cost can mean either a cheaper total cost or a cheaper startup
399 : * cost; if one path is cheaper in one of these aspects and another is
400 : * cheaper in the other, we keep both. However, when some path type is
401 : * disabled (e.g. due to enable_seqscan=false), the number of times that
402 : * a disabled path type is used is considered to be a higher-order
403 : * component of the cost. Hence, if path A uses no disabled path type,
404 : * and path B uses 1 or more disabled path types, A is cheaper, no matter
405 : * what we estimate for the startup and total costs. The startup and total
406 : * cost essentially act as a tiebreak when comparing paths that use equal
407 : * numbers of disabled path nodes; but in practice this tiebreak is almost
408 : * always used, since normally no path types are disabled.
409 : *
410 : * In addition to possibly adding new_path, we also remove from the rel's
411 : * pathlist any old paths that are dominated by new_path --- that is,
412 : * new_path is cheaper, at least as well ordered, generates no more rows,
413 : * requires no outer rels not required by the old path, and is no less
414 : * parallel-safe.
415 : *
416 : * In most cases, a path with a superset parameterization will generate
417 : * fewer rows (since it has more join clauses to apply), so that those two
418 : * figures of merit move in opposite directions; this means that a path of
419 : * one parameterization can seldom dominate a path of another. But such
420 : * cases do arise, so we make the full set of checks anyway.
421 : *
422 : * There are two policy decisions embedded in this function, along with
423 : * its sibling add_path_precheck. First, we treat all parameterized paths
424 : * as having NIL pathkeys, so that they cannot win comparisons on the
425 : * basis of sort order. This is to reduce the number of parameterized
426 : * paths that are kept; see discussion in src/backend/optimizer/README.
427 : *
428 : * Second, we only consider cheap startup cost to be interesting if
429 : * parent_rel->consider_startup is true for an unparameterized path, or
430 : * parent_rel->consider_param_startup is true for a parameterized one.
431 : * Again, this allows discarding useless paths sooner.
432 : *
433 : * The pathlist is kept sorted by disabled_nodes and then by total_cost,
434 : * with cheaper paths at the front. Within this routine, that's simply a
435 : * speed hack: doing it that way makes it more likely that we will reject
436 : * an inferior path after a few comparisons, rather than many comparisons.
437 : * However, add_path_precheck relies on this ordering to exit early
438 : * when possible.
439 : *
440 : * NOTE: discarded Path objects are immediately pfree'd to reduce planner
441 : * memory consumption. We dare not try to free the substructure of a Path,
442 : * since much of it may be shared with other Paths or the query tree itself;
443 : * but just recycling discarded Path nodes is a very useful savings in
444 : * a large join tree. We can recycle the List nodes of pathlist, too.
445 : *
446 : * As noted in optimizer/README, deleting a previously-accepted Path is
447 : * safe because we know that Paths of this rel cannot yet be referenced
448 : * from any other rel, such as a higher-level join. However, in some cases
449 : * it is possible that a Path is referenced by another Path for its own
450 : * rel; we must not delete such a Path, even if it is dominated by the new
451 : * Path. Currently this occurs only for IndexPath objects, which may be
452 : * referenced as children of BitmapHeapPaths as well as being paths in
453 : * their own right. Hence, we don't pfree IndexPaths when rejecting them.
454 : *
455 : * 'parent_rel' is the relation entry to which the path corresponds.
456 : * 'new_path' is a potential path for parent_rel.
457 : *
458 : * Returns nothing, but modifies parent_rel->pathlist.
459 : */
460 : void
461 3793922 : add_path(RelOptInfo *parent_rel, Path *new_path)
462 : {
463 3793922 : bool accept_new = true; /* unless we find a superior old path */
464 3793922 : int insert_at = 0; /* where to insert new item */
465 : List *new_path_pathkeys;
466 : ListCell *p1;
467 :
468 : /*
469 : * This is a convenient place to check for query cancel --- no part of the
470 : * planner goes very long without calling add_path().
471 : */
472 3793922 : CHECK_FOR_INTERRUPTS();
473 :
474 : /* Pretend parameterized paths have no pathkeys, per comment above */
475 3793922 : new_path_pathkeys = new_path->param_info ? NIL : new_path->pathkeys;
476 :
477 : /*
478 : * Loop to check proposed new path against old paths. Note it is possible
479 : * for more than one old path to be tossed out because new_path dominates
480 : * it.
481 : */
482 5732134 : foreach(p1, parent_rel->pathlist)
483 : {
484 3274596 : Path *old_path = (Path *) lfirst(p1);
485 3274596 : bool remove_old = false; /* unless new proves superior */
486 : PathCostComparison costcmp;
487 : PathKeysComparison keyscmp;
488 : BMS_Comparison outercmp;
489 :
490 : /*
491 : * Do a fuzzy cost comparison with standard fuzziness limit.
492 : */
493 3274596 : costcmp = compare_path_costs_fuzzily(new_path, old_path,
494 : STD_FUZZ_FACTOR);
495 :
496 : /*
497 : * If the two paths compare differently for startup and total cost,
498 : * then we want to keep both, and we can skip comparing pathkeys and
499 : * required_outer rels. If they compare the same, proceed with the
500 : * other comparisons. Row count is checked last. (We make the tests
501 : * in this order because the cost comparison is most likely to turn
502 : * out "different", and the pathkeys comparison next most likely. As
503 : * explained above, row count very seldom makes a difference, so even
504 : * though it's cheap to compare there's not much point in checking it
505 : * earlier.)
506 : */
507 3274596 : if (costcmp != COSTS_DIFFERENT)
508 : {
509 : /* Similarly check to see if either dominates on pathkeys */
510 : List *old_path_pathkeys;
511 :
512 3155870 : old_path_pathkeys = old_path->param_info ? NIL : old_path->pathkeys;
513 3155870 : keyscmp = compare_pathkeys(new_path_pathkeys,
514 : old_path_pathkeys);
515 3155870 : if (keyscmp != PATHKEYS_DIFFERENT)
516 : {
517 3010708 : switch (costcmp)
518 : {
519 309240 : case COSTS_EQUAL:
520 309240 : outercmp = bms_subset_compare(PATH_REQ_OUTER(new_path),
521 309240 : PATH_REQ_OUTER(old_path));
522 309240 : if (keyscmp == PATHKEYS_BETTER1)
523 : {
524 3010 : if ((outercmp == BMS_EQUAL ||
525 3010 : outercmp == BMS_SUBSET1) &&
526 3010 : new_path->rows <= old_path->rows &&
527 3002 : new_path->parallel_safe >= old_path->parallel_safe)
528 3002 : remove_old = true; /* new dominates old */
529 : }
530 306230 : else if (keyscmp == PATHKEYS_BETTER2)
531 : {
532 14660 : if ((outercmp == BMS_EQUAL ||
533 14660 : outercmp == BMS_SUBSET2) &&
534 14660 : new_path->rows >= old_path->rows &&
535 14660 : new_path->parallel_safe <= old_path->parallel_safe)
536 14660 : accept_new = false; /* old dominates new */
537 : }
538 : else /* keyscmp == PATHKEYS_EQUAL */
539 : {
540 291570 : if (outercmp == BMS_EQUAL)
541 : {
542 : /*
543 : * Same pathkeys and outer rels, and fuzzily
544 : * the same cost, so keep just one; to decide
545 : * which, first check parallel-safety, then
546 : * rows, then do a fuzzy cost comparison with
547 : * very small fuzz limit. (We used to do an
548 : * exact cost comparison, but that results in
549 : * annoying platform-specific plan variations
550 : * due to roundoff in the cost estimates.) If
551 : * things are still tied, arbitrarily keep
552 : * only the old path. Notice that we will
553 : * keep only the old path even if the
554 : * less-fuzzy comparison decides the startup
555 : * and total costs compare differently.
556 : */
557 287594 : if (new_path->parallel_safe >
558 287594 : old_path->parallel_safe)
559 42 : remove_old = true; /* new dominates old */
560 287552 : else if (new_path->parallel_safe <
561 287552 : old_path->parallel_safe)
562 54 : accept_new = false; /* old dominates new */
563 287498 : else if (new_path->rows < old_path->rows)
564 4 : remove_old = true; /* new dominates old */
565 287494 : else if (new_path->rows > old_path->rows)
566 280 : accept_new = false; /* old dominates new */
567 287214 : else if (compare_path_costs_fuzzily(new_path,
568 : old_path,
569 : 1.0000000001) == COSTS_BETTER1)
570 12964 : remove_old = true; /* new dominates old */
571 : else
572 274250 : accept_new = false; /* old equals or
573 : * dominates new */
574 : }
575 3976 : else if (outercmp == BMS_SUBSET1 &&
576 1162 : new_path->rows <= old_path->rows &&
577 1158 : new_path->parallel_safe >= old_path->parallel_safe)
578 1158 : remove_old = true; /* new dominates old */
579 2818 : else if (outercmp == BMS_SUBSET2 &&
580 2576 : new_path->rows >= old_path->rows &&
581 2562 : new_path->parallel_safe <= old_path->parallel_safe)
582 2562 : accept_new = false; /* old dominates new */
583 : /* else different parameterizations, keep both */
584 : }
585 309240 : break;
586 891936 : case COSTS_BETTER1:
587 891936 : if (keyscmp != PATHKEYS_BETTER2)
588 : {
589 630016 : outercmp = bms_subset_compare(PATH_REQ_OUTER(new_path),
590 630016 : PATH_REQ_OUTER(old_path));
591 630016 : if ((outercmp == BMS_EQUAL ||
592 540402 : outercmp == BMS_SUBSET1) &&
593 540402 : new_path->rows <= old_path->rows &&
594 535660 : new_path->parallel_safe >= old_path->parallel_safe)
595 533178 : remove_old = true; /* new dominates old */
596 : }
597 891936 : break;
598 1809532 : case COSTS_BETTER2:
599 1809532 : if (keyscmp != PATHKEYS_BETTER1)
600 : {
601 1171566 : outercmp = bms_subset_compare(PATH_REQ_OUTER(new_path),
602 1171566 : PATH_REQ_OUTER(old_path));
603 1171566 : if ((outercmp == BMS_EQUAL ||
604 1098516 : outercmp == BMS_SUBSET2) &&
605 1098516 : new_path->rows >= old_path->rows &&
606 1046286 : new_path->parallel_safe <= old_path->parallel_safe)
607 1044578 : accept_new = false; /* old dominates new */
608 : }
609 1809532 : break;
610 0 : case COSTS_DIFFERENT:
611 :
612 : /*
613 : * can't get here, but keep this case to keep compiler
614 : * quiet
615 : */
616 0 : break;
617 : }
618 263888 : }
619 : }
620 :
621 : /*
622 : * Remove current element from pathlist if dominated by new.
623 : */
624 3274596 : if (remove_old)
625 : {
626 550348 : parent_rel->pathlist = foreach_delete_current(parent_rel->pathlist,
627 : p1);
628 :
629 : /*
630 : * Delete the data pointed-to by the deleted cell, if possible
631 : */
632 550348 : if (!IsA(old_path, IndexPath))
633 531714 : pfree(old_path);
634 : }
635 : else
636 : {
637 : /*
638 : * new belongs after this old path if it has more disabled nodes
639 : * or if it has the same number of nodes but a greater total cost
640 : */
641 2724248 : if (new_path->disabled_nodes > old_path->disabled_nodes ||
642 2710782 : (new_path->disabled_nodes == old_path->disabled_nodes &&
643 2709980 : new_path->total_cost >= old_path->total_cost))
644 2269904 : insert_at = foreach_current_index(p1) + 1;
645 : }
646 :
647 : /*
648 : * If we found an old path that dominates new_path, we can quit
649 : * scanning the pathlist; we will not add new_path, and we assume
650 : * new_path cannot dominate any other elements of the pathlist.
651 : */
652 3274596 : if (!accept_new)
653 1336384 : break;
654 : }
655 :
656 3793922 : if (accept_new)
657 : {
658 : /* Accept the new path: insert it at proper place in pathlist */
659 2457538 : parent_rel->pathlist =
660 2457538 : list_insert_nth(parent_rel->pathlist, insert_at, new_path);
661 : }
662 : else
663 : {
664 : /* Reject and recycle the new path */
665 1336384 : if (!IsA(new_path, IndexPath))
666 1253802 : pfree(new_path);
667 : }
668 3793922 : }
669 :
670 : /*
671 : * add_path_precheck
672 : * Check whether a proposed new path could possibly get accepted.
673 : * We assume we know the path's pathkeys and parameterization accurately,
674 : * and have lower bounds for its costs.
675 : *
676 : * Note that we do not know the path's rowcount, since getting an estimate for
677 : * that is too expensive to do before prechecking. We assume here that paths
678 : * of a superset parameterization will generate fewer rows; if that holds,
679 : * then paths with different parameterizations cannot dominate each other
680 : * and so we can simply ignore existing paths of another parameterization.
681 : * (In the infrequent cases where that rule of thumb fails, add_path will
682 : * get rid of the inferior path.)
683 : *
684 : * At the time this is called, we haven't actually built a Path structure,
685 : * so the required information has to be passed piecemeal.
686 : */
687 : bool
688 3804068 : add_path_precheck(RelOptInfo *parent_rel, int disabled_nodes,
689 : Cost startup_cost, Cost total_cost,
690 : List *pathkeys, Relids required_outer)
691 : {
692 : List *new_path_pathkeys;
693 : bool consider_startup;
694 : ListCell *p1;
695 :
696 : /* Pretend parameterized paths have no pathkeys, per add_path policy */
697 3804068 : new_path_pathkeys = required_outer ? NIL : pathkeys;
698 :
699 : /* Decide whether new path's startup cost is interesting */
700 3804068 : consider_startup = required_outer ? parent_rel->consider_param_startup : parent_rel->consider_startup;
701 :
702 4955492 : foreach(p1, parent_rel->pathlist)
703 : {
704 4699018 : Path *old_path = (Path *) lfirst(p1);
705 : PathKeysComparison keyscmp;
706 :
707 : /*
708 : * Since the pathlist is sorted by disabled_nodes and then by
709 : * total_cost, we can stop looking once we reach a path with more
710 : * disabled nodes, or the same number of disabled nodes plus a
711 : * total_cost larger than the new path's.
712 : */
713 4699018 : if (unlikely(old_path->disabled_nodes != disabled_nodes))
714 : {
715 11280 : if (disabled_nodes < old_path->disabled_nodes)
716 298 : break;
717 : }
718 4687738 : else if (total_cost <= old_path->total_cost * STD_FUZZ_FACTOR)
719 1372496 : break;
720 :
721 : /*
722 : * We are looking for an old_path with the same parameterization (and
723 : * by assumption the same rowcount) that dominates the new path on
724 : * pathkeys as well as both cost metrics. If we find one, we can
725 : * reject the new path.
726 : *
727 : * Cost comparisons here should match compare_path_costs_fuzzily.
728 : */
729 : /* new path can win on startup cost only if consider_startup */
730 3326224 : if (startup_cost > old_path->startup_cost * STD_FUZZ_FACTOR ||
731 1590206 : !consider_startup)
732 : {
733 : /* new path loses on cost, so check pathkeys... */
734 : List *old_path_pathkeys;
735 :
736 3228758 : old_path_pathkeys = old_path->param_info ? NIL : old_path->pathkeys;
737 3228758 : keyscmp = compare_pathkeys(new_path_pathkeys,
738 : old_path_pathkeys);
739 3228758 : if (keyscmp == PATHKEYS_EQUAL ||
740 : keyscmp == PATHKEYS_BETTER2)
741 : {
742 : /* new path does not win on pathkeys... */
743 2225694 : if (bms_equal(required_outer, PATH_REQ_OUTER(old_path)))
744 : {
745 : /* Found an old path that dominates the new one */
746 2174800 : return false;
747 : }
748 : }
749 : }
750 : }
751 :
752 1629268 : return true;
753 : }
754 :
755 : /*
756 : * add_partial_path
757 : * Like add_path, our goal here is to consider whether a path is worthy
758 : * of being kept around, but the considerations here are a bit different.
759 : * A partial path is one which can be executed in any number of workers in
760 : * parallel such that each worker will generate a subset of the path's
761 : * overall result.
762 : *
763 : * As in add_path, the partial_pathlist is kept sorted with the cheapest
764 : * total path in front. This is depended on by multiple places, which
765 : * just take the front entry as the cheapest path without searching.
766 : *
767 : * We don't generate parameterized partial paths for several reasons. Most
768 : * importantly, they're not safe to execute, because there's nothing to
769 : * make sure that a parallel scan within the parameterized portion of the
770 : * plan is running with the same value in every worker at the same time.
771 : * Fortunately, it seems unlikely to be worthwhile anyway, because having
772 : * each worker scan the entire outer relation and a subset of the inner
773 : * relation will generally be a terrible plan. The inner (parameterized)
774 : * side of the plan will be small anyway. There could be rare cases where
775 : * this wins big - e.g. if join order constraints put a 1-row relation on
776 : * the outer side of the topmost join with a parameterized plan on the inner
777 : * side - but we'll have to be content not to handle such cases until
778 : * somebody builds an executor infrastructure that can cope with them.
779 : *
780 : * Because we don't consider parameterized paths here, we also don't
781 : * need to consider the row counts as a measure of quality: every path will
782 : * produce the same number of rows. Neither do we need to consider startup
783 : * costs: parallelism is only used for plans that will be run to completion.
784 : * Therefore, this routine is much simpler than add_path: it needs to
785 : * consider only disabled nodes, pathkeys and total cost.
786 : *
787 : * As with add_path, we pfree paths that are found to be dominated by
788 : * another partial path; this requires that there be no other references to
789 : * such paths yet. Hence, GatherPaths must not be created for a rel until
790 : * we're done creating all partial paths for it. Unlike add_path, we don't
791 : * take an exception for IndexPaths as partial index paths won't be
792 : * referenced by partial BitmapHeapPaths.
793 : */
794 : void
795 98078 : add_partial_path(RelOptInfo *parent_rel, Path *new_path)
796 : {
797 98078 : bool accept_new = true; /* unless we find a superior old path */
798 98078 : int insert_at = 0; /* where to insert new item */
799 : ListCell *p1;
800 :
801 : /* Check for query cancel. */
802 98078 : CHECK_FOR_INTERRUPTS();
803 :
804 : /* Path to be added must be parallel safe. */
805 : Assert(new_path->parallel_safe);
806 :
807 : /* Relation should be OK for parallelism, too. */
808 : Assert(parent_rel->consider_parallel);
809 :
810 : /*
811 : * As in add_path, throw out any paths which are dominated by the new
812 : * path, but throw out the new path if some existing path dominates it.
813 : */
814 131972 : foreach(p1, parent_rel->partial_pathlist)
815 : {
816 52292 : Path *old_path = (Path *) lfirst(p1);
817 52292 : bool remove_old = false; /* unless new proves superior */
818 : PathKeysComparison keyscmp;
819 :
820 : /* Compare pathkeys. */
821 52292 : keyscmp = compare_pathkeys(new_path->pathkeys, old_path->pathkeys);
822 :
823 : /* Unless pathkeys are incompatible, keep just one of the two paths. */
824 52292 : if (keyscmp != PATHKEYS_DIFFERENT)
825 : {
826 52094 : if (unlikely(new_path->disabled_nodes != old_path->disabled_nodes))
827 : {
828 1452 : if (new_path->disabled_nodes > old_path->disabled_nodes)
829 956 : accept_new = false;
830 : else
831 496 : remove_old = true;
832 : }
833 50642 : else if (new_path->total_cost > old_path->total_cost
834 50642 : * STD_FUZZ_FACTOR)
835 : {
836 : /* New path costs more; keep it only if pathkeys are better. */
837 17544 : if (keyscmp != PATHKEYS_BETTER1)
838 9002 : accept_new = false;
839 : }
840 33098 : else if (old_path->total_cost > new_path->total_cost
841 33098 : * STD_FUZZ_FACTOR)
842 : {
843 : /* Old path costs more; keep it only if pathkeys are better. */
844 24244 : if (keyscmp != PATHKEYS_BETTER2)
845 12474 : remove_old = true;
846 : }
847 8854 : else if (keyscmp == PATHKEYS_BETTER1)
848 : {
849 : /* Costs are about the same, new path has better pathkeys. */
850 0 : remove_old = true;
851 : }
852 8854 : else if (keyscmp == PATHKEYS_BETTER2)
853 : {
854 : /* Costs are about the same, old path has better pathkeys. */
855 1740 : accept_new = false;
856 : }
857 7114 : else if (old_path->total_cost > new_path->total_cost * 1.0000000001)
858 : {
859 : /* Pathkeys are the same, and the old path costs more. */
860 414 : remove_old = true;
861 : }
862 : else
863 : {
864 : /*
865 : * Pathkeys are the same, and new path isn't materially
866 : * cheaper.
867 : */
868 6700 : accept_new = false;
869 : }
870 : }
871 :
872 : /*
873 : * Remove current element from partial_pathlist if dominated by new.
874 : */
875 52292 : if (remove_old)
876 : {
877 13384 : parent_rel->partial_pathlist =
878 13384 : foreach_delete_current(parent_rel->partial_pathlist, p1);
879 13384 : pfree(old_path);
880 : }
881 : else
882 : {
883 : /* new belongs after this old path if it has cost >= old's */
884 38908 : if (new_path->total_cost >= old_path->total_cost)
885 26152 : insert_at = foreach_current_index(p1) + 1;
886 : }
887 :
888 : /*
889 : * If we found an old path that dominates new_path, we can quit
890 : * scanning the partial_pathlist; we will not add new_path, and we
891 : * assume new_path cannot dominate any later path.
892 : */
893 52292 : if (!accept_new)
894 18398 : break;
895 : }
896 :
897 98078 : if (accept_new)
898 : {
899 : /* Accept the new path: insert it at proper place */
900 79680 : parent_rel->partial_pathlist =
901 79680 : list_insert_nth(parent_rel->partial_pathlist, insert_at, new_path);
902 : }
903 : else
904 : {
905 : /* Reject and recycle the new path */
906 18398 : pfree(new_path);
907 : }
908 98078 : }
909 :
910 : /*
911 : * add_partial_path_precheck
912 : * Check whether a proposed new partial path could possibly get accepted.
913 : *
914 : * Unlike add_path_precheck, we can ignore startup cost and parameterization,
915 : * since they don't matter for partial paths (see add_partial_path). But
916 : * we do want to make sure we don't add a partial path if there's already
917 : * a complete path that dominates it, since in that case the proposed path
918 : * is surely a loser.
919 : */
920 : bool
921 78932 : add_partial_path_precheck(RelOptInfo *parent_rel, int disabled_nodes,
922 : Cost total_cost, List *pathkeys)
923 : {
924 : ListCell *p1;
925 :
926 : /*
927 : * Our goal here is twofold. First, we want to find out whether this path
928 : * is clearly inferior to some existing partial path. If so, we want to
929 : * reject it immediately. Second, we want to find out whether this path
930 : * is clearly superior to some existing partial path -- at least, modulo
931 : * final cost computations. If so, we definitely want to consider it.
932 : *
933 : * Unlike add_path(), we always compare pathkeys here. This is because we
934 : * expect partial_pathlist to be very short, and getting a definitive
935 : * answer at this stage avoids the need to call add_path_precheck.
936 : */
937 108628 : foreach(p1, parent_rel->partial_pathlist)
938 : {
939 88126 : Path *old_path = (Path *) lfirst(p1);
940 : PathKeysComparison keyscmp;
941 :
942 88126 : keyscmp = compare_pathkeys(pathkeys, old_path->pathkeys);
943 88126 : if (keyscmp != PATHKEYS_DIFFERENT)
944 : {
945 87934 : if (total_cost > old_path->total_cost * STD_FUZZ_FACTOR &&
946 : keyscmp != PATHKEYS_BETTER1)
947 58430 : return false;
948 43430 : if (old_path->total_cost > total_cost * STD_FUZZ_FACTOR &&
949 : keyscmp != PATHKEYS_BETTER2)
950 13926 : return true;
951 : }
952 : }
953 :
954 : /*
955 : * This path is neither clearly inferior to an existing partial path nor
956 : * clearly good enough that it might replace one. Compare it to
957 : * non-parallel plans. If it loses even before accounting for the cost of
958 : * the Gather node, we should definitely reject it.
959 : *
960 : * Note that we pass the total_cost to add_path_precheck twice. This is
961 : * because it's never advantageous to consider the startup cost of a
962 : * partial path; the resulting plans, if run in parallel, will be run to
963 : * completion.
964 : */
965 20502 : if (!add_path_precheck(parent_rel, disabled_nodes, total_cost, total_cost,
966 : pathkeys, NULL))
967 2062 : return false;
968 :
969 18440 : return true;
970 : }
971 :
972 :
973 : /*****************************************************************************
974 : * PATH NODE CREATION ROUTINES
975 : *****************************************************************************/
976 :
977 : /*
978 : * create_seqscan_path
979 : * Creates a path corresponding to a sequential scan, returning the
980 : * pathnode.
981 : */
982 : Path *
983 387850 : create_seqscan_path(PlannerInfo *root, RelOptInfo *rel,
984 : Relids required_outer, int parallel_workers)
985 : {
986 387850 : Path *pathnode = makeNode(Path);
987 :
988 387850 : pathnode->pathtype = T_SeqScan;
989 387850 : pathnode->parent = rel;
990 387850 : pathnode->pathtarget = rel->reltarget;
991 387850 : pathnode->param_info = get_baserel_parampathinfo(root, rel,
992 : required_outer);
993 387850 : pathnode->parallel_aware = (parallel_workers > 0);
994 387850 : pathnode->parallel_safe = rel->consider_parallel;
995 387850 : pathnode->parallel_workers = parallel_workers;
996 387850 : pathnode->pathkeys = NIL; /* seqscan has unordered result */
997 :
998 387850 : cost_seqscan(pathnode, root, rel, pathnode->param_info);
999 :
1000 387850 : return pathnode;
1001 : }
1002 :
1003 : /*
1004 : * create_samplescan_path
1005 : * Creates a path node for a sampled table scan.
1006 : */
1007 : Path *
1008 300 : create_samplescan_path(PlannerInfo *root, RelOptInfo *rel, Relids required_outer)
1009 : {
1010 300 : Path *pathnode = makeNode(Path);
1011 :
1012 300 : pathnode->pathtype = T_SampleScan;
1013 300 : pathnode->parent = rel;
1014 300 : pathnode->pathtarget = rel->reltarget;
1015 300 : pathnode->param_info = get_baserel_parampathinfo(root, rel,
1016 : required_outer);
1017 300 : pathnode->parallel_aware = false;
1018 300 : pathnode->parallel_safe = rel->consider_parallel;
1019 300 : pathnode->parallel_workers = 0;
1020 300 : pathnode->pathkeys = NIL; /* samplescan has unordered result */
1021 :
1022 300 : cost_samplescan(pathnode, root, rel, pathnode->param_info);
1023 :
1024 300 : return pathnode;
1025 : }
1026 :
1027 : /*
1028 : * create_index_path
1029 : * Creates a path node for an index scan.
1030 : *
1031 : * 'index' is a usable index.
1032 : * 'indexclauses' is a list of IndexClause nodes representing clauses
1033 : * to be enforced as qual conditions in the scan.
1034 : * 'indexorderbys' is a list of bare expressions (no RestrictInfos)
1035 : * to be used as index ordering operators in the scan.
1036 : * 'indexorderbycols' is an integer list of index column numbers (zero based)
1037 : * the ordering operators can be used with.
1038 : * 'pathkeys' describes the ordering of the path.
1039 : * 'indexscandir' is either ForwardScanDirection or BackwardScanDirection.
1040 : * 'indexonly' is true if an index-only scan is wanted.
1041 : * 'required_outer' is the set of outer relids for a parameterized path.
1042 : * 'loop_count' is the number of repetitions of the indexscan to factor into
1043 : * estimates of caching behavior.
1044 : * 'partial_path' is true if constructing a parallel index scan path.
1045 : *
1046 : * Returns the new path node.
1047 : */
1048 : IndexPath *
1049 685644 : create_index_path(PlannerInfo *root,
1050 : IndexOptInfo *index,
1051 : List *indexclauses,
1052 : List *indexorderbys,
1053 : List *indexorderbycols,
1054 : List *pathkeys,
1055 : ScanDirection indexscandir,
1056 : bool indexonly,
1057 : Relids required_outer,
1058 : double loop_count,
1059 : bool partial_path)
1060 : {
1061 685644 : IndexPath *pathnode = makeNode(IndexPath);
1062 685644 : RelOptInfo *rel = index->rel;
1063 :
1064 685644 : pathnode->path.pathtype = indexonly ? T_IndexOnlyScan : T_IndexScan;
1065 685644 : pathnode->path.parent = rel;
1066 685644 : pathnode->path.pathtarget = rel->reltarget;
1067 685644 : pathnode->path.param_info = get_baserel_parampathinfo(root, rel,
1068 : required_outer);
1069 685644 : pathnode->path.parallel_aware = false;
1070 685644 : pathnode->path.parallel_safe = rel->consider_parallel;
1071 685644 : pathnode->path.parallel_workers = 0;
1072 685644 : pathnode->path.pathkeys = pathkeys;
1073 :
1074 685644 : pathnode->indexinfo = index;
1075 685644 : pathnode->indexclauses = indexclauses;
1076 685644 : pathnode->indexorderbys = indexorderbys;
1077 685644 : pathnode->indexorderbycols = indexorderbycols;
1078 685644 : pathnode->indexscandir = indexscandir;
1079 :
1080 685644 : cost_index(pathnode, root, loop_count, partial_path);
1081 :
1082 685644 : return pathnode;
1083 : }
1084 :
1085 : /*
1086 : * create_bitmap_heap_path
1087 : * Creates a path node for a bitmap scan.
1088 : *
1089 : * 'bitmapqual' is a tree of IndexPath, BitmapAndPath, and BitmapOrPath nodes.
1090 : * 'required_outer' is the set of outer relids for a parameterized path.
1091 : * 'loop_count' is the number of repetitions of the indexscan to factor into
1092 : * estimates of caching behavior.
1093 : *
1094 : * loop_count should match the value used when creating the component
1095 : * IndexPaths.
1096 : */
1097 : BitmapHeapPath *
1098 310746 : create_bitmap_heap_path(PlannerInfo *root,
1099 : RelOptInfo *rel,
1100 : Path *bitmapqual,
1101 : Relids required_outer,
1102 : double loop_count,
1103 : int parallel_degree)
1104 : {
1105 310746 : BitmapHeapPath *pathnode = makeNode(BitmapHeapPath);
1106 :
1107 310746 : pathnode->path.pathtype = T_BitmapHeapScan;
1108 310746 : pathnode->path.parent = rel;
1109 310746 : pathnode->path.pathtarget = rel->reltarget;
1110 310746 : pathnode->path.param_info = get_baserel_parampathinfo(root, rel,
1111 : required_outer);
1112 310746 : pathnode->path.parallel_aware = (parallel_degree > 0);
1113 310746 : pathnode->path.parallel_safe = rel->consider_parallel;
1114 310746 : pathnode->path.parallel_workers = parallel_degree;
1115 310746 : pathnode->path.pathkeys = NIL; /* always unordered */
1116 :
1117 310746 : pathnode->bitmapqual = bitmapqual;
1118 :
1119 310746 : cost_bitmap_heap_scan(&pathnode->path, root, rel,
1120 : pathnode->path.param_info,
1121 : bitmapqual, loop_count);
1122 :
1123 310746 : return pathnode;
1124 : }
1125 :
1126 : /*
1127 : * create_bitmap_and_path
1128 : * Creates a path node representing a BitmapAnd.
1129 : */
1130 : BitmapAndPath *
1131 44756 : create_bitmap_and_path(PlannerInfo *root,
1132 : RelOptInfo *rel,
1133 : List *bitmapquals)
1134 : {
1135 44756 : BitmapAndPath *pathnode = makeNode(BitmapAndPath);
1136 44756 : Relids required_outer = NULL;
1137 : ListCell *lc;
1138 :
1139 44756 : pathnode->path.pathtype = T_BitmapAnd;
1140 44756 : pathnode->path.parent = rel;
1141 44756 : pathnode->path.pathtarget = rel->reltarget;
1142 :
1143 : /*
1144 : * Identify the required outer rels as the union of what the child paths
1145 : * depend on. (Alternatively, we could insist that the caller pass this
1146 : * in, but it's more convenient and reliable to compute it here.)
1147 : */
1148 134268 : foreach(lc, bitmapquals)
1149 : {
1150 89512 : Path *bitmapqual = (Path *) lfirst(lc);
1151 :
1152 89512 : required_outer = bms_add_members(required_outer,
1153 89512 : PATH_REQ_OUTER(bitmapqual));
1154 : }
1155 44756 : pathnode->path.param_info = get_baserel_parampathinfo(root, rel,
1156 : required_outer);
1157 :
1158 : /*
1159 : * Currently, a BitmapHeapPath, BitmapAndPath, or BitmapOrPath will be
1160 : * parallel-safe if and only if rel->consider_parallel is set. So, we can
1161 : * set the flag for this path based only on the relation-level flag,
1162 : * without actually iterating over the list of children.
1163 : */
1164 44756 : pathnode->path.parallel_aware = false;
1165 44756 : pathnode->path.parallel_safe = rel->consider_parallel;
1166 44756 : pathnode->path.parallel_workers = 0;
1167 :
1168 44756 : pathnode->path.pathkeys = NIL; /* always unordered */
1169 :
1170 44756 : pathnode->bitmapquals = bitmapquals;
1171 :
1172 : /* this sets bitmapselectivity as well as the regular cost fields: */
1173 44756 : cost_bitmap_and_node(pathnode, root);
1174 :
1175 44756 : return pathnode;
1176 : }
1177 :
1178 : /*
1179 : * create_bitmap_or_path
1180 : * Creates a path node representing a BitmapOr.
1181 : */
1182 : BitmapOrPath *
1183 906 : create_bitmap_or_path(PlannerInfo *root,
1184 : RelOptInfo *rel,
1185 : List *bitmapquals)
1186 : {
1187 906 : BitmapOrPath *pathnode = makeNode(BitmapOrPath);
1188 906 : Relids required_outer = NULL;
1189 : ListCell *lc;
1190 :
1191 906 : pathnode->path.pathtype = T_BitmapOr;
1192 906 : pathnode->path.parent = rel;
1193 906 : pathnode->path.pathtarget = rel->reltarget;
1194 :
1195 : /*
1196 : * Identify the required outer rels as the union of what the child paths
1197 : * depend on. (Alternatively, we could insist that the caller pass this
1198 : * in, but it's more convenient and reliable to compute it here.)
1199 : */
1200 2778 : foreach(lc, bitmapquals)
1201 : {
1202 1872 : Path *bitmapqual = (Path *) lfirst(lc);
1203 :
1204 1872 : required_outer = bms_add_members(required_outer,
1205 1872 : PATH_REQ_OUTER(bitmapqual));
1206 : }
1207 906 : pathnode->path.param_info = get_baserel_parampathinfo(root, rel,
1208 : required_outer);
1209 :
1210 : /*
1211 : * Currently, a BitmapHeapPath, BitmapAndPath, or BitmapOrPath will be
1212 : * parallel-safe if and only if rel->consider_parallel is set. So, we can
1213 : * set the flag for this path based only on the relation-level flag,
1214 : * without actually iterating over the list of children.
1215 : */
1216 906 : pathnode->path.parallel_aware = false;
1217 906 : pathnode->path.parallel_safe = rel->consider_parallel;
1218 906 : pathnode->path.parallel_workers = 0;
1219 :
1220 906 : pathnode->path.pathkeys = NIL; /* always unordered */
1221 :
1222 906 : pathnode->bitmapquals = bitmapquals;
1223 :
1224 : /* this sets bitmapselectivity as well as the regular cost fields: */
1225 906 : cost_bitmap_or_node(pathnode, root);
1226 :
1227 906 : return pathnode;
1228 : }
1229 :
1230 : /*
1231 : * create_tidscan_path
1232 : * Creates a path corresponding to a scan by TID, returning the pathnode.
1233 : */
1234 : TidPath *
1235 792 : create_tidscan_path(PlannerInfo *root, RelOptInfo *rel, List *tidquals,
1236 : Relids required_outer)
1237 : {
1238 792 : TidPath *pathnode = makeNode(TidPath);
1239 :
1240 792 : pathnode->path.pathtype = T_TidScan;
1241 792 : pathnode->path.parent = rel;
1242 792 : pathnode->path.pathtarget = rel->reltarget;
1243 792 : pathnode->path.param_info = get_baserel_parampathinfo(root, rel,
1244 : required_outer);
1245 792 : pathnode->path.parallel_aware = false;
1246 792 : pathnode->path.parallel_safe = rel->consider_parallel;
1247 792 : pathnode->path.parallel_workers = 0;
1248 792 : pathnode->path.pathkeys = NIL; /* always unordered */
1249 :
1250 792 : pathnode->tidquals = tidquals;
1251 :
1252 792 : cost_tidscan(&pathnode->path, root, rel, tidquals,
1253 : pathnode->path.param_info);
1254 :
1255 792 : return pathnode;
1256 : }
1257 :
1258 : /*
1259 : * create_tidrangescan_path
1260 : * Creates a path corresponding to a scan by a range of TIDs, returning
1261 : * the pathnode.
1262 : */
1263 : TidRangePath *
1264 202 : create_tidrangescan_path(PlannerInfo *root, RelOptInfo *rel,
1265 : List *tidrangequals, Relids required_outer)
1266 : {
1267 202 : TidRangePath *pathnode = makeNode(TidRangePath);
1268 :
1269 202 : pathnode->path.pathtype = T_TidRangeScan;
1270 202 : pathnode->path.parent = rel;
1271 202 : pathnode->path.pathtarget = rel->reltarget;
1272 202 : pathnode->path.param_info = get_baserel_parampathinfo(root, rel,
1273 : required_outer);
1274 202 : pathnode->path.parallel_aware = false;
1275 202 : pathnode->path.parallel_safe = rel->consider_parallel;
1276 202 : pathnode->path.parallel_workers = 0;
1277 202 : pathnode->path.pathkeys = NIL; /* always unordered */
1278 :
1279 202 : pathnode->tidrangequals = tidrangequals;
1280 :
1281 202 : cost_tidrangescan(&pathnode->path, root, rel, tidrangequals,
1282 : pathnode->path.param_info);
1283 :
1284 202 : return pathnode;
1285 : }
1286 :
1287 : /*
1288 : * create_append_path
1289 : * Creates a path corresponding to an Append plan, returning the
1290 : * pathnode.
1291 : *
1292 : * Note that we must handle subpaths = NIL, representing a dummy access path.
1293 : * Also, there are callers that pass root = NULL.
1294 : *
1295 : * 'rows', when passed as a non-negative number, will be used to overwrite the
1296 : * returned path's row estimate. Otherwise, the row estimate is calculated
1297 : * by totalling the row estimates from the 'subpaths' list.
1298 : */
1299 : AppendPath *
1300 69202 : create_append_path(PlannerInfo *root,
1301 : RelOptInfo *rel,
1302 : List *subpaths, List *partial_subpaths,
1303 : List *pathkeys, Relids required_outer,
1304 : int parallel_workers, bool parallel_aware,
1305 : double rows)
1306 : {
1307 69202 : AppendPath *pathnode = makeNode(AppendPath);
1308 : ListCell *l;
1309 :
1310 : Assert(!parallel_aware || parallel_workers > 0);
1311 :
1312 69202 : pathnode->path.pathtype = T_Append;
1313 69202 : pathnode->path.parent = rel;
1314 69202 : pathnode->path.pathtarget = rel->reltarget;
1315 :
1316 : /*
1317 : * If this is for a baserel (not a join or non-leaf partition), we prefer
1318 : * to apply get_baserel_parampathinfo to construct a full ParamPathInfo
1319 : * for the path. This supports building a Memoize path atop this path,
1320 : * and if this is a partitioned table the info may be useful for run-time
1321 : * pruning (cf make_partition_pruneinfo()).
1322 : *
1323 : * However, if we don't have "root" then that won't work and we fall back
1324 : * on the simpler get_appendrel_parampathinfo. There's no point in doing
1325 : * the more expensive thing for a dummy path, either.
1326 : */
1327 69202 : if (rel->reloptkind == RELOPT_BASEREL && root && subpaths != NIL)
1328 31482 : pathnode->path.param_info = get_baserel_parampathinfo(root,
1329 : rel,
1330 : required_outer);
1331 : else
1332 37720 : pathnode->path.param_info = get_appendrel_parampathinfo(rel,
1333 : required_outer);
1334 :
1335 69202 : pathnode->path.parallel_aware = parallel_aware;
1336 69202 : pathnode->path.parallel_safe = rel->consider_parallel;
1337 69202 : pathnode->path.parallel_workers = parallel_workers;
1338 69202 : pathnode->path.pathkeys = pathkeys;
1339 :
1340 : /*
1341 : * For parallel append, non-partial paths are sorted by descending total
1342 : * costs. That way, the total time to finish all non-partial paths is
1343 : * minimized. Also, the partial paths are sorted by descending startup
1344 : * costs. There may be some paths that require to do startup work by a
1345 : * single worker. In such case, it's better for workers to choose the
1346 : * expensive ones first, whereas the leader should choose the cheapest
1347 : * startup plan.
1348 : */
1349 69202 : if (pathnode->path.parallel_aware)
1350 : {
1351 : /*
1352 : * We mustn't fiddle with the order of subpaths when the Append has
1353 : * pathkeys. The order they're listed in is critical to keeping the
1354 : * pathkeys valid.
1355 : */
1356 : Assert(pathkeys == NIL);
1357 :
1358 22826 : list_sort(subpaths, append_total_cost_compare);
1359 22826 : list_sort(partial_subpaths, append_startup_cost_compare);
1360 : }
1361 69202 : pathnode->first_partial_path = list_length(subpaths);
1362 69202 : pathnode->subpaths = list_concat(subpaths, partial_subpaths);
1363 :
1364 : /*
1365 : * Apply query-wide LIMIT if known and path is for sole base relation.
1366 : * (Handling this at this low level is a bit klugy.)
1367 : */
1368 69202 : if (root != NULL && bms_equal(rel->relids, root->all_query_rels))
1369 38242 : pathnode->limit_tuples = root->limit_tuples;
1370 : else
1371 30960 : pathnode->limit_tuples = -1.0;
1372 :
1373 227704 : foreach(l, pathnode->subpaths)
1374 : {
1375 158502 : Path *subpath = (Path *) lfirst(l);
1376 :
1377 280306 : pathnode->path.parallel_safe = pathnode->path.parallel_safe &&
1378 121804 : subpath->parallel_safe;
1379 :
1380 : /* All child paths must have same parameterization */
1381 : Assert(bms_equal(PATH_REQ_OUTER(subpath), required_outer));
1382 : }
1383 :
1384 : Assert(!parallel_aware || pathnode->path.parallel_safe);
1385 :
1386 : /*
1387 : * If there's exactly one child path then the output of the Append is
1388 : * necessarily ordered the same as the child's, so we can inherit the
1389 : * child's pathkeys if any, overriding whatever the caller might've said.
1390 : * Furthermore, if the child's parallel awareness matches the Append's,
1391 : * then the Append is a no-op and will be discarded later (in setrefs.c).
1392 : * Then we can inherit the child's size and cost too, effectively charging
1393 : * zero for the Append. Otherwise, we must do the normal costsize
1394 : * calculation.
1395 : */
1396 69202 : if (list_length(pathnode->subpaths) == 1)
1397 : {
1398 21918 : Path *child = (Path *) linitial(pathnode->subpaths);
1399 :
1400 21918 : if (child->parallel_aware == parallel_aware)
1401 : {
1402 21480 : pathnode->path.rows = child->rows;
1403 21480 : pathnode->path.startup_cost = child->startup_cost;
1404 21480 : pathnode->path.total_cost = child->total_cost;
1405 : }
1406 : else
1407 438 : cost_append(pathnode);
1408 : /* Must do this last, else cost_append complains */
1409 21918 : pathnode->path.pathkeys = child->pathkeys;
1410 : }
1411 : else
1412 47284 : cost_append(pathnode);
1413 :
1414 : /* If the caller provided a row estimate, override the computed value. */
1415 69202 : if (rows >= 0)
1416 576 : pathnode->path.rows = rows;
1417 :
1418 69202 : return pathnode;
1419 : }
1420 :
1421 : /*
1422 : * append_total_cost_compare
1423 : * list_sort comparator for sorting append child paths
1424 : * by total_cost descending
1425 : *
1426 : * For equal total costs, we fall back to comparing startup costs; if those
1427 : * are equal too, break ties using bms_compare on the paths' relids.
1428 : * (This is to avoid getting unpredictable results from list_sort.)
1429 : */
1430 : static int
1431 2004 : append_total_cost_compare(const ListCell *a, const ListCell *b)
1432 : {
1433 2004 : Path *path1 = (Path *) lfirst(a);
1434 2004 : Path *path2 = (Path *) lfirst(b);
1435 : int cmp;
1436 :
1437 2004 : cmp = compare_path_costs(path1, path2, TOTAL_COST);
1438 2004 : if (cmp != 0)
1439 1734 : return -cmp;
1440 270 : return bms_compare(path1->parent->relids, path2->parent->relids);
1441 : }
1442 :
1443 : /*
1444 : * append_startup_cost_compare
1445 : * list_sort comparator for sorting append child paths
1446 : * by startup_cost descending
1447 : *
1448 : * For equal startup costs, we fall back to comparing total costs; if those
1449 : * are equal too, break ties using bms_compare on the paths' relids.
1450 : * (This is to avoid getting unpredictable results from list_sort.)
1451 : */
1452 : static int
1453 33518 : append_startup_cost_compare(const ListCell *a, const ListCell *b)
1454 : {
1455 33518 : Path *path1 = (Path *) lfirst(a);
1456 33518 : Path *path2 = (Path *) lfirst(b);
1457 : int cmp;
1458 :
1459 33518 : cmp = compare_path_costs(path1, path2, STARTUP_COST);
1460 33518 : if (cmp != 0)
1461 13220 : return -cmp;
1462 20298 : return bms_compare(path1->parent->relids, path2->parent->relids);
1463 : }
1464 :
1465 : /*
1466 : * create_merge_append_path
1467 : * Creates a path corresponding to a MergeAppend plan, returning the
1468 : * pathnode.
1469 : */
1470 : MergeAppendPath *
1471 4114 : create_merge_append_path(PlannerInfo *root,
1472 : RelOptInfo *rel,
1473 : List *subpaths,
1474 : List *pathkeys,
1475 : Relids required_outer)
1476 : {
1477 4114 : MergeAppendPath *pathnode = makeNode(MergeAppendPath);
1478 : int input_disabled_nodes;
1479 : Cost input_startup_cost;
1480 : Cost input_total_cost;
1481 : ListCell *l;
1482 :
1483 : /*
1484 : * We don't currently support parameterized MergeAppend paths, as
1485 : * explained in the comments for generate_orderedappend_paths.
1486 : */
1487 : Assert(bms_is_empty(rel->lateral_relids) && bms_is_empty(required_outer));
1488 :
1489 4114 : pathnode->path.pathtype = T_MergeAppend;
1490 4114 : pathnode->path.parent = rel;
1491 4114 : pathnode->path.pathtarget = rel->reltarget;
1492 4114 : pathnode->path.param_info = NULL;
1493 4114 : pathnode->path.parallel_aware = false;
1494 4114 : pathnode->path.parallel_safe = rel->consider_parallel;
1495 4114 : pathnode->path.parallel_workers = 0;
1496 4114 : pathnode->path.pathkeys = pathkeys;
1497 4114 : pathnode->subpaths = subpaths;
1498 :
1499 : /*
1500 : * Apply query-wide LIMIT if known and path is for sole base relation.
1501 : * (Handling this at this low level is a bit klugy.)
1502 : */
1503 4114 : if (bms_equal(rel->relids, root->all_query_rels))
1504 2136 : pathnode->limit_tuples = root->limit_tuples;
1505 : else
1506 1978 : pathnode->limit_tuples = -1.0;
1507 :
1508 : /*
1509 : * Add up the sizes and costs of the input paths.
1510 : */
1511 4114 : pathnode->path.rows = 0;
1512 4114 : input_disabled_nodes = 0;
1513 4114 : input_startup_cost = 0;
1514 4114 : input_total_cost = 0;
1515 15336 : foreach(l, subpaths)
1516 : {
1517 11222 : Path *subpath = (Path *) lfirst(l);
1518 :
1519 : /* All child paths should be unparameterized */
1520 : Assert(bms_is_empty(PATH_REQ_OUTER(subpath)));
1521 :
1522 11222 : pathnode->path.rows += subpath->rows;
1523 20026 : pathnode->path.parallel_safe = pathnode->path.parallel_safe &&
1524 8804 : subpath->parallel_safe;
1525 :
1526 11222 : if (pathkeys_contained_in(pathkeys, subpath->pathkeys))
1527 : {
1528 : /* Subpath is adequately ordered, we won't need to sort it */
1529 10924 : input_disabled_nodes += subpath->disabled_nodes;
1530 10924 : input_startup_cost += subpath->startup_cost;
1531 10924 : input_total_cost += subpath->total_cost;
1532 : }
1533 : else
1534 : {
1535 : /* We'll need to insert a Sort node, so include cost for that */
1536 : Path sort_path; /* dummy for result of cost_sort */
1537 :
1538 298 : cost_sort(&sort_path,
1539 : root,
1540 : pathkeys,
1541 : subpath->disabled_nodes,
1542 : subpath->total_cost,
1543 : subpath->rows,
1544 298 : subpath->pathtarget->width,
1545 : 0.0,
1546 : work_mem,
1547 : pathnode->limit_tuples);
1548 298 : input_disabled_nodes += sort_path.disabled_nodes;
1549 298 : input_startup_cost += sort_path.startup_cost;
1550 298 : input_total_cost += sort_path.total_cost;
1551 : }
1552 : }
1553 :
1554 : /*
1555 : * Now we can compute total costs of the MergeAppend. If there's exactly
1556 : * one child path and its parallel awareness matches that of the
1557 : * MergeAppend, then the MergeAppend is a no-op and will be discarded
1558 : * later (in setrefs.c); otherwise we do the normal cost calculation.
1559 : */
1560 4114 : if (list_length(subpaths) == 1 &&
1561 110 : ((Path *) linitial(subpaths))->parallel_aware ==
1562 110 : pathnode->path.parallel_aware)
1563 : {
1564 110 : pathnode->path.disabled_nodes = input_disabled_nodes;
1565 110 : pathnode->path.startup_cost = input_startup_cost;
1566 110 : pathnode->path.total_cost = input_total_cost;
1567 : }
1568 : else
1569 4004 : cost_merge_append(&pathnode->path, root,
1570 : pathkeys, list_length(subpaths),
1571 : input_disabled_nodes,
1572 : input_startup_cost, input_total_cost,
1573 : pathnode->path.rows);
1574 :
1575 4114 : return pathnode;
1576 : }
1577 :
1578 : /*
1579 : * create_group_result_path
1580 : * Creates a path representing a Result-and-nothing-else plan.
1581 : *
1582 : * This is only used for degenerate grouping cases, in which we know we
1583 : * need to produce one result row, possibly filtered by a HAVING qual.
1584 : */
1585 : GroupResultPath *
1586 214844 : create_group_result_path(PlannerInfo *root, RelOptInfo *rel,
1587 : PathTarget *target, List *havingqual)
1588 : {
1589 214844 : GroupResultPath *pathnode = makeNode(GroupResultPath);
1590 :
1591 214844 : pathnode->path.pathtype = T_Result;
1592 214844 : pathnode->path.parent = rel;
1593 214844 : pathnode->path.pathtarget = target;
1594 214844 : pathnode->path.param_info = NULL; /* there are no other rels... */
1595 214844 : pathnode->path.parallel_aware = false;
1596 214844 : pathnode->path.parallel_safe = rel->consider_parallel;
1597 214844 : pathnode->path.parallel_workers = 0;
1598 214844 : pathnode->path.pathkeys = NIL;
1599 214844 : pathnode->quals = havingqual;
1600 :
1601 : /*
1602 : * We can't quite use cost_resultscan() because the quals we want to
1603 : * account for are not baserestrict quals of the rel. Might as well just
1604 : * hack it here.
1605 : */
1606 214844 : pathnode->path.rows = 1;
1607 214844 : pathnode->path.startup_cost = target->cost.startup;
1608 214844 : pathnode->path.total_cost = target->cost.startup +
1609 214844 : cpu_tuple_cost + target->cost.per_tuple;
1610 :
1611 : /*
1612 : * Add cost of qual, if any --- but we ignore its selectivity, since our
1613 : * rowcount estimate should be 1 no matter what the qual is.
1614 : */
1615 214844 : if (havingqual)
1616 : {
1617 : QualCost qual_cost;
1618 :
1619 528 : cost_qual_eval(&qual_cost, havingqual, root);
1620 : /* havingqual is evaluated once at startup */
1621 528 : pathnode->path.startup_cost += qual_cost.startup + qual_cost.per_tuple;
1622 528 : pathnode->path.total_cost += qual_cost.startup + qual_cost.per_tuple;
1623 : }
1624 :
1625 214844 : return pathnode;
1626 : }
1627 :
1628 : /*
1629 : * create_material_path
1630 : * Creates a path corresponding to a Material plan, returning the
1631 : * pathnode.
1632 : */
1633 : MaterialPath *
1634 444174 : create_material_path(RelOptInfo *rel, Path *subpath)
1635 : {
1636 444174 : MaterialPath *pathnode = makeNode(MaterialPath);
1637 :
1638 : Assert(subpath->parent == rel);
1639 :
1640 444174 : pathnode->path.pathtype = T_Material;
1641 444174 : pathnode->path.parent = rel;
1642 444174 : pathnode->path.pathtarget = rel->reltarget;
1643 444174 : pathnode->path.param_info = subpath->param_info;
1644 444174 : pathnode->path.parallel_aware = false;
1645 839024 : pathnode->path.parallel_safe = rel->consider_parallel &&
1646 394850 : subpath->parallel_safe;
1647 444174 : pathnode->path.parallel_workers = subpath->parallel_workers;
1648 444174 : pathnode->path.pathkeys = subpath->pathkeys;
1649 :
1650 444174 : pathnode->subpath = subpath;
1651 :
1652 444174 : cost_material(&pathnode->path,
1653 : subpath->disabled_nodes,
1654 : subpath->startup_cost,
1655 : subpath->total_cost,
1656 : subpath->rows,
1657 444174 : subpath->pathtarget->width);
1658 :
1659 444174 : return pathnode;
1660 : }
1661 :
1662 : /*
1663 : * create_memoize_path
1664 : * Creates a path corresponding to a Memoize plan, returning the pathnode.
1665 : */
1666 : MemoizePath *
1667 248334 : create_memoize_path(PlannerInfo *root, RelOptInfo *rel, Path *subpath,
1668 : List *param_exprs, List *hash_operators,
1669 : bool singlerow, bool binary_mode, double calls)
1670 : {
1671 248334 : MemoizePath *pathnode = makeNode(MemoizePath);
1672 :
1673 : Assert(subpath->parent == rel);
1674 :
1675 248334 : pathnode->path.pathtype = T_Memoize;
1676 248334 : pathnode->path.parent = rel;
1677 248334 : pathnode->path.pathtarget = rel->reltarget;
1678 248334 : pathnode->path.param_info = subpath->param_info;
1679 248334 : pathnode->path.parallel_aware = false;
1680 485084 : pathnode->path.parallel_safe = rel->consider_parallel &&
1681 236750 : subpath->parallel_safe;
1682 248334 : pathnode->path.parallel_workers = subpath->parallel_workers;
1683 248334 : pathnode->path.pathkeys = subpath->pathkeys;
1684 :
1685 248334 : pathnode->subpath = subpath;
1686 248334 : pathnode->hash_operators = hash_operators;
1687 248334 : pathnode->param_exprs = param_exprs;
1688 248334 : pathnode->singlerow = singlerow;
1689 248334 : pathnode->binary_mode = binary_mode;
1690 248334 : pathnode->calls = clamp_row_est(calls);
1691 :
1692 : /*
1693 : * For now we set est_entries to 0. cost_memoize_rescan() does all the
1694 : * hard work to determine how many cache entries there are likely to be,
1695 : * so it seems best to leave it up to that function to fill this field in.
1696 : * If left at 0, the executor will make a guess at a good value.
1697 : */
1698 248334 : pathnode->est_entries = 0;
1699 :
1700 : /* we should not generate this path type when enable_memoize=false */
1701 : Assert(enable_memoize);
1702 248334 : pathnode->path.disabled_nodes = subpath->disabled_nodes;
1703 :
1704 : /*
1705 : * Add a small additional charge for caching the first entry. All the
1706 : * harder calculations for rescans are performed in cost_memoize_rescan().
1707 : */
1708 248334 : pathnode->path.startup_cost = subpath->startup_cost + cpu_tuple_cost;
1709 248334 : pathnode->path.total_cost = subpath->total_cost + cpu_tuple_cost;
1710 248334 : pathnode->path.rows = subpath->rows;
1711 :
1712 248334 : return pathnode;
1713 : }
1714 :
1715 : /*
1716 : * create_unique_path
1717 : * Creates a path representing elimination of distinct rows from the
1718 : * input data. Distinct-ness is defined according to the needs of the
1719 : * semijoin represented by sjinfo. If it is not possible to identify
1720 : * how to make the data unique, NULL is returned.
1721 : *
1722 : * If used at all, this is likely to be called repeatedly on the same rel;
1723 : * and the input subpath should always be the same (the cheapest_total path
1724 : * for the rel). So we cache the result.
1725 : */
1726 : UniquePath *
1727 25396 : create_unique_path(PlannerInfo *root, RelOptInfo *rel, Path *subpath,
1728 : SpecialJoinInfo *sjinfo)
1729 : {
1730 : UniquePath *pathnode;
1731 : Path sort_path; /* dummy for result of cost_sort */
1732 : Path agg_path; /* dummy for result of cost_agg */
1733 : MemoryContext oldcontext;
1734 : int numCols;
1735 :
1736 : /* Caller made a mistake if subpath isn't cheapest_total ... */
1737 : Assert(subpath == rel->cheapest_total_path);
1738 : Assert(subpath->parent == rel);
1739 : /* ... or if SpecialJoinInfo is the wrong one */
1740 : Assert(sjinfo->jointype == JOIN_SEMI);
1741 : Assert(bms_equal(rel->relids, sjinfo->syn_righthand));
1742 :
1743 : /* If result already cached, return it */
1744 25396 : if (rel->cheapest_unique_path)
1745 21776 : return (UniquePath *) rel->cheapest_unique_path;
1746 :
1747 : /* If it's not possible to unique-ify, return NULL */
1748 3620 : if (!(sjinfo->semi_can_btree || sjinfo->semi_can_hash))
1749 120 : return NULL;
1750 :
1751 : /*
1752 : * When called during GEQO join planning, we are in a short-lived memory
1753 : * context. We must make sure that the path and any subsidiary data
1754 : * structures created for a baserel survive the GEQO cycle, else the
1755 : * baserel is trashed for future GEQO cycles. On the other hand, when we
1756 : * are creating those for a joinrel during GEQO, we don't want them to
1757 : * clutter the main planning context. Upshot is that the best solution is
1758 : * to explicitly allocate memory in the same context the given RelOptInfo
1759 : * is in.
1760 : */
1761 3500 : oldcontext = MemoryContextSwitchTo(GetMemoryChunkContext(rel));
1762 :
1763 3500 : pathnode = makeNode(UniquePath);
1764 :
1765 3500 : pathnode->path.pathtype = T_Unique;
1766 3500 : pathnode->path.parent = rel;
1767 3500 : pathnode->path.pathtarget = rel->reltarget;
1768 3500 : pathnode->path.param_info = subpath->param_info;
1769 3500 : pathnode->path.parallel_aware = false;
1770 6552 : pathnode->path.parallel_safe = rel->consider_parallel &&
1771 3052 : subpath->parallel_safe;
1772 3500 : pathnode->path.parallel_workers = subpath->parallel_workers;
1773 :
1774 : /*
1775 : * Assume the output is unsorted, since we don't necessarily have pathkeys
1776 : * to represent it. (This might get overridden below.)
1777 : */
1778 3500 : pathnode->path.pathkeys = NIL;
1779 :
1780 3500 : pathnode->subpath = subpath;
1781 :
1782 : /*
1783 : * Under GEQO and when planning child joins, the sjinfo might be
1784 : * short-lived, so we'd better make copies of data structures we extract
1785 : * from it.
1786 : */
1787 3500 : pathnode->in_operators = copyObject(sjinfo->semi_operators);
1788 3500 : pathnode->uniq_exprs = copyObject(sjinfo->semi_rhs_exprs);
1789 :
1790 : /*
1791 : * If the input is a relation and it has a unique index that proves the
1792 : * semi_rhs_exprs are unique, then we don't need to do anything. Note
1793 : * that relation_has_unique_index_for automatically considers restriction
1794 : * clauses for the rel, as well.
1795 : */
1796 4386 : if (rel->rtekind == RTE_RELATION && sjinfo->semi_can_btree &&
1797 886 : relation_has_unique_index_for(root, rel, NIL,
1798 : sjinfo->semi_rhs_exprs,
1799 : sjinfo->semi_operators))
1800 : {
1801 0 : pathnode->umethod = UNIQUE_PATH_NOOP;
1802 0 : pathnode->path.rows = rel->rows;
1803 0 : pathnode->path.disabled_nodes = subpath->disabled_nodes;
1804 0 : pathnode->path.startup_cost = subpath->startup_cost;
1805 0 : pathnode->path.total_cost = subpath->total_cost;
1806 0 : pathnode->path.pathkeys = subpath->pathkeys;
1807 :
1808 0 : rel->cheapest_unique_path = (Path *) pathnode;
1809 :
1810 0 : MemoryContextSwitchTo(oldcontext);
1811 :
1812 0 : return pathnode;
1813 : }
1814 :
1815 : /*
1816 : * If the input is a subquery whose output must be unique already, then we
1817 : * don't need to do anything. The test for uniqueness has to consider
1818 : * exactly which columns we are extracting; for example "SELECT DISTINCT
1819 : * x,y" doesn't guarantee that x alone is distinct. So we cannot check for
1820 : * this optimization unless semi_rhs_exprs consists only of simple Vars
1821 : * referencing subquery outputs. (Possibly we could do something with
1822 : * expressions in the subquery outputs, too, but for now keep it simple.)
1823 : */
1824 3500 : if (rel->rtekind == RTE_SUBQUERY)
1825 : {
1826 648 : RangeTblEntry *rte = planner_rt_fetch(rel->relid, root);
1827 :
1828 648 : if (query_supports_distinctness(rte->subquery))
1829 : {
1830 : List *sub_tlist_colnos;
1831 :
1832 612 : sub_tlist_colnos = translate_sub_tlist(sjinfo->semi_rhs_exprs,
1833 612 : rel->relid);
1834 :
1835 722 : if (sub_tlist_colnos &&
1836 110 : query_is_distinct_for(rte->subquery,
1837 : sub_tlist_colnos,
1838 : sjinfo->semi_operators))
1839 : {
1840 0 : pathnode->umethod = UNIQUE_PATH_NOOP;
1841 0 : pathnode->path.rows = rel->rows;
1842 0 : pathnode->path.disabled_nodes = subpath->disabled_nodes;
1843 0 : pathnode->path.startup_cost = subpath->startup_cost;
1844 0 : pathnode->path.total_cost = subpath->total_cost;
1845 0 : pathnode->path.pathkeys = subpath->pathkeys;
1846 :
1847 0 : rel->cheapest_unique_path = (Path *) pathnode;
1848 :
1849 0 : MemoryContextSwitchTo(oldcontext);
1850 :
1851 0 : return pathnode;
1852 : }
1853 : }
1854 : }
1855 :
1856 : /* Estimate number of output rows */
1857 3500 : pathnode->path.rows = estimate_num_groups(root,
1858 : sjinfo->semi_rhs_exprs,
1859 : rel->rows,
1860 : NULL,
1861 : NULL);
1862 3500 : numCols = list_length(sjinfo->semi_rhs_exprs);
1863 :
1864 3500 : if (sjinfo->semi_can_btree)
1865 : {
1866 : /*
1867 : * Estimate cost for sort+unique implementation
1868 : */
1869 3500 : cost_sort(&sort_path, root, NIL,
1870 : subpath->disabled_nodes,
1871 : subpath->total_cost,
1872 : rel->rows,
1873 3500 : subpath->pathtarget->width,
1874 : 0.0,
1875 : work_mem,
1876 : -1.0);
1877 :
1878 : /*
1879 : * Charge one cpu_operator_cost per comparison per input tuple. We
1880 : * assume all columns get compared at most of the tuples. (XXX
1881 : * probably this is an overestimate.) This should agree with
1882 : * create_upper_unique_path.
1883 : */
1884 3500 : sort_path.total_cost += cpu_operator_cost * rel->rows * numCols;
1885 : }
1886 :
1887 3500 : if (sjinfo->semi_can_hash)
1888 : {
1889 : /*
1890 : * Estimate the overhead per hashtable entry at 64 bytes (same as in
1891 : * planner.c).
1892 : */
1893 3500 : int hashentrysize = subpath->pathtarget->width + 64;
1894 :
1895 3500 : if (hashentrysize * pathnode->path.rows > get_hash_memory_limit())
1896 : {
1897 : /*
1898 : * We should not try to hash. Hack the SpecialJoinInfo to
1899 : * remember this, in case we come through here again.
1900 : */
1901 0 : sjinfo->semi_can_hash = false;
1902 : }
1903 : else
1904 3500 : cost_agg(&agg_path, root,
1905 : AGG_HASHED, NULL,
1906 : numCols, pathnode->path.rows,
1907 : NIL,
1908 : subpath->disabled_nodes,
1909 : subpath->startup_cost,
1910 : subpath->total_cost,
1911 : rel->rows,
1912 3500 : subpath->pathtarget->width);
1913 : }
1914 :
1915 3500 : if (sjinfo->semi_can_btree && sjinfo->semi_can_hash)
1916 : {
1917 3500 : if (agg_path.disabled_nodes < sort_path.disabled_nodes ||
1918 3494 : (agg_path.disabled_nodes == sort_path.disabled_nodes &&
1919 3494 : agg_path.total_cost < sort_path.total_cost))
1920 3376 : pathnode->umethod = UNIQUE_PATH_HASH;
1921 : else
1922 124 : pathnode->umethod = UNIQUE_PATH_SORT;
1923 : }
1924 0 : else if (sjinfo->semi_can_btree)
1925 0 : pathnode->umethod = UNIQUE_PATH_SORT;
1926 0 : else if (sjinfo->semi_can_hash)
1927 0 : pathnode->umethod = UNIQUE_PATH_HASH;
1928 : else
1929 : {
1930 : /* we can get here only if we abandoned hashing above */
1931 0 : MemoryContextSwitchTo(oldcontext);
1932 0 : return NULL;
1933 : }
1934 :
1935 3500 : if (pathnode->umethod == UNIQUE_PATH_HASH)
1936 : {
1937 3376 : pathnode->path.disabled_nodes = agg_path.disabled_nodes;
1938 3376 : pathnode->path.startup_cost = agg_path.startup_cost;
1939 3376 : pathnode->path.total_cost = agg_path.total_cost;
1940 : }
1941 : else
1942 : {
1943 124 : pathnode->path.disabled_nodes = sort_path.disabled_nodes;
1944 124 : pathnode->path.startup_cost = sort_path.startup_cost;
1945 124 : pathnode->path.total_cost = sort_path.total_cost;
1946 : }
1947 :
1948 3500 : rel->cheapest_unique_path = (Path *) pathnode;
1949 :
1950 3500 : MemoryContextSwitchTo(oldcontext);
1951 :
1952 3500 : return pathnode;
1953 : }
1954 :
1955 : /*
1956 : * create_gather_merge_path
1957 : *
1958 : * Creates a path corresponding to a gather merge scan, returning
1959 : * the pathnode.
1960 : */
1961 : GatherMergePath *
1962 9896 : create_gather_merge_path(PlannerInfo *root, RelOptInfo *rel, Path *subpath,
1963 : PathTarget *target, List *pathkeys,
1964 : Relids required_outer, double *rows)
1965 : {
1966 9896 : GatherMergePath *pathnode = makeNode(GatherMergePath);
1967 9896 : int input_disabled_nodes = 0;
1968 9896 : Cost input_startup_cost = 0;
1969 9896 : Cost input_total_cost = 0;
1970 :
1971 : Assert(subpath->parallel_safe);
1972 : Assert(pathkeys);
1973 :
1974 : /*
1975 : * The subpath should guarantee that it is adequately ordered either by
1976 : * adding an explicit sort node or by using presorted input. We cannot
1977 : * add an explicit Sort node for the subpath in createplan.c on additional
1978 : * pathkeys, because we can't guarantee the sort would be safe. For
1979 : * example, expressions may be volatile or otherwise parallel unsafe.
1980 : */
1981 9896 : if (!pathkeys_contained_in(pathkeys, subpath->pathkeys))
1982 0 : elog(ERROR, "gather merge input not sufficiently sorted");
1983 :
1984 9896 : pathnode->path.pathtype = T_GatherMerge;
1985 9896 : pathnode->path.parent = rel;
1986 9896 : pathnode->path.param_info = get_baserel_parampathinfo(root, rel,
1987 : required_outer);
1988 9896 : pathnode->path.parallel_aware = false;
1989 :
1990 9896 : pathnode->subpath = subpath;
1991 9896 : pathnode->num_workers = subpath->parallel_workers;
1992 9896 : pathnode->path.pathkeys = pathkeys;
1993 9896 : pathnode->path.pathtarget = target ? target : rel->reltarget;
1994 :
1995 9896 : input_disabled_nodes += subpath->disabled_nodes;
1996 9896 : input_startup_cost += subpath->startup_cost;
1997 9896 : input_total_cost += subpath->total_cost;
1998 :
1999 9896 : cost_gather_merge(pathnode, root, rel, pathnode->path.param_info,
2000 : input_disabled_nodes, input_startup_cost,
2001 : input_total_cost, rows);
2002 :
2003 9896 : return pathnode;
2004 : }
2005 :
2006 : /*
2007 : * translate_sub_tlist - get subquery column numbers represented by tlist
2008 : *
2009 : * The given targetlist usually contains only Vars referencing the given relid.
2010 : * Extract their varattnos (ie, the column numbers of the subquery) and return
2011 : * as an integer List.
2012 : *
2013 : * If any of the tlist items is not a simple Var, we cannot determine whether
2014 : * the subquery's uniqueness condition (if any) matches ours, so punt and
2015 : * return NIL.
2016 : */
2017 : static List *
2018 612 : translate_sub_tlist(List *tlist, int relid)
2019 : {
2020 612 : List *result = NIL;
2021 : ListCell *l;
2022 :
2023 722 : foreach(l, tlist)
2024 : {
2025 612 : Var *var = (Var *) lfirst(l);
2026 :
2027 612 : if (!var || !IsA(var, Var) ||
2028 110 : var->varno != relid)
2029 502 : return NIL; /* punt */
2030 :
2031 110 : result = lappend_int(result, var->varattno);
2032 : }
2033 110 : return result;
2034 : }
2035 :
2036 : /*
2037 : * create_gather_path
2038 : * Creates a path corresponding to a gather scan, returning the
2039 : * pathnode.
2040 : *
2041 : * 'rows' may optionally be set to override row estimates from other sources.
2042 : */
2043 : GatherPath *
2044 16232 : create_gather_path(PlannerInfo *root, RelOptInfo *rel, Path *subpath,
2045 : PathTarget *target, Relids required_outer, double *rows)
2046 : {
2047 16232 : GatherPath *pathnode = makeNode(GatherPath);
2048 :
2049 : Assert(subpath->parallel_safe);
2050 :
2051 16232 : pathnode->path.pathtype = T_Gather;
2052 16232 : pathnode->path.parent = rel;
2053 16232 : pathnode->path.pathtarget = target;
2054 16232 : pathnode->path.param_info = get_baserel_parampathinfo(root, rel,
2055 : required_outer);
2056 16232 : pathnode->path.parallel_aware = false;
2057 16232 : pathnode->path.parallel_safe = false;
2058 16232 : pathnode->path.parallel_workers = 0;
2059 16232 : pathnode->path.pathkeys = NIL; /* Gather has unordered result */
2060 :
2061 16232 : pathnode->subpath = subpath;
2062 16232 : pathnode->num_workers = subpath->parallel_workers;
2063 16232 : pathnode->single_copy = false;
2064 :
2065 16232 : if (pathnode->num_workers == 0)
2066 : {
2067 0 : pathnode->path.pathkeys = subpath->pathkeys;
2068 0 : pathnode->num_workers = 1;
2069 0 : pathnode->single_copy = true;
2070 : }
2071 :
2072 16232 : cost_gather(pathnode, root, rel, pathnode->path.param_info, rows);
2073 :
2074 16232 : return pathnode;
2075 : }
2076 :
2077 : /*
2078 : * create_subqueryscan_path
2079 : * Creates a path corresponding to a scan of a subquery,
2080 : * returning the pathnode.
2081 : *
2082 : * Caller must pass trivial_pathtarget = true if it believes rel->reltarget to
2083 : * be trivial, ie just a fetch of all the subquery output columns in order.
2084 : * While we could determine that here, the caller can usually do it more
2085 : * efficiently (or at least amortize it over multiple calls).
2086 : */
2087 : SubqueryScanPath *
2088 41004 : create_subqueryscan_path(PlannerInfo *root, RelOptInfo *rel, Path *subpath,
2089 : bool trivial_pathtarget,
2090 : List *pathkeys, Relids required_outer)
2091 : {
2092 41004 : SubqueryScanPath *pathnode = makeNode(SubqueryScanPath);
2093 :
2094 41004 : pathnode->path.pathtype = T_SubqueryScan;
2095 41004 : pathnode->path.parent = rel;
2096 41004 : pathnode->path.pathtarget = rel->reltarget;
2097 41004 : pathnode->path.param_info = get_baserel_parampathinfo(root, rel,
2098 : required_outer);
2099 41004 : pathnode->path.parallel_aware = false;
2100 67546 : pathnode->path.parallel_safe = rel->consider_parallel &&
2101 26542 : subpath->parallel_safe;
2102 41004 : pathnode->path.parallel_workers = subpath->parallel_workers;
2103 41004 : pathnode->path.pathkeys = pathkeys;
2104 41004 : pathnode->subpath = subpath;
2105 :
2106 41004 : cost_subqueryscan(pathnode, root, rel, pathnode->path.param_info,
2107 : trivial_pathtarget);
2108 :
2109 41004 : return pathnode;
2110 : }
2111 :
2112 : /*
2113 : * create_functionscan_path
2114 : * Creates a path corresponding to a sequential scan of a function,
2115 : * returning the pathnode.
2116 : */
2117 : Path *
2118 43978 : create_functionscan_path(PlannerInfo *root, RelOptInfo *rel,
2119 : List *pathkeys, Relids required_outer)
2120 : {
2121 43978 : Path *pathnode = makeNode(Path);
2122 :
2123 43978 : pathnode->pathtype = T_FunctionScan;
2124 43978 : pathnode->parent = rel;
2125 43978 : pathnode->pathtarget = rel->reltarget;
2126 43978 : pathnode->param_info = get_baserel_parampathinfo(root, rel,
2127 : required_outer);
2128 43978 : pathnode->parallel_aware = false;
2129 43978 : pathnode->parallel_safe = rel->consider_parallel;
2130 43978 : pathnode->parallel_workers = 0;
2131 43978 : pathnode->pathkeys = pathkeys;
2132 :
2133 43978 : cost_functionscan(pathnode, root, rel, pathnode->param_info);
2134 :
2135 43978 : return pathnode;
2136 : }
2137 :
2138 : /*
2139 : * create_tablefuncscan_path
2140 : * Creates a path corresponding to a sequential scan of a table function,
2141 : * returning the pathnode.
2142 : */
2143 : Path *
2144 626 : create_tablefuncscan_path(PlannerInfo *root, RelOptInfo *rel,
2145 : Relids required_outer)
2146 : {
2147 626 : Path *pathnode = makeNode(Path);
2148 :
2149 626 : pathnode->pathtype = T_TableFuncScan;
2150 626 : pathnode->parent = rel;
2151 626 : pathnode->pathtarget = rel->reltarget;
2152 626 : pathnode->param_info = get_baserel_parampathinfo(root, rel,
2153 : required_outer);
2154 626 : pathnode->parallel_aware = false;
2155 626 : pathnode->parallel_safe = rel->consider_parallel;
2156 626 : pathnode->parallel_workers = 0;
2157 626 : pathnode->pathkeys = NIL; /* result is always unordered */
2158 :
2159 626 : cost_tablefuncscan(pathnode, root, rel, pathnode->param_info);
2160 :
2161 626 : return pathnode;
2162 : }
2163 :
2164 : /*
2165 : * create_valuesscan_path
2166 : * Creates a path corresponding to a scan of a VALUES list,
2167 : * returning the pathnode.
2168 : */
2169 : Path *
2170 7896 : create_valuesscan_path(PlannerInfo *root, RelOptInfo *rel,
2171 : Relids required_outer)
2172 : {
2173 7896 : Path *pathnode = makeNode(Path);
2174 :
2175 7896 : pathnode->pathtype = T_ValuesScan;
2176 7896 : pathnode->parent = rel;
2177 7896 : pathnode->pathtarget = rel->reltarget;
2178 7896 : pathnode->param_info = get_baserel_parampathinfo(root, rel,
2179 : required_outer);
2180 7896 : pathnode->parallel_aware = false;
2181 7896 : pathnode->parallel_safe = rel->consider_parallel;
2182 7896 : pathnode->parallel_workers = 0;
2183 7896 : pathnode->pathkeys = NIL; /* result is always unordered */
2184 :
2185 7896 : cost_valuesscan(pathnode, root, rel, pathnode->param_info);
2186 :
2187 7896 : return pathnode;
2188 : }
2189 :
2190 : /*
2191 : * create_ctescan_path
2192 : * Creates a path corresponding to a scan of a non-self-reference CTE,
2193 : * returning the pathnode.
2194 : */
2195 : Path *
2196 3196 : create_ctescan_path(PlannerInfo *root, RelOptInfo *rel,
2197 : List *pathkeys, Relids required_outer)
2198 : {
2199 3196 : Path *pathnode = makeNode(Path);
2200 :
2201 3196 : pathnode->pathtype = T_CteScan;
2202 3196 : pathnode->parent = rel;
2203 3196 : pathnode->pathtarget = rel->reltarget;
2204 3196 : pathnode->param_info = get_baserel_parampathinfo(root, rel,
2205 : required_outer);
2206 3196 : pathnode->parallel_aware = false;
2207 3196 : pathnode->parallel_safe = rel->consider_parallel;
2208 3196 : pathnode->parallel_workers = 0;
2209 3196 : pathnode->pathkeys = pathkeys;
2210 :
2211 3196 : cost_ctescan(pathnode, root, rel, pathnode->param_info);
2212 :
2213 3196 : return pathnode;
2214 : }
2215 :
2216 : /*
2217 : * create_namedtuplestorescan_path
2218 : * Creates a path corresponding to a scan of a named tuplestore, returning
2219 : * the pathnode.
2220 : */
2221 : Path *
2222 446 : create_namedtuplestorescan_path(PlannerInfo *root, RelOptInfo *rel,
2223 : Relids required_outer)
2224 : {
2225 446 : Path *pathnode = makeNode(Path);
2226 :
2227 446 : pathnode->pathtype = T_NamedTuplestoreScan;
2228 446 : pathnode->parent = rel;
2229 446 : pathnode->pathtarget = rel->reltarget;
2230 446 : pathnode->param_info = get_baserel_parampathinfo(root, rel,
2231 : required_outer);
2232 446 : pathnode->parallel_aware = false;
2233 446 : pathnode->parallel_safe = rel->consider_parallel;
2234 446 : pathnode->parallel_workers = 0;
2235 446 : pathnode->pathkeys = NIL; /* result is always unordered */
2236 :
2237 446 : cost_namedtuplestorescan(pathnode, root, rel, pathnode->param_info);
2238 :
2239 446 : return pathnode;
2240 : }
2241 :
2242 : /*
2243 : * create_resultscan_path
2244 : * Creates a path corresponding to a scan of an RTE_RESULT relation,
2245 : * returning the pathnode.
2246 : */
2247 : Path *
2248 1646 : create_resultscan_path(PlannerInfo *root, RelOptInfo *rel,
2249 : Relids required_outer)
2250 : {
2251 1646 : Path *pathnode = makeNode(Path);
2252 :
2253 1646 : pathnode->pathtype = T_Result;
2254 1646 : pathnode->parent = rel;
2255 1646 : pathnode->pathtarget = rel->reltarget;
2256 1646 : pathnode->param_info = get_baserel_parampathinfo(root, rel,
2257 : required_outer);
2258 1646 : pathnode->parallel_aware = false;
2259 1646 : pathnode->parallel_safe = rel->consider_parallel;
2260 1646 : pathnode->parallel_workers = 0;
2261 1646 : pathnode->pathkeys = NIL; /* result is always unordered */
2262 :
2263 1646 : cost_resultscan(pathnode, root, rel, pathnode->param_info);
2264 :
2265 1646 : return pathnode;
2266 : }
2267 :
2268 : /*
2269 : * create_worktablescan_path
2270 : * Creates a path corresponding to a scan of a self-reference CTE,
2271 : * returning the pathnode.
2272 : */
2273 : Path *
2274 812 : create_worktablescan_path(PlannerInfo *root, RelOptInfo *rel,
2275 : Relids required_outer)
2276 : {
2277 812 : Path *pathnode = makeNode(Path);
2278 :
2279 812 : pathnode->pathtype = T_WorkTableScan;
2280 812 : pathnode->parent = rel;
2281 812 : pathnode->pathtarget = rel->reltarget;
2282 812 : pathnode->param_info = get_baserel_parampathinfo(root, rel,
2283 : required_outer);
2284 812 : pathnode->parallel_aware = false;
2285 812 : pathnode->parallel_safe = rel->consider_parallel;
2286 812 : pathnode->parallel_workers = 0;
2287 812 : pathnode->pathkeys = NIL; /* result is always unordered */
2288 :
2289 : /* Cost is the same as for a regular CTE scan */
2290 812 : cost_ctescan(pathnode, root, rel, pathnode->param_info);
2291 :
2292 812 : return pathnode;
2293 : }
2294 :
2295 : /*
2296 : * create_foreignscan_path
2297 : * Creates a path corresponding to a scan of a foreign base table,
2298 : * returning the pathnode.
2299 : *
2300 : * This function is never called from core Postgres; rather, it's expected
2301 : * to be called by the GetForeignPaths function of a foreign data wrapper.
2302 : * We make the FDW supply all fields of the path, since we do not have any way
2303 : * to calculate them in core. However, there is a usually-sane default for
2304 : * the pathtarget (rel->reltarget), so we let a NULL for "target" select that.
2305 : */
2306 : ForeignPath *
2307 3566 : create_foreignscan_path(PlannerInfo *root, RelOptInfo *rel,
2308 : PathTarget *target,
2309 : double rows, int disabled_nodes,
2310 : Cost startup_cost, Cost total_cost,
2311 : List *pathkeys,
2312 : Relids required_outer,
2313 : Path *fdw_outerpath,
2314 : List *fdw_restrictinfo,
2315 : List *fdw_private)
2316 : {
2317 3566 : ForeignPath *pathnode = makeNode(ForeignPath);
2318 :
2319 : /* Historically some FDWs were confused about when to use this */
2320 : Assert(IS_SIMPLE_REL(rel));
2321 :
2322 3566 : pathnode->path.pathtype = T_ForeignScan;
2323 3566 : pathnode->path.parent = rel;
2324 3566 : pathnode->path.pathtarget = target ? target : rel->reltarget;
2325 3566 : pathnode->path.param_info = get_baserel_parampathinfo(root, rel,
2326 : required_outer);
2327 3566 : pathnode->path.parallel_aware = false;
2328 3566 : pathnode->path.parallel_safe = rel->consider_parallel;
2329 3566 : pathnode->path.parallel_workers = 0;
2330 3566 : pathnode->path.rows = rows;
2331 3566 : pathnode->path.disabled_nodes = disabled_nodes;
2332 3566 : pathnode->path.startup_cost = startup_cost;
2333 3566 : pathnode->path.total_cost = total_cost;
2334 3566 : pathnode->path.pathkeys = pathkeys;
2335 :
2336 3566 : pathnode->fdw_outerpath = fdw_outerpath;
2337 3566 : pathnode->fdw_restrictinfo = fdw_restrictinfo;
2338 3566 : pathnode->fdw_private = fdw_private;
2339 :
2340 3566 : return pathnode;
2341 : }
2342 :
2343 : /*
2344 : * create_foreign_join_path
2345 : * Creates a path corresponding to a scan of a foreign join,
2346 : * returning the pathnode.
2347 : *
2348 : * This function is never called from core Postgres; rather, it's expected
2349 : * to be called by the GetForeignJoinPaths function of a foreign data wrapper.
2350 : * We make the FDW supply all fields of the path, since we do not have any way
2351 : * to calculate them in core. However, there is a usually-sane default for
2352 : * the pathtarget (rel->reltarget), so we let a NULL for "target" select that.
2353 : */
2354 : ForeignPath *
2355 1124 : create_foreign_join_path(PlannerInfo *root, RelOptInfo *rel,
2356 : PathTarget *target,
2357 : double rows, int disabled_nodes,
2358 : Cost startup_cost, Cost total_cost,
2359 : List *pathkeys,
2360 : Relids required_outer,
2361 : Path *fdw_outerpath,
2362 : List *fdw_restrictinfo,
2363 : List *fdw_private)
2364 : {
2365 1124 : ForeignPath *pathnode = makeNode(ForeignPath);
2366 :
2367 : /*
2368 : * We should use get_joinrel_parampathinfo to handle parameterized paths,
2369 : * but the API of this function doesn't support it, and existing
2370 : * extensions aren't yet trying to build such paths anyway. For the
2371 : * moment just throw an error if someone tries it; eventually we should
2372 : * revisit this.
2373 : */
2374 1124 : if (!bms_is_empty(required_outer) || !bms_is_empty(rel->lateral_relids))
2375 0 : elog(ERROR, "parameterized foreign joins are not supported yet");
2376 :
2377 1124 : pathnode->path.pathtype = T_ForeignScan;
2378 1124 : pathnode->path.parent = rel;
2379 1124 : pathnode->path.pathtarget = target ? target : rel->reltarget;
2380 1124 : pathnode->path.param_info = NULL; /* XXX see above */
2381 1124 : pathnode->path.parallel_aware = false;
2382 1124 : pathnode->path.parallel_safe = rel->consider_parallel;
2383 1124 : pathnode->path.parallel_workers = 0;
2384 1124 : pathnode->path.rows = rows;
2385 1124 : pathnode->path.disabled_nodes = disabled_nodes;
2386 1124 : pathnode->path.startup_cost = startup_cost;
2387 1124 : pathnode->path.total_cost = total_cost;
2388 1124 : pathnode->path.pathkeys = pathkeys;
2389 :
2390 1124 : pathnode->fdw_outerpath = fdw_outerpath;
2391 1124 : pathnode->fdw_restrictinfo = fdw_restrictinfo;
2392 1124 : pathnode->fdw_private = fdw_private;
2393 :
2394 1124 : return pathnode;
2395 : }
2396 :
2397 : /*
2398 : * create_foreign_upper_path
2399 : * Creates a path corresponding to an upper relation that's computed
2400 : * directly by an FDW, returning the pathnode.
2401 : *
2402 : * This function is never called from core Postgres; rather, it's expected to
2403 : * be called by the GetForeignUpperPaths function of a foreign data wrapper.
2404 : * We make the FDW supply all fields of the path, since we do not have any way
2405 : * to calculate them in core. However, there is a usually-sane default for
2406 : * the pathtarget (rel->reltarget), so we let a NULL for "target" select that.
2407 : */
2408 : ForeignPath *
2409 576 : create_foreign_upper_path(PlannerInfo *root, RelOptInfo *rel,
2410 : PathTarget *target,
2411 : double rows, int disabled_nodes,
2412 : Cost startup_cost, Cost total_cost,
2413 : List *pathkeys,
2414 : Path *fdw_outerpath,
2415 : List *fdw_restrictinfo,
2416 : List *fdw_private)
2417 : {
2418 576 : ForeignPath *pathnode = makeNode(ForeignPath);
2419 :
2420 : /*
2421 : * Upper relations should never have any lateral references, since joining
2422 : * is complete.
2423 : */
2424 : Assert(bms_is_empty(rel->lateral_relids));
2425 :
2426 576 : pathnode->path.pathtype = T_ForeignScan;
2427 576 : pathnode->path.parent = rel;
2428 576 : pathnode->path.pathtarget = target ? target : rel->reltarget;
2429 576 : pathnode->path.param_info = NULL;
2430 576 : pathnode->path.parallel_aware = false;
2431 576 : pathnode->path.parallel_safe = rel->consider_parallel;
2432 576 : pathnode->path.parallel_workers = 0;
2433 576 : pathnode->path.rows = rows;
2434 576 : pathnode->path.disabled_nodes = disabled_nodes;
2435 576 : pathnode->path.startup_cost = startup_cost;
2436 576 : pathnode->path.total_cost = total_cost;
2437 576 : pathnode->path.pathkeys = pathkeys;
2438 :
2439 576 : pathnode->fdw_outerpath = fdw_outerpath;
2440 576 : pathnode->fdw_restrictinfo = fdw_restrictinfo;
2441 576 : pathnode->fdw_private = fdw_private;
2442 :
2443 576 : return pathnode;
2444 : }
2445 :
2446 : /*
2447 : * calc_nestloop_required_outer
2448 : * Compute the required_outer set for a nestloop join path
2449 : *
2450 : * Note: when considering a child join, the inputs nonetheless use top-level
2451 : * parent relids
2452 : *
2453 : * Note: result must not share storage with either input
2454 : */
2455 : Relids
2456 2501880 : calc_nestloop_required_outer(Relids outerrelids,
2457 : Relids outer_paramrels,
2458 : Relids innerrelids,
2459 : Relids inner_paramrels)
2460 : {
2461 : Relids required_outer;
2462 :
2463 : /* inner_path can require rels from outer path, but not vice versa */
2464 : Assert(!bms_overlap(outer_paramrels, innerrelids));
2465 : /* easy case if inner path is not parameterized */
2466 2501880 : if (!inner_paramrels)
2467 1699978 : return bms_copy(outer_paramrels);
2468 : /* else, form the union ... */
2469 801902 : required_outer = bms_union(outer_paramrels, inner_paramrels);
2470 : /* ... and remove any mention of now-satisfied outer rels */
2471 801902 : required_outer = bms_del_members(required_outer,
2472 : outerrelids);
2473 801902 : return required_outer;
2474 : }
2475 :
2476 : /*
2477 : * calc_non_nestloop_required_outer
2478 : * Compute the required_outer set for a merge or hash join path
2479 : *
2480 : * Note: result must not share storage with either input
2481 : */
2482 : Relids
2483 1629762 : calc_non_nestloop_required_outer(Path *outer_path, Path *inner_path)
2484 : {
2485 1629762 : Relids outer_paramrels = PATH_REQ_OUTER(outer_path);
2486 1629762 : Relids inner_paramrels = PATH_REQ_OUTER(inner_path);
2487 : Relids innerrelids PG_USED_FOR_ASSERTS_ONLY;
2488 : Relids outerrelids PG_USED_FOR_ASSERTS_ONLY;
2489 : Relids required_outer;
2490 :
2491 : /*
2492 : * Any parameterization of the input paths refers to topmost parents of
2493 : * the relevant relations, because reparameterize_path_by_child() hasn't
2494 : * been called yet. So we must consider topmost parents of the relations
2495 : * being joined, too, while checking for disallowed parameterization
2496 : * cases.
2497 : */
2498 1629762 : if (inner_path->parent->top_parent_relids)
2499 35708 : innerrelids = inner_path->parent->top_parent_relids;
2500 : else
2501 1594054 : innerrelids = inner_path->parent->relids;
2502 :
2503 1629762 : if (outer_path->parent->top_parent_relids)
2504 35708 : outerrelids = outer_path->parent->top_parent_relids;
2505 : else
2506 1594054 : outerrelids = outer_path->parent->relids;
2507 :
2508 : /* neither path can require rels from the other */
2509 : Assert(!bms_overlap(outer_paramrels, innerrelids));
2510 : Assert(!bms_overlap(inner_paramrels, outerrelids));
2511 : /* form the union ... */
2512 1629762 : required_outer = bms_union(outer_paramrels, inner_paramrels);
2513 : /* we do not need an explicit test for empty; bms_union gets it right */
2514 1629762 : return required_outer;
2515 : }
2516 :
2517 : /*
2518 : * create_nestloop_path
2519 : * Creates a pathnode corresponding to a nestloop join between two
2520 : * relations.
2521 : *
2522 : * 'joinrel' is the join relation.
2523 : * 'jointype' is the type of join required
2524 : * 'workspace' is the result from initial_cost_nestloop
2525 : * 'extra' contains various information about the join
2526 : * 'outer_path' is the outer path
2527 : * 'inner_path' is the inner path
2528 : * 'restrict_clauses' are the RestrictInfo nodes to apply at the join
2529 : * 'pathkeys' are the path keys of the new join path
2530 : * 'required_outer' is the set of required outer rels
2531 : *
2532 : * Returns the resulting path node.
2533 : */
2534 : NestPath *
2535 1141484 : create_nestloop_path(PlannerInfo *root,
2536 : RelOptInfo *joinrel,
2537 : JoinType jointype,
2538 : JoinCostWorkspace *workspace,
2539 : JoinPathExtraData *extra,
2540 : Path *outer_path,
2541 : Path *inner_path,
2542 : List *restrict_clauses,
2543 : List *pathkeys,
2544 : Relids required_outer)
2545 : {
2546 1141484 : NestPath *pathnode = makeNode(NestPath);
2547 1141484 : Relids inner_req_outer = PATH_REQ_OUTER(inner_path);
2548 : Relids outerrelids;
2549 :
2550 : /*
2551 : * Paths are parameterized by top-level parents, so run parameterization
2552 : * tests on the parent relids.
2553 : */
2554 1141484 : if (outer_path->parent->top_parent_relids)
2555 17506 : outerrelids = outer_path->parent->top_parent_relids;
2556 : else
2557 1123978 : outerrelids = outer_path->parent->relids;
2558 :
2559 : /*
2560 : * If the inner path is parameterized by the outer, we must drop any
2561 : * restrict_clauses that are due to be moved into the inner path. We have
2562 : * to do this now, rather than postpone the work till createplan time,
2563 : * because the restrict_clauses list can affect the size and cost
2564 : * estimates for this path. We detect such clauses by checking for serial
2565 : * number match to clauses already enforced in the inner path.
2566 : */
2567 1141484 : if (bms_overlap(inner_req_outer, outerrelids))
2568 : {
2569 309936 : Bitmapset *enforced_serials = get_param_path_clause_serials(inner_path);
2570 309936 : List *jclauses = NIL;
2571 : ListCell *lc;
2572 :
2573 673660 : foreach(lc, restrict_clauses)
2574 : {
2575 363724 : RestrictInfo *rinfo = (RestrictInfo *) lfirst(lc);
2576 :
2577 363724 : if (!bms_is_member(rinfo->rinfo_serial, enforced_serials))
2578 47958 : jclauses = lappend(jclauses, rinfo);
2579 : }
2580 309936 : restrict_clauses = jclauses;
2581 : }
2582 :
2583 1141484 : pathnode->jpath.path.pathtype = T_NestLoop;
2584 1141484 : pathnode->jpath.path.parent = joinrel;
2585 1141484 : pathnode->jpath.path.pathtarget = joinrel->reltarget;
2586 1141484 : pathnode->jpath.path.param_info =
2587 1141484 : get_joinrel_parampathinfo(root,
2588 : joinrel,
2589 : outer_path,
2590 : inner_path,
2591 : extra->sjinfo,
2592 : required_outer,
2593 : &restrict_clauses);
2594 1141484 : pathnode->jpath.path.parallel_aware = false;
2595 3319450 : pathnode->jpath.path.parallel_safe = joinrel->consider_parallel &&
2596 1141484 : outer_path->parallel_safe && inner_path->parallel_safe;
2597 : /* This is a foolish way to estimate parallel_workers, but for now... */
2598 1141484 : pathnode->jpath.path.parallel_workers = outer_path->parallel_workers;
2599 1141484 : pathnode->jpath.path.pathkeys = pathkeys;
2600 1141484 : pathnode->jpath.jointype = jointype;
2601 1141484 : pathnode->jpath.inner_unique = extra->inner_unique;
2602 1141484 : pathnode->jpath.outerjoinpath = outer_path;
2603 1141484 : pathnode->jpath.innerjoinpath = inner_path;
2604 1141484 : pathnode->jpath.joinrestrictinfo = restrict_clauses;
2605 :
2606 1141484 : final_cost_nestloop(root, pathnode, workspace, extra);
2607 :
2608 1141484 : return pathnode;
2609 : }
2610 :
2611 : /*
2612 : * create_mergejoin_path
2613 : * Creates a pathnode corresponding to a mergejoin join between
2614 : * two relations
2615 : *
2616 : * 'joinrel' is the join relation
2617 : * 'jointype' is the type of join required
2618 : * 'workspace' is the result from initial_cost_mergejoin
2619 : * 'extra' contains various information about the join
2620 : * 'outer_path' is the outer path
2621 : * 'inner_path' is the inner path
2622 : * 'restrict_clauses' are the RestrictInfo nodes to apply at the join
2623 : * 'pathkeys' are the path keys of the new join path
2624 : * 'required_outer' is the set of required outer rels
2625 : * 'mergeclauses' are the RestrictInfo nodes to use as merge clauses
2626 : * (this should be a subset of the restrict_clauses list)
2627 : * 'outersortkeys' are the sort varkeys for the outer relation
2628 : * 'innersortkeys' are the sort varkeys for the inner relation
2629 : */
2630 : MergePath *
2631 266912 : create_mergejoin_path(PlannerInfo *root,
2632 : RelOptInfo *joinrel,
2633 : JoinType jointype,
2634 : JoinCostWorkspace *workspace,
2635 : JoinPathExtraData *extra,
2636 : Path *outer_path,
2637 : Path *inner_path,
2638 : List *restrict_clauses,
2639 : List *pathkeys,
2640 : Relids required_outer,
2641 : List *mergeclauses,
2642 : List *outersortkeys,
2643 : List *innersortkeys)
2644 : {
2645 266912 : MergePath *pathnode = makeNode(MergePath);
2646 :
2647 266912 : pathnode->jpath.path.pathtype = T_MergeJoin;
2648 266912 : pathnode->jpath.path.parent = joinrel;
2649 266912 : pathnode->jpath.path.pathtarget = joinrel->reltarget;
2650 266912 : pathnode->jpath.path.param_info =
2651 266912 : get_joinrel_parampathinfo(root,
2652 : joinrel,
2653 : outer_path,
2654 : inner_path,
2655 : extra->sjinfo,
2656 : required_outer,
2657 : &restrict_clauses);
2658 266912 : pathnode->jpath.path.parallel_aware = false;
2659 770026 : pathnode->jpath.path.parallel_safe = joinrel->consider_parallel &&
2660 266912 : outer_path->parallel_safe && inner_path->parallel_safe;
2661 : /* This is a foolish way to estimate parallel_workers, but for now... */
2662 266912 : pathnode->jpath.path.parallel_workers = outer_path->parallel_workers;
2663 266912 : pathnode->jpath.path.pathkeys = pathkeys;
2664 266912 : pathnode->jpath.jointype = jointype;
2665 266912 : pathnode->jpath.inner_unique = extra->inner_unique;
2666 266912 : pathnode->jpath.outerjoinpath = outer_path;
2667 266912 : pathnode->jpath.innerjoinpath = inner_path;
2668 266912 : pathnode->jpath.joinrestrictinfo = restrict_clauses;
2669 266912 : pathnode->path_mergeclauses = mergeclauses;
2670 266912 : pathnode->outersortkeys = outersortkeys;
2671 266912 : pathnode->innersortkeys = innersortkeys;
2672 : /* pathnode->skip_mark_restore will be set by final_cost_mergejoin */
2673 : /* pathnode->materialize_inner will be set by final_cost_mergejoin */
2674 :
2675 266912 : final_cost_mergejoin(root, pathnode, workspace, extra);
2676 :
2677 266912 : return pathnode;
2678 : }
2679 :
2680 : /*
2681 : * create_hashjoin_path
2682 : * Creates a pathnode corresponding to a hash join between two relations.
2683 : *
2684 : * 'joinrel' is the join relation
2685 : * 'jointype' is the type of join required
2686 : * 'workspace' is the result from initial_cost_hashjoin
2687 : * 'extra' contains various information about the join
2688 : * 'outer_path' is the cheapest outer path
2689 : * 'inner_path' is the cheapest inner path
2690 : * 'parallel_hash' to select Parallel Hash of inner path (shared hash table)
2691 : * 'restrict_clauses' are the RestrictInfo nodes to apply at the join
2692 : * 'required_outer' is the set of required outer rels
2693 : * 'hashclauses' are the RestrictInfo nodes to use as hash clauses
2694 : * (this should be a subset of the restrict_clauses list)
2695 : */
2696 : HashPath *
2697 234798 : create_hashjoin_path(PlannerInfo *root,
2698 : RelOptInfo *joinrel,
2699 : JoinType jointype,
2700 : JoinCostWorkspace *workspace,
2701 : JoinPathExtraData *extra,
2702 : Path *outer_path,
2703 : Path *inner_path,
2704 : bool parallel_hash,
2705 : List *restrict_clauses,
2706 : Relids required_outer,
2707 : List *hashclauses)
2708 : {
2709 234798 : HashPath *pathnode = makeNode(HashPath);
2710 :
2711 234798 : pathnode->jpath.path.pathtype = T_HashJoin;
2712 234798 : pathnode->jpath.path.parent = joinrel;
2713 234798 : pathnode->jpath.path.pathtarget = joinrel->reltarget;
2714 234798 : pathnode->jpath.path.param_info =
2715 234798 : get_joinrel_parampathinfo(root,
2716 : joinrel,
2717 : outer_path,
2718 : inner_path,
2719 : extra->sjinfo,
2720 : required_outer,
2721 : &restrict_clauses);
2722 234798 : pathnode->jpath.path.parallel_aware =
2723 234798 : joinrel->consider_parallel && parallel_hash;
2724 675078 : pathnode->jpath.path.parallel_safe = joinrel->consider_parallel &&
2725 234798 : outer_path->parallel_safe && inner_path->parallel_safe;
2726 : /* This is a foolish way to estimate parallel_workers, but for now... */
2727 234798 : pathnode->jpath.path.parallel_workers = outer_path->parallel_workers;
2728 :
2729 : /*
2730 : * A hashjoin never has pathkeys, since its output ordering is
2731 : * unpredictable due to possible batching. XXX If the inner relation is
2732 : * small enough, we could instruct the executor that it must not batch,
2733 : * and then we could assume that the output inherits the outer relation's
2734 : * ordering, which might save a sort step. However there is considerable
2735 : * downside if our estimate of the inner relation size is badly off. For
2736 : * the moment we don't risk it. (Note also that if we wanted to take this
2737 : * seriously, joinpath.c would have to consider many more paths for the
2738 : * outer rel than it does now.)
2739 : */
2740 234798 : pathnode->jpath.path.pathkeys = NIL;
2741 234798 : pathnode->jpath.jointype = jointype;
2742 234798 : pathnode->jpath.inner_unique = extra->inner_unique;
2743 234798 : pathnode->jpath.outerjoinpath = outer_path;
2744 234798 : pathnode->jpath.innerjoinpath = inner_path;
2745 234798 : pathnode->jpath.joinrestrictinfo = restrict_clauses;
2746 234798 : pathnode->path_hashclauses = hashclauses;
2747 : /* final_cost_hashjoin will fill in pathnode->num_batches */
2748 :
2749 234798 : final_cost_hashjoin(root, pathnode, workspace, extra);
2750 :
2751 234798 : return pathnode;
2752 : }
2753 :
2754 : /*
2755 : * create_projection_path
2756 : * Creates a pathnode that represents performing a projection.
2757 : *
2758 : * 'rel' is the parent relation associated with the result
2759 : * 'subpath' is the path representing the source of data
2760 : * 'target' is the PathTarget to be computed
2761 : */
2762 : ProjectionPath *
2763 374708 : create_projection_path(PlannerInfo *root,
2764 : RelOptInfo *rel,
2765 : Path *subpath,
2766 : PathTarget *target)
2767 : {
2768 374708 : ProjectionPath *pathnode = makeNode(ProjectionPath);
2769 : PathTarget *oldtarget;
2770 :
2771 : /*
2772 : * We mustn't put a ProjectionPath directly above another; it's useless
2773 : * and will confuse create_projection_plan. Rather than making sure all
2774 : * callers handle that, let's implement it here, by stripping off any
2775 : * ProjectionPath in what we're given. Given this rule, there won't be
2776 : * more than one.
2777 : */
2778 374708 : if (IsA(subpath, ProjectionPath))
2779 : {
2780 12 : ProjectionPath *subpp = (ProjectionPath *) subpath;
2781 :
2782 : Assert(subpp->path.parent == rel);
2783 12 : subpath = subpp->subpath;
2784 : Assert(!IsA(subpath, ProjectionPath));
2785 : }
2786 :
2787 374708 : pathnode->path.pathtype = T_Result;
2788 374708 : pathnode->path.parent = rel;
2789 374708 : pathnode->path.pathtarget = target;
2790 : /* For now, assume we are above any joins, so no parameterization */
2791 374708 : pathnode->path.param_info = NULL;
2792 374708 : pathnode->path.parallel_aware = false;
2793 832416 : pathnode->path.parallel_safe = rel->consider_parallel &&
2794 457312 : subpath->parallel_safe &&
2795 82604 : is_parallel_safe(root, (Node *) target->exprs);
2796 374708 : pathnode->path.parallel_workers = subpath->parallel_workers;
2797 : /* Projection does not change the sort order */
2798 374708 : pathnode->path.pathkeys = subpath->pathkeys;
2799 :
2800 374708 : pathnode->subpath = subpath;
2801 :
2802 : /*
2803 : * We might not need a separate Result node. If the input plan node type
2804 : * can project, we can just tell it to project something else. Or, if it
2805 : * can't project but the desired target has the same expression list as
2806 : * what the input will produce anyway, we can still give it the desired
2807 : * tlist (possibly changing its ressortgroupref labels, but nothing else).
2808 : * Note: in the latter case, create_projection_plan has to recheck our
2809 : * conclusion; see comments therein.
2810 : */
2811 374708 : oldtarget = subpath->pathtarget;
2812 376990 : if (is_projection_capable_path(subpath) ||
2813 2282 : equal(oldtarget->exprs, target->exprs))
2814 : {
2815 : /* No separate Result node needed */
2816 372552 : pathnode->dummypp = true;
2817 :
2818 : /*
2819 : * Set cost of plan as subpath's cost, adjusted for tlist replacement.
2820 : */
2821 372552 : pathnode->path.rows = subpath->rows;
2822 372552 : pathnode->path.disabled_nodes = subpath->disabled_nodes;
2823 372552 : pathnode->path.startup_cost = subpath->startup_cost +
2824 372552 : (target->cost.startup - oldtarget->cost.startup);
2825 372552 : pathnode->path.total_cost = subpath->total_cost +
2826 372552 : (target->cost.startup - oldtarget->cost.startup) +
2827 372552 : (target->cost.per_tuple - oldtarget->cost.per_tuple) * subpath->rows;
2828 : }
2829 : else
2830 : {
2831 : /* We really do need the Result node */
2832 2156 : pathnode->dummypp = false;
2833 :
2834 : /*
2835 : * The Result node's cost is cpu_tuple_cost per row, plus the cost of
2836 : * evaluating the tlist. There is no qual to worry about.
2837 : */
2838 2156 : pathnode->path.rows = subpath->rows;
2839 2156 : pathnode->path.disabled_nodes = subpath->disabled_nodes;
2840 2156 : pathnode->path.startup_cost = subpath->startup_cost +
2841 2156 : target->cost.startup;
2842 2156 : pathnode->path.total_cost = subpath->total_cost +
2843 2156 : target->cost.startup +
2844 2156 : (cpu_tuple_cost + target->cost.per_tuple) * subpath->rows;
2845 : }
2846 :
2847 374708 : return pathnode;
2848 : }
2849 :
2850 : /*
2851 : * apply_projection_to_path
2852 : * Add a projection step, or just apply the target directly to given path.
2853 : *
2854 : * This has the same net effect as create_projection_path(), except that if
2855 : * a separate Result plan node isn't needed, we just replace the given path's
2856 : * pathtarget with the desired one. This must be used only when the caller
2857 : * knows that the given path isn't referenced elsewhere and so can be modified
2858 : * in-place.
2859 : *
2860 : * If the input path is a GatherPath or GatherMergePath, we try to push the
2861 : * new target down to its input as well; this is a yet more invasive
2862 : * modification of the input path, which create_projection_path() can't do.
2863 : *
2864 : * Note that we mustn't change the source path's parent link; so when it is
2865 : * add_path'd to "rel" things will be a bit inconsistent. So far that has
2866 : * not caused any trouble.
2867 : *
2868 : * 'rel' is the parent relation associated with the result
2869 : * 'path' is the path representing the source of data
2870 : * 'target' is the PathTarget to be computed
2871 : */
2872 : Path *
2873 10626 : apply_projection_to_path(PlannerInfo *root,
2874 : RelOptInfo *rel,
2875 : Path *path,
2876 : PathTarget *target)
2877 : {
2878 : QualCost oldcost;
2879 :
2880 : /*
2881 : * If given path can't project, we might need a Result node, so make a
2882 : * separate ProjectionPath.
2883 : */
2884 10626 : if (!is_projection_capable_path(path))
2885 1526 : return (Path *) create_projection_path(root, rel, path, target);
2886 :
2887 : /*
2888 : * We can just jam the desired tlist into the existing path, being sure to
2889 : * update its cost estimates appropriately.
2890 : */
2891 9100 : oldcost = path->pathtarget->cost;
2892 9100 : path->pathtarget = target;
2893 :
2894 9100 : path->startup_cost += target->cost.startup - oldcost.startup;
2895 9100 : path->total_cost += target->cost.startup - oldcost.startup +
2896 9100 : (target->cost.per_tuple - oldcost.per_tuple) * path->rows;
2897 :
2898 : /*
2899 : * If the path happens to be a Gather or GatherMerge path, we'd like to
2900 : * arrange for the subpath to return the required target list so that
2901 : * workers can help project. But if there is something that is not
2902 : * parallel-safe in the target expressions, then we can't.
2903 : */
2904 9124 : if ((IsA(path, GatherPath) || IsA(path, GatherMergePath)) &&
2905 24 : is_parallel_safe(root, (Node *) target->exprs))
2906 : {
2907 : /*
2908 : * We always use create_projection_path here, even if the subpath is
2909 : * projection-capable, so as to avoid modifying the subpath in place.
2910 : * It seems unlikely at present that there could be any other
2911 : * references to the subpath, but better safe than sorry.
2912 : *
2913 : * Note that we don't change the parallel path's cost estimates; it
2914 : * might be appropriate to do so, to reflect the fact that the bulk of
2915 : * the target evaluation will happen in workers.
2916 : */
2917 24 : if (IsA(path, GatherPath))
2918 : {
2919 0 : GatherPath *gpath = (GatherPath *) path;
2920 :
2921 0 : gpath->subpath = (Path *)
2922 0 : create_projection_path(root,
2923 0 : gpath->subpath->parent,
2924 : gpath->subpath,
2925 : target);
2926 : }
2927 : else
2928 : {
2929 24 : GatherMergePath *gmpath = (GatherMergePath *) path;
2930 :
2931 24 : gmpath->subpath = (Path *)
2932 24 : create_projection_path(root,
2933 24 : gmpath->subpath->parent,
2934 : gmpath->subpath,
2935 : target);
2936 : }
2937 : }
2938 9076 : else if (path->parallel_safe &&
2939 1906 : !is_parallel_safe(root, (Node *) target->exprs))
2940 : {
2941 : /*
2942 : * We're inserting a parallel-restricted target list into a path
2943 : * currently marked parallel-safe, so we have to mark it as no longer
2944 : * safe.
2945 : */
2946 12 : path->parallel_safe = false;
2947 : }
2948 :
2949 9100 : return path;
2950 : }
2951 :
2952 : /*
2953 : * create_set_projection_path
2954 : * Creates a pathnode that represents performing a projection that
2955 : * includes set-returning functions.
2956 : *
2957 : * 'rel' is the parent relation associated with the result
2958 : * 'subpath' is the path representing the source of data
2959 : * 'target' is the PathTarget to be computed
2960 : */
2961 : ProjectSetPath *
2962 8756 : create_set_projection_path(PlannerInfo *root,
2963 : RelOptInfo *rel,
2964 : Path *subpath,
2965 : PathTarget *target)
2966 : {
2967 8756 : ProjectSetPath *pathnode = makeNode(ProjectSetPath);
2968 : double tlist_rows;
2969 : ListCell *lc;
2970 :
2971 8756 : pathnode->path.pathtype = T_ProjectSet;
2972 8756 : pathnode->path.parent = rel;
2973 8756 : pathnode->path.pathtarget = target;
2974 : /* For now, assume we are above any joins, so no parameterization */
2975 8756 : pathnode->path.param_info = NULL;
2976 8756 : pathnode->path.parallel_aware = false;
2977 19250 : pathnode->path.parallel_safe = rel->consider_parallel &&
2978 10458 : subpath->parallel_safe &&
2979 1702 : is_parallel_safe(root, (Node *) target->exprs);
2980 8756 : pathnode->path.parallel_workers = subpath->parallel_workers;
2981 : /* Projection does not change the sort order XXX? */
2982 8756 : pathnode->path.pathkeys = subpath->pathkeys;
2983 :
2984 8756 : pathnode->subpath = subpath;
2985 :
2986 : /*
2987 : * Estimate number of rows produced by SRFs for each row of input; if
2988 : * there's more than one in this node, use the maximum.
2989 : */
2990 8756 : tlist_rows = 1;
2991 19394 : foreach(lc, target->exprs)
2992 : {
2993 10638 : Node *node = (Node *) lfirst(lc);
2994 : double itemrows;
2995 :
2996 10638 : itemrows = expression_returns_set_rows(root, node);
2997 10638 : if (tlist_rows < itemrows)
2998 8490 : tlist_rows = itemrows;
2999 : }
3000 :
3001 : /*
3002 : * In addition to the cost of evaluating the tlist, charge cpu_tuple_cost
3003 : * per input row, and half of cpu_tuple_cost for each added output row.
3004 : * This is slightly bizarre maybe, but it's what 9.6 did; we may revisit
3005 : * this estimate later.
3006 : */
3007 8756 : pathnode->path.disabled_nodes = subpath->disabled_nodes;
3008 8756 : pathnode->path.rows = subpath->rows * tlist_rows;
3009 8756 : pathnode->path.startup_cost = subpath->startup_cost +
3010 8756 : target->cost.startup;
3011 8756 : pathnode->path.total_cost = subpath->total_cost +
3012 8756 : target->cost.startup +
3013 8756 : (cpu_tuple_cost + target->cost.per_tuple) * subpath->rows +
3014 8756 : (pathnode->path.rows - subpath->rows) * cpu_tuple_cost / 2;
3015 :
3016 8756 : return pathnode;
3017 : }
3018 :
3019 : /*
3020 : * create_incremental_sort_path
3021 : * Creates a pathnode that represents performing an incremental sort.
3022 : *
3023 : * 'rel' is the parent relation associated with the result
3024 : * 'subpath' is the path representing the source of data
3025 : * 'pathkeys' represents the desired sort order
3026 : * 'presorted_keys' is the number of keys by which the input path is
3027 : * already sorted
3028 : * 'limit_tuples' is the estimated bound on the number of output tuples,
3029 : * or -1 if no LIMIT or couldn't estimate
3030 : */
3031 : IncrementalSortPath *
3032 8014 : create_incremental_sort_path(PlannerInfo *root,
3033 : RelOptInfo *rel,
3034 : Path *subpath,
3035 : List *pathkeys,
3036 : int presorted_keys,
3037 : double limit_tuples)
3038 : {
3039 8014 : IncrementalSortPath *sort = makeNode(IncrementalSortPath);
3040 8014 : SortPath *pathnode = &sort->spath;
3041 :
3042 8014 : pathnode->path.pathtype = T_IncrementalSort;
3043 8014 : pathnode->path.parent = rel;
3044 : /* Sort doesn't project, so use source path's pathtarget */
3045 8014 : pathnode->path.pathtarget = subpath->pathtarget;
3046 : /* For now, assume we are above any joins, so no parameterization */
3047 8014 : pathnode->path.param_info = NULL;
3048 8014 : pathnode->path.parallel_aware = false;
3049 11554 : pathnode->path.parallel_safe = rel->consider_parallel &&
3050 3540 : subpath->parallel_safe;
3051 8014 : pathnode->path.parallel_workers = subpath->parallel_workers;
3052 8014 : pathnode->path.pathkeys = pathkeys;
3053 :
3054 8014 : pathnode->subpath = subpath;
3055 :
3056 8014 : cost_incremental_sort(&pathnode->path,
3057 : root, pathkeys, presorted_keys,
3058 : subpath->disabled_nodes,
3059 : subpath->startup_cost,
3060 : subpath->total_cost,
3061 : subpath->rows,
3062 8014 : subpath->pathtarget->width,
3063 : 0.0, /* XXX comparison_cost shouldn't be 0? */
3064 : work_mem, limit_tuples);
3065 :
3066 8014 : sort->nPresortedCols = presorted_keys;
3067 :
3068 8014 : return sort;
3069 : }
3070 :
3071 : /*
3072 : * create_sort_path
3073 : * Creates a pathnode that represents performing an explicit sort.
3074 : *
3075 : * 'rel' is the parent relation associated with the result
3076 : * 'subpath' is the path representing the source of data
3077 : * 'pathkeys' represents the desired sort order
3078 : * 'limit_tuples' is the estimated bound on the number of output tuples,
3079 : * or -1 if no LIMIT or couldn't estimate
3080 : */
3081 : SortPath *
3082 87516 : create_sort_path(PlannerInfo *root,
3083 : RelOptInfo *rel,
3084 : Path *subpath,
3085 : List *pathkeys,
3086 : double limit_tuples)
3087 : {
3088 87516 : SortPath *pathnode = makeNode(SortPath);
3089 :
3090 87516 : pathnode->path.pathtype = T_Sort;
3091 87516 : pathnode->path.parent = rel;
3092 : /* Sort doesn't project, so use source path's pathtarget */
3093 87516 : pathnode->path.pathtarget = subpath->pathtarget;
3094 : /* For now, assume we are above any joins, so no parameterization */
3095 87516 : pathnode->path.param_info = NULL;
3096 87516 : pathnode->path.parallel_aware = false;
3097 147840 : pathnode->path.parallel_safe = rel->consider_parallel &&
3098 60324 : subpath->parallel_safe;
3099 87516 : pathnode->path.parallel_workers = subpath->parallel_workers;
3100 87516 : pathnode->path.pathkeys = pathkeys;
3101 :
3102 87516 : pathnode->subpath = subpath;
3103 :
3104 87516 : cost_sort(&pathnode->path, root, pathkeys,
3105 : subpath->disabled_nodes,
3106 : subpath->total_cost,
3107 : subpath->rows,
3108 87516 : subpath->pathtarget->width,
3109 : 0.0, /* XXX comparison_cost shouldn't be 0? */
3110 : work_mem, limit_tuples);
3111 :
3112 87516 : return pathnode;
3113 : }
3114 :
3115 : /*
3116 : * create_group_path
3117 : * Creates a pathnode that represents performing grouping of presorted input
3118 : *
3119 : * 'rel' is the parent relation associated with the result
3120 : * 'subpath' is the path representing the source of data
3121 : * 'target' is the PathTarget to be computed
3122 : * 'groupClause' is a list of SortGroupClause's representing the grouping
3123 : * 'qual' is the HAVING quals if any
3124 : * 'numGroups' is the estimated number of groups
3125 : */
3126 : GroupPath *
3127 1112 : create_group_path(PlannerInfo *root,
3128 : RelOptInfo *rel,
3129 : Path *subpath,
3130 : List *groupClause,
3131 : List *qual,
3132 : double numGroups)
3133 : {
3134 1112 : GroupPath *pathnode = makeNode(GroupPath);
3135 1112 : PathTarget *target = rel->reltarget;
3136 :
3137 1112 : pathnode->path.pathtype = T_Group;
3138 1112 : pathnode->path.parent = rel;
3139 1112 : pathnode->path.pathtarget = target;
3140 : /* For now, assume we are above any joins, so no parameterization */
3141 1112 : pathnode->path.param_info = NULL;
3142 1112 : pathnode->path.parallel_aware = false;
3143 1820 : pathnode->path.parallel_safe = rel->consider_parallel &&
3144 708 : subpath->parallel_safe;
3145 1112 : pathnode->path.parallel_workers = subpath->parallel_workers;
3146 : /* Group doesn't change sort ordering */
3147 1112 : pathnode->path.pathkeys = subpath->pathkeys;
3148 :
3149 1112 : pathnode->subpath = subpath;
3150 :
3151 1112 : pathnode->groupClause = groupClause;
3152 1112 : pathnode->qual = qual;
3153 :
3154 1112 : cost_group(&pathnode->path, root,
3155 : list_length(groupClause),
3156 : numGroups,
3157 : qual,
3158 : subpath->disabled_nodes,
3159 : subpath->startup_cost, subpath->total_cost,
3160 : subpath->rows);
3161 :
3162 : /* add tlist eval cost for each output row */
3163 1112 : pathnode->path.startup_cost += target->cost.startup;
3164 1112 : pathnode->path.total_cost += target->cost.startup +
3165 1112 : target->cost.per_tuple * pathnode->path.rows;
3166 :
3167 1112 : return pathnode;
3168 : }
3169 :
3170 : /*
3171 : * create_upper_unique_path
3172 : * Creates a pathnode that represents performing an explicit Unique step
3173 : * on presorted input.
3174 : *
3175 : * This produces a Unique plan node, but the use-case is so different from
3176 : * create_unique_path that it doesn't seem worth trying to merge the two.
3177 : *
3178 : * 'rel' is the parent relation associated with the result
3179 : * 'subpath' is the path representing the source of data
3180 : * 'numCols' is the number of grouping columns
3181 : * 'numGroups' is the estimated number of groups
3182 : *
3183 : * The input path must be sorted on the grouping columns, plus possibly
3184 : * additional columns; so the first numCols pathkeys are the grouping columns
3185 : */
3186 : UpperUniquePath *
3187 6854 : create_upper_unique_path(PlannerInfo *root,
3188 : RelOptInfo *rel,
3189 : Path *subpath,
3190 : int numCols,
3191 : double numGroups)
3192 : {
3193 6854 : UpperUniquePath *pathnode = makeNode(UpperUniquePath);
3194 :
3195 6854 : pathnode->path.pathtype = T_Unique;
3196 6854 : pathnode->path.parent = rel;
3197 : /* Unique doesn't project, so use source path's pathtarget */
3198 6854 : pathnode->path.pathtarget = subpath->pathtarget;
3199 : /* For now, assume we are above any joins, so no parameterization */
3200 6854 : pathnode->path.param_info = NULL;
3201 6854 : pathnode->path.parallel_aware = false;
3202 10506 : pathnode->path.parallel_safe = rel->consider_parallel &&
3203 3652 : subpath->parallel_safe;
3204 6854 : pathnode->path.parallel_workers = subpath->parallel_workers;
3205 : /* Unique doesn't change the input ordering */
3206 6854 : pathnode->path.pathkeys = subpath->pathkeys;
3207 :
3208 6854 : pathnode->subpath = subpath;
3209 6854 : pathnode->numkeys = numCols;
3210 :
3211 : /*
3212 : * Charge one cpu_operator_cost per comparison per input tuple. We assume
3213 : * all columns get compared at most of the tuples. (XXX probably this is
3214 : * an overestimate.)
3215 : */
3216 6854 : pathnode->path.disabled_nodes = subpath->disabled_nodes;
3217 6854 : pathnode->path.startup_cost = subpath->startup_cost;
3218 6854 : pathnode->path.total_cost = subpath->total_cost +
3219 6854 : cpu_operator_cost * subpath->rows * numCols;
3220 6854 : pathnode->path.rows = numGroups;
3221 :
3222 6854 : return pathnode;
3223 : }
3224 :
3225 : /*
3226 : * create_agg_path
3227 : * Creates a pathnode that represents performing aggregation/grouping
3228 : *
3229 : * 'rel' is the parent relation associated with the result
3230 : * 'subpath' is the path representing the source of data
3231 : * 'target' is the PathTarget to be computed
3232 : * 'aggstrategy' is the Agg node's basic implementation strategy
3233 : * 'aggsplit' is the Agg node's aggregate-splitting mode
3234 : * 'groupClause' is a list of SortGroupClause's representing the grouping
3235 : * 'qual' is the HAVING quals if any
3236 : * 'aggcosts' contains cost info about the aggregate functions to be computed
3237 : * 'numGroups' is the estimated number of groups (1 if not grouping)
3238 : */
3239 : AggPath *
3240 57520 : create_agg_path(PlannerInfo *root,
3241 : RelOptInfo *rel,
3242 : Path *subpath,
3243 : PathTarget *target,
3244 : AggStrategy aggstrategy,
3245 : AggSplit aggsplit,
3246 : List *groupClause,
3247 : List *qual,
3248 : const AggClauseCosts *aggcosts,
3249 : double numGroups)
3250 : {
3251 57520 : AggPath *pathnode = makeNode(AggPath);
3252 :
3253 57520 : pathnode->path.pathtype = T_Agg;
3254 57520 : pathnode->path.parent = rel;
3255 57520 : pathnode->path.pathtarget = target;
3256 : /* For now, assume we are above any joins, so no parameterization */
3257 57520 : pathnode->path.param_info = NULL;
3258 57520 : pathnode->path.parallel_aware = false;
3259 98068 : pathnode->path.parallel_safe = rel->consider_parallel &&
3260 40548 : subpath->parallel_safe;
3261 57520 : pathnode->path.parallel_workers = subpath->parallel_workers;
3262 :
3263 57520 : if (aggstrategy == AGG_SORTED)
3264 : {
3265 : /*
3266 : * Attempt to preserve the order of the subpath. Additional pathkeys
3267 : * may have been added in adjust_group_pathkeys_for_groupagg() to
3268 : * support ORDER BY / DISTINCT aggregates. Pathkeys added there
3269 : * belong to columns within the aggregate function, so we must strip
3270 : * these additional pathkeys off as those columns are unavailable
3271 : * above the aggregate node.
3272 : */
3273 7784 : if (list_length(subpath->pathkeys) > root->num_groupby_pathkeys)
3274 326 : pathnode->path.pathkeys = list_copy_head(subpath->pathkeys,
3275 : root->num_groupby_pathkeys);
3276 : else
3277 7458 : pathnode->path.pathkeys = subpath->pathkeys; /* preserves order */
3278 : }
3279 : else
3280 49736 : pathnode->path.pathkeys = NIL; /* output is unordered */
3281 :
3282 57520 : pathnode->subpath = subpath;
3283 :
3284 57520 : pathnode->aggstrategy = aggstrategy;
3285 57520 : pathnode->aggsplit = aggsplit;
3286 57520 : pathnode->numGroups = numGroups;
3287 57520 : pathnode->transitionSpace = aggcosts ? aggcosts->transitionSpace : 0;
3288 57520 : pathnode->groupClause = groupClause;
3289 57520 : pathnode->qual = qual;
3290 :
3291 57520 : cost_agg(&pathnode->path, root,
3292 : aggstrategy, aggcosts,
3293 : list_length(groupClause), numGroups,
3294 : qual,
3295 : subpath->disabled_nodes,
3296 : subpath->startup_cost, subpath->total_cost,
3297 57520 : subpath->rows, subpath->pathtarget->width);
3298 :
3299 : /* add tlist eval cost for each output row */
3300 57520 : pathnode->path.startup_cost += target->cost.startup;
3301 57520 : pathnode->path.total_cost += target->cost.startup +
3302 57520 : target->cost.per_tuple * pathnode->path.rows;
3303 :
3304 57520 : return pathnode;
3305 : }
3306 :
3307 : /*
3308 : * create_groupingsets_path
3309 : * Creates a pathnode that represents performing GROUPING SETS aggregation
3310 : *
3311 : * GroupingSetsPath represents sorted grouping with one or more grouping sets.
3312 : * The input path's result must be sorted to match the last entry in
3313 : * rollup_groupclauses.
3314 : *
3315 : * 'rel' is the parent relation associated with the result
3316 : * 'subpath' is the path representing the source of data
3317 : * 'target' is the PathTarget to be computed
3318 : * 'having_qual' is the HAVING quals if any
3319 : * 'rollups' is a list of RollupData nodes
3320 : * 'agg_costs' contains cost info about the aggregate functions to be computed
3321 : */
3322 : GroupingSetsPath *
3323 2056 : create_groupingsets_path(PlannerInfo *root,
3324 : RelOptInfo *rel,
3325 : Path *subpath,
3326 : List *having_qual,
3327 : AggStrategy aggstrategy,
3328 : List *rollups,
3329 : const AggClauseCosts *agg_costs)
3330 : {
3331 2056 : GroupingSetsPath *pathnode = makeNode(GroupingSetsPath);
3332 2056 : PathTarget *target = rel->reltarget;
3333 : ListCell *lc;
3334 2056 : bool is_first = true;
3335 2056 : bool is_first_sort = true;
3336 :
3337 : /* The topmost generated Plan node will be an Agg */
3338 2056 : pathnode->path.pathtype = T_Agg;
3339 2056 : pathnode->path.parent = rel;
3340 2056 : pathnode->path.pathtarget = target;
3341 2056 : pathnode->path.param_info = subpath->param_info;
3342 2056 : pathnode->path.parallel_aware = false;
3343 2974 : pathnode->path.parallel_safe = rel->consider_parallel &&
3344 918 : subpath->parallel_safe;
3345 2056 : pathnode->path.parallel_workers = subpath->parallel_workers;
3346 2056 : pathnode->subpath = subpath;
3347 :
3348 : /*
3349 : * Simplify callers by downgrading AGG_SORTED to AGG_PLAIN, and AGG_MIXED
3350 : * to AGG_HASHED, here if possible.
3351 : */
3352 2940 : if (aggstrategy == AGG_SORTED &&
3353 884 : list_length(rollups) == 1 &&
3354 458 : ((RollupData *) linitial(rollups))->groupClause == NIL)
3355 42 : aggstrategy = AGG_PLAIN;
3356 :
3357 2948 : if (aggstrategy == AGG_MIXED &&
3358 892 : list_length(rollups) == 1)
3359 0 : aggstrategy = AGG_HASHED;
3360 :
3361 : /*
3362 : * Output will be in sorted order by group_pathkeys if, and only if, there
3363 : * is a single rollup operation on a non-empty list of grouping
3364 : * expressions.
3365 : */
3366 2056 : if (aggstrategy == AGG_SORTED && list_length(rollups) == 1)
3367 416 : pathnode->path.pathkeys = root->group_pathkeys;
3368 : else
3369 1640 : pathnode->path.pathkeys = NIL;
3370 :
3371 2056 : pathnode->aggstrategy = aggstrategy;
3372 2056 : pathnode->rollups = rollups;
3373 2056 : pathnode->qual = having_qual;
3374 2056 : pathnode->transitionSpace = agg_costs ? agg_costs->transitionSpace : 0;
3375 :
3376 : Assert(rollups != NIL);
3377 : Assert(aggstrategy != AGG_PLAIN || list_length(rollups) == 1);
3378 : Assert(aggstrategy != AGG_MIXED || list_length(rollups) > 1);
3379 :
3380 7092 : foreach(lc, rollups)
3381 : {
3382 5036 : RollupData *rollup = lfirst(lc);
3383 5036 : List *gsets = rollup->gsets;
3384 5036 : int numGroupCols = list_length(linitial(gsets));
3385 :
3386 : /*
3387 : * In AGG_SORTED or AGG_PLAIN mode, the first rollup takes the
3388 : * (already-sorted) input, and following ones do their own sort.
3389 : *
3390 : * In AGG_HASHED mode, there is one rollup for each grouping set.
3391 : *
3392 : * In AGG_MIXED mode, the first rollups are hashed, the first
3393 : * non-hashed one takes the (already-sorted) input, and following ones
3394 : * do their own sort.
3395 : */
3396 5036 : if (is_first)
3397 : {
3398 2056 : cost_agg(&pathnode->path, root,
3399 : aggstrategy,
3400 : agg_costs,
3401 : numGroupCols,
3402 : rollup->numGroups,
3403 : having_qual,
3404 : subpath->disabled_nodes,
3405 : subpath->startup_cost,
3406 : subpath->total_cost,
3407 : subpath->rows,
3408 2056 : subpath->pathtarget->width);
3409 2056 : is_first = false;
3410 2056 : if (!rollup->is_hashed)
3411 884 : is_first_sort = false;
3412 : }
3413 : else
3414 : {
3415 : Path sort_path; /* dummy for result of cost_sort */
3416 : Path agg_path; /* dummy for result of cost_agg */
3417 :
3418 2980 : if (rollup->is_hashed || is_first_sort)
3419 : {
3420 : /*
3421 : * Account for cost of aggregation, but don't charge input
3422 : * cost again
3423 : */
3424 2302 : cost_agg(&agg_path, root,
3425 2302 : rollup->is_hashed ? AGG_HASHED : AGG_SORTED,
3426 : agg_costs,
3427 : numGroupCols,
3428 : rollup->numGroups,
3429 : having_qual,
3430 : 0, 0.0, 0.0,
3431 : subpath->rows,
3432 2302 : subpath->pathtarget->width);
3433 2302 : if (!rollup->is_hashed)
3434 892 : is_first_sort = false;
3435 : }
3436 : else
3437 : {
3438 : /* Account for cost of sort, but don't charge input cost again */
3439 678 : cost_sort(&sort_path, root, NIL, 0,
3440 : 0.0,
3441 : subpath->rows,
3442 678 : subpath->pathtarget->width,
3443 : 0.0,
3444 : work_mem,
3445 : -1.0);
3446 :
3447 : /* Account for cost of aggregation */
3448 :
3449 678 : cost_agg(&agg_path, root,
3450 : AGG_SORTED,
3451 : agg_costs,
3452 : numGroupCols,
3453 : rollup->numGroups,
3454 : having_qual,
3455 : sort_path.disabled_nodes,
3456 : sort_path.startup_cost,
3457 : sort_path.total_cost,
3458 : sort_path.rows,
3459 678 : subpath->pathtarget->width);
3460 : }
3461 :
3462 2980 : pathnode->path.disabled_nodes += agg_path.disabled_nodes;
3463 2980 : pathnode->path.total_cost += agg_path.total_cost;
3464 2980 : pathnode->path.rows += agg_path.rows;
3465 : }
3466 : }
3467 :
3468 : /* add tlist eval cost for each output row */
3469 2056 : pathnode->path.startup_cost += target->cost.startup;
3470 2056 : pathnode->path.total_cost += target->cost.startup +
3471 2056 : target->cost.per_tuple * pathnode->path.rows;
3472 :
3473 2056 : return pathnode;
3474 : }
3475 :
3476 : /*
3477 : * create_minmaxagg_path
3478 : * Creates a pathnode that represents computation of MIN/MAX aggregates
3479 : *
3480 : * 'rel' is the parent relation associated with the result
3481 : * 'target' is the PathTarget to be computed
3482 : * 'mmaggregates' is a list of MinMaxAggInfo structs
3483 : * 'quals' is the HAVING quals if any
3484 : */
3485 : MinMaxAggPath *
3486 410 : create_minmaxagg_path(PlannerInfo *root,
3487 : RelOptInfo *rel,
3488 : PathTarget *target,
3489 : List *mmaggregates,
3490 : List *quals)
3491 : {
3492 410 : MinMaxAggPath *pathnode = makeNode(MinMaxAggPath);
3493 : Cost initplan_cost;
3494 410 : int initplan_disabled_nodes = 0;
3495 : ListCell *lc;
3496 :
3497 : /* The topmost generated Plan node will be a Result */
3498 410 : pathnode->path.pathtype = T_Result;
3499 410 : pathnode->path.parent = rel;
3500 410 : pathnode->path.pathtarget = target;
3501 : /* For now, assume we are above any joins, so no parameterization */
3502 410 : pathnode->path.param_info = NULL;
3503 410 : pathnode->path.parallel_aware = false;
3504 410 : pathnode->path.parallel_safe = true; /* might change below */
3505 410 : pathnode->path.parallel_workers = 0;
3506 : /* Result is one unordered row */
3507 410 : pathnode->path.rows = 1;
3508 410 : pathnode->path.pathkeys = NIL;
3509 :
3510 410 : pathnode->mmaggregates = mmaggregates;
3511 410 : pathnode->quals = quals;
3512 :
3513 : /* Calculate cost of all the initplans, and check parallel safety */
3514 410 : initplan_cost = 0;
3515 856 : foreach(lc, mmaggregates)
3516 : {
3517 446 : MinMaxAggInfo *mminfo = (MinMaxAggInfo *) lfirst(lc);
3518 :
3519 446 : initplan_disabled_nodes += mminfo->path->disabled_nodes;
3520 446 : initplan_cost += mminfo->pathcost;
3521 446 : if (!mminfo->path->parallel_safe)
3522 110 : pathnode->path.parallel_safe = false;
3523 : }
3524 :
3525 : /* add tlist eval cost for each output row, plus cpu_tuple_cost */
3526 410 : pathnode->path.disabled_nodes = initplan_disabled_nodes;
3527 410 : pathnode->path.startup_cost = initplan_cost + target->cost.startup;
3528 410 : pathnode->path.total_cost = initplan_cost + target->cost.startup +
3529 410 : target->cost.per_tuple + cpu_tuple_cost;
3530 :
3531 : /*
3532 : * Add cost of qual, if any --- but we ignore its selectivity, since our
3533 : * rowcount estimate should be 1 no matter what the qual is.
3534 : */
3535 410 : if (quals)
3536 : {
3537 : QualCost qual_cost;
3538 :
3539 0 : cost_qual_eval(&qual_cost, quals, root);
3540 0 : pathnode->path.startup_cost += qual_cost.startup;
3541 0 : pathnode->path.total_cost += qual_cost.startup + qual_cost.per_tuple;
3542 : }
3543 :
3544 : /*
3545 : * If the initplans were all parallel-safe, also check safety of the
3546 : * target and quals. (The Result node itself isn't parallelizable, but if
3547 : * we are in a subquery then it can be useful for the outer query to know
3548 : * that this one is parallel-safe.)
3549 : */
3550 410 : if (pathnode->path.parallel_safe)
3551 300 : pathnode->path.parallel_safe =
3552 600 : is_parallel_safe(root, (Node *) target->exprs) &&
3553 300 : is_parallel_safe(root, (Node *) quals);
3554 :
3555 410 : return pathnode;
3556 : }
3557 :
3558 : /*
3559 : * create_windowagg_path
3560 : * Creates a pathnode that represents computation of window functions
3561 : *
3562 : * 'rel' is the parent relation associated with the result
3563 : * 'subpath' is the path representing the source of data
3564 : * 'target' is the PathTarget to be computed
3565 : * 'windowFuncs' is a list of WindowFunc structs
3566 : * 'runCondition' is a list of OpExprs to short-circuit WindowAgg execution
3567 : * 'winclause' is a WindowClause that is common to all the WindowFuncs
3568 : * 'qual' WindowClause.runconditions from lower-level WindowAggPaths.
3569 : * Must always be NIL when topwindow == false
3570 : * 'topwindow' pass as true only for the top-level WindowAgg. False for all
3571 : * intermediate WindowAggs.
3572 : *
3573 : * The input must be sorted according to the WindowClause's PARTITION keys
3574 : * plus ORDER BY keys.
3575 : */
3576 : WindowAggPath *
3577 2678 : create_windowagg_path(PlannerInfo *root,
3578 : RelOptInfo *rel,
3579 : Path *subpath,
3580 : PathTarget *target,
3581 : List *windowFuncs,
3582 : List *runCondition,
3583 : WindowClause *winclause,
3584 : List *qual,
3585 : bool topwindow)
3586 : {
3587 2678 : WindowAggPath *pathnode = makeNode(WindowAggPath);
3588 :
3589 : /* qual can only be set for the topwindow */
3590 : Assert(qual == NIL || topwindow);
3591 :
3592 2678 : pathnode->path.pathtype = T_WindowAgg;
3593 2678 : pathnode->path.parent = rel;
3594 2678 : pathnode->path.pathtarget = target;
3595 : /* For now, assume we are above any joins, so no parameterization */
3596 2678 : pathnode->path.param_info = NULL;
3597 2678 : pathnode->path.parallel_aware = false;
3598 2678 : pathnode->path.parallel_safe = rel->consider_parallel &&
3599 0 : subpath->parallel_safe;
3600 2678 : pathnode->path.parallel_workers = subpath->parallel_workers;
3601 : /* WindowAgg preserves the input sort order */
3602 2678 : pathnode->path.pathkeys = subpath->pathkeys;
3603 :
3604 2678 : pathnode->subpath = subpath;
3605 2678 : pathnode->winclause = winclause;
3606 2678 : pathnode->qual = qual;
3607 2678 : pathnode->runCondition = runCondition;
3608 2678 : pathnode->topwindow = topwindow;
3609 :
3610 : /*
3611 : * For costing purposes, assume that there are no redundant partitioning
3612 : * or ordering columns; it's not worth the trouble to deal with that
3613 : * corner case here. So we just pass the unmodified list lengths to
3614 : * cost_windowagg.
3615 : */
3616 2678 : cost_windowagg(&pathnode->path, root,
3617 : windowFuncs,
3618 : winclause,
3619 : subpath->disabled_nodes,
3620 : subpath->startup_cost,
3621 : subpath->total_cost,
3622 : subpath->rows);
3623 :
3624 : /* add tlist eval cost for each output row */
3625 2678 : pathnode->path.startup_cost += target->cost.startup;
3626 2678 : pathnode->path.total_cost += target->cost.startup +
3627 2678 : target->cost.per_tuple * pathnode->path.rows;
3628 :
3629 2678 : return pathnode;
3630 : }
3631 :
3632 : /*
3633 : * create_setop_path
3634 : * Creates a pathnode that represents computation of INTERSECT or EXCEPT
3635 : *
3636 : * 'rel' is the parent relation associated with the result
3637 : * 'subpath' is the path representing the source of data
3638 : * 'cmd' is the specific semantics (INTERSECT or EXCEPT, with/without ALL)
3639 : * 'strategy' is the implementation strategy (sorted or hashed)
3640 : * 'distinctList' is a list of SortGroupClause's representing the grouping
3641 : * 'flagColIdx' is the column number where the flag column will be, if any
3642 : * 'firstFlag' is the flag value for the first input relation when hashing;
3643 : * or -1 when sorting
3644 : * 'numGroups' is the estimated number of distinct groups
3645 : * 'outputRows' is the estimated number of output rows
3646 : */
3647 : SetOpPath *
3648 650 : create_setop_path(PlannerInfo *root,
3649 : RelOptInfo *rel,
3650 : Path *subpath,
3651 : SetOpCmd cmd,
3652 : SetOpStrategy strategy,
3653 : List *distinctList,
3654 : AttrNumber flagColIdx,
3655 : int firstFlag,
3656 : double numGroups,
3657 : double outputRows)
3658 : {
3659 650 : SetOpPath *pathnode = makeNode(SetOpPath);
3660 :
3661 650 : pathnode->path.pathtype = T_SetOp;
3662 650 : pathnode->path.parent = rel;
3663 : /* SetOp doesn't project, so use source path's pathtarget */
3664 650 : pathnode->path.pathtarget = subpath->pathtarget;
3665 : /* For now, assume we are above any joins, so no parameterization */
3666 650 : pathnode->path.param_info = NULL;
3667 650 : pathnode->path.parallel_aware = false;
3668 650 : pathnode->path.parallel_safe = rel->consider_parallel &&
3669 0 : subpath->parallel_safe;
3670 650 : pathnode->path.parallel_workers = subpath->parallel_workers;
3671 : /* SetOp preserves the input sort order if in sort mode */
3672 650 : pathnode->path.pathkeys =
3673 650 : (strategy == SETOP_SORTED) ? subpath->pathkeys : NIL;
3674 :
3675 650 : pathnode->subpath = subpath;
3676 650 : pathnode->cmd = cmd;
3677 650 : pathnode->strategy = strategy;
3678 650 : pathnode->distinctList = distinctList;
3679 650 : pathnode->flagColIdx = flagColIdx;
3680 650 : pathnode->firstFlag = firstFlag;
3681 650 : pathnode->numGroups = numGroups;
3682 :
3683 : /*
3684 : * Charge one cpu_operator_cost per comparison per input tuple. We assume
3685 : * all columns get compared at most of the tuples.
3686 : */
3687 650 : pathnode->path.disabled_nodes = subpath->disabled_nodes;
3688 650 : pathnode->path.startup_cost = subpath->startup_cost;
3689 1300 : pathnode->path.total_cost = subpath->total_cost +
3690 650 : cpu_operator_cost * subpath->rows * list_length(distinctList);
3691 650 : pathnode->path.rows = outputRows;
3692 :
3693 650 : return pathnode;
3694 : }
3695 :
3696 : /*
3697 : * create_recursiveunion_path
3698 : * Creates a pathnode that represents a recursive UNION node
3699 : *
3700 : * 'rel' is the parent relation associated with the result
3701 : * 'leftpath' is the source of data for the non-recursive term
3702 : * 'rightpath' is the source of data for the recursive term
3703 : * 'target' is the PathTarget to be computed
3704 : * 'distinctList' is a list of SortGroupClause's representing the grouping
3705 : * 'wtParam' is the ID of Param representing work table
3706 : * 'numGroups' is the estimated number of groups
3707 : *
3708 : * For recursive UNION ALL, distinctList is empty and numGroups is zero
3709 : */
3710 : RecursiveUnionPath *
3711 806 : create_recursiveunion_path(PlannerInfo *root,
3712 : RelOptInfo *rel,
3713 : Path *leftpath,
3714 : Path *rightpath,
3715 : PathTarget *target,
3716 : List *distinctList,
3717 : int wtParam,
3718 : double numGroups)
3719 : {
3720 806 : RecursiveUnionPath *pathnode = makeNode(RecursiveUnionPath);
3721 :
3722 806 : pathnode->path.pathtype = T_RecursiveUnion;
3723 806 : pathnode->path.parent = rel;
3724 806 : pathnode->path.pathtarget = target;
3725 : /* For now, assume we are above any joins, so no parameterization */
3726 806 : pathnode->path.param_info = NULL;
3727 806 : pathnode->path.parallel_aware = false;
3728 1612 : pathnode->path.parallel_safe = rel->consider_parallel &&
3729 806 : leftpath->parallel_safe && rightpath->parallel_safe;
3730 : /* Foolish, but we'll do it like joins for now: */
3731 806 : pathnode->path.parallel_workers = leftpath->parallel_workers;
3732 : /* RecursiveUnion result is always unsorted */
3733 806 : pathnode->path.pathkeys = NIL;
3734 :
3735 806 : pathnode->leftpath = leftpath;
3736 806 : pathnode->rightpath = rightpath;
3737 806 : pathnode->distinctList = distinctList;
3738 806 : pathnode->wtParam = wtParam;
3739 806 : pathnode->numGroups = numGroups;
3740 :
3741 806 : cost_recursive_union(&pathnode->path, leftpath, rightpath);
3742 :
3743 806 : return pathnode;
3744 : }
3745 :
3746 : /*
3747 : * create_lockrows_path
3748 : * Creates a pathnode that represents acquiring row locks
3749 : *
3750 : * 'rel' is the parent relation associated with the result
3751 : * 'subpath' is the path representing the source of data
3752 : * 'rowMarks' is a list of PlanRowMark's
3753 : * 'epqParam' is the ID of Param for EvalPlanQual re-eval
3754 : */
3755 : LockRowsPath *
3756 8194 : create_lockrows_path(PlannerInfo *root, RelOptInfo *rel,
3757 : Path *subpath, List *rowMarks, int epqParam)
3758 : {
3759 8194 : LockRowsPath *pathnode = makeNode(LockRowsPath);
3760 :
3761 8194 : pathnode->path.pathtype = T_LockRows;
3762 8194 : pathnode->path.parent = rel;
3763 : /* LockRows doesn't project, so use source path's pathtarget */
3764 8194 : pathnode->path.pathtarget = subpath->pathtarget;
3765 : /* For now, assume we are above any joins, so no parameterization */
3766 8194 : pathnode->path.param_info = NULL;
3767 8194 : pathnode->path.parallel_aware = false;
3768 8194 : pathnode->path.parallel_safe = false;
3769 8194 : pathnode->path.parallel_workers = 0;
3770 8194 : pathnode->path.rows = subpath->rows;
3771 :
3772 : /*
3773 : * The result cannot be assumed sorted, since locking might cause the sort
3774 : * key columns to be replaced with new values.
3775 : */
3776 8194 : pathnode->path.pathkeys = NIL;
3777 :
3778 8194 : pathnode->subpath = subpath;
3779 8194 : pathnode->rowMarks = rowMarks;
3780 8194 : pathnode->epqParam = epqParam;
3781 :
3782 : /*
3783 : * We should charge something extra for the costs of row locking and
3784 : * possible refetches, but it's hard to say how much. For now, use
3785 : * cpu_tuple_cost per row.
3786 : */
3787 8194 : pathnode->path.disabled_nodes = subpath->disabled_nodes;
3788 8194 : pathnode->path.startup_cost = subpath->startup_cost;
3789 8194 : pathnode->path.total_cost = subpath->total_cost +
3790 8194 : cpu_tuple_cost * subpath->rows;
3791 :
3792 8194 : return pathnode;
3793 : }
3794 :
3795 : /*
3796 : * create_modifytable_path
3797 : * Creates a pathnode that represents performing INSERT/UPDATE/DELETE/MERGE
3798 : * mods
3799 : *
3800 : * 'rel' is the parent relation associated with the result
3801 : * 'subpath' is a Path producing source data
3802 : * 'operation' is the operation type
3803 : * 'canSetTag' is true if we set the command tag/es_processed
3804 : * 'nominalRelation' is the parent RT index for use of EXPLAIN
3805 : * 'rootRelation' is the partitioned/inherited table root RTI, or 0 if none
3806 : * 'partColsUpdated' is true if any partitioning columns are being updated,
3807 : * either from the target relation or a descendent partitioned table.
3808 : * 'resultRelations' is an integer list of actual RT indexes of target rel(s)
3809 : * 'updateColnosLists' is a list of UPDATE target column number lists
3810 : * (one sublist per rel); or NIL if not an UPDATE
3811 : * 'withCheckOptionLists' is a list of WCO lists (one per rel)
3812 : * 'returningLists' is a list of RETURNING tlists (one per rel)
3813 : * 'rowMarks' is a list of PlanRowMarks (non-locking only)
3814 : * 'onconflict' is the ON CONFLICT clause, or NULL
3815 : * 'epqParam' is the ID of Param for EvalPlanQual re-eval
3816 : * 'mergeActionLists' is a list of lists of MERGE actions (one per rel)
3817 : * 'mergeJoinConditions' is a list of join conditions for MERGE (one per rel)
3818 : */
3819 : ModifyTablePath *
3820 90332 : create_modifytable_path(PlannerInfo *root, RelOptInfo *rel,
3821 : Path *subpath,
3822 : CmdType operation, bool canSetTag,
3823 : Index nominalRelation, Index rootRelation,
3824 : bool partColsUpdated,
3825 : List *resultRelations,
3826 : List *updateColnosLists,
3827 : List *withCheckOptionLists, List *returningLists,
3828 : List *rowMarks, OnConflictExpr *onconflict,
3829 : List *mergeActionLists, List *mergeJoinConditions,
3830 : int epqParam)
3831 : {
3832 90332 : ModifyTablePath *pathnode = makeNode(ModifyTablePath);
3833 :
3834 : Assert(operation == CMD_MERGE ||
3835 : (operation == CMD_UPDATE ?
3836 : list_length(resultRelations) == list_length(updateColnosLists) :
3837 : updateColnosLists == NIL));
3838 : Assert(withCheckOptionLists == NIL ||
3839 : list_length(resultRelations) == list_length(withCheckOptionLists));
3840 : Assert(returningLists == NIL ||
3841 : list_length(resultRelations) == list_length(returningLists));
3842 :
3843 90332 : pathnode->path.pathtype = T_ModifyTable;
3844 90332 : pathnode->path.parent = rel;
3845 : /* pathtarget is not interesting, just make it minimally valid */
3846 90332 : pathnode->path.pathtarget = rel->reltarget;
3847 : /* For now, assume we are above any joins, so no parameterization */
3848 90332 : pathnode->path.param_info = NULL;
3849 90332 : pathnode->path.parallel_aware = false;
3850 90332 : pathnode->path.parallel_safe = false;
3851 90332 : pathnode->path.parallel_workers = 0;
3852 90332 : pathnode->path.pathkeys = NIL;
3853 :
3854 : /*
3855 : * Compute cost & rowcount as subpath cost & rowcount (if RETURNING)
3856 : *
3857 : * Currently, we don't charge anything extra for the actual table
3858 : * modification work, nor for the WITH CHECK OPTIONS or RETURNING
3859 : * expressions if any. It would only be window dressing, since
3860 : * ModifyTable is always a top-level node and there is no way for the
3861 : * costs to change any higher-level planning choices. But we might want
3862 : * to make it look better sometime.
3863 : */
3864 90332 : pathnode->path.disabled_nodes = subpath->disabled_nodes;
3865 90332 : pathnode->path.startup_cost = subpath->startup_cost;
3866 90332 : pathnode->path.total_cost = subpath->total_cost;
3867 90332 : if (returningLists != NIL)
3868 : {
3869 2464 : pathnode->path.rows = subpath->rows;
3870 :
3871 : /*
3872 : * Set width to match the subpath output. XXX this is totally wrong:
3873 : * we should return an average of the RETURNING tlist widths. But
3874 : * it's what happened historically, and improving it is a task for
3875 : * another day. (Again, it's mostly window dressing.)
3876 : */
3877 2464 : pathnode->path.pathtarget->width = subpath->pathtarget->width;
3878 : }
3879 : else
3880 : {
3881 87868 : pathnode->path.rows = 0;
3882 87868 : pathnode->path.pathtarget->width = 0;
3883 : }
3884 :
3885 90332 : pathnode->subpath = subpath;
3886 90332 : pathnode->operation = operation;
3887 90332 : pathnode->canSetTag = canSetTag;
3888 90332 : pathnode->nominalRelation = nominalRelation;
3889 90332 : pathnode->rootRelation = rootRelation;
3890 90332 : pathnode->partColsUpdated = partColsUpdated;
3891 90332 : pathnode->resultRelations = resultRelations;
3892 90332 : pathnode->updateColnosLists = updateColnosLists;
3893 90332 : pathnode->withCheckOptionLists = withCheckOptionLists;
3894 90332 : pathnode->returningLists = returningLists;
3895 90332 : pathnode->rowMarks = rowMarks;
3896 90332 : pathnode->onconflict = onconflict;
3897 90332 : pathnode->epqParam = epqParam;
3898 90332 : pathnode->mergeActionLists = mergeActionLists;
3899 90332 : pathnode->mergeJoinConditions = mergeJoinConditions;
3900 :
3901 90332 : return pathnode;
3902 : }
3903 :
3904 : /*
3905 : * create_limit_path
3906 : * Creates a pathnode that represents performing LIMIT/OFFSET
3907 : *
3908 : * In addition to providing the actual OFFSET and LIMIT expressions,
3909 : * the caller must provide estimates of their values for costing purposes.
3910 : * The estimates are as computed by preprocess_limit(), ie, 0 represents
3911 : * the clause not being present, and -1 means it's present but we could
3912 : * not estimate its value.
3913 : *
3914 : * 'rel' is the parent relation associated with the result
3915 : * 'subpath' is the path representing the source of data
3916 : * 'limitOffset' is the actual OFFSET expression, or NULL
3917 : * 'limitCount' is the actual LIMIT expression, or NULL
3918 : * 'offset_est' is the estimated value of the OFFSET expression
3919 : * 'count_est' is the estimated value of the LIMIT expression
3920 : */
3921 : LimitPath *
3922 5706 : create_limit_path(PlannerInfo *root, RelOptInfo *rel,
3923 : Path *subpath,
3924 : Node *limitOffset, Node *limitCount,
3925 : LimitOption limitOption,
3926 : int64 offset_est, int64 count_est)
3927 : {
3928 5706 : LimitPath *pathnode = makeNode(LimitPath);
3929 :
3930 5706 : pathnode->path.pathtype = T_Limit;
3931 5706 : pathnode->path.parent = rel;
3932 : /* Limit doesn't project, so use source path's pathtarget */
3933 5706 : pathnode->path.pathtarget = subpath->pathtarget;
3934 : /* For now, assume we are above any joins, so no parameterization */
3935 5706 : pathnode->path.param_info = NULL;
3936 5706 : pathnode->path.parallel_aware = false;
3937 7960 : pathnode->path.parallel_safe = rel->consider_parallel &&
3938 2254 : subpath->parallel_safe;
3939 5706 : pathnode->path.parallel_workers = subpath->parallel_workers;
3940 5706 : pathnode->path.rows = subpath->rows;
3941 5706 : pathnode->path.disabled_nodes = subpath->disabled_nodes;
3942 5706 : pathnode->path.startup_cost = subpath->startup_cost;
3943 5706 : pathnode->path.total_cost = subpath->total_cost;
3944 5706 : pathnode->path.pathkeys = subpath->pathkeys;
3945 5706 : pathnode->subpath = subpath;
3946 5706 : pathnode->limitOffset = limitOffset;
3947 5706 : pathnode->limitCount = limitCount;
3948 5706 : pathnode->limitOption = limitOption;
3949 :
3950 : /*
3951 : * Adjust the output rows count and costs according to the offset/limit.
3952 : */
3953 5706 : adjust_limit_rows_costs(&pathnode->path.rows,
3954 : &pathnode->path.startup_cost,
3955 : &pathnode->path.total_cost,
3956 : offset_est, count_est);
3957 :
3958 5706 : return pathnode;
3959 : }
3960 :
3961 : /*
3962 : * adjust_limit_rows_costs
3963 : * Adjust the size and cost estimates for a LimitPath node according to the
3964 : * offset/limit.
3965 : *
3966 : * This is only a cosmetic issue if we are at top level, but if we are
3967 : * building a subquery then it's important to report correct info to the outer
3968 : * planner.
3969 : *
3970 : * When the offset or count couldn't be estimated, use 10% of the estimated
3971 : * number of rows emitted from the subpath.
3972 : *
3973 : * XXX we don't bother to add eval costs of the offset/limit expressions
3974 : * themselves to the path costs. In theory we should, but in most cases those
3975 : * expressions are trivial and it's just not worth the trouble.
3976 : */
3977 : void
3978 5890 : adjust_limit_rows_costs(double *rows, /* in/out parameter */
3979 : Cost *startup_cost, /* in/out parameter */
3980 : Cost *total_cost, /* in/out parameter */
3981 : int64 offset_est,
3982 : int64 count_est)
3983 : {
3984 5890 : double input_rows = *rows;
3985 5890 : Cost input_startup_cost = *startup_cost;
3986 5890 : Cost input_total_cost = *total_cost;
3987 :
3988 5890 : if (offset_est != 0)
3989 : {
3990 : double offset_rows;
3991 :
3992 706 : if (offset_est > 0)
3993 682 : offset_rows = (double) offset_est;
3994 : else
3995 24 : offset_rows = clamp_row_est(input_rows * 0.10);
3996 706 : if (offset_rows > *rows)
3997 30 : offset_rows = *rows;
3998 706 : if (input_rows > 0)
3999 706 : *startup_cost +=
4000 706 : (input_total_cost - input_startup_cost)
4001 706 : * offset_rows / input_rows;
4002 706 : *rows -= offset_rows;
4003 706 : if (*rows < 1)
4004 30 : *rows = 1;
4005 : }
4006 :
4007 5890 : if (count_est != 0)
4008 : {
4009 : double count_rows;
4010 :
4011 5842 : if (count_est > 0)
4012 5836 : count_rows = (double) count_est;
4013 : else
4014 6 : count_rows = clamp_row_est(input_rows * 0.10);
4015 5842 : if (count_rows > *rows)
4016 234 : count_rows = *rows;
4017 5842 : if (input_rows > 0)
4018 5842 : *total_cost = *startup_cost +
4019 5842 : (input_total_cost - input_startup_cost)
4020 5842 : * count_rows / input_rows;
4021 5842 : *rows = count_rows;
4022 5842 : if (*rows < 1)
4023 0 : *rows = 1;
4024 : }
4025 5890 : }
4026 :
4027 :
4028 : /*
4029 : * reparameterize_path
4030 : * Attempt to modify a Path to have greater parameterization
4031 : *
4032 : * We use this to attempt to bring all child paths of an appendrel to the
4033 : * same parameterization level, ensuring that they all enforce the same set
4034 : * of join quals (and thus that that parameterization can be attributed to
4035 : * an append path built from such paths). Currently, only a few path types
4036 : * are supported here, though more could be added at need. We return NULL
4037 : * if we can't reparameterize the given path.
4038 : *
4039 : * Note: we intentionally do not pass created paths to add_path(); it would
4040 : * possibly try to delete them on the grounds of being cost-inferior to the
4041 : * paths they were made from, and we don't want that. Paths made here are
4042 : * not necessarily of general-purpose usefulness, but they can be useful
4043 : * as members of an append path.
4044 : */
4045 : Path *
4046 344 : reparameterize_path(PlannerInfo *root, Path *path,
4047 : Relids required_outer,
4048 : double loop_count)
4049 : {
4050 344 : RelOptInfo *rel = path->parent;
4051 :
4052 : /* Can only increase, not decrease, path's parameterization */
4053 344 : if (!bms_is_subset(PATH_REQ_OUTER(path), required_outer))
4054 0 : return NULL;
4055 344 : switch (path->pathtype)
4056 : {
4057 264 : case T_SeqScan:
4058 264 : return create_seqscan_path(root, rel, required_outer, 0);
4059 0 : case T_SampleScan:
4060 0 : return (Path *) create_samplescan_path(root, rel, required_outer);
4061 0 : case T_IndexScan:
4062 : case T_IndexOnlyScan:
4063 : {
4064 0 : IndexPath *ipath = (IndexPath *) path;
4065 0 : IndexPath *newpath = makeNode(IndexPath);
4066 :
4067 : /*
4068 : * We can't use create_index_path directly, and would not want
4069 : * to because it would re-compute the indexqual conditions
4070 : * which is wasted effort. Instead we hack things a bit:
4071 : * flat-copy the path node, revise its param_info, and redo
4072 : * the cost estimate.
4073 : */
4074 0 : memcpy(newpath, ipath, sizeof(IndexPath));
4075 0 : newpath->path.param_info =
4076 0 : get_baserel_parampathinfo(root, rel, required_outer);
4077 0 : cost_index(newpath, root, loop_count, false);
4078 0 : return (Path *) newpath;
4079 : }
4080 0 : case T_BitmapHeapScan:
4081 : {
4082 0 : BitmapHeapPath *bpath = (BitmapHeapPath *) path;
4083 :
4084 0 : return (Path *) create_bitmap_heap_path(root,
4085 : rel,
4086 : bpath->bitmapqual,
4087 : required_outer,
4088 : loop_count, 0);
4089 : }
4090 0 : case T_SubqueryScan:
4091 : {
4092 0 : SubqueryScanPath *spath = (SubqueryScanPath *) path;
4093 0 : Path *subpath = spath->subpath;
4094 : bool trivial_pathtarget;
4095 :
4096 : /*
4097 : * If existing node has zero extra cost, we must have decided
4098 : * its target is trivial. (The converse is not true, because
4099 : * it might have a trivial target but quals to enforce; but in
4100 : * that case the new node will too, so it doesn't matter
4101 : * whether we get the right answer here.)
4102 : */
4103 0 : trivial_pathtarget =
4104 0 : (subpath->total_cost == spath->path.total_cost);
4105 :
4106 0 : return (Path *) create_subqueryscan_path(root,
4107 : rel,
4108 : subpath,
4109 : trivial_pathtarget,
4110 : spath->path.pathkeys,
4111 : required_outer);
4112 : }
4113 48 : case T_Result:
4114 : /* Supported only for RTE_RESULT scan paths */
4115 48 : if (IsA(path, Path))
4116 48 : return create_resultscan_path(root, rel, required_outer);
4117 0 : break;
4118 0 : case T_Append:
4119 : {
4120 0 : AppendPath *apath = (AppendPath *) path;
4121 0 : List *childpaths = NIL;
4122 0 : List *partialpaths = NIL;
4123 : int i;
4124 : ListCell *lc;
4125 :
4126 : /* Reparameterize the children */
4127 0 : i = 0;
4128 0 : foreach(lc, apath->subpaths)
4129 : {
4130 0 : Path *spath = (Path *) lfirst(lc);
4131 :
4132 0 : spath = reparameterize_path(root, spath,
4133 : required_outer,
4134 : loop_count);
4135 0 : if (spath == NULL)
4136 0 : return NULL;
4137 : /* We have to re-split the regular and partial paths */
4138 0 : if (i < apath->first_partial_path)
4139 0 : childpaths = lappend(childpaths, spath);
4140 : else
4141 0 : partialpaths = lappend(partialpaths, spath);
4142 0 : i++;
4143 : }
4144 0 : return (Path *)
4145 0 : create_append_path(root, rel, childpaths, partialpaths,
4146 : apath->path.pathkeys, required_outer,
4147 : apath->path.parallel_workers,
4148 0 : apath->path.parallel_aware,
4149 : -1);
4150 : }
4151 0 : case T_Material:
4152 : {
4153 0 : MaterialPath *mpath = (MaterialPath *) path;
4154 0 : Path *spath = mpath->subpath;
4155 :
4156 0 : spath = reparameterize_path(root, spath,
4157 : required_outer,
4158 : loop_count);
4159 0 : if (spath == NULL)
4160 0 : return NULL;
4161 0 : return (Path *) create_material_path(rel, spath);
4162 : }
4163 0 : case T_Memoize:
4164 : {
4165 0 : MemoizePath *mpath = (MemoizePath *) path;
4166 0 : Path *spath = mpath->subpath;
4167 :
4168 0 : spath = reparameterize_path(root, spath,
4169 : required_outer,
4170 : loop_count);
4171 0 : if (spath == NULL)
4172 0 : return NULL;
4173 0 : return (Path *) create_memoize_path(root, rel,
4174 : spath,
4175 : mpath->param_exprs,
4176 : mpath->hash_operators,
4177 0 : mpath->singlerow,
4178 0 : mpath->binary_mode,
4179 : mpath->calls);
4180 : }
4181 32 : default:
4182 32 : break;
4183 : }
4184 32 : return NULL;
4185 : }
4186 :
4187 : /*
4188 : * reparameterize_path_by_child
4189 : * Given a path parameterized by the parent of the given child relation,
4190 : * translate the path to be parameterized by the given child relation.
4191 : *
4192 : * Most fields in the path are not changed, but any expressions must be
4193 : * adjusted to refer to the correct varnos, and any subpaths must be
4194 : * recursively reparameterized. Other fields that refer to specific relids
4195 : * also need adjustment.
4196 : *
4197 : * The cost, number of rows, width and parallel path properties depend upon
4198 : * path->parent, which does not change during the translation. So we need
4199 : * not change those.
4200 : *
4201 : * Currently, only a few path types are supported here, though more could be
4202 : * added at need. We return NULL if we can't reparameterize the given path.
4203 : *
4204 : * Note that this function can change referenced RangeTblEntries, RelOptInfos
4205 : * and IndexOptInfos as well as the Path structures. Therefore, it's only safe
4206 : * to call during create_plan(), when we have made a final choice of which Path
4207 : * to use for each RangeTblEntry/RelOptInfo/IndexOptInfo.
4208 : *
4209 : * Keep this code in sync with path_is_reparameterizable_by_child()!
4210 : */
4211 : Path *
4212 82084 : reparameterize_path_by_child(PlannerInfo *root, Path *path,
4213 : RelOptInfo *child_rel)
4214 : {
4215 : Path *new_path;
4216 : ParamPathInfo *new_ppi;
4217 : ParamPathInfo *old_ppi;
4218 : Relids required_outer;
4219 :
4220 : #define ADJUST_CHILD_ATTRS(node) \
4221 : ((node) = (void *) adjust_appendrel_attrs_multilevel(root, \
4222 : (Node *) (node), \
4223 : child_rel, \
4224 : child_rel->top_parent))
4225 :
4226 : #define REPARAMETERIZE_CHILD_PATH(path) \
4227 : do { \
4228 : (path) = reparameterize_path_by_child(root, (path), child_rel); \
4229 : if ((path) == NULL) \
4230 : return NULL; \
4231 : } while(0)
4232 :
4233 : #define REPARAMETERIZE_CHILD_PATH_LIST(pathlist) \
4234 : do { \
4235 : if ((pathlist) != NIL) \
4236 : { \
4237 : (pathlist) = reparameterize_pathlist_by_child(root, (pathlist), \
4238 : child_rel); \
4239 : if ((pathlist) == NIL) \
4240 : return NULL; \
4241 : } \
4242 : } while(0)
4243 :
4244 : /*
4245 : * If the path is not parameterized by the parent of the given relation,
4246 : * it doesn't need reparameterization.
4247 : */
4248 82084 : if (!path->param_info ||
4249 42630 : !bms_overlap(PATH_REQ_OUTER(path), child_rel->top_parent_relids))
4250 81196 : return path;
4251 :
4252 : /*
4253 : * If possible, reparameterize the given path.
4254 : *
4255 : * This function is currently only applied to the inner side of a nestloop
4256 : * join that is being partitioned by the partitionwise-join code. Hence,
4257 : * we need only support path types that plausibly arise in that context.
4258 : * (In particular, supporting sorted path types would be a waste of code
4259 : * and cycles: even if we translated them here, they'd just lose in
4260 : * subsequent cost comparisons.) If we do see an unsupported path type,
4261 : * that just means we won't be able to generate a partitionwise-join plan
4262 : * using that path type.
4263 : */
4264 888 : switch (nodeTag(path))
4265 : {
4266 228 : case T_Path:
4267 228 : new_path = path;
4268 228 : ADJUST_CHILD_ATTRS(new_path->parent->baserestrictinfo);
4269 228 : if (path->pathtype == T_SampleScan)
4270 : {
4271 48 : Index scan_relid = path->parent->relid;
4272 : RangeTblEntry *rte;
4273 :
4274 : /* it should be a base rel with a tablesample clause... */
4275 : Assert(scan_relid > 0);
4276 48 : rte = planner_rt_fetch(scan_relid, root);
4277 : Assert(rte->rtekind == RTE_RELATION);
4278 : Assert(rte->tablesample != NULL);
4279 :
4280 48 : ADJUST_CHILD_ATTRS(rte->tablesample);
4281 : }
4282 228 : break;
4283 :
4284 444 : case T_IndexPath:
4285 : {
4286 444 : IndexPath *ipath = (IndexPath *) path;
4287 :
4288 444 : ADJUST_CHILD_ATTRS(ipath->indexinfo->indrestrictinfo);
4289 444 : ADJUST_CHILD_ATTRS(ipath->indexclauses);
4290 444 : new_path = (Path *) ipath;
4291 : }
4292 444 : break;
4293 :
4294 48 : case T_BitmapHeapPath:
4295 : {
4296 48 : BitmapHeapPath *bhpath = (BitmapHeapPath *) path;
4297 :
4298 48 : ADJUST_CHILD_ATTRS(bhpath->path.parent->baserestrictinfo);
4299 48 : REPARAMETERIZE_CHILD_PATH(bhpath->bitmapqual);
4300 48 : new_path = (Path *) bhpath;
4301 : }
4302 48 : break;
4303 :
4304 24 : case T_BitmapAndPath:
4305 : {
4306 24 : BitmapAndPath *bapath = (BitmapAndPath *) path;
4307 :
4308 24 : REPARAMETERIZE_CHILD_PATH_LIST(bapath->bitmapquals);
4309 24 : new_path = (Path *) bapath;
4310 : }
4311 24 : break;
4312 :
4313 24 : case T_BitmapOrPath:
4314 : {
4315 24 : BitmapOrPath *bopath = (BitmapOrPath *) path;
4316 :
4317 24 : REPARAMETERIZE_CHILD_PATH_LIST(bopath->bitmapquals);
4318 24 : new_path = (Path *) bopath;
4319 : }
4320 24 : break;
4321 :
4322 0 : case T_ForeignPath:
4323 : {
4324 0 : ForeignPath *fpath = (ForeignPath *) path;
4325 : ReparameterizeForeignPathByChild_function rfpc_func;
4326 :
4327 0 : ADJUST_CHILD_ATTRS(fpath->path.parent->baserestrictinfo);
4328 0 : if (fpath->fdw_outerpath)
4329 0 : REPARAMETERIZE_CHILD_PATH(fpath->fdw_outerpath);
4330 0 : if (fpath->fdw_restrictinfo)
4331 0 : ADJUST_CHILD_ATTRS(fpath->fdw_restrictinfo);
4332 :
4333 : /* Hand over to FDW if needed. */
4334 0 : rfpc_func =
4335 0 : path->parent->fdwroutine->ReparameterizeForeignPathByChild;
4336 0 : if (rfpc_func)
4337 0 : fpath->fdw_private = rfpc_func(root, fpath->fdw_private,
4338 : child_rel);
4339 0 : new_path = (Path *) fpath;
4340 : }
4341 0 : break;
4342 :
4343 0 : case T_CustomPath:
4344 : {
4345 0 : CustomPath *cpath = (CustomPath *) path;
4346 :
4347 0 : ADJUST_CHILD_ATTRS(cpath->path.parent->baserestrictinfo);
4348 0 : REPARAMETERIZE_CHILD_PATH_LIST(cpath->custom_paths);
4349 0 : if (cpath->custom_restrictinfo)
4350 0 : ADJUST_CHILD_ATTRS(cpath->custom_restrictinfo);
4351 0 : if (cpath->methods &&
4352 0 : cpath->methods->ReparameterizeCustomPathByChild)
4353 0 : cpath->custom_private =
4354 0 : cpath->methods->ReparameterizeCustomPathByChild(root,
4355 : cpath->custom_private,
4356 : child_rel);
4357 0 : new_path = (Path *) cpath;
4358 : }
4359 0 : break;
4360 :
4361 36 : case T_NestPath:
4362 : {
4363 36 : NestPath *npath = (NestPath *) path;
4364 36 : JoinPath *jpath = (JoinPath *) npath;
4365 :
4366 36 : REPARAMETERIZE_CHILD_PATH(jpath->outerjoinpath);
4367 36 : REPARAMETERIZE_CHILD_PATH(jpath->innerjoinpath);
4368 36 : ADJUST_CHILD_ATTRS(jpath->joinrestrictinfo);
4369 36 : new_path = (Path *) npath;
4370 : }
4371 36 : break;
4372 :
4373 0 : case T_MergePath:
4374 : {
4375 0 : MergePath *mpath = (MergePath *) path;
4376 0 : JoinPath *jpath = (JoinPath *) mpath;
4377 :
4378 0 : REPARAMETERIZE_CHILD_PATH(jpath->outerjoinpath);
4379 0 : REPARAMETERIZE_CHILD_PATH(jpath->innerjoinpath);
4380 0 : ADJUST_CHILD_ATTRS(jpath->joinrestrictinfo);
4381 0 : ADJUST_CHILD_ATTRS(mpath->path_mergeclauses);
4382 0 : new_path = (Path *) mpath;
4383 : }
4384 0 : break;
4385 :
4386 48 : case T_HashPath:
4387 : {
4388 48 : HashPath *hpath = (HashPath *) path;
4389 48 : JoinPath *jpath = (JoinPath *) hpath;
4390 :
4391 48 : REPARAMETERIZE_CHILD_PATH(jpath->outerjoinpath);
4392 48 : REPARAMETERIZE_CHILD_PATH(jpath->innerjoinpath);
4393 48 : ADJUST_CHILD_ATTRS(jpath->joinrestrictinfo);
4394 48 : ADJUST_CHILD_ATTRS(hpath->path_hashclauses);
4395 48 : new_path = (Path *) hpath;
4396 : }
4397 48 : break;
4398 :
4399 24 : case T_AppendPath:
4400 : {
4401 24 : AppendPath *apath = (AppendPath *) path;
4402 :
4403 24 : REPARAMETERIZE_CHILD_PATH_LIST(apath->subpaths);
4404 24 : new_path = (Path *) apath;
4405 : }
4406 24 : break;
4407 :
4408 0 : case T_MaterialPath:
4409 : {
4410 0 : MaterialPath *mpath = (MaterialPath *) path;
4411 :
4412 0 : REPARAMETERIZE_CHILD_PATH(mpath->subpath);
4413 0 : new_path = (Path *) mpath;
4414 : }
4415 0 : break;
4416 :
4417 12 : case T_MemoizePath:
4418 : {
4419 12 : MemoizePath *mpath = (MemoizePath *) path;
4420 :
4421 12 : REPARAMETERIZE_CHILD_PATH(mpath->subpath);
4422 12 : ADJUST_CHILD_ATTRS(mpath->param_exprs);
4423 12 : new_path = (Path *) mpath;
4424 : }
4425 12 : break;
4426 :
4427 0 : case T_GatherPath:
4428 : {
4429 0 : GatherPath *gpath = (GatherPath *) path;
4430 :
4431 0 : REPARAMETERIZE_CHILD_PATH(gpath->subpath);
4432 0 : new_path = (Path *) gpath;
4433 : }
4434 0 : break;
4435 :
4436 0 : default:
4437 : /* We don't know how to reparameterize this path. */
4438 0 : return NULL;
4439 : }
4440 :
4441 : /*
4442 : * Adjust the parameterization information, which refers to the topmost
4443 : * parent. The topmost parent can be multiple levels away from the given
4444 : * child, hence use multi-level expression adjustment routines.
4445 : */
4446 888 : old_ppi = new_path->param_info;
4447 : required_outer =
4448 888 : adjust_child_relids_multilevel(root, old_ppi->ppi_req_outer,
4449 : child_rel,
4450 888 : child_rel->top_parent);
4451 :
4452 : /* If we already have a PPI for this parameterization, just return it */
4453 888 : new_ppi = find_param_path_info(new_path->parent, required_outer);
4454 :
4455 : /*
4456 : * If not, build a new one and link it to the list of PPIs. For the same
4457 : * reason as explained in mark_dummy_rel(), allocate new PPI in the same
4458 : * context the given RelOptInfo is in.
4459 : */
4460 888 : if (new_ppi == NULL)
4461 : {
4462 : MemoryContext oldcontext;
4463 780 : RelOptInfo *rel = path->parent;
4464 :
4465 780 : oldcontext = MemoryContextSwitchTo(GetMemoryChunkContext(rel));
4466 :
4467 780 : new_ppi = makeNode(ParamPathInfo);
4468 780 : new_ppi->ppi_req_outer = bms_copy(required_outer);
4469 780 : new_ppi->ppi_rows = old_ppi->ppi_rows;
4470 780 : new_ppi->ppi_clauses = old_ppi->ppi_clauses;
4471 780 : ADJUST_CHILD_ATTRS(new_ppi->ppi_clauses);
4472 780 : new_ppi->ppi_serials = bms_copy(old_ppi->ppi_serials);
4473 780 : rel->ppilist = lappend(rel->ppilist, new_ppi);
4474 :
4475 780 : MemoryContextSwitchTo(oldcontext);
4476 : }
4477 888 : bms_free(required_outer);
4478 :
4479 888 : new_path->param_info = new_ppi;
4480 :
4481 : /*
4482 : * Adjust the path target if the parent of the outer relation is
4483 : * referenced in the targetlist. This can happen when only the parent of
4484 : * outer relation is laterally referenced in this relation.
4485 : */
4486 888 : if (bms_overlap(path->parent->lateral_relids,
4487 888 : child_rel->top_parent_relids))
4488 : {
4489 480 : new_path->pathtarget = copy_pathtarget(new_path->pathtarget);
4490 480 : ADJUST_CHILD_ATTRS(new_path->pathtarget->exprs);
4491 : }
4492 :
4493 888 : return new_path;
4494 : }
4495 :
4496 : /*
4497 : * path_is_reparameterizable_by_child
4498 : * Given a path parameterized by the parent of the given child relation,
4499 : * see if it can be translated to be parameterized by the child relation.
4500 : *
4501 : * This must return true if and only if reparameterize_path_by_child()
4502 : * would succeed on this path. Currently it's sufficient to verify that
4503 : * the path and all of its subpaths (if any) are of the types handled by
4504 : * that function. However, subpaths that are not parameterized can be
4505 : * disregarded since they won't require translation.
4506 : */
4507 : bool
4508 31764 : path_is_reparameterizable_by_child(Path *path, RelOptInfo *child_rel)
4509 : {
4510 : #define REJECT_IF_PATH_NOT_REPARAMETERIZABLE(path) \
4511 : do { \
4512 : if (!path_is_reparameterizable_by_child(path, child_rel)) \
4513 : return false; \
4514 : } while(0)
4515 :
4516 : #define REJECT_IF_PATH_LIST_NOT_REPARAMETERIZABLE(pathlist) \
4517 : do { \
4518 : if (!pathlist_is_reparameterizable_by_child(pathlist, child_rel)) \
4519 : return false; \
4520 : } while(0)
4521 :
4522 : /*
4523 : * If the path is not parameterized by the parent of the given relation,
4524 : * it doesn't need reparameterization.
4525 : */
4526 31764 : if (!path->param_info ||
4527 31428 : !bms_overlap(PATH_REQ_OUTER(path), child_rel->top_parent_relids))
4528 912 : return true;
4529 :
4530 : /*
4531 : * Check that the path type is one that reparameterize_path_by_child() can
4532 : * handle, and recursively check subpaths.
4533 : */
4534 30852 : switch (nodeTag(path))
4535 : {
4536 20640 : case T_Path:
4537 : case T_IndexPath:
4538 20640 : break;
4539 :
4540 48 : case T_BitmapHeapPath:
4541 : {
4542 48 : BitmapHeapPath *bhpath = (BitmapHeapPath *) path;
4543 :
4544 48 : REJECT_IF_PATH_NOT_REPARAMETERIZABLE(bhpath->bitmapqual);
4545 : }
4546 48 : break;
4547 :
4548 24 : case T_BitmapAndPath:
4549 : {
4550 24 : BitmapAndPath *bapath = (BitmapAndPath *) path;
4551 :
4552 24 : REJECT_IF_PATH_LIST_NOT_REPARAMETERIZABLE(bapath->bitmapquals);
4553 : }
4554 24 : break;
4555 :
4556 24 : case T_BitmapOrPath:
4557 : {
4558 24 : BitmapOrPath *bopath = (BitmapOrPath *) path;
4559 :
4560 24 : REJECT_IF_PATH_LIST_NOT_REPARAMETERIZABLE(bopath->bitmapquals);
4561 : }
4562 24 : break;
4563 :
4564 148 : case T_ForeignPath:
4565 : {
4566 148 : ForeignPath *fpath = (ForeignPath *) path;
4567 :
4568 148 : if (fpath->fdw_outerpath)
4569 0 : REJECT_IF_PATH_NOT_REPARAMETERIZABLE(fpath->fdw_outerpath);
4570 : }
4571 148 : break;
4572 :
4573 0 : case T_CustomPath:
4574 : {
4575 0 : CustomPath *cpath = (CustomPath *) path;
4576 :
4577 0 : REJECT_IF_PATH_LIST_NOT_REPARAMETERIZABLE(cpath->custom_paths);
4578 : }
4579 0 : break;
4580 :
4581 1176 : case T_NestPath:
4582 : case T_MergePath:
4583 : case T_HashPath:
4584 : {
4585 1176 : JoinPath *jpath = (JoinPath *) path;
4586 :
4587 1176 : REJECT_IF_PATH_NOT_REPARAMETERIZABLE(jpath->outerjoinpath);
4588 1176 : REJECT_IF_PATH_NOT_REPARAMETERIZABLE(jpath->innerjoinpath);
4589 : }
4590 1176 : break;
4591 :
4592 192 : case T_AppendPath:
4593 : {
4594 192 : AppendPath *apath = (AppendPath *) path;
4595 :
4596 192 : REJECT_IF_PATH_LIST_NOT_REPARAMETERIZABLE(apath->subpaths);
4597 : }
4598 192 : break;
4599 :
4600 0 : case T_MaterialPath:
4601 : {
4602 0 : MaterialPath *mpath = (MaterialPath *) path;
4603 :
4604 0 : REJECT_IF_PATH_NOT_REPARAMETERIZABLE(mpath->subpath);
4605 : }
4606 0 : break;
4607 :
4608 8600 : case T_MemoizePath:
4609 : {
4610 8600 : MemoizePath *mpath = (MemoizePath *) path;
4611 :
4612 8600 : REJECT_IF_PATH_NOT_REPARAMETERIZABLE(mpath->subpath);
4613 : }
4614 8600 : break;
4615 :
4616 0 : case T_GatherPath:
4617 : {
4618 0 : GatherPath *gpath = (GatherPath *) path;
4619 :
4620 0 : REJECT_IF_PATH_NOT_REPARAMETERIZABLE(gpath->subpath);
4621 : }
4622 0 : break;
4623 :
4624 0 : default:
4625 : /* We don't know how to reparameterize this path. */
4626 0 : return false;
4627 : }
4628 :
4629 30852 : return true;
4630 : }
4631 :
4632 : /*
4633 : * reparameterize_pathlist_by_child
4634 : * Helper function to reparameterize a list of paths by given child rel.
4635 : *
4636 : * Returns NIL to indicate failure, so pathlist had better not be NIL.
4637 : */
4638 : static List *
4639 72 : reparameterize_pathlist_by_child(PlannerInfo *root,
4640 : List *pathlist,
4641 : RelOptInfo *child_rel)
4642 : {
4643 : ListCell *lc;
4644 72 : List *result = NIL;
4645 :
4646 216 : foreach(lc, pathlist)
4647 : {
4648 144 : Path *path = reparameterize_path_by_child(root, lfirst(lc),
4649 : child_rel);
4650 :
4651 144 : if (path == NULL)
4652 : {
4653 0 : list_free(result);
4654 0 : return NIL;
4655 : }
4656 :
4657 144 : result = lappend(result, path);
4658 : }
4659 :
4660 72 : return result;
4661 : }
4662 :
4663 : /*
4664 : * pathlist_is_reparameterizable_by_child
4665 : * Helper function to check if a list of paths can be reparameterized.
4666 : */
4667 : static bool
4668 240 : pathlist_is_reparameterizable_by_child(List *pathlist, RelOptInfo *child_rel)
4669 : {
4670 : ListCell *lc;
4671 :
4672 720 : foreach(lc, pathlist)
4673 : {
4674 480 : Path *path = (Path *) lfirst(lc);
4675 :
4676 480 : if (!path_is_reparameterizable_by_child(path, child_rel))
4677 0 : return false;
4678 : }
4679 :
4680 240 : return true;
4681 : }
|
